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 /*
 144  * This works like sb_bread() except it uses ERR_PTR for error
 145  * returns.  Currently with sb_bread it's impossible to distinguish
 146  * between ENOMEM and EIO situations (since both result in a NULL
 147  * return.
 148  */
 149 struct buffer_head *
 150 ext4_sb_bread(struct super_block *sb, sector_t block, int op_flags)
 151 {
 152         struct buffer_head *bh = sb_getblk(sb, block);
 153
 154         if (bh == NULL)
 155                 return ERR_PTR(-ENOMEM);
 156         if (buffer_uptodate(bh))
 157                 return bh;
 158         ll_rw_block(REQ_OP_READ, REQ_META | op_flags, 1, &bh);
 159         wait_on_buffer(bh);
 160         if (buffer_uptodate(bh))
 161                 return bh;
 162         put_bh(bh);
 163         return ERR_PTR(-EIO);
 164 }
 165
 166 static int ext4_verify_csum_type(struct super_block *sb,
 167                                  struct ext4_super_block *es)
 168 {
 169         if (!ext4_has_feature_metadata_csum(sb))
 170                 return 1;
 171
 172         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 173 }
 174
 175 static __le32 ext4_superblock_csum(struct super_block *sb,
 176                                    struct ext4_super_block *es)
 177 {
 178         struct ext4_sb_info *sbi = EXT4_SB(sb);
 179         int offset = offsetof(struct ext4_super_block, s_checksum);
 180         __u32 csum;
 181
 182         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 183
 184         return cpu_to_le32(csum);
 185 }
 186
 187 static int ext4_superblock_csum_verify(struct super_block *sb,
 188                                        struct ext4_super_block *es)
 189 {
 190         if (!ext4_has_metadata_csum(sb))
 191                 return 1;
 192
 193         return es->s_checksum == ext4_superblock_csum(sb, es);
 194 }
 195
 196 void ext4_superblock_csum_set(struct super_block *sb)
 197 {
 198         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 199
 200         if (!ext4_has_metadata_csum(sb))
 201                 return;
 202
 203         es->s_checksum = ext4_superblock_csum(sb, es);
 204 }
 205
 206 void *ext4_kvmalloc(size_t size, gfp_t flags)
 207 {
 208         void *ret;
 209
 210         ret = kmalloc(size, flags | __GFP_NOWARN);
 211         if (!ret)
 212                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 213         return ret;
 214 }
 215
 216 void *ext4_kvzalloc(size_t size, gfp_t flags)
 217 {
 218         void *ret;
 219
 220         ret = kzalloc(size, flags | __GFP_NOWARN);
 221         if (!ret)
 222                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 223         return ret;
 224 }
 225
 226 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 227                                struct ext4_group_desc *bg)
 228 {
 229         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 230                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 231                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 232 }
 233
 234 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 235                                struct ext4_group_desc *bg)
 236 {
 237         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 238                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 239                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 240 }
 241
 242 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 243                               struct ext4_group_desc *bg)
 244 {
 245         return le32_to_cpu(bg->bg_inode_table_lo) |
 246                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 247                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 248 }
 249
 250 __u32 ext4_free_group_clusters(struct super_block *sb,
 251                                struct ext4_group_desc *bg)
 252 {
 253         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 254                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 255                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 256 }
 257
 258 __u32 ext4_free_inodes_count(struct super_block *sb,
 259                               struct ext4_group_desc *bg)
 260 {
 261         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 262                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 263                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 264 }
 265
 266 __u32 ext4_used_dirs_count(struct super_block *sb,
 267                               struct ext4_group_desc *bg)
 268 {
 269         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 270                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 271                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 272 }
 273
 274 __u32 ext4_itable_unused_count(struct super_block *sb,
 275                               struct ext4_group_desc *bg)
 276 {
 277         return le16_to_cpu(bg->bg_itable_unused_lo) |
 278                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 279                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 280 }
 281
 282 void ext4_block_bitmap_set(struct super_block *sb,
 283                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 284 {
 285         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 286         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 287                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 288 }
 289
 290 void ext4_inode_bitmap_set(struct super_block *sb,
 291                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 292 {
 293         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 294         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 295                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 296 }
 297
 298 void ext4_inode_table_set(struct super_block *sb,
 299                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 300 {
 301         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 302         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 303                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 304 }
 305
 306 void ext4_free_group_clusters_set(struct super_block *sb,
 307                                   struct ext4_group_desc *bg, __u32 count)
 308 {
 309         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 310         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 311                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 312 }
 313
 314 void ext4_free_inodes_set(struct super_block *sb,
 315                           struct ext4_group_desc *bg, __u32 count)
 316 {
 317         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 318         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 319                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 320 }
 321
 322 void ext4_used_dirs_set(struct super_block *sb,
 323                           struct ext4_group_desc *bg, __u32 count)
 324 {
 325         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 326         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 327                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 328 }
 329
 330 void ext4_itable_unused_set(struct super_block *sb,
 331                           struct ext4_group_desc *bg, __u32 count)
 332 {
 333         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 334         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 335                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 336 }
 337
 338 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi)
 339 {
 340         time64_t now = ktime_get_real_seconds();
 341
 342         now = clamp_val(now, 0, (1ull << 40) - 1);
 343
 344         *lo = cpu_to_le32(lower_32_bits(now));
 345         *hi = upper_32_bits(now);
 346 }
 347
 348 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 349 {
 350         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 351 }
 352 #define ext4_update_tstamp(es, tstamp) \
 353         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 354 #define ext4_get_tstamp(es, tstamp) \
 355         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 356
 357 static void __save_error_info(struct super_block *sb, const char *func,
 358                             unsigned int line)
 359 {
 360         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 361
 362         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 363         if (bdev_read_only(sb->s_bdev))
 364                 return;
 365         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 366         ext4_update_tstamp(es, s_last_error_time);
 367         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 368         es->s_last_error_line = cpu_to_le32(line);
 369         if (!es->s_first_error_time) {
 370                 es->s_first_error_time = es->s_last_error_time;
 371                 es->s_first_error_time_hi = es->s_last_error_time_hi;
 372                 strncpy(es->s_first_error_func, func,
 373                         sizeof(es->s_first_error_func));
 374                 es->s_first_error_line = cpu_to_le32(line);
 375                 es->s_first_error_ino = es->s_last_error_ino;
 376                 es->s_first_error_block = es->s_last_error_block;
 377         }
 378         /*
 379          * Start the daily error reporting function if it hasn't been
 380          * started already
 381          */
 382         if (!es->s_error_count)
 383                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 384         le32_add_cpu(&es->s_error_count, 1);
 385 }
 386
 387 static void save_error_info(struct super_block *sb, const char *func,
 388                             unsigned int line)
 389 {
 390         __save_error_info(sb, func, line);
 391         ext4_commit_super(sb, 1);
 392 }
 393
 394 /*
 395  * The del_gendisk() function uninitializes the disk-specific data
 396  * structures, including the bdi structure, without telling anyone
 397  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 398  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 399  * This is a kludge to prevent these oops until we can put in a proper
 400  * hook in del_gendisk() to inform the VFS and file system layers.
 401  */
 402 static int block_device_ejected(struct super_block *sb)
 403 {
 404         struct inode *bd_inode = sb->s_bdev->bd_inode;
 405         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 406
 407         return bdi->dev == NULL;
 408 }
 409
 410 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 411 {
 412         struct super_block              *sb = journal->j_private;
 413         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 414         int                             error = is_journal_aborted(journal);
 415         struct ext4_journal_cb_entry    *jce;
 416
 417         BUG_ON(txn->t_state == T_FINISHED);
 418
 419         ext4_process_freed_data(sb, txn->t_tid);
 420
 421         spin_lock(&sbi->s_md_lock);
 422         while (!list_empty(&txn->t_private_list)) {
 423                 jce = list_entry(txn->t_private_list.next,
 424                                  struct ext4_journal_cb_entry, jce_list);
 425                 list_del_init(&jce->jce_list);
 426                 spin_unlock(&sbi->s_md_lock);
 427                 jce->jce_func(sb, jce, error);
 428                 spin_lock(&sbi->s_md_lock);
 429         }
 430         spin_unlock(&sbi->s_md_lock);
 431 }
 432
 433 /* Deal with the reporting of failure conditions on a filesystem such as
 434  * inconsistencies detected or read IO failures.
 435  *
 436  * On ext2, we can store the error state of the filesystem in the
 437  * superblock.  That is not possible on ext4, because we may have other
 438  * write ordering constraints on the superblock which prevent us from
 439  * writing it out straight away; and given that the journal is about to
 440  * be aborted, we can't rely on the current, or future, transactions to
 441  * write out the superblock safely.
 442  *
 443  * We'll just use the jbd2_journal_abort() error code to record an error in
 444  * the journal instead.  On recovery, the journal will complain about
 445  * that error until we've noted it down and cleared it.
 446  */
 447
 448 static void ext4_handle_error(struct super_block *sb)
 449 {
 450         if (test_opt(sb, WARN_ON_ERROR))
 451                 WARN_ON_ONCE(1);
 452
 453         if (sb_rdonly(sb))
 454                 return;
 455
 456         if (!test_opt(sb, ERRORS_CONT)) {
 457                 journal_t *journal = EXT4_SB(sb)->s_journal;
 458
 459                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 460                 if (journal)
 461                         jbd2_journal_abort(journal, -EIO);
 462         }
 463         if (test_opt(sb, ERRORS_RO)) {
 464                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 465                 /*
 466                  * Make sure updated value of ->s_mount_flags will be visible
 467                  * before ->s_flags update
 468                  */
 469                 smp_wmb();
 470                 sb->s_flags |= SB_RDONLY;
 471         }
 472         if (test_opt(sb, ERRORS_PANIC)) {
 473                 if (EXT4_SB(sb)->s_journal &&
 474                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 475                         return;
 476                 panic("EXT4-fs (device %s): panic forced after error\n",
 477                         sb->s_id);
 478         }
 479 }
 480
 481 #define ext4_error_ratelimit(sb)                                        \
 482                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 483                              "EXT4-fs error")
 484
 485 void __ext4_error(struct super_block *sb, const char *function,
 486                   unsigned int line, const char *fmt, ...)
 487 {
 488         struct va_format vaf;
 489         va_list args;
 490
 491         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 492                 return;
 493
 494         trace_ext4_error(sb, function, line);
 495         if (ext4_error_ratelimit(sb)) {
 496                 va_start(args, fmt);
 497                 vaf.fmt = fmt;
 498                 vaf.va = &args;
 499                 printk(KERN_CRIT
 500                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 501                        sb->s_id, function, line, current->comm, &vaf);
 502                 va_end(args);
 503         }
 504         save_error_info(sb, function, line);
 505         ext4_handle_error(sb);
 506 }
 507
 508 void __ext4_error_inode(struct inode *inode, const char *function,
 509                         unsigned int line, ext4_fsblk_t block,
 510                         const char *fmt, ...)
 511 {
 512         va_list args;
 513         struct va_format vaf;
 514         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 515
 516         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 517                 return;
 518
 519         trace_ext4_error(inode->i_sb, function, line);
 520         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 521         es->s_last_error_block = cpu_to_le64(block);
 522         if (ext4_error_ratelimit(inode->i_sb)) {
 523                 va_start(args, fmt);
 524                 vaf.fmt = fmt;
 525                 vaf.va = &args;
 526                 if (block)
 527                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 528                                "inode #%lu: block %llu: comm %s: %pV\n",
 529                                inode->i_sb->s_id, function, line, inode->i_ino,
 530                                block, current->comm, &vaf);
 531                 else
 532                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 533                                "inode #%lu: comm %s: %pV\n",
 534                                inode->i_sb->s_id, function, line, inode->i_ino,
 535                                current->comm, &vaf);
 536                 va_end(args);
 537         }
 538         save_error_info(inode->i_sb, function, line);
 539         ext4_handle_error(inode->i_sb);
 540 }
 541
 542 void __ext4_error_file(struct file *file, const char *function,
 543                        unsigned int line, ext4_fsblk_t block,
 544                        const char *fmt, ...)
 545 {
 546         va_list args;
 547         struct va_format vaf;
 548         struct ext4_super_block *es;
 549         struct inode *inode = file_inode(file);
 550         char pathname[80], *path;
 551
 552         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 553                 return;
 554
 555         trace_ext4_error(inode->i_sb, function, line);
 556         es = EXT4_SB(inode->i_sb)->s_es;
 557         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 558         if (ext4_error_ratelimit(inode->i_sb)) {
 559                 path = file_path(file, pathname, sizeof(pathname));
 560                 if (IS_ERR(path))
 561                         path = "(unknown)";
 562                 va_start(args, fmt);
 563                 vaf.fmt = fmt;
 564                 vaf.va = &args;
 565                 if (block)
 566                         printk(KERN_CRIT
 567                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 568                                "block %llu: comm %s: path %s: %pV\n",
 569                                inode->i_sb->s_id, function, line, inode->i_ino,
 570                                block, current->comm, path, &vaf);
 571                 else
 572                         printk(KERN_CRIT
 573                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 574                                "comm %s: path %s: %pV\n",
 575                                inode->i_sb->s_id, function, line, inode->i_ino,
 576                                current->comm, path, &vaf);
 577                 va_end(args);
 578         }
 579         save_error_info(inode->i_sb, function, line);
 580         ext4_handle_error(inode->i_sb);
 581 }
 582
 583 const char *ext4_decode_error(struct super_block *sb, int errno,
 584                               char nbuf[16])
 585 {
 586         char *errstr = NULL;
 587
 588         switch (errno) {
 589         case -EFSCORRUPTED:
 590                 errstr = "Corrupt filesystem";
 591                 break;
 592         case -EFSBADCRC:
 593                 errstr = "Filesystem failed CRC";
 594                 break;
 595         case -EIO:
 596                 errstr = "IO failure";
 597                 break;
 598         case -ENOMEM:
 599                 errstr = "Out of memory";
 600                 break;
 601         case -EROFS:
 602                 if (!sb || (EXT4_SB(sb)->s_journal &&
 603                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 604                         errstr = "Journal has aborted";
 605                 else
 606                         errstr = "Readonly filesystem";
 607                 break;
 608         default:
 609                 /* If the caller passed in an extra buffer for unknown
 610                  * errors, textualise them now.  Else we just return
 611                  * NULL. */
 612                 if (nbuf) {
 613                         /* Check for truncated error codes... */
 614                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 615                                 errstr = nbuf;
 616                 }
 617                 break;
 618         }
 619
 620         return errstr;
 621 }
 622
 623 /* __ext4_std_error decodes expected errors from journaling functions
 624  * automatically and invokes the appropriate error response.  */
 625
 626 void __ext4_std_error(struct super_block *sb, const char *function,
 627                       unsigned int line, int errno)
 628 {
 629         char nbuf[16];
 630         const char *errstr;
 631
 632         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 633                 return;
 634
 635         /* Special case: if the error is EROFS, and we're not already
 636          * inside a transaction, then there's really no point in logging
 637          * an error. */
 638         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 639                 return;
 640
 641         if (ext4_error_ratelimit(sb)) {
 642                 errstr = ext4_decode_error(sb, errno, nbuf);
 643                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 644                        sb->s_id, function, line, errstr);
 645         }
 646
 647         save_error_info(sb, function, line);
 648         ext4_handle_error(sb);
 649 }
 650
 651 /*
 652  * ext4_abort is a much stronger failure handler than ext4_error.  The
 653  * abort function may be used to deal with unrecoverable failures such
 654  * as journal IO errors or ENOMEM at a critical moment in log management.
 655  *
 656  * We unconditionally force the filesystem into an ABORT|READONLY state,
 657  * unless the error response on the fs has been set to panic in which
 658  * case we take the easy way out and panic immediately.
 659  */
 660
 661 void __ext4_abort(struct super_block *sb, const char *function,
 662                 unsigned int line, const char *fmt, ...)
 663 {
 664         struct va_format vaf;
 665         va_list args;
 666
 667         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 668                 return;
 669
 670         save_error_info(sb, function, line);
 671         va_start(args, fmt);
 672         vaf.fmt = fmt;
 673         vaf.va = &args;
 674         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 675                sb->s_id, function, line, &vaf);
 676         va_end(args);
 677
 678         if (sb_rdonly(sb) == 0) {
 679                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 680                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 681                 /*
 682                  * Make sure updated value of ->s_mount_flags will be visible
 683                  * before ->s_flags update
 684                  */
 685                 smp_wmb();
 686                 sb->s_flags |= SB_RDONLY;
 687                 if (EXT4_SB(sb)->s_journal)
 688                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 689                 save_error_info(sb, function, line);
 690         }
 691         if (test_opt(sb, ERRORS_PANIC)) {
 692                 if (EXT4_SB(sb)->s_journal &&
 693                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 694                         return;
 695                 panic("EXT4-fs panic from previous error\n");
 696         }
 697 }
 698
 699 void __ext4_msg(struct super_block *sb,
 700                 const char *prefix, const char *fmt, ...)
 701 {
 702         struct va_format vaf;
 703         va_list args;
 704
 705         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 706                 return;
 707
 708         va_start(args, fmt);
 709         vaf.fmt = fmt;
 710         vaf.va = &args;
 711         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 712         va_end(args);
 713 }
 714
 715 #define ext4_warning_ratelimit(sb)                                      \
 716                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 717                              "EXT4-fs warning")
 718
 719 void __ext4_warning(struct super_block *sb, const char *function,
 720                     unsigned int line, const char *fmt, ...)
 721 {
 722         struct va_format vaf;
 723         va_list args;
 724
 725         if (!ext4_warning_ratelimit(sb))
 726                 return;
 727
 728         va_start(args, fmt);
 729         vaf.fmt = fmt;
 730         vaf.va = &args;
 731         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 732                sb->s_id, function, line, &vaf);
 733         va_end(args);
 734 }
 735
 736 void __ext4_warning_inode(const struct inode *inode, const char *function,
 737                           unsigned int line, const char *fmt, ...)
 738 {
 739         struct va_format vaf;
 740         va_list args;
 741
 742         if (!ext4_warning_ratelimit(inode->i_sb))
 743                 return;
 744
 745         va_start(args, fmt);
 746         vaf.fmt = fmt;
 747         vaf.va = &args;
 748         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 749                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 750                function, line, inode->i_ino, current->comm, &vaf);
 751         va_end(args);
 752 }
 753
 754 void __ext4_grp_locked_error(const char *function, unsigned int line,
 755                              struct super_block *sb, ext4_group_t grp,
 756                              unsigned long ino, ext4_fsblk_t block,
 757                              const char *fmt, ...)
 758 __releases(bitlock)
 759 __acquires(bitlock)
 760 {
 761         struct va_format vaf;
 762         va_list args;
 763         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 764
 765         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 766                 return;
 767
 768         trace_ext4_error(sb, function, line);
 769         es->s_last_error_ino = cpu_to_le32(ino);
 770         es->s_last_error_block = cpu_to_le64(block);
 771         __save_error_info(sb, function, line);
 772
 773         if (ext4_error_ratelimit(sb)) {
 774                 va_start(args, fmt);
 775                 vaf.fmt = fmt;
 776                 vaf.va = &args;
 777                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 778                        sb->s_id, function, line, grp);
 779                 if (ino)
 780                         printk(KERN_CONT "inode %lu: ", ino);
 781                 if (block)
 782                         printk(KERN_CONT "block %llu:",
 783                                (unsigned long long) block);
 784                 printk(KERN_CONT "%pV\n", &vaf);
 785                 va_end(args);
 786         }
 787
 788         if (test_opt(sb, WARN_ON_ERROR))
 789                 WARN_ON_ONCE(1);
 790
 791         if (test_opt(sb, ERRORS_CONT)) {
 792                 ext4_commit_super(sb, 0);
 793                 return;
 794         }
 795
 796         ext4_unlock_group(sb, grp);
 797         ext4_commit_super(sb, 1);
 798         ext4_handle_error(sb);
 799         /*
 800          * We only get here in the ERRORS_RO case; relocking the group
 801          * may be dangerous, but nothing bad will happen since the
 802          * filesystem will have already been marked read/only and the
 803          * journal has been aborted.  We return 1 as a hint to callers
 804          * who might what to use the return value from
 805          * ext4_grp_locked_error() to distinguish between the
 806          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 807          * aggressively from the ext4 function in question, with a
 808          * more appropriate error code.
 809          */
 810         ext4_lock_group(sb, grp);
 811         return;
 812 }
 813
 814 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 815                                      ext4_group_t group,
 816                                      unsigned int flags)
 817 {
 818         struct ext4_sb_info *sbi = EXT4_SB(sb);
 819         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 820         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 821         int ret;
 822
 823         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
 824                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 825                                             &grp->bb_state);
 826                 if (!ret)
 827                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 828                                            grp->bb_free);
 829         }
 830
 831         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
 832                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 833                                             &grp->bb_state);
 834                 if (!ret && gdp) {
 835                         int count;
 836
 837                         count = ext4_free_inodes_count(sb, gdp);
 838                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 839                                            count);
 840                 }
 841         }
 842 }
 843
 844 void ext4_update_dynamic_rev(struct super_block *sb)
 845 {
 846         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 847
 848         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 849                 return;
 850
 851         ext4_warning(sb,
 852                      "updating to rev %d because of new feature flag, "
 853                      "running e2fsck is recommended",
 854                      EXT4_DYNAMIC_REV);
 855
 856         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 857         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 858         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 859         /* leave es->s_feature_*compat flags alone */
 860         /* es->s_uuid will be set by e2fsck if empty */
 861
 862         /*
 863          * The rest of the superblock fields should be zero, and if not it
 864          * means they are likely already in use, so leave them alone.  We
 865          * can leave it up to e2fsck to clean up any inconsistencies there.
 866          */
 867 }
 868
 869 /*
 870  * Open the external journal device
 871  */
 872 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 873 {
 874         struct block_device *bdev;
 875         char b[BDEVNAME_SIZE];
 876
 877         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 878         if (IS_ERR(bdev))
 879                 goto fail;
 880         return bdev;
 881
 882 fail:
 883         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 884                         __bdevname(dev, b), PTR_ERR(bdev));
 885         return NULL;
 886 }
 887
 888 /*
 889  * Release the journal device
 890  */
 891 static void ext4_blkdev_put(struct block_device *bdev)
 892 {
 893         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 894 }
 895
 896 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 897 {
 898         struct block_device *bdev;
 899         bdev = sbi->journal_bdev;
 900         if (bdev) {
 901                 ext4_blkdev_put(bdev);
 902                 sbi->journal_bdev = NULL;
 903         }
 904 }
 905
 906 static inline struct inode *orphan_list_entry(struct list_head *l)
 907 {
 908         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 909 }
 910
 911 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 912 {
 913         struct list_head *l;
 914
 915         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 916                  le32_to_cpu(sbi->s_es->s_last_orphan));
 917
 918         printk(KERN_ERR "sb_info orphan list:\n");
 919         list_for_each(l, &sbi->s_orphan) {
 920                 struct inode *inode = orphan_list_entry(l);
 921                 printk(KERN_ERR "  "
 922                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 923                        inode->i_sb->s_id, inode->i_ino, inode,
 924                        inode->i_mode, inode->i_nlink,
 925                        NEXT_ORPHAN(inode));
 926         }
 927 }
 928
 929 #ifdef CONFIG_QUOTA
 930 static int ext4_quota_off(struct super_block *sb, int type);
 931
 932 static inline void ext4_quota_off_umount(struct super_block *sb)
 933 {
 934         int type;
 935
 936         /* Use our quota_off function to clear inode flags etc. */
 937         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 938                 ext4_quota_off(sb, type);
 939 }
 940
 941 /*
 942  * This is a helper function which is used in the mount/remount
 943  * codepaths (which holds s_umount) to fetch the quota file name.
 944  */
 945 static inline char *get_qf_name(struct super_block *sb,
 946                                 struct ext4_sb_info *sbi,
 947                                 int type)
 948 {
 949         return rcu_dereference_protected(sbi->s_qf_names[type],
 950                                          lockdep_is_held(&sb->s_umount));
 951 }
 952 #else
 953 static inline void ext4_quota_off_umount(struct super_block *sb)
 954 {
 955 }
 956 #endif
 957
 958 static void ext4_put_super(struct super_block *sb)
 959 {
 960         struct ext4_sb_info *sbi = EXT4_SB(sb);
 961         struct ext4_super_block *es = sbi->s_es;
 962         int aborted = 0;
 963         int i, err;
 964
 965         ext4_unregister_li_request(sb);
 966         ext4_quota_off_umount(sb);
 967
 968         destroy_workqueue(sbi->rsv_conversion_wq);
 969
 970         if (sbi->s_journal) {
 971                 aborted = is_journal_aborted(sbi->s_journal);
 972                 err = jbd2_journal_destroy(sbi->s_journal);
 973                 sbi->s_journal = NULL;
 974                 if ((err < 0) && !aborted)
 975                         ext4_abort(sb, "Couldn't clean up the journal");
 976         }
 977
 978         ext4_unregister_sysfs(sb);
 979         ext4_es_unregister_shrinker(sbi);
 980         del_timer_sync(&sbi->s_err_report);
 981         ext4_release_system_zone(sb);
 982         ext4_mb_release(sb);
 983         ext4_ext_release(sb);
 984
 985         if (!sb_rdonly(sb) && !aborted) {
 986                 ext4_clear_feature_journal_needs_recovery(sb);
 987                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 988         }
 989         if (!sb_rdonly(sb))
 990                 ext4_commit_super(sb, 1);
 991
 992         for (i = 0; i < sbi->s_gdb_count; i++)
 993                 brelse(sbi->s_group_desc[i]);
 994         kvfree(sbi->s_group_desc);
 995         kvfree(sbi->s_flex_groups);
 996         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 997         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 998         percpu_counter_destroy(&sbi->s_dirs_counter);
 999         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1000         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
1001 #ifdef CONFIG_QUOTA
1002         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1003                 kfree(get_qf_name(sb, sbi, i));
1004 #endif
1005
1006         /* Debugging code just in case the in-memory inode orphan list
1007          * isn't empty.  The on-disk one can be non-empty if we've
1008          * detected an error and taken the fs readonly, but the
1009          * in-memory list had better be clean by this point. */
1010         if (!list_empty(&sbi->s_orphan))
1011                 dump_orphan_list(sb, sbi);
1012         J_ASSERT(list_empty(&sbi->s_orphan));
1013
1014         sync_blockdev(sb->s_bdev);
1015         invalidate_bdev(sb->s_bdev);
1016         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
1017                 /*
1018                  * Invalidate the journal device's buffers.  We don't want them
1019                  * floating about in memory - the physical journal device may
1020                  * hotswapped, and it breaks the `ro-after' testing code.
1021                  */
1022                 sync_blockdev(sbi->journal_bdev);
1023                 invalidate_bdev(sbi->journal_bdev);
1024                 ext4_blkdev_remove(sbi);
1025         }
1026
1027         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1028         sbi->s_ea_inode_cache = NULL;
1029
1030         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1031         sbi->s_ea_block_cache = NULL;
1032
1033         if (sbi->s_mmp_tsk)
1034                 kthread_stop(sbi->s_mmp_tsk);
1035         brelse(sbi->s_sbh);
1036         sb->s_fs_info = NULL;
1037         /*
1038          * Now that we are completely done shutting down the
1039          * superblock, we need to actually destroy the kobject.
1040          */
1041         kobject_put(&sbi->s_kobj);
1042         wait_for_completion(&sbi->s_kobj_unregister);
1043         if (sbi->s_chksum_driver)
1044                 crypto_free_shash(sbi->s_chksum_driver);
1045         kfree(sbi->s_blockgroup_lock);
1046         fs_put_dax(sbi->s_daxdev);
1047         kfree(sbi);
1048 }
1049
1050 static struct kmem_cache *ext4_inode_cachep;
1051
1052 /*
1053  * Called inside transaction, so use GFP_NOFS
1054  */
1055 static struct inode *ext4_alloc_inode(struct super_block *sb)
1056 {
1057         struct ext4_inode_info *ei;
1058
1059         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1060         if (!ei)
1061                 return NULL;
1062
1063         inode_set_iversion(&ei->vfs_inode, 1);
1064         spin_lock_init(&ei->i_raw_lock);
1065         INIT_LIST_HEAD(&ei->i_prealloc_list);
1066         spin_lock_init(&ei->i_prealloc_lock);
1067         ext4_es_init_tree(&ei->i_es_tree);
1068         rwlock_init(&ei->i_es_lock);
1069         INIT_LIST_HEAD(&ei->i_es_list);
1070         ei->i_es_all_nr = 0;
1071         ei->i_es_shk_nr = 0;
1072         ei->i_es_shrink_lblk = 0;
1073         ei->i_reserved_data_blocks = 0;
1074         ei->i_da_metadata_calc_len = 0;
1075         ei->i_da_metadata_calc_last_lblock = 0;
1076         spin_lock_init(&(ei->i_block_reservation_lock));
1077         ext4_init_pending_tree(&ei->i_pending_tree);
1078 #ifdef CONFIG_QUOTA
1079         ei->i_reserved_quota = 0;
1080         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1081 #endif
1082         ei->jinode = NULL;
1083         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1084         spin_lock_init(&ei->i_completed_io_lock);
1085         ei->i_sync_tid = 0;
1086         ei->i_datasync_tid = 0;
1087         atomic_set(&ei->i_unwritten, 0);
1088         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1089         return &ei->vfs_inode;
1090 }
1091
1092 static int ext4_drop_inode(struct inode *inode)
1093 {
1094         int drop = generic_drop_inode(inode);
1095
1096         trace_ext4_drop_inode(inode, drop);
1097         return drop;
1098 }
1099
1100 static void ext4_i_callback(struct rcu_head *head)
1101 {
1102         struct inode *inode = container_of(head, struct inode, i_rcu);
1103         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1104 }
1105
1106 static void ext4_destroy_inode(struct inode *inode)
1107 {
1108         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1109                 ext4_msg(inode->i_sb, KERN_ERR,
1110                          "Inode %lu (%p): orphan list check failed!",
1111                          inode->i_ino, EXT4_I(inode));
1112                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1113                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1114                                 true);
1115                 dump_stack();
1116         }
1117         call_rcu(&inode->i_rcu, ext4_i_callback);
1118 }
1119
1120 static void init_once(void *foo)
1121 {
1122         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1123
1124         INIT_LIST_HEAD(&ei->i_orphan);
1125         init_rwsem(&ei->xattr_sem);
1126         init_rwsem(&ei->i_data_sem);
1127         init_rwsem(&ei->i_mmap_sem);
1128         inode_init_once(&ei->vfs_inode);
1129 }
1130
1131 static int __init init_inodecache(void)
1132 {
1133         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1134                                 sizeof(struct ext4_inode_info), 0,
1135                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1136                                         SLAB_ACCOUNT),
1137                                 offsetof(struct ext4_inode_info, i_data),
1138                                 sizeof_field(struct ext4_inode_info, i_data),
1139                                 init_once);
1140         if (ext4_inode_cachep == NULL)
1141                 return -ENOMEM;
1142         return 0;
1143 }
1144
1145 static void destroy_inodecache(void)
1146 {
1147         /*
1148          * Make sure all delayed rcu free inodes are flushed before we
1149          * destroy cache.
1150          */
1151         rcu_barrier();
1152         kmem_cache_destroy(ext4_inode_cachep);
1153 }
1154
1155 void ext4_clear_inode(struct inode *inode)
1156 {
1157         invalidate_inode_buffers(inode);
1158         clear_inode(inode);
1159         dquot_drop(inode);
1160         ext4_discard_preallocations(inode);
1161         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1162         if (EXT4_I(inode)->jinode) {
1163                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1164                                                EXT4_I(inode)->jinode);
1165                 jbd2_free_inode(EXT4_I(inode)->jinode);
1166                 EXT4_I(inode)->jinode = NULL;
1167         }
1168         fscrypt_put_encryption_info(inode);
1169 }
1170
1171 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1172                                         u64 ino, u32 generation)
1173 {
1174         struct inode *inode;
1175
1176         /*
1177          * Currently we don't know the generation for parent directory, so
1178          * a generation of 0 means "accept any"
1179          */
1180         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1181         if (IS_ERR(inode))
1182                 return ERR_CAST(inode);
1183         if (generation && inode->i_generation != generation) {
1184                 iput(inode);
1185                 return ERR_PTR(-ESTALE);
1186         }
1187
1188         return inode;
1189 }
1190
1191 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1192                                         int fh_len, int fh_type)
1193 {
1194         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1195                                     ext4_nfs_get_inode);
1196 }
1197
1198 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1199                                         int fh_len, int fh_type)
1200 {
1201         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1202                                     ext4_nfs_get_inode);
1203 }
1204
1205 static int ext4_nfs_commit_metadata(struct inode *inode)
1206 {
1207         struct writeback_control wbc = {
1208                 .sync_mode = WB_SYNC_ALL
1209         };
1210
1211         trace_ext4_nfs_commit_metadata(inode);
1212         return ext4_write_inode(inode, &wbc);
1213 }
1214
1215 /*
1216  * Try to release metadata pages (indirect blocks, directories) which are
1217  * mapped via the block device.  Since these pages could have journal heads
1218  * which would prevent try_to_free_buffers() from freeing them, we must use
1219  * jbd2 layer's try_to_free_buffers() function to release them.
1220  */
1221 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1222                                  gfp_t wait)
1223 {
1224         journal_t *journal = EXT4_SB(sb)->s_journal;
1225
1226         WARN_ON(PageChecked(page));
1227         if (!page_has_buffers(page))
1228                 return 0;
1229         if (journal)
1230                 return jbd2_journal_try_to_free_buffers(journal, page,
1231                                                 wait & ~__GFP_DIRECT_RECLAIM);
1232         return try_to_free_buffers(page);
1233 }
1234
1235 #ifdef CONFIG_FS_ENCRYPTION
1236 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1237 {
1238         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1239                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1240 }
1241
1242 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1243                                                         void *fs_data)
1244 {
1245         handle_t *handle = fs_data;
1246         int res, res2, credits, retries = 0;
1247
1248         /*
1249          * Encrypting the root directory is not allowed because e2fsck expects
1250          * lost+found to exist and be unencrypted, and encrypting the root
1251          * directory would imply encrypting the lost+found directory as well as
1252          * the filename "lost+found" itself.
1253          */
1254         if (inode->i_ino == EXT4_ROOT_INO)
1255                 return -EPERM;
1256
1257         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1258                 return -EINVAL;
1259
1260         res = ext4_convert_inline_data(inode);
1261         if (res)
1262                 return res;
1263
1264         /*
1265          * If a journal handle was specified, then the encryption context is
1266          * being set on a new inode via inheritance and is part of a larger
1267          * transaction to create the inode.  Otherwise the encryption context is
1268          * being set on an existing inode in its own transaction.  Only in the
1269          * latter case should the "retry on ENOSPC" logic be used.
1270          */
1271
1272         if (handle) {
1273                 res = ext4_xattr_set_handle(handle, inode,
1274                                             EXT4_XATTR_INDEX_ENCRYPTION,
1275                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1276                                             ctx, len, 0);
1277                 if (!res) {
1278                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1279                         ext4_clear_inode_state(inode,
1280                                         EXT4_STATE_MAY_INLINE_DATA);
1281                         /*
1282                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1283                          * S_DAX may be disabled
1284                          */
1285                         ext4_set_inode_flags(inode);
1286                 }
1287                 return res;
1288         }
1289
1290         res = dquot_initialize(inode);
1291         if (res)
1292                 return res;
1293 retry:
1294         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1295                                      &credits);
1296         if (res)
1297                 return res;
1298
1299         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1300         if (IS_ERR(handle))
1301                 return PTR_ERR(handle);
1302
1303         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1304                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1305                                     ctx, len, 0);
1306         if (!res) {
1307                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1308                 /*
1309                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1310                  * S_DAX may be disabled
1311                  */
1312                 ext4_set_inode_flags(inode);
1313                 res = ext4_mark_inode_dirty(handle, inode);
1314                 if (res)
1315                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1316         }
1317         res2 = ext4_journal_stop(handle);
1318
1319         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1320                 goto retry;
1321         if (!res)
1322                 res = res2;
1323         return res;
1324 }
1325
1326 static bool ext4_dummy_context(struct inode *inode)
1327 {
1328         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1329 }
1330
1331 static const struct fscrypt_operations ext4_cryptops = {
1332         .key_prefix             = "ext4:",
1333         .get_context            = ext4_get_context,
1334         .set_context            = ext4_set_context,
1335         .dummy_context          = ext4_dummy_context,
1336         .empty_dir              = ext4_empty_dir,
1337         .max_namelen            = EXT4_NAME_LEN,
1338 };
1339 #endif
1340
1341 #ifdef CONFIG_QUOTA
1342 static const char * const quotatypes[] = INITQFNAMES;
1343 #define QTYPE2NAME(t) (quotatypes[t])
1344
1345 static int ext4_write_dquot(struct dquot *dquot);
1346 static int ext4_acquire_dquot(struct dquot *dquot);
1347 static int ext4_release_dquot(struct dquot *dquot);
1348 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1349 static int ext4_write_info(struct super_block *sb, int type);
1350 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1351                          const struct path *path);
1352 static int ext4_quota_on_mount(struct super_block *sb, int type);
1353 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1354                                size_t len, loff_t off);
1355 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1356                                 const char *data, size_t len, loff_t off);
1357 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1358                              unsigned int flags);
1359 static int ext4_enable_quotas(struct super_block *sb);
1360 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1361
1362 static struct dquot **ext4_get_dquots(struct inode *inode)
1363 {
1364         return EXT4_I(inode)->i_dquot;
1365 }
1366
1367 static const struct dquot_operations ext4_quota_operations = {
1368         .get_reserved_space     = ext4_get_reserved_space,
1369         .write_dquot            = ext4_write_dquot,
1370         .acquire_dquot          = ext4_acquire_dquot,
1371         .release_dquot          = ext4_release_dquot,
1372         .mark_dirty             = ext4_mark_dquot_dirty,
1373         .write_info             = ext4_write_info,
1374         .alloc_dquot            = dquot_alloc,
1375         .destroy_dquot          = dquot_destroy,
1376         .get_projid             = ext4_get_projid,
1377         .get_inode_usage        = ext4_get_inode_usage,
1378         .get_next_id            = ext4_get_next_id,
1379 };
1380
1381 static const struct quotactl_ops ext4_qctl_operations = {
1382         .quota_on       = ext4_quota_on,
1383         .quota_off      = ext4_quota_off,
1384         .quota_sync     = dquot_quota_sync,
1385         .get_state      = dquot_get_state,
1386         .set_info       = dquot_set_dqinfo,
1387         .get_dqblk      = dquot_get_dqblk,
1388         .set_dqblk      = dquot_set_dqblk,
1389         .get_nextdqblk  = dquot_get_next_dqblk,
1390 };
1391 #endif
1392
1393 static const struct super_operations ext4_sops = {
1394         .alloc_inode    = ext4_alloc_inode,
1395         .destroy_inode  = ext4_destroy_inode,
1396         .write_inode    = ext4_write_inode,
1397         .dirty_inode    = ext4_dirty_inode,
1398         .drop_inode     = ext4_drop_inode,
1399         .evict_inode    = ext4_evict_inode,
1400         .put_super      = ext4_put_super,
1401         .sync_fs        = ext4_sync_fs,
1402         .freeze_fs      = ext4_freeze,
1403         .unfreeze_fs    = ext4_unfreeze,
1404         .statfs         = ext4_statfs,
1405         .remount_fs     = ext4_remount,
1406         .show_options   = ext4_show_options,
1407 #ifdef CONFIG_QUOTA
1408         .quota_read     = ext4_quota_read,
1409         .quota_write    = ext4_quota_write,
1410         .get_dquots     = ext4_get_dquots,
1411 #endif
1412         .bdev_try_to_free_page = bdev_try_to_free_page,
1413 };
1414
1415 static const struct export_operations ext4_export_ops = {
1416         .fh_to_dentry = ext4_fh_to_dentry,
1417         .fh_to_parent = ext4_fh_to_parent,
1418         .get_parent = ext4_get_parent,
1419         .commit_metadata = ext4_nfs_commit_metadata,
1420 };
1421
1422 enum {
1423         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1424         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1425         Opt_nouid32, Opt_debug, Opt_removed,
1426         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1427         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1428         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1429         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1430         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1431         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1432         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1433         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1434         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1435         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1436         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1437         Opt_nowarn_on_error, Opt_mblk_io_submit,
1438         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1439         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1440         Opt_inode_readahead_blks, Opt_journal_ioprio,
1441         Opt_dioread_nolock, Opt_dioread_lock,
1442         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1443         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1444 };
1445
1446 static const match_table_t tokens = {
1447         {Opt_bsd_df, "bsddf"},
1448         {Opt_minix_df, "minixdf"},
1449         {Opt_grpid, "grpid"},
1450         {Opt_grpid, "bsdgroups"},
1451         {Opt_nogrpid, "nogrpid"},
1452         {Opt_nogrpid, "sysvgroups"},
1453         {Opt_resgid, "resgid=%u"},
1454         {Opt_resuid, "resuid=%u"},
1455         {Opt_sb, "sb=%u"},
1456         {Opt_err_cont, "errors=continue"},
1457         {Opt_err_panic, "errors=panic"},
1458         {Opt_err_ro, "errors=remount-ro"},
1459         {Opt_nouid32, "nouid32"},
1460         {Opt_debug, "debug"},
1461         {Opt_removed, "oldalloc"},
1462         {Opt_removed, "orlov"},
1463         {Opt_user_xattr, "user_xattr"},
1464         {Opt_nouser_xattr, "nouser_xattr"},
1465         {Opt_acl, "acl"},
1466         {Opt_noacl, "noacl"},
1467         {Opt_noload, "norecovery"},
1468         {Opt_noload, "noload"},
1469         {Opt_removed, "nobh"},
1470         {Opt_removed, "bh"},
1471         {Opt_commit, "commit=%u"},
1472         {Opt_min_batch_time, "min_batch_time=%u"},
1473         {Opt_max_batch_time, "max_batch_time=%u"},
1474         {Opt_journal_dev, "journal_dev=%u"},
1475         {Opt_journal_path, "journal_path=%s"},
1476         {Opt_journal_checksum, "journal_checksum"},
1477         {Opt_nojournal_checksum, "nojournal_checksum"},
1478         {Opt_journal_async_commit, "journal_async_commit"},
1479         {Opt_abort, "abort"},
1480         {Opt_data_journal, "data=journal"},
1481         {Opt_data_ordered, "data=ordered"},
1482         {Opt_data_writeback, "data=writeback"},
1483         {Opt_data_err_abort, "data_err=abort"},
1484         {Opt_data_err_ignore, "data_err=ignore"},
1485         {Opt_offusrjquota, "usrjquota="},
1486         {Opt_usrjquota, "usrjquota=%s"},
1487         {Opt_offgrpjquota, "grpjquota="},
1488         {Opt_grpjquota, "grpjquota=%s"},
1489         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1490         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1491         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1492         {Opt_grpquota, "grpquota"},
1493         {Opt_noquota, "noquota"},
1494         {Opt_quota, "quota"},
1495         {Opt_usrquota, "usrquota"},
1496         {Opt_prjquota, "prjquota"},
1497         {Opt_barrier, "barrier=%u"},
1498         {Opt_barrier, "barrier"},
1499         {Opt_nobarrier, "nobarrier"},
1500         {Opt_i_version, "i_version"},
1501         {Opt_dax, "dax"},
1502         {Opt_stripe, "stripe=%u"},
1503         {Opt_delalloc, "delalloc"},
1504         {Opt_warn_on_error, "warn_on_error"},
1505         {Opt_nowarn_on_error, "nowarn_on_error"},
1506         {Opt_lazytime, "lazytime"},
1507         {Opt_nolazytime, "nolazytime"},
1508         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1509         {Opt_nodelalloc, "nodelalloc"},
1510         {Opt_removed, "mblk_io_submit"},
1511         {Opt_removed, "nomblk_io_submit"},
1512         {Opt_block_validity, "block_validity"},
1513         {Opt_noblock_validity, "noblock_validity"},
1514         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1515         {Opt_journal_ioprio, "journal_ioprio=%u"},
1516         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1517         {Opt_auto_da_alloc, "auto_da_alloc"},
1518         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1519         {Opt_dioread_nolock, "dioread_nolock"},
1520         {Opt_dioread_lock, "dioread_lock"},
1521         {Opt_discard, "discard"},
1522         {Opt_nodiscard, "nodiscard"},
1523         {Opt_init_itable, "init_itable=%u"},
1524         {Opt_init_itable, "init_itable"},
1525         {Opt_noinit_itable, "noinit_itable"},
1526         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1527         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1528         {Opt_nombcache, "nombcache"},
1529         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1530         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1531         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1532         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1533         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1534         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1535         {Opt_err, NULL},
1536 };
1537
1538 static ext4_fsblk_t get_sb_block(void **data)
1539 {
1540         ext4_fsblk_t    sb_block;
1541         char            *options = (char *) *data;
1542
1543         if (!options || strncmp(options, "sb=", 3) != 0)
1544                 return 1;       /* Default location */
1545
1546         options += 3;
1547         /* TODO: use simple_strtoll with >32bit ext4 */
1548         sb_block = simple_strtoul(options, &options, 0);
1549         if (*options && *options != ',') {
1550                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1551                        (char *) *data);
1552                 return 1;
1553         }
1554         if (*options == ',')
1555                 options++;
1556         *data = (void *) options;
1557
1558         return sb_block;
1559 }
1560
1561 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1562 static const char deprecated_msg[] =
1563         "Mount option \"%s\" will be removed by %s\n"
1564         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1565
1566 #ifdef CONFIG_QUOTA
1567 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1568 {
1569         struct ext4_sb_info *sbi = EXT4_SB(sb);
1570         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1571         int ret = -1;
1572
1573         if (sb_any_quota_loaded(sb) && !old_qname) {
1574                 ext4_msg(sb, KERN_ERR,
1575                         "Cannot change journaled "
1576                         "quota options when quota turned on");
1577                 return -1;
1578         }
1579         if (ext4_has_feature_quota(sb)) {
1580                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1581                          "ignored when QUOTA feature is enabled");
1582                 return 1;
1583         }
1584         qname = match_strdup(args);
1585         if (!qname) {
1586                 ext4_msg(sb, KERN_ERR,
1587                         "Not enough memory for storing quotafile name");
1588                 return -1;
1589         }
1590         if (old_qname) {
1591                 if (strcmp(old_qname, qname) == 0)
1592                         ret = 1;
1593                 else
1594                         ext4_msg(sb, KERN_ERR,
1595                                  "%s quota file already specified",
1596                                  QTYPE2NAME(qtype));
1597                 goto errout;
1598         }
1599         if (strchr(qname, '/')) {
1600                 ext4_msg(sb, KERN_ERR,
1601                         "quotafile must be on filesystem root");
1602                 goto errout;
1603         }
1604         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1605         set_opt(sb, QUOTA);
1606         return 1;
1607 errout:
1608         kfree(qname);
1609         return ret;
1610 }
1611
1612 static int clear_qf_name(struct super_block *sb, int qtype)
1613 {
1614
1615         struct ext4_sb_info *sbi = EXT4_SB(sb);
1616         char *old_qname = get_qf_name(sb, sbi, qtype);
1617
1618         if (sb_any_quota_loaded(sb) && old_qname) {
1619                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1620                         " when quota turned on");
1621                 return -1;
1622         }
1623         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1624         synchronize_rcu();
1625         kfree(old_qname);
1626         return 1;
1627 }
1628 #endif
1629
1630 #define MOPT_SET        0x0001
1631 #define MOPT_CLEAR      0x0002
1632 #define MOPT_NOSUPPORT  0x0004
1633 #define MOPT_EXPLICIT   0x0008
1634 #define MOPT_CLEAR_ERR  0x0010
1635 #define MOPT_GTE0       0x0020
1636 #ifdef CONFIG_QUOTA
1637 #define MOPT_Q          0
1638 #define MOPT_QFMT       0x0040
1639 #else
1640 #define MOPT_Q          MOPT_NOSUPPORT
1641 #define MOPT_QFMT       MOPT_NOSUPPORT
1642 #endif
1643 #define MOPT_DATAJ      0x0080
1644 #define MOPT_NO_EXT2    0x0100
1645 #define MOPT_NO_EXT3    0x0200
1646 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1647 #define MOPT_STRING     0x0400
1648
1649 static const struct mount_opts {
1650         int     token;
1651         int     mount_opt;
1652         int     flags;
1653 } ext4_mount_opts[] = {
1654         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1655         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1656         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1657         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1658         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1659         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1660         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1661          MOPT_EXT4_ONLY | MOPT_SET},
1662         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1663          MOPT_EXT4_ONLY | MOPT_CLEAR},
1664         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1665         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1666         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1667          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1668         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1669          MOPT_EXT4_ONLY | MOPT_CLEAR},
1670         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1671         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1672         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1673          MOPT_EXT4_ONLY | MOPT_CLEAR},
1674         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1675          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1676         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1677                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1678          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1679         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1680         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1681         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1682         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1683         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1684          MOPT_NO_EXT2},
1685         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1686          MOPT_NO_EXT2},
1687         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1688         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1689         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1690         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1691         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1692         {Opt_commit, 0, MOPT_GTE0},
1693         {Opt_max_batch_time, 0, MOPT_GTE0},
1694         {Opt_min_batch_time, 0, MOPT_GTE0},
1695         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1696         {Opt_init_itable, 0, MOPT_GTE0},
1697         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1698         {Opt_stripe, 0, MOPT_GTE0},
1699         {Opt_resuid, 0, MOPT_GTE0},
1700         {Opt_resgid, 0, MOPT_GTE0},
1701         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1702         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1703         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1704         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1705         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1706         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1707          MOPT_NO_EXT2 | MOPT_DATAJ},
1708         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1709         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1710 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1711         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1712         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1713 #else
1714         {Opt_acl, 0, MOPT_NOSUPPORT},
1715         {Opt_noacl, 0, MOPT_NOSUPPORT},
1716 #endif
1717         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1718         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1719         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1720         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1721         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1722                                                         MOPT_SET | MOPT_Q},
1723         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1724                                                         MOPT_SET | MOPT_Q},
1725         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1726                                                         MOPT_SET | MOPT_Q},
1727         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1728                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1729                                                         MOPT_CLEAR | MOPT_Q},
1730         {Opt_usrjquota, 0, MOPT_Q},
1731         {Opt_grpjquota, 0, MOPT_Q},
1732         {Opt_offusrjquota, 0, MOPT_Q},
1733         {Opt_offgrpjquota, 0, MOPT_Q},
1734         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1735         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1736         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1737         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1738         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1739         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1740         {Opt_err, 0, 0}
1741 };
1742
1743 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1744                             substring_t *args, unsigned long *journal_devnum,
1745                             unsigned int *journal_ioprio, int is_remount)
1746 {
1747         struct ext4_sb_info *sbi = EXT4_SB(sb);
1748         const struct mount_opts *m;
1749         kuid_t uid;
1750         kgid_t gid;
1751         int arg = 0;
1752
1753 #ifdef CONFIG_QUOTA
1754         if (token == Opt_usrjquota)
1755                 return set_qf_name(sb, USRQUOTA, &args[0]);
1756         else if (token == Opt_grpjquota)
1757                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1758         else if (token == Opt_offusrjquota)
1759                 return clear_qf_name(sb, USRQUOTA);
1760         else if (token == Opt_offgrpjquota)
1761                 return clear_qf_name(sb, GRPQUOTA);
1762 #endif
1763         switch (token) {
1764         case Opt_noacl:
1765         case Opt_nouser_xattr:
1766                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1767                 break;
1768         case Opt_sb:
1769                 return 1;       /* handled by get_sb_block() */
1770         case Opt_removed:
1771                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1772                 return 1;
1773         case Opt_abort:
1774                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1775                 return 1;
1776         case Opt_i_version:
1777                 sb->s_flags |= SB_I_VERSION;
1778                 return 1;
1779         case Opt_lazytime:
1780                 sb->s_flags |= SB_LAZYTIME;
1781                 return 1;
1782         case Opt_nolazytime:
1783                 sb->s_flags &= ~SB_LAZYTIME;
1784                 return 1;
1785         }
1786
1787         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1788                 if (token == m->token)
1789                         break;
1790
1791         if (m->token == Opt_err) {
1792                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1793                          "or missing value", opt);
1794                 return -1;
1795         }
1796
1797         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1798                 ext4_msg(sb, KERN_ERR,
1799                          "Mount option \"%s\" incompatible with ext2", opt);
1800                 return -1;
1801         }
1802         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1803                 ext4_msg(sb, KERN_ERR,
1804                          "Mount option \"%s\" incompatible with ext3", opt);
1805                 return -1;
1806         }
1807
1808         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1809                 return -1;
1810         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1811                 return -1;
1812         if (m->flags & MOPT_EXPLICIT) {
1813                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1814                         set_opt2(sb, EXPLICIT_DELALLOC);
1815                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1816                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1817                 } else
1818                         return -1;
1819         }
1820         if (m->flags & MOPT_CLEAR_ERR)
1821                 clear_opt(sb, ERRORS_MASK);
1822         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1823                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1824                          "options when quota turned on");
1825                 return -1;
1826         }
1827
1828         if (m->flags & MOPT_NOSUPPORT) {
1829                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1830         } else if (token == Opt_commit) {
1831                 if (arg == 0)
1832                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1833                 sbi->s_commit_interval = HZ * arg;
1834         } else if (token == Opt_debug_want_extra_isize) {
1835                 sbi->s_want_extra_isize = arg;
1836         } else if (token == Opt_max_batch_time) {
1837                 sbi->s_max_batch_time = arg;
1838         } else if (token == Opt_min_batch_time) {
1839                 sbi->s_min_batch_time = arg;
1840         } else if (token == Opt_inode_readahead_blks) {
1841                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1842                         ext4_msg(sb, KERN_ERR,
1843                                  "EXT4-fs: inode_readahead_blks must be "
1844                                  "0 or a power of 2 smaller than 2^31");
1845                         return -1;
1846                 }
1847                 sbi->s_inode_readahead_blks = arg;
1848         } else if (token == Opt_init_itable) {
1849                 set_opt(sb, INIT_INODE_TABLE);
1850                 if (!args->from)
1851                         arg = EXT4_DEF_LI_WAIT_MULT;
1852                 sbi->s_li_wait_mult = arg;
1853         } else if (token == Opt_max_dir_size_kb) {
1854                 sbi->s_max_dir_size_kb = arg;
1855         } else if (token == Opt_stripe) {
1856                 sbi->s_stripe = arg;
1857         } else if (token == Opt_resuid) {
1858                 uid = make_kuid(current_user_ns(), arg);
1859                 if (!uid_valid(uid)) {
1860                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1861                         return -1;
1862                 }
1863                 sbi->s_resuid = uid;
1864         } else if (token == Opt_resgid) {
1865                 gid = make_kgid(current_user_ns(), arg);
1866                 if (!gid_valid(gid)) {
1867                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1868                         return -1;
1869                 }
1870                 sbi->s_resgid = gid;
1871         } else if (token == Opt_journal_dev) {
1872                 if (is_remount) {
1873                         ext4_msg(sb, KERN_ERR,
1874                                  "Cannot specify journal on remount");
1875                         return -1;
1876                 }
1877                 *journal_devnum = arg;
1878         } else if (token == Opt_journal_path) {
1879                 char *journal_path;
1880                 struct inode *journal_inode;
1881                 struct path path;
1882                 int error;
1883
1884                 if (is_remount) {
1885                         ext4_msg(sb, KERN_ERR,
1886                                  "Cannot specify journal on remount");
1887                         return -1;
1888                 }
1889                 journal_path = match_strdup(&args[0]);
1890                 if (!journal_path) {
1891                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1892                                 "journal device string");
1893                         return -1;
1894                 }
1895
1896                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1897                 if (error) {
1898                         ext4_msg(sb, KERN_ERR, "error: could not find "
1899                                 "journal device path: error %d", error);
1900                         kfree(journal_path);
1901                         return -1;
1902                 }
1903
1904                 journal_inode = d_inode(path.dentry);
1905                 if (!S_ISBLK(journal_inode->i_mode)) {
1906                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1907                                 "is not a block device", journal_path);
1908                         path_put(&path);
1909                         kfree(journal_path);
1910                         return -1;
1911                 }
1912
1913                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1914                 path_put(&path);
1915                 kfree(journal_path);
1916         } else if (token == Opt_journal_ioprio) {
1917                 if (arg > 7) {
1918                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1919                                  " (must be 0-7)");
1920                         return -1;
1921                 }
1922                 *journal_ioprio =
1923                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1924         } else if (token == Opt_test_dummy_encryption) {
1925 #ifdef CONFIG_FS_ENCRYPTION
1926                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1927                 ext4_msg(sb, KERN_WARNING,
1928                          "Test dummy encryption mode enabled");
1929 #else
1930                 ext4_msg(sb, KERN_WARNING,
1931                          "Test dummy encryption mount option ignored");
1932 #endif
1933         } else if (m->flags & MOPT_DATAJ) {
1934                 if (is_remount) {
1935                         if (!sbi->s_journal)
1936                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1937                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1938                                 ext4_msg(sb, KERN_ERR,
1939                                          "Cannot change data mode on remount");
1940                                 return -1;
1941                         }
1942                 } else {
1943                         clear_opt(sb, DATA_FLAGS);
1944                         sbi->s_mount_opt |= m->mount_opt;
1945                 }
1946 #ifdef CONFIG_QUOTA
1947         } else if (m->flags & MOPT_QFMT) {
1948                 if (sb_any_quota_loaded(sb) &&
1949                     sbi->s_jquota_fmt != m->mount_opt) {
1950                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1951                                  "quota options when quota turned on");
1952                         return -1;
1953                 }
1954                 if (ext4_has_feature_quota(sb)) {
1955                         ext4_msg(sb, KERN_INFO,
1956                                  "Quota format mount options ignored "
1957                                  "when QUOTA feature is enabled");
1958                         return 1;
1959                 }
1960                 sbi->s_jquota_fmt = m->mount_opt;
1961 #endif
1962         } else if (token == Opt_dax) {
1963 #ifdef CONFIG_FS_DAX
1964                 ext4_msg(sb, KERN_WARNING,
1965                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1966                 sbi->s_mount_opt |= m->mount_opt;
1967 #else
1968                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1969                 return -1;
1970 #endif
1971         } else if (token == Opt_data_err_abort) {
1972                 sbi->s_mount_opt |= m->mount_opt;
1973         } else if (token == Opt_data_err_ignore) {
1974                 sbi->s_mount_opt &= ~m->mount_opt;
1975         } else {
1976                 if (!args->from)
1977                         arg = 1;
1978                 if (m->flags & MOPT_CLEAR)
1979                         arg = !arg;
1980                 else if (unlikely(!(m->flags & MOPT_SET))) {
1981                         ext4_msg(sb, KERN_WARNING,
1982                                  "buggy handling of option %s", opt);
1983                         WARN_ON(1);
1984                         return -1;
1985                 }
1986                 if (arg != 0)
1987                         sbi->s_mount_opt |= m->mount_opt;
1988                 else
1989                         sbi->s_mount_opt &= ~m->mount_opt;
1990         }
1991         return 1;
1992 }
1993
1994 static int parse_options(char *options, struct super_block *sb,
1995                          unsigned long *journal_devnum,
1996                          unsigned int *journal_ioprio,
1997                          int is_remount)
1998 {
1999         struct ext4_sb_info *sbi = EXT4_SB(sb);
2000         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2001         substring_t args[MAX_OPT_ARGS];
2002         int token;
2003
2004         if (!options)
2005                 return 1;
2006
2007         while ((p = strsep(&options, ",")) != NULL) {
2008                 if (!*p)
2009                         continue;
2010                 /*
2011                  * Initialize args struct so we know whether arg was
2012                  * found; some options take optional arguments.
2013                  */
2014                 args[0].to = args[0].from = NULL;
2015                 token = match_token(p, tokens, args);
2016                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2017                                      journal_ioprio, is_remount) < 0)
2018                         return 0;
2019         }
2020 #ifdef CONFIG_QUOTA
2021         /*
2022          * We do the test below only for project quotas. 'usrquota' and
2023          * 'grpquota' mount options are allowed even without quota feature
2024          * to support legacy quotas in quota files.
2025          */
2026         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2027                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2028                          "Cannot enable project quota enforcement.");
2029                 return 0;
2030         }
2031         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2032         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2033         if (usr_qf_name || grp_qf_name) {
2034                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2035                         clear_opt(sb, USRQUOTA);
2036
2037                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2038                         clear_opt(sb, GRPQUOTA);
2039
2040                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2041                         ext4_msg(sb, KERN_ERR, "old and new quota "
2042                                         "format mixing");
2043                         return 0;
2044                 }
2045
2046                 if (!sbi->s_jquota_fmt) {
2047                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2048                                         "not specified");
2049                         return 0;
2050                 }
2051         }
2052 #endif
2053         if (test_opt(sb, DIOREAD_NOLOCK)) {
2054                 int blocksize =
2055                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2056
2057                 if (blocksize < PAGE_SIZE) {
2058                         ext4_msg(sb, KERN_ERR, "can't mount with "
2059                                  "dioread_nolock if block size != PAGE_SIZE");
2060                         return 0;
2061                 }
2062         }
2063         return 1;
2064 }
2065
2066 static inline void ext4_show_quota_options(struct seq_file *seq,
2067                                            struct super_block *sb)
2068 {
2069 #if defined(CONFIG_QUOTA)
2070         struct ext4_sb_info *sbi = EXT4_SB(sb);
2071         char *usr_qf_name, *grp_qf_name;
2072
2073         if (sbi->s_jquota_fmt) {
2074                 char *fmtname = "";
2075
2076                 switch (sbi->s_jquota_fmt) {
2077                 case QFMT_VFS_OLD:
2078                         fmtname = "vfsold";
2079                         break;
2080                 case QFMT_VFS_V0:
2081                         fmtname = "vfsv0";
2082                         break;
2083                 case QFMT_VFS_V1:
2084                         fmtname = "vfsv1";
2085                         break;
2086                 }
2087                 seq_printf(seq, ",jqfmt=%s", fmtname);
2088         }
2089
2090         rcu_read_lock();
2091         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2092         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2093         if (usr_qf_name)
2094                 seq_show_option(seq, "usrjquota", usr_qf_name);
2095         if (grp_qf_name)
2096                 seq_show_option(seq, "grpjquota", grp_qf_name);
2097         rcu_read_unlock();
2098 #endif
2099 }
2100
2101 static const char *token2str(int token)
2102 {
2103         const struct match_token *t;
2104
2105         for (t = tokens; t->token != Opt_err; t++)
2106                 if (t->token == token && !strchr(t->pattern, '='))
2107                         break;
2108         return t->pattern;
2109 }
2110
2111 /*
2112  * Show an option if
2113  *  - it's set to a non-default value OR
2114  *  - if the per-sb default is different from the global default
2115  */
2116 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2117                               int nodefs)
2118 {
2119         struct ext4_sb_info *sbi = EXT4_SB(sb);
2120         struct ext4_super_block *es = sbi->s_es;
2121         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2122         const struct mount_opts *m;
2123         char sep = nodefs ? '\n' : ',';
2124
2125 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2126 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2127
2128         if (sbi->s_sb_block != 1)
2129                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2130
2131         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2132                 int want_set = m->flags & MOPT_SET;
2133                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2134                     (m->flags & MOPT_CLEAR_ERR))
2135                         continue;
2136                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2137                         continue; /* skip if same as the default */
2138                 if ((want_set &&
2139                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2140                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2141                         continue; /* select Opt_noFoo vs Opt_Foo */
2142                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2143         }
2144
2145         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2146             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2147                 SEQ_OPTS_PRINT("resuid=%u",
2148                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2149         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2150             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2151                 SEQ_OPTS_PRINT("resgid=%u",
2152                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2153         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2154         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2155                 SEQ_OPTS_PUTS("errors=remount-ro");
2156         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2157                 SEQ_OPTS_PUTS("errors=continue");
2158         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2159                 SEQ_OPTS_PUTS("errors=panic");
2160         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2161                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2162         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2163                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2164         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2165                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2166         if (sb->s_flags & SB_I_VERSION)
2167                 SEQ_OPTS_PUTS("i_version");
2168         if (nodefs || sbi->s_stripe)
2169                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2170         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2171                         (sbi->s_mount_opt ^ def_mount_opt)) {
2172                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2173                         SEQ_OPTS_PUTS("data=journal");
2174                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2175                         SEQ_OPTS_PUTS("data=ordered");
2176                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2177                         SEQ_OPTS_PUTS("data=writeback");
2178         }
2179         if (nodefs ||
2180             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2181                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2182                                sbi->s_inode_readahead_blks);
2183
2184         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2185                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2186                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2187         if (nodefs || sbi->s_max_dir_size_kb)
2188                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2189         if (test_opt(sb, DATA_ERR_ABORT))
2190                 SEQ_OPTS_PUTS("data_err=abort");
2191         if (DUMMY_ENCRYPTION_ENABLED(sbi))
2192                 SEQ_OPTS_PUTS("test_dummy_encryption");
2193
2194         ext4_show_quota_options(seq, sb);
2195         return 0;
2196 }
2197
2198 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2199 {
2200         return _ext4_show_options(seq, root->d_sb, 0);
2201 }
2202
2203 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2204 {
2205         struct super_block *sb = seq->private;
2206         int rc;
2207
2208         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2209         rc = _ext4_show_options(seq, sb, 1);
2210         seq_puts(seq, "\n");
2211         return rc;
2212 }
2213
2214 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2215                             int read_only)
2216 {
2217         struct ext4_sb_info *sbi = EXT4_SB(sb);
2218         int err = 0;
2219
2220         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2221                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2222                          "forcing read-only mode");
2223                 err = -EROFS;
2224         }
2225         if (read_only)
2226                 goto done;
2227         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2228                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2229                          "running e2fsck is recommended");
2230         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2231                 ext4_msg(sb, KERN_WARNING,
2232                          "warning: mounting fs with errors, "
2233                          "running e2fsck is recommended");
2234         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2235                  le16_to_cpu(es->s_mnt_count) >=
2236                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2237                 ext4_msg(sb, KERN_WARNING,
2238                          "warning: maximal mount count reached, "
2239                          "running e2fsck is recommended");
2240         else if (le32_to_cpu(es->s_checkinterval) &&
2241                  (ext4_get_tstamp(es, s_lastcheck) +
2242                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2243                 ext4_msg(sb, KERN_WARNING,
2244                          "warning: checktime reached, "
2245                          "running e2fsck is recommended");
2246         if (!sbi->s_journal)
2247                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2248         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2249                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2250         le16_add_cpu(&es->s_mnt_count, 1);
2251         ext4_update_tstamp(es, s_mtime);
2252         if (sbi->s_journal)
2253                 ext4_set_feature_journal_needs_recovery(sb);
2254
2255         err = ext4_commit_super(sb, 1);
2256 done:
2257         if (test_opt(sb, DEBUG))
2258                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2259                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2260                         sb->s_blocksize,
2261                         sbi->s_groups_count,
2262                         EXT4_BLOCKS_PER_GROUP(sb),
2263                         EXT4_INODES_PER_GROUP(sb),
2264                         sbi->s_mount_opt, sbi->s_mount_opt2);
2265
2266         cleancache_init_fs(sb);
2267         return err;
2268 }
2269
2270 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2271 {
2272         struct ext4_sb_info *sbi = EXT4_SB(sb);
2273         struct flex_groups *new_groups;
2274         int size;
2275
2276         if (!sbi->s_log_groups_per_flex)
2277                 return 0;
2278
2279         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2280         if (size <= sbi->s_flex_groups_allocated)
2281                 return 0;
2282
2283         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2284         new_groups = kvzalloc(size, GFP_KERNEL);
2285         if (!new_groups) {
2286                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2287                          size / (int) sizeof(struct flex_groups));
2288                 return -ENOMEM;
2289         }
2290
2291         if (sbi->s_flex_groups) {
2292                 memcpy(new_groups, sbi->s_flex_groups,
2293                        (sbi->s_flex_groups_allocated *
2294                         sizeof(struct flex_groups)));
2295                 kvfree(sbi->s_flex_groups);
2296         }
2297         sbi->s_flex_groups = new_groups;
2298         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2299         return 0;
2300 }
2301
2302 static int ext4_fill_flex_info(struct super_block *sb)
2303 {
2304         struct ext4_sb_info *sbi = EXT4_SB(sb);
2305         struct ext4_group_desc *gdp = NULL;
2306         ext4_group_t flex_group;
2307         int i, err;
2308
2309         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2310         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2311                 sbi->s_log_groups_per_flex = 0;
2312                 return 1;
2313         }
2314
2315         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2316         if (err)
2317                 goto failed;
2318
2319         for (i = 0; i < sbi->s_groups_count; i++) {
2320                 gdp = ext4_get_group_desc(sb, i, NULL);
2321
2322                 flex_group = ext4_flex_group(sbi, i);
2323                 atomic_add(ext4_free_inodes_count(sb, gdp),
2324                            &sbi->s_flex_groups[flex_group].free_inodes);
2325                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2326                              &sbi->s_flex_groups[flex_group].free_clusters);
2327                 atomic_add(ext4_used_dirs_count(sb, gdp),
2328                            &sbi->s_flex_groups[flex_group].used_dirs);
2329         }
2330
2331         return 1;
2332 failed:
2333         return 0;
2334 }
2335
2336 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2337                                    struct ext4_group_desc *gdp)
2338 {
2339         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2340         __u16 crc = 0;
2341         __le32 le_group = cpu_to_le32(block_group);
2342         struct ext4_sb_info *sbi = EXT4_SB(sb);
2343
2344         if (ext4_has_metadata_csum(sbi->s_sb)) {
2345                 /* Use new metadata_csum algorithm */
2346                 __u32 csum32;
2347                 __u16 dummy_csum = 0;
2348
2349                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2350                                      sizeof(le_group));
2351                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2352                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2353                                      sizeof(dummy_csum));
2354                 offset += sizeof(dummy_csum);
2355                 if (offset < sbi->s_desc_size)
2356                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2357                                              sbi->s_desc_size - offset);
2358
2359                 crc = csum32 & 0xFFFF;
2360                 goto out;
2361         }
2362
2363         /* old crc16 code */
2364         if (!ext4_has_feature_gdt_csum(sb))
2365                 return 0;
2366
2367         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2368         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2369         crc = crc16(crc, (__u8 *)gdp, offset);
2370         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2371         /* for checksum of struct ext4_group_desc do the rest...*/
2372         if (ext4_has_feature_64bit(sb) &&
2373             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2374                 crc = crc16(crc, (__u8 *)gdp + offset,
2375                             le16_to_cpu(sbi->s_es->s_desc_size) -
2376                                 offset);
2377
2378 out:
2379         return cpu_to_le16(crc);
2380 }
2381
2382 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2383                                 struct ext4_group_desc *gdp)
2384 {
2385         if (ext4_has_group_desc_csum(sb) &&
2386             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2387                 return 0;
2388
2389         return 1;
2390 }
2391
2392 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2393                               struct ext4_group_desc *gdp)
2394 {
2395         if (!ext4_has_group_desc_csum(sb))
2396                 return;
2397         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2398 }
2399
2400 /* Called at mount-time, super-block is locked */
2401 static int ext4_check_descriptors(struct super_block *sb,
2402                                   ext4_fsblk_t sb_block,
2403                                   ext4_group_t *first_not_zeroed)
2404 {
2405         struct ext4_sb_info *sbi = EXT4_SB(sb);
2406         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2407         ext4_fsblk_t last_block;
2408         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2409         ext4_fsblk_t block_bitmap;
2410         ext4_fsblk_t inode_bitmap;
2411         ext4_fsblk_t inode_table;
2412         int flexbg_flag = 0;
2413         ext4_group_t i, grp = sbi->s_groups_count;
2414
2415         if (ext4_has_feature_flex_bg(sb))
2416                 flexbg_flag = 1;
2417
2418         ext4_debug("Checking group descriptors");
2419
2420         for (i = 0; i < sbi->s_groups_count; i++) {
2421                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2422
2423                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2424                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2425                 else
2426                         last_block = first_block +
2427                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2428
2429                 if ((grp == sbi->s_groups_count) &&
2430                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2431                         grp = i;
2432
2433                 block_bitmap = ext4_block_bitmap(sb, gdp);
2434                 if (block_bitmap == sb_block) {
2435                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2436                                  "Block bitmap for group %u overlaps "
2437                                  "superblock", i);
2438                         if (!sb_rdonly(sb))
2439                                 return 0;
2440                 }
2441                 if (block_bitmap >= sb_block + 1 &&
2442                     block_bitmap <= last_bg_block) {
2443                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2444                                  "Block bitmap for group %u overlaps "
2445                                  "block group descriptors", i);
2446                         if (!sb_rdonly(sb))
2447                                 return 0;
2448                 }
2449                 if (block_bitmap < first_block || block_bitmap > last_block) {
2450                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2451                                "Block bitmap for group %u not in group "
2452                                "(block %llu)!", i, block_bitmap);
2453                         return 0;
2454                 }
2455                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2456                 if (inode_bitmap == sb_block) {
2457                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2458                                  "Inode bitmap for group %u overlaps "
2459                                  "superblock", i);
2460                         if (!sb_rdonly(sb))
2461                                 return 0;
2462                 }
2463                 if (inode_bitmap >= sb_block + 1 &&
2464                     inode_bitmap <= last_bg_block) {
2465                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2466                                  "Inode bitmap for group %u overlaps "
2467                                  "block group descriptors", i);
2468                         if (!sb_rdonly(sb))
2469                                 return 0;
2470                 }
2471                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2472                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2473                                "Inode bitmap for group %u not in group "
2474                                "(block %llu)!", i, inode_bitmap);
2475                         return 0;
2476                 }
2477                 inode_table = ext4_inode_table(sb, gdp);
2478                 if (inode_table == sb_block) {
2479                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2480                                  "Inode table for group %u overlaps "
2481                                  "superblock", i);
2482                         if (!sb_rdonly(sb))
2483                                 return 0;
2484                 }
2485                 if (inode_table >= sb_block + 1 &&
2486                     inode_table <= last_bg_block) {
2487                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2488                                  "Inode table for group %u overlaps "
2489                                  "block group descriptors", i);
2490                         if (!sb_rdonly(sb))
2491                                 return 0;
2492                 }
2493                 if (inode_table < first_block ||
2494                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2495                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2496                                "Inode table for group %u not in group "
2497                                "(block %llu)!", i, inode_table);
2498                         return 0;
2499                 }
2500                 ext4_lock_group(sb, i);
2501                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2502                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2503                                  "Checksum for group %u failed (%u!=%u)",
2504                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2505                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2506                         if (!sb_rdonly(sb)) {
2507                                 ext4_unlock_group(sb, i);
2508                                 return 0;
2509                         }
2510                 }
2511                 ext4_unlock_group(sb, i);
2512                 if (!flexbg_flag)
2513                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2514         }
2515         if (NULL != first_not_zeroed)
2516                 *first_not_zeroed = grp;
2517         return 1;
2518 }
2519
2520 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2521  * the superblock) which were deleted from all directories, but held open by
2522  * a process at the time of a crash.  We walk the list and try to delete these
2523  * inodes at recovery time (only with a read-write filesystem).
2524  *
2525  * In order to keep the orphan inode chain consistent during traversal (in
2526  * case of crash during recovery), we link each inode into the superblock
2527  * orphan list_head and handle it the same way as an inode deletion during
2528  * normal operation (which journals the operations for us).
2529  *
2530  * We only do an iget() and an iput() on each inode, which is very safe if we
2531  * accidentally point at an in-use or already deleted inode.  The worst that
2532  * can happen in this case is that we get a "bit already cleared" message from
2533  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2534  * e2fsck was run on this filesystem, and it must have already done the orphan
2535  * inode cleanup for us, so we can safely abort without any further action.
2536  */
2537 static void ext4_orphan_cleanup(struct super_block *sb,
2538                                 struct ext4_super_block *es)
2539 {
2540         unsigned int s_flags = sb->s_flags;
2541         int ret, nr_orphans = 0, nr_truncates = 0;
2542 #ifdef CONFIG_QUOTA
2543         int quota_update = 0;
2544         int i;
2545 #endif
2546         if (!es->s_last_orphan) {
2547                 jbd_debug(4, "no orphan inodes to clean up\n");
2548                 return;
2549         }
2550
2551         if (bdev_read_only(sb->s_bdev)) {
2552                 ext4_msg(sb, KERN_ERR, "write access "
2553                         "unavailable, skipping orphan cleanup");
2554                 return;
2555         }
2556
2557         /* Check if feature set would not allow a r/w mount */
2558         if (!ext4_feature_set_ok(sb, 0)) {
2559                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2560                          "unknown ROCOMPAT features");
2561                 return;
2562         }
2563
2564         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2565                 /* don't clear list on RO mount w/ errors */
2566                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2567                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2568                                   "clearing orphan list.\n");
2569                         es->s_last_orphan = 0;
2570                 }
2571                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2572                 return;
2573         }
2574
2575         if (s_flags & SB_RDONLY) {
2576                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2577                 sb->s_flags &= ~SB_RDONLY;
2578         }
2579 #ifdef CONFIG_QUOTA
2580         /* Needed for iput() to work correctly and not trash data */
2581         sb->s_flags |= SB_ACTIVE;
2582
2583         /*
2584          * Turn on quotas which were not enabled for read-only mounts if
2585          * filesystem has quota feature, so that they are updated correctly.
2586          */
2587         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2588                 int ret = ext4_enable_quotas(sb);
2589
2590                 if (!ret)
2591                         quota_update = 1;
2592                 else
2593                         ext4_msg(sb, KERN_ERR,
2594                                 "Cannot turn on quotas: error %d", ret);
2595         }
2596
2597         /* Turn on journaled quotas used for old sytle */
2598         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2599                 if (EXT4_SB(sb)->s_qf_names[i]) {
2600                         int ret = ext4_quota_on_mount(sb, i);
2601
2602                         if (!ret)
2603                                 quota_update = 1;
2604                         else
2605                                 ext4_msg(sb, KERN_ERR,
2606                                         "Cannot turn on journaled "
2607                                         "quota: type %d: error %d", i, ret);
2608                 }
2609         }
2610 #endif
2611
2612         while (es->s_last_orphan) {
2613                 struct inode *inode;
2614
2615                 /*
2616                  * We may have encountered an error during cleanup; if
2617                  * so, skip the rest.
2618                  */
2619                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2620                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2621                         es->s_last_orphan = 0;
2622                         break;
2623                 }
2624
2625                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2626                 if (IS_ERR(inode)) {
2627                         es->s_last_orphan = 0;
2628                         break;
2629                 }
2630
2631                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2632                 dquot_initialize(inode);
2633                 if (inode->i_nlink) {
2634                         if (test_opt(sb, DEBUG))
2635                                 ext4_msg(sb, KERN_DEBUG,
2636                                         "%s: truncating inode %lu to %lld bytes",
2637                                         __func__, inode->i_ino, inode->i_size);
2638                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2639                                   inode->i_ino, inode->i_size);
2640                         inode_lock(inode);
2641                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2642                         ret = ext4_truncate(inode);
2643                         if (ret)
2644                                 ext4_std_error(inode->i_sb, ret);
2645                         inode_unlock(inode);
2646                         nr_truncates++;
2647                 } else {
2648                         if (test_opt(sb, DEBUG))
2649                                 ext4_msg(sb, KERN_DEBUG,
2650                                         "%s: deleting unreferenced inode %lu",
2651                                         __func__, inode->i_ino);
2652                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2653                                   inode->i_ino);
2654                         nr_orphans++;
2655                 }
2656                 iput(inode);  /* The delete magic happens here! */
2657         }
2658
2659 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2660
2661         if (nr_orphans)
2662                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2663                        PLURAL(nr_orphans));
2664         if (nr_truncates)
2665                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2666                        PLURAL(nr_truncates));
2667 #ifdef CONFIG_QUOTA
2668         /* Turn off quotas if they were enabled for orphan cleanup */
2669         if (quota_update) {
2670                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2671                         if (sb_dqopt(sb)->files[i])
2672                                 dquot_quota_off(sb, i);
2673                 }
2674         }
2675 #endif
2676         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2677 }
2678
2679 /*
2680  * Maximal extent format file size.
2681  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2682  * extent format containers, within a sector_t, and within i_blocks
2683  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2684  * so that won't be a limiting factor.
2685  *
2686  * However there is other limiting factor. We do store extents in the form
2687  * of starting block and length, hence the resulting length of the extent
2688  * covering maximum file size must fit into on-disk format containers as
2689  * well. Given that length is always by 1 unit bigger than max unit (because
2690  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2691  *
2692  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2693  */
2694 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2695 {
2696         loff_t res;
2697         loff_t upper_limit = MAX_LFS_FILESIZE;
2698
2699         /* small i_blocks in vfs inode? */
2700         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2701                 /*
2702                  * CONFIG_LBDAF is not enabled implies the inode
2703                  * i_block represent total blocks in 512 bytes
2704                  * 32 == size of vfs inode i_blocks * 8
2705                  */
2706                 upper_limit = (1LL << 32) - 1;
2707
2708                 /* total blocks in file system block size */
2709                 upper_limit >>= (blkbits - 9);
2710                 upper_limit <<= blkbits;
2711         }
2712
2713         /*
2714          * 32-bit extent-start container, ee_block. We lower the maxbytes
2715          * by one fs block, so ee_len can cover the extent of maximum file
2716          * size
2717          */
2718         res = (1LL << 32) - 1;
2719         res <<= blkbits;
2720
2721         /* Sanity check against vm- & vfs- imposed limits */
2722         if (res > upper_limit)
2723                 res = upper_limit;
2724
2725         return res;
2726 }
2727
2728 /*
2729  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2730  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2731  * We need to be 1 filesystem block less than the 2^48 sector limit.
2732  */
2733 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2734 {
2735         loff_t res = EXT4_NDIR_BLOCKS;
2736         int meta_blocks;
2737         loff_t upper_limit;
2738         /* This is calculated to be the largest file size for a dense, block
2739          * mapped file such that the file's total number of 512-byte sectors,
2740          * including data and all indirect blocks, does not exceed (2^48 - 1).
2741          *
2742          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2743          * number of 512-byte sectors of the file.
2744          */
2745
2746         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2747                 /*
2748                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2749                  * the inode i_block field represents total file blocks in
2750                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2751                  */
2752                 upper_limit = (1LL << 32) - 1;
2753
2754                 /* total blocks in file system block size */
2755                 upper_limit >>= (bits - 9);
2756
2757         } else {
2758                 /*
2759                  * We use 48 bit ext4_inode i_blocks
2760                  * With EXT4_HUGE_FILE_FL set the i_blocks
2761                  * represent total number of blocks in
2762                  * file system block size
2763                  */
2764                 upper_limit = (1LL << 48) - 1;
2765
2766         }
2767
2768         /* indirect blocks */
2769         meta_blocks = 1;
2770         /* double indirect blocks */
2771         meta_blocks += 1 + (1LL << (bits-2));
2772         /* tripple indirect blocks */
2773         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2774
2775         upper_limit -= meta_blocks;
2776         upper_limit <<= bits;
2777
2778         res += 1LL << (bits-2);
2779         res += 1LL << (2*(bits-2));
2780         res += 1LL << (3*(bits-2));
2781         res <<= bits;
2782         if (res > upper_limit)
2783                 res = upper_limit;
2784
2785         if (res > MAX_LFS_FILESIZE)
2786                 res = MAX_LFS_FILESIZE;
2787
2788         return res;
2789 }
2790
2791 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2792                                    ext4_fsblk_t logical_sb_block, int nr)
2793 {
2794         struct ext4_sb_info *sbi = EXT4_SB(sb);
2795         ext4_group_t bg, first_meta_bg;
2796         int has_super = 0;
2797
2798         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2799
2800         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2801                 return logical_sb_block + nr + 1;
2802         bg = sbi->s_desc_per_block * nr;
2803         if (ext4_bg_has_super(sb, bg))
2804                 has_super = 1;
2805
2806         /*
2807          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2808          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2809          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2810          * compensate.
2811          */
2812         if (sb->s_blocksize == 1024 && nr == 0 &&
2813             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2814                 has_super++;
2815
2816         return (has_super + ext4_group_first_block_no(sb, bg));
2817 }
2818
2819 /**
2820  * ext4_get_stripe_size: Get the stripe size.
2821  * @sbi: In memory super block info
2822  *
2823  * If we have specified it via mount option, then
2824  * use the mount option value. If the value specified at mount time is
2825  * greater than the blocks per group use the super block value.
2826  * If the super block value is greater than blocks per group return 0.
2827  * Allocator needs it be less than blocks per group.
2828  *
2829  */
2830 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2831 {
2832         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2833         unsigned long stripe_width =
2834                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2835         int ret;
2836
2837         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2838                 ret = sbi->s_stripe;
2839         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2840                 ret = stripe_width;
2841         else if (stride && stride <= sbi->s_blocks_per_group)
2842                 ret = stride;
2843         else
2844                 ret = 0;
2845
2846         /*
2847          * If the stripe width is 1, this makes no sense and
2848          * we set it to 0 to turn off stripe handling code.
2849          */
2850         if (ret <= 1)
2851                 ret = 0;
2852
2853         return ret;
2854 }
2855
2856 /*
2857  * Check whether this filesystem can be mounted based on
2858  * the features present and the RDONLY/RDWR mount requested.
2859  * Returns 1 if this filesystem can be mounted as requested,
2860  * 0 if it cannot be.
2861  */
2862 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2863 {
2864         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2865                 ext4_msg(sb, KERN_ERR,
2866                         "Couldn't mount because of "
2867                         "unsupported optional features (%x)",
2868                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2869                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2870                 return 0;
2871         }
2872
2873         if (readonly)
2874                 return 1;
2875
2876         if (ext4_has_feature_readonly(sb)) {
2877                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2878                 sb->s_flags |= SB_RDONLY;
2879                 return 1;
2880         }
2881
2882         /* Check that feature set is OK for a read-write mount */
2883         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2884                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2885                          "unsupported optional features (%x)",
2886                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2887                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2888                 return 0;
2889         }
2890         /*
2891          * Large file size enabled file system can only be mounted
2892          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2893          */
2894         if (ext4_has_feature_huge_file(sb)) {
2895                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2896                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2897                                  "cannot be mounted RDWR without "
2898                                  "CONFIG_LBDAF");
2899                         return 0;
2900                 }
2901         }
2902         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2903                 ext4_msg(sb, KERN_ERR,
2904                          "Can't support bigalloc feature without "
2905                          "extents feature\n");
2906                 return 0;
2907         }
2908
2909 #ifndef CONFIG_QUOTA
2910         if (ext4_has_feature_quota(sb) && !readonly) {
2911                 ext4_msg(sb, KERN_ERR,
2912                          "Filesystem with quota feature cannot be mounted RDWR "
2913                          "without CONFIG_QUOTA");
2914                 return 0;
2915         }
2916         if (ext4_has_feature_project(sb) && !readonly) {
2917                 ext4_msg(sb, KERN_ERR,
2918                          "Filesystem with project quota feature cannot be mounted RDWR "
2919                          "without CONFIG_QUOTA");
2920                 return 0;
2921         }
2922 #endif  /* CONFIG_QUOTA */
2923         return 1;
2924 }
2925
2926 /*
2927  * This function is called once a day if we have errors logged
2928  * on the file system
2929  */
2930 static void print_daily_error_info(struct timer_list *t)
2931 {
2932         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
2933         struct super_block *sb = sbi->s_sb;
2934         struct ext4_super_block *es = sbi->s_es;
2935
2936         if (es->s_error_count)
2937                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2938                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2939                          le32_to_cpu(es->s_error_count));
2940         if (es->s_first_error_time) {
2941                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
2942                        sb->s_id,
2943                        ext4_get_tstamp(es, s_first_error_time),
2944                        (int) sizeof(es->s_first_error_func),
2945                        es->s_first_error_func,
2946                        le32_to_cpu(es->s_first_error_line));
2947                 if (es->s_first_error_ino)
2948                         printk(KERN_CONT ": inode %u",
2949                                le32_to_cpu(es->s_first_error_ino));
2950                 if (es->s_first_error_block)
2951                         printk(KERN_CONT ": block %llu", (unsigned long long)
2952                                le64_to_cpu(es->s_first_error_block));
2953                 printk(KERN_CONT "\n");
2954         }
2955         if (es->s_last_error_time) {
2956                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
2957                        sb->s_id,
2958                        ext4_get_tstamp(es, s_last_error_time),
2959                        (int) sizeof(es->s_last_error_func),
2960                        es->s_last_error_func,
2961                        le32_to_cpu(es->s_last_error_line));
2962                 if (es->s_last_error_ino)
2963                         printk(KERN_CONT ": inode %u",
2964                                le32_to_cpu(es->s_last_error_ino));
2965                 if (es->s_last_error_block)
2966                         printk(KERN_CONT ": block %llu", (unsigned long long)
2967                                le64_to_cpu(es->s_last_error_block));
2968                 printk(KERN_CONT "\n");
2969         }
2970         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2971 }
2972
2973 /* Find next suitable group and run ext4_init_inode_table */
2974 static int ext4_run_li_request(struct ext4_li_request *elr)
2975 {
2976         struct ext4_group_desc *gdp = NULL;
2977         ext4_group_t group, ngroups;
2978         struct super_block *sb;
2979         unsigned long timeout = 0;
2980         int ret = 0;
2981
2982         sb = elr->lr_super;
2983         ngroups = EXT4_SB(sb)->s_groups_count;
2984
2985         for (group = elr->lr_next_group; group < ngroups; group++) {
2986                 gdp = ext4_get_group_desc(sb, group, NULL);
2987                 if (!gdp) {
2988                         ret = 1;
2989                         break;
2990                 }
2991
2992                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2993                         break;
2994         }
2995
2996         if (group >= ngroups)
2997                 ret = 1;
2998
2999         if (!ret) {
3000                 timeout = jiffies;
3001                 ret = ext4_init_inode_table(sb, group,
3002                                             elr->lr_timeout ? 0 : 1);
3003                 if (elr->lr_timeout == 0) {
3004                         timeout = (jiffies - timeout) *
3005                                   elr->lr_sbi->s_li_wait_mult;
3006                         elr->lr_timeout = timeout;
3007                 }
3008                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3009                 elr->lr_next_group = group + 1;
3010         }
3011         return ret;
3012 }
3013
3014 /*
3015  * Remove lr_request from the list_request and free the
3016  * request structure. Should be called with li_list_mtx held
3017  */
3018 static void ext4_remove_li_request(struct ext4_li_request *elr)
3019 {
3020         struct ext4_sb_info *sbi;
3021
3022         if (!elr)
3023                 return;
3024
3025         sbi = elr->lr_sbi;
3026
3027         list_del(&elr->lr_request);
3028         sbi->s_li_request = NULL;
3029         kfree(elr);
3030 }
3031
3032 static void ext4_unregister_li_request(struct super_block *sb)
3033 {
3034         mutex_lock(&ext4_li_mtx);
3035         if (!ext4_li_info) {
3036                 mutex_unlock(&ext4_li_mtx);
3037                 return;
3038         }
3039
3040         mutex_lock(&ext4_li_info->li_list_mtx);
3041         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3042         mutex_unlock(&ext4_li_info->li_list_mtx);
3043         mutex_unlock(&ext4_li_mtx);
3044 }
3045
3046 static struct task_struct *ext4_lazyinit_task;
3047
3048 /*
3049  * This is the function where ext4lazyinit thread lives. It walks
3050  * through the request list searching for next scheduled filesystem.
3051  * When such a fs is found, run the lazy initialization request
3052  * (ext4_rn_li_request) and keep track of the time spend in this
3053  * function. Based on that time we compute next schedule time of
3054  * the request. When walking through the list is complete, compute
3055  * next waking time and put itself into sleep.
3056  */
3057 static int ext4_lazyinit_thread(void *arg)
3058 {
3059         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3060         struct list_head *pos, *n;
3061         struct ext4_li_request *elr;
3062         unsigned long next_wakeup, cur;
3063
3064         BUG_ON(NULL == eli);
3065
3066 cont_thread:
3067         while (true) {
3068                 next_wakeup = MAX_JIFFY_OFFSET;
3069
3070                 mutex_lock(&eli->li_list_mtx);
3071                 if (list_empty(&eli->li_request_list)) {
3072                         mutex_unlock(&eli->li_list_mtx);
3073                         goto exit_thread;
3074                 }
3075                 list_for_each_safe(pos, n, &eli->li_request_list) {
3076                         int err = 0;
3077                         int progress = 0;
3078                         elr = list_entry(pos, struct ext4_li_request,
3079                                          lr_request);
3080
3081                         if (time_before(jiffies, elr->lr_next_sched)) {
3082                                 if (time_before(elr->lr_next_sched, next_wakeup))
3083                                         next_wakeup = elr->lr_next_sched;
3084                                 continue;
3085                         }
3086                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3087                                 if (sb_start_write_trylock(elr->lr_super)) {
3088                                         progress = 1;
3089                                         /*
3090                                          * We hold sb->s_umount, sb can not
3091                                          * be removed from the list, it is
3092                                          * now safe to drop li_list_mtx
3093                                          */
3094                                         mutex_unlock(&eli->li_list_mtx);
3095                                         err = ext4_run_li_request(elr);
3096                                         sb_end_write(elr->lr_super);
3097                                         mutex_lock(&eli->li_list_mtx);
3098                                         n = pos->next;
3099                                 }
3100                                 up_read((&elr->lr_super->s_umount));
3101                         }
3102                         /* error, remove the lazy_init job */
3103                         if (err) {
3104                                 ext4_remove_li_request(elr);
3105                                 continue;
3106                         }
3107                         if (!progress) {
3108                                 elr->lr_next_sched = jiffies +
3109                                         (prandom_u32()
3110                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3111                         }
3112                         if (time_before(elr->lr_next_sched, next_wakeup))
3113                                 next_wakeup = elr->lr_next_sched;
3114                 }
3115                 mutex_unlock(&eli->li_list_mtx);
3116
3117                 try_to_freeze();
3118
3119                 cur = jiffies;
3120                 if ((time_after_eq(cur, next_wakeup)) ||
3121                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3122                         cond_resched();
3123                         continue;
3124                 }
3125
3126                 schedule_timeout_interruptible(next_wakeup - cur);
3127
3128                 if (kthread_should_stop()) {
3129                         ext4_clear_request_list();
3130                         goto exit_thread;
3131                 }
3132         }
3133
3134 exit_thread:
3135         /*
3136          * It looks like the request list is empty, but we need
3137          * to check it under the li_list_mtx lock, to prevent any
3138          * additions into it, and of course we should lock ext4_li_mtx
3139          * to atomically free the list and ext4_li_info, because at
3140          * this point another ext4 filesystem could be registering
3141          * new one.
3142          */
3143         mutex_lock(&ext4_li_mtx);
3144         mutex_lock(&eli->li_list_mtx);
3145         if (!list_empty(&eli->li_request_list)) {
3146                 mutex_unlock(&eli->li_list_mtx);
3147                 mutex_unlock(&ext4_li_mtx);
3148                 goto cont_thread;
3149         }
3150         mutex_unlock(&eli->li_list_mtx);
3151         kfree(ext4_li_info);
3152         ext4_li_info = NULL;
3153         mutex_unlock(&ext4_li_mtx);
3154
3155         return 0;
3156 }
3157
3158 static void ext4_clear_request_list(void)
3159 {
3160         struct list_head *pos, *n;
3161         struct ext4_li_request *elr;
3162
3163         mutex_lock(&ext4_li_info->li_list_mtx);
3164         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3165                 elr = list_entry(pos, struct ext4_li_request,
3166                                  lr_request);
3167                 ext4_remove_li_request(elr);
3168         }
3169         mutex_unlock(&ext4_li_info->li_list_mtx);
3170 }
3171
3172 static int ext4_run_lazyinit_thread(void)
3173 {
3174         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3175                                          ext4_li_info, "ext4lazyinit");
3176         if (IS_ERR(ext4_lazyinit_task)) {
3177                 int err = PTR_ERR(ext4_lazyinit_task);
3178                 ext4_clear_request_list();
3179                 kfree(ext4_li_info);
3180                 ext4_li_info = NULL;
3181                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3182                                  "initialization thread\n",
3183                                  err);
3184                 return err;
3185         }
3186         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3187         return 0;
3188 }
3189
3190 /*
3191  * Check whether it make sense to run itable init. thread or not.
3192  * If there is at least one uninitialized inode table, return
3193  * corresponding group number, else the loop goes through all
3194  * groups and return total number of groups.
3195  */
3196 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3197 {
3198         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3199         struct ext4_group_desc *gdp = NULL;
3200
3201         if (!ext4_has_group_desc_csum(sb))
3202                 return ngroups;
3203
3204         for (group = 0; group < ngroups; group++) {
3205                 gdp = ext4_get_group_desc(sb, group, NULL);
3206                 if (!gdp)
3207                         continue;
3208
3209                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3210                         break;
3211         }
3212
3213         return group;
3214 }
3215
3216 static int ext4_li_info_new(void)
3217 {
3218         struct ext4_lazy_init *eli = NULL;
3219
3220         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3221         if (!eli)
3222                 return -ENOMEM;
3223
3224         INIT_LIST_HEAD(&eli->li_request_list);
3225         mutex_init(&eli->li_list_mtx);
3226
3227         eli->li_state |= EXT4_LAZYINIT_QUIT;
3228
3229         ext4_li_info = eli;
3230
3231         return 0;
3232 }
3233
3234 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3235                                             ext4_group_t start)
3236 {
3237         struct ext4_sb_info *sbi = EXT4_SB(sb);
3238         struct ext4_li_request *elr;
3239
3240         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3241         if (!elr)
3242                 return NULL;
3243
3244         elr->lr_super = sb;
3245         elr->lr_sbi = sbi;
3246         elr->lr_next_group = start;
3247
3248         /*
3249          * Randomize first schedule time of the request to
3250          * spread the inode table initialization requests
3251          * better.
3252          */
3253         elr->lr_next_sched = jiffies + (prandom_u32() %
3254                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3255         return elr;
3256 }
3257
3258 int ext4_register_li_request(struct super_block *sb,
3259                              ext4_group_t first_not_zeroed)
3260 {
3261         struct ext4_sb_info *sbi = EXT4_SB(sb);
3262         struct ext4_li_request *elr = NULL;
3263         ext4_group_t ngroups = sbi->s_groups_count;
3264         int ret = 0;
3265
3266         mutex_lock(&ext4_li_mtx);
3267         if (sbi->s_li_request != NULL) {
3268                 /*
3269                  * Reset timeout so it can be computed again, because
3270                  * s_li_wait_mult might have changed.
3271                  */
3272                 sbi->s_li_request->lr_timeout = 0;
3273                 goto out;
3274         }
3275
3276         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3277             !test_opt(sb, INIT_INODE_TABLE))
3278                 goto out;
3279
3280         elr = ext4_li_request_new(sb, first_not_zeroed);
3281         if (!elr) {
3282                 ret = -ENOMEM;
3283                 goto out;
3284         }
3285
3286         if (NULL == ext4_li_info) {
3287                 ret = ext4_li_info_new();
3288                 if (ret)
3289                         goto out;
3290         }
3291
3292         mutex_lock(&ext4_li_info->li_list_mtx);
3293         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3294         mutex_unlock(&ext4_li_info->li_list_mtx);
3295
3296         sbi->s_li_request = elr;
3297         /*
3298          * set elr to NULL here since it has been inserted to
3299          * the request_list and the removal and free of it is
3300          * handled by ext4_clear_request_list from now on.
3301          */
3302         elr = NULL;
3303
3304         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3305                 ret = ext4_run_lazyinit_thread();
3306                 if (ret)
3307                         goto out;
3308         }
3309 out:
3310         mutex_unlock(&ext4_li_mtx);
3311         if (ret)
3312                 kfree(elr);
3313         return ret;
3314 }
3315
3316 /*
3317  * We do not need to lock anything since this is called on
3318  * module unload.
3319  */
3320 static void ext4_destroy_lazyinit_thread(void)
3321 {
3322         /*
3323          * If thread exited earlier
3324          * there's nothing to be done.
3325          */
3326         if (!ext4_li_info || !ext4_lazyinit_task)
3327                 return;
3328
3329         kthread_stop(ext4_lazyinit_task);
3330 }
3331
3332 static int set_journal_csum_feature_set(struct super_block *sb)
3333 {
3334         int ret = 1;
3335         int compat, incompat;
3336         struct ext4_sb_info *sbi = EXT4_SB(sb);
3337
3338         if (ext4_has_metadata_csum(sb)) {
3339                 /* journal checksum v3 */
3340                 compat = 0;
3341                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3342         } else {
3343                 /* journal checksum v1 */
3344                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3345                 incompat = 0;
3346         }
3347
3348         jbd2_journal_clear_features(sbi->s_journal,
3349                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3350                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3351                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3352         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3353                 ret = jbd2_journal_set_features(sbi->s_journal,
3354                                 compat, 0,
3355                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3356                                 incompat);
3357         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3358                 ret = jbd2_journal_set_features(sbi->s_journal,
3359                                 compat, 0,
3360                                 incompat);
3361                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3362                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3363         } else {
3364                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3365                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3366         }
3367
3368         return ret;
3369 }
3370
3371 /*
3372  * Note: calculating the overhead so we can be compatible with
3373  * historical BSD practice is quite difficult in the face of
3374  * clusters/bigalloc.  This is because multiple metadata blocks from
3375  * different block group can end up in the same allocation cluster.
3376  * Calculating the exact overhead in the face of clustered allocation
3377  * requires either O(all block bitmaps) in memory or O(number of block
3378  * groups**2) in time.  We will still calculate the superblock for
3379  * older file systems --- and if we come across with a bigalloc file
3380  * system with zero in s_overhead_clusters the estimate will be close to
3381  * correct especially for very large cluster sizes --- but for newer
3382  * file systems, it's better to calculate this figure once at mkfs
3383  * time, and store it in the superblock.  If the superblock value is
3384  * present (even for non-bigalloc file systems), we will use it.
3385  */
3386 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3387                           char *buf)
3388 {
3389         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3390         struct ext4_group_desc  *gdp;
3391         ext4_fsblk_t            first_block, last_block, b;
3392         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3393         int                     s, j, count = 0;
3394
3395         if (!ext4_has_feature_bigalloc(sb))
3396                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3397                         sbi->s_itb_per_group + 2);
3398
3399         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3400                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3401         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3402         for (i = 0; i < ngroups; i++) {
3403                 gdp = ext4_get_group_desc(sb, i, NULL);
3404                 b = ext4_block_bitmap(sb, gdp);
3405                 if (b >= first_block && b <= last_block) {
3406                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3407                         count++;
3408                 }
3409                 b = ext4_inode_bitmap(sb, gdp);
3410                 if (b >= first_block && b <= last_block) {
3411                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3412                         count++;
3413                 }
3414                 b = ext4_inode_table(sb, gdp);
3415                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3416                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3417                                 int c = EXT4_B2C(sbi, b - first_block);
3418                                 ext4_set_bit(c, buf);
3419                                 count++;
3420                         }
3421                 if (i != grp)
3422                         continue;
3423                 s = 0;
3424                 if (ext4_bg_has_super(sb, grp)) {
3425                         ext4_set_bit(s++, buf);
3426                         count++;
3427                 }
3428                 j = ext4_bg_num_gdb(sb, grp);
3429                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3430                         ext4_error(sb, "Invalid number of block group "
3431                                    "descriptor blocks: %d", j);
3432                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3433                 }
3434                 count += j;
3435                 for (; j > 0; j--)
3436                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3437         }
3438         if (!count)
3439                 return 0;
3440         return EXT4_CLUSTERS_PER_GROUP(sb) -
3441                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3442 }
3443
3444 /*
3445  * Compute the overhead and stash it in sbi->s_overhead
3446  */
3447 int ext4_calculate_overhead(struct super_block *sb)
3448 {
3449         struct ext4_sb_info *sbi = EXT4_SB(sb);
3450         struct ext4_super_block *es = sbi->s_es;
3451         struct inode *j_inode;
3452         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3453         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3454         ext4_fsblk_t overhead = 0;
3455         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3456
3457         if (!buf)
3458                 return -ENOMEM;
3459
3460         /*
3461          * Compute the overhead (FS structures).  This is constant
3462          * for a given filesystem unless the number of block groups
3463          * changes so we cache the previous value until it does.
3464          */
3465
3466         /*
3467          * All of the blocks before first_data_block are overhead
3468          */
3469         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3470
3471         /*
3472          * Add the overhead found in each block group
3473          */
3474         for (i = 0; i < ngroups; i++) {
3475                 int blks;
3476
3477                 blks = count_overhead(sb, i, buf);
3478                 overhead += blks;
3479                 if (blks)
3480                         memset(buf, 0, PAGE_SIZE);
3481                 cond_resched();
3482         }
3483
3484         /*
3485          * Add the internal journal blocks whether the journal has been
3486          * loaded or not
3487          */
3488         if (sbi->s_journal && !sbi->journal_bdev)
3489                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3490         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3491                 j_inode = ext4_get_journal_inode(sb, j_inum);
3492                 if (j_inode) {
3493                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3494                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3495                         iput(j_inode);
3496                 } else {
3497                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3498                 }
3499         }
3500         sbi->s_overhead = overhead;
3501         smp_wmb();
3502         free_page((unsigned long) buf);
3503         return 0;
3504 }
3505
3506 static void ext4_set_resv_clusters(struct super_block *sb)
3507 {
3508         ext4_fsblk_t resv_clusters;
3509         struct ext4_sb_info *sbi = EXT4_SB(sb);
3510
3511         /*
3512          * There's no need to reserve anything when we aren't using extents.
3513          * The space estimates are exact, there are no unwritten extents,
3514          * hole punching doesn't need new metadata... This is needed especially
3515          * to keep ext2/3 backward compatibility.
3516          */
3517         if (!ext4_has_feature_extents(sb))
3518                 return;
3519         /*
3520          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3521          * This should cover the situations where we can not afford to run
3522          * out of space like for example punch hole, or converting
3523          * unwritten extents in delalloc path. In most cases such
3524          * allocation would require 1, or 2 blocks, higher numbers are
3525          * very rare.
3526          */
3527         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3528                          sbi->s_cluster_bits);
3529
3530         do_div(resv_clusters, 50);
3531         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3532
3533         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3534 }
3535
3536 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3537 {
3538         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3539         char *orig_data = kstrdup(data, GFP_KERNEL);
3540         struct buffer_head *bh;
3541         struct ext4_super_block *es = NULL;
3542         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3543         ext4_fsblk_t block;
3544         ext4_fsblk_t sb_block = get_sb_block(&data);
3545         ext4_fsblk_t logical_sb_block;
3546         unsigned long offset = 0;
3547         unsigned long journal_devnum = 0;
3548         unsigned long def_mount_opts;
3549         struct inode *root;
3550         const char *descr;
3551         int ret = -ENOMEM;
3552         int blocksize, clustersize;
3553         unsigned int db_count;
3554         unsigned int i;
3555         int needs_recovery, has_huge_files, has_bigalloc;
3556         __u64 blocks_count;
3557         int err = 0;
3558         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3559         ext4_group_t first_not_zeroed;
3560
3561         if ((data && !orig_data) || !sbi)
3562                 goto out_free_base;
3563
3564         sbi->s_daxdev = dax_dev;
3565         sbi->s_blockgroup_lock =
3566                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3567         if (!sbi->s_blockgroup_lock)
3568                 goto out_free_base;
3569
3570         sb->s_fs_info = sbi;
3571         sbi->s_sb = sb;
3572         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3573         sbi->s_sb_block = sb_block;
3574         if (sb->s_bdev->bd_part)
3575                 sbi->s_sectors_written_start =
3576                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3577
3578         /* Cleanup superblock name */
3579         strreplace(sb->s_id, '/', '!');
3580
3581         /* -EINVAL is default */
3582         ret = -EINVAL;
3583         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3584         if (!blocksize) {
3585                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3586                 goto out_fail;
3587         }
3588
3589         /*
3590          * The ext4 superblock will not be buffer aligned for other than 1kB
3591          * block sizes.  We need to calculate the offset from buffer start.
3592          */
3593         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3594                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3595                 offset = do_div(logical_sb_block, blocksize);
3596         } else {
3597                 logical_sb_block = sb_block;
3598         }
3599
3600         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3601                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3602                 goto out_fail;
3603         }
3604         /*
3605          * Note: s_es must be initialized as soon as possible because
3606          *       some ext4 macro-instructions depend on its value
3607          */
3608         es = (struct ext4_super_block *) (bh->b_data + offset);
3609         sbi->s_es = es;
3610         sb->s_magic = le16_to_cpu(es->s_magic);
3611         if (sb->s_magic != EXT4_SUPER_MAGIC)
3612                 goto cantfind_ext4;
3613         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3614
3615         /* Warn if metadata_csum and gdt_csum are both set. */
3616         if (ext4_has_feature_metadata_csum(sb) &&
3617             ext4_has_feature_gdt_csum(sb))
3618                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3619                              "redundant flags; please run fsck.");
3620
3621         /* Check for a known checksum algorithm */
3622         if (!ext4_verify_csum_type(sb, es)) {
3623                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3624                          "unknown checksum algorithm.");
3625                 silent = 1;
3626                 goto cantfind_ext4;
3627         }
3628
3629         /* Load the checksum driver */
3630         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3631         if (IS_ERR(sbi->s_chksum_driver)) {
3632                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3633                 ret = PTR_ERR(sbi->s_chksum_driver);
3634                 sbi->s_chksum_driver = NULL;
3635                 goto failed_mount;
3636         }
3637
3638         /* Check superblock checksum */
3639         if (!ext4_superblock_csum_verify(sb, es)) {
3640                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3641                          "invalid superblock checksum.  Run e2fsck?");
3642                 silent = 1;
3643                 ret = -EFSBADCRC;
3644                 goto cantfind_ext4;
3645         }
3646
3647         /* Precompute checksum seed for all metadata */
3648         if (ext4_has_feature_csum_seed(sb))
3649                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3650         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3651                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3652                                                sizeof(es->s_uuid));
3653
3654         /* Set defaults before we parse the mount options */
3655         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3656         set_opt(sb, INIT_INODE_TABLE);
3657         if (def_mount_opts & EXT4_DEFM_DEBUG)
3658                 set_opt(sb, DEBUG);
3659         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3660                 set_opt(sb, GRPID);
3661         if (def_mount_opts & EXT4_DEFM_UID16)
3662                 set_opt(sb, NO_UID32);
3663         /* xattr user namespace & acls are now defaulted on */
3664         set_opt(sb, XATTR_USER);
3665 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3666         set_opt(sb, POSIX_ACL);
3667 #endif
3668         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3669         if (ext4_has_metadata_csum(sb))
3670                 set_opt(sb, JOURNAL_CHECKSUM);
3671
3672         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3673                 set_opt(sb, JOURNAL_DATA);
3674         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3675                 set_opt(sb, ORDERED_DATA);
3676         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3677                 set_opt(sb, WRITEBACK_DATA);
3678
3679         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3680                 set_opt(sb, ERRORS_PANIC);
3681         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3682                 set_opt(sb, ERRORS_CONT);
3683         else
3684                 set_opt(sb, ERRORS_RO);
3685         /* block_validity enabled by default; disable with noblock_validity */
3686         set_opt(sb, BLOCK_VALIDITY);
3687         if (def_mount_opts & EXT4_DEFM_DISCARD)
3688                 set_opt(sb, DISCARD);
3689
3690         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3691         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3692         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3693         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3694         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3695
3696         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3697                 set_opt(sb, BARRIER);
3698
3699         /*
3700          * enable delayed allocation by default
3701          * Use -o nodelalloc to turn it off
3702          */
3703         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3704             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3705                 set_opt(sb, DELALLOC);
3706
3707         /*
3708          * set default s_li_wait_mult for lazyinit, for the case there is
3709          * no mount option specified.
3710          */
3711         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3712
3713         if (sbi->s_es->s_mount_opts[0]) {
3714                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3715                                               sizeof(sbi->s_es->s_mount_opts),
3716                                               GFP_KERNEL);
3717                 if (!s_mount_opts)
3718                         goto failed_mount;
3719                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3720                                    &journal_ioprio, 0)) {
3721                         ext4_msg(sb, KERN_WARNING,
3722                                  "failed to parse options in superblock: %s",
3723                                  s_mount_opts);
3724                 }
3725                 kfree(s_mount_opts);
3726         }
3727         sbi->s_def_mount_opt = sbi->s_mount_opt;
3728         if (!parse_options((char *) data, sb, &journal_devnum,
3729                            &journal_ioprio, 0))
3730                 goto failed_mount;
3731
3732         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3733                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3734                             "with data=journal disables delayed "
3735                             "allocation and O_DIRECT support!\n");
3736                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3737                         ext4_msg(sb, KERN_ERR, "can't mount with "
3738                                  "both data=journal and delalloc");
3739                         goto failed_mount;
3740                 }
3741                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3742                         ext4_msg(sb, KERN_ERR, "can't mount with "
3743                                  "both data=journal and dioread_nolock");
3744                         goto failed_mount;
3745                 }
3746                 if (test_opt(sb, DAX)) {
3747                         ext4_msg(sb, KERN_ERR, "can't mount with "
3748                                  "both data=journal and dax");
3749                         goto failed_mount;
3750                 }
3751                 if (ext4_has_feature_encrypt(sb)) {
3752                         ext4_msg(sb, KERN_WARNING,
3753                                  "encrypted files will use data=ordered "
3754                                  "instead of data journaling mode");
3755                 }
3756                 if (test_opt(sb, DELALLOC))
3757                         clear_opt(sb, DELALLOC);
3758         } else {
3759                 sb->s_iflags |= SB_I_CGROUPWB;
3760         }
3761
3762         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3763                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3764
3765         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3766             (ext4_has_compat_features(sb) ||
3767              ext4_has_ro_compat_features(sb) ||
3768              ext4_has_incompat_features(sb)))
3769                 ext4_msg(sb, KERN_WARNING,
3770                        "feature flags set on rev 0 fs, "
3771                        "running e2fsck is recommended");
3772
3773         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3774                 set_opt2(sb, HURD_COMPAT);
3775                 if (ext4_has_feature_64bit(sb)) {
3776                         ext4_msg(sb, KERN_ERR,
3777                                  "The Hurd can't support 64-bit file systems");
3778                         goto failed_mount;
3779                 }
3780
3781                 /*
3782                  * ea_inode feature uses l_i_version field which is not
3783                  * available in HURD_COMPAT mode.
3784                  */
3785                 if (ext4_has_feature_ea_inode(sb)) {
3786                         ext4_msg(sb, KERN_ERR,
3787                                  "ea_inode feature is not supported for Hurd");
3788                         goto failed_mount;
3789                 }
3790         }
3791
3792         if (IS_EXT2_SB(sb)) {
3793                 if (ext2_feature_set_ok(sb))
3794                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3795                                  "using the ext4 subsystem");
3796                 else {
3797                         /*
3798                          * If we're probing be silent, if this looks like
3799                          * it's actually an ext[34] filesystem.
3800                          */
3801                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3802                                 goto failed_mount;
3803                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3804                                  "to feature incompatibilities");
3805                         goto failed_mount;
3806                 }
3807         }
3808
3809         if (IS_EXT3_SB(sb)) {
3810                 if (ext3_feature_set_ok(sb))
3811                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3812                                  "using the ext4 subsystem");
3813                 else {
3814                         /*
3815                          * If we're probing be silent, if this looks like
3816                          * it's actually an ext4 filesystem.
3817                          */
3818                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3819                                 goto failed_mount;
3820                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3821                                  "to feature incompatibilities");
3822                         goto failed_mount;
3823                 }
3824         }
3825
3826         /*
3827          * Check feature flags regardless of the revision level, since we
3828          * previously didn't change the revision level when setting the flags,
3829          * so there is a chance incompat flags are set on a rev 0 filesystem.
3830          */
3831         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3832                 goto failed_mount;
3833
3834         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3835         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3836             blocksize > EXT4_MAX_BLOCK_SIZE) {
3837                 ext4_msg(sb, KERN_ERR,
3838                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3839                          blocksize, le32_to_cpu(es->s_log_block_size));
3840                 goto failed_mount;
3841         }
3842         if (le32_to_cpu(es->s_log_block_size) >
3843             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3844                 ext4_msg(sb, KERN_ERR,
3845                          "Invalid log block size: %u",
3846                          le32_to_cpu(es->s_log_block_size));
3847                 goto failed_mount;
3848         }
3849         if (le32_to_cpu(es->s_log_cluster_size) >
3850             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3851                 ext4_msg(sb, KERN_ERR,
3852                          "Invalid log cluster size: %u",
3853                          le32_to_cpu(es->s_log_cluster_size));
3854                 goto failed_mount;
3855         }
3856
3857         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3858                 ext4_msg(sb, KERN_ERR,
3859                          "Number of reserved GDT blocks insanely large: %d",
3860                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3861                 goto failed_mount;
3862         }
3863
3864         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3865                 if (ext4_has_feature_inline_data(sb)) {
3866                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3867                                         " that may contain inline data");
3868                         goto failed_mount;
3869                 }
3870                 if (!bdev_dax_supported(sb->s_bdev, blocksize)) {
3871                         ext4_msg(sb, KERN_ERR,
3872                                 "DAX unsupported by block device.");
3873                         goto failed_mount;
3874                 }
3875         }
3876
3877         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3878                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3879                          es->s_encryption_level);
3880                 goto failed_mount;
3881         }
3882
3883         if (sb->s_blocksize != blocksize) {
3884                 /* Validate the filesystem blocksize */
3885                 if (!sb_set_blocksize(sb, blocksize)) {
3886                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3887                                         blocksize);
3888                         goto failed_mount;
3889                 }
3890
3891                 brelse(bh);
3892                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3893                 offset = do_div(logical_sb_block, blocksize);
3894                 bh = sb_bread_unmovable(sb, logical_sb_block);
3895                 if (!bh) {
3896                         ext4_msg(sb, KERN_ERR,
3897                                "Can't read superblock on 2nd try");
3898                         goto failed_mount;
3899                 }
3900                 es = (struct ext4_super_block *)(bh->b_data + offset);
3901                 sbi->s_es = es;
3902                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3903                         ext4_msg(sb, KERN_ERR,
3904                                "Magic mismatch, very weird!");
3905                         goto failed_mount;
3906                 }
3907         }
3908
3909         has_huge_files = ext4_has_feature_huge_file(sb);
3910         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3911                                                       has_huge_files);
3912         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3913
3914         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3915                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3916                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3917         } else {
3918                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3919                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3920                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3921                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3922                                  sbi->s_first_ino);
3923                         goto failed_mount;
3924                 }
3925                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3926                     (!is_power_of_2(sbi->s_inode_size)) ||
3927                     (sbi->s_inode_size > blocksize)) {
3928                         ext4_msg(sb, KERN_ERR,
3929                                "unsupported inode size: %d",
3930                                sbi->s_inode_size);
3931                         goto failed_mount;
3932                 }
3933                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3934                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3935         }
3936
3937         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3938         if (ext4_has_feature_64bit(sb)) {
3939                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3940                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3941                     !is_power_of_2(sbi->s_desc_size)) {
3942                         ext4_msg(sb, KERN_ERR,
3943                                "unsupported descriptor size %lu",
3944                                sbi->s_desc_size);
3945                         goto failed_mount;
3946                 }
3947         } else
3948                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3949
3950         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3951         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3952
3953         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3954         if (sbi->s_inodes_per_block == 0)
3955                 goto cantfind_ext4;
3956         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3957             sbi->s_inodes_per_group > blocksize * 8) {
3958                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3959                          sbi->s_blocks_per_group);
3960                 goto failed_mount;
3961         }
3962         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3963                                         sbi->s_inodes_per_block;
3964         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3965         sbi->s_sbh = bh;
3966         sbi->s_mount_state = le16_to_cpu(es->s_state);
3967         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3968         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3969
3970         for (i = 0; i < 4; i++)
3971                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3972         sbi->s_def_hash_version = es->s_def_hash_version;
3973         if (ext4_has_feature_dir_index(sb)) {
3974                 i = le32_to_cpu(es->s_flags);
3975                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3976                         sbi->s_hash_unsigned = 3;
3977                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3978 #ifdef __CHAR_UNSIGNED__
3979                         if (!sb_rdonly(sb))
3980                                 es->s_flags |=
3981                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3982                         sbi->s_hash_unsigned = 3;
3983 #else
3984                         if (!sb_rdonly(sb))
3985                                 es->s_flags |=
3986                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3987 #endif
3988                 }
3989         }
3990
3991         /* Handle clustersize */
3992         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3993         has_bigalloc = ext4_has_feature_bigalloc(sb);
3994         if (has_bigalloc) {
3995                 if (clustersize < blocksize) {
3996                         ext4_msg(sb, KERN_ERR,
3997                                  "cluster size (%d) smaller than "
3998                                  "block size (%d)", clustersize, blocksize);
3999                         goto failed_mount;
4000                 }
4001                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4002                         le32_to_cpu(es->s_log_block_size);
4003                 sbi->s_clusters_per_group =
4004                         le32_to_cpu(es->s_clusters_per_group);
4005                 if (sbi->s_clusters_per_group > blocksize * 8) {
4006                         ext4_msg(sb, KERN_ERR,
4007                                  "#clusters per group too big: %lu",
4008                                  sbi->s_clusters_per_group);
4009                         goto failed_mount;
4010                 }
4011                 if (sbi->s_blocks_per_group !=
4012                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4013                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4014                                  "clusters per group (%lu) inconsistent",
4015                                  sbi->s_blocks_per_group,
4016                                  sbi->s_clusters_per_group);
4017                         goto failed_mount;
4018                 }
4019         } else {
4020                 if (clustersize != blocksize) {
4021                         ext4_msg(sb, KERN_ERR,
4022                                  "fragment/cluster size (%d) != "
4023                                  "block size (%d)", clustersize, blocksize);
4024                         goto failed_mount;
4025                 }
4026                 if (sbi->s_blocks_per_group > blocksize * 8) {
4027                         ext4_msg(sb, KERN_ERR,
4028                                  "#blocks per group too big: %lu",
4029                                  sbi->s_blocks_per_group);
4030                         goto failed_mount;
4031                 }
4032                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4033                 sbi->s_cluster_bits = 0;
4034         }
4035         sbi->s_cluster_ratio = clustersize / blocksize;
4036
4037         /* Do we have standard group size of clustersize * 8 blocks ? */
4038         if (sbi->s_blocks_per_group == clustersize << 3)
4039                 set_opt2(sb, STD_GROUP_SIZE);
4040
4041         /*
4042          * Test whether we have more sectors than will fit in sector_t,
4043          * and whether the max offset is addressable by the page cache.
4044          */
4045         err = generic_check_addressable(sb->s_blocksize_bits,
4046                                         ext4_blocks_count(es));
4047         if (err) {
4048                 ext4_msg(sb, KERN_ERR, "filesystem"
4049                          " too large to mount safely on this system");
4050                 if (sizeof(sector_t) < 8)
4051                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
4052                 goto failed_mount;
4053         }
4054
4055         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4056                 goto cantfind_ext4;
4057
4058         /* check blocks count against device size */
4059         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4060         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4061                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4062                        "exceeds size of device (%llu blocks)",
4063                        ext4_blocks_count(es), blocks_count);
4064                 goto failed_mount;
4065         }
4066
4067         /*
4068          * It makes no sense for the first data block to be beyond the end
4069          * of the filesystem.
4070          */
4071         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4072                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4073                          "block %u is beyond end of filesystem (%llu)",
4074                          le32_to_cpu(es->s_first_data_block),
4075                          ext4_blocks_count(es));
4076                 goto failed_mount;
4077         }
4078         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4079             (sbi->s_cluster_ratio == 1)) {
4080                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4081                          "block is 0 with a 1k block and cluster size");
4082                 goto failed_mount;
4083         }
4084
4085         blocks_count = (ext4_blocks_count(es) -
4086                         le32_to_cpu(es->s_first_data_block) +
4087                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4088         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4089         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4090                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
4091                        "(block count %llu, first data block %u, "
4092                        "blocks per group %lu)", sbi->s_groups_count,
4093                        ext4_blocks_count(es),
4094                        le32_to_cpu(es->s_first_data_block),
4095                        EXT4_BLOCKS_PER_GROUP(sb));
4096                 goto failed_mount;
4097         }
4098         sbi->s_groups_count = blocks_count;
4099         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4100                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4101         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4102             le32_to_cpu(es->s_inodes_count)) {
4103                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4104                          le32_to_cpu(es->s_inodes_count),
4105                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4106                 ret = -EINVAL;
4107                 goto failed_mount;
4108         }
4109         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4110                    EXT4_DESC_PER_BLOCK(sb);
4111         if (ext4_has_feature_meta_bg(sb)) {
4112                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4113                         ext4_msg(sb, KERN_WARNING,
4114                                  "first meta block group too large: %u "
4115                                  "(group descriptor block count %u)",
4116                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4117                         goto failed_mount;
4118                 }
4119         }
4120         sbi->s_group_desc = kvmalloc_array(db_count,
4121                                            sizeof(struct buffer_head *),
4122                                            GFP_KERNEL);
4123         if (sbi->s_group_desc == NULL) {
4124                 ext4_msg(sb, KERN_ERR, "not enough memory");
4125                 ret = -ENOMEM;
4126                 goto failed_mount;
4127         }
4128
4129         bgl_lock_init(sbi->s_blockgroup_lock);
4130
4131         /* Pre-read the descriptors into the buffer cache */
4132         for (i = 0; i < db_count; i++) {
4133                 block = descriptor_loc(sb, logical_sb_block, i);
4134                 sb_breadahead(sb, block);
4135         }
4136
4137         for (i = 0; i < db_count; i++) {
4138                 block = descriptor_loc(sb, logical_sb_block, i);
4139                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
4140                 if (!sbi->s_group_desc[i]) {
4141                         ext4_msg(sb, KERN_ERR,
4142                                "can't read group descriptor %d", i);
4143                         db_count = i;
4144                         goto failed_mount2;
4145                 }
4146         }
4147         sbi->s_gdb_count = db_count;
4148         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4149                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4150                 ret = -EFSCORRUPTED;
4151                 goto failed_mount2;
4152         }
4153
4154         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4155
4156         /* Register extent status tree shrinker */
4157         if (ext4_es_register_shrinker(sbi))
4158                 goto failed_mount3;
4159
4160         sbi->s_stripe = ext4_get_stripe_size(sbi);
4161         sbi->s_extent_max_zeroout_kb = 32;
4162
4163         /*
4164          * set up enough so that it can read an inode
4165          */
4166         sb->s_op = &ext4_sops;
4167         sb->s_export_op = &ext4_export_ops;
4168         sb->s_xattr = ext4_xattr_handlers;
4169 #ifdef CONFIG_FS_ENCRYPTION
4170         sb->s_cop = &ext4_cryptops;
4171 #endif
4172 #ifdef CONFIG_QUOTA
4173         sb->dq_op = &ext4_quota_operations;
4174         if (ext4_has_feature_quota(sb))
4175                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4176         else
4177                 sb->s_qcop = &ext4_qctl_operations;
4178         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4179 #endif
4180         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4181
4182         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4183         mutex_init(&sbi->s_orphan_lock);
4184
4185         sb->s_root = NULL;
4186
4187         needs_recovery = (es->s_last_orphan != 0 ||
4188                           ext4_has_feature_journal_needs_recovery(sb));
4189
4190         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4191                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4192                         goto failed_mount3a;
4193
4194         /*
4195          * The first inode we look at is the journal inode.  Don't try
4196          * root first: it may be modified in the journal!
4197          */
4198         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4199                 err = ext4_load_journal(sb, es, journal_devnum);
4200                 if (err)
4201                         goto failed_mount3a;
4202         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4203                    ext4_has_feature_journal_needs_recovery(sb)) {
4204                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4205                        "suppressed and not mounted read-only");
4206                 goto failed_mount_wq;
4207         } else {
4208                 /* Nojournal mode, all journal mount options are illegal */
4209                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4210                         ext4_msg(sb, KERN_ERR, "can't mount with "
4211                                  "journal_checksum, fs mounted w/o journal");
4212                         goto failed_mount_wq;
4213                 }
4214                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4215                         ext4_msg(sb, KERN_ERR, "can't mount with "
4216                                  "journal_async_commit, fs mounted w/o journal");
4217                         goto failed_mount_wq;
4218                 }
4219                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4220                         ext4_msg(sb, KERN_ERR, "can't mount with "
4221                                  "commit=%lu, fs mounted w/o journal",
4222                                  sbi->s_commit_interval / HZ);
4223                         goto failed_mount_wq;
4224                 }
4225                 if (EXT4_MOUNT_DATA_FLAGS &
4226                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4227                         ext4_msg(sb, KERN_ERR, "can't mount with "
4228                                  "data=, fs mounted w/o journal");
4229                         goto failed_mount_wq;
4230                 }
4231                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4232                 clear_opt(sb, JOURNAL_CHECKSUM);
4233                 clear_opt(sb, DATA_FLAGS);
4234                 sbi->s_journal = NULL;
4235                 needs_recovery = 0;
4236                 goto no_journal;
4237         }
4238
4239         if (ext4_has_feature_64bit(sb) &&
4240             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4241                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4242                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4243                 goto failed_mount_wq;
4244         }
4245
4246         if (!set_journal_csum_feature_set(sb)) {
4247                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4248                          "feature set");
4249                 goto failed_mount_wq;
4250         }
4251
4252         /* We have now updated the journal if required, so we can
4253          * validate the data journaling mode. */
4254         switch (test_opt(sb, DATA_FLAGS)) {
4255         case 0:
4256                 /* No mode set, assume a default based on the journal
4257                  * capabilities: ORDERED_DATA if the journal can
4258                  * cope, else JOURNAL_DATA
4259                  */
4260                 if (jbd2_journal_check_available_features
4261                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4262                         set_opt(sb, ORDERED_DATA);
4263                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4264                 } else {
4265                         set_opt(sb, JOURNAL_DATA);
4266                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4267                 }
4268                 break;
4269
4270         case EXT4_MOUNT_ORDERED_DATA:
4271         case EXT4_MOUNT_WRITEBACK_DATA:
4272                 if (!jbd2_journal_check_available_features
4273                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4274                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4275                                "requested data journaling mode");
4276                         goto failed_mount_wq;
4277                 }
4278         default:
4279                 break;
4280         }
4281
4282         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4283             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4284                 ext4_msg(sb, KERN_ERR, "can't mount with "
4285                         "journal_async_commit in data=ordered mode");
4286                 goto failed_mount_wq;
4287         }
4288
4289         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4290
4291         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4292
4293 no_journal:
4294         if (!test_opt(sb, NO_MBCACHE)) {
4295                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4296                 if (!sbi->s_ea_block_cache) {
4297                         ext4_msg(sb, KERN_ERR,
4298                                  "Failed to create ea_block_cache");
4299                         goto failed_mount_wq;
4300                 }
4301
4302                 if (ext4_has_feature_ea_inode(sb)) {
4303                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4304                         if (!sbi->s_ea_inode_cache) {
4305                                 ext4_msg(sb, KERN_ERR,
4306                                          "Failed to create ea_inode_cache");
4307                                 goto failed_mount_wq;
4308                         }
4309                 }
4310         }
4311
4312         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4313             (blocksize != PAGE_SIZE)) {
4314                 ext4_msg(sb, KERN_ERR,
4315                          "Unsupported blocksize for fs encryption");
4316                 goto failed_mount_wq;
4317         }
4318
4319         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4320             !ext4_has_feature_encrypt(sb)) {
4321                 ext4_set_feature_encrypt(sb);
4322                 ext4_commit_super(sb, 1);
4323         }
4324
4325         /*
4326          * Get the # of file system overhead blocks from the
4327          * superblock if present.
4328          */
4329         if (es->s_overhead_clusters)
4330                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4331         else {
4332                 err = ext4_calculate_overhead(sb);
4333                 if (err)
4334                         goto failed_mount_wq;
4335         }
4336
4337         /*
4338          * The maximum number of concurrent works can be high and
4339          * concurrency isn't really necessary.  Limit it to 1.
4340          */
4341         EXT4_SB(sb)->rsv_conversion_wq =
4342                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4343         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4344                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4345                 ret = -ENOMEM;
4346                 goto failed_mount4;
4347         }
4348
4349         /*
4350          * The jbd2_journal_load will have done any necessary log recovery,
4351          * so we can safely mount the rest of the filesystem now.
4352          */
4353
4354         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4355         if (IS_ERR(root)) {
4356                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4357                 ret = PTR_ERR(root);
4358                 root = NULL;
4359                 goto failed_mount4;
4360         }
4361         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4362                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4363                 iput(root);
4364                 goto failed_mount4;
4365         }
4366         sb->s_root = d_make_root(root);
4367         if (!sb->s_root) {
4368                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4369                 ret = -ENOMEM;
4370                 goto failed_mount4;
4371         }
4372
4373         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4374         if (ret == -EROFS) {
4375                 sb->s_flags |= SB_RDONLY;
4376                 ret = 0;
4377         } else if (ret)
4378                 goto failed_mount4a;
4379
4380         /* determine the minimum size of new large inodes, if present */
4381         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4382             sbi->s_want_extra_isize == 0) {
4383                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4384                                                      EXT4_GOOD_OLD_INODE_SIZE;
4385                 if (ext4_has_feature_extra_isize(sb)) {
4386                         if (sbi->s_want_extra_isize <
4387                             le16_to_cpu(es->s_want_extra_isize))
4388                                 sbi->s_want_extra_isize =
4389                                         le16_to_cpu(es->s_want_extra_isize);
4390                         if (sbi->s_want_extra_isize <
4391                             le16_to_cpu(es->s_min_extra_isize))
4392                                 sbi->s_want_extra_isize =
4393                                         le16_to_cpu(es->s_min_extra_isize);
4394                 }
4395         }
4396         /* Check if enough inode space is available */
4397         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4398                                                         sbi->s_inode_size) {
4399                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4400                                                        EXT4_GOOD_OLD_INODE_SIZE;
4401                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4402                          "available");
4403         }
4404
4405         ext4_set_resv_clusters(sb);
4406
4407         err = ext4_setup_system_zone(sb);
4408         if (err) {
4409                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4410                          "zone (%d)", err);
4411                 goto failed_mount4a;
4412         }
4413
4414         ext4_ext_init(sb);
4415         err = ext4_mb_init(sb);
4416         if (err) {
4417                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4418                          err);
4419                 goto failed_mount5;
4420         }
4421
4422         block = ext4_count_free_clusters(sb);
4423         ext4_free_blocks_count_set(sbi->s_es,
4424                                    EXT4_C2B(sbi, block));
4425         ext4_superblock_csum_set(sb);
4426         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4427                                   GFP_KERNEL);
4428         if (!err) {
4429                 unsigned long freei = ext4_count_free_inodes(sb);
4430                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4431                 ext4_superblock_csum_set(sb);
4432                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4433                                           GFP_KERNEL);
4434         }
4435         if (!err)
4436                 err = percpu_counter_init(&sbi->s_dirs_counter,
4437                                           ext4_count_dirs(sb), GFP_KERNEL);
4438         if (!err)
4439                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4440                                           GFP_KERNEL);
4441         if (!err)
4442                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4443
4444         if (err) {
4445                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4446                 goto failed_mount6;
4447         }
4448
4449         if (ext4_has_feature_flex_bg(sb))
4450                 if (!ext4_fill_flex_info(sb)) {
4451                         ext4_msg(sb, KERN_ERR,
4452                                "unable to initialize "
4453                                "flex_bg meta info!");
4454                         goto failed_mount6;
4455                 }
4456
4457         err = ext4_register_li_request(sb, first_not_zeroed);
4458         if (err)
4459                 goto failed_mount6;
4460
4461         err = ext4_register_sysfs(sb);
4462         if (err)
4463                 goto failed_mount7;
4464
4465 #ifdef CONFIG_QUOTA
4466         /* Enable quota usage during mount. */
4467         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4468                 err = ext4_enable_quotas(sb);
4469                 if (err)
4470                         goto failed_mount8;
4471         }
4472 #endif  /* CONFIG_QUOTA */
4473
4474         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4475         ext4_orphan_cleanup(sb, es);
4476         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4477         if (needs_recovery) {
4478                 ext4_msg(sb, KERN_INFO, "recovery complete");
4479                 ext4_mark_recovery_complete(sb, es);
4480         }
4481         if (EXT4_SB(sb)->s_journal) {
4482                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4483                         descr = " journalled data mode";
4484                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4485                         descr = " ordered data mode";
4486                 else
4487                         descr = " writeback data mode";
4488         } else
4489                 descr = "out journal";
4490
4491         if (test_opt(sb, DISCARD)) {
4492                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4493                 if (!blk_queue_discard(q))
4494                         ext4_msg(sb, KERN_WARNING,
4495                                  "mounting with \"discard\" option, but "
4496                                  "the device does not support discard");
4497         }
4498
4499         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4500                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4501                          "Opts: %.*s%s%s", descr,
4502                          (int) sizeof(sbi->s_es->s_mount_opts),
4503                          sbi->s_es->s_mount_opts,
4504                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4505
4506         if (es->s_error_count)
4507                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4508
4509         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4510         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4511         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4512         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4513
4514         kfree(orig_data);
4515         return 0;
4516
4517 cantfind_ext4:
4518         if (!silent)
4519                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4520         goto failed_mount;
4521
4522 #ifdef CONFIG_QUOTA
4523 failed_mount8:
4524         ext4_unregister_sysfs(sb);
4525 #endif
4526 failed_mount7:
4527         ext4_unregister_li_request(sb);
4528 failed_mount6:
4529         ext4_mb_release(sb);
4530         if (sbi->s_flex_groups)
4531                 kvfree(sbi->s_flex_groups);
4532         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4533         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4534         percpu_counter_destroy(&sbi->s_dirs_counter);
4535         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4536         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
4537 failed_mount5:
4538         ext4_ext_release(sb);
4539         ext4_release_system_zone(sb);
4540 failed_mount4a:
4541         dput(sb->s_root);
4542         sb->s_root = NULL;
4543 failed_mount4:
4544         ext4_msg(sb, KERN_ERR, "mount failed");
4545         if (EXT4_SB(sb)->rsv_conversion_wq)
4546                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4547 failed_mount_wq:
4548         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4549         sbi->s_ea_inode_cache = NULL;
4550
4551         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4552         sbi->s_ea_block_cache = NULL;
4553
4554         if (sbi->s_journal) {
4555                 jbd2_journal_destroy(sbi->s_journal);
4556                 sbi->s_journal = NULL;
4557         }
4558 failed_mount3a:
4559         ext4_es_unregister_shrinker(sbi);
4560 failed_mount3:
4561         del_timer_sync(&sbi->s_err_report);
4562         if (sbi->s_mmp_tsk)
4563                 kthread_stop(sbi->s_mmp_tsk);
4564 failed_mount2:
4565         for (i = 0; i < db_count; i++)
4566                 brelse(sbi->s_group_desc[i]);
4567         kvfree(sbi->s_group_desc);
4568 failed_mount:
4569         if (sbi->s_chksum_driver)
4570                 crypto_free_shash(sbi->s_chksum_driver);
4571 #ifdef CONFIG_QUOTA
4572         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4573                 kfree(sbi->s_qf_names[i]);
4574 #endif
4575         ext4_blkdev_remove(sbi);
4576         brelse(bh);
4577 out_fail:
4578         sb->s_fs_info = NULL;
4579         kfree(sbi->s_blockgroup_lock);
4580 out_free_base:
4581         kfree(sbi);
4582         kfree(orig_data);
4583         fs_put_dax(dax_dev);
4584         return err ? err : ret;
4585 }
4586
4587 /*
4588  * Setup any per-fs journal parameters now.  We'll do this both on
4589  * initial mount, once the journal has been initialised but before we've
4590  * done any recovery; and again on any subsequent remount.
4591  */
4592 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4593 {
4594         struct ext4_sb_info *sbi = EXT4_SB(sb);
4595
4596         journal->j_commit_interval = sbi->s_commit_interval;
4597         journal->j_min_batch_time = sbi->s_min_batch_time;
4598         journal->j_max_batch_time = sbi->s_max_batch_time;
4599
4600         write_lock(&journal->j_state_lock);
4601         if (test_opt(sb, BARRIER))
4602                 journal->j_flags |= JBD2_BARRIER;
4603         else
4604                 journal->j_flags &= ~JBD2_BARRIER;
4605         if (test_opt(sb, DATA_ERR_ABORT))
4606                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4607         else
4608                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4609         write_unlock(&journal->j_state_lock);
4610 }
4611
4612 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4613                                              unsigned int journal_inum)
4614 {
4615         struct inode *journal_inode;
4616
4617         /*
4618          * Test for the existence of a valid inode on disk.  Bad things
4619          * happen if we iget() an unused inode, as the subsequent iput()
4620          * will try to delete it.
4621          */
4622         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4623         if (IS_ERR(journal_inode)) {
4624                 ext4_msg(sb, KERN_ERR, "no journal found");
4625                 return NULL;
4626         }
4627         if (!journal_inode->i_nlink) {
4628                 make_bad_inode(journal_inode);
4629                 iput(journal_inode);
4630                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4631                 return NULL;
4632         }
4633
4634         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4635                   journal_inode, journal_inode->i_size);
4636         if (!S_ISREG(journal_inode->i_mode)) {
4637                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4638                 iput(journal_inode);
4639                 return NULL;
4640         }
4641         return journal_inode;
4642 }
4643
4644 static journal_t *ext4_get_journal(struct super_block *sb,
4645                                    unsigned int journal_inum)
4646 {
4647         struct inode *journal_inode;
4648         journal_t *journal;
4649
4650         BUG_ON(!ext4_has_feature_journal(sb));
4651
4652         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4653         if (!journal_inode)
4654                 return NULL;
4655
4656         journal = jbd2_journal_init_inode(journal_inode);
4657         if (!journal) {
4658                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4659                 iput(journal_inode);
4660                 return NULL;
4661         }
4662         journal->j_private = sb;
4663         ext4_init_journal_params(sb, journal);
4664         return journal;
4665 }
4666
4667 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4668                                        dev_t j_dev)
4669 {
4670         struct buffer_head *bh;
4671         journal_t *journal;
4672         ext4_fsblk_t start;
4673         ext4_fsblk_t len;
4674         int hblock, blocksize;
4675         ext4_fsblk_t sb_block;
4676         unsigned long offset;
4677         struct ext4_super_block *es;
4678         struct block_device *bdev;
4679
4680         BUG_ON(!ext4_has_feature_journal(sb));
4681
4682         bdev = ext4_blkdev_get(j_dev, sb);
4683         if (bdev == NULL)
4684                 return NULL;
4685
4686         blocksize = sb->s_blocksize;
4687         hblock = bdev_logical_block_size(bdev);
4688         if (blocksize < hblock) {
4689                 ext4_msg(sb, KERN_ERR,
4690                         "blocksize too small for journal device");
4691                 goto out_bdev;
4692         }
4693
4694         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4695         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4696         set_blocksize(bdev, blocksize);
4697         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4698                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4699                        "external journal");
4700                 goto out_bdev;
4701         }
4702
4703         es = (struct ext4_super_block *) (bh->b_data + offset);
4704         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4705             !(le32_to_cpu(es->s_feature_incompat) &
4706               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4707                 ext4_msg(sb, KERN_ERR, "external journal has "
4708                                         "bad superblock");
4709                 brelse(bh);
4710                 goto out_bdev;
4711         }
4712
4713         if ((le32_to_cpu(es->s_feature_ro_compat) &
4714              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4715             es->s_checksum != ext4_superblock_csum(sb, es)) {
4716                 ext4_msg(sb, KERN_ERR, "external journal has "
4717                                        "corrupt superblock");
4718                 brelse(bh);
4719                 goto out_bdev;
4720         }
4721
4722         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4723                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4724                 brelse(bh);
4725                 goto out_bdev;
4726         }
4727
4728         len = ext4_blocks_count(es);
4729         start = sb_block + 1;
4730         brelse(bh);     /* we're done with the superblock */
4731
4732         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4733                                         start, len, blocksize);
4734         if (!journal) {
4735                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4736                 goto out_bdev;
4737         }
4738         journal->j_private = sb;
4739         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4740         wait_on_buffer(journal->j_sb_buffer);
4741         if (!buffer_uptodate(journal->j_sb_buffer)) {
4742                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4743                 goto out_journal;
4744         }
4745         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4746                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4747                                         "user (unsupported) - %d",
4748                         be32_to_cpu(journal->j_superblock->s_nr_users));
4749                 goto out_journal;
4750         }
4751         EXT4_SB(sb)->journal_bdev = bdev;
4752         ext4_init_journal_params(sb, journal);
4753         return journal;
4754
4755 out_journal:
4756         jbd2_journal_destroy(journal);
4757 out_bdev:
4758         ext4_blkdev_put(bdev);
4759         return NULL;
4760 }
4761
4762 static int ext4_load_journal(struct super_block *sb,
4763                              struct ext4_super_block *es,
4764                              unsigned long journal_devnum)
4765 {
4766         journal_t *journal;
4767         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4768         dev_t journal_dev;
4769         int err = 0;
4770         int really_read_only;
4771
4772         BUG_ON(!ext4_has_feature_journal(sb));
4773
4774         if (journal_devnum &&
4775             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4776                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4777                         "numbers have changed");
4778                 journal_dev = new_decode_dev(journal_devnum);
4779         } else
4780                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4781
4782         really_read_only = bdev_read_only(sb->s_bdev);
4783
4784         /*
4785          * Are we loading a blank journal or performing recovery after a
4786          * crash?  For recovery, we need to check in advance whether we
4787          * can get read-write access to the device.
4788          */
4789         if (ext4_has_feature_journal_needs_recovery(sb)) {
4790                 if (sb_rdonly(sb)) {
4791                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4792                                         "required on readonly filesystem");
4793                         if (really_read_only) {
4794                                 ext4_msg(sb, KERN_ERR, "write access "
4795                                         "unavailable, cannot proceed "
4796                                         "(try mounting with noload)");
4797                                 return -EROFS;
4798                         }
4799                         ext4_msg(sb, KERN_INFO, "write access will "
4800                                "be enabled during recovery");
4801                 }
4802         }
4803
4804         if (journal_inum && journal_dev) {
4805                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4806                        "and inode journals!");
4807                 return -EINVAL;
4808         }
4809
4810         if (journal_inum) {
4811                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4812                         return -EINVAL;
4813         } else {
4814                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4815                         return -EINVAL;
4816         }
4817
4818         if (!(journal->j_flags & JBD2_BARRIER))
4819                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4820
4821         if (!ext4_has_feature_journal_needs_recovery(sb))
4822                 err = jbd2_journal_wipe(journal, !really_read_only);
4823         if (!err) {
4824                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4825                 if (save)
4826                         memcpy(save, ((char *) es) +
4827                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4828                 err = jbd2_journal_load(journal);
4829                 if (save)
4830                         memcpy(((char *) es) + EXT4_S_ERR_START,
4831                                save, EXT4_S_ERR_LEN);
4832                 kfree(save);
4833         }
4834
4835         if (err) {
4836                 ext4_msg(sb, KERN_ERR, "error loading journal");
4837                 jbd2_journal_destroy(journal);
4838                 return err;
4839         }
4840
4841         EXT4_SB(sb)->s_journal = journal;
4842         ext4_clear_journal_err(sb, es);
4843
4844         if (!really_read_only && journal_devnum &&
4845             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4846                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4847
4848                 /* Make sure we flush the recovery flag to disk. */
4849                 ext4_commit_super(sb, 1);
4850         }
4851
4852         return 0;
4853 }
4854
4855 static int ext4_commit_super(struct super_block *sb, int sync)
4856 {
4857         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4858         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4859         int error = 0;
4860
4861         if (!sbh || block_device_ejected(sb))
4862                 return error;
4863
4864         /*
4865          * The superblock bh should be mapped, but it might not be if the
4866          * device was hot-removed. Not much we can do but fail the I/O.
4867          */
4868         if (!buffer_mapped(sbh))
4869                 return error;
4870
4871         /*
4872          * If the file system is mounted read-only, don't update the
4873          * superblock write time.  This avoids updating the superblock
4874          * write time when we are mounting the root file system
4875          * read/only but we need to replay the journal; at that point,
4876          * for people who are east of GMT and who make their clock
4877          * tick in localtime for Windows bug-for-bug compatibility,
4878          * the clock is set in the future, and this will cause e2fsck
4879          * to complain and force a full file system check.
4880          */
4881         if (!(sb->s_flags & SB_RDONLY))
4882                 ext4_update_tstamp(es, s_wtime);
4883         if (sb->s_bdev->bd_part)
4884                 es->s_kbytes_written =
4885                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4886                             ((part_stat_read(sb->s_bdev->bd_part,
4887                                              sectors[STAT_WRITE]) -
4888                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4889         else
4890                 es->s_kbytes_written =
4891                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4892         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4893                 ext4_free_blocks_count_set(es,
4894                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4895                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4896         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4897                 es->s_free_inodes_count =
4898                         cpu_to_le32(percpu_counter_sum_positive(
4899                                 &EXT4_SB(sb)->s_freeinodes_counter));
4900         BUFFER_TRACE(sbh, "marking dirty");
4901         ext4_superblock_csum_set(sb);
4902         if (sync)
4903                 lock_buffer(sbh);
4904         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
4905                 /*
4906                  * Oh, dear.  A previous attempt to write the
4907                  * superblock failed.  This could happen because the
4908                  * USB device was yanked out.  Or it could happen to
4909                  * be a transient write error and maybe the block will
4910                  * be remapped.  Nothing we can do but to retry the
4911                  * write and hope for the best.
4912                  */
4913                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4914                        "superblock detected");
4915                 clear_buffer_write_io_error(sbh);
4916                 set_buffer_uptodate(sbh);
4917         }
4918         mark_buffer_dirty(sbh);
4919         if (sync) {
4920                 unlock_buffer(sbh);
4921                 error = __sync_dirty_buffer(sbh,
4922                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4923                 if (buffer_write_io_error(sbh)) {
4924                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4925                                "superblock");
4926                         clear_buffer_write_io_error(sbh);
4927                         set_buffer_uptodate(sbh);
4928                 }
4929         }
4930         return error;
4931 }
4932
4933 /*
4934  * Have we just finished recovery?  If so, and if we are mounting (or
4935  * remounting) the filesystem readonly, then we will end up with a
4936  * consistent fs on disk.  Record that fact.
4937  */
4938 static void ext4_mark_recovery_complete(struct super_block *sb,
4939                                         struct ext4_super_block *es)
4940 {
4941         journal_t *journal = EXT4_SB(sb)->s_journal;
4942
4943         if (!ext4_has_feature_journal(sb)) {
4944                 BUG_ON(journal != NULL);
4945                 return;
4946         }
4947         jbd2_journal_lock_updates(journal);
4948         if (jbd2_journal_flush(journal) < 0)
4949                 goto out;
4950
4951         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4952                 ext4_clear_feature_journal_needs_recovery(sb);
4953                 ext4_commit_super(sb, 1);
4954         }
4955
4956 out:
4957         jbd2_journal_unlock_updates(journal);
4958 }
4959
4960 /*
4961  * If we are mounting (or read-write remounting) a filesystem whose journal
4962  * has recorded an error from a previous lifetime, move that error to the
4963  * main filesystem now.
4964  */
4965 static void ext4_clear_journal_err(struct super_block *sb,
4966                                    struct ext4_super_block *es)
4967 {
4968         journal_t *journal;
4969         int j_errno;
4970         const char *errstr;
4971
4972         BUG_ON(!ext4_has_feature_journal(sb));
4973
4974         journal = EXT4_SB(sb)->s_journal;
4975
4976         /*
4977          * Now check for any error status which may have been recorded in the
4978          * journal by a prior ext4_error() or ext4_abort()
4979          */
4980
4981         j_errno = jbd2_journal_errno(journal);
4982         if (j_errno) {
4983                 char nbuf[16];
4984
4985                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4986                 ext4_warning(sb, "Filesystem error recorded "
4987                              "from previous mount: %s", errstr);
4988                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4989
4990                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4991                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4992                 ext4_commit_super(sb, 1);
4993
4994                 jbd2_journal_clear_err(journal);
4995                 jbd2_journal_update_sb_errno(journal);
4996         }
4997 }
4998
4999 /*
5000  * Force the running and committing transactions to commit,
5001  * and wait on the commit.
5002  */
5003 int ext4_force_commit(struct super_block *sb)
5004 {
5005         journal_t *journal;
5006
5007         if (sb_rdonly(sb))
5008                 return 0;
5009
5010         journal = EXT4_SB(sb)->s_journal;
5011         return ext4_journal_force_commit(journal);
5012 }
5013
5014 static int ext4_sync_fs(struct super_block *sb, int wait)
5015 {
5016         int ret = 0;
5017         tid_t target;
5018         bool needs_barrier = false;
5019         struct ext4_sb_info *sbi = EXT4_SB(sb);
5020
5021         if (unlikely(ext4_forced_shutdown(sbi)))
5022                 return 0;
5023
5024         trace_ext4_sync_fs(sb, wait);
5025         flush_workqueue(sbi->rsv_conversion_wq);
5026         /*
5027          * Writeback quota in non-journalled quota case - journalled quota has
5028          * no dirty dquots
5029          */
5030         dquot_writeback_dquots(sb, -1);
5031         /*
5032          * Data writeback is possible w/o journal transaction, so barrier must
5033          * being sent at the end of the function. But we can skip it if
5034          * transaction_commit will do it for us.
5035          */
5036         if (sbi->s_journal) {
5037                 target = jbd2_get_latest_transaction(sbi->s_journal);
5038                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5039                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5040                         needs_barrier = true;
5041
5042                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5043                         if (wait)
5044                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5045                                                            target);
5046                 }
5047         } else if (wait && test_opt(sb, BARRIER))
5048                 needs_barrier = true;
5049         if (needs_barrier) {
5050                 int err;
5051                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5052                 if (!ret)
5053                         ret = err;
5054         }
5055
5056         return ret;
5057 }
5058
5059 /*
5060  * LVM calls this function before a (read-only) snapshot is created.  This
5061  * gives us a chance to flush the journal completely and mark the fs clean.
5062  *
5063  * Note that only this function cannot bring a filesystem to be in a clean
5064  * state independently. It relies on upper layer to stop all data & metadata
5065  * modifications.
5066  */
5067 static int ext4_freeze(struct super_block *sb)
5068 {
5069         int error = 0;
5070         journal_t *journal;
5071
5072         if (sb_rdonly(sb))
5073                 return 0;
5074
5075         journal = EXT4_SB(sb)->s_journal;
5076
5077         if (journal) {
5078                 /* Now we set up the journal barrier. */
5079                 jbd2_journal_lock_updates(journal);
5080
5081                 /*
5082                  * Don't clear the needs_recovery flag if we failed to
5083                  * flush the journal.
5084                  */
5085                 error = jbd2_journal_flush(journal);
5086                 if (error < 0)
5087                         goto out;
5088
5089                 /* Journal blocked and flushed, clear needs_recovery flag. */
5090                 ext4_clear_feature_journal_needs_recovery(sb);
5091         }
5092
5093         error = ext4_commit_super(sb, 1);
5094 out:
5095         if (journal)
5096                 /* we rely on upper layer to stop further updates */
5097                 jbd2_journal_unlock_updates(journal);
5098         return error;
5099 }
5100
5101 /*
5102  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5103  * flag here, even though the filesystem is not technically dirty yet.
5104  */
5105 static int ext4_unfreeze(struct super_block *sb)
5106 {
5107         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5108                 return 0;
5109
5110         if (EXT4_SB(sb)->s_journal) {
5111                 /* Reset the needs_recovery flag before the fs is unlocked. */
5112                 ext4_set_feature_journal_needs_recovery(sb);
5113         }
5114
5115         ext4_commit_super(sb, 1);
5116         return 0;
5117 }
5118
5119 /*
5120  * Structure to save mount options for ext4_remount's benefit
5121  */
5122 struct ext4_mount_options {
5123         unsigned long s_mount_opt;
5124         unsigned long s_mount_opt2;
5125         kuid_t s_resuid;
5126         kgid_t s_resgid;
5127         unsigned long s_commit_interval;
5128         u32 s_min_batch_time, s_max_batch_time;
5129 #ifdef CONFIG_QUOTA
5130         int s_jquota_fmt;
5131         char *s_qf_names[EXT4_MAXQUOTAS];
5132 #endif
5133 };
5134
5135 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5136 {
5137         struct ext4_super_block *es;
5138         struct ext4_sb_info *sbi = EXT4_SB(sb);
5139         unsigned long old_sb_flags;
5140         struct ext4_mount_options old_opts;
5141         int enable_quota = 0;
5142         ext4_group_t g;
5143         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5144         int err = 0;
5145 #ifdef CONFIG_QUOTA
5146         int i, j;
5147         char *to_free[EXT4_MAXQUOTAS];
5148 #endif
5149         char *orig_data = kstrdup(data, GFP_KERNEL);
5150
5151         if (data && !orig_data)
5152                 return -ENOMEM;
5153
5154         /* Store the original options */
5155         old_sb_flags = sb->s_flags;
5156         old_opts.s_mount_opt = sbi->s_mount_opt;
5157         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5158         old_opts.s_resuid = sbi->s_resuid;
5159         old_opts.s_resgid = sbi->s_resgid;
5160         old_opts.s_commit_interval = sbi->s_commit_interval;
5161         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5162         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5163 #ifdef CONFIG_QUOTA
5164         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5165         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5166                 if (sbi->s_qf_names[i]) {
5167                         char *qf_name = get_qf_name(sb, sbi, i);
5168
5169                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5170                         if (!old_opts.s_qf_names[i]) {
5171                                 for (j = 0; j < i; j++)
5172                                         kfree(old_opts.s_qf_names[j]);
5173                                 kfree(orig_data);
5174                                 return -ENOMEM;
5175                         }
5176                 } else
5177                         old_opts.s_qf_names[i] = NULL;
5178 #endif
5179         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5180                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5181
5182         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5183                 err = -EINVAL;
5184                 goto restore_opts;
5185         }
5186
5187         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5188             test_opt(sb, JOURNAL_CHECKSUM)) {
5189                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5190                          "during remount not supported; ignoring");
5191                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5192         }
5193
5194         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5195                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5196                         ext4_msg(sb, KERN_ERR, "can't mount with "
5197                                  "both data=journal and delalloc");
5198                         err = -EINVAL;
5199                         goto restore_opts;
5200                 }
5201                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5202                         ext4_msg(sb, KERN_ERR, "can't mount with "
5203                                  "both data=journal and dioread_nolock");
5204                         err = -EINVAL;
5205                         goto restore_opts;
5206                 }
5207                 if (test_opt(sb, DAX)) {
5208                         ext4_msg(sb, KERN_ERR, "can't mount with "
5209                                  "both data=journal and dax");
5210                         err = -EINVAL;
5211                         goto restore_opts;
5212                 }
5213         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5214                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5215                         ext4_msg(sb, KERN_ERR, "can't mount with "
5216                                 "journal_async_commit in data=ordered mode");
5217                         err = -EINVAL;
5218                         goto restore_opts;
5219                 }
5220         }
5221
5222         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5223                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5224                 err = -EINVAL;
5225                 goto restore_opts;
5226         }
5227
5228         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5229                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5230                         "dax flag with busy inodes while remounting");
5231                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5232         }
5233
5234         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5235                 ext4_abort(sb, "Abort forced by user");
5236
5237         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5238                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5239
5240         es = sbi->s_es;
5241
5242         if (sbi->s_journal) {
5243                 ext4_init_journal_params(sb, sbi->s_journal);
5244                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5245         }
5246
5247         if (*flags & SB_LAZYTIME)
5248                 sb->s_flags |= SB_LAZYTIME;
5249
5250         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5251                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5252                         err = -EROFS;
5253                         goto restore_opts;
5254                 }
5255
5256                 if (*flags & SB_RDONLY) {
5257                         err = sync_filesystem(sb);
5258                         if (err < 0)
5259                                 goto restore_opts;
5260                         err = dquot_suspend(sb, -1);
5261                         if (err < 0)
5262                                 goto restore_opts;
5263
5264                         /*
5265                          * First of all, the unconditional stuff we have to do
5266                          * to disable replay of the journal when we next remount
5267                          */
5268                         sb->s_flags |= SB_RDONLY;
5269
5270                         /*
5271                          * OK, test if we are remounting a valid rw partition
5272                          * readonly, and if so set the rdonly flag and then
5273                          * mark the partition as valid again.
5274                          */
5275                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5276                             (sbi->s_mount_state & EXT4_VALID_FS))
5277                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5278
5279                         if (sbi->s_journal)
5280                                 ext4_mark_recovery_complete(sb, es);
5281                         if (sbi->s_mmp_tsk)
5282                                 kthread_stop(sbi->s_mmp_tsk);
5283                 } else {
5284                         /* Make sure we can mount this feature set readwrite */
5285                         if (ext4_has_feature_readonly(sb) ||
5286                             !ext4_feature_set_ok(sb, 0)) {
5287                                 err = -EROFS;
5288                                 goto restore_opts;
5289                         }
5290                         /*
5291                          * Make sure the group descriptor checksums
5292                          * are sane.  If they aren't, refuse to remount r/w.
5293                          */
5294                         for (g = 0; g < sbi->s_groups_count; g++) {
5295                                 struct ext4_group_desc *gdp =
5296                                         ext4_get_group_desc(sb, g, NULL);
5297
5298                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5299                                         ext4_msg(sb, KERN_ERR,
5300                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5301                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5302                                                le16_to_cpu(gdp->bg_checksum));
5303                                         err = -EFSBADCRC;
5304                                         goto restore_opts;
5305                                 }
5306                         }
5307
5308                         /*
5309                          * If we have an unprocessed orphan list hanging
5310                          * around from a previously readonly bdev mount,
5311                          * require a full umount/remount for now.
5312                          */
5313                         if (es->s_last_orphan) {
5314                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5315                                        "remount RDWR because of unprocessed "
5316                                        "orphan inode list.  Please "
5317                                        "umount/remount instead");
5318                                 err = -EINVAL;
5319                                 goto restore_opts;
5320                         }
5321
5322                         /*
5323                          * Mounting a RDONLY partition read-write, so reread
5324                          * and store the current valid flag.  (It may have
5325                          * been changed by e2fsck since we originally mounted
5326                          * the partition.)
5327                          */
5328                         if (sbi->s_journal)
5329                                 ext4_clear_journal_err(sb, es);
5330                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5331
5332                         err = ext4_setup_super(sb, es, 0);
5333                         if (err)
5334                                 goto restore_opts;
5335
5336                         sb->s_flags &= ~SB_RDONLY;
5337                         if (ext4_has_feature_mmp(sb))
5338                                 if (ext4_multi_mount_protect(sb,
5339                                                 le64_to_cpu(es->s_mmp_block))) {
5340                                         err = -EROFS;
5341                                         goto restore_opts;
5342                                 }
5343                         enable_quota = 1;
5344                 }
5345         }
5346
5347         /*
5348          * Reinitialize lazy itable initialization thread based on
5349          * current settings
5350          */
5351         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5352                 ext4_unregister_li_request(sb);
5353         else {
5354                 ext4_group_t first_not_zeroed;
5355                 first_not_zeroed = ext4_has_uninit_itable(sb);
5356                 ext4_register_li_request(sb, first_not_zeroed);
5357         }
5358
5359         ext4_setup_system_zone(sb);
5360         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5361                 err = ext4_commit_super(sb, 1);
5362                 if (err)
5363                         goto restore_opts;
5364         }
5365
5366 #ifdef CONFIG_QUOTA
5367         /* Release old quota file names */
5368         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5369                 kfree(old_opts.s_qf_names[i]);
5370         if (enable_quota) {
5371                 if (sb_any_quota_suspended(sb))
5372                         dquot_resume(sb, -1);
5373                 else if (ext4_has_feature_quota(sb)) {
5374                         err = ext4_enable_quotas(sb);
5375                         if (err)
5376                                 goto restore_opts;
5377                 }
5378         }
5379 #endif
5380
5381         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5382         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5383         kfree(orig_data);
5384         return 0;
5385
5386 restore_opts:
5387         sb->s_flags = old_sb_flags;
5388         sbi->s_mount_opt = old_opts.s_mount_opt;
5389         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5390         sbi->s_resuid = old_opts.s_resuid;
5391         sbi->s_resgid = old_opts.s_resgid;
5392         sbi->s_commit_interval = old_opts.s_commit_interval;
5393         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5394         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5395 #ifdef CONFIG_QUOTA
5396         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5397         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5398                 to_free[i] = get_qf_name(sb, sbi, i);
5399                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5400         }
5401         synchronize_rcu();
5402         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5403                 kfree(to_free[i]);
5404 #endif
5405         kfree(orig_data);
5406         return err;
5407 }
5408
5409 #ifdef CONFIG_QUOTA
5410 static int ext4_statfs_project(struct super_block *sb,
5411                                kprojid_t projid, struct kstatfs *buf)
5412 {
5413         struct kqid qid;
5414         struct dquot *dquot;
5415         u64 limit;
5416         u64 curblock;
5417
5418         qid = make_kqid_projid(projid);
5419         dquot = dqget(sb, qid);
5420         if (IS_ERR(dquot))
5421                 return PTR_ERR(dquot);
5422         spin_lock(&dquot->dq_dqb_lock);
5423
5424         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5425                  dquot->dq_dqb.dqb_bsoftlimit :
5426                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5427         if (limit && buf->f_blocks > limit) {
5428                 curblock = (dquot->dq_dqb.dqb_curspace +
5429                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5430                 buf->f_blocks = limit;
5431                 buf->f_bfree = buf->f_bavail =
5432                         (buf->f_blocks > curblock) ?
5433                          (buf->f_blocks - curblock) : 0;
5434         }
5435
5436         limit = dquot->dq_dqb.dqb_isoftlimit ?
5437                 dquot->dq_dqb.dqb_isoftlimit :
5438                 dquot->dq_dqb.dqb_ihardlimit;
5439         if (limit && buf->f_files > limit) {
5440                 buf->f_files = limit;
5441                 buf->f_ffree =
5442                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5443                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5444         }
5445
5446         spin_unlock(&dquot->dq_dqb_lock);
5447         dqput(dquot);
5448         return 0;
5449 }
5450 #endif
5451
5452 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5453 {
5454         struct super_block *sb = dentry->d_sb;
5455         struct ext4_sb_info *sbi = EXT4_SB(sb);
5456         struct ext4_super_block *es = sbi->s_es;
5457         ext4_fsblk_t overhead = 0, resv_blocks;
5458         u64 fsid;
5459         s64 bfree;
5460         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5461
5462         if (!test_opt(sb, MINIX_DF))
5463                 overhead = sbi->s_overhead;
5464
5465         buf->f_type = EXT4_SUPER_MAGIC;
5466         buf->f_bsize = sb->s_blocksize;
5467         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5468         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5469                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5470         /* prevent underflow in case that few free space is available */
5471         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5472         buf->f_bavail = buf->f_bfree -
5473                         (ext4_r_blocks_count(es) + resv_blocks);
5474         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5475                 buf->f_bavail = 0;
5476         buf->f_files = le32_to_cpu(es->s_inodes_count);
5477         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5478         buf->f_namelen = EXT4_NAME_LEN;
5479         fsid = le64_to_cpup((void *)es->s_uuid) ^
5480                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5481         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5482         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5483
5484 #ifdef CONFIG_QUOTA
5485         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5486             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5487                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5488 #endif
5489         return 0;
5490 }
5491
5492
5493 #ifdef CONFIG_QUOTA
5494
5495 /*
5496  * Helper functions so that transaction is started before we acquire dqio_sem
5497  * to keep correct lock ordering of transaction > dqio_sem
5498  */
5499 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5500 {
5501         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5502 }
5503
5504 static int ext4_write_dquot(struct dquot *dquot)
5505 {
5506         int ret, err;
5507         handle_t *handle;
5508         struct inode *inode;
5509
5510         inode = dquot_to_inode(dquot);
5511         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5512                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5513         if (IS_ERR(handle))
5514                 return PTR_ERR(handle);
5515         ret = dquot_commit(dquot);
5516         err = ext4_journal_stop(handle);
5517         if (!ret)
5518                 ret = err;
5519         return ret;
5520 }
5521
5522 static int ext4_acquire_dquot(struct dquot *dquot)
5523 {
5524         int ret, err;
5525         handle_t *handle;
5526
5527         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5528                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5529         if (IS_ERR(handle))
5530                 return PTR_ERR(handle);
5531         ret = dquot_acquire(dquot);
5532         err = ext4_journal_stop(handle);
5533         if (!ret)
5534                 ret = err;
5535         return ret;
5536 }
5537
5538 static int ext4_release_dquot(struct dquot *dquot)
5539 {
5540         int ret, err;
5541         handle_t *handle;
5542
5543         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5544                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5545         if (IS_ERR(handle)) {
5546                 /* Release dquot anyway to avoid endless cycle in dqput() */
5547                 dquot_release(dquot);
5548                 return PTR_ERR(handle);
5549         }
5550         ret = dquot_release(dquot);
5551         err = ext4_journal_stop(handle);
5552         if (!ret)
5553                 ret = err;
5554         return ret;
5555 }
5556
5557 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5558 {
5559         struct super_block *sb = dquot->dq_sb;
5560         struct ext4_sb_info *sbi = EXT4_SB(sb);
5561
5562         /* Are we journaling quotas? */
5563         if (ext4_has_feature_quota(sb) ||
5564             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5565                 dquot_mark_dquot_dirty(dquot);
5566                 return ext4_write_dquot(dquot);
5567         } else {
5568                 return dquot_mark_dquot_dirty(dquot);
5569         }
5570 }
5571
5572 static int ext4_write_info(struct super_block *sb, int type)
5573 {
5574         int ret, err;
5575         handle_t *handle;
5576
5577         /* Data block + inode block */
5578         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5579         if (IS_ERR(handle))
5580                 return PTR_ERR(handle);
5581         ret = dquot_commit_info(sb, type);
5582         err = ext4_journal_stop(handle);
5583         if (!ret)
5584                 ret = err;
5585         return ret;
5586 }
5587
5588 /*
5589  * Turn on quotas during mount time - we need to find
5590  * the quota file and such...
5591  */
5592 static int ext4_quota_on_mount(struct super_block *sb, int type)
5593 {
5594         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5595                                         EXT4_SB(sb)->s_jquota_fmt, type);
5596 }
5597
5598 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5599 {
5600         struct ext4_inode_info *ei = EXT4_I(inode);
5601
5602         /* The first argument of lockdep_set_subclass has to be
5603          * *exactly* the same as the argument to init_rwsem() --- in
5604          * this case, in init_once() --- or lockdep gets unhappy
5605          * because the name of the lock is set using the
5606          * stringification of the argument to init_rwsem().
5607          */
5608         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5609         lockdep_set_subclass(&ei->i_data_sem, subclass);
5610 }
5611
5612 /*
5613  * Standard function to be called on quota_on
5614  */
5615 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5616                          const struct path *path)
5617 {
5618         int err;
5619
5620         if (!test_opt(sb, QUOTA))
5621                 return -EINVAL;
5622
5623         /* Quotafile not on the same filesystem? */
5624         if (path->dentry->d_sb != sb)
5625                 return -EXDEV;
5626         /* Journaling quota? */
5627         if (EXT4_SB(sb)->s_qf_names[type]) {
5628                 /* Quotafile not in fs root? */
5629                 if (path->dentry->d_parent != sb->s_root)
5630                         ext4_msg(sb, KERN_WARNING,
5631                                 "Quota file not on filesystem root. "
5632                                 "Journaled quota will not work");
5633                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5634         } else {
5635                 /*
5636                  * Clear the flag just in case mount options changed since
5637                  * last time.
5638                  */
5639                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5640         }
5641
5642         /*
5643          * When we journal data on quota file, we have to flush journal to see
5644          * all updates to the file when we bypass pagecache...
5645          */
5646         if (EXT4_SB(sb)->s_journal &&
5647             ext4_should_journal_data(d_inode(path->dentry))) {
5648                 /*
5649                  * We don't need to lock updates but journal_flush() could
5650                  * otherwise be livelocked...
5651                  */
5652                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5653                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5654                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5655                 if (err)
5656                         return err;
5657         }
5658
5659         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5660         err = dquot_quota_on(sb, type, format_id, path);
5661         if (err) {
5662                 lockdep_set_quota_inode(path->dentry->d_inode,
5663                                              I_DATA_SEM_NORMAL);
5664         } else {
5665                 struct inode *inode = d_inode(path->dentry);
5666                 handle_t *handle;
5667
5668                 /*
5669                  * Set inode flags to prevent userspace from messing with quota
5670                  * files. If this fails, we return success anyway since quotas
5671                  * are already enabled and this is not a hard failure.
5672                  */
5673                 inode_lock(inode);
5674                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5675                 if (IS_ERR(handle))
5676                         goto unlock_inode;
5677                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5678                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5679                                 S_NOATIME | S_IMMUTABLE);
5680                 ext4_mark_inode_dirty(handle, inode);
5681                 ext4_journal_stop(handle);
5682         unlock_inode:
5683                 inode_unlock(inode);
5684         }
5685         return err;
5686 }
5687
5688 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5689                              unsigned int flags)
5690 {
5691         int err;
5692         struct inode *qf_inode;
5693         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5694                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5695                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5696                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5697         };
5698
5699         BUG_ON(!ext4_has_feature_quota(sb));
5700
5701         if (!qf_inums[type])
5702                 return -EPERM;
5703
5704         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
5705         if (IS_ERR(qf_inode)) {
5706                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5707                 return PTR_ERR(qf_inode);
5708         }
5709
5710         /* Don't account quota for quota files to avoid recursion */
5711         qf_inode->i_flags |= S_NOQUOTA;
5712         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5713         err = dquot_enable(qf_inode, type, format_id, flags);
5714         if (err)
5715                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5716         iput(qf_inode);
5717
5718         return err;
5719 }
5720
5721 /* Enable usage tracking for all quota types. */
5722 static int ext4_enable_quotas(struct super_block *sb)
5723 {
5724         int type, err = 0;
5725         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5726                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5727                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5728                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5729         };
5730         bool quota_mopt[EXT4_MAXQUOTAS] = {
5731                 test_opt(sb, USRQUOTA),
5732                 test_opt(sb, GRPQUOTA),
5733                 test_opt(sb, PRJQUOTA),
5734         };
5735
5736         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5737         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5738                 if (qf_inums[type]) {
5739                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5740                                 DQUOT_USAGE_ENABLED |
5741                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5742                         if (err) {
5743                                 ext4_warning(sb,
5744                                         "Failed to enable quota tracking "
5745                                         "(type=%d, err=%d). Please run "
5746                                         "e2fsck to fix.", type, err);
5747                                 for (type--; type >= 0; type--)
5748                                         dquot_quota_off(sb, type);
5749
5750                                 return err;
5751                         }
5752                 }
5753         }
5754         return 0;
5755 }
5756
5757 static int ext4_quota_off(struct super_block *sb, int type)
5758 {
5759         struct inode *inode = sb_dqopt(sb)->files[type];
5760         handle_t *handle;
5761         int err;
5762
5763         /* Force all delayed allocation blocks to be allocated.
5764          * Caller already holds s_umount sem */
5765         if (test_opt(sb, DELALLOC))
5766                 sync_filesystem(sb);
5767
5768         if (!inode || !igrab(inode))
5769                 goto out;
5770
5771         err = dquot_quota_off(sb, type);
5772         if (err || ext4_has_feature_quota(sb))
5773                 goto out_put;
5774
5775         inode_lock(inode);
5776         /*
5777          * Update modification times of quota files when userspace can
5778          * start looking at them. If we fail, we return success anyway since
5779          * this is not a hard failure and quotas are already disabled.
5780          */
5781         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5782         if (IS_ERR(handle))
5783                 goto out_unlock;
5784         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5785         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5786         inode->i_mtime = inode->i_ctime = current_time(inode);
5787         ext4_mark_inode_dirty(handle, inode);
5788         ext4_journal_stop(handle);
5789 out_unlock:
5790         inode_unlock(inode);
5791 out_put:
5792         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5793         iput(inode);
5794         return err;
5795 out:
5796         return dquot_quota_off(sb, type);
5797 }
5798
5799 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5800  * acquiring the locks... As quota files are never truncated and quota code
5801  * itself serializes the operations (and no one else should touch the files)
5802  * we don't have to be afraid of races */
5803 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5804                                size_t len, loff_t off)
5805 {
5806         struct inode *inode = sb_dqopt(sb)->files[type];
5807         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5808         int offset = off & (sb->s_blocksize - 1);
5809         int tocopy;
5810         size_t toread;
5811         struct buffer_head *bh;
5812         loff_t i_size = i_size_read(inode);
5813
5814         if (off > i_size)
5815                 return 0;
5816         if (off+len > i_size)
5817                 len = i_size-off;
5818         toread = len;
5819         while (toread > 0) {
5820                 tocopy = sb->s_blocksize - offset < toread ?
5821                                 sb->s_blocksize - offset : toread;
5822                 bh = ext4_bread(NULL, inode, blk, 0);
5823                 if (IS_ERR(bh))
5824                         return PTR_ERR(bh);
5825                 if (!bh)        /* A hole? */
5826                         memset(data, 0, tocopy);
5827                 else
5828                         memcpy(data, bh->b_data+offset, tocopy);
5829                 brelse(bh);
5830                 offset = 0;
5831                 toread -= tocopy;
5832                 data += tocopy;
5833                 blk++;
5834         }
5835         return len;
5836 }
5837
5838 /* Write to quotafile (we know the transaction is already started and has
5839  * enough credits) */
5840 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5841                                 const char *data, size_t len, loff_t off)
5842 {
5843         struct inode *inode = sb_dqopt(sb)->files[type];
5844         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5845         int err, offset = off & (sb->s_blocksize - 1);
5846         int retries = 0;
5847         struct buffer_head *bh;
5848         handle_t *handle = journal_current_handle();
5849
5850         if (EXT4_SB(sb)->s_journal && !handle) {
5851                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5852                         " cancelled because transaction is not started",
5853                         (unsigned long long)off, (unsigned long long)len);
5854                 return -EIO;
5855         }
5856         /*
5857          * Since we account only one data block in transaction credits,
5858          * then it is impossible to cross a block boundary.
5859          */
5860         if (sb->s_blocksize - offset < len) {
5861                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5862                         " cancelled because not block aligned",
5863                         (unsigned long long)off, (unsigned long long)len);
5864                 return -EIO;
5865         }
5866
5867         do {
5868                 bh = ext4_bread(handle, inode, blk,
5869                                 EXT4_GET_BLOCKS_CREATE |
5870                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5871         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5872                  ext4_should_retry_alloc(inode->i_sb, &retries));
5873         if (IS_ERR(bh))
5874                 return PTR_ERR(bh);
5875         if (!bh)
5876                 goto out;
5877         BUFFER_TRACE(bh, "get write access");
5878         err = ext4_journal_get_write_access(handle, bh);
5879         if (err) {
5880                 brelse(bh);
5881                 return err;
5882         }
5883         lock_buffer(bh);
5884         memcpy(bh->b_data+offset, data, len);
5885         flush_dcache_page(bh->b_page);
5886         unlock_buffer(bh);
5887         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5888         brelse(bh);
5889 out:
5890         if (inode->i_size < off + len) {
5891                 i_size_write(inode, off + len);
5892                 EXT4_I(inode)->i_disksize = inode->i_size;
5893                 ext4_mark_inode_dirty(handle, inode);
5894         }
5895         return len;
5896 }
5897
5898 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5899 {
5900         const struct quota_format_ops   *ops;
5901
5902         if (!sb_has_quota_loaded(sb, qid->type))
5903                 return -ESRCH;
5904         ops = sb_dqopt(sb)->ops[qid->type];
5905         if (!ops || !ops->get_next_id)
5906                 return -ENOSYS;
5907         return dquot_get_next_id(sb, qid);
5908 }
5909 #endif
5910
5911 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5912                        const char *dev_name, void *data)
5913 {
5914         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5915 }
5916
5917 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5918 static inline void register_as_ext2(void)
5919 {
5920         int err = register_filesystem(&ext2_fs_type);
5921         if (err)
5922                 printk(KERN_WARNING
5923                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5924 }
5925
5926 static inline void unregister_as_ext2(void)
5927 {
5928         unregister_filesystem(&ext2_fs_type);
5929 }
5930
5931 static inline int ext2_feature_set_ok(struct super_block *sb)
5932 {
5933         if (ext4_has_unknown_ext2_incompat_features(sb))
5934                 return 0;
5935         if (sb_rdonly(sb))
5936                 return 1;
5937         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5938                 return 0;
5939         return 1;
5940 }
5941 #else
5942 static inline void register_as_ext2(void) { }
5943 static inline void unregister_as_ext2(void) { }
5944 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5945 #endif
5946
5947 static inline void register_as_ext3(void)
5948 {
5949         int err = register_filesystem(&ext3_fs_type);
5950         if (err)
5951                 printk(KERN_WARNING
5952                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5953 }
5954
5955 static inline void unregister_as_ext3(void)
5956 {
5957         unregister_filesystem(&ext3_fs_type);
5958 }
5959
5960 static inline int ext3_feature_set_ok(struct super_block *sb)
5961 {
5962         if (ext4_has_unknown_ext3_incompat_features(sb))
5963                 return 0;
5964         if (!ext4_has_feature_journal(sb))
5965                 return 0;
5966         if (sb_rdonly(sb))
5967                 return 1;
5968         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5969                 return 0;
5970         return 1;
5971 }
5972
5973 static struct file_system_type ext4_fs_type = {
5974         .owner          = THIS_MODULE,
5975         .name           = "ext4",
5976         .mount          = ext4_mount,
5977         .kill_sb        = kill_block_super,
5978         .fs_flags       = FS_REQUIRES_DEV,
5979 };
5980 MODULE_ALIAS_FS("ext4");
5981
5982 /* Shared across all ext4 file systems */
5983 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5984
5985 static int __init ext4_init_fs(void)
5986 {
5987         int i, err;
5988
5989         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5990         ext4_li_info = NULL;
5991         mutex_init(&ext4_li_mtx);
5992
5993         /* Build-time check for flags consistency */
5994         ext4_check_flag_values();
5995
5996         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5997                 init_waitqueue_head(&ext4__ioend_wq[i]);
5998
5999         err = ext4_init_es();
6000         if (err)
6001                 return err;
6002
6003         err = ext4_init_pending();
6004         if (err)
6005                 goto out6;
6006
6007         err = ext4_init_pageio();
6008         if (err)
6009                 goto out5;
6010
6011         err = ext4_init_system_zone();
6012         if (err)
6013                 goto out4;
6014
6015         err = ext4_init_sysfs();
6016         if (err)
6017                 goto out3;
6018
6019         err = ext4_init_mballoc();
6020         if (err)
6021                 goto out2;
6022         err = init_inodecache();
6023         if (err)
6024                 goto out1;
6025         register_as_ext3();
6026         register_as_ext2();
6027         err = register_filesystem(&ext4_fs_type);
6028         if (err)
6029                 goto out;
6030
6031         return 0;
6032 out:
6033         unregister_as_ext2();
6034         unregister_as_ext3();
6035         destroy_inodecache();
6036 out1:
6037         ext4_exit_mballoc();
6038 out2:
6039         ext4_exit_sysfs();
6040 out3:
6041         ext4_exit_system_zone();
6042 out4:
6043         ext4_exit_pageio();
6044 out5:
6045         ext4_exit_pending();
6046 out6:
6047         ext4_exit_es();
6048
6049         return err;
6050 }
6051
6052 static void __exit ext4_exit_fs(void)
6053 {
6054         ext4_destroy_lazyinit_thread();
6055         unregister_as_ext2();
6056         unregister_as_ext3();
6057         unregister_filesystem(&ext4_fs_type);
6058         destroy_inodecache();
6059         ext4_exit_mballoc();
6060         ext4_exit_sysfs();
6061         ext4_exit_system_zone();
6062         ext4_exit_pageio();
6063         ext4_exit_es();
6064         ext4_exit_pending();
6065 }
6066
6067 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6068 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6069 MODULE_LICENSE("GPL");
6070 MODULE_SOFTDEP("pre: crc32c");
6071 module_init(ext4_init_fs)
6072 module_exit(ext4_exit_fs)