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         ext4_init_pending_tree(&ei->i_pending_tree);
1056 #ifdef CONFIG_QUOTA
1057         ei->i_reserved_quota = 0;
1058         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1059 #endif
1060         ei->jinode = NULL;
1061         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1062         spin_lock_init(&ei->i_completed_io_lock);
1063         ei->i_sync_tid = 0;
1064         ei->i_datasync_tid = 0;
1065         atomic_set(&ei->i_unwritten, 0);
1066         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1067         return &ei->vfs_inode;
1068 }
1069
1070 static int ext4_drop_inode(struct inode *inode)
1071 {
1072         int drop = generic_drop_inode(inode);
1073
1074         trace_ext4_drop_inode(inode, drop);
1075         return drop;
1076 }
1077
1078 static void ext4_i_callback(struct rcu_head *head)
1079 {
1080         struct inode *inode = container_of(head, struct inode, i_rcu);
1081         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1082 }
1083
1084 static void ext4_destroy_inode(struct inode *inode)
1085 {
1086         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1087                 ext4_msg(inode->i_sb, KERN_ERR,
1088                          "Inode %lu (%p): orphan list check failed!",
1089                          inode->i_ino, EXT4_I(inode));
1090                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1091                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1092                                 true);
1093                 dump_stack();
1094         }
1095         call_rcu(&inode->i_rcu, ext4_i_callback);
1096 }
1097
1098 static void init_once(void *foo)
1099 {
1100         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1101
1102         INIT_LIST_HEAD(&ei->i_orphan);
1103         init_rwsem(&ei->xattr_sem);
1104         init_rwsem(&ei->i_data_sem);
1105         init_rwsem(&ei->i_mmap_sem);
1106         inode_init_once(&ei->vfs_inode);
1107 }
1108
1109 static int __init init_inodecache(void)
1110 {
1111         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1112                                 sizeof(struct ext4_inode_info), 0,
1113                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1114                                         SLAB_ACCOUNT),
1115                                 offsetof(struct ext4_inode_info, i_data),
1116                                 sizeof_field(struct ext4_inode_info, i_data),
1117                                 init_once);
1118         if (ext4_inode_cachep == NULL)
1119                 return -ENOMEM;
1120         return 0;
1121 }
1122
1123 static void destroy_inodecache(void)
1124 {
1125         /*
1126          * Make sure all delayed rcu free inodes are flushed before we
1127          * destroy cache.
1128          */
1129         rcu_barrier();
1130         kmem_cache_destroy(ext4_inode_cachep);
1131 }
1132
1133 void ext4_clear_inode(struct inode *inode)
1134 {
1135         invalidate_inode_buffers(inode);
1136         clear_inode(inode);
1137         dquot_drop(inode);
1138         ext4_discard_preallocations(inode);
1139         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1140         if (EXT4_I(inode)->jinode) {
1141                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1142                                                EXT4_I(inode)->jinode);
1143                 jbd2_free_inode(EXT4_I(inode)->jinode);
1144                 EXT4_I(inode)->jinode = NULL;
1145         }
1146         fscrypt_put_encryption_info(inode);
1147 }
1148
1149 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1150                                         u64 ino, u32 generation)
1151 {
1152         struct inode *inode;
1153
1154         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1155                 return ERR_PTR(-ESTALE);
1156         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1157                 return ERR_PTR(-ESTALE);
1158
1159         /* iget isn't really right if the inode is currently unallocated!!
1160          *
1161          * ext4_read_inode will return a bad_inode if the inode had been
1162          * deleted, so we should be safe.
1163          *
1164          * Currently we don't know the generation for parent directory, so
1165          * a generation of 0 means "accept any"
1166          */
1167         inode = ext4_iget_normal(sb, ino);
1168         if (IS_ERR(inode))
1169                 return ERR_CAST(inode);
1170         if (generation && inode->i_generation != generation) {
1171                 iput(inode);
1172                 return ERR_PTR(-ESTALE);
1173         }
1174
1175         return inode;
1176 }
1177
1178 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1179                                         int fh_len, int fh_type)
1180 {
1181         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1182                                     ext4_nfs_get_inode);
1183 }
1184
1185 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1186                                         int fh_len, int fh_type)
1187 {
1188         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1189                                     ext4_nfs_get_inode);
1190 }
1191
1192 /*
1193  * Try to release metadata pages (indirect blocks, directories) which are
1194  * mapped via the block device.  Since these pages could have journal heads
1195  * which would prevent try_to_free_buffers() from freeing them, we must use
1196  * jbd2 layer's try_to_free_buffers() function to release them.
1197  */
1198 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1199                                  gfp_t wait)
1200 {
1201         journal_t *journal = EXT4_SB(sb)->s_journal;
1202
1203         WARN_ON(PageChecked(page));
1204         if (!page_has_buffers(page))
1205                 return 0;
1206         if (journal)
1207                 return jbd2_journal_try_to_free_buffers(journal, page,
1208                                                 wait & ~__GFP_DIRECT_RECLAIM);
1209         return try_to_free_buffers(page);
1210 }
1211
1212 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1213 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1214 {
1215         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1216                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1217 }
1218
1219 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1220                                                         void *fs_data)
1221 {
1222         handle_t *handle = fs_data;
1223         int res, res2, credits, retries = 0;
1224
1225         /*
1226          * Encrypting the root directory is not allowed because e2fsck expects
1227          * lost+found to exist and be unencrypted, and encrypting the root
1228          * directory would imply encrypting the lost+found directory as well as
1229          * the filename "lost+found" itself.
1230          */
1231         if (inode->i_ino == EXT4_ROOT_INO)
1232                 return -EPERM;
1233
1234         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1235                 return -EINVAL;
1236
1237         res = ext4_convert_inline_data(inode);
1238         if (res)
1239                 return res;
1240
1241         /*
1242          * If a journal handle was specified, then the encryption context is
1243          * being set on a new inode via inheritance and is part of a larger
1244          * transaction to create the inode.  Otherwise the encryption context is
1245          * being set on an existing inode in its own transaction.  Only in the
1246          * latter case should the "retry on ENOSPC" logic be used.
1247          */
1248
1249         if (handle) {
1250                 res = ext4_xattr_set_handle(handle, inode,
1251                                             EXT4_XATTR_INDEX_ENCRYPTION,
1252                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1253                                             ctx, len, 0);
1254                 if (!res) {
1255                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1256                         ext4_clear_inode_state(inode,
1257                                         EXT4_STATE_MAY_INLINE_DATA);
1258                         /*
1259                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1260                          * S_DAX may be disabled
1261                          */
1262                         ext4_set_inode_flags(inode);
1263                 }
1264                 return res;
1265         }
1266
1267         res = dquot_initialize(inode);
1268         if (res)
1269                 return res;
1270 retry:
1271         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1272                                      &credits);
1273         if (res)
1274                 return res;
1275
1276         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1277         if (IS_ERR(handle))
1278                 return PTR_ERR(handle);
1279
1280         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1281                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1282                                     ctx, len, 0);
1283         if (!res) {
1284                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1285                 /*
1286                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1287                  * S_DAX may be disabled
1288                  */
1289                 ext4_set_inode_flags(inode);
1290                 res = ext4_mark_inode_dirty(handle, inode);
1291                 if (res)
1292                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1293         }
1294         res2 = ext4_journal_stop(handle);
1295
1296         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1297                 goto retry;
1298         if (!res)
1299                 res = res2;
1300         return res;
1301 }
1302
1303 static bool ext4_dummy_context(struct inode *inode)
1304 {
1305         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1306 }
1307
1308 static const struct fscrypt_operations ext4_cryptops = {
1309         .key_prefix             = "ext4:",
1310         .get_context            = ext4_get_context,
1311         .set_context            = ext4_set_context,
1312         .dummy_context          = ext4_dummy_context,
1313         .empty_dir              = ext4_empty_dir,
1314         .max_namelen            = EXT4_NAME_LEN,
1315 };
1316 #endif
1317
1318 #ifdef CONFIG_QUOTA
1319 static const char * const quotatypes[] = INITQFNAMES;
1320 #define QTYPE2NAME(t) (quotatypes[t])
1321
1322 static int ext4_write_dquot(struct dquot *dquot);
1323 static int ext4_acquire_dquot(struct dquot *dquot);
1324 static int ext4_release_dquot(struct dquot *dquot);
1325 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1326 static int ext4_write_info(struct super_block *sb, int type);
1327 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1328                          const struct path *path);
1329 static int ext4_quota_on_mount(struct super_block *sb, int type);
1330 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1331                                size_t len, loff_t off);
1332 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1333                                 const char *data, size_t len, loff_t off);
1334 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1335                              unsigned int flags);
1336 static int ext4_enable_quotas(struct super_block *sb);
1337 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1338
1339 static struct dquot **ext4_get_dquots(struct inode *inode)
1340 {
1341         return EXT4_I(inode)->i_dquot;
1342 }
1343
1344 static const struct dquot_operations ext4_quota_operations = {
1345         .get_reserved_space     = ext4_get_reserved_space,
1346         .write_dquot            = ext4_write_dquot,
1347         .acquire_dquot          = ext4_acquire_dquot,
1348         .release_dquot          = ext4_release_dquot,
1349         .mark_dirty             = ext4_mark_dquot_dirty,
1350         .write_info             = ext4_write_info,
1351         .alloc_dquot            = dquot_alloc,
1352         .destroy_dquot          = dquot_destroy,
1353         .get_projid             = ext4_get_projid,
1354         .get_inode_usage        = ext4_get_inode_usage,
1355         .get_next_id            = ext4_get_next_id,
1356 };
1357
1358 static const struct quotactl_ops ext4_qctl_operations = {
1359         .quota_on       = ext4_quota_on,
1360         .quota_off      = ext4_quota_off,
1361         .quota_sync     = dquot_quota_sync,
1362         .get_state      = dquot_get_state,
1363         .set_info       = dquot_set_dqinfo,
1364         .get_dqblk      = dquot_get_dqblk,
1365         .set_dqblk      = dquot_set_dqblk,
1366         .get_nextdqblk  = dquot_get_next_dqblk,
1367 };
1368 #endif
1369
1370 static const struct super_operations ext4_sops = {
1371         .alloc_inode    = ext4_alloc_inode,
1372         .destroy_inode  = ext4_destroy_inode,
1373         .write_inode    = ext4_write_inode,
1374         .dirty_inode    = ext4_dirty_inode,
1375         .drop_inode     = ext4_drop_inode,
1376         .evict_inode    = ext4_evict_inode,
1377         .put_super      = ext4_put_super,
1378         .sync_fs        = ext4_sync_fs,
1379         .freeze_fs      = ext4_freeze,
1380         .unfreeze_fs    = ext4_unfreeze,
1381         .statfs         = ext4_statfs,
1382         .remount_fs     = ext4_remount,
1383         .show_options   = ext4_show_options,
1384 #ifdef CONFIG_QUOTA
1385         .quota_read     = ext4_quota_read,
1386         .quota_write    = ext4_quota_write,
1387         .get_dquots     = ext4_get_dquots,
1388 #endif
1389         .bdev_try_to_free_page = bdev_try_to_free_page,
1390 };
1391
1392 static const struct export_operations ext4_export_ops = {
1393         .fh_to_dentry = ext4_fh_to_dentry,
1394         .fh_to_parent = ext4_fh_to_parent,
1395         .get_parent = ext4_get_parent,
1396 };
1397
1398 enum {
1399         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1400         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1401         Opt_nouid32, Opt_debug, Opt_removed,
1402         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1403         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1404         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1405         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1406         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1407         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1408         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1409         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1410         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1411         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1412         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1413         Opt_nowarn_on_error, Opt_mblk_io_submit,
1414         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1415         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1416         Opt_inode_readahead_blks, Opt_journal_ioprio,
1417         Opt_dioread_nolock, Opt_dioread_lock,
1418         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1419         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1420 };
1421
1422 static const match_table_t tokens = {
1423         {Opt_bsd_df, "bsddf"},
1424         {Opt_minix_df, "minixdf"},
1425         {Opt_grpid, "grpid"},
1426         {Opt_grpid, "bsdgroups"},
1427         {Opt_nogrpid, "nogrpid"},
1428         {Opt_nogrpid, "sysvgroups"},
1429         {Opt_resgid, "resgid=%u"},
1430         {Opt_resuid, "resuid=%u"},
1431         {Opt_sb, "sb=%u"},
1432         {Opt_err_cont, "errors=continue"},
1433         {Opt_err_panic, "errors=panic"},
1434         {Opt_err_ro, "errors=remount-ro"},
1435         {Opt_nouid32, "nouid32"},
1436         {Opt_debug, "debug"},
1437         {Opt_removed, "oldalloc"},
1438         {Opt_removed, "orlov"},
1439         {Opt_user_xattr, "user_xattr"},
1440         {Opt_nouser_xattr, "nouser_xattr"},
1441         {Opt_acl, "acl"},
1442         {Opt_noacl, "noacl"},
1443         {Opt_noload, "norecovery"},
1444         {Opt_noload, "noload"},
1445         {Opt_removed, "nobh"},
1446         {Opt_removed, "bh"},
1447         {Opt_commit, "commit=%u"},
1448         {Opt_min_batch_time, "min_batch_time=%u"},
1449         {Opt_max_batch_time, "max_batch_time=%u"},
1450         {Opt_journal_dev, "journal_dev=%u"},
1451         {Opt_journal_path, "journal_path=%s"},
1452         {Opt_journal_checksum, "journal_checksum"},
1453         {Opt_nojournal_checksum, "nojournal_checksum"},
1454         {Opt_journal_async_commit, "journal_async_commit"},
1455         {Opt_abort, "abort"},
1456         {Opt_data_journal, "data=journal"},
1457         {Opt_data_ordered, "data=ordered"},
1458         {Opt_data_writeback, "data=writeback"},
1459         {Opt_data_err_abort, "data_err=abort"},
1460         {Opt_data_err_ignore, "data_err=ignore"},
1461         {Opt_offusrjquota, "usrjquota="},
1462         {Opt_usrjquota, "usrjquota=%s"},
1463         {Opt_offgrpjquota, "grpjquota="},
1464         {Opt_grpjquota, "grpjquota=%s"},
1465         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1466         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1467         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1468         {Opt_grpquota, "grpquota"},
1469         {Opt_noquota, "noquota"},
1470         {Opt_quota, "quota"},
1471         {Opt_usrquota, "usrquota"},
1472         {Opt_prjquota, "prjquota"},
1473         {Opt_barrier, "barrier=%u"},
1474         {Opt_barrier, "barrier"},
1475         {Opt_nobarrier, "nobarrier"},
1476         {Opt_i_version, "i_version"},
1477         {Opt_dax, "dax"},
1478         {Opt_stripe, "stripe=%u"},
1479         {Opt_delalloc, "delalloc"},
1480         {Opt_warn_on_error, "warn_on_error"},
1481         {Opt_nowarn_on_error, "nowarn_on_error"},
1482         {Opt_lazytime, "lazytime"},
1483         {Opt_nolazytime, "nolazytime"},
1484         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1485         {Opt_nodelalloc, "nodelalloc"},
1486         {Opt_removed, "mblk_io_submit"},
1487         {Opt_removed, "nomblk_io_submit"},
1488         {Opt_block_validity, "block_validity"},
1489         {Opt_noblock_validity, "noblock_validity"},
1490         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1491         {Opt_journal_ioprio, "journal_ioprio=%u"},
1492         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1493         {Opt_auto_da_alloc, "auto_da_alloc"},
1494         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1495         {Opt_dioread_nolock, "dioread_nolock"},
1496         {Opt_dioread_lock, "dioread_lock"},
1497         {Opt_discard, "discard"},
1498         {Opt_nodiscard, "nodiscard"},
1499         {Opt_init_itable, "init_itable=%u"},
1500         {Opt_init_itable, "init_itable"},
1501         {Opt_noinit_itable, "noinit_itable"},
1502         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1503         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1504         {Opt_nombcache, "nombcache"},
1505         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1506         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1507         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1508         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1509         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1510         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1511         {Opt_err, NULL},
1512 };
1513
1514 static ext4_fsblk_t get_sb_block(void **data)
1515 {
1516         ext4_fsblk_t    sb_block;
1517         char            *options = (char *) *data;
1518
1519         if (!options || strncmp(options, "sb=", 3) != 0)
1520                 return 1;       /* Default location */
1521
1522         options += 3;
1523         /* TODO: use simple_strtoll with >32bit ext4 */
1524         sb_block = simple_strtoul(options, &options, 0);
1525         if (*options && *options != ',') {
1526                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1527                        (char *) *data);
1528                 return 1;
1529         }
1530         if (*options == ',')
1531                 options++;
1532         *data = (void *) options;
1533
1534         return sb_block;
1535 }
1536
1537 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1538 static const char deprecated_msg[] =
1539         "Mount option \"%s\" will be removed by %s\n"
1540         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1541
1542 #ifdef CONFIG_QUOTA
1543 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1544 {
1545         struct ext4_sb_info *sbi = EXT4_SB(sb);
1546         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1547         int ret = -1;
1548
1549         if (sb_any_quota_loaded(sb) && !old_qname) {
1550                 ext4_msg(sb, KERN_ERR,
1551                         "Cannot change journaled "
1552                         "quota options when quota turned on");
1553                 return -1;
1554         }
1555         if (ext4_has_feature_quota(sb)) {
1556                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1557                          "ignored when QUOTA feature is enabled");
1558                 return 1;
1559         }
1560         qname = match_strdup(args);
1561         if (!qname) {
1562                 ext4_msg(sb, KERN_ERR,
1563                         "Not enough memory for storing quotafile name");
1564                 return -1;
1565         }
1566         if (old_qname) {
1567                 if (strcmp(old_qname, qname) == 0)
1568                         ret = 1;
1569                 else
1570                         ext4_msg(sb, KERN_ERR,
1571                                  "%s quota file already specified",
1572                                  QTYPE2NAME(qtype));
1573                 goto errout;
1574         }
1575         if (strchr(qname, '/')) {
1576                 ext4_msg(sb, KERN_ERR,
1577                         "quotafile must be on filesystem root");
1578                 goto errout;
1579         }
1580         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1581         set_opt(sb, QUOTA);
1582         return 1;
1583 errout:
1584         kfree(qname);
1585         return ret;
1586 }
1587
1588 static int clear_qf_name(struct super_block *sb, int qtype)
1589 {
1590
1591         struct ext4_sb_info *sbi = EXT4_SB(sb);
1592         char *old_qname = get_qf_name(sb, sbi, qtype);
1593
1594         if (sb_any_quota_loaded(sb) && old_qname) {
1595                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1596                         " when quota turned on");
1597                 return -1;
1598         }
1599         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1600         synchronize_rcu();
1601         kfree(old_qname);
1602         return 1;
1603 }
1604 #endif
1605
1606 #define MOPT_SET        0x0001
1607 #define MOPT_CLEAR      0x0002
1608 #define MOPT_NOSUPPORT  0x0004
1609 #define MOPT_EXPLICIT   0x0008
1610 #define MOPT_CLEAR_ERR  0x0010
1611 #define MOPT_GTE0       0x0020
1612 #ifdef CONFIG_QUOTA
1613 #define MOPT_Q          0
1614 #define MOPT_QFMT       0x0040
1615 #else
1616 #define MOPT_Q          MOPT_NOSUPPORT
1617 #define MOPT_QFMT       MOPT_NOSUPPORT
1618 #endif
1619 #define MOPT_DATAJ      0x0080
1620 #define MOPT_NO_EXT2    0x0100
1621 #define MOPT_NO_EXT3    0x0200
1622 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1623 #define MOPT_STRING     0x0400
1624
1625 static const struct mount_opts {
1626         int     token;
1627         int     mount_opt;
1628         int     flags;
1629 } ext4_mount_opts[] = {
1630         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1631         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1632         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1633         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1634         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1635         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1636         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1637          MOPT_EXT4_ONLY | MOPT_SET},
1638         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1639          MOPT_EXT4_ONLY | MOPT_CLEAR},
1640         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1641         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1642         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1643          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1644         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1645          MOPT_EXT4_ONLY | MOPT_CLEAR},
1646         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1647         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1648         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1649          MOPT_EXT4_ONLY | MOPT_CLEAR},
1650         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1651          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1652         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1653                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1654          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1655         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1656         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1657         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1658         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1659         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1660          MOPT_NO_EXT2},
1661         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1662          MOPT_NO_EXT2},
1663         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1664         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1665         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1666         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1667         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1668         {Opt_commit, 0, MOPT_GTE0},
1669         {Opt_max_batch_time, 0, MOPT_GTE0},
1670         {Opt_min_batch_time, 0, MOPT_GTE0},
1671         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1672         {Opt_init_itable, 0, MOPT_GTE0},
1673         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1674         {Opt_stripe, 0, MOPT_GTE0},
1675         {Opt_resuid, 0, MOPT_GTE0},
1676         {Opt_resgid, 0, MOPT_GTE0},
1677         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1678         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1679         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1680         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1681         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1682         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1683          MOPT_NO_EXT2 | MOPT_DATAJ},
1684         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1685         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1686 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1687         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1688         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1689 #else
1690         {Opt_acl, 0, MOPT_NOSUPPORT},
1691         {Opt_noacl, 0, MOPT_NOSUPPORT},
1692 #endif
1693         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1694         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1695         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1696         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1697         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1698                                                         MOPT_SET | MOPT_Q},
1699         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1700                                                         MOPT_SET | MOPT_Q},
1701         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1702                                                         MOPT_SET | MOPT_Q},
1703         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1704                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1705                                                         MOPT_CLEAR | MOPT_Q},
1706         {Opt_usrjquota, 0, MOPT_Q},
1707         {Opt_grpjquota, 0, MOPT_Q},
1708         {Opt_offusrjquota, 0, MOPT_Q},
1709         {Opt_offgrpjquota, 0, MOPT_Q},
1710         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1711         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1712         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1713         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1714         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1715         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1716         {Opt_err, 0, 0}
1717 };
1718
1719 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1720                             substring_t *args, unsigned long *journal_devnum,
1721                             unsigned int *journal_ioprio, int is_remount)
1722 {
1723         struct ext4_sb_info *sbi = EXT4_SB(sb);
1724         const struct mount_opts *m;
1725         kuid_t uid;
1726         kgid_t gid;
1727         int arg = 0;
1728
1729 #ifdef CONFIG_QUOTA
1730         if (token == Opt_usrjquota)
1731                 return set_qf_name(sb, USRQUOTA, &args[0]);
1732         else if (token == Opt_grpjquota)
1733                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1734         else if (token == Opt_offusrjquota)
1735                 return clear_qf_name(sb, USRQUOTA);
1736         else if (token == Opt_offgrpjquota)
1737                 return clear_qf_name(sb, GRPQUOTA);
1738 #endif
1739         switch (token) {
1740         case Opt_noacl:
1741         case Opt_nouser_xattr:
1742                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1743                 break;
1744         case Opt_sb:
1745                 return 1;       /* handled by get_sb_block() */
1746         case Opt_removed:
1747                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1748                 return 1;
1749         case Opt_abort:
1750                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1751                 return 1;
1752         case Opt_i_version:
1753                 sb->s_flags |= SB_I_VERSION;
1754                 return 1;
1755         case Opt_lazytime:
1756                 sb->s_flags |= SB_LAZYTIME;
1757                 return 1;
1758         case Opt_nolazytime:
1759                 sb->s_flags &= ~SB_LAZYTIME;
1760                 return 1;
1761         }
1762
1763         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1764                 if (token == m->token)
1765                         break;
1766
1767         if (m->token == Opt_err) {
1768                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1769                          "or missing value", opt);
1770                 return -1;
1771         }
1772
1773         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1774                 ext4_msg(sb, KERN_ERR,
1775                          "Mount option \"%s\" incompatible with ext2", opt);
1776                 return -1;
1777         }
1778         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1779                 ext4_msg(sb, KERN_ERR,
1780                          "Mount option \"%s\" incompatible with ext3", opt);
1781                 return -1;
1782         }
1783
1784         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1785                 return -1;
1786         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1787                 return -1;
1788         if (m->flags & MOPT_EXPLICIT) {
1789                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1790                         set_opt2(sb, EXPLICIT_DELALLOC);
1791                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1792                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1793                 } else
1794                         return -1;
1795         }
1796         if (m->flags & MOPT_CLEAR_ERR)
1797                 clear_opt(sb, ERRORS_MASK);
1798         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1799                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1800                          "options when quota turned on");
1801                 return -1;
1802         }
1803
1804         if (m->flags & MOPT_NOSUPPORT) {
1805                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1806         } else if (token == Opt_commit) {
1807                 if (arg == 0)
1808                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1809                 sbi->s_commit_interval = HZ * arg;
1810         } else if (token == Opt_debug_want_extra_isize) {
1811                 sbi->s_want_extra_isize = arg;
1812         } else if (token == Opt_max_batch_time) {
1813                 sbi->s_max_batch_time = arg;
1814         } else if (token == Opt_min_batch_time) {
1815                 sbi->s_min_batch_time = arg;
1816         } else if (token == Opt_inode_readahead_blks) {
1817                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1818                         ext4_msg(sb, KERN_ERR,
1819                                  "EXT4-fs: inode_readahead_blks must be "
1820                                  "0 or a power of 2 smaller than 2^31");
1821                         return -1;
1822                 }
1823                 sbi->s_inode_readahead_blks = arg;
1824         } else if (token == Opt_init_itable) {
1825                 set_opt(sb, INIT_INODE_TABLE);
1826                 if (!args->from)
1827                         arg = EXT4_DEF_LI_WAIT_MULT;
1828                 sbi->s_li_wait_mult = arg;
1829         } else if (token == Opt_max_dir_size_kb) {
1830                 sbi->s_max_dir_size_kb = arg;
1831         } else if (token == Opt_stripe) {
1832                 sbi->s_stripe = arg;
1833         } else if (token == Opt_resuid) {
1834                 uid = make_kuid(current_user_ns(), arg);
1835                 if (!uid_valid(uid)) {
1836                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1837                         return -1;
1838                 }
1839                 sbi->s_resuid = uid;
1840         } else if (token == Opt_resgid) {
1841                 gid = make_kgid(current_user_ns(), arg);
1842                 if (!gid_valid(gid)) {
1843                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1844                         return -1;
1845                 }
1846                 sbi->s_resgid = gid;
1847         } else if (token == Opt_journal_dev) {
1848                 if (is_remount) {
1849                         ext4_msg(sb, KERN_ERR,
1850                                  "Cannot specify journal on remount");
1851                         return -1;
1852                 }
1853                 *journal_devnum = arg;
1854         } else if (token == Opt_journal_path) {
1855                 char *journal_path;
1856                 struct inode *journal_inode;
1857                 struct path path;
1858                 int error;
1859
1860                 if (is_remount) {
1861                         ext4_msg(sb, KERN_ERR,
1862                                  "Cannot specify journal on remount");
1863                         return -1;
1864                 }
1865                 journal_path = match_strdup(&args[0]);
1866                 if (!journal_path) {
1867                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1868                                 "journal device string");
1869                         return -1;
1870                 }
1871
1872                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1873                 if (error) {
1874                         ext4_msg(sb, KERN_ERR, "error: could not find "
1875                                 "journal device path: error %d", error);
1876                         kfree(journal_path);
1877                         return -1;
1878                 }
1879
1880                 journal_inode = d_inode(path.dentry);
1881                 if (!S_ISBLK(journal_inode->i_mode)) {
1882                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1883                                 "is not a block device", journal_path);
1884                         path_put(&path);
1885                         kfree(journal_path);
1886                         return -1;
1887                 }
1888
1889                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1890                 path_put(&path);
1891                 kfree(journal_path);
1892         } else if (token == Opt_journal_ioprio) {
1893                 if (arg > 7) {
1894                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1895                                  " (must be 0-7)");
1896                         return -1;
1897                 }
1898                 *journal_ioprio =
1899                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1900         } else if (token == Opt_test_dummy_encryption) {
1901 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1902                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1903                 ext4_msg(sb, KERN_WARNING,
1904                          "Test dummy encryption mode enabled");
1905 #else
1906                 ext4_msg(sb, KERN_WARNING,
1907                          "Test dummy encryption mount option ignored");
1908 #endif
1909         } else if (m->flags & MOPT_DATAJ) {
1910                 if (is_remount) {
1911                         if (!sbi->s_journal)
1912                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1913                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1914                                 ext4_msg(sb, KERN_ERR,
1915                                          "Cannot change data mode on remount");
1916                                 return -1;
1917                         }
1918                 } else {
1919                         clear_opt(sb, DATA_FLAGS);
1920                         sbi->s_mount_opt |= m->mount_opt;
1921                 }
1922 #ifdef CONFIG_QUOTA
1923         } else if (m->flags & MOPT_QFMT) {
1924                 if (sb_any_quota_loaded(sb) &&
1925                     sbi->s_jquota_fmt != m->mount_opt) {
1926                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1927                                  "quota options when quota turned on");
1928                         return -1;
1929                 }
1930                 if (ext4_has_feature_quota(sb)) {
1931                         ext4_msg(sb, KERN_INFO,
1932                                  "Quota format mount options ignored "
1933                                  "when QUOTA feature is enabled");
1934                         return 1;
1935                 }
1936                 sbi->s_jquota_fmt = m->mount_opt;
1937 #endif
1938         } else if (token == Opt_dax) {
1939 #ifdef CONFIG_FS_DAX
1940                 ext4_msg(sb, KERN_WARNING,
1941                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1942                         sbi->s_mount_opt |= m->mount_opt;
1943 #else
1944                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1945                 return -1;
1946 #endif
1947         } else if (token == Opt_data_err_abort) {
1948                 sbi->s_mount_opt |= m->mount_opt;
1949         } else if (token == Opt_data_err_ignore) {
1950                 sbi->s_mount_opt &= ~m->mount_opt;
1951         } else {
1952                 if (!args->from)
1953                         arg = 1;
1954                 if (m->flags & MOPT_CLEAR)
1955                         arg = !arg;
1956                 else if (unlikely(!(m->flags & MOPT_SET))) {
1957                         ext4_msg(sb, KERN_WARNING,
1958                                  "buggy handling of option %s", opt);
1959                         WARN_ON(1);
1960                         return -1;
1961                 }
1962                 if (arg != 0)
1963                         sbi->s_mount_opt |= m->mount_opt;
1964                 else
1965                         sbi->s_mount_opt &= ~m->mount_opt;
1966         }
1967         return 1;
1968 }
1969
1970 static int parse_options(char *options, struct super_block *sb,
1971                          unsigned long *journal_devnum,
1972                          unsigned int *journal_ioprio,
1973                          int is_remount)
1974 {
1975         struct ext4_sb_info *sbi = EXT4_SB(sb);
1976         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
1977         substring_t args[MAX_OPT_ARGS];
1978         int token;
1979
1980         if (!options)
1981                 return 1;
1982
1983         while ((p = strsep(&options, ",")) != NULL) {
1984                 if (!*p)
1985                         continue;
1986                 /*
1987                  * Initialize args struct so we know whether arg was
1988                  * found; some options take optional arguments.
1989                  */
1990                 args[0].to = args[0].from = NULL;
1991                 token = match_token(p, tokens, args);
1992                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1993                                      journal_ioprio, is_remount) < 0)
1994                         return 0;
1995         }
1996 #ifdef CONFIG_QUOTA
1997         /*
1998          * We do the test below only for project quotas. 'usrquota' and
1999          * 'grpquota' mount options are allowed even without quota feature
2000          * to support legacy quotas in quota files.
2001          */
2002         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2003                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2004                          "Cannot enable project quota enforcement.");
2005                 return 0;
2006         }
2007         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2008         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2009         if (usr_qf_name || grp_qf_name) {
2010                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2011                         clear_opt(sb, USRQUOTA);
2012
2013                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2014                         clear_opt(sb, GRPQUOTA);
2015
2016                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2017                         ext4_msg(sb, KERN_ERR, "old and new quota "
2018                                         "format mixing");
2019                         return 0;
2020                 }
2021
2022                 if (!sbi->s_jquota_fmt) {
2023                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2024                                         "not specified");
2025                         return 0;
2026                 }
2027         }
2028 #endif
2029         if (test_opt(sb, DIOREAD_NOLOCK)) {
2030                 int blocksize =
2031                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2032
2033                 if (blocksize < PAGE_SIZE) {
2034                         ext4_msg(sb, KERN_ERR, "can't mount with "
2035                                  "dioread_nolock if block size != PAGE_SIZE");
2036                         return 0;
2037                 }
2038         }
2039         return 1;
2040 }
2041
2042 static inline void ext4_show_quota_options(struct seq_file *seq,
2043                                            struct super_block *sb)
2044 {
2045 #if defined(CONFIG_QUOTA)
2046         struct ext4_sb_info *sbi = EXT4_SB(sb);
2047         char *usr_qf_name, *grp_qf_name;
2048
2049         if (sbi->s_jquota_fmt) {
2050                 char *fmtname = "";
2051
2052                 switch (sbi->s_jquota_fmt) {
2053                 case QFMT_VFS_OLD:
2054                         fmtname = "vfsold";
2055                         break;
2056                 case QFMT_VFS_V0:
2057                         fmtname = "vfsv0";
2058                         break;
2059                 case QFMT_VFS_V1:
2060                         fmtname = "vfsv1";
2061                         break;
2062                 }
2063                 seq_printf(seq, ",jqfmt=%s", fmtname);
2064         }
2065
2066         rcu_read_lock();
2067         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2068         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2069         if (usr_qf_name)
2070                 seq_show_option(seq, "usrjquota", usr_qf_name);
2071         if (grp_qf_name)
2072                 seq_show_option(seq, "grpjquota", grp_qf_name);
2073         rcu_read_unlock();
2074 #endif
2075 }
2076
2077 static const char *token2str(int token)
2078 {
2079         const struct match_token *t;
2080
2081         for (t = tokens; t->token != Opt_err; t++)
2082                 if (t->token == token && !strchr(t->pattern, '='))
2083                         break;
2084         return t->pattern;
2085 }
2086
2087 /*
2088  * Show an option if
2089  *  - it's set to a non-default value OR
2090  *  - if the per-sb default is different from the global default
2091  */
2092 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2093                               int nodefs)
2094 {
2095         struct ext4_sb_info *sbi = EXT4_SB(sb);
2096         struct ext4_super_block *es = sbi->s_es;
2097         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2098         const struct mount_opts *m;
2099         char sep = nodefs ? '\n' : ',';
2100
2101 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2102 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2103
2104         if (sbi->s_sb_block != 1)
2105                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2106
2107         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2108                 int want_set = m->flags & MOPT_SET;
2109                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2110                     (m->flags & MOPT_CLEAR_ERR))
2111                         continue;
2112                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2113                         continue; /* skip if same as the default */
2114                 if ((want_set &&
2115                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2116                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2117                         continue; /* select Opt_noFoo vs Opt_Foo */
2118                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2119         }
2120
2121         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2122             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2123                 SEQ_OPTS_PRINT("resuid=%u",
2124                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2125         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2126             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2127                 SEQ_OPTS_PRINT("resgid=%u",
2128                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2129         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2130         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2131                 SEQ_OPTS_PUTS("errors=remount-ro");
2132         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2133                 SEQ_OPTS_PUTS("errors=continue");
2134         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2135                 SEQ_OPTS_PUTS("errors=panic");
2136         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2137                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2138         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2139                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2140         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2141                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2142         if (sb->s_flags & SB_I_VERSION)
2143                 SEQ_OPTS_PUTS("i_version");
2144         if (nodefs || sbi->s_stripe)
2145                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2146         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2147                         (sbi->s_mount_opt ^ def_mount_opt)) {
2148                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2149                         SEQ_OPTS_PUTS("data=journal");
2150                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2151                         SEQ_OPTS_PUTS("data=ordered");
2152                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2153                         SEQ_OPTS_PUTS("data=writeback");
2154         }
2155         if (nodefs ||
2156             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2157                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2158                                sbi->s_inode_readahead_blks);
2159
2160         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2161                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2162                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2163         if (nodefs || sbi->s_max_dir_size_kb)
2164                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2165         if (test_opt(sb, DATA_ERR_ABORT))
2166                 SEQ_OPTS_PUTS("data_err=abort");
2167         if (DUMMY_ENCRYPTION_ENABLED(sbi))
2168                 SEQ_OPTS_PUTS("test_dummy_encryption");
2169
2170         ext4_show_quota_options(seq, sb);
2171         return 0;
2172 }
2173
2174 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2175 {
2176         return _ext4_show_options(seq, root->d_sb, 0);
2177 }
2178
2179 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2180 {
2181         struct super_block *sb = seq->private;
2182         int rc;
2183
2184         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2185         rc = _ext4_show_options(seq, sb, 1);
2186         seq_puts(seq, "\n");
2187         return rc;
2188 }
2189
2190 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2191                             int read_only)
2192 {
2193         struct ext4_sb_info *sbi = EXT4_SB(sb);
2194         int err = 0;
2195
2196         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2197                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2198                          "forcing read-only mode");
2199                 err = -EROFS;
2200         }
2201         if (read_only)
2202                 goto done;
2203         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2204                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2205                          "running e2fsck is recommended");
2206         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2207                 ext4_msg(sb, KERN_WARNING,
2208                          "warning: mounting fs with errors, "
2209                          "running e2fsck is recommended");
2210         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2211                  le16_to_cpu(es->s_mnt_count) >=
2212                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2213                 ext4_msg(sb, KERN_WARNING,
2214                          "warning: maximal mount count reached, "
2215                          "running e2fsck is recommended");
2216         else if (le32_to_cpu(es->s_checkinterval) &&
2217                  (ext4_get_tstamp(es, s_lastcheck) +
2218                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2219                 ext4_msg(sb, KERN_WARNING,
2220                          "warning: checktime reached, "
2221                          "running e2fsck is recommended");
2222         if (!sbi->s_journal)
2223                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2224         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2225                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2226         le16_add_cpu(&es->s_mnt_count, 1);
2227         ext4_update_tstamp(es, s_mtime);
2228         ext4_update_dynamic_rev(sb);
2229         if (sbi->s_journal)
2230                 ext4_set_feature_journal_needs_recovery(sb);
2231
2232         err = ext4_commit_super(sb, 1);
2233 done:
2234         if (test_opt(sb, DEBUG))
2235                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2236                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2237                         sb->s_blocksize,
2238                         sbi->s_groups_count,
2239                         EXT4_BLOCKS_PER_GROUP(sb),
2240                         EXT4_INODES_PER_GROUP(sb),
2241                         sbi->s_mount_opt, sbi->s_mount_opt2);
2242
2243         cleancache_init_fs(sb);
2244         return err;
2245 }
2246
2247 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2248 {
2249         struct ext4_sb_info *sbi = EXT4_SB(sb);
2250         struct flex_groups *new_groups;
2251         int size;
2252
2253         if (!sbi->s_log_groups_per_flex)
2254                 return 0;
2255
2256         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2257         if (size <= sbi->s_flex_groups_allocated)
2258                 return 0;
2259
2260         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2261         new_groups = kvzalloc(size, GFP_KERNEL);
2262         if (!new_groups) {
2263                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2264                          size / (int) sizeof(struct flex_groups));
2265                 return -ENOMEM;
2266         }
2267
2268         if (sbi->s_flex_groups) {
2269                 memcpy(new_groups, sbi->s_flex_groups,
2270                        (sbi->s_flex_groups_allocated *
2271                         sizeof(struct flex_groups)));
2272                 kvfree(sbi->s_flex_groups);
2273         }
2274         sbi->s_flex_groups = new_groups;
2275         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2276         return 0;
2277 }
2278
2279 static int ext4_fill_flex_info(struct super_block *sb)
2280 {
2281         struct ext4_sb_info *sbi = EXT4_SB(sb);
2282         struct ext4_group_desc *gdp = NULL;
2283         ext4_group_t flex_group;
2284         int i, err;
2285
2286         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2287         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2288                 sbi->s_log_groups_per_flex = 0;
2289                 return 1;
2290         }
2291
2292         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2293         if (err)
2294                 goto failed;
2295
2296         for (i = 0; i < sbi->s_groups_count; i++) {
2297                 gdp = ext4_get_group_desc(sb, i, NULL);
2298
2299                 flex_group = ext4_flex_group(sbi, i);
2300                 atomic_add(ext4_free_inodes_count(sb, gdp),
2301                            &sbi->s_flex_groups[flex_group].free_inodes);
2302                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2303                              &sbi->s_flex_groups[flex_group].free_clusters);
2304                 atomic_add(ext4_used_dirs_count(sb, gdp),
2305                            &sbi->s_flex_groups[flex_group].used_dirs);
2306         }
2307
2308         return 1;
2309 failed:
2310         return 0;
2311 }
2312
2313 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2314                                    struct ext4_group_desc *gdp)
2315 {
2316         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2317         __u16 crc = 0;
2318         __le32 le_group = cpu_to_le32(block_group);
2319         struct ext4_sb_info *sbi = EXT4_SB(sb);
2320
2321         if (ext4_has_metadata_csum(sbi->s_sb)) {
2322                 /* Use new metadata_csum algorithm */
2323                 __u32 csum32;
2324                 __u16 dummy_csum = 0;
2325
2326                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2327                                      sizeof(le_group));
2328                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2329                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2330                                      sizeof(dummy_csum));
2331                 offset += sizeof(dummy_csum);
2332                 if (offset < sbi->s_desc_size)
2333                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2334                                              sbi->s_desc_size - offset);
2335
2336                 crc = csum32 & 0xFFFF;
2337                 goto out;
2338         }
2339
2340         /* old crc16 code */
2341         if (!ext4_has_feature_gdt_csum(sb))
2342                 return 0;
2343
2344         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2345         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2346         crc = crc16(crc, (__u8 *)gdp, offset);
2347         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2348         /* for checksum of struct ext4_group_desc do the rest...*/
2349         if (ext4_has_feature_64bit(sb) &&
2350             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2351                 crc = crc16(crc, (__u8 *)gdp + offset,
2352                             le16_to_cpu(sbi->s_es->s_desc_size) -
2353                                 offset);
2354
2355 out:
2356         return cpu_to_le16(crc);
2357 }
2358
2359 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2360                                 struct ext4_group_desc *gdp)
2361 {
2362         if (ext4_has_group_desc_csum(sb) &&
2363             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2364                 return 0;
2365
2366         return 1;
2367 }
2368
2369 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2370                               struct ext4_group_desc *gdp)
2371 {
2372         if (!ext4_has_group_desc_csum(sb))
2373                 return;
2374         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2375 }
2376
2377 /* Called at mount-time, super-block is locked */
2378 static int ext4_check_descriptors(struct super_block *sb,
2379                                   ext4_fsblk_t sb_block,
2380                                   ext4_group_t *first_not_zeroed)
2381 {
2382         struct ext4_sb_info *sbi = EXT4_SB(sb);
2383         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2384         ext4_fsblk_t last_block;
2385         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2386         ext4_fsblk_t block_bitmap;
2387         ext4_fsblk_t inode_bitmap;
2388         ext4_fsblk_t inode_table;
2389         int flexbg_flag = 0;
2390         ext4_group_t i, grp = sbi->s_groups_count;
2391
2392         if (ext4_has_feature_flex_bg(sb))
2393                 flexbg_flag = 1;
2394
2395         ext4_debug("Checking group descriptors");
2396
2397         for (i = 0; i < sbi->s_groups_count; i++) {
2398                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2399
2400                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2401                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2402                 else
2403                         last_block = first_block +
2404                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2405
2406                 if ((grp == sbi->s_groups_count) &&
2407                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2408                         grp = i;
2409
2410                 block_bitmap = ext4_block_bitmap(sb, gdp);
2411                 if (block_bitmap == sb_block) {
2412                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2413                                  "Block bitmap for group %u overlaps "
2414                                  "superblock", i);
2415                         if (!sb_rdonly(sb))
2416                                 return 0;
2417                 }
2418                 if (block_bitmap >= sb_block + 1 &&
2419                     block_bitmap <= last_bg_block) {
2420                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2421                                  "Block bitmap for group %u overlaps "
2422                                  "block group descriptors", i);
2423                         if (!sb_rdonly(sb))
2424                                 return 0;
2425                 }
2426                 if (block_bitmap < first_block || block_bitmap > last_block) {
2427                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2428                                "Block bitmap for group %u not in group "
2429                                "(block %llu)!", i, block_bitmap);
2430                         return 0;
2431                 }
2432                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2433                 if (inode_bitmap == sb_block) {
2434                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2435                                  "Inode bitmap for group %u overlaps "
2436                                  "superblock", i);
2437                         if (!sb_rdonly(sb))
2438                                 return 0;
2439                 }
2440                 if (inode_bitmap >= sb_block + 1 &&
2441                     inode_bitmap <= last_bg_block) {
2442                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2443                                  "Inode bitmap for group %u overlaps "
2444                                  "block group descriptors", i);
2445                         if (!sb_rdonly(sb))
2446                                 return 0;
2447                 }
2448                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2449                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2450                                "Inode bitmap for group %u not in group "
2451                                "(block %llu)!", i, inode_bitmap);
2452                         return 0;
2453                 }
2454                 inode_table = ext4_inode_table(sb, gdp);
2455                 if (inode_table == sb_block) {
2456                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2457                                  "Inode table for group %u overlaps "
2458                                  "superblock", i);
2459                         if (!sb_rdonly(sb))
2460                                 return 0;
2461                 }
2462                 if (inode_table >= sb_block + 1 &&
2463                     inode_table <= last_bg_block) {
2464                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2465                                  "Inode table for group %u overlaps "
2466                                  "block group descriptors", i);
2467                         if (!sb_rdonly(sb))
2468                                 return 0;
2469                 }
2470                 if (inode_table < first_block ||
2471                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2472                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2473                                "Inode table for group %u not in group "
2474                                "(block %llu)!", i, inode_table);
2475                         return 0;
2476                 }
2477                 ext4_lock_group(sb, i);
2478                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2479                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2480                                  "Checksum for group %u failed (%u!=%u)",
2481                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2482                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2483                         if (!sb_rdonly(sb)) {
2484                                 ext4_unlock_group(sb, i);
2485                                 return 0;
2486                         }
2487                 }
2488                 ext4_unlock_group(sb, i);
2489                 if (!flexbg_flag)
2490                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2491         }
2492         if (NULL != first_not_zeroed)
2493                 *first_not_zeroed = grp;
2494         return 1;
2495 }
2496
2497 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2498  * the superblock) which were deleted from all directories, but held open by
2499  * a process at the time of a crash.  We walk the list and try to delete these
2500  * inodes at recovery time (only with a read-write filesystem).
2501  *
2502  * In order to keep the orphan inode chain consistent during traversal (in
2503  * case of crash during recovery), we link each inode into the superblock
2504  * orphan list_head and handle it the same way as an inode deletion during
2505  * normal operation (which journals the operations for us).
2506  *
2507  * We only do an iget() and an iput() on each inode, which is very safe if we
2508  * accidentally point at an in-use or already deleted inode.  The worst that
2509  * can happen in this case is that we get a "bit already cleared" message from
2510  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2511  * e2fsck was run on this filesystem, and it must have already done the orphan
2512  * inode cleanup for us, so we can safely abort without any further action.
2513  */
2514 static void ext4_orphan_cleanup(struct super_block *sb,
2515                                 struct ext4_super_block *es)
2516 {
2517         unsigned int s_flags = sb->s_flags;
2518         int ret, nr_orphans = 0, nr_truncates = 0;
2519 #ifdef CONFIG_QUOTA
2520         int quota_update = 0;
2521         int i;
2522 #endif
2523         if (!es->s_last_orphan) {
2524                 jbd_debug(4, "no orphan inodes to clean up\n");
2525                 return;
2526         }
2527
2528         if (bdev_read_only(sb->s_bdev)) {
2529                 ext4_msg(sb, KERN_ERR, "write access "
2530                         "unavailable, skipping orphan cleanup");
2531                 return;
2532         }
2533
2534         /* Check if feature set would not allow a r/w mount */
2535         if (!ext4_feature_set_ok(sb, 0)) {
2536                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2537                          "unknown ROCOMPAT features");
2538                 return;
2539         }
2540
2541         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2542                 /* don't clear list on RO mount w/ errors */
2543                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2544                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2545                                   "clearing orphan list.\n");
2546                         es->s_last_orphan = 0;
2547                 }
2548                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2549                 return;
2550         }
2551
2552         if (s_flags & SB_RDONLY) {
2553                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2554                 sb->s_flags &= ~SB_RDONLY;
2555         }
2556 #ifdef CONFIG_QUOTA
2557         /* Needed for iput() to work correctly and not trash data */
2558         sb->s_flags |= SB_ACTIVE;
2559
2560         /*
2561          * Turn on quotas which were not enabled for read-only mounts if
2562          * filesystem has quota feature, so that they are updated correctly.
2563          */
2564         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2565                 int ret = ext4_enable_quotas(sb);
2566
2567                 if (!ret)
2568                         quota_update = 1;
2569                 else
2570                         ext4_msg(sb, KERN_ERR,
2571                                 "Cannot turn on quotas: error %d", ret);
2572         }
2573
2574         /* Turn on journaled quotas used for old sytle */
2575         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2576                 if (EXT4_SB(sb)->s_qf_names[i]) {
2577                         int ret = ext4_quota_on_mount(sb, i);
2578
2579                         if (!ret)
2580                                 quota_update = 1;
2581                         else
2582                                 ext4_msg(sb, KERN_ERR,
2583                                         "Cannot turn on journaled "
2584                                         "quota: type %d: error %d", i, ret);
2585                 }
2586         }
2587 #endif
2588
2589         while (es->s_last_orphan) {
2590                 struct inode *inode;
2591
2592                 /*
2593                  * We may have encountered an error during cleanup; if
2594                  * so, skip the rest.
2595                  */
2596                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2597                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2598                         es->s_last_orphan = 0;
2599                         break;
2600                 }
2601
2602                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2603                 if (IS_ERR(inode)) {
2604                         es->s_last_orphan = 0;
2605                         break;
2606                 }
2607
2608                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2609                 dquot_initialize(inode);
2610                 if (inode->i_nlink) {
2611                         if (test_opt(sb, DEBUG))
2612                                 ext4_msg(sb, KERN_DEBUG,
2613                                         "%s: truncating inode %lu to %lld bytes",
2614                                         __func__, inode->i_ino, inode->i_size);
2615                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2616                                   inode->i_ino, inode->i_size);
2617                         inode_lock(inode);
2618                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2619                         ret = ext4_truncate(inode);
2620                         if (ret)
2621                                 ext4_std_error(inode->i_sb, ret);
2622                         inode_unlock(inode);
2623                         nr_truncates++;
2624                 } else {
2625                         if (test_opt(sb, DEBUG))
2626                                 ext4_msg(sb, KERN_DEBUG,
2627                                         "%s: deleting unreferenced inode %lu",
2628                                         __func__, inode->i_ino);
2629                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2630                                   inode->i_ino);
2631                         nr_orphans++;
2632                 }
2633                 iput(inode);  /* The delete magic happens here! */
2634         }
2635
2636 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2637
2638         if (nr_orphans)
2639                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2640                        PLURAL(nr_orphans));
2641         if (nr_truncates)
2642                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2643                        PLURAL(nr_truncates));
2644 #ifdef CONFIG_QUOTA
2645         /* Turn off quotas if they were enabled for orphan cleanup */
2646         if (quota_update) {
2647                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2648                         if (sb_dqopt(sb)->files[i])
2649                                 dquot_quota_off(sb, i);
2650                 }
2651         }
2652 #endif
2653         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2654 }
2655
2656 /*
2657  * Maximal extent format file size.
2658  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2659  * extent format containers, within a sector_t, and within i_blocks
2660  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2661  * so that won't be a limiting factor.
2662  *
2663  * However there is other limiting factor. We do store extents in the form
2664  * of starting block and length, hence the resulting length of the extent
2665  * covering maximum file size must fit into on-disk format containers as
2666  * well. Given that length is always by 1 unit bigger than max unit (because
2667  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2668  *
2669  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2670  */
2671 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2672 {
2673         loff_t res;
2674         loff_t upper_limit = MAX_LFS_FILESIZE;
2675
2676         /* small i_blocks in vfs inode? */
2677         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2678                 /*
2679                  * CONFIG_LBDAF is not enabled implies the inode
2680                  * i_block represent total blocks in 512 bytes
2681                  * 32 == size of vfs inode i_blocks * 8
2682                  */
2683                 upper_limit = (1LL << 32) - 1;
2684
2685                 /* total blocks in file system block size */
2686                 upper_limit >>= (blkbits - 9);
2687                 upper_limit <<= blkbits;
2688         }
2689
2690         /*
2691          * 32-bit extent-start container, ee_block. We lower the maxbytes
2692          * by one fs block, so ee_len can cover the extent of maximum file
2693          * size
2694          */
2695         res = (1LL << 32) - 1;
2696         res <<= blkbits;
2697
2698         /* Sanity check against vm- & vfs- imposed limits */
2699         if (res > upper_limit)
2700                 res = upper_limit;
2701
2702         return res;
2703 }
2704
2705 /*
2706  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2707  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2708  * We need to be 1 filesystem block less than the 2^48 sector limit.
2709  */
2710 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2711 {
2712         loff_t res = EXT4_NDIR_BLOCKS;
2713         int meta_blocks;
2714         loff_t upper_limit;
2715         /* This is calculated to be the largest file size for a dense, block
2716          * mapped file such that the file's total number of 512-byte sectors,
2717          * including data and all indirect blocks, does not exceed (2^48 - 1).
2718          *
2719          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2720          * number of 512-byte sectors of the file.
2721          */
2722
2723         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2724                 /*
2725                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2726                  * the inode i_block field represents total file blocks in
2727                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2728                  */
2729                 upper_limit = (1LL << 32) - 1;
2730
2731                 /* total blocks in file system block size */
2732                 upper_limit >>= (bits - 9);
2733
2734         } else {
2735                 /*
2736                  * We use 48 bit ext4_inode i_blocks
2737                  * With EXT4_HUGE_FILE_FL set the i_blocks
2738                  * represent total number of blocks in
2739                  * file system block size
2740                  */
2741                 upper_limit = (1LL << 48) - 1;
2742
2743         }
2744
2745         /* indirect blocks */
2746         meta_blocks = 1;
2747         /* double indirect blocks */
2748         meta_blocks += 1 + (1LL << (bits-2));
2749         /* tripple indirect blocks */
2750         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2751
2752         upper_limit -= meta_blocks;
2753         upper_limit <<= bits;
2754
2755         res += 1LL << (bits-2);
2756         res += 1LL << (2*(bits-2));
2757         res += 1LL << (3*(bits-2));
2758         res <<= bits;
2759         if (res > upper_limit)
2760                 res = upper_limit;
2761
2762         if (res > MAX_LFS_FILESIZE)
2763                 res = MAX_LFS_FILESIZE;
2764
2765         return res;
2766 }
2767
2768 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2769                                    ext4_fsblk_t logical_sb_block, int nr)
2770 {
2771         struct ext4_sb_info *sbi = EXT4_SB(sb);
2772         ext4_group_t bg, first_meta_bg;
2773         int has_super = 0;
2774
2775         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2776
2777         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2778                 return logical_sb_block + nr + 1;
2779         bg = sbi->s_desc_per_block * nr;
2780         if (ext4_bg_has_super(sb, bg))
2781                 has_super = 1;
2782
2783         /*
2784          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2785          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2786          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2787          * compensate.
2788          */
2789         if (sb->s_blocksize == 1024 && nr == 0 &&
2790             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2791                 has_super++;
2792
2793         return (has_super + ext4_group_first_block_no(sb, bg));
2794 }
2795
2796 /**
2797  * ext4_get_stripe_size: Get the stripe size.
2798  * @sbi: In memory super block info
2799  *
2800  * If we have specified it via mount option, then
2801  * use the mount option value. If the value specified at mount time is
2802  * greater than the blocks per group use the super block value.
2803  * If the super block value is greater than blocks per group return 0.
2804  * Allocator needs it be less than blocks per group.
2805  *
2806  */
2807 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2808 {
2809         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2810         unsigned long stripe_width =
2811                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2812         int ret;
2813
2814         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2815                 ret = sbi->s_stripe;
2816         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2817                 ret = stripe_width;
2818         else if (stride && stride <= sbi->s_blocks_per_group)
2819                 ret = stride;
2820         else
2821                 ret = 0;
2822
2823         /*
2824          * If the stripe width is 1, this makes no sense and
2825          * we set it to 0 to turn off stripe handling code.
2826          */
2827         if (ret <= 1)
2828                 ret = 0;
2829
2830         return ret;
2831 }
2832
2833 /*
2834  * Check whether this filesystem can be mounted based on
2835  * the features present and the RDONLY/RDWR mount requested.
2836  * Returns 1 if this filesystem can be mounted as requested,
2837  * 0 if it cannot be.
2838  */
2839 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2840 {
2841         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2842                 ext4_msg(sb, KERN_ERR,
2843                         "Couldn't mount because of "
2844                         "unsupported optional features (%x)",
2845                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2846                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2847                 return 0;
2848         }
2849
2850         if (readonly)
2851                 return 1;
2852
2853         if (ext4_has_feature_readonly(sb)) {
2854                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2855                 sb->s_flags |= SB_RDONLY;
2856                 return 1;
2857         }
2858
2859         /* Check that feature set is OK for a read-write mount */
2860         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2861                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2862                          "unsupported optional features (%x)",
2863                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2864                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2865                 return 0;
2866         }
2867         /*
2868          * Large file size enabled file system can only be mounted
2869          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2870          */
2871         if (ext4_has_feature_huge_file(sb)) {
2872                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2873                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2874                                  "cannot be mounted RDWR without "
2875                                  "CONFIG_LBDAF");
2876                         return 0;
2877                 }
2878         }
2879         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2880                 ext4_msg(sb, KERN_ERR,
2881                          "Can't support bigalloc feature without "
2882                          "extents feature\n");
2883                 return 0;
2884         }
2885
2886 #ifndef CONFIG_QUOTA
2887         if (ext4_has_feature_quota(sb) && !readonly) {
2888                 ext4_msg(sb, KERN_ERR,
2889                          "Filesystem with quota feature cannot be mounted RDWR "
2890                          "without CONFIG_QUOTA");
2891                 return 0;
2892         }
2893         if (ext4_has_feature_project(sb) && !readonly) {
2894                 ext4_msg(sb, KERN_ERR,
2895                          "Filesystem with project quota feature cannot be mounted RDWR "
2896                          "without CONFIG_QUOTA");
2897                 return 0;
2898         }
2899 #endif  /* CONFIG_QUOTA */
2900         return 1;
2901 }
2902
2903 /*
2904  * This function is called once a day if we have errors logged
2905  * on the file system
2906  */
2907 static void print_daily_error_info(struct timer_list *t)
2908 {
2909         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
2910         struct super_block *sb = sbi->s_sb;
2911         struct ext4_super_block *es = sbi->s_es;
2912
2913         if (es->s_error_count)
2914                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2915                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2916                          le32_to_cpu(es->s_error_count));
2917         if (es->s_first_error_time) {
2918                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
2919                        sb->s_id,
2920                        ext4_get_tstamp(es, s_first_error_time),
2921                        (int) sizeof(es->s_first_error_func),
2922                        es->s_first_error_func,
2923                        le32_to_cpu(es->s_first_error_line));
2924                 if (es->s_first_error_ino)
2925                         printk(KERN_CONT ": inode %u",
2926                                le32_to_cpu(es->s_first_error_ino));
2927                 if (es->s_first_error_block)
2928                         printk(KERN_CONT ": block %llu", (unsigned long long)
2929                                le64_to_cpu(es->s_first_error_block));
2930                 printk(KERN_CONT "\n");
2931         }
2932         if (es->s_last_error_time) {
2933                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
2934                        sb->s_id,
2935                        ext4_get_tstamp(es, s_last_error_time),
2936                        (int) sizeof(es->s_last_error_func),
2937                        es->s_last_error_func,
2938                        le32_to_cpu(es->s_last_error_line));
2939                 if (es->s_last_error_ino)
2940                         printk(KERN_CONT ": inode %u",
2941                                le32_to_cpu(es->s_last_error_ino));
2942                 if (es->s_last_error_block)
2943                         printk(KERN_CONT ": block %llu", (unsigned long long)
2944                                le64_to_cpu(es->s_last_error_block));
2945                 printk(KERN_CONT "\n");
2946         }
2947         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2948 }
2949
2950 /* Find next suitable group and run ext4_init_inode_table */
2951 static int ext4_run_li_request(struct ext4_li_request *elr)
2952 {
2953         struct ext4_group_desc *gdp = NULL;
2954         ext4_group_t group, ngroups;
2955         struct super_block *sb;
2956         unsigned long timeout = 0;
2957         int ret = 0;
2958
2959         sb = elr->lr_super;
2960         ngroups = EXT4_SB(sb)->s_groups_count;
2961
2962         for (group = elr->lr_next_group; group < ngroups; group++) {
2963                 gdp = ext4_get_group_desc(sb, group, NULL);
2964                 if (!gdp) {
2965                         ret = 1;
2966                         break;
2967                 }
2968
2969                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2970                         break;
2971         }
2972
2973         if (group >= ngroups)
2974                 ret = 1;
2975
2976         if (!ret) {
2977                 timeout = jiffies;
2978                 ret = ext4_init_inode_table(sb, group,
2979                                             elr->lr_timeout ? 0 : 1);
2980                 if (elr->lr_timeout == 0) {
2981                         timeout = (jiffies - timeout) *
2982                                   elr->lr_sbi->s_li_wait_mult;
2983                         elr->lr_timeout = timeout;
2984                 }
2985                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2986                 elr->lr_next_group = group + 1;
2987         }
2988         return ret;
2989 }
2990
2991 /*
2992  * Remove lr_request from the list_request and free the
2993  * request structure. Should be called with li_list_mtx held
2994  */
2995 static void ext4_remove_li_request(struct ext4_li_request *elr)
2996 {
2997         struct ext4_sb_info *sbi;
2998
2999         if (!elr)
3000                 return;
3001
3002         sbi = elr->lr_sbi;
3003
3004         list_del(&elr->lr_request);
3005         sbi->s_li_request = NULL;
3006         kfree(elr);
3007 }
3008
3009 static void ext4_unregister_li_request(struct super_block *sb)
3010 {
3011         mutex_lock(&ext4_li_mtx);
3012         if (!ext4_li_info) {
3013                 mutex_unlock(&ext4_li_mtx);
3014                 return;
3015         }
3016
3017         mutex_lock(&ext4_li_info->li_list_mtx);
3018         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3019         mutex_unlock(&ext4_li_info->li_list_mtx);
3020         mutex_unlock(&ext4_li_mtx);
3021 }
3022
3023 static struct task_struct *ext4_lazyinit_task;
3024
3025 /*
3026  * This is the function where ext4lazyinit thread lives. It walks
3027  * through the request list searching for next scheduled filesystem.
3028  * When such a fs is found, run the lazy initialization request
3029  * (ext4_rn_li_request) and keep track of the time spend in this
3030  * function. Based on that time we compute next schedule time of
3031  * the request. When walking through the list is complete, compute
3032  * next waking time and put itself into sleep.
3033  */
3034 static int ext4_lazyinit_thread(void *arg)
3035 {
3036         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3037         struct list_head *pos, *n;
3038         struct ext4_li_request *elr;
3039         unsigned long next_wakeup, cur;
3040
3041         BUG_ON(NULL == eli);
3042
3043 cont_thread:
3044         while (true) {
3045                 next_wakeup = MAX_JIFFY_OFFSET;
3046
3047                 mutex_lock(&eli->li_list_mtx);
3048                 if (list_empty(&eli->li_request_list)) {
3049                         mutex_unlock(&eli->li_list_mtx);
3050                         goto exit_thread;
3051                 }
3052                 list_for_each_safe(pos, n, &eli->li_request_list) {
3053                         int err = 0;
3054                         int progress = 0;
3055                         elr = list_entry(pos, struct ext4_li_request,
3056                                          lr_request);
3057
3058                         if (time_before(jiffies, elr->lr_next_sched)) {
3059                                 if (time_before(elr->lr_next_sched, next_wakeup))
3060                                         next_wakeup = elr->lr_next_sched;
3061                                 continue;
3062                         }
3063                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3064                                 if (sb_start_write_trylock(elr->lr_super)) {
3065                                         progress = 1;
3066                                         /*
3067                                          * We hold sb->s_umount, sb can not
3068                                          * be removed from the list, it is
3069                                          * now safe to drop li_list_mtx
3070                                          */
3071                                         mutex_unlock(&eli->li_list_mtx);
3072                                         err = ext4_run_li_request(elr);
3073                                         sb_end_write(elr->lr_super);
3074                                         mutex_lock(&eli->li_list_mtx);
3075                                         n = pos->next;
3076                                 }
3077                                 up_read((&elr->lr_super->s_umount));
3078                         }
3079                         /* error, remove the lazy_init job */
3080                         if (err) {
3081                                 ext4_remove_li_request(elr);
3082                                 continue;
3083                         }
3084                         if (!progress) {
3085                                 elr->lr_next_sched = jiffies +
3086                                         (prandom_u32()
3087                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3088                         }
3089                         if (time_before(elr->lr_next_sched, next_wakeup))
3090                                 next_wakeup = elr->lr_next_sched;
3091                 }
3092                 mutex_unlock(&eli->li_list_mtx);
3093
3094                 try_to_freeze();
3095
3096                 cur = jiffies;
3097                 if ((time_after_eq(cur, next_wakeup)) ||
3098                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3099                         cond_resched();
3100                         continue;
3101                 }
3102
3103                 schedule_timeout_interruptible(next_wakeup - cur);
3104
3105                 if (kthread_should_stop()) {
3106                         ext4_clear_request_list();
3107                         goto exit_thread;
3108                 }
3109         }
3110
3111 exit_thread:
3112         /*
3113          * It looks like the request list is empty, but we need
3114          * to check it under the li_list_mtx lock, to prevent any
3115          * additions into it, and of course we should lock ext4_li_mtx
3116          * to atomically free the list and ext4_li_info, because at
3117          * this point another ext4 filesystem could be registering
3118          * new one.
3119          */
3120         mutex_lock(&ext4_li_mtx);
3121         mutex_lock(&eli->li_list_mtx);
3122         if (!list_empty(&eli->li_request_list)) {
3123                 mutex_unlock(&eli->li_list_mtx);
3124                 mutex_unlock(&ext4_li_mtx);
3125                 goto cont_thread;
3126         }
3127         mutex_unlock(&eli->li_list_mtx);
3128         kfree(ext4_li_info);
3129         ext4_li_info = NULL;
3130         mutex_unlock(&ext4_li_mtx);
3131
3132         return 0;
3133 }
3134
3135 static void ext4_clear_request_list(void)
3136 {
3137         struct list_head *pos, *n;
3138         struct ext4_li_request *elr;
3139
3140         mutex_lock(&ext4_li_info->li_list_mtx);
3141         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3142                 elr = list_entry(pos, struct ext4_li_request,
3143                                  lr_request);
3144                 ext4_remove_li_request(elr);
3145         }
3146         mutex_unlock(&ext4_li_info->li_list_mtx);
3147 }
3148
3149 static int ext4_run_lazyinit_thread(void)
3150 {
3151         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3152                                          ext4_li_info, "ext4lazyinit");
3153         if (IS_ERR(ext4_lazyinit_task)) {
3154                 int err = PTR_ERR(ext4_lazyinit_task);
3155                 ext4_clear_request_list();
3156                 kfree(ext4_li_info);
3157                 ext4_li_info = NULL;
3158                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3159                                  "initialization thread\n",
3160                                  err);
3161                 return err;
3162         }
3163         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3164         return 0;
3165 }
3166
3167 /*
3168  * Check whether it make sense to run itable init. thread or not.
3169  * If there is at least one uninitialized inode table, return
3170  * corresponding group number, else the loop goes through all
3171  * groups and return total number of groups.
3172  */
3173 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3174 {
3175         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3176         struct ext4_group_desc *gdp = NULL;
3177
3178         if (!ext4_has_group_desc_csum(sb))
3179                 return ngroups;
3180
3181         for (group = 0; group < ngroups; group++) {
3182                 gdp = ext4_get_group_desc(sb, group, NULL);
3183                 if (!gdp)
3184                         continue;
3185
3186                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3187                         break;
3188         }
3189
3190         return group;
3191 }
3192
3193 static int ext4_li_info_new(void)
3194 {
3195         struct ext4_lazy_init *eli = NULL;
3196
3197         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3198         if (!eli)
3199                 return -ENOMEM;
3200
3201         INIT_LIST_HEAD(&eli->li_request_list);
3202         mutex_init(&eli->li_list_mtx);
3203
3204         eli->li_state |= EXT4_LAZYINIT_QUIT;
3205
3206         ext4_li_info = eli;
3207
3208         return 0;
3209 }
3210
3211 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3212                                             ext4_group_t start)
3213 {
3214         struct ext4_sb_info *sbi = EXT4_SB(sb);
3215         struct ext4_li_request *elr;
3216
3217         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3218         if (!elr)
3219                 return NULL;
3220
3221         elr->lr_super = sb;
3222         elr->lr_sbi = sbi;
3223         elr->lr_next_group = start;
3224
3225         /*
3226          * Randomize first schedule time of the request to
3227          * spread the inode table initialization requests
3228          * better.
3229          */
3230         elr->lr_next_sched = jiffies + (prandom_u32() %
3231                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3232         return elr;
3233 }
3234
3235 int ext4_register_li_request(struct super_block *sb,
3236                              ext4_group_t first_not_zeroed)
3237 {
3238         struct ext4_sb_info *sbi = EXT4_SB(sb);
3239         struct ext4_li_request *elr = NULL;
3240         ext4_group_t ngroups = sbi->s_groups_count;
3241         int ret = 0;
3242
3243         mutex_lock(&ext4_li_mtx);
3244         if (sbi->s_li_request != NULL) {
3245                 /*
3246                  * Reset timeout so it can be computed again, because
3247                  * s_li_wait_mult might have changed.
3248                  */
3249                 sbi->s_li_request->lr_timeout = 0;
3250                 goto out;
3251         }
3252
3253         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3254             !test_opt(sb, INIT_INODE_TABLE))
3255                 goto out;
3256
3257         elr = ext4_li_request_new(sb, first_not_zeroed);
3258         if (!elr) {
3259                 ret = -ENOMEM;
3260                 goto out;
3261         }
3262
3263         if (NULL == ext4_li_info) {
3264                 ret = ext4_li_info_new();
3265                 if (ret)
3266                         goto out;
3267         }
3268
3269         mutex_lock(&ext4_li_info->li_list_mtx);
3270         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3271         mutex_unlock(&ext4_li_info->li_list_mtx);
3272
3273         sbi->s_li_request = elr;
3274         /*
3275          * set elr to NULL here since it has been inserted to
3276          * the request_list and the removal and free of it is
3277          * handled by ext4_clear_request_list from now on.
3278          */
3279         elr = NULL;
3280
3281         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3282                 ret = ext4_run_lazyinit_thread();
3283                 if (ret)
3284                         goto out;
3285         }
3286 out:
3287         mutex_unlock(&ext4_li_mtx);
3288         if (ret)
3289                 kfree(elr);
3290         return ret;
3291 }
3292
3293 /*
3294  * We do not need to lock anything since this is called on
3295  * module unload.
3296  */
3297 static void ext4_destroy_lazyinit_thread(void)
3298 {
3299         /*
3300          * If thread exited earlier
3301          * there's nothing to be done.
3302          */
3303         if (!ext4_li_info || !ext4_lazyinit_task)
3304                 return;
3305
3306         kthread_stop(ext4_lazyinit_task);
3307 }
3308
3309 static int set_journal_csum_feature_set(struct super_block *sb)
3310 {
3311         int ret = 1;
3312         int compat, incompat;
3313         struct ext4_sb_info *sbi = EXT4_SB(sb);
3314
3315         if (ext4_has_metadata_csum(sb)) {
3316                 /* journal checksum v3 */
3317                 compat = 0;
3318                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3319         } else {
3320                 /* journal checksum v1 */
3321                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3322                 incompat = 0;
3323         }
3324
3325         jbd2_journal_clear_features(sbi->s_journal,
3326                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3327                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3328                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3329         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3330                 ret = jbd2_journal_set_features(sbi->s_journal,
3331                                 compat, 0,
3332                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3333                                 incompat);
3334         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3335                 ret = jbd2_journal_set_features(sbi->s_journal,
3336                                 compat, 0,
3337                                 incompat);
3338                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3339                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3340         } else {
3341                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3342                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3343         }
3344
3345         return ret;
3346 }
3347
3348 /*
3349  * Note: calculating the overhead so we can be compatible with
3350  * historical BSD practice is quite difficult in the face of
3351  * clusters/bigalloc.  This is because multiple metadata blocks from
3352  * different block group can end up in the same allocation cluster.
3353  * Calculating the exact overhead in the face of clustered allocation
3354  * requires either O(all block bitmaps) in memory or O(number of block
3355  * groups**2) in time.  We will still calculate the superblock for
3356  * older file systems --- and if we come across with a bigalloc file
3357  * system with zero in s_overhead_clusters the estimate will be close to
3358  * correct especially for very large cluster sizes --- but for newer
3359  * file systems, it's better to calculate this figure once at mkfs
3360  * time, and store it in the superblock.  If the superblock value is
3361  * present (even for non-bigalloc file systems), we will use it.
3362  */
3363 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3364                           char *buf)
3365 {
3366         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3367         struct ext4_group_desc  *gdp;
3368         ext4_fsblk_t            first_block, last_block, b;
3369         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3370         int                     s, j, count = 0;
3371
3372         if (!ext4_has_feature_bigalloc(sb))
3373                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3374                         sbi->s_itb_per_group + 2);
3375
3376         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3377                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3378         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3379         for (i = 0; i < ngroups; i++) {
3380                 gdp = ext4_get_group_desc(sb, i, NULL);
3381                 b = ext4_block_bitmap(sb, gdp);
3382                 if (b >= first_block && b <= last_block) {
3383                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3384                         count++;
3385                 }
3386                 b = ext4_inode_bitmap(sb, gdp);
3387                 if (b >= first_block && b <= last_block) {
3388                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3389                         count++;
3390                 }
3391                 b = ext4_inode_table(sb, gdp);
3392                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3393                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3394                                 int c = EXT4_B2C(sbi, b - first_block);
3395                                 ext4_set_bit(c, buf);
3396                                 count++;
3397                         }
3398                 if (i != grp)
3399                         continue;
3400                 s = 0;
3401                 if (ext4_bg_has_super(sb, grp)) {
3402                         ext4_set_bit(s++, buf);
3403                         count++;
3404                 }
3405                 j = ext4_bg_num_gdb(sb, grp);
3406                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3407                         ext4_error(sb, "Invalid number of block group "
3408                                    "descriptor blocks: %d", j);
3409                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3410                 }
3411                 count += j;
3412                 for (; j > 0; j--)
3413                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3414         }
3415         if (!count)
3416                 return 0;
3417         return EXT4_CLUSTERS_PER_GROUP(sb) -
3418                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3419 }
3420
3421 /*
3422  * Compute the overhead and stash it in sbi->s_overhead
3423  */
3424 int ext4_calculate_overhead(struct super_block *sb)
3425 {
3426         struct ext4_sb_info *sbi = EXT4_SB(sb);
3427         struct ext4_super_block *es = sbi->s_es;
3428         struct inode *j_inode;
3429         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3430         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3431         ext4_fsblk_t overhead = 0;
3432         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3433
3434         if (!buf)
3435                 return -ENOMEM;
3436
3437         /*
3438          * Compute the overhead (FS structures).  This is constant
3439          * for a given filesystem unless the number of block groups
3440          * changes so we cache the previous value until it does.
3441          */
3442
3443         /*
3444          * All of the blocks before first_data_block are overhead
3445          */
3446         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3447
3448         /*
3449          * Add the overhead found in each block group
3450          */
3451         for (i = 0; i < ngroups; i++) {
3452                 int blks;
3453
3454                 blks = count_overhead(sb, i, buf);
3455                 overhead += blks;
3456                 if (blks)
3457                         memset(buf, 0, PAGE_SIZE);
3458                 cond_resched();
3459         }
3460
3461         /*
3462          * Add the internal journal blocks whether the journal has been
3463          * loaded or not
3464          */
3465         if (sbi->s_journal && !sbi->journal_bdev)
3466                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3467         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3468                 j_inode = ext4_get_journal_inode(sb, j_inum);
3469                 if (j_inode) {
3470                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3471                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3472                         iput(j_inode);
3473                 } else {
3474                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3475                 }
3476         }
3477         sbi->s_overhead = overhead;
3478         smp_wmb();
3479         free_page((unsigned long) buf);
3480         return 0;
3481 }
3482
3483 static void ext4_set_resv_clusters(struct super_block *sb)
3484 {
3485         ext4_fsblk_t resv_clusters;
3486         struct ext4_sb_info *sbi = EXT4_SB(sb);
3487
3488         /*
3489          * There's no need to reserve anything when we aren't using extents.
3490          * The space estimates are exact, there are no unwritten extents,
3491          * hole punching doesn't need new metadata... This is needed especially
3492          * to keep ext2/3 backward compatibility.
3493          */
3494         if (!ext4_has_feature_extents(sb))
3495                 return;
3496         /*
3497          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3498          * This should cover the situations where we can not afford to run
3499          * out of space like for example punch hole, or converting
3500          * unwritten extents in delalloc path. In most cases such
3501          * allocation would require 1, or 2 blocks, higher numbers are
3502          * very rare.
3503          */
3504         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3505                          sbi->s_cluster_bits);
3506
3507         do_div(resv_clusters, 50);
3508         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3509
3510         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3511 }
3512
3513 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3514 {
3515         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3516         char *orig_data = kstrdup(data, GFP_KERNEL);
3517         struct buffer_head *bh;
3518         struct ext4_super_block *es = NULL;
3519         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3520         ext4_fsblk_t block;
3521         ext4_fsblk_t sb_block = get_sb_block(&data);
3522         ext4_fsblk_t logical_sb_block;
3523         unsigned long offset = 0;
3524         unsigned long journal_devnum = 0;
3525         unsigned long def_mount_opts;
3526         struct inode *root;
3527         const char *descr;
3528         int ret = -ENOMEM;
3529         int blocksize, clustersize;
3530         unsigned int db_count;
3531         unsigned int i;
3532         int needs_recovery, has_huge_files, has_bigalloc;
3533         __u64 blocks_count;
3534         int err = 0;
3535         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3536         ext4_group_t first_not_zeroed;
3537
3538         if ((data && !orig_data) || !sbi)
3539                 goto out_free_base;
3540
3541         sbi->s_daxdev = dax_dev;
3542         sbi->s_blockgroup_lock =
3543                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3544         if (!sbi->s_blockgroup_lock)
3545                 goto out_free_base;
3546
3547         sb->s_fs_info = sbi;
3548         sbi->s_sb = sb;
3549         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3550         sbi->s_sb_block = sb_block;
3551         if (sb->s_bdev->bd_part)
3552                 sbi->s_sectors_written_start =
3553                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3554
3555         /* Cleanup superblock name */
3556         strreplace(sb->s_id, '/', '!');
3557
3558         /* -EINVAL is default */
3559         ret = -EINVAL;
3560         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3561         if (!blocksize) {
3562                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3563                 goto out_fail;
3564         }
3565
3566         /*
3567          * The ext4 superblock will not be buffer aligned for other than 1kB
3568          * block sizes.  We need to calculate the offset from buffer start.
3569          */
3570         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3571                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3572                 offset = do_div(logical_sb_block, blocksize);
3573         } else {
3574                 logical_sb_block = sb_block;
3575         }
3576
3577         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3578                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3579                 goto out_fail;
3580         }
3581         /*
3582          * Note: s_es must be initialized as soon as possible because
3583          *       some ext4 macro-instructions depend on its value
3584          */
3585         es = (struct ext4_super_block *) (bh->b_data + offset);
3586         sbi->s_es = es;
3587         sb->s_magic = le16_to_cpu(es->s_magic);
3588         if (sb->s_magic != EXT4_SUPER_MAGIC)
3589                 goto cantfind_ext4;
3590         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3591
3592         /* Warn if metadata_csum and gdt_csum are both set. */
3593         if (ext4_has_feature_metadata_csum(sb) &&
3594             ext4_has_feature_gdt_csum(sb))
3595                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3596                              "redundant flags; please run fsck.");
3597
3598         /* Check for a known checksum algorithm */
3599         if (!ext4_verify_csum_type(sb, es)) {
3600                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3601                          "unknown checksum algorithm.");
3602                 silent = 1;
3603                 goto cantfind_ext4;
3604         }
3605
3606         /* Load the checksum driver */
3607         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3608         if (IS_ERR(sbi->s_chksum_driver)) {
3609                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3610                 ret = PTR_ERR(sbi->s_chksum_driver);
3611                 sbi->s_chksum_driver = NULL;
3612                 goto failed_mount;
3613         }
3614
3615         /* Check superblock checksum */
3616         if (!ext4_superblock_csum_verify(sb, es)) {
3617                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3618                          "invalid superblock checksum.  Run e2fsck?");
3619                 silent = 1;
3620                 ret = -EFSBADCRC;
3621                 goto cantfind_ext4;
3622         }
3623
3624         /* Precompute checksum seed for all metadata */
3625         if (ext4_has_feature_csum_seed(sb))
3626                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3627         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3628                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3629                                                sizeof(es->s_uuid));
3630
3631         /* Set defaults before we parse the mount options */
3632         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3633         set_opt(sb, INIT_INODE_TABLE);
3634         if (def_mount_opts & EXT4_DEFM_DEBUG)
3635                 set_opt(sb, DEBUG);
3636         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3637                 set_opt(sb, GRPID);
3638         if (def_mount_opts & EXT4_DEFM_UID16)
3639                 set_opt(sb, NO_UID32);
3640         /* xattr user namespace & acls are now defaulted on */
3641         set_opt(sb, XATTR_USER);
3642 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3643         set_opt(sb, POSIX_ACL);
3644 #endif
3645         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3646         if (ext4_has_metadata_csum(sb))
3647                 set_opt(sb, JOURNAL_CHECKSUM);
3648
3649         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3650                 set_opt(sb, JOURNAL_DATA);
3651         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3652                 set_opt(sb, ORDERED_DATA);
3653         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3654                 set_opt(sb, WRITEBACK_DATA);
3655
3656         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3657                 set_opt(sb, ERRORS_PANIC);
3658         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3659                 set_opt(sb, ERRORS_CONT);
3660         else
3661                 set_opt(sb, ERRORS_RO);
3662         /* block_validity enabled by default; disable with noblock_validity */
3663         set_opt(sb, BLOCK_VALIDITY);
3664         if (def_mount_opts & EXT4_DEFM_DISCARD)
3665                 set_opt(sb, DISCARD);
3666
3667         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3668         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3669         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3670         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3671         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3672
3673         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3674                 set_opt(sb, BARRIER);
3675
3676         /*
3677          * enable delayed allocation by default
3678          * Use -o nodelalloc to turn it off
3679          */
3680         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3681             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3682                 set_opt(sb, DELALLOC);
3683
3684         /*
3685          * set default s_li_wait_mult for lazyinit, for the case there is
3686          * no mount option specified.
3687          */
3688         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3689
3690         if (sbi->s_es->s_mount_opts[0]) {
3691                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3692                                               sizeof(sbi->s_es->s_mount_opts),
3693                                               GFP_KERNEL);
3694                 if (!s_mount_opts)
3695                         goto failed_mount;
3696                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3697                                    &journal_ioprio, 0)) {
3698                         ext4_msg(sb, KERN_WARNING,
3699                                  "failed to parse options in superblock: %s",
3700                                  s_mount_opts);
3701                 }
3702                 kfree(s_mount_opts);
3703         }
3704         sbi->s_def_mount_opt = sbi->s_mount_opt;
3705         if (!parse_options((char *) data, sb, &journal_devnum,
3706                            &journal_ioprio, 0))
3707                 goto failed_mount;
3708
3709         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3710                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3711                             "with data=journal disables delayed "
3712                             "allocation and O_DIRECT support!\n");
3713                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3714                         ext4_msg(sb, KERN_ERR, "can't mount with "
3715                                  "both data=journal and delalloc");
3716                         goto failed_mount;
3717                 }
3718                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3719                         ext4_msg(sb, KERN_ERR, "can't mount with "
3720                                  "both data=journal and dioread_nolock");
3721                         goto failed_mount;
3722                 }
3723                 if (test_opt(sb, DAX)) {
3724                         ext4_msg(sb, KERN_ERR, "can't mount with "
3725                                  "both data=journal and dax");
3726                         goto failed_mount;
3727                 }
3728                 if (ext4_has_feature_encrypt(sb)) {
3729                         ext4_msg(sb, KERN_WARNING,
3730                                  "encrypted files will use data=ordered "
3731                                  "instead of data journaling mode");
3732                 }
3733                 if (test_opt(sb, DELALLOC))
3734                         clear_opt(sb, DELALLOC);
3735         } else {
3736                 sb->s_iflags |= SB_I_CGROUPWB;
3737         }
3738
3739         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3740                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3741
3742         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3743             (ext4_has_compat_features(sb) ||
3744              ext4_has_ro_compat_features(sb) ||
3745              ext4_has_incompat_features(sb)))
3746                 ext4_msg(sb, KERN_WARNING,
3747                        "feature flags set on rev 0 fs, "
3748                        "running e2fsck is recommended");
3749
3750         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3751                 set_opt2(sb, HURD_COMPAT);
3752                 if (ext4_has_feature_64bit(sb)) {
3753                         ext4_msg(sb, KERN_ERR,
3754                                  "The Hurd can't support 64-bit file systems");
3755                         goto failed_mount;
3756                 }
3757
3758                 /*
3759                  * ea_inode feature uses l_i_version field which is not
3760                  * available in HURD_COMPAT mode.
3761                  */
3762                 if (ext4_has_feature_ea_inode(sb)) {
3763                         ext4_msg(sb, KERN_ERR,
3764                                  "ea_inode feature is not supported for Hurd");
3765                         goto failed_mount;
3766                 }
3767         }
3768
3769         if (IS_EXT2_SB(sb)) {
3770                 if (ext2_feature_set_ok(sb))
3771                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3772                                  "using the ext4 subsystem");
3773                 else {
3774                         /*
3775                          * If we're probing be silent, if this looks like
3776                          * it's actually an ext[34] filesystem.
3777                          */
3778                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3779                                 goto failed_mount;
3780                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3781                                  "to feature incompatibilities");
3782                         goto failed_mount;
3783                 }
3784         }
3785
3786         if (IS_EXT3_SB(sb)) {
3787                 if (ext3_feature_set_ok(sb))
3788                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3789                                  "using the ext4 subsystem");
3790                 else {
3791                         /*
3792                          * If we're probing be silent, if this looks like
3793                          * it's actually an ext4 filesystem.
3794                          */
3795                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3796                                 goto failed_mount;
3797                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3798                                  "to feature incompatibilities");
3799                         goto failed_mount;
3800                 }
3801         }
3802
3803         /*
3804          * Check feature flags regardless of the revision level, since we
3805          * previously didn't change the revision level when setting the flags,
3806          * so there is a chance incompat flags are set on a rev 0 filesystem.
3807          */
3808         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3809                 goto failed_mount;
3810
3811         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3812         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3813             blocksize > EXT4_MAX_BLOCK_SIZE) {
3814                 ext4_msg(sb, KERN_ERR,
3815                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3816                          blocksize, le32_to_cpu(es->s_log_block_size));
3817                 goto failed_mount;
3818         }
3819         if (le32_to_cpu(es->s_log_block_size) >
3820             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3821                 ext4_msg(sb, KERN_ERR,
3822                          "Invalid log block size: %u",
3823                          le32_to_cpu(es->s_log_block_size));
3824                 goto failed_mount;
3825         }
3826         if (le32_to_cpu(es->s_log_cluster_size) >
3827             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3828                 ext4_msg(sb, KERN_ERR,
3829                          "Invalid log cluster size: %u",
3830                          le32_to_cpu(es->s_log_cluster_size));
3831                 goto failed_mount;
3832         }
3833
3834         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3835                 ext4_msg(sb, KERN_ERR,
3836                          "Number of reserved GDT blocks insanely large: %d",
3837                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3838                 goto failed_mount;
3839         }
3840
3841         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3842                 if (ext4_has_feature_inline_data(sb)) {
3843                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3844                                         " that may contain inline data");
3845                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3846                 }
3847                 if (!bdev_dax_supported(sb->s_bdev, blocksize)) {
3848                         ext4_msg(sb, KERN_ERR,
3849                                 "DAX unsupported by block device. Turning off DAX.");
3850                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3851                 }
3852         }
3853
3854         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3855                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3856                          es->s_encryption_level);
3857                 goto failed_mount;
3858         }
3859
3860         if (sb->s_blocksize != blocksize) {
3861                 /* Validate the filesystem blocksize */
3862                 if (!sb_set_blocksize(sb, blocksize)) {
3863                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3864                                         blocksize);
3865                         goto failed_mount;
3866                 }
3867
3868                 brelse(bh);
3869                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3870                 offset = do_div(logical_sb_block, blocksize);
3871                 bh = sb_bread_unmovable(sb, logical_sb_block);
3872                 if (!bh) {
3873                         ext4_msg(sb, KERN_ERR,
3874                                "Can't read superblock on 2nd try");
3875                         goto failed_mount;
3876                 }
3877                 es = (struct ext4_super_block *)(bh->b_data + offset);
3878                 sbi->s_es = es;
3879                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3880                         ext4_msg(sb, KERN_ERR,
3881                                "Magic mismatch, very weird!");
3882                         goto failed_mount;
3883                 }
3884         }
3885
3886         has_huge_files = ext4_has_feature_huge_file(sb);
3887         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3888                                                       has_huge_files);
3889         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3890
3891         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3892                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3893                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3894         } else {
3895                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3896                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3897                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3898                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3899                                  sbi->s_first_ino);
3900                         goto failed_mount;
3901                 }
3902                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3903                     (!is_power_of_2(sbi->s_inode_size)) ||
3904                     (sbi->s_inode_size > blocksize)) {
3905                         ext4_msg(sb, KERN_ERR,
3906                                "unsupported inode size: %d",
3907                                sbi->s_inode_size);
3908                         goto failed_mount;
3909                 }
3910                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3911                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3912         }
3913
3914         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3915         if (ext4_has_feature_64bit(sb)) {
3916                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3917                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3918                     !is_power_of_2(sbi->s_desc_size)) {
3919                         ext4_msg(sb, KERN_ERR,
3920                                "unsupported descriptor size %lu",
3921                                sbi->s_desc_size);
3922                         goto failed_mount;
3923                 }
3924         } else
3925                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3926
3927         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3928         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3929
3930         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3931         if (sbi->s_inodes_per_block == 0)
3932                 goto cantfind_ext4;
3933         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3934             sbi->s_inodes_per_group > blocksize * 8) {
3935                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3936                          sbi->s_blocks_per_group);
3937                 goto failed_mount;
3938         }
3939         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3940                                         sbi->s_inodes_per_block;
3941         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3942         sbi->s_sbh = bh;
3943         sbi->s_mount_state = le16_to_cpu(es->s_state);
3944         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3945         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3946
3947         for (i = 0; i < 4; i++)
3948                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3949         sbi->s_def_hash_version = es->s_def_hash_version;
3950         if (ext4_has_feature_dir_index(sb)) {
3951                 i = le32_to_cpu(es->s_flags);
3952                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3953                         sbi->s_hash_unsigned = 3;
3954                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3955 #ifdef __CHAR_UNSIGNED__
3956                         if (!sb_rdonly(sb))
3957                                 es->s_flags |=
3958                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3959                         sbi->s_hash_unsigned = 3;
3960 #else
3961                         if (!sb_rdonly(sb))
3962                                 es->s_flags |=
3963                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3964 #endif
3965                 }
3966         }
3967
3968         /* Handle clustersize */
3969         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3970         has_bigalloc = ext4_has_feature_bigalloc(sb);
3971         if (has_bigalloc) {
3972                 if (clustersize < blocksize) {
3973                         ext4_msg(sb, KERN_ERR,
3974                                  "cluster size (%d) smaller than "
3975                                  "block size (%d)", clustersize, blocksize);
3976                         goto failed_mount;
3977                 }
3978                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3979                         le32_to_cpu(es->s_log_block_size);
3980                 sbi->s_clusters_per_group =
3981                         le32_to_cpu(es->s_clusters_per_group);
3982                 if (sbi->s_clusters_per_group > blocksize * 8) {
3983                         ext4_msg(sb, KERN_ERR,
3984                                  "#clusters per group too big: %lu",
3985                                  sbi->s_clusters_per_group);
3986                         goto failed_mount;
3987                 }
3988                 if (sbi->s_blocks_per_group !=
3989                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3990                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3991                                  "clusters per group (%lu) inconsistent",
3992                                  sbi->s_blocks_per_group,
3993                                  sbi->s_clusters_per_group);
3994                         goto failed_mount;
3995                 }
3996         } else {
3997                 if (clustersize != blocksize) {
3998                         ext4_msg(sb, KERN_ERR,
3999                                  "fragment/cluster size (%d) != "
4000                                  "block size (%d)", clustersize, blocksize);
4001                         goto failed_mount;
4002                 }
4003                 if (sbi->s_blocks_per_group > blocksize * 8) {
4004                         ext4_msg(sb, KERN_ERR,
4005                                  "#blocks per group too big: %lu",
4006                                  sbi->s_blocks_per_group);
4007                         goto failed_mount;
4008                 }
4009                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4010                 sbi->s_cluster_bits = 0;
4011         }
4012         sbi->s_cluster_ratio = clustersize / blocksize;
4013
4014         /* Do we have standard group size of clustersize * 8 blocks ? */
4015         if (sbi->s_blocks_per_group == clustersize << 3)
4016                 set_opt2(sb, STD_GROUP_SIZE);
4017
4018         /*
4019          * Test whether we have more sectors than will fit in sector_t,
4020          * and whether the max offset is addressable by the page cache.
4021          */
4022         err = generic_check_addressable(sb->s_blocksize_bits,
4023                                         ext4_blocks_count(es));
4024         if (err) {
4025                 ext4_msg(sb, KERN_ERR, "filesystem"
4026                          " too large to mount safely on this system");
4027                 if (sizeof(sector_t) < 8)
4028                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
4029                 goto failed_mount;
4030         }
4031
4032         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4033                 goto cantfind_ext4;
4034
4035         /* check blocks count against device size */
4036         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4037         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4038                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4039                        "exceeds size of device (%llu blocks)",
4040                        ext4_blocks_count(es), blocks_count);
4041                 goto failed_mount;
4042         }
4043
4044         /*
4045          * It makes no sense for the first data block to be beyond the end
4046          * of the filesystem.
4047          */
4048         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4049                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4050                          "block %u is beyond end of filesystem (%llu)",
4051                          le32_to_cpu(es->s_first_data_block),
4052                          ext4_blocks_count(es));
4053                 goto failed_mount;
4054         }
4055         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4056             (sbi->s_cluster_ratio == 1)) {
4057                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4058                          "block is 0 with a 1k block and cluster size");
4059                 goto failed_mount;
4060         }
4061
4062         blocks_count = (ext4_blocks_count(es) -
4063                         le32_to_cpu(es->s_first_data_block) +
4064                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4065         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4066         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4067                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
4068                        "(block count %llu, first data block %u, "
4069                        "blocks per group %lu)", sbi->s_groups_count,
4070                        ext4_blocks_count(es),
4071                        le32_to_cpu(es->s_first_data_block),
4072                        EXT4_BLOCKS_PER_GROUP(sb));
4073                 goto failed_mount;
4074         }
4075         sbi->s_groups_count = blocks_count;
4076         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4077                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4078         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4079                    EXT4_DESC_PER_BLOCK(sb);
4080         if (ext4_has_feature_meta_bg(sb)) {
4081                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4082                         ext4_msg(sb, KERN_WARNING,
4083                                  "first meta block group too large: %u "
4084                                  "(group descriptor block count %u)",
4085                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4086                         goto failed_mount;
4087                 }
4088         }
4089         sbi->s_group_desc = kvmalloc_array(db_count,
4090                                            sizeof(struct buffer_head *),
4091                                            GFP_KERNEL);
4092         if (sbi->s_group_desc == NULL) {
4093                 ext4_msg(sb, KERN_ERR, "not enough memory");
4094                 ret = -ENOMEM;
4095                 goto failed_mount;
4096         }
4097         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4098             le32_to_cpu(es->s_inodes_count)) {
4099                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4100                          le32_to_cpu(es->s_inodes_count),
4101                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4102                 ret = -EINVAL;
4103                 goto failed_mount;
4104         }
4105
4106         bgl_lock_init(sbi->s_blockgroup_lock);
4107
4108         /* Pre-read the descriptors into the buffer cache */
4109         for (i = 0; i < db_count; i++) {
4110                 block = descriptor_loc(sb, logical_sb_block, i);
4111                 sb_breadahead(sb, block);
4112         }
4113
4114         for (i = 0; i < db_count; i++) {
4115                 block = descriptor_loc(sb, logical_sb_block, i);
4116                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
4117                 if (!sbi->s_group_desc[i]) {
4118                         ext4_msg(sb, KERN_ERR,
4119                                "can't read group descriptor %d", i);
4120                         db_count = i;
4121                         goto failed_mount2;
4122                 }
4123         }
4124         sbi->s_gdb_count = db_count;
4125         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4126                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4127                 ret = -EFSCORRUPTED;
4128                 goto failed_mount2;
4129         }
4130
4131         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4132
4133         /* Register extent status tree shrinker */
4134         if (ext4_es_register_shrinker(sbi))
4135                 goto failed_mount3;
4136
4137         sbi->s_stripe = ext4_get_stripe_size(sbi);
4138         sbi->s_extent_max_zeroout_kb = 32;
4139
4140         /*
4141          * set up enough so that it can read an inode
4142          */
4143         sb->s_op = &ext4_sops;
4144         sb->s_export_op = &ext4_export_ops;
4145         sb->s_xattr = ext4_xattr_handlers;
4146 #ifdef CONFIG_EXT4_FS_ENCRYPTION
4147         sb->s_cop = &ext4_cryptops;
4148 #endif
4149 #ifdef CONFIG_QUOTA
4150         sb->dq_op = &ext4_quota_operations;
4151         if (ext4_has_feature_quota(sb))
4152                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4153         else
4154                 sb->s_qcop = &ext4_qctl_operations;
4155         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4156 #endif
4157         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4158
4159         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4160         mutex_init(&sbi->s_orphan_lock);
4161
4162         sb->s_root = NULL;
4163
4164         needs_recovery = (es->s_last_orphan != 0 ||
4165                           ext4_has_feature_journal_needs_recovery(sb));
4166
4167         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4168                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4169                         goto failed_mount3a;
4170
4171         /*
4172          * The first inode we look at is the journal inode.  Don't try
4173          * root first: it may be modified in the journal!
4174          */
4175         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4176                 err = ext4_load_journal(sb, es, journal_devnum);
4177                 if (err)
4178                         goto failed_mount3a;
4179         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4180                    ext4_has_feature_journal_needs_recovery(sb)) {
4181                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4182                        "suppressed and not mounted read-only");
4183                 goto failed_mount_wq;
4184         } else {
4185                 /* Nojournal mode, all journal mount options are illegal */
4186                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4187                         ext4_msg(sb, KERN_ERR, "can't mount with "
4188                                  "journal_checksum, fs mounted w/o journal");
4189                         goto failed_mount_wq;
4190                 }
4191                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4192                         ext4_msg(sb, KERN_ERR, "can't mount with "
4193                                  "journal_async_commit, fs mounted w/o journal");
4194                         goto failed_mount_wq;
4195                 }
4196                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4197                         ext4_msg(sb, KERN_ERR, "can't mount with "
4198                                  "commit=%lu, fs mounted w/o journal",
4199                                  sbi->s_commit_interval / HZ);
4200                         goto failed_mount_wq;
4201                 }
4202                 if (EXT4_MOUNT_DATA_FLAGS &
4203                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4204                         ext4_msg(sb, KERN_ERR, "can't mount with "
4205                                  "data=, fs mounted w/o journal");
4206                         goto failed_mount_wq;
4207                 }
4208                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4209                 clear_opt(sb, JOURNAL_CHECKSUM);
4210                 clear_opt(sb, DATA_FLAGS);
4211                 sbi->s_journal = NULL;
4212                 needs_recovery = 0;
4213                 goto no_journal;
4214         }
4215
4216         if (ext4_has_feature_64bit(sb) &&
4217             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4218                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4219                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4220                 goto failed_mount_wq;
4221         }
4222
4223         if (!set_journal_csum_feature_set(sb)) {
4224                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4225                          "feature set");
4226                 goto failed_mount_wq;
4227         }
4228
4229         /* We have now updated the journal if required, so we can
4230          * validate the data journaling mode. */
4231         switch (test_opt(sb, DATA_FLAGS)) {
4232         case 0:
4233                 /* No mode set, assume a default based on the journal
4234                  * capabilities: ORDERED_DATA if the journal can
4235                  * cope, else JOURNAL_DATA
4236                  */
4237                 if (jbd2_journal_check_available_features
4238                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4239                         set_opt(sb, ORDERED_DATA);
4240                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4241                 } else {
4242                         set_opt(sb, JOURNAL_DATA);
4243                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4244                 }
4245                 break;
4246
4247         case EXT4_MOUNT_ORDERED_DATA:
4248         case EXT4_MOUNT_WRITEBACK_DATA:
4249                 if (!jbd2_journal_check_available_features
4250                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4251                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4252                                "requested data journaling mode");
4253                         goto failed_mount_wq;
4254                 }
4255         default:
4256                 break;
4257         }
4258
4259         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4260             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4261                 ext4_msg(sb, KERN_ERR, "can't mount with "
4262                         "journal_async_commit in data=ordered mode");
4263                 goto failed_mount_wq;
4264         }
4265
4266         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4267
4268         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4269
4270 no_journal:
4271         if (!test_opt(sb, NO_MBCACHE)) {
4272                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4273                 if (!sbi->s_ea_block_cache) {
4274                         ext4_msg(sb, KERN_ERR,
4275                                  "Failed to create ea_block_cache");
4276                         goto failed_mount_wq;
4277                 }
4278
4279                 if (ext4_has_feature_ea_inode(sb)) {
4280                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4281                         if (!sbi->s_ea_inode_cache) {
4282                                 ext4_msg(sb, KERN_ERR,
4283                                          "Failed to create ea_inode_cache");
4284                                 goto failed_mount_wq;
4285                         }
4286                 }
4287         }
4288
4289         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4290             (blocksize != PAGE_SIZE)) {
4291                 ext4_msg(sb, KERN_ERR,
4292                          "Unsupported blocksize for fs encryption");
4293                 goto failed_mount_wq;
4294         }
4295
4296         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4297             !ext4_has_feature_encrypt(sb)) {
4298                 ext4_set_feature_encrypt(sb);
4299                 ext4_commit_super(sb, 1);
4300         }
4301
4302         /*
4303          * Get the # of file system overhead blocks from the
4304          * superblock if present.
4305          */
4306         if (es->s_overhead_clusters)
4307                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4308         else {
4309                 err = ext4_calculate_overhead(sb);
4310                 if (err)
4311                         goto failed_mount_wq;
4312         }
4313
4314         /*
4315          * The maximum number of concurrent works can be high and
4316          * concurrency isn't really necessary.  Limit it to 1.
4317          */
4318         EXT4_SB(sb)->rsv_conversion_wq =
4319                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4320         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4321                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4322                 ret = -ENOMEM;
4323                 goto failed_mount4;
4324         }
4325
4326         /*
4327          * The jbd2_journal_load will have done any necessary log recovery,
4328          * so we can safely mount the rest of the filesystem now.
4329          */
4330
4331         root = ext4_iget(sb, EXT4_ROOT_INO);
4332         if (IS_ERR(root)) {
4333                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4334                 ret = PTR_ERR(root);
4335                 root = NULL;
4336                 goto failed_mount4;
4337         }
4338         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4339                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4340                 iput(root);
4341                 goto failed_mount4;
4342         }
4343         sb->s_root = d_make_root(root);
4344         if (!sb->s_root) {
4345                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4346                 ret = -ENOMEM;
4347                 goto failed_mount4;
4348         }
4349
4350         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4351         if (ret == -EROFS) {
4352                 sb->s_flags |= SB_RDONLY;
4353                 ret = 0;
4354         } else if (ret)
4355                 goto failed_mount4a;
4356
4357         /* determine the minimum size of new large inodes, if present */
4358         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4359             sbi->s_want_extra_isize == 0) {
4360                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4361                                                      EXT4_GOOD_OLD_INODE_SIZE;
4362                 if (ext4_has_feature_extra_isize(sb)) {
4363                         if (sbi->s_want_extra_isize <
4364                             le16_to_cpu(es->s_want_extra_isize))
4365                                 sbi->s_want_extra_isize =
4366                                         le16_to_cpu(es->s_want_extra_isize);
4367                         if (sbi->s_want_extra_isize <
4368                             le16_to_cpu(es->s_min_extra_isize))
4369                                 sbi->s_want_extra_isize =
4370                                         le16_to_cpu(es->s_min_extra_isize);
4371                 }
4372         }
4373         /* Check if enough inode space is available */
4374         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4375                                                         sbi->s_inode_size) {
4376                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4377                                                        EXT4_GOOD_OLD_INODE_SIZE;
4378                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4379                          "available");
4380         }
4381
4382         ext4_set_resv_clusters(sb);
4383
4384         err = ext4_setup_system_zone(sb);
4385         if (err) {
4386                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4387                          "zone (%d)", err);
4388                 goto failed_mount4a;
4389         }
4390
4391         ext4_ext_init(sb);
4392         err = ext4_mb_init(sb);
4393         if (err) {
4394                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4395                          err);
4396                 goto failed_mount5;
4397         }
4398
4399         block = ext4_count_free_clusters(sb);
4400         ext4_free_blocks_count_set(sbi->s_es,
4401                                    EXT4_C2B(sbi, block));
4402         ext4_superblock_csum_set(sb);
4403         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4404                                   GFP_KERNEL);
4405         if (!err) {
4406                 unsigned long freei = ext4_count_free_inodes(sb);
4407                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4408                 ext4_superblock_csum_set(sb);
4409                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4410                                           GFP_KERNEL);
4411         }
4412         if (!err)
4413                 err = percpu_counter_init(&sbi->s_dirs_counter,
4414                                           ext4_count_dirs(sb), GFP_KERNEL);
4415         if (!err)
4416                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4417                                           GFP_KERNEL);
4418         if (!err)
4419                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4420
4421         if (err) {
4422                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4423                 goto failed_mount6;
4424         }
4425
4426         if (ext4_has_feature_flex_bg(sb))
4427                 if (!ext4_fill_flex_info(sb)) {
4428                         ext4_msg(sb, KERN_ERR,
4429                                "unable to initialize "
4430                                "flex_bg meta info!");
4431                         goto failed_mount6;
4432                 }
4433
4434         err = ext4_register_li_request(sb, first_not_zeroed);
4435         if (err)
4436                 goto failed_mount6;
4437
4438         err = ext4_register_sysfs(sb);
4439         if (err)
4440                 goto failed_mount7;
4441
4442 #ifdef CONFIG_QUOTA
4443         /* Enable quota usage during mount. */
4444         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4445                 err = ext4_enable_quotas(sb);
4446                 if (err)
4447                         goto failed_mount8;
4448         }
4449 #endif  /* CONFIG_QUOTA */
4450
4451         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4452         ext4_orphan_cleanup(sb, es);
4453         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4454         if (needs_recovery) {
4455                 ext4_msg(sb, KERN_INFO, "recovery complete");
4456                 ext4_mark_recovery_complete(sb, es);
4457         }
4458         if (EXT4_SB(sb)->s_journal) {
4459                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4460                         descr = " journalled data mode";
4461                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4462                         descr = " ordered data mode";
4463                 else
4464                         descr = " writeback data mode";
4465         } else
4466                 descr = "out journal";
4467
4468         if (test_opt(sb, DISCARD)) {
4469                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4470                 if (!blk_queue_discard(q))
4471                         ext4_msg(sb, KERN_WARNING,
4472                                  "mounting with \"discard\" option, but "
4473                                  "the device does not support discard");
4474         }
4475
4476         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4477                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4478                          "Opts: %.*s%s%s", descr,
4479                          (int) sizeof(sbi->s_es->s_mount_opts),
4480                          sbi->s_es->s_mount_opts,
4481                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4482
4483         if (es->s_error_count)
4484                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4485
4486         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4487         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4488         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4489         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4490
4491         kfree(orig_data);
4492         return 0;
4493
4494 cantfind_ext4:
4495         if (!silent)
4496                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4497         goto failed_mount;
4498
4499 #ifdef CONFIG_QUOTA
4500 failed_mount8:
4501         ext4_unregister_sysfs(sb);
4502 #endif
4503 failed_mount7:
4504         ext4_unregister_li_request(sb);
4505 failed_mount6:
4506         ext4_mb_release(sb);
4507         if (sbi->s_flex_groups)
4508                 kvfree(sbi->s_flex_groups);
4509         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4510         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4511         percpu_counter_destroy(&sbi->s_dirs_counter);
4512         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4513 failed_mount5:
4514         ext4_ext_release(sb);
4515         ext4_release_system_zone(sb);
4516 failed_mount4a:
4517         dput(sb->s_root);
4518         sb->s_root = NULL;
4519 failed_mount4:
4520         ext4_msg(sb, KERN_ERR, "mount failed");
4521         if (EXT4_SB(sb)->rsv_conversion_wq)
4522                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4523 failed_mount_wq:
4524         if (sbi->s_ea_inode_cache) {
4525                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4526                 sbi->s_ea_inode_cache = NULL;
4527         }
4528         if (sbi->s_ea_block_cache) {
4529                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4530                 sbi->s_ea_block_cache = NULL;
4531         }
4532         if (sbi->s_journal) {
4533                 jbd2_journal_destroy(sbi->s_journal);
4534                 sbi->s_journal = NULL;
4535         }
4536 failed_mount3a:
4537         ext4_es_unregister_shrinker(sbi);
4538 failed_mount3:
4539         del_timer_sync(&sbi->s_err_report);
4540         if (sbi->s_mmp_tsk)
4541                 kthread_stop(sbi->s_mmp_tsk);
4542 failed_mount2:
4543         for (i = 0; i < db_count; i++)
4544                 brelse(sbi->s_group_desc[i]);
4545         kvfree(sbi->s_group_desc);
4546 failed_mount:
4547         if (sbi->s_chksum_driver)
4548                 crypto_free_shash(sbi->s_chksum_driver);
4549 #ifdef CONFIG_QUOTA
4550         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4551                 kfree(sbi->s_qf_names[i]);
4552 #endif
4553         ext4_blkdev_remove(sbi);
4554         brelse(bh);
4555 out_fail:
4556         sb->s_fs_info = NULL;
4557         kfree(sbi->s_blockgroup_lock);
4558 out_free_base:
4559         kfree(sbi);
4560         kfree(orig_data);
4561         fs_put_dax(dax_dev);
4562         return err ? err : ret;
4563 }
4564
4565 /*
4566  * Setup any per-fs journal parameters now.  We'll do this both on
4567  * initial mount, once the journal has been initialised but before we've
4568  * done any recovery; and again on any subsequent remount.
4569  */
4570 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4571 {
4572         struct ext4_sb_info *sbi = EXT4_SB(sb);
4573
4574         journal->j_commit_interval = sbi->s_commit_interval;
4575         journal->j_min_batch_time = sbi->s_min_batch_time;
4576         journal->j_max_batch_time = sbi->s_max_batch_time;
4577
4578         write_lock(&journal->j_state_lock);
4579         if (test_opt(sb, BARRIER))
4580                 journal->j_flags |= JBD2_BARRIER;
4581         else
4582                 journal->j_flags &= ~JBD2_BARRIER;
4583         if (test_opt(sb, DATA_ERR_ABORT))
4584                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4585         else
4586                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4587         write_unlock(&journal->j_state_lock);
4588 }
4589
4590 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4591                                              unsigned int journal_inum)
4592 {
4593         struct inode *journal_inode;
4594
4595         /*
4596          * Test for the existence of a valid inode on disk.  Bad things
4597          * happen if we iget() an unused inode, as the subsequent iput()
4598          * will try to delete it.
4599          */
4600         journal_inode = ext4_iget(sb, journal_inum);
4601         if (IS_ERR(journal_inode)) {
4602                 ext4_msg(sb, KERN_ERR, "no journal found");
4603                 return NULL;
4604         }
4605         if (!journal_inode->i_nlink) {
4606                 make_bad_inode(journal_inode);
4607                 iput(journal_inode);
4608                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4609                 return NULL;
4610         }
4611
4612         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4613                   journal_inode, journal_inode->i_size);
4614         if (!S_ISREG(journal_inode->i_mode)) {
4615                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4616                 iput(journal_inode);
4617                 return NULL;
4618         }
4619         return journal_inode;
4620 }
4621
4622 static journal_t *ext4_get_journal(struct super_block *sb,
4623                                    unsigned int journal_inum)
4624 {
4625         struct inode *journal_inode;
4626         journal_t *journal;
4627
4628         BUG_ON(!ext4_has_feature_journal(sb));
4629
4630         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4631         if (!journal_inode)
4632                 return NULL;
4633
4634         journal = jbd2_journal_init_inode(journal_inode);
4635         if (!journal) {
4636                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4637                 iput(journal_inode);
4638                 return NULL;
4639         }
4640         journal->j_private = sb;
4641         ext4_init_journal_params(sb, journal);
4642         return journal;
4643 }
4644
4645 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4646                                        dev_t j_dev)
4647 {
4648         struct buffer_head *bh;
4649         journal_t *journal;
4650         ext4_fsblk_t start;
4651         ext4_fsblk_t len;
4652         int hblock, blocksize;
4653         ext4_fsblk_t sb_block;
4654         unsigned long offset;
4655         struct ext4_super_block *es;
4656         struct block_device *bdev;
4657
4658         BUG_ON(!ext4_has_feature_journal(sb));
4659
4660         bdev = ext4_blkdev_get(j_dev, sb);
4661         if (bdev == NULL)
4662                 return NULL;
4663
4664         blocksize = sb->s_blocksize;
4665         hblock = bdev_logical_block_size(bdev);
4666         if (blocksize < hblock) {
4667                 ext4_msg(sb, KERN_ERR,
4668                         "blocksize too small for journal device");
4669                 goto out_bdev;
4670         }
4671
4672         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4673         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4674         set_blocksize(bdev, blocksize);
4675         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4676                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4677                        "external journal");
4678                 goto out_bdev;
4679         }
4680
4681         es = (struct ext4_super_block *) (bh->b_data + offset);
4682         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4683             !(le32_to_cpu(es->s_feature_incompat) &
4684               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4685                 ext4_msg(sb, KERN_ERR, "external journal has "
4686                                         "bad superblock");
4687                 brelse(bh);
4688                 goto out_bdev;
4689         }
4690
4691         if ((le32_to_cpu(es->s_feature_ro_compat) &
4692              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4693             es->s_checksum != ext4_superblock_csum(sb, es)) {
4694                 ext4_msg(sb, KERN_ERR, "external journal has "
4695                                        "corrupt superblock");
4696                 brelse(bh);
4697                 goto out_bdev;
4698         }
4699
4700         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4701                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4702                 brelse(bh);
4703                 goto out_bdev;
4704         }
4705
4706         len = ext4_blocks_count(es);
4707         start = sb_block + 1;
4708         brelse(bh);     /* we're done with the superblock */
4709
4710         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4711                                         start, len, blocksize);
4712         if (!journal) {
4713                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4714                 goto out_bdev;
4715         }
4716         journal->j_private = sb;
4717         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4718         wait_on_buffer(journal->j_sb_buffer);
4719         if (!buffer_uptodate(journal->j_sb_buffer)) {
4720                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4721                 goto out_journal;
4722         }
4723         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4724                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4725                                         "user (unsupported) - %d",
4726                         be32_to_cpu(journal->j_superblock->s_nr_users));
4727                 goto out_journal;
4728         }
4729         EXT4_SB(sb)->journal_bdev = bdev;
4730         ext4_init_journal_params(sb, journal);
4731         return journal;
4732
4733 out_journal:
4734         jbd2_journal_destroy(journal);
4735 out_bdev:
4736         ext4_blkdev_put(bdev);
4737         return NULL;
4738 }
4739
4740 static int ext4_load_journal(struct super_block *sb,
4741                              struct ext4_super_block *es,
4742                              unsigned long journal_devnum)
4743 {
4744         journal_t *journal;
4745         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4746         dev_t journal_dev;
4747         int err = 0;
4748         int really_read_only;
4749
4750         BUG_ON(!ext4_has_feature_journal(sb));
4751
4752         if (journal_devnum &&
4753             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4754                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4755                         "numbers have changed");
4756                 journal_dev = new_decode_dev(journal_devnum);
4757         } else
4758                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4759
4760         really_read_only = bdev_read_only(sb->s_bdev);
4761
4762         /*
4763          * Are we loading a blank journal or performing recovery after a
4764          * crash?  For recovery, we need to check in advance whether we
4765          * can get read-write access to the device.
4766          */
4767         if (ext4_has_feature_journal_needs_recovery(sb)) {
4768                 if (sb_rdonly(sb)) {
4769                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4770                                         "required on readonly filesystem");
4771                         if (really_read_only) {
4772                                 ext4_msg(sb, KERN_ERR, "write access "
4773                                         "unavailable, cannot proceed "
4774                                         "(try mounting with noload)");
4775                                 return -EROFS;
4776                         }
4777                         ext4_msg(sb, KERN_INFO, "write access will "
4778                                "be enabled during recovery");
4779                 }
4780         }
4781
4782         if (journal_inum && journal_dev) {
4783                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4784                        "and inode journals!");
4785                 return -EINVAL;
4786         }
4787
4788         if (journal_inum) {
4789                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4790                         return -EINVAL;
4791         } else {
4792                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4793                         return -EINVAL;
4794         }
4795
4796         if (!(journal->j_flags & JBD2_BARRIER))
4797                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4798
4799         if (!ext4_has_feature_journal_needs_recovery(sb))
4800                 err = jbd2_journal_wipe(journal, !really_read_only);
4801         if (!err) {
4802                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4803                 if (save)
4804                         memcpy(save, ((char *) es) +
4805                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4806                 err = jbd2_journal_load(journal);
4807                 if (save)
4808                         memcpy(((char *) es) + EXT4_S_ERR_START,
4809                                save, EXT4_S_ERR_LEN);
4810                 kfree(save);
4811         }
4812
4813         if (err) {
4814                 ext4_msg(sb, KERN_ERR, "error loading journal");
4815                 jbd2_journal_destroy(journal);
4816                 return err;
4817         }
4818
4819         EXT4_SB(sb)->s_journal = journal;
4820         ext4_clear_journal_err(sb, es);
4821
4822         if (!really_read_only && journal_devnum &&
4823             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4824                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4825
4826                 /* Make sure we flush the recovery flag to disk. */
4827                 ext4_commit_super(sb, 1);
4828         }
4829
4830         return 0;
4831 }
4832
4833 static int ext4_commit_super(struct super_block *sb, int sync)
4834 {
4835         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4836         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4837         int error = 0;
4838
4839         if (!sbh || block_device_ejected(sb))
4840                 return error;
4841
4842         /*
4843          * The superblock bh should be mapped, but it might not be if the
4844          * device was hot-removed. Not much we can do but fail the I/O.
4845          */
4846         if (!buffer_mapped(sbh))
4847                 return error;
4848
4849         /*
4850          * If the file system is mounted read-only, don't update the
4851          * superblock write time.  This avoids updating the superblock
4852          * write time when we are mounting the root file system
4853          * read/only but we need to replay the journal; at that point,
4854          * for people who are east of GMT and who make their clock
4855          * tick in localtime for Windows bug-for-bug compatibility,
4856          * the clock is set in the future, and this will cause e2fsck
4857          * to complain and force a full file system check.
4858          */
4859         if (!(sb->s_flags & SB_RDONLY))
4860                 ext4_update_tstamp(es, s_wtime);
4861         if (sb->s_bdev->bd_part)
4862                 es->s_kbytes_written =
4863                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4864                             ((part_stat_read(sb->s_bdev->bd_part,
4865                                              sectors[STAT_WRITE]) -
4866                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4867         else
4868                 es->s_kbytes_written =
4869                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4870         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4871                 ext4_free_blocks_count_set(es,
4872                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4873                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4874         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4875                 es->s_free_inodes_count =
4876                         cpu_to_le32(percpu_counter_sum_positive(
4877                                 &EXT4_SB(sb)->s_freeinodes_counter));
4878         BUFFER_TRACE(sbh, "marking dirty");
4879         ext4_superblock_csum_set(sb);
4880         if (sync)
4881                 lock_buffer(sbh);
4882         if (buffer_write_io_error(sbh)) {
4883                 /*
4884                  * Oh, dear.  A previous attempt to write the
4885                  * superblock failed.  This could happen because the
4886                  * USB device was yanked out.  Or it could happen to
4887                  * be a transient write error and maybe the block will
4888                  * be remapped.  Nothing we can do but to retry the
4889                  * write and hope for the best.
4890                  */
4891                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4892                        "superblock detected");
4893                 clear_buffer_write_io_error(sbh);
4894                 set_buffer_uptodate(sbh);
4895         }
4896         mark_buffer_dirty(sbh);
4897         if (sync) {
4898                 unlock_buffer(sbh);
4899                 error = __sync_dirty_buffer(sbh,
4900                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4901                 if (buffer_write_io_error(sbh)) {
4902                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4903                                "superblock");
4904                         clear_buffer_write_io_error(sbh);
4905                         set_buffer_uptodate(sbh);
4906                 }
4907         }
4908         return error;
4909 }
4910
4911 /*
4912  * Have we just finished recovery?  If so, and if we are mounting (or
4913  * remounting) the filesystem readonly, then we will end up with a
4914  * consistent fs on disk.  Record that fact.
4915  */
4916 static void ext4_mark_recovery_complete(struct super_block *sb,
4917                                         struct ext4_super_block *es)
4918 {
4919         journal_t *journal = EXT4_SB(sb)->s_journal;
4920
4921         if (!ext4_has_feature_journal(sb)) {
4922                 BUG_ON(journal != NULL);
4923                 return;
4924         }
4925         jbd2_journal_lock_updates(journal);
4926         if (jbd2_journal_flush(journal) < 0)
4927                 goto out;
4928
4929         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4930                 ext4_clear_feature_journal_needs_recovery(sb);
4931                 ext4_commit_super(sb, 1);
4932         }
4933
4934 out:
4935         jbd2_journal_unlock_updates(journal);
4936 }
4937
4938 /*
4939  * If we are mounting (or read-write remounting) a filesystem whose journal
4940  * has recorded an error from a previous lifetime, move that error to the
4941  * main filesystem now.
4942  */
4943 static void ext4_clear_journal_err(struct super_block *sb,
4944                                    struct ext4_super_block *es)
4945 {
4946         journal_t *journal;
4947         int j_errno;
4948         const char *errstr;
4949
4950         BUG_ON(!ext4_has_feature_journal(sb));
4951
4952         journal = EXT4_SB(sb)->s_journal;
4953
4954         /*
4955          * Now check for any error status which may have been recorded in the
4956          * journal by a prior ext4_error() or ext4_abort()
4957          */
4958
4959         j_errno = jbd2_journal_errno(journal);
4960         if (j_errno) {
4961                 char nbuf[16];
4962
4963                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4964                 ext4_warning(sb, "Filesystem error recorded "
4965                              "from previous mount: %s", errstr);
4966                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4967
4968                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4969                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4970                 ext4_commit_super(sb, 1);
4971
4972                 jbd2_journal_clear_err(journal);
4973                 jbd2_journal_update_sb_errno(journal);
4974         }
4975 }
4976
4977 /*
4978  * Force the running and committing transactions to commit,
4979  * and wait on the commit.
4980  */
4981 int ext4_force_commit(struct super_block *sb)
4982 {
4983         journal_t *journal;
4984
4985         if (sb_rdonly(sb))
4986                 return 0;
4987
4988         journal = EXT4_SB(sb)->s_journal;
4989         return ext4_journal_force_commit(journal);
4990 }
4991
4992 static int ext4_sync_fs(struct super_block *sb, int wait)
4993 {
4994         int ret = 0;
4995         tid_t target;
4996         bool needs_barrier = false;
4997         struct ext4_sb_info *sbi = EXT4_SB(sb);
4998
4999         if (unlikely(ext4_forced_shutdown(sbi)))
5000                 return 0;
5001
5002         trace_ext4_sync_fs(sb, wait);
5003         flush_workqueue(sbi->rsv_conversion_wq);
5004         /*
5005          * Writeback quota in non-journalled quota case - journalled quota has
5006          * no dirty dquots
5007          */
5008         dquot_writeback_dquots(sb, -1);
5009         /*
5010          * Data writeback is possible w/o journal transaction, so barrier must
5011          * being sent at the end of the function. But we can skip it if
5012          * transaction_commit will do it for us.
5013          */
5014         if (sbi->s_journal) {
5015                 target = jbd2_get_latest_transaction(sbi->s_journal);
5016                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5017                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5018                         needs_barrier = true;
5019
5020                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5021                         if (wait)
5022                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5023                                                            target);
5024                 }
5025         } else if (wait && test_opt(sb, BARRIER))
5026                 needs_barrier = true;
5027         if (needs_barrier) {
5028                 int err;
5029                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5030                 if (!ret)
5031                         ret = err;
5032         }
5033
5034         return ret;
5035 }
5036
5037 /*
5038  * LVM calls this function before a (read-only) snapshot is created.  This
5039  * gives us a chance to flush the journal completely and mark the fs clean.
5040  *
5041  * Note that only this function cannot bring a filesystem to be in a clean
5042  * state independently. It relies on upper layer to stop all data & metadata
5043  * modifications.
5044  */
5045 static int ext4_freeze(struct super_block *sb)
5046 {
5047         int error = 0;
5048         journal_t *journal;
5049
5050         if (sb_rdonly(sb))
5051                 return 0;
5052
5053         journal = EXT4_SB(sb)->s_journal;
5054
5055         if (journal) {
5056                 /* Now we set up the journal barrier. */
5057                 jbd2_journal_lock_updates(journal);
5058
5059                 /*
5060                  * Don't clear the needs_recovery flag if we failed to
5061                  * flush the journal.
5062                  */
5063                 error = jbd2_journal_flush(journal);
5064                 if (error < 0)
5065                         goto out;
5066
5067                 /* Journal blocked and flushed, clear needs_recovery flag. */
5068                 ext4_clear_feature_journal_needs_recovery(sb);
5069         }
5070
5071         error = ext4_commit_super(sb, 1);
5072 out:
5073         if (journal)
5074                 /* we rely on upper layer to stop further updates */
5075                 jbd2_journal_unlock_updates(journal);
5076         return error;
5077 }
5078
5079 /*
5080  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5081  * flag here, even though the filesystem is not technically dirty yet.
5082  */
5083 static int ext4_unfreeze(struct super_block *sb)
5084 {
5085         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5086                 return 0;
5087
5088         if (EXT4_SB(sb)->s_journal) {
5089                 /* Reset the needs_recovery flag before the fs is unlocked. */
5090                 ext4_set_feature_journal_needs_recovery(sb);
5091         }
5092
5093         ext4_commit_super(sb, 1);
5094         return 0;
5095 }
5096
5097 /*
5098  * Structure to save mount options for ext4_remount's benefit
5099  */
5100 struct ext4_mount_options {
5101         unsigned long s_mount_opt;
5102         unsigned long s_mount_opt2;
5103         kuid_t s_resuid;
5104         kgid_t s_resgid;
5105         unsigned long s_commit_interval;
5106         u32 s_min_batch_time, s_max_batch_time;
5107 #ifdef CONFIG_QUOTA
5108         int s_jquota_fmt;
5109         char *s_qf_names[EXT4_MAXQUOTAS];
5110 #endif
5111 };
5112
5113 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5114 {
5115         struct ext4_super_block *es;
5116         struct ext4_sb_info *sbi = EXT4_SB(sb);
5117         unsigned long old_sb_flags;
5118         struct ext4_mount_options old_opts;
5119         int enable_quota = 0;
5120         ext4_group_t g;
5121         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5122         int err = 0;
5123 #ifdef CONFIG_QUOTA
5124         int i, j;
5125         char *to_free[EXT4_MAXQUOTAS];
5126 #endif
5127         char *orig_data = kstrdup(data, GFP_KERNEL);
5128
5129         if (data && !orig_data)
5130                 return -ENOMEM;
5131
5132         /* Store the original options */
5133         old_sb_flags = sb->s_flags;
5134         old_opts.s_mount_opt = sbi->s_mount_opt;
5135         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5136         old_opts.s_resuid = sbi->s_resuid;
5137         old_opts.s_resgid = sbi->s_resgid;
5138         old_opts.s_commit_interval = sbi->s_commit_interval;
5139         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5140         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5141 #ifdef CONFIG_QUOTA
5142         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5143         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5144                 if (sbi->s_qf_names[i]) {
5145                         char *qf_name = get_qf_name(sb, sbi, i);
5146
5147                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5148                         if (!old_opts.s_qf_names[i]) {
5149                                 for (j = 0; j < i; j++)
5150                                         kfree(old_opts.s_qf_names[j]);
5151                                 kfree(orig_data);
5152                                 return -ENOMEM;
5153                         }
5154                 } else
5155                         old_opts.s_qf_names[i] = NULL;
5156 #endif
5157         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5158                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5159
5160         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5161                 err = -EINVAL;
5162                 goto restore_opts;
5163         }
5164
5165         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5166             test_opt(sb, JOURNAL_CHECKSUM)) {
5167                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5168                          "during remount not supported; ignoring");
5169                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5170         }
5171
5172         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5173                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5174                         ext4_msg(sb, KERN_ERR, "can't mount with "
5175                                  "both data=journal and delalloc");
5176                         err = -EINVAL;
5177                         goto restore_opts;
5178                 }
5179                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5180                         ext4_msg(sb, KERN_ERR, "can't mount with "
5181                                  "both data=journal and dioread_nolock");
5182                         err = -EINVAL;
5183                         goto restore_opts;
5184                 }
5185                 if (test_opt(sb, DAX)) {
5186                         ext4_msg(sb, KERN_ERR, "can't mount with "
5187                                  "both data=journal and dax");
5188                         err = -EINVAL;
5189                         goto restore_opts;
5190                 }
5191         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5192                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5193                         ext4_msg(sb, KERN_ERR, "can't mount with "
5194                                 "journal_async_commit in data=ordered mode");
5195                         err = -EINVAL;
5196                         goto restore_opts;
5197                 }
5198         }
5199
5200         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5201                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5202                 err = -EINVAL;
5203                 goto restore_opts;
5204         }
5205
5206         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5207                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5208                         "dax flag with busy inodes while remounting");
5209                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5210         }
5211
5212         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5213                 ext4_abort(sb, "Abort forced by user");
5214
5215         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5216                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5217
5218         es = sbi->s_es;
5219
5220         if (sbi->s_journal) {
5221                 ext4_init_journal_params(sb, sbi->s_journal);
5222                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5223         }
5224
5225         if (*flags & SB_LAZYTIME)
5226                 sb->s_flags |= SB_LAZYTIME;
5227
5228         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5229                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5230                         err = -EROFS;
5231                         goto restore_opts;
5232                 }
5233
5234                 if (*flags & SB_RDONLY) {
5235                         err = sync_filesystem(sb);
5236                         if (err < 0)
5237                                 goto restore_opts;
5238                         err = dquot_suspend(sb, -1);
5239                         if (err < 0)
5240                                 goto restore_opts;
5241
5242                         /*
5243                          * First of all, the unconditional stuff we have to do
5244                          * to disable replay of the journal when we next remount
5245                          */
5246                         sb->s_flags |= SB_RDONLY;
5247
5248                         /*
5249                          * OK, test if we are remounting a valid rw partition
5250                          * readonly, and if so set the rdonly flag and then
5251                          * mark the partition as valid again.
5252                          */
5253                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5254                             (sbi->s_mount_state & EXT4_VALID_FS))
5255                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5256
5257                         if (sbi->s_journal)
5258                                 ext4_mark_recovery_complete(sb, es);
5259                         if (sbi->s_mmp_tsk)
5260                                 kthread_stop(sbi->s_mmp_tsk);
5261                 } else {
5262                         /* Make sure we can mount this feature set readwrite */
5263                         if (ext4_has_feature_readonly(sb) ||
5264                             !ext4_feature_set_ok(sb, 0)) {
5265                                 err = -EROFS;
5266                                 goto restore_opts;
5267                         }
5268                         /*
5269                          * Make sure the group descriptor checksums
5270                          * are sane.  If they aren't, refuse to remount r/w.
5271                          */
5272                         for (g = 0; g < sbi->s_groups_count; g++) {
5273                                 struct ext4_group_desc *gdp =
5274                                         ext4_get_group_desc(sb, g, NULL);
5275
5276                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5277                                         ext4_msg(sb, KERN_ERR,
5278                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5279                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5280                                                le16_to_cpu(gdp->bg_checksum));
5281                                         err = -EFSBADCRC;
5282                                         goto restore_opts;
5283                                 }
5284                         }
5285
5286                         /*
5287                          * If we have an unprocessed orphan list hanging
5288                          * around from a previously readonly bdev mount,
5289                          * require a full umount/remount for now.
5290                          */
5291                         if (es->s_last_orphan) {
5292                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5293                                        "remount RDWR because of unprocessed "
5294                                        "orphan inode list.  Please "
5295                                        "umount/remount instead");
5296                                 err = -EINVAL;
5297                                 goto restore_opts;
5298                         }
5299
5300                         /*
5301                          * Mounting a RDONLY partition read-write, so reread
5302                          * and store the current valid flag.  (It may have
5303                          * been changed by e2fsck since we originally mounted
5304                          * the partition.)
5305                          */
5306                         if (sbi->s_journal)
5307                                 ext4_clear_journal_err(sb, es);
5308                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5309
5310                         err = ext4_setup_super(sb, es, 0);
5311                         if (err)
5312                                 goto restore_opts;
5313
5314                         sb->s_flags &= ~SB_RDONLY;
5315                         if (ext4_has_feature_mmp(sb))
5316                                 if (ext4_multi_mount_protect(sb,
5317                                                 le64_to_cpu(es->s_mmp_block))) {
5318                                         err = -EROFS;
5319                                         goto restore_opts;
5320                                 }
5321                         enable_quota = 1;
5322                 }
5323         }
5324
5325         /*
5326          * Reinitialize lazy itable initialization thread based on
5327          * current settings
5328          */
5329         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5330                 ext4_unregister_li_request(sb);
5331         else {
5332                 ext4_group_t first_not_zeroed;
5333                 first_not_zeroed = ext4_has_uninit_itable(sb);
5334                 ext4_register_li_request(sb, first_not_zeroed);
5335         }
5336
5337         ext4_setup_system_zone(sb);
5338         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5339                 err = ext4_commit_super(sb, 1);
5340                 if (err)
5341                         goto restore_opts;
5342         }
5343
5344 #ifdef CONFIG_QUOTA
5345         /* Release old quota file names */
5346         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5347                 kfree(old_opts.s_qf_names[i]);
5348         if (enable_quota) {
5349                 if (sb_any_quota_suspended(sb))
5350                         dquot_resume(sb, -1);
5351                 else if (ext4_has_feature_quota(sb)) {
5352                         err = ext4_enable_quotas(sb);
5353                         if (err)
5354                                 goto restore_opts;
5355                 }
5356         }
5357 #endif
5358
5359         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5360         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5361         kfree(orig_data);
5362         return 0;
5363
5364 restore_opts:
5365         sb->s_flags = old_sb_flags;
5366         sbi->s_mount_opt = old_opts.s_mount_opt;
5367         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5368         sbi->s_resuid = old_opts.s_resuid;
5369         sbi->s_resgid = old_opts.s_resgid;
5370         sbi->s_commit_interval = old_opts.s_commit_interval;
5371         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5372         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5373 #ifdef CONFIG_QUOTA
5374         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5375         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5376                 to_free[i] = get_qf_name(sb, sbi, i);
5377                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5378         }
5379         synchronize_rcu();
5380         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5381                 kfree(to_free[i]);
5382 #endif
5383         kfree(orig_data);
5384         return err;
5385 }
5386
5387 #ifdef CONFIG_QUOTA
5388 static int ext4_statfs_project(struct super_block *sb,
5389                                kprojid_t projid, struct kstatfs *buf)
5390 {
5391         struct kqid qid;
5392         struct dquot *dquot;
5393         u64 limit;
5394         u64 curblock;
5395
5396         qid = make_kqid_projid(projid);
5397         dquot = dqget(sb, qid);
5398         if (IS_ERR(dquot))
5399                 return PTR_ERR(dquot);
5400         spin_lock(&dquot->dq_dqb_lock);
5401
5402         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5403                  dquot->dq_dqb.dqb_bsoftlimit :
5404                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5405         if (limit && buf->f_blocks > limit) {
5406                 curblock = (dquot->dq_dqb.dqb_curspace +
5407                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5408                 buf->f_blocks = limit;
5409                 buf->f_bfree = buf->f_bavail =
5410                         (buf->f_blocks > curblock) ?
5411                          (buf->f_blocks - curblock) : 0;
5412         }
5413
5414         limit = dquot->dq_dqb.dqb_isoftlimit ?
5415                 dquot->dq_dqb.dqb_isoftlimit :
5416                 dquot->dq_dqb.dqb_ihardlimit;
5417         if (limit && buf->f_files > limit) {
5418                 buf->f_files = limit;
5419                 buf->f_ffree =
5420                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5421                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5422         }
5423
5424         spin_unlock(&dquot->dq_dqb_lock);
5425         dqput(dquot);
5426         return 0;
5427 }
5428 #endif
5429
5430 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5431 {
5432         struct super_block *sb = dentry->d_sb;
5433         struct ext4_sb_info *sbi = EXT4_SB(sb);
5434         struct ext4_super_block *es = sbi->s_es;
5435         ext4_fsblk_t overhead = 0, resv_blocks;
5436         u64 fsid;
5437         s64 bfree;
5438         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5439
5440         if (!test_opt(sb, MINIX_DF))
5441                 overhead = sbi->s_overhead;
5442
5443         buf->f_type = EXT4_SUPER_MAGIC;
5444         buf->f_bsize = sb->s_blocksize;
5445         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5446         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5447                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5448         /* prevent underflow in case that few free space is available */
5449         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5450         buf->f_bavail = buf->f_bfree -
5451                         (ext4_r_blocks_count(es) + resv_blocks);
5452         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5453                 buf->f_bavail = 0;
5454         buf->f_files = le32_to_cpu(es->s_inodes_count);
5455         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5456         buf->f_namelen = EXT4_NAME_LEN;
5457         fsid = le64_to_cpup((void *)es->s_uuid) ^
5458                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5459         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5460         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5461
5462 #ifdef CONFIG_QUOTA
5463         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5464             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5465                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5466 #endif
5467         return 0;
5468 }
5469
5470
5471 #ifdef CONFIG_QUOTA
5472
5473 /*
5474  * Helper functions so that transaction is started before we acquire dqio_sem
5475  * to keep correct lock ordering of transaction > dqio_sem
5476  */
5477 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5478 {
5479         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5480 }
5481
5482 static int ext4_write_dquot(struct dquot *dquot)
5483 {
5484         int ret, err;
5485         handle_t *handle;
5486         struct inode *inode;
5487
5488         inode = dquot_to_inode(dquot);
5489         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5490                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5491         if (IS_ERR(handle))
5492                 return PTR_ERR(handle);
5493         ret = dquot_commit(dquot);
5494         err = ext4_journal_stop(handle);
5495         if (!ret)
5496                 ret = err;
5497         return ret;
5498 }
5499
5500 static int ext4_acquire_dquot(struct dquot *dquot)
5501 {
5502         int ret, err;
5503         handle_t *handle;
5504
5505         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5506                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5507         if (IS_ERR(handle))
5508                 return PTR_ERR(handle);
5509         ret = dquot_acquire(dquot);
5510         err = ext4_journal_stop(handle);
5511         if (!ret)
5512                 ret = err;
5513         return ret;
5514 }
5515
5516 static int ext4_release_dquot(struct dquot *dquot)
5517 {
5518         int ret, err;
5519         handle_t *handle;
5520
5521         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5522                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5523         if (IS_ERR(handle)) {
5524                 /* Release dquot anyway to avoid endless cycle in dqput() */
5525                 dquot_release(dquot);
5526                 return PTR_ERR(handle);
5527         }
5528         ret = dquot_release(dquot);
5529         err = ext4_journal_stop(handle);
5530         if (!ret)
5531                 ret = err;
5532         return ret;
5533 }
5534
5535 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5536 {
5537         struct super_block *sb = dquot->dq_sb;
5538         struct ext4_sb_info *sbi = EXT4_SB(sb);
5539
5540         /* Are we journaling quotas? */
5541         if (ext4_has_feature_quota(sb) ||
5542             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5543                 dquot_mark_dquot_dirty(dquot);
5544                 return ext4_write_dquot(dquot);
5545         } else {
5546                 return dquot_mark_dquot_dirty(dquot);
5547         }
5548 }
5549
5550 static int ext4_write_info(struct super_block *sb, int type)
5551 {
5552         int ret, err;
5553         handle_t *handle;
5554
5555         /* Data block + inode block */
5556         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5557         if (IS_ERR(handle))
5558                 return PTR_ERR(handle);
5559         ret = dquot_commit_info(sb, type);
5560         err = ext4_journal_stop(handle);
5561         if (!ret)
5562                 ret = err;
5563         return ret;
5564 }
5565
5566 /*
5567  * Turn on quotas during mount time - we need to find
5568  * the quota file and such...
5569  */
5570 static int ext4_quota_on_mount(struct super_block *sb, int type)
5571 {
5572         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5573                                         EXT4_SB(sb)->s_jquota_fmt, type);
5574 }
5575
5576 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5577 {
5578         struct ext4_inode_info *ei = EXT4_I(inode);
5579
5580         /* The first argument of lockdep_set_subclass has to be
5581          * *exactly* the same as the argument to init_rwsem() --- in
5582          * this case, in init_once() --- or lockdep gets unhappy
5583          * because the name of the lock is set using the
5584          * stringification of the argument to init_rwsem().
5585          */
5586         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5587         lockdep_set_subclass(&ei->i_data_sem, subclass);
5588 }
5589
5590 /*
5591  * Standard function to be called on quota_on
5592  */
5593 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5594                          const struct path *path)
5595 {
5596         int err;
5597
5598         if (!test_opt(sb, QUOTA))
5599                 return -EINVAL;
5600
5601         /* Quotafile not on the same filesystem? */
5602         if (path->dentry->d_sb != sb)
5603                 return -EXDEV;
5604         /* Journaling quota? */
5605         if (EXT4_SB(sb)->s_qf_names[type]) {
5606                 /* Quotafile not in fs root? */
5607                 if (path->dentry->d_parent != sb->s_root)
5608                         ext4_msg(sb, KERN_WARNING,
5609                                 "Quota file not on filesystem root. "
5610                                 "Journaled quota will not work");
5611                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5612         } else {
5613                 /*
5614                  * Clear the flag just in case mount options changed since
5615                  * last time.
5616                  */
5617                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5618         }
5619
5620         /*
5621          * When we journal data on quota file, we have to flush journal to see
5622          * all updates to the file when we bypass pagecache...
5623          */
5624         if (EXT4_SB(sb)->s_journal &&
5625             ext4_should_journal_data(d_inode(path->dentry))) {
5626                 /*
5627                  * We don't need to lock updates but journal_flush() could
5628                  * otherwise be livelocked...
5629                  */
5630                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5631                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5632                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5633                 if (err)
5634                         return err;
5635         }
5636
5637         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5638         err = dquot_quota_on(sb, type, format_id, path);
5639         if (err) {
5640                 lockdep_set_quota_inode(path->dentry->d_inode,
5641                                              I_DATA_SEM_NORMAL);
5642         } else {
5643                 struct inode *inode = d_inode(path->dentry);
5644                 handle_t *handle;
5645
5646                 /*
5647                  * Set inode flags to prevent userspace from messing with quota
5648                  * files. If this fails, we return success anyway since quotas
5649                  * are already enabled and this is not a hard failure.
5650                  */
5651                 inode_lock(inode);
5652                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5653                 if (IS_ERR(handle))
5654                         goto unlock_inode;
5655                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5656                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5657                                 S_NOATIME | S_IMMUTABLE);
5658                 ext4_mark_inode_dirty(handle, inode);
5659                 ext4_journal_stop(handle);
5660         unlock_inode:
5661                 inode_unlock(inode);
5662         }
5663         return err;
5664 }
5665
5666 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5667                              unsigned int flags)
5668 {
5669         int err;
5670         struct inode *qf_inode;
5671         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5672                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5673                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5674                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5675         };
5676
5677         BUG_ON(!ext4_has_feature_quota(sb));
5678
5679         if (!qf_inums[type])
5680                 return -EPERM;
5681
5682         qf_inode = ext4_iget(sb, qf_inums[type]);
5683         if (IS_ERR(qf_inode)) {
5684                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5685                 return PTR_ERR(qf_inode);
5686         }
5687
5688         /* Don't account quota for quota files to avoid recursion */
5689         qf_inode->i_flags |= S_NOQUOTA;
5690         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5691         err = dquot_enable(qf_inode, type, format_id, flags);
5692         iput(qf_inode);
5693         if (err)
5694                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5695
5696         return err;
5697 }
5698
5699 /* Enable usage tracking for all quota types. */
5700 static int ext4_enable_quotas(struct super_block *sb)
5701 {
5702         int type, err = 0;
5703         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5704                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5705                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5706                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5707         };
5708         bool quota_mopt[EXT4_MAXQUOTAS] = {
5709                 test_opt(sb, USRQUOTA),
5710                 test_opt(sb, GRPQUOTA),
5711                 test_opt(sb, PRJQUOTA),
5712         };
5713
5714         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5715         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5716                 if (qf_inums[type]) {
5717                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5718                                 DQUOT_USAGE_ENABLED |
5719                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5720                         if (err) {
5721                                 ext4_warning(sb,
5722                                         "Failed to enable quota tracking "
5723                                         "(type=%d, err=%d). Please run "
5724                                         "e2fsck to fix.", type, err);
5725                                 for (type--; type >= 0; type--)
5726                                         dquot_quota_off(sb, type);
5727
5728                                 return err;
5729                         }
5730                 }
5731         }
5732         return 0;
5733 }
5734
5735 static int ext4_quota_off(struct super_block *sb, int type)
5736 {
5737         struct inode *inode = sb_dqopt(sb)->files[type];
5738         handle_t *handle;
5739         int err;
5740
5741         /* Force all delayed allocation blocks to be allocated.
5742          * Caller already holds s_umount sem */
5743         if (test_opt(sb, DELALLOC))
5744                 sync_filesystem(sb);
5745
5746         if (!inode || !igrab(inode))
5747                 goto out;
5748
5749         err = dquot_quota_off(sb, type);
5750         if (err || ext4_has_feature_quota(sb))
5751                 goto out_put;
5752
5753         inode_lock(inode);
5754         /*
5755          * Update modification times of quota files when userspace can
5756          * start looking at them. If we fail, we return success anyway since
5757          * this is not a hard failure and quotas are already disabled.
5758          */
5759         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5760         if (IS_ERR(handle))
5761                 goto out_unlock;
5762         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5763         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5764         inode->i_mtime = inode->i_ctime = current_time(inode);
5765         ext4_mark_inode_dirty(handle, inode);
5766         ext4_journal_stop(handle);
5767 out_unlock:
5768         inode_unlock(inode);
5769 out_put:
5770         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5771         iput(inode);
5772         return err;
5773 out:
5774         return dquot_quota_off(sb, type);
5775 }
5776
5777 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5778  * acquiring the locks... As quota files are never truncated and quota code
5779  * itself serializes the operations (and no one else should touch the files)
5780  * we don't have to be afraid of races */
5781 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5782                                size_t len, loff_t off)
5783 {
5784         struct inode *inode = sb_dqopt(sb)->files[type];
5785         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5786         int offset = off & (sb->s_blocksize - 1);
5787         int tocopy;
5788         size_t toread;
5789         struct buffer_head *bh;
5790         loff_t i_size = i_size_read(inode);
5791
5792         if (off > i_size)
5793                 return 0;
5794         if (off+len > i_size)
5795                 len = i_size-off;
5796         toread = len;
5797         while (toread > 0) {
5798                 tocopy = sb->s_blocksize - offset < toread ?
5799                                 sb->s_blocksize - offset : toread;
5800                 bh = ext4_bread(NULL, inode, blk, 0);
5801                 if (IS_ERR(bh))
5802                         return PTR_ERR(bh);
5803                 if (!bh)        /* A hole? */
5804                         memset(data, 0, tocopy);
5805                 else
5806                         memcpy(data, bh->b_data+offset, tocopy);
5807                 brelse(bh);
5808                 offset = 0;
5809                 toread -= tocopy;
5810                 data += tocopy;
5811                 blk++;
5812         }
5813         return len;
5814 }
5815
5816 /* Write to quotafile (we know the transaction is already started and has
5817  * enough credits) */
5818 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5819                                 const char *data, size_t len, loff_t off)
5820 {
5821         struct inode *inode = sb_dqopt(sb)->files[type];
5822         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5823         int err, offset = off & (sb->s_blocksize - 1);
5824         int retries = 0;
5825         struct buffer_head *bh;
5826         handle_t *handle = journal_current_handle();
5827
5828         if (EXT4_SB(sb)->s_journal && !handle) {
5829                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5830                         " cancelled because transaction is not started",
5831                         (unsigned long long)off, (unsigned long long)len);
5832                 return -EIO;
5833         }
5834         /*
5835          * Since we account only one data block in transaction credits,
5836          * then it is impossible to cross a block boundary.
5837          */
5838         if (sb->s_blocksize - offset < len) {
5839                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5840                         " cancelled because not block aligned",
5841                         (unsigned long long)off, (unsigned long long)len);
5842                 return -EIO;
5843         }
5844
5845         do {
5846                 bh = ext4_bread(handle, inode, blk,
5847                                 EXT4_GET_BLOCKS_CREATE |
5848                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5849         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5850                  ext4_should_retry_alloc(inode->i_sb, &retries));
5851         if (IS_ERR(bh))
5852                 return PTR_ERR(bh);
5853         if (!bh)
5854                 goto out;
5855         BUFFER_TRACE(bh, "get write access");
5856         err = ext4_journal_get_write_access(handle, bh);
5857         if (err) {
5858                 brelse(bh);
5859                 return err;
5860         }
5861         lock_buffer(bh);
5862         memcpy(bh->b_data+offset, data, len);
5863         flush_dcache_page(bh->b_page);
5864         unlock_buffer(bh);
5865         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5866         brelse(bh);
5867 out:
5868         if (inode->i_size < off + len) {
5869                 i_size_write(inode, off + len);
5870                 EXT4_I(inode)->i_disksize = inode->i_size;
5871                 ext4_mark_inode_dirty(handle, inode);
5872         }
5873         return len;
5874 }
5875
5876 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5877 {
5878         const struct quota_format_ops   *ops;
5879
5880         if (!sb_has_quota_loaded(sb, qid->type))
5881                 return -ESRCH;
5882         ops = sb_dqopt(sb)->ops[qid->type];
5883         if (!ops || !ops->get_next_id)
5884                 return -ENOSYS;
5885         return dquot_get_next_id(sb, qid);
5886 }
5887 #endif
5888
5889 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5890                        const char *dev_name, void *data)
5891 {
5892         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5893 }
5894
5895 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5896 static inline void register_as_ext2(void)
5897 {
5898         int err = register_filesystem(&ext2_fs_type);
5899         if (err)
5900                 printk(KERN_WARNING
5901                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5902 }
5903
5904 static inline void unregister_as_ext2(void)
5905 {
5906         unregister_filesystem(&ext2_fs_type);
5907 }
5908
5909 static inline int ext2_feature_set_ok(struct super_block *sb)
5910 {
5911         if (ext4_has_unknown_ext2_incompat_features(sb))
5912                 return 0;
5913         if (sb_rdonly(sb))
5914                 return 1;
5915         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5916                 return 0;
5917         return 1;
5918 }
5919 #else
5920 static inline void register_as_ext2(void) { }
5921 static inline void unregister_as_ext2(void) { }
5922 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5923 #endif
5924
5925 static inline void register_as_ext3(void)
5926 {
5927         int err = register_filesystem(&ext3_fs_type);
5928         if (err)
5929                 printk(KERN_WARNING
5930                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5931 }
5932
5933 static inline void unregister_as_ext3(void)
5934 {
5935         unregister_filesystem(&ext3_fs_type);
5936 }
5937
5938 static inline int ext3_feature_set_ok(struct super_block *sb)
5939 {
5940         if (ext4_has_unknown_ext3_incompat_features(sb))
5941                 return 0;
5942         if (!ext4_has_feature_journal(sb))
5943                 return 0;
5944         if (sb_rdonly(sb))
5945                 return 1;
5946         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5947                 return 0;
5948         return 1;
5949 }
5950
5951 static struct file_system_type ext4_fs_type = {
5952         .owner          = THIS_MODULE,
5953         .name           = "ext4",
5954         .mount          = ext4_mount,
5955         .kill_sb        = kill_block_super,
5956         .fs_flags       = FS_REQUIRES_DEV,
5957 };
5958 MODULE_ALIAS_FS("ext4");
5959
5960 /* Shared across all ext4 file systems */
5961 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5962
5963 static int __init ext4_init_fs(void)
5964 {
5965         int i, err;
5966
5967         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5968         ext4_li_info = NULL;
5969         mutex_init(&ext4_li_mtx);
5970
5971         /* Build-time check for flags consistency */
5972         ext4_check_flag_values();
5973
5974         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5975                 init_waitqueue_head(&ext4__ioend_wq[i]);
5976
5977         err = ext4_init_es();
5978         if (err)
5979                 return err;
5980
5981         err = ext4_init_pending();
5982         if (err)
5983                 goto out6;
5984
5985         err = ext4_init_pageio();
5986         if (err)
5987                 goto out5;
5988
5989         err = ext4_init_system_zone();
5990         if (err)
5991                 goto out4;
5992
5993         err = ext4_init_sysfs();
5994         if (err)
5995                 goto out3;
5996
5997         err = ext4_init_mballoc();
5998         if (err)
5999                 goto out2;
6000         err = init_inodecache();
6001         if (err)
6002                 goto out1;
6003         register_as_ext3();
6004         register_as_ext2();
6005         err = register_filesystem(&ext4_fs_type);
6006         if (err)
6007                 goto out;
6008
6009         return 0;
6010 out:
6011         unregister_as_ext2();
6012         unregister_as_ext3();
6013         destroy_inodecache();
6014 out1:
6015         ext4_exit_mballoc();
6016 out2:
6017         ext4_exit_sysfs();
6018 out3:
6019         ext4_exit_system_zone();
6020 out4:
6021         ext4_exit_pageio();
6022 out5:
6023         ext4_exit_pending();
6024 out6:
6025         ext4_exit_es();
6026
6027         return err;
6028 }
6029
6030 static void __exit ext4_exit_fs(void)
6031 {
6032         ext4_destroy_lazyinit_thread();
6033         unregister_as_ext2();
6034         unregister_as_ext3();
6035         unregister_filesystem(&ext4_fs_type);
6036         destroy_inodecache();
6037         ext4_exit_mballoc();
6038         ext4_exit_sysfs();
6039         ext4_exit_system_zone();
6040         ext4_exit_pageio();
6041         ext4_exit_es();
6042         ext4_exit_pending();
6043 }
6044
6045 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6046 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6047 MODULE_LICENSE("GPL");
6048 MODULE_SOFTDEP("pre: crc32c");
6049 module_init(ext4_init_fs)
6050 module_exit(ext4_exit_fs)