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_EXT4_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_EXT4_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         ext4_update_dynamic_rev(sb);
2253         if (sbi->s_journal)
2254                 ext4_set_feature_journal_needs_recovery(sb);
2255
2256         err = ext4_commit_super(sb, 1);
2257 done:
2258         if (test_opt(sb, DEBUG))
2259                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2260                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2261                         sb->s_blocksize,
2262                         sbi->s_groups_count,
2263                         EXT4_BLOCKS_PER_GROUP(sb),
2264                         EXT4_INODES_PER_GROUP(sb),
2265                         sbi->s_mount_opt, sbi->s_mount_opt2);
2266
2267         cleancache_init_fs(sb);
2268         return err;
2269 }
2270
2271 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2272 {
2273         struct ext4_sb_info *sbi = EXT4_SB(sb);
2274         struct flex_groups *new_groups;
2275         int size;
2276
2277         if (!sbi->s_log_groups_per_flex)
2278                 return 0;
2279
2280         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2281         if (size <= sbi->s_flex_groups_allocated)
2282                 return 0;
2283
2284         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2285         new_groups = kvzalloc(size, GFP_KERNEL);
2286         if (!new_groups) {
2287                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2288                          size / (int) sizeof(struct flex_groups));
2289                 return -ENOMEM;
2290         }
2291
2292         if (sbi->s_flex_groups) {
2293                 memcpy(new_groups, sbi->s_flex_groups,
2294                        (sbi->s_flex_groups_allocated *
2295                         sizeof(struct flex_groups)));
2296                 kvfree(sbi->s_flex_groups);
2297         }
2298         sbi->s_flex_groups = new_groups;
2299         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2300         return 0;
2301 }
2302
2303 static int ext4_fill_flex_info(struct super_block *sb)
2304 {
2305         struct ext4_sb_info *sbi = EXT4_SB(sb);
2306         struct ext4_group_desc *gdp = NULL;
2307         ext4_group_t flex_group;
2308         int i, err;
2309
2310         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2311         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2312                 sbi->s_log_groups_per_flex = 0;
2313                 return 1;
2314         }
2315
2316         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2317         if (err)
2318                 goto failed;
2319
2320         for (i = 0; i < sbi->s_groups_count; i++) {
2321                 gdp = ext4_get_group_desc(sb, i, NULL);
2322
2323                 flex_group = ext4_flex_group(sbi, i);
2324                 atomic_add(ext4_free_inodes_count(sb, gdp),
2325                            &sbi->s_flex_groups[flex_group].free_inodes);
2326                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2327                              &sbi->s_flex_groups[flex_group].free_clusters);
2328                 atomic_add(ext4_used_dirs_count(sb, gdp),
2329                            &sbi->s_flex_groups[flex_group].used_dirs);
2330         }
2331
2332         return 1;
2333 failed:
2334         return 0;
2335 }
2336
2337 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2338                                    struct ext4_group_desc *gdp)
2339 {
2340         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2341         __u16 crc = 0;
2342         __le32 le_group = cpu_to_le32(block_group);
2343         struct ext4_sb_info *sbi = EXT4_SB(sb);
2344
2345         if (ext4_has_metadata_csum(sbi->s_sb)) {
2346                 /* Use new metadata_csum algorithm */
2347                 __u32 csum32;
2348                 __u16 dummy_csum = 0;
2349
2350                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2351                                      sizeof(le_group));
2352                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2353                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2354                                      sizeof(dummy_csum));
2355                 offset += sizeof(dummy_csum);
2356                 if (offset < sbi->s_desc_size)
2357                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2358                                              sbi->s_desc_size - offset);
2359
2360                 crc = csum32 & 0xFFFF;
2361                 goto out;
2362         }
2363
2364         /* old crc16 code */
2365         if (!ext4_has_feature_gdt_csum(sb))
2366                 return 0;
2367
2368         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2369         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2370         crc = crc16(crc, (__u8 *)gdp, offset);
2371         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2372         /* for checksum of struct ext4_group_desc do the rest...*/
2373         if (ext4_has_feature_64bit(sb) &&
2374             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2375                 crc = crc16(crc, (__u8 *)gdp + offset,
2376                             le16_to_cpu(sbi->s_es->s_desc_size) -
2377                                 offset);
2378
2379 out:
2380         return cpu_to_le16(crc);
2381 }
2382
2383 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2384                                 struct ext4_group_desc *gdp)
2385 {
2386         if (ext4_has_group_desc_csum(sb) &&
2387             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2388                 return 0;
2389
2390         return 1;
2391 }
2392
2393 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2394                               struct ext4_group_desc *gdp)
2395 {
2396         if (!ext4_has_group_desc_csum(sb))
2397                 return;
2398         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2399 }
2400
2401 /* Called at mount-time, super-block is locked */
2402 static int ext4_check_descriptors(struct super_block *sb,
2403                                   ext4_fsblk_t sb_block,
2404                                   ext4_group_t *first_not_zeroed)
2405 {
2406         struct ext4_sb_info *sbi = EXT4_SB(sb);
2407         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2408         ext4_fsblk_t last_block;
2409         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2410         ext4_fsblk_t block_bitmap;
2411         ext4_fsblk_t inode_bitmap;
2412         ext4_fsblk_t inode_table;
2413         int flexbg_flag = 0;
2414         ext4_group_t i, grp = sbi->s_groups_count;
2415
2416         if (ext4_has_feature_flex_bg(sb))
2417                 flexbg_flag = 1;
2418
2419         ext4_debug("Checking group descriptors");
2420
2421         for (i = 0; i < sbi->s_groups_count; i++) {
2422                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2423
2424                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2425                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2426                 else
2427                         last_block = first_block +
2428                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2429
2430                 if ((grp == sbi->s_groups_count) &&
2431                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2432                         grp = i;
2433
2434                 block_bitmap = ext4_block_bitmap(sb, gdp);
2435                 if (block_bitmap == sb_block) {
2436                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2437                                  "Block bitmap for group %u overlaps "
2438                                  "superblock", i);
2439                         if (!sb_rdonly(sb))
2440                                 return 0;
2441                 }
2442                 if (block_bitmap >= sb_block + 1 &&
2443                     block_bitmap <= last_bg_block) {
2444                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2445                                  "Block bitmap for group %u overlaps "
2446                                  "block group descriptors", i);
2447                         if (!sb_rdonly(sb))
2448                                 return 0;
2449                 }
2450                 if (block_bitmap < first_block || block_bitmap > last_block) {
2451                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2452                                "Block bitmap for group %u not in group "
2453                                "(block %llu)!", i, block_bitmap);
2454                         return 0;
2455                 }
2456                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2457                 if (inode_bitmap == sb_block) {
2458                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2459                                  "Inode bitmap for group %u overlaps "
2460                                  "superblock", i);
2461                         if (!sb_rdonly(sb))
2462                                 return 0;
2463                 }
2464                 if (inode_bitmap >= sb_block + 1 &&
2465                     inode_bitmap <= last_bg_block) {
2466                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2467                                  "Inode bitmap for group %u overlaps "
2468                                  "block group descriptors", i);
2469                         if (!sb_rdonly(sb))
2470                                 return 0;
2471                 }
2472                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2473                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2474                                "Inode bitmap for group %u not in group "
2475                                "(block %llu)!", i, inode_bitmap);
2476                         return 0;
2477                 }
2478                 inode_table = ext4_inode_table(sb, gdp);
2479                 if (inode_table == sb_block) {
2480                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2481                                  "Inode table for group %u overlaps "
2482                                  "superblock", i);
2483                         if (!sb_rdonly(sb))
2484                                 return 0;
2485                 }
2486                 if (inode_table >= sb_block + 1 &&
2487                     inode_table <= last_bg_block) {
2488                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2489                                  "Inode table for group %u overlaps "
2490                                  "block group descriptors", i);
2491                         if (!sb_rdonly(sb))
2492                                 return 0;
2493                 }
2494                 if (inode_table < first_block ||
2495                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2496                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2497                                "Inode table for group %u not in group "
2498                                "(block %llu)!", i, inode_table);
2499                         return 0;
2500                 }
2501                 ext4_lock_group(sb, i);
2502                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2503                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2504                                  "Checksum for group %u failed (%u!=%u)",
2505                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2506                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2507                         if (!sb_rdonly(sb)) {
2508                                 ext4_unlock_group(sb, i);
2509                                 return 0;
2510                         }
2511                 }
2512                 ext4_unlock_group(sb, i);
2513                 if (!flexbg_flag)
2514                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2515         }
2516         if (NULL != first_not_zeroed)
2517                 *first_not_zeroed = grp;
2518         return 1;
2519 }
2520
2521 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2522  * the superblock) which were deleted from all directories, but held open by
2523  * a process at the time of a crash.  We walk the list and try to delete these
2524  * inodes at recovery time (only with a read-write filesystem).
2525  *
2526  * In order to keep the orphan inode chain consistent during traversal (in
2527  * case of crash during recovery), we link each inode into the superblock
2528  * orphan list_head and handle it the same way as an inode deletion during
2529  * normal operation (which journals the operations for us).
2530  *
2531  * We only do an iget() and an iput() on each inode, which is very safe if we
2532  * accidentally point at an in-use or already deleted inode.  The worst that
2533  * can happen in this case is that we get a "bit already cleared" message from
2534  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2535  * e2fsck was run on this filesystem, and it must have already done the orphan
2536  * inode cleanup for us, so we can safely abort without any further action.
2537  */
2538 static void ext4_orphan_cleanup(struct super_block *sb,
2539                                 struct ext4_super_block *es)
2540 {
2541         unsigned int s_flags = sb->s_flags;
2542         int ret, nr_orphans = 0, nr_truncates = 0;
2543 #ifdef CONFIG_QUOTA
2544         int quota_update = 0;
2545         int i;
2546 #endif
2547         if (!es->s_last_orphan) {
2548                 jbd_debug(4, "no orphan inodes to clean up\n");
2549                 return;
2550         }
2551
2552         if (bdev_read_only(sb->s_bdev)) {
2553                 ext4_msg(sb, KERN_ERR, "write access "
2554                         "unavailable, skipping orphan cleanup");
2555                 return;
2556         }
2557
2558         /* Check if feature set would not allow a r/w mount */
2559         if (!ext4_feature_set_ok(sb, 0)) {
2560                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2561                          "unknown ROCOMPAT features");
2562                 return;
2563         }
2564
2565         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2566                 /* don't clear list on RO mount w/ errors */
2567                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2568                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2569                                   "clearing orphan list.\n");
2570                         es->s_last_orphan = 0;
2571                 }
2572                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2573                 return;
2574         }
2575
2576         if (s_flags & SB_RDONLY) {
2577                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2578                 sb->s_flags &= ~SB_RDONLY;
2579         }
2580 #ifdef CONFIG_QUOTA
2581         /* Needed for iput() to work correctly and not trash data */
2582         sb->s_flags |= SB_ACTIVE;
2583
2584         /*
2585          * Turn on quotas which were not enabled for read-only mounts if
2586          * filesystem has quota feature, so that they are updated correctly.
2587          */
2588         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2589                 int ret = ext4_enable_quotas(sb);
2590
2591                 if (!ret)
2592                         quota_update = 1;
2593                 else
2594                         ext4_msg(sb, KERN_ERR,
2595                                 "Cannot turn on quotas: error %d", ret);
2596         }
2597
2598         /* Turn on journaled quotas used for old sytle */
2599         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2600                 if (EXT4_SB(sb)->s_qf_names[i]) {
2601                         int ret = ext4_quota_on_mount(sb, i);
2602
2603                         if (!ret)
2604                                 quota_update = 1;
2605                         else
2606                                 ext4_msg(sb, KERN_ERR,
2607                                         "Cannot turn on journaled "
2608                                         "quota: type %d: error %d", i, ret);
2609                 }
2610         }
2611 #endif
2612
2613         while (es->s_last_orphan) {
2614                 struct inode *inode;
2615
2616                 /*
2617                  * We may have encountered an error during cleanup; if
2618                  * so, skip the rest.
2619                  */
2620                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2621                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2622                         es->s_last_orphan = 0;
2623                         break;
2624                 }
2625
2626                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2627                 if (IS_ERR(inode)) {
2628                         es->s_last_orphan = 0;
2629                         break;
2630                 }
2631
2632                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2633                 dquot_initialize(inode);
2634                 if (inode->i_nlink) {
2635                         if (test_opt(sb, DEBUG))
2636                                 ext4_msg(sb, KERN_DEBUG,
2637                                         "%s: truncating inode %lu to %lld bytes",
2638                                         __func__, inode->i_ino, inode->i_size);
2639                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2640                                   inode->i_ino, inode->i_size);
2641                         inode_lock(inode);
2642                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2643                         ret = ext4_truncate(inode);
2644                         if (ret)
2645                                 ext4_std_error(inode->i_sb, ret);
2646                         inode_unlock(inode);
2647                         nr_truncates++;
2648                 } else {
2649                         if (test_opt(sb, DEBUG))
2650                                 ext4_msg(sb, KERN_DEBUG,
2651                                         "%s: deleting unreferenced inode %lu",
2652                                         __func__, inode->i_ino);
2653                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2654                                   inode->i_ino);
2655                         nr_orphans++;
2656                 }
2657                 iput(inode);  /* The delete magic happens here! */
2658         }
2659
2660 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2661
2662         if (nr_orphans)
2663                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2664                        PLURAL(nr_orphans));
2665         if (nr_truncates)
2666                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2667                        PLURAL(nr_truncates));
2668 #ifdef CONFIG_QUOTA
2669         /* Turn off quotas if they were enabled for orphan cleanup */
2670         if (quota_update) {
2671                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2672                         if (sb_dqopt(sb)->files[i])
2673                                 dquot_quota_off(sb, i);
2674                 }
2675         }
2676 #endif
2677         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2678 }
2679
2680 /*
2681  * Maximal extent format file size.
2682  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2683  * extent format containers, within a sector_t, and within i_blocks
2684  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2685  * so that won't be a limiting factor.
2686  *
2687  * However there is other limiting factor. We do store extents in the form
2688  * of starting block and length, hence the resulting length of the extent
2689  * covering maximum file size must fit into on-disk format containers as
2690  * well. Given that length is always by 1 unit bigger than max unit (because
2691  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2692  *
2693  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2694  */
2695 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2696 {
2697         loff_t res;
2698         loff_t upper_limit = MAX_LFS_FILESIZE;
2699
2700         /* small i_blocks in vfs inode? */
2701         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2702                 /*
2703                  * CONFIG_LBDAF is not enabled implies the inode
2704                  * i_block represent total blocks in 512 bytes
2705                  * 32 == size of vfs inode i_blocks * 8
2706                  */
2707                 upper_limit = (1LL << 32) - 1;
2708
2709                 /* total blocks in file system block size */
2710                 upper_limit >>= (blkbits - 9);
2711                 upper_limit <<= blkbits;
2712         }
2713
2714         /*
2715          * 32-bit extent-start container, ee_block. We lower the maxbytes
2716          * by one fs block, so ee_len can cover the extent of maximum file
2717          * size
2718          */
2719         res = (1LL << 32) - 1;
2720         res <<= blkbits;
2721
2722         /* Sanity check against vm- & vfs- imposed limits */
2723         if (res > upper_limit)
2724                 res = upper_limit;
2725
2726         return res;
2727 }
2728
2729 /*
2730  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2731  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2732  * We need to be 1 filesystem block less than the 2^48 sector limit.
2733  */
2734 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2735 {
2736         loff_t res = EXT4_NDIR_BLOCKS;
2737         int meta_blocks;
2738         loff_t upper_limit;
2739         /* This is calculated to be the largest file size for a dense, block
2740          * mapped file such that the file's total number of 512-byte sectors,
2741          * including data and all indirect blocks, does not exceed (2^48 - 1).
2742          *
2743          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2744          * number of 512-byte sectors of the file.
2745          */
2746
2747         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2748                 /*
2749                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2750                  * the inode i_block field represents total file blocks in
2751                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2752                  */
2753                 upper_limit = (1LL << 32) - 1;
2754
2755                 /* total blocks in file system block size */
2756                 upper_limit >>= (bits - 9);
2757
2758         } else {
2759                 /*
2760                  * We use 48 bit ext4_inode i_blocks
2761                  * With EXT4_HUGE_FILE_FL set the i_blocks
2762                  * represent total number of blocks in
2763                  * file system block size
2764                  */
2765                 upper_limit = (1LL << 48) - 1;
2766
2767         }
2768
2769         /* indirect blocks */
2770         meta_blocks = 1;
2771         /* double indirect blocks */
2772         meta_blocks += 1 + (1LL << (bits-2));
2773         /* tripple indirect blocks */
2774         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2775
2776         upper_limit -= meta_blocks;
2777         upper_limit <<= bits;
2778
2779         res += 1LL << (bits-2);
2780         res += 1LL << (2*(bits-2));
2781         res += 1LL << (3*(bits-2));
2782         res <<= bits;
2783         if (res > upper_limit)
2784                 res = upper_limit;
2785
2786         if (res > MAX_LFS_FILESIZE)
2787                 res = MAX_LFS_FILESIZE;
2788
2789         return res;
2790 }
2791
2792 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2793                                    ext4_fsblk_t logical_sb_block, int nr)
2794 {
2795         struct ext4_sb_info *sbi = EXT4_SB(sb);
2796         ext4_group_t bg, first_meta_bg;
2797         int has_super = 0;
2798
2799         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2800
2801         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2802                 return logical_sb_block + nr + 1;
2803         bg = sbi->s_desc_per_block * nr;
2804         if (ext4_bg_has_super(sb, bg))
2805                 has_super = 1;
2806
2807         /*
2808          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2809          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2810          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2811          * compensate.
2812          */
2813         if (sb->s_blocksize == 1024 && nr == 0 &&
2814             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2815                 has_super++;
2816
2817         return (has_super + ext4_group_first_block_no(sb, bg));
2818 }
2819
2820 /**
2821  * ext4_get_stripe_size: Get the stripe size.
2822  * @sbi: In memory super block info
2823  *
2824  * If we have specified it via mount option, then
2825  * use the mount option value. If the value specified at mount time is
2826  * greater than the blocks per group use the super block value.
2827  * If the super block value is greater than blocks per group return 0.
2828  * Allocator needs it be less than blocks per group.
2829  *
2830  */
2831 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2832 {
2833         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2834         unsigned long stripe_width =
2835                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2836         int ret;
2837
2838         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2839                 ret = sbi->s_stripe;
2840         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2841                 ret = stripe_width;
2842         else if (stride && stride <= sbi->s_blocks_per_group)
2843                 ret = stride;
2844         else
2845                 ret = 0;
2846
2847         /*
2848          * If the stripe width is 1, this makes no sense and
2849          * we set it to 0 to turn off stripe handling code.
2850          */
2851         if (ret <= 1)
2852                 ret = 0;
2853
2854         return ret;
2855 }
2856
2857 /*
2858  * Check whether this filesystem can be mounted based on
2859  * the features present and the RDONLY/RDWR mount requested.
2860  * Returns 1 if this filesystem can be mounted as requested,
2861  * 0 if it cannot be.
2862  */
2863 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2864 {
2865         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2866                 ext4_msg(sb, KERN_ERR,
2867                         "Couldn't mount because of "
2868                         "unsupported optional features (%x)",
2869                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2870                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2871                 return 0;
2872         }
2873
2874         if (readonly)
2875                 return 1;
2876
2877         if (ext4_has_feature_readonly(sb)) {
2878                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2879                 sb->s_flags |= SB_RDONLY;
2880                 return 1;
2881         }
2882
2883         /* Check that feature set is OK for a read-write mount */
2884         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2885                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2886                          "unsupported optional features (%x)",
2887                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2888                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2889                 return 0;
2890         }
2891         /*
2892          * Large file size enabled file system can only be mounted
2893          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2894          */
2895         if (ext4_has_feature_huge_file(sb)) {
2896                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2897                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2898                                  "cannot be mounted RDWR without "
2899                                  "CONFIG_LBDAF");
2900                         return 0;
2901                 }
2902         }
2903         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2904                 ext4_msg(sb, KERN_ERR,
2905                          "Can't support bigalloc feature without "
2906                          "extents feature\n");
2907                 return 0;
2908         }
2909
2910 #ifndef CONFIG_QUOTA
2911         if (ext4_has_feature_quota(sb) && !readonly) {
2912                 ext4_msg(sb, KERN_ERR,
2913                          "Filesystem with quota feature cannot be mounted RDWR "
2914                          "without CONFIG_QUOTA");
2915                 return 0;
2916         }
2917         if (ext4_has_feature_project(sb) && !readonly) {
2918                 ext4_msg(sb, KERN_ERR,
2919                          "Filesystem with project quota feature cannot be mounted RDWR "
2920                          "without CONFIG_QUOTA");
2921                 return 0;
2922         }
2923 #endif  /* CONFIG_QUOTA */
2924         return 1;
2925 }
2926
2927 /*
2928  * This function is called once a day if we have errors logged
2929  * on the file system
2930  */
2931 static void print_daily_error_info(struct timer_list *t)
2932 {
2933         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
2934         struct super_block *sb = sbi->s_sb;
2935         struct ext4_super_block *es = sbi->s_es;
2936
2937         if (es->s_error_count)
2938                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2939                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2940                          le32_to_cpu(es->s_error_count));
2941         if (es->s_first_error_time) {
2942                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
2943                        sb->s_id,
2944                        ext4_get_tstamp(es, s_first_error_time),
2945                        (int) sizeof(es->s_first_error_func),
2946                        es->s_first_error_func,
2947                        le32_to_cpu(es->s_first_error_line));
2948                 if (es->s_first_error_ino)
2949                         printk(KERN_CONT ": inode %u",
2950                                le32_to_cpu(es->s_first_error_ino));
2951                 if (es->s_first_error_block)
2952                         printk(KERN_CONT ": block %llu", (unsigned long long)
2953                                le64_to_cpu(es->s_first_error_block));
2954                 printk(KERN_CONT "\n");
2955         }
2956         if (es->s_last_error_time) {
2957                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
2958                        sb->s_id,
2959                        ext4_get_tstamp(es, s_last_error_time),
2960                        (int) sizeof(es->s_last_error_func),
2961                        es->s_last_error_func,
2962                        le32_to_cpu(es->s_last_error_line));
2963                 if (es->s_last_error_ino)
2964                         printk(KERN_CONT ": inode %u",
2965                                le32_to_cpu(es->s_last_error_ino));
2966                 if (es->s_last_error_block)
2967                         printk(KERN_CONT ": block %llu", (unsigned long long)
2968                                le64_to_cpu(es->s_last_error_block));
2969                 printk(KERN_CONT "\n");
2970         }
2971         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2972 }
2973
2974 /* Find next suitable group and run ext4_init_inode_table */
2975 static int ext4_run_li_request(struct ext4_li_request *elr)
2976 {
2977         struct ext4_group_desc *gdp = NULL;
2978         ext4_group_t group, ngroups;
2979         struct super_block *sb;
2980         unsigned long timeout = 0;
2981         int ret = 0;
2982
2983         sb = elr->lr_super;
2984         ngroups = EXT4_SB(sb)->s_groups_count;
2985
2986         for (group = elr->lr_next_group; group < ngroups; group++) {
2987                 gdp = ext4_get_group_desc(sb, group, NULL);
2988                 if (!gdp) {
2989                         ret = 1;
2990                         break;
2991                 }
2992
2993                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2994                         break;
2995         }
2996
2997         if (group >= ngroups)
2998                 ret = 1;
2999
3000         if (!ret) {
3001                 timeout = jiffies;
3002                 ret = ext4_init_inode_table(sb, group,
3003                                             elr->lr_timeout ? 0 : 1);
3004                 if (elr->lr_timeout == 0) {
3005                         timeout = (jiffies - timeout) *
3006                                   elr->lr_sbi->s_li_wait_mult;
3007                         elr->lr_timeout = timeout;
3008                 }
3009                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3010                 elr->lr_next_group = group + 1;
3011         }
3012         return ret;
3013 }
3014
3015 /*
3016  * Remove lr_request from the list_request and free the
3017  * request structure. Should be called with li_list_mtx held
3018  */
3019 static void ext4_remove_li_request(struct ext4_li_request *elr)
3020 {
3021         struct ext4_sb_info *sbi;
3022
3023         if (!elr)
3024                 return;
3025
3026         sbi = elr->lr_sbi;
3027
3028         list_del(&elr->lr_request);
3029         sbi->s_li_request = NULL;
3030         kfree(elr);
3031 }
3032
3033 static void ext4_unregister_li_request(struct super_block *sb)
3034 {
3035         mutex_lock(&ext4_li_mtx);
3036         if (!ext4_li_info) {
3037                 mutex_unlock(&ext4_li_mtx);
3038                 return;
3039         }
3040
3041         mutex_lock(&ext4_li_info->li_list_mtx);
3042         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3043         mutex_unlock(&ext4_li_info->li_list_mtx);
3044         mutex_unlock(&ext4_li_mtx);
3045 }
3046
3047 static struct task_struct *ext4_lazyinit_task;
3048
3049 /*
3050  * This is the function where ext4lazyinit thread lives. It walks
3051  * through the request list searching for next scheduled filesystem.
3052  * When such a fs is found, run the lazy initialization request
3053  * (ext4_rn_li_request) and keep track of the time spend in this
3054  * function. Based on that time we compute next schedule time of
3055  * the request. When walking through the list is complete, compute
3056  * next waking time and put itself into sleep.
3057  */
3058 static int ext4_lazyinit_thread(void *arg)
3059 {
3060         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3061         struct list_head *pos, *n;
3062         struct ext4_li_request *elr;
3063         unsigned long next_wakeup, cur;
3064
3065         BUG_ON(NULL == eli);
3066
3067 cont_thread:
3068         while (true) {
3069                 next_wakeup = MAX_JIFFY_OFFSET;
3070
3071                 mutex_lock(&eli->li_list_mtx);
3072                 if (list_empty(&eli->li_request_list)) {
3073                         mutex_unlock(&eli->li_list_mtx);
3074                         goto exit_thread;
3075                 }
3076                 list_for_each_safe(pos, n, &eli->li_request_list) {
3077                         int err = 0;
3078                         int progress = 0;
3079                         elr = list_entry(pos, struct ext4_li_request,
3080                                          lr_request);
3081
3082                         if (time_before(jiffies, elr->lr_next_sched)) {
3083                                 if (time_before(elr->lr_next_sched, next_wakeup))
3084                                         next_wakeup = elr->lr_next_sched;
3085                                 continue;
3086                         }
3087                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3088                                 if (sb_start_write_trylock(elr->lr_super)) {
3089                                         progress = 1;
3090                                         /*
3091                                          * We hold sb->s_umount, sb can not
3092                                          * be removed from the list, it is
3093                                          * now safe to drop li_list_mtx
3094                                          */
3095                                         mutex_unlock(&eli->li_list_mtx);
3096                                         err = ext4_run_li_request(elr);
3097                                         sb_end_write(elr->lr_super);
3098                                         mutex_lock(&eli->li_list_mtx);
3099                                         n = pos->next;
3100                                 }
3101                                 up_read((&elr->lr_super->s_umount));
3102                         }
3103                         /* error, remove the lazy_init job */
3104                         if (err) {
3105                                 ext4_remove_li_request(elr);
3106                                 continue;
3107                         }
3108                         if (!progress) {
3109                                 elr->lr_next_sched = jiffies +
3110                                         (prandom_u32()
3111                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3112                         }
3113                         if (time_before(elr->lr_next_sched, next_wakeup))
3114                                 next_wakeup = elr->lr_next_sched;
3115                 }
3116                 mutex_unlock(&eli->li_list_mtx);
3117
3118                 try_to_freeze();
3119
3120                 cur = jiffies;
3121                 if ((time_after_eq(cur, next_wakeup)) ||
3122                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3123                         cond_resched();
3124                         continue;
3125                 }
3126
3127                 schedule_timeout_interruptible(next_wakeup - cur);
3128
3129                 if (kthread_should_stop()) {
3130                         ext4_clear_request_list();
3131                         goto exit_thread;
3132                 }
3133         }
3134
3135 exit_thread:
3136         /*
3137          * It looks like the request list is empty, but we need
3138          * to check it under the li_list_mtx lock, to prevent any
3139          * additions into it, and of course we should lock ext4_li_mtx
3140          * to atomically free the list and ext4_li_info, because at
3141          * this point another ext4 filesystem could be registering
3142          * new one.
3143          */
3144         mutex_lock(&ext4_li_mtx);
3145         mutex_lock(&eli->li_list_mtx);
3146         if (!list_empty(&eli->li_request_list)) {
3147                 mutex_unlock(&eli->li_list_mtx);
3148                 mutex_unlock(&ext4_li_mtx);
3149                 goto cont_thread;
3150         }
3151         mutex_unlock(&eli->li_list_mtx);
3152         kfree(ext4_li_info);
3153         ext4_li_info = NULL;
3154         mutex_unlock(&ext4_li_mtx);
3155
3156         return 0;
3157 }
3158
3159 static void ext4_clear_request_list(void)
3160 {
3161         struct list_head *pos, *n;
3162         struct ext4_li_request *elr;
3163
3164         mutex_lock(&ext4_li_info->li_list_mtx);
3165         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3166                 elr = list_entry(pos, struct ext4_li_request,
3167                                  lr_request);
3168                 ext4_remove_li_request(elr);
3169         }
3170         mutex_unlock(&ext4_li_info->li_list_mtx);
3171 }
3172
3173 static int ext4_run_lazyinit_thread(void)
3174 {
3175         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3176                                          ext4_li_info, "ext4lazyinit");
3177         if (IS_ERR(ext4_lazyinit_task)) {
3178                 int err = PTR_ERR(ext4_lazyinit_task);
3179                 ext4_clear_request_list();
3180                 kfree(ext4_li_info);
3181                 ext4_li_info = NULL;
3182                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3183                                  "initialization thread\n",
3184                                  err);
3185                 return err;
3186         }
3187         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3188         return 0;
3189 }
3190
3191 /*
3192  * Check whether it make sense to run itable init. thread or not.
3193  * If there is at least one uninitialized inode table, return
3194  * corresponding group number, else the loop goes through all
3195  * groups and return total number of groups.
3196  */
3197 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3198 {
3199         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3200         struct ext4_group_desc *gdp = NULL;
3201
3202         if (!ext4_has_group_desc_csum(sb))
3203                 return ngroups;
3204
3205         for (group = 0; group < ngroups; group++) {
3206                 gdp = ext4_get_group_desc(sb, group, NULL);
3207                 if (!gdp)
3208                         continue;
3209
3210                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3211                         break;
3212         }
3213
3214         return group;
3215 }
3216
3217 static int ext4_li_info_new(void)
3218 {
3219         struct ext4_lazy_init *eli = NULL;
3220
3221         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3222         if (!eli)
3223                 return -ENOMEM;
3224
3225         INIT_LIST_HEAD(&eli->li_request_list);
3226         mutex_init(&eli->li_list_mtx);
3227
3228         eli->li_state |= EXT4_LAZYINIT_QUIT;
3229
3230         ext4_li_info = eli;
3231
3232         return 0;
3233 }
3234
3235 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3236                                             ext4_group_t start)
3237 {
3238         struct ext4_sb_info *sbi = EXT4_SB(sb);
3239         struct ext4_li_request *elr;
3240
3241         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3242         if (!elr)
3243                 return NULL;
3244
3245         elr->lr_super = sb;
3246         elr->lr_sbi = sbi;
3247         elr->lr_next_group = start;
3248
3249         /*
3250          * Randomize first schedule time of the request to
3251          * spread the inode table initialization requests
3252          * better.
3253          */
3254         elr->lr_next_sched = jiffies + (prandom_u32() %
3255                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3256         return elr;
3257 }
3258
3259 int ext4_register_li_request(struct super_block *sb,
3260                              ext4_group_t first_not_zeroed)
3261 {
3262         struct ext4_sb_info *sbi = EXT4_SB(sb);
3263         struct ext4_li_request *elr = NULL;
3264         ext4_group_t ngroups = sbi->s_groups_count;
3265         int ret = 0;
3266
3267         mutex_lock(&ext4_li_mtx);
3268         if (sbi->s_li_request != NULL) {
3269                 /*
3270                  * Reset timeout so it can be computed again, because
3271                  * s_li_wait_mult might have changed.
3272                  */
3273                 sbi->s_li_request->lr_timeout = 0;
3274                 goto out;
3275         }
3276
3277         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3278             !test_opt(sb, INIT_INODE_TABLE))
3279                 goto out;
3280
3281         elr = ext4_li_request_new(sb, first_not_zeroed);
3282         if (!elr) {
3283                 ret = -ENOMEM;
3284                 goto out;
3285         }
3286
3287         if (NULL == ext4_li_info) {
3288                 ret = ext4_li_info_new();
3289                 if (ret)
3290                         goto out;
3291         }
3292
3293         mutex_lock(&ext4_li_info->li_list_mtx);
3294         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3295         mutex_unlock(&ext4_li_info->li_list_mtx);
3296
3297         sbi->s_li_request = elr;
3298         /*
3299          * set elr to NULL here since it has been inserted to
3300          * the request_list and the removal and free of it is
3301          * handled by ext4_clear_request_list from now on.
3302          */
3303         elr = NULL;
3304
3305         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3306                 ret = ext4_run_lazyinit_thread();
3307                 if (ret)
3308                         goto out;
3309         }
3310 out:
3311         mutex_unlock(&ext4_li_mtx);
3312         if (ret)
3313                 kfree(elr);
3314         return ret;
3315 }
3316
3317 /*
3318  * We do not need to lock anything since this is called on
3319  * module unload.
3320  */
3321 static void ext4_destroy_lazyinit_thread(void)
3322 {
3323         /*
3324          * If thread exited earlier
3325          * there's nothing to be done.
3326          */
3327         if (!ext4_li_info || !ext4_lazyinit_task)
3328                 return;
3329
3330         kthread_stop(ext4_lazyinit_task);
3331 }
3332
3333 static int set_journal_csum_feature_set(struct super_block *sb)
3334 {
3335         int ret = 1;
3336         int compat, incompat;
3337         struct ext4_sb_info *sbi = EXT4_SB(sb);
3338
3339         if (ext4_has_metadata_csum(sb)) {
3340                 /* journal checksum v3 */
3341                 compat = 0;
3342                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3343         } else {
3344                 /* journal checksum v1 */
3345                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3346                 incompat = 0;
3347         }
3348
3349         jbd2_journal_clear_features(sbi->s_journal,
3350                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3351                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3352                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3353         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3354                 ret = jbd2_journal_set_features(sbi->s_journal,
3355                                 compat, 0,
3356                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3357                                 incompat);
3358         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3359                 ret = jbd2_journal_set_features(sbi->s_journal,
3360                                 compat, 0,
3361                                 incompat);
3362                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3363                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3364         } else {
3365                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3366                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3367         }
3368
3369         return ret;
3370 }
3371
3372 /*
3373  * Note: calculating the overhead so we can be compatible with
3374  * historical BSD practice is quite difficult in the face of
3375  * clusters/bigalloc.  This is because multiple metadata blocks from
3376  * different block group can end up in the same allocation cluster.
3377  * Calculating the exact overhead in the face of clustered allocation
3378  * requires either O(all block bitmaps) in memory or O(number of block
3379  * groups**2) in time.  We will still calculate the superblock for
3380  * older file systems --- and if we come across with a bigalloc file
3381  * system with zero in s_overhead_clusters the estimate will be close to
3382  * correct especially for very large cluster sizes --- but for newer
3383  * file systems, it's better to calculate this figure once at mkfs
3384  * time, and store it in the superblock.  If the superblock value is
3385  * present (even for non-bigalloc file systems), we will use it.
3386  */
3387 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3388                           char *buf)
3389 {
3390         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3391         struct ext4_group_desc  *gdp;
3392         ext4_fsblk_t            first_block, last_block, b;
3393         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3394         int                     s, j, count = 0;
3395
3396         if (!ext4_has_feature_bigalloc(sb))
3397                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3398                         sbi->s_itb_per_group + 2);
3399
3400         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3401                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3402         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3403         for (i = 0; i < ngroups; i++) {
3404                 gdp = ext4_get_group_desc(sb, i, NULL);
3405                 b = ext4_block_bitmap(sb, gdp);
3406                 if (b >= first_block && b <= last_block) {
3407                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3408                         count++;
3409                 }
3410                 b = ext4_inode_bitmap(sb, gdp);
3411                 if (b >= first_block && b <= last_block) {
3412                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3413                         count++;
3414                 }
3415                 b = ext4_inode_table(sb, gdp);
3416                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3417                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3418                                 int c = EXT4_B2C(sbi, b - first_block);
3419                                 ext4_set_bit(c, buf);
3420                                 count++;
3421                         }
3422                 if (i != grp)
3423                         continue;
3424                 s = 0;
3425                 if (ext4_bg_has_super(sb, grp)) {
3426                         ext4_set_bit(s++, buf);
3427                         count++;
3428                 }
3429                 j = ext4_bg_num_gdb(sb, grp);
3430                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3431                         ext4_error(sb, "Invalid number of block group "
3432                                    "descriptor blocks: %d", j);
3433                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3434                 }
3435                 count += j;
3436                 for (; j > 0; j--)
3437                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3438         }
3439         if (!count)
3440                 return 0;
3441         return EXT4_CLUSTERS_PER_GROUP(sb) -
3442                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3443 }
3444
3445 /*
3446  * Compute the overhead and stash it in sbi->s_overhead
3447  */
3448 int ext4_calculate_overhead(struct super_block *sb)
3449 {
3450         struct ext4_sb_info *sbi = EXT4_SB(sb);
3451         struct ext4_super_block *es = sbi->s_es;
3452         struct inode *j_inode;
3453         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3454         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3455         ext4_fsblk_t overhead = 0;
3456         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3457
3458         if (!buf)
3459                 return -ENOMEM;
3460
3461         /*
3462          * Compute the overhead (FS structures).  This is constant
3463          * for a given filesystem unless the number of block groups
3464          * changes so we cache the previous value until it does.
3465          */
3466
3467         /*
3468          * All of the blocks before first_data_block are overhead
3469          */
3470         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3471
3472         /*
3473          * Add the overhead found in each block group
3474          */
3475         for (i = 0; i < ngroups; i++) {
3476                 int blks;
3477
3478                 blks = count_overhead(sb, i, buf);
3479                 overhead += blks;
3480                 if (blks)
3481                         memset(buf, 0, PAGE_SIZE);
3482                 cond_resched();
3483         }
3484
3485         /*
3486          * Add the internal journal blocks whether the journal has been
3487          * loaded or not
3488          */
3489         if (sbi->s_journal && !sbi->journal_bdev)
3490                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3491         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3492                 j_inode = ext4_get_journal_inode(sb, j_inum);
3493                 if (j_inode) {
3494                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3495                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3496                         iput(j_inode);
3497                 } else {
3498                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3499                 }
3500         }
3501         sbi->s_overhead = overhead;
3502         smp_wmb();
3503         free_page((unsigned long) buf);
3504         return 0;
3505 }
3506
3507 static void ext4_set_resv_clusters(struct super_block *sb)
3508 {
3509         ext4_fsblk_t resv_clusters;
3510         struct ext4_sb_info *sbi = EXT4_SB(sb);
3511
3512         /*
3513          * There's no need to reserve anything when we aren't using extents.
3514          * The space estimates are exact, there are no unwritten extents,
3515          * hole punching doesn't need new metadata... This is needed especially
3516          * to keep ext2/3 backward compatibility.
3517          */
3518         if (!ext4_has_feature_extents(sb))
3519                 return;
3520         /*
3521          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3522          * This should cover the situations where we can not afford to run
3523          * out of space like for example punch hole, or converting
3524          * unwritten extents in delalloc path. In most cases such
3525          * allocation would require 1, or 2 blocks, higher numbers are
3526          * very rare.
3527          */
3528         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3529                          sbi->s_cluster_bits);
3530
3531         do_div(resv_clusters, 50);
3532         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3533
3534         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3535 }
3536
3537 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3538 {
3539         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3540         char *orig_data = kstrdup(data, GFP_KERNEL);
3541         struct buffer_head *bh;
3542         struct ext4_super_block *es = NULL;
3543         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3544         ext4_fsblk_t block;
3545         ext4_fsblk_t sb_block = get_sb_block(&data);
3546         ext4_fsblk_t logical_sb_block;
3547         unsigned long offset = 0;
3548         unsigned long journal_devnum = 0;
3549         unsigned long def_mount_opts;
3550         struct inode *root;
3551         const char *descr;
3552         int ret = -ENOMEM;
3553         int blocksize, clustersize;
3554         unsigned int db_count;
3555         unsigned int i;
3556         int needs_recovery, has_huge_files, has_bigalloc;
3557         __u64 blocks_count;
3558         int err = 0;
3559         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3560         ext4_group_t first_not_zeroed;
3561
3562         if ((data && !orig_data) || !sbi)
3563                 goto out_free_base;
3564
3565         sbi->s_daxdev = dax_dev;
3566         sbi->s_blockgroup_lock =
3567                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3568         if (!sbi->s_blockgroup_lock)
3569                 goto out_free_base;
3570
3571         sb->s_fs_info = sbi;
3572         sbi->s_sb = sb;
3573         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3574         sbi->s_sb_block = sb_block;
3575         if (sb->s_bdev->bd_part)
3576                 sbi->s_sectors_written_start =
3577                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3578
3579         /* Cleanup superblock name */
3580         strreplace(sb->s_id, '/', '!');
3581
3582         /* -EINVAL is default */
3583         ret = -EINVAL;
3584         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3585         if (!blocksize) {
3586                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3587                 goto out_fail;
3588         }
3589
3590         /*
3591          * The ext4 superblock will not be buffer aligned for other than 1kB
3592          * block sizes.  We need to calculate the offset from buffer start.
3593          */
3594         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3595                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3596                 offset = do_div(logical_sb_block, blocksize);
3597         } else {
3598                 logical_sb_block = sb_block;
3599         }
3600
3601         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3602                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3603                 goto out_fail;
3604         }
3605         /*
3606          * Note: s_es must be initialized as soon as possible because
3607          *       some ext4 macro-instructions depend on its value
3608          */
3609         es = (struct ext4_super_block *) (bh->b_data + offset);
3610         sbi->s_es = es;
3611         sb->s_magic = le16_to_cpu(es->s_magic);
3612         if (sb->s_magic != EXT4_SUPER_MAGIC)
3613                 goto cantfind_ext4;
3614         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3615
3616         /* Warn if metadata_csum and gdt_csum are both set. */
3617         if (ext4_has_feature_metadata_csum(sb) &&
3618             ext4_has_feature_gdt_csum(sb))
3619                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3620                              "redundant flags; please run fsck.");
3621
3622         /* Check for a known checksum algorithm */
3623         if (!ext4_verify_csum_type(sb, es)) {
3624                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3625                          "unknown checksum algorithm.");
3626                 silent = 1;
3627                 goto cantfind_ext4;
3628         }
3629
3630         /* Load the checksum driver */
3631         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3632         if (IS_ERR(sbi->s_chksum_driver)) {
3633                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3634                 ret = PTR_ERR(sbi->s_chksum_driver);
3635                 sbi->s_chksum_driver = NULL;
3636                 goto failed_mount;
3637         }
3638
3639         /* Check superblock checksum */
3640         if (!ext4_superblock_csum_verify(sb, es)) {
3641                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3642                          "invalid superblock checksum.  Run e2fsck?");
3643                 silent = 1;
3644                 ret = -EFSBADCRC;
3645                 goto cantfind_ext4;
3646         }
3647
3648         /* Precompute checksum seed for all metadata */
3649         if (ext4_has_feature_csum_seed(sb))
3650                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3651         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3652                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3653                                                sizeof(es->s_uuid));
3654
3655         /* Set defaults before we parse the mount options */
3656         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3657         set_opt(sb, INIT_INODE_TABLE);
3658         if (def_mount_opts & EXT4_DEFM_DEBUG)
3659                 set_opt(sb, DEBUG);
3660         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3661                 set_opt(sb, GRPID);
3662         if (def_mount_opts & EXT4_DEFM_UID16)
3663                 set_opt(sb, NO_UID32);
3664         /* xattr user namespace & acls are now defaulted on */
3665         set_opt(sb, XATTR_USER);
3666 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3667         set_opt(sb, POSIX_ACL);
3668 #endif
3669         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3670         if (ext4_has_metadata_csum(sb))
3671                 set_opt(sb, JOURNAL_CHECKSUM);
3672
3673         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3674                 set_opt(sb, JOURNAL_DATA);
3675         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3676                 set_opt(sb, ORDERED_DATA);
3677         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3678                 set_opt(sb, WRITEBACK_DATA);
3679
3680         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3681                 set_opt(sb, ERRORS_PANIC);
3682         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3683                 set_opt(sb, ERRORS_CONT);
3684         else
3685                 set_opt(sb, ERRORS_RO);
3686         /* block_validity enabled by default; disable with noblock_validity */
3687         set_opt(sb, BLOCK_VALIDITY);
3688         if (def_mount_opts & EXT4_DEFM_DISCARD)
3689                 set_opt(sb, DISCARD);
3690
3691         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3692         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3693         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3694         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3695         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3696
3697         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3698                 set_opt(sb, BARRIER);
3699
3700         /*
3701          * enable delayed allocation by default
3702          * Use -o nodelalloc to turn it off
3703          */
3704         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3705             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3706                 set_opt(sb, DELALLOC);
3707
3708         /*
3709          * set default s_li_wait_mult for lazyinit, for the case there is
3710          * no mount option specified.
3711          */
3712         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3713
3714         if (sbi->s_es->s_mount_opts[0]) {
3715                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3716                                               sizeof(sbi->s_es->s_mount_opts),
3717                                               GFP_KERNEL);
3718                 if (!s_mount_opts)
3719                         goto failed_mount;
3720                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3721                                    &journal_ioprio, 0)) {
3722                         ext4_msg(sb, KERN_WARNING,
3723                                  "failed to parse options in superblock: %s",
3724                                  s_mount_opts);
3725                 }
3726                 kfree(s_mount_opts);
3727         }
3728         sbi->s_def_mount_opt = sbi->s_mount_opt;
3729         if (!parse_options((char *) data, sb, &journal_devnum,
3730                            &journal_ioprio, 0))
3731                 goto failed_mount;
3732
3733         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3734                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3735                             "with data=journal disables delayed "
3736                             "allocation and O_DIRECT support!\n");
3737                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3738                         ext4_msg(sb, KERN_ERR, "can't mount with "
3739                                  "both data=journal and delalloc");
3740                         goto failed_mount;
3741                 }
3742                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3743                         ext4_msg(sb, KERN_ERR, "can't mount with "
3744                                  "both data=journal and dioread_nolock");
3745                         goto failed_mount;
3746                 }
3747                 if (test_opt(sb, DAX)) {
3748                         ext4_msg(sb, KERN_ERR, "can't mount with "
3749                                  "both data=journal and dax");
3750                         goto failed_mount;
3751                 }
3752                 if (ext4_has_feature_encrypt(sb)) {
3753                         ext4_msg(sb, KERN_WARNING,
3754                                  "encrypted files will use data=ordered "
3755                                  "instead of data journaling mode");
3756                 }
3757                 if (test_opt(sb, DELALLOC))
3758                         clear_opt(sb, DELALLOC);
3759         } else {
3760                 sb->s_iflags |= SB_I_CGROUPWB;
3761         }
3762
3763         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3764                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3765
3766         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3767             (ext4_has_compat_features(sb) ||
3768              ext4_has_ro_compat_features(sb) ||
3769              ext4_has_incompat_features(sb)))
3770                 ext4_msg(sb, KERN_WARNING,
3771                        "feature flags set on rev 0 fs, "
3772                        "running e2fsck is recommended");
3773
3774         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3775                 set_opt2(sb, HURD_COMPAT);
3776                 if (ext4_has_feature_64bit(sb)) {
3777                         ext4_msg(sb, KERN_ERR,
3778                                  "The Hurd can't support 64-bit file systems");
3779                         goto failed_mount;
3780                 }
3781
3782                 /*
3783                  * ea_inode feature uses l_i_version field which is not
3784                  * available in HURD_COMPAT mode.
3785                  */
3786                 if (ext4_has_feature_ea_inode(sb)) {
3787                         ext4_msg(sb, KERN_ERR,
3788                                  "ea_inode feature is not supported for Hurd");
3789                         goto failed_mount;
3790                 }
3791         }
3792
3793         if (IS_EXT2_SB(sb)) {
3794                 if (ext2_feature_set_ok(sb))
3795                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3796                                  "using the ext4 subsystem");
3797                 else {
3798                         /*
3799                          * If we're probing be silent, if this looks like
3800                          * it's actually an ext[34] filesystem.
3801                          */
3802                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3803                                 goto failed_mount;
3804                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3805                                  "to feature incompatibilities");
3806                         goto failed_mount;
3807                 }
3808         }
3809
3810         if (IS_EXT3_SB(sb)) {
3811                 if (ext3_feature_set_ok(sb))
3812                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3813                                  "using the ext4 subsystem");
3814                 else {
3815                         /*
3816                          * If we're probing be silent, if this looks like
3817                          * it's actually an ext4 filesystem.
3818                          */
3819                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3820                                 goto failed_mount;
3821                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3822                                  "to feature incompatibilities");
3823                         goto failed_mount;
3824                 }
3825         }
3826
3827         /*
3828          * Check feature flags regardless of the revision level, since we
3829          * previously didn't change the revision level when setting the flags,
3830          * so there is a chance incompat flags are set on a rev 0 filesystem.
3831          */
3832         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3833                 goto failed_mount;
3834
3835         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3836         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3837             blocksize > EXT4_MAX_BLOCK_SIZE) {
3838                 ext4_msg(sb, KERN_ERR,
3839                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3840                          blocksize, le32_to_cpu(es->s_log_block_size));
3841                 goto failed_mount;
3842         }
3843         if (le32_to_cpu(es->s_log_block_size) >
3844             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3845                 ext4_msg(sb, KERN_ERR,
3846                          "Invalid log block size: %u",
3847                          le32_to_cpu(es->s_log_block_size));
3848                 goto failed_mount;
3849         }
3850         if (le32_to_cpu(es->s_log_cluster_size) >
3851             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3852                 ext4_msg(sb, KERN_ERR,
3853                          "Invalid log cluster size: %u",
3854                          le32_to_cpu(es->s_log_cluster_size));
3855                 goto failed_mount;
3856         }
3857
3858         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3859                 ext4_msg(sb, KERN_ERR,
3860                          "Number of reserved GDT blocks insanely large: %d",
3861                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3862                 goto failed_mount;
3863         }
3864
3865         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3866                 if (ext4_has_feature_inline_data(sb)) {
3867                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3868                                         " that may contain inline data");
3869                         goto failed_mount;
3870                 }
3871                 if (!bdev_dax_supported(sb->s_bdev, blocksize)) {
3872                         ext4_msg(sb, KERN_ERR,
3873                                 "DAX unsupported by block device.");
3874                         goto failed_mount;
3875                 }
3876         }
3877
3878         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3879                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3880                          es->s_encryption_level);
3881                 goto failed_mount;
3882         }
3883
3884         if (sb->s_blocksize != blocksize) {
3885                 /* Validate the filesystem blocksize */
3886                 if (!sb_set_blocksize(sb, blocksize)) {
3887                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3888                                         blocksize);
3889                         goto failed_mount;
3890                 }
3891
3892                 brelse(bh);
3893                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3894                 offset = do_div(logical_sb_block, blocksize);
3895                 bh = sb_bread_unmovable(sb, logical_sb_block);
3896                 if (!bh) {
3897                         ext4_msg(sb, KERN_ERR,
3898                                "Can't read superblock on 2nd try");
3899                         goto failed_mount;
3900                 }
3901                 es = (struct ext4_super_block *)(bh->b_data + offset);
3902                 sbi->s_es = es;
3903                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3904                         ext4_msg(sb, KERN_ERR,
3905                                "Magic mismatch, very weird!");
3906                         goto failed_mount;
3907                 }
3908         }
3909
3910         has_huge_files = ext4_has_feature_huge_file(sb);
3911         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3912                                                       has_huge_files);
3913         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3914
3915         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3916                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3917                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3918         } else {
3919                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3920                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3921                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3922                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3923                                  sbi->s_first_ino);
3924                         goto failed_mount;
3925                 }
3926                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3927                     (!is_power_of_2(sbi->s_inode_size)) ||
3928                     (sbi->s_inode_size > blocksize)) {
3929                         ext4_msg(sb, KERN_ERR,
3930                                "unsupported inode size: %d",
3931                                sbi->s_inode_size);
3932                         goto failed_mount;
3933                 }
3934                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3935                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3936         }
3937
3938         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3939         if (ext4_has_feature_64bit(sb)) {
3940                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3941                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3942                     !is_power_of_2(sbi->s_desc_size)) {
3943                         ext4_msg(sb, KERN_ERR,
3944                                "unsupported descriptor size %lu",
3945                                sbi->s_desc_size);
3946                         goto failed_mount;
3947                 }
3948         } else
3949                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3950
3951         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3952         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3953
3954         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3955         if (sbi->s_inodes_per_block == 0)
3956                 goto cantfind_ext4;
3957         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3958             sbi->s_inodes_per_group > blocksize * 8) {
3959                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3960                          sbi->s_blocks_per_group);
3961                 goto failed_mount;
3962         }
3963         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3964                                         sbi->s_inodes_per_block;
3965         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3966         sbi->s_sbh = bh;
3967         sbi->s_mount_state = le16_to_cpu(es->s_state);
3968         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3969         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3970
3971         for (i = 0; i < 4; i++)
3972                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3973         sbi->s_def_hash_version = es->s_def_hash_version;
3974         if (ext4_has_feature_dir_index(sb)) {
3975                 i = le32_to_cpu(es->s_flags);
3976                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3977                         sbi->s_hash_unsigned = 3;
3978                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3979 #ifdef __CHAR_UNSIGNED__
3980                         if (!sb_rdonly(sb))
3981                                 es->s_flags |=
3982                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3983                         sbi->s_hash_unsigned = 3;
3984 #else
3985                         if (!sb_rdonly(sb))
3986                                 es->s_flags |=
3987                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3988 #endif
3989                 }
3990         }
3991
3992         /* Handle clustersize */
3993         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3994         has_bigalloc = ext4_has_feature_bigalloc(sb);
3995         if (has_bigalloc) {
3996                 if (clustersize < blocksize) {
3997                         ext4_msg(sb, KERN_ERR,
3998                                  "cluster size (%d) smaller than "
3999                                  "block size (%d)", clustersize, blocksize);
4000                         goto failed_mount;
4001                 }
4002                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4003                         le32_to_cpu(es->s_log_block_size);
4004                 sbi->s_clusters_per_group =
4005                         le32_to_cpu(es->s_clusters_per_group);
4006                 if (sbi->s_clusters_per_group > blocksize * 8) {
4007                         ext4_msg(sb, KERN_ERR,
4008                                  "#clusters per group too big: %lu",
4009                                  sbi->s_clusters_per_group);
4010                         goto failed_mount;
4011                 }
4012                 if (sbi->s_blocks_per_group !=
4013                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4014                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4015                                  "clusters per group (%lu) inconsistent",
4016                                  sbi->s_blocks_per_group,
4017                                  sbi->s_clusters_per_group);
4018                         goto failed_mount;
4019                 }
4020         } else {
4021                 if (clustersize != blocksize) {
4022                         ext4_msg(sb, KERN_ERR,
4023                                  "fragment/cluster size (%d) != "
4024                                  "block size (%d)", clustersize, blocksize);
4025                         goto failed_mount;
4026                 }
4027                 if (sbi->s_blocks_per_group > blocksize * 8) {
4028                         ext4_msg(sb, KERN_ERR,
4029                                  "#blocks per group too big: %lu",
4030                                  sbi->s_blocks_per_group);
4031                         goto failed_mount;
4032                 }
4033                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4034                 sbi->s_cluster_bits = 0;
4035         }
4036         sbi->s_cluster_ratio = clustersize / blocksize;
4037
4038         /* Do we have standard group size of clustersize * 8 blocks ? */
4039         if (sbi->s_blocks_per_group == clustersize << 3)
4040                 set_opt2(sb, STD_GROUP_SIZE);
4041
4042         /*
4043          * Test whether we have more sectors than will fit in sector_t,
4044          * and whether the max offset is addressable by the page cache.
4045          */
4046         err = generic_check_addressable(sb->s_blocksize_bits,
4047                                         ext4_blocks_count(es));
4048         if (err) {
4049                 ext4_msg(sb, KERN_ERR, "filesystem"
4050                          " too large to mount safely on this system");
4051                 if (sizeof(sector_t) < 8)
4052                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
4053                 goto failed_mount;
4054         }
4055
4056         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4057                 goto cantfind_ext4;
4058
4059         /* check blocks count against device size */
4060         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4061         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4062                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4063                        "exceeds size of device (%llu blocks)",
4064                        ext4_blocks_count(es), blocks_count);
4065                 goto failed_mount;
4066         }
4067
4068         /*
4069          * It makes no sense for the first data block to be beyond the end
4070          * of the filesystem.
4071          */
4072         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4073                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4074                          "block %u is beyond end of filesystem (%llu)",
4075                          le32_to_cpu(es->s_first_data_block),
4076                          ext4_blocks_count(es));
4077                 goto failed_mount;
4078         }
4079         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4080             (sbi->s_cluster_ratio == 1)) {
4081                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4082                          "block is 0 with a 1k block and cluster size");
4083                 goto failed_mount;
4084         }
4085
4086         blocks_count = (ext4_blocks_count(es) -
4087                         le32_to_cpu(es->s_first_data_block) +
4088                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4089         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4090         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4091                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
4092                        "(block count %llu, first data block %u, "
4093                        "blocks per group %lu)", sbi->s_groups_count,
4094                        ext4_blocks_count(es),
4095                        le32_to_cpu(es->s_first_data_block),
4096                        EXT4_BLOCKS_PER_GROUP(sb));
4097                 goto failed_mount;
4098         }
4099         sbi->s_groups_count = blocks_count;
4100         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4101                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4102         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4103             le32_to_cpu(es->s_inodes_count)) {
4104                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4105                          le32_to_cpu(es->s_inodes_count),
4106                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4107                 ret = -EINVAL;
4108                 goto failed_mount;
4109         }
4110         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4111                    EXT4_DESC_PER_BLOCK(sb);
4112         if (ext4_has_feature_meta_bg(sb)) {
4113                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4114                         ext4_msg(sb, KERN_WARNING,
4115                                  "first meta block group too large: %u "
4116                                  "(group descriptor block count %u)",
4117                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4118                         goto failed_mount;
4119                 }
4120         }
4121         sbi->s_group_desc = kvmalloc_array(db_count,
4122                                            sizeof(struct buffer_head *),
4123                                            GFP_KERNEL);
4124         if (sbi->s_group_desc == NULL) {
4125                 ext4_msg(sb, KERN_ERR, "not enough memory");
4126                 ret = -ENOMEM;
4127                 goto failed_mount;
4128         }
4129
4130         bgl_lock_init(sbi->s_blockgroup_lock);
4131
4132         /* Pre-read the descriptors into the buffer cache */
4133         for (i = 0; i < db_count; i++) {
4134                 block = descriptor_loc(sb, logical_sb_block, i);
4135                 sb_breadahead(sb, block);
4136         }
4137
4138         for (i = 0; i < db_count; i++) {
4139                 block = descriptor_loc(sb, logical_sb_block, i);
4140                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
4141                 if (!sbi->s_group_desc[i]) {
4142                         ext4_msg(sb, KERN_ERR,
4143                                "can't read group descriptor %d", i);
4144                         db_count = i;
4145                         goto failed_mount2;
4146                 }
4147         }
4148         sbi->s_gdb_count = db_count;
4149         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4150                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4151                 ret = -EFSCORRUPTED;
4152                 goto failed_mount2;
4153         }
4154
4155         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4156
4157         /* Register extent status tree shrinker */
4158         if (ext4_es_register_shrinker(sbi))
4159                 goto failed_mount3;
4160
4161         sbi->s_stripe = ext4_get_stripe_size(sbi);
4162         sbi->s_extent_max_zeroout_kb = 32;
4163
4164         /*
4165          * set up enough so that it can read an inode
4166          */
4167         sb->s_op = &ext4_sops;
4168         sb->s_export_op = &ext4_export_ops;
4169         sb->s_xattr = ext4_xattr_handlers;
4170 #ifdef CONFIG_EXT4_FS_ENCRYPTION
4171         sb->s_cop = &ext4_cryptops;
4172 #endif
4173 #ifdef CONFIG_QUOTA
4174         sb->dq_op = &ext4_quota_operations;
4175         if (ext4_has_feature_quota(sb))
4176                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4177         else
4178                 sb->s_qcop = &ext4_qctl_operations;
4179         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4180 #endif
4181         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4182
4183         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4184         mutex_init(&sbi->s_orphan_lock);
4185
4186         sb->s_root = NULL;
4187
4188         needs_recovery = (es->s_last_orphan != 0 ||
4189                           ext4_has_feature_journal_needs_recovery(sb));
4190
4191         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4192                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4193                         goto failed_mount3a;
4194
4195         /*
4196          * The first inode we look at is the journal inode.  Don't try
4197          * root first: it may be modified in the journal!
4198          */
4199         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4200                 err = ext4_load_journal(sb, es, journal_devnum);
4201                 if (err)
4202                         goto failed_mount3a;
4203         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4204                    ext4_has_feature_journal_needs_recovery(sb)) {
4205                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4206                        "suppressed and not mounted read-only");
4207                 goto failed_mount_wq;
4208         } else {
4209                 /* Nojournal mode, all journal mount options are illegal */
4210                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4211                         ext4_msg(sb, KERN_ERR, "can't mount with "
4212                                  "journal_checksum, fs mounted w/o journal");
4213                         goto failed_mount_wq;
4214                 }
4215                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4216                         ext4_msg(sb, KERN_ERR, "can't mount with "
4217                                  "journal_async_commit, fs mounted w/o journal");
4218                         goto failed_mount_wq;
4219                 }
4220                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4221                         ext4_msg(sb, KERN_ERR, "can't mount with "
4222                                  "commit=%lu, fs mounted w/o journal",
4223                                  sbi->s_commit_interval / HZ);
4224                         goto failed_mount_wq;
4225                 }
4226                 if (EXT4_MOUNT_DATA_FLAGS &
4227                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4228                         ext4_msg(sb, KERN_ERR, "can't mount with "
4229                                  "data=, fs mounted w/o journal");
4230                         goto failed_mount_wq;
4231                 }
4232                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4233                 clear_opt(sb, JOURNAL_CHECKSUM);
4234                 clear_opt(sb, DATA_FLAGS);
4235                 sbi->s_journal = NULL;
4236                 needs_recovery = 0;
4237                 goto no_journal;
4238         }
4239
4240         if (ext4_has_feature_64bit(sb) &&
4241             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4242                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4243                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4244                 goto failed_mount_wq;
4245         }
4246
4247         if (!set_journal_csum_feature_set(sb)) {
4248                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4249                          "feature set");
4250                 goto failed_mount_wq;
4251         }
4252
4253         /* We have now updated the journal if required, so we can
4254          * validate the data journaling mode. */
4255         switch (test_opt(sb, DATA_FLAGS)) {
4256         case 0:
4257                 /* No mode set, assume a default based on the journal
4258                  * capabilities: ORDERED_DATA if the journal can
4259                  * cope, else JOURNAL_DATA
4260                  */
4261                 if (jbd2_journal_check_available_features
4262                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4263                         set_opt(sb, ORDERED_DATA);
4264                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4265                 } else {
4266                         set_opt(sb, JOURNAL_DATA);
4267                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4268                 }
4269                 break;
4270
4271         case EXT4_MOUNT_ORDERED_DATA:
4272         case EXT4_MOUNT_WRITEBACK_DATA:
4273                 if (!jbd2_journal_check_available_features
4274                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4275                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4276                                "requested data journaling mode");
4277                         goto failed_mount_wq;
4278                 }
4279         default:
4280                 break;
4281         }
4282
4283         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4284             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4285                 ext4_msg(sb, KERN_ERR, "can't mount with "
4286                         "journal_async_commit in data=ordered mode");
4287                 goto failed_mount_wq;
4288         }
4289
4290         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4291
4292         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4293
4294 no_journal:
4295         if (!test_opt(sb, NO_MBCACHE)) {
4296                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4297                 if (!sbi->s_ea_block_cache) {
4298                         ext4_msg(sb, KERN_ERR,
4299                                  "Failed to create ea_block_cache");
4300                         goto failed_mount_wq;
4301                 }
4302
4303                 if (ext4_has_feature_ea_inode(sb)) {
4304                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4305                         if (!sbi->s_ea_inode_cache) {
4306                                 ext4_msg(sb, KERN_ERR,
4307                                          "Failed to create ea_inode_cache");
4308                                 goto failed_mount_wq;
4309                         }
4310                 }
4311         }
4312
4313         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4314             (blocksize != PAGE_SIZE)) {
4315                 ext4_msg(sb, KERN_ERR,
4316                          "Unsupported blocksize for fs encryption");
4317                 goto failed_mount_wq;
4318         }
4319
4320         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4321             !ext4_has_feature_encrypt(sb)) {
4322                 ext4_set_feature_encrypt(sb);
4323                 ext4_commit_super(sb, 1);
4324         }
4325
4326         /*
4327          * Get the # of file system overhead blocks from the
4328          * superblock if present.
4329          */
4330         if (es->s_overhead_clusters)
4331                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4332         else {
4333                 err = ext4_calculate_overhead(sb);
4334                 if (err)
4335                         goto failed_mount_wq;
4336         }
4337
4338         /*
4339          * The maximum number of concurrent works can be high and
4340          * concurrency isn't really necessary.  Limit it to 1.
4341          */
4342         EXT4_SB(sb)->rsv_conversion_wq =
4343                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4344         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4345                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4346                 ret = -ENOMEM;
4347                 goto failed_mount4;
4348         }
4349
4350         /*
4351          * The jbd2_journal_load will have done any necessary log recovery,
4352          * so we can safely mount the rest of the filesystem now.
4353          */
4354
4355         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4356         if (IS_ERR(root)) {
4357                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4358                 ret = PTR_ERR(root);
4359                 root = NULL;
4360                 goto failed_mount4;
4361         }
4362         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4363                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4364                 iput(root);
4365                 goto failed_mount4;
4366         }
4367         sb->s_root = d_make_root(root);
4368         if (!sb->s_root) {
4369                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4370                 ret = -ENOMEM;
4371                 goto failed_mount4;
4372         }
4373
4374         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4375         if (ret == -EROFS) {
4376                 sb->s_flags |= SB_RDONLY;
4377                 ret = 0;
4378         } else if (ret)
4379                 goto failed_mount4a;
4380
4381         /* determine the minimum size of new large inodes, if present */
4382         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4383             sbi->s_want_extra_isize == 0) {
4384                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4385                                                      EXT4_GOOD_OLD_INODE_SIZE;
4386                 if (ext4_has_feature_extra_isize(sb)) {
4387                         if (sbi->s_want_extra_isize <
4388                             le16_to_cpu(es->s_want_extra_isize))
4389                                 sbi->s_want_extra_isize =
4390                                         le16_to_cpu(es->s_want_extra_isize);
4391                         if (sbi->s_want_extra_isize <
4392                             le16_to_cpu(es->s_min_extra_isize))
4393                                 sbi->s_want_extra_isize =
4394                                         le16_to_cpu(es->s_min_extra_isize);
4395                 }
4396         }
4397         /* Check if enough inode space is available */
4398         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4399                                                         sbi->s_inode_size) {
4400                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4401                                                        EXT4_GOOD_OLD_INODE_SIZE;
4402                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4403                          "available");
4404         }
4405
4406         ext4_set_resv_clusters(sb);
4407
4408         err = ext4_setup_system_zone(sb);
4409         if (err) {
4410                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4411                          "zone (%d)", err);
4412                 goto failed_mount4a;
4413         }
4414
4415         ext4_ext_init(sb);
4416         err = ext4_mb_init(sb);
4417         if (err) {
4418                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4419                          err);
4420                 goto failed_mount5;
4421         }
4422
4423         block = ext4_count_free_clusters(sb);
4424         ext4_free_blocks_count_set(sbi->s_es,
4425                                    EXT4_C2B(sbi, block));
4426         ext4_superblock_csum_set(sb);
4427         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4428                                   GFP_KERNEL);
4429         if (!err) {
4430                 unsigned long freei = ext4_count_free_inodes(sb);
4431                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4432                 ext4_superblock_csum_set(sb);
4433                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4434                                           GFP_KERNEL);
4435         }
4436         if (!err)
4437                 err = percpu_counter_init(&sbi->s_dirs_counter,
4438                                           ext4_count_dirs(sb), GFP_KERNEL);
4439         if (!err)
4440                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4441                                           GFP_KERNEL);
4442         if (!err)
4443                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4444
4445         if (err) {
4446                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4447                 goto failed_mount6;
4448         }
4449
4450         if (ext4_has_feature_flex_bg(sb))
4451                 if (!ext4_fill_flex_info(sb)) {
4452                         ext4_msg(sb, KERN_ERR,
4453                                "unable to initialize "
4454                                "flex_bg meta info!");
4455                         goto failed_mount6;
4456                 }
4457
4458         err = ext4_register_li_request(sb, first_not_zeroed);
4459         if (err)
4460                 goto failed_mount6;
4461
4462         err = ext4_register_sysfs(sb);
4463         if (err)
4464                 goto failed_mount7;
4465
4466 #ifdef CONFIG_QUOTA
4467         /* Enable quota usage during mount. */
4468         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4469                 err = ext4_enable_quotas(sb);
4470                 if (err)
4471                         goto failed_mount8;
4472         }
4473 #endif  /* CONFIG_QUOTA */
4474
4475         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4476         ext4_orphan_cleanup(sb, es);
4477         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4478         if (needs_recovery) {
4479                 ext4_msg(sb, KERN_INFO, "recovery complete");
4480                 ext4_mark_recovery_complete(sb, es);
4481         }
4482         if (EXT4_SB(sb)->s_journal) {
4483                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4484                         descr = " journalled data mode";
4485                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4486                         descr = " ordered data mode";
4487                 else
4488                         descr = " writeback data mode";
4489         } else
4490                 descr = "out journal";
4491
4492         if (test_opt(sb, DISCARD)) {
4493                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4494                 if (!blk_queue_discard(q))
4495                         ext4_msg(sb, KERN_WARNING,
4496                                  "mounting with \"discard\" option, but "
4497                                  "the device does not support discard");
4498         }
4499
4500         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4501                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4502                          "Opts: %.*s%s%s", descr,
4503                          (int) sizeof(sbi->s_es->s_mount_opts),
4504                          sbi->s_es->s_mount_opts,
4505                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4506
4507         if (es->s_error_count)
4508                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4509
4510         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4511         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4512         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4513         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4514
4515         kfree(orig_data);
4516         return 0;
4517
4518 cantfind_ext4:
4519         if (!silent)
4520                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4521         goto failed_mount;
4522
4523 #ifdef CONFIG_QUOTA
4524 failed_mount8:
4525         ext4_unregister_sysfs(sb);
4526 #endif
4527 failed_mount7:
4528         ext4_unregister_li_request(sb);
4529 failed_mount6:
4530         ext4_mb_release(sb);
4531         if (sbi->s_flex_groups)
4532                 kvfree(sbi->s_flex_groups);
4533         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4534         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4535         percpu_counter_destroy(&sbi->s_dirs_counter);
4536         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4537         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
4538 failed_mount5:
4539         ext4_ext_release(sb);
4540         ext4_release_system_zone(sb);
4541 failed_mount4a:
4542         dput(sb->s_root);
4543         sb->s_root = NULL;
4544 failed_mount4:
4545         ext4_msg(sb, KERN_ERR, "mount failed");
4546         if (EXT4_SB(sb)->rsv_conversion_wq)
4547                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4548 failed_mount_wq:
4549         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4550         sbi->s_ea_inode_cache = NULL;
4551
4552         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4553         sbi->s_ea_block_cache = NULL;
4554
4555         if (sbi->s_journal) {
4556                 jbd2_journal_destroy(sbi->s_journal);
4557                 sbi->s_journal = NULL;
4558         }
4559 failed_mount3a:
4560         ext4_es_unregister_shrinker(sbi);
4561 failed_mount3:
4562         del_timer_sync(&sbi->s_err_report);
4563         if (sbi->s_mmp_tsk)
4564                 kthread_stop(sbi->s_mmp_tsk);
4565 failed_mount2:
4566         for (i = 0; i < db_count; i++)
4567                 brelse(sbi->s_group_desc[i]);
4568         kvfree(sbi->s_group_desc);
4569 failed_mount:
4570         if (sbi->s_chksum_driver)
4571                 crypto_free_shash(sbi->s_chksum_driver);
4572 #ifdef CONFIG_QUOTA
4573         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4574                 kfree(sbi->s_qf_names[i]);
4575 #endif
4576         ext4_blkdev_remove(sbi);
4577         brelse(bh);
4578 out_fail:
4579         sb->s_fs_info = NULL;
4580         kfree(sbi->s_blockgroup_lock);
4581 out_free_base:
4582         kfree(sbi);
4583         kfree(orig_data);
4584         fs_put_dax(dax_dev);
4585         return err ? err : ret;
4586 }
4587
4588 /*
4589  * Setup any per-fs journal parameters now.  We'll do this both on
4590  * initial mount, once the journal has been initialised but before we've
4591  * done any recovery; and again on any subsequent remount.
4592  */
4593 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4594 {
4595         struct ext4_sb_info *sbi = EXT4_SB(sb);
4596
4597         journal->j_commit_interval = sbi->s_commit_interval;
4598         journal->j_min_batch_time = sbi->s_min_batch_time;
4599         journal->j_max_batch_time = sbi->s_max_batch_time;
4600
4601         write_lock(&journal->j_state_lock);
4602         if (test_opt(sb, BARRIER))
4603                 journal->j_flags |= JBD2_BARRIER;
4604         else
4605                 journal->j_flags &= ~JBD2_BARRIER;
4606         if (test_opt(sb, DATA_ERR_ABORT))
4607                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4608         else
4609                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4610         write_unlock(&journal->j_state_lock);
4611 }
4612
4613 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4614                                              unsigned int journal_inum)
4615 {
4616         struct inode *journal_inode;
4617
4618         /*
4619          * Test for the existence of a valid inode on disk.  Bad things
4620          * happen if we iget() an unused inode, as the subsequent iput()
4621          * will try to delete it.
4622          */
4623         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4624         if (IS_ERR(journal_inode)) {
4625                 ext4_msg(sb, KERN_ERR, "no journal found");
4626                 return NULL;
4627         }
4628         if (!journal_inode->i_nlink) {
4629                 make_bad_inode(journal_inode);
4630                 iput(journal_inode);
4631                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4632                 return NULL;
4633         }
4634
4635         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4636                   journal_inode, journal_inode->i_size);
4637         if (!S_ISREG(journal_inode->i_mode)) {
4638                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4639                 iput(journal_inode);
4640                 return NULL;
4641         }
4642         return journal_inode;
4643 }
4644
4645 static journal_t *ext4_get_journal(struct super_block *sb,
4646                                    unsigned int journal_inum)
4647 {
4648         struct inode *journal_inode;
4649         journal_t *journal;
4650
4651         BUG_ON(!ext4_has_feature_journal(sb));
4652
4653         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4654         if (!journal_inode)
4655                 return NULL;
4656
4657         journal = jbd2_journal_init_inode(journal_inode);
4658         if (!journal) {
4659                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4660                 iput(journal_inode);
4661                 return NULL;
4662         }
4663         journal->j_private = sb;
4664         ext4_init_journal_params(sb, journal);
4665         return journal;
4666 }
4667
4668 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4669                                        dev_t j_dev)
4670 {
4671         struct buffer_head *bh;
4672         journal_t *journal;
4673         ext4_fsblk_t start;
4674         ext4_fsblk_t len;
4675         int hblock, blocksize;
4676         ext4_fsblk_t sb_block;
4677         unsigned long offset;
4678         struct ext4_super_block *es;
4679         struct block_device *bdev;
4680
4681         BUG_ON(!ext4_has_feature_journal(sb));
4682
4683         bdev = ext4_blkdev_get(j_dev, sb);
4684         if (bdev == NULL)
4685                 return NULL;
4686
4687         blocksize = sb->s_blocksize;
4688         hblock = bdev_logical_block_size(bdev);
4689         if (blocksize < hblock) {
4690                 ext4_msg(sb, KERN_ERR,
4691                         "blocksize too small for journal device");
4692                 goto out_bdev;
4693         }
4694
4695         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4696         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4697         set_blocksize(bdev, blocksize);
4698         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4699                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4700                        "external journal");
4701                 goto out_bdev;
4702         }
4703
4704         es = (struct ext4_super_block *) (bh->b_data + offset);
4705         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4706             !(le32_to_cpu(es->s_feature_incompat) &
4707               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4708                 ext4_msg(sb, KERN_ERR, "external journal has "
4709                                         "bad superblock");
4710                 brelse(bh);
4711                 goto out_bdev;
4712         }
4713
4714         if ((le32_to_cpu(es->s_feature_ro_compat) &
4715              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4716             es->s_checksum != ext4_superblock_csum(sb, es)) {
4717                 ext4_msg(sb, KERN_ERR, "external journal has "
4718                                        "corrupt superblock");
4719                 brelse(bh);
4720                 goto out_bdev;
4721         }
4722
4723         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4724                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4725                 brelse(bh);
4726                 goto out_bdev;
4727         }
4728
4729         len = ext4_blocks_count(es);
4730         start = sb_block + 1;
4731         brelse(bh);     /* we're done with the superblock */
4732
4733         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4734                                         start, len, blocksize);
4735         if (!journal) {
4736                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4737                 goto out_bdev;
4738         }
4739         journal->j_private = sb;
4740         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4741         wait_on_buffer(journal->j_sb_buffer);
4742         if (!buffer_uptodate(journal->j_sb_buffer)) {
4743                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4744                 goto out_journal;
4745         }
4746         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4747                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4748                                         "user (unsupported) - %d",
4749                         be32_to_cpu(journal->j_superblock->s_nr_users));
4750                 goto out_journal;
4751         }
4752         EXT4_SB(sb)->journal_bdev = bdev;
4753         ext4_init_journal_params(sb, journal);
4754         return journal;
4755
4756 out_journal:
4757         jbd2_journal_destroy(journal);
4758 out_bdev:
4759         ext4_blkdev_put(bdev);
4760         return NULL;
4761 }
4762
4763 static int ext4_load_journal(struct super_block *sb,
4764                              struct ext4_super_block *es,
4765                              unsigned long journal_devnum)
4766 {
4767         journal_t *journal;
4768         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4769         dev_t journal_dev;
4770         int err = 0;
4771         int really_read_only;
4772
4773         BUG_ON(!ext4_has_feature_journal(sb));
4774
4775         if (journal_devnum &&
4776             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4777                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4778                         "numbers have changed");
4779                 journal_dev = new_decode_dev(journal_devnum);
4780         } else
4781                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4782
4783         really_read_only = bdev_read_only(sb->s_bdev);
4784
4785         /*
4786          * Are we loading a blank journal or performing recovery after a
4787          * crash?  For recovery, we need to check in advance whether we
4788          * can get read-write access to the device.
4789          */
4790         if (ext4_has_feature_journal_needs_recovery(sb)) {
4791                 if (sb_rdonly(sb)) {
4792                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4793                                         "required on readonly filesystem");
4794                         if (really_read_only) {
4795                                 ext4_msg(sb, KERN_ERR, "write access "
4796                                         "unavailable, cannot proceed "
4797                                         "(try mounting with noload)");
4798                                 return -EROFS;
4799                         }
4800                         ext4_msg(sb, KERN_INFO, "write access will "
4801                                "be enabled during recovery");
4802                 }
4803         }
4804
4805         if (journal_inum && journal_dev) {
4806                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4807                        "and inode journals!");
4808                 return -EINVAL;
4809         }
4810
4811         if (journal_inum) {
4812                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4813                         return -EINVAL;
4814         } else {
4815                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4816                         return -EINVAL;
4817         }
4818
4819         if (!(journal->j_flags & JBD2_BARRIER))
4820                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4821
4822         if (!ext4_has_feature_journal_needs_recovery(sb))
4823                 err = jbd2_journal_wipe(journal, !really_read_only);
4824         if (!err) {
4825                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4826                 if (save)
4827                         memcpy(save, ((char *) es) +
4828                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4829                 err = jbd2_journal_load(journal);
4830                 if (save)
4831                         memcpy(((char *) es) + EXT4_S_ERR_START,
4832                                save, EXT4_S_ERR_LEN);
4833                 kfree(save);
4834         }
4835
4836         if (err) {
4837                 ext4_msg(sb, KERN_ERR, "error loading journal");
4838                 jbd2_journal_destroy(journal);
4839                 return err;
4840         }
4841
4842         EXT4_SB(sb)->s_journal = journal;
4843         ext4_clear_journal_err(sb, es);
4844
4845         if (!really_read_only && journal_devnum &&
4846             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4847                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4848
4849                 /* Make sure we flush the recovery flag to disk. */
4850                 ext4_commit_super(sb, 1);
4851         }
4852
4853         return 0;
4854 }
4855
4856 static int ext4_commit_super(struct super_block *sb, int sync)
4857 {
4858         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4859         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4860         int error = 0;
4861
4862         if (!sbh || block_device_ejected(sb))
4863                 return error;
4864
4865         /*
4866          * The superblock bh should be mapped, but it might not be if the
4867          * device was hot-removed. Not much we can do but fail the I/O.
4868          */
4869         if (!buffer_mapped(sbh))
4870                 return error;
4871
4872         /*
4873          * If the file system is mounted read-only, don't update the
4874          * superblock write time.  This avoids updating the superblock
4875          * write time when we are mounting the root file system
4876          * read/only but we need to replay the journal; at that point,
4877          * for people who are east of GMT and who make their clock
4878          * tick in localtime for Windows bug-for-bug compatibility,
4879          * the clock is set in the future, and this will cause e2fsck
4880          * to complain and force a full file system check.
4881          */
4882         if (!(sb->s_flags & SB_RDONLY))
4883                 ext4_update_tstamp(es, s_wtime);
4884         if (sb->s_bdev->bd_part)
4885                 es->s_kbytes_written =
4886                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4887                             ((part_stat_read(sb->s_bdev->bd_part,
4888                                              sectors[STAT_WRITE]) -
4889                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4890         else
4891                 es->s_kbytes_written =
4892                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4893         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4894                 ext4_free_blocks_count_set(es,
4895                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4896                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4897         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4898                 es->s_free_inodes_count =
4899                         cpu_to_le32(percpu_counter_sum_positive(
4900                                 &EXT4_SB(sb)->s_freeinodes_counter));
4901         BUFFER_TRACE(sbh, "marking dirty");
4902         ext4_superblock_csum_set(sb);
4903         if (sync)
4904                 lock_buffer(sbh);
4905         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
4906                 /*
4907                  * Oh, dear.  A previous attempt to write the
4908                  * superblock failed.  This could happen because the
4909                  * USB device was yanked out.  Or it could happen to
4910                  * be a transient write error and maybe the block will
4911                  * be remapped.  Nothing we can do but to retry the
4912                  * write and hope for the best.
4913                  */
4914                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4915                        "superblock detected");
4916                 clear_buffer_write_io_error(sbh);
4917                 set_buffer_uptodate(sbh);
4918         }
4919         mark_buffer_dirty(sbh);
4920         if (sync) {
4921                 unlock_buffer(sbh);
4922                 error = __sync_dirty_buffer(sbh,
4923                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4924                 if (buffer_write_io_error(sbh)) {
4925                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4926                                "superblock");
4927                         clear_buffer_write_io_error(sbh);
4928                         set_buffer_uptodate(sbh);
4929                 }
4930         }
4931         return error;
4932 }
4933
4934 /*
4935  * Have we just finished recovery?  If so, and if we are mounting (or
4936  * remounting) the filesystem readonly, then we will end up with a
4937  * consistent fs on disk.  Record that fact.
4938  */
4939 static void ext4_mark_recovery_complete(struct super_block *sb,
4940                                         struct ext4_super_block *es)
4941 {
4942         journal_t *journal = EXT4_SB(sb)->s_journal;
4943
4944         if (!ext4_has_feature_journal(sb)) {
4945                 BUG_ON(journal != NULL);
4946                 return;
4947         }
4948         jbd2_journal_lock_updates(journal);
4949         if (jbd2_journal_flush(journal) < 0)
4950                 goto out;
4951
4952         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4953                 ext4_clear_feature_journal_needs_recovery(sb);
4954                 ext4_commit_super(sb, 1);
4955         }
4956
4957 out:
4958         jbd2_journal_unlock_updates(journal);
4959 }
4960
4961 /*
4962  * If we are mounting (or read-write remounting) a filesystem whose journal
4963  * has recorded an error from a previous lifetime, move that error to the
4964  * main filesystem now.
4965  */
4966 static void ext4_clear_journal_err(struct super_block *sb,
4967                                    struct ext4_super_block *es)
4968 {
4969         journal_t *journal;
4970         int j_errno;
4971         const char *errstr;
4972
4973         BUG_ON(!ext4_has_feature_journal(sb));
4974
4975         journal = EXT4_SB(sb)->s_journal;
4976
4977         /*
4978          * Now check for any error status which may have been recorded in the
4979          * journal by a prior ext4_error() or ext4_abort()
4980          */
4981
4982         j_errno = jbd2_journal_errno(journal);
4983         if (j_errno) {
4984                 char nbuf[16];
4985
4986                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4987                 ext4_warning(sb, "Filesystem error recorded "
4988                              "from previous mount: %s", errstr);
4989                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4990
4991                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4992                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4993                 ext4_commit_super(sb, 1);
4994
4995                 jbd2_journal_clear_err(journal);
4996                 jbd2_journal_update_sb_errno(journal);
4997         }
4998 }
4999
5000 /*
5001  * Force the running and committing transactions to commit,
5002  * and wait on the commit.
5003  */
5004 int ext4_force_commit(struct super_block *sb)
5005 {
5006         journal_t *journal;
5007
5008         if (sb_rdonly(sb))
5009                 return 0;
5010
5011         journal = EXT4_SB(sb)->s_journal;
5012         return ext4_journal_force_commit(journal);
5013 }
5014
5015 static int ext4_sync_fs(struct super_block *sb, int wait)
5016 {
5017         int ret = 0;
5018         tid_t target;
5019         bool needs_barrier = false;
5020         struct ext4_sb_info *sbi = EXT4_SB(sb);
5021
5022         if (unlikely(ext4_forced_shutdown(sbi)))
5023                 return 0;
5024
5025         trace_ext4_sync_fs(sb, wait);
5026         flush_workqueue(sbi->rsv_conversion_wq);
5027         /*
5028          * Writeback quota in non-journalled quota case - journalled quota has
5029          * no dirty dquots
5030          */
5031         dquot_writeback_dquots(sb, -1);
5032         /*
5033          * Data writeback is possible w/o journal transaction, so barrier must
5034          * being sent at the end of the function. But we can skip it if
5035          * transaction_commit will do it for us.
5036          */
5037         if (sbi->s_journal) {
5038                 target = jbd2_get_latest_transaction(sbi->s_journal);
5039                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5040                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5041                         needs_barrier = true;
5042
5043                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5044                         if (wait)
5045                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5046                                                            target);
5047                 }
5048         } else if (wait && test_opt(sb, BARRIER))
5049                 needs_barrier = true;
5050         if (needs_barrier) {
5051                 int err;
5052                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5053                 if (!ret)
5054                         ret = err;
5055         }
5056
5057         return ret;
5058 }
5059
5060 /*
5061  * LVM calls this function before a (read-only) snapshot is created.  This
5062  * gives us a chance to flush the journal completely and mark the fs clean.
5063  *
5064  * Note that only this function cannot bring a filesystem to be in a clean
5065  * state independently. It relies on upper layer to stop all data & metadata
5066  * modifications.
5067  */
5068 static int ext4_freeze(struct super_block *sb)
5069 {
5070         int error = 0;
5071         journal_t *journal;
5072
5073         if (sb_rdonly(sb))
5074                 return 0;
5075
5076         journal = EXT4_SB(sb)->s_journal;
5077
5078         if (journal) {
5079                 /* Now we set up the journal barrier. */
5080                 jbd2_journal_lock_updates(journal);
5081
5082                 /*
5083                  * Don't clear the needs_recovery flag if we failed to
5084                  * flush the journal.
5085                  */
5086                 error = jbd2_journal_flush(journal);
5087                 if (error < 0)
5088                         goto out;
5089
5090                 /* Journal blocked and flushed, clear needs_recovery flag. */
5091                 ext4_clear_feature_journal_needs_recovery(sb);
5092         }
5093
5094         error = ext4_commit_super(sb, 1);
5095 out:
5096         if (journal)
5097                 /* we rely on upper layer to stop further updates */
5098                 jbd2_journal_unlock_updates(journal);
5099         return error;
5100 }
5101
5102 /*
5103  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5104  * flag here, even though the filesystem is not technically dirty yet.
5105  */
5106 static int ext4_unfreeze(struct super_block *sb)
5107 {
5108         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5109                 return 0;
5110
5111         if (EXT4_SB(sb)->s_journal) {
5112                 /* Reset the needs_recovery flag before the fs is unlocked. */
5113                 ext4_set_feature_journal_needs_recovery(sb);
5114         }
5115
5116         ext4_commit_super(sb, 1);
5117         return 0;
5118 }
5119
5120 /*
5121  * Structure to save mount options for ext4_remount's benefit
5122  */
5123 struct ext4_mount_options {
5124         unsigned long s_mount_opt;
5125         unsigned long s_mount_opt2;
5126         kuid_t s_resuid;
5127         kgid_t s_resgid;
5128         unsigned long s_commit_interval;
5129         u32 s_min_batch_time, s_max_batch_time;
5130 #ifdef CONFIG_QUOTA
5131         int s_jquota_fmt;
5132         char *s_qf_names[EXT4_MAXQUOTAS];
5133 #endif
5134 };
5135
5136 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5137 {
5138         struct ext4_super_block *es;
5139         struct ext4_sb_info *sbi = EXT4_SB(sb);
5140         unsigned long old_sb_flags;
5141         struct ext4_mount_options old_opts;
5142         int enable_quota = 0;
5143         ext4_group_t g;
5144         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5145         int err = 0;
5146 #ifdef CONFIG_QUOTA
5147         int i, j;
5148         char *to_free[EXT4_MAXQUOTAS];
5149 #endif
5150         char *orig_data = kstrdup(data, GFP_KERNEL);
5151
5152         if (data && !orig_data)
5153                 return -ENOMEM;
5154
5155         /* Store the original options */
5156         old_sb_flags = sb->s_flags;
5157         old_opts.s_mount_opt = sbi->s_mount_opt;
5158         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5159         old_opts.s_resuid = sbi->s_resuid;
5160         old_opts.s_resgid = sbi->s_resgid;
5161         old_opts.s_commit_interval = sbi->s_commit_interval;
5162         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5163         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5164 #ifdef CONFIG_QUOTA
5165         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5166         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5167                 if (sbi->s_qf_names[i]) {
5168                         char *qf_name = get_qf_name(sb, sbi, i);
5169
5170                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5171                         if (!old_opts.s_qf_names[i]) {
5172                                 for (j = 0; j < i; j++)
5173                                         kfree(old_opts.s_qf_names[j]);
5174                                 kfree(orig_data);
5175                                 return -ENOMEM;
5176                         }
5177                 } else
5178                         old_opts.s_qf_names[i] = NULL;
5179 #endif
5180         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5181                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5182
5183         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5184                 err = -EINVAL;
5185                 goto restore_opts;
5186         }
5187
5188         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5189             test_opt(sb, JOURNAL_CHECKSUM)) {
5190                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5191                          "during remount not supported; ignoring");
5192                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5193         }
5194
5195         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5196                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5197                         ext4_msg(sb, KERN_ERR, "can't mount with "
5198                                  "both data=journal and delalloc");
5199                         err = -EINVAL;
5200                         goto restore_opts;
5201                 }
5202                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5203                         ext4_msg(sb, KERN_ERR, "can't mount with "
5204                                  "both data=journal and dioread_nolock");
5205                         err = -EINVAL;
5206                         goto restore_opts;
5207                 }
5208                 if (test_opt(sb, DAX)) {
5209                         ext4_msg(sb, KERN_ERR, "can't mount with "
5210                                  "both data=journal and dax");
5211                         err = -EINVAL;
5212                         goto restore_opts;
5213                 }
5214         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5215                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5216                         ext4_msg(sb, KERN_ERR, "can't mount with "
5217                                 "journal_async_commit in data=ordered mode");
5218                         err = -EINVAL;
5219                         goto restore_opts;
5220                 }
5221         }
5222
5223         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5224                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5225                 err = -EINVAL;
5226                 goto restore_opts;
5227         }
5228
5229         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5230                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5231                         "dax flag with busy inodes while remounting");
5232                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5233         }
5234
5235         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5236                 ext4_abort(sb, "Abort forced by user");
5237
5238         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5239                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5240
5241         es = sbi->s_es;
5242
5243         if (sbi->s_journal) {
5244                 ext4_init_journal_params(sb, sbi->s_journal);
5245                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5246         }
5247
5248         if (*flags & SB_LAZYTIME)
5249                 sb->s_flags |= SB_LAZYTIME;
5250
5251         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5252                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5253                         err = -EROFS;
5254                         goto restore_opts;
5255                 }
5256
5257                 if (*flags & SB_RDONLY) {
5258                         err = sync_filesystem(sb);
5259                         if (err < 0)
5260                                 goto restore_opts;
5261                         err = dquot_suspend(sb, -1);
5262                         if (err < 0)
5263                                 goto restore_opts;
5264
5265                         /*
5266                          * First of all, the unconditional stuff we have to do
5267                          * to disable replay of the journal when we next remount
5268                          */
5269                         sb->s_flags |= SB_RDONLY;
5270
5271                         /*
5272                          * OK, test if we are remounting a valid rw partition
5273                          * readonly, and if so set the rdonly flag and then
5274                          * mark the partition as valid again.
5275                          */
5276                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5277                             (sbi->s_mount_state & EXT4_VALID_FS))
5278                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5279
5280                         if (sbi->s_journal)
5281                                 ext4_mark_recovery_complete(sb, es);
5282                         if (sbi->s_mmp_tsk)
5283                                 kthread_stop(sbi->s_mmp_tsk);
5284                 } else {
5285                         /* Make sure we can mount this feature set readwrite */
5286                         if (ext4_has_feature_readonly(sb) ||
5287                             !ext4_feature_set_ok(sb, 0)) {
5288                                 err = -EROFS;
5289                                 goto restore_opts;
5290                         }
5291                         /*
5292                          * Make sure the group descriptor checksums
5293                          * are sane.  If they aren't, refuse to remount r/w.
5294                          */
5295                         for (g = 0; g < sbi->s_groups_count; g++) {
5296                                 struct ext4_group_desc *gdp =
5297                                         ext4_get_group_desc(sb, g, NULL);
5298
5299                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5300                                         ext4_msg(sb, KERN_ERR,
5301                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5302                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5303                                                le16_to_cpu(gdp->bg_checksum));
5304                                         err = -EFSBADCRC;
5305                                         goto restore_opts;
5306                                 }
5307                         }
5308
5309                         /*
5310                          * If we have an unprocessed orphan list hanging
5311                          * around from a previously readonly bdev mount,
5312                          * require a full umount/remount for now.
5313                          */
5314                         if (es->s_last_orphan) {
5315                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5316                                        "remount RDWR because of unprocessed "
5317                                        "orphan inode list.  Please "
5318                                        "umount/remount instead");
5319                                 err = -EINVAL;
5320                                 goto restore_opts;
5321                         }
5322
5323                         /*
5324                          * Mounting a RDONLY partition read-write, so reread
5325                          * and store the current valid flag.  (It may have
5326                          * been changed by e2fsck since we originally mounted
5327                          * the partition.)
5328                          */
5329                         if (sbi->s_journal)
5330                                 ext4_clear_journal_err(sb, es);
5331                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5332
5333                         err = ext4_setup_super(sb, es, 0);
5334                         if (err)
5335                                 goto restore_opts;
5336
5337                         sb->s_flags &= ~SB_RDONLY;
5338                         if (ext4_has_feature_mmp(sb))
5339                                 if (ext4_multi_mount_protect(sb,
5340                                                 le64_to_cpu(es->s_mmp_block))) {
5341                                         err = -EROFS;
5342                                         goto restore_opts;
5343                                 }
5344                         enable_quota = 1;
5345                 }
5346         }
5347
5348         /*
5349          * Reinitialize lazy itable initialization thread based on
5350          * current settings
5351          */
5352         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5353                 ext4_unregister_li_request(sb);
5354         else {
5355                 ext4_group_t first_not_zeroed;
5356                 first_not_zeroed = ext4_has_uninit_itable(sb);
5357                 ext4_register_li_request(sb, first_not_zeroed);
5358         }
5359
5360         ext4_setup_system_zone(sb);
5361         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5362                 err = ext4_commit_super(sb, 1);
5363                 if (err)
5364                         goto restore_opts;
5365         }
5366
5367 #ifdef CONFIG_QUOTA
5368         /* Release old quota file names */
5369         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5370                 kfree(old_opts.s_qf_names[i]);
5371         if (enable_quota) {
5372                 if (sb_any_quota_suspended(sb))
5373                         dquot_resume(sb, -1);
5374                 else if (ext4_has_feature_quota(sb)) {
5375                         err = ext4_enable_quotas(sb);
5376                         if (err)
5377                                 goto restore_opts;
5378                 }
5379         }
5380 #endif
5381
5382         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5383         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5384         kfree(orig_data);
5385         return 0;
5386
5387 restore_opts:
5388         sb->s_flags = old_sb_flags;
5389         sbi->s_mount_opt = old_opts.s_mount_opt;
5390         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5391         sbi->s_resuid = old_opts.s_resuid;
5392         sbi->s_resgid = old_opts.s_resgid;
5393         sbi->s_commit_interval = old_opts.s_commit_interval;
5394         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5395         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5396 #ifdef CONFIG_QUOTA
5397         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5398         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5399                 to_free[i] = get_qf_name(sb, sbi, i);
5400                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5401         }
5402         synchronize_rcu();
5403         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5404                 kfree(to_free[i]);
5405 #endif
5406         kfree(orig_data);
5407         return err;
5408 }
5409
5410 #ifdef CONFIG_QUOTA
5411 static int ext4_statfs_project(struct super_block *sb,
5412                                kprojid_t projid, struct kstatfs *buf)
5413 {
5414         struct kqid qid;
5415         struct dquot *dquot;
5416         u64 limit;
5417         u64 curblock;
5418
5419         qid = make_kqid_projid(projid);
5420         dquot = dqget(sb, qid);
5421         if (IS_ERR(dquot))
5422                 return PTR_ERR(dquot);
5423         spin_lock(&dquot->dq_dqb_lock);
5424
5425         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5426                  dquot->dq_dqb.dqb_bsoftlimit :
5427                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5428         if (limit && buf->f_blocks > limit) {
5429                 curblock = (dquot->dq_dqb.dqb_curspace +
5430                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5431                 buf->f_blocks = limit;
5432                 buf->f_bfree = buf->f_bavail =
5433                         (buf->f_blocks > curblock) ?
5434                          (buf->f_blocks - curblock) : 0;
5435         }
5436
5437         limit = dquot->dq_dqb.dqb_isoftlimit ?
5438                 dquot->dq_dqb.dqb_isoftlimit :
5439                 dquot->dq_dqb.dqb_ihardlimit;
5440         if (limit && buf->f_files > limit) {
5441                 buf->f_files = limit;
5442                 buf->f_ffree =
5443                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5444                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5445         }
5446
5447         spin_unlock(&dquot->dq_dqb_lock);
5448         dqput(dquot);
5449         return 0;
5450 }
5451 #endif
5452
5453 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5454 {
5455         struct super_block *sb = dentry->d_sb;
5456         struct ext4_sb_info *sbi = EXT4_SB(sb);
5457         struct ext4_super_block *es = sbi->s_es;
5458         ext4_fsblk_t overhead = 0, resv_blocks;
5459         u64 fsid;
5460         s64 bfree;
5461         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5462
5463         if (!test_opt(sb, MINIX_DF))
5464                 overhead = sbi->s_overhead;
5465
5466         buf->f_type = EXT4_SUPER_MAGIC;
5467         buf->f_bsize = sb->s_blocksize;
5468         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5469         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5470                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5471         /* prevent underflow in case that few free space is available */
5472         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5473         buf->f_bavail = buf->f_bfree -
5474                         (ext4_r_blocks_count(es) + resv_blocks);
5475         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5476                 buf->f_bavail = 0;
5477         buf->f_files = le32_to_cpu(es->s_inodes_count);
5478         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5479         buf->f_namelen = EXT4_NAME_LEN;
5480         fsid = le64_to_cpup((void *)es->s_uuid) ^
5481                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5482         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5483         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5484
5485 #ifdef CONFIG_QUOTA
5486         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5487             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5488                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5489 #endif
5490         return 0;
5491 }
5492
5493
5494 #ifdef CONFIG_QUOTA
5495
5496 /*
5497  * Helper functions so that transaction is started before we acquire dqio_sem
5498  * to keep correct lock ordering of transaction > dqio_sem
5499  */
5500 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5501 {
5502         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5503 }
5504
5505 static int ext4_write_dquot(struct dquot *dquot)
5506 {
5507         int ret, err;
5508         handle_t *handle;
5509         struct inode *inode;
5510
5511         inode = dquot_to_inode(dquot);
5512         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5513                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5514         if (IS_ERR(handle))
5515                 return PTR_ERR(handle);
5516         ret = dquot_commit(dquot);
5517         err = ext4_journal_stop(handle);
5518         if (!ret)
5519                 ret = err;
5520         return ret;
5521 }
5522
5523 static int ext4_acquire_dquot(struct dquot *dquot)
5524 {
5525         int ret, err;
5526         handle_t *handle;
5527
5528         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5529                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5530         if (IS_ERR(handle))
5531                 return PTR_ERR(handle);
5532         ret = dquot_acquire(dquot);
5533         err = ext4_journal_stop(handle);
5534         if (!ret)
5535                 ret = err;
5536         return ret;
5537 }
5538
5539 static int ext4_release_dquot(struct dquot *dquot)
5540 {
5541         int ret, err;
5542         handle_t *handle;
5543
5544         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5545                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5546         if (IS_ERR(handle)) {
5547                 /* Release dquot anyway to avoid endless cycle in dqput() */
5548                 dquot_release(dquot);
5549                 return PTR_ERR(handle);
5550         }
5551         ret = dquot_release(dquot);
5552         err = ext4_journal_stop(handle);
5553         if (!ret)
5554                 ret = err;
5555         return ret;
5556 }
5557
5558 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5559 {
5560         struct super_block *sb = dquot->dq_sb;
5561         struct ext4_sb_info *sbi = EXT4_SB(sb);
5562
5563         /* Are we journaling quotas? */
5564         if (ext4_has_feature_quota(sb) ||
5565             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5566                 dquot_mark_dquot_dirty(dquot);
5567                 return ext4_write_dquot(dquot);
5568         } else {
5569                 return dquot_mark_dquot_dirty(dquot);
5570         }
5571 }
5572
5573 static int ext4_write_info(struct super_block *sb, int type)
5574 {
5575         int ret, err;
5576         handle_t *handle;
5577
5578         /* Data block + inode block */
5579         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5580         if (IS_ERR(handle))
5581                 return PTR_ERR(handle);
5582         ret = dquot_commit_info(sb, type);
5583         err = ext4_journal_stop(handle);
5584         if (!ret)
5585                 ret = err;
5586         return ret;
5587 }
5588
5589 /*
5590  * Turn on quotas during mount time - we need to find
5591  * the quota file and such...
5592  */
5593 static int ext4_quota_on_mount(struct super_block *sb, int type)
5594 {
5595         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5596                                         EXT4_SB(sb)->s_jquota_fmt, type);
5597 }
5598
5599 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5600 {
5601         struct ext4_inode_info *ei = EXT4_I(inode);
5602
5603         /* The first argument of lockdep_set_subclass has to be
5604          * *exactly* the same as the argument to init_rwsem() --- in
5605          * this case, in init_once() --- or lockdep gets unhappy
5606          * because the name of the lock is set using the
5607          * stringification of the argument to init_rwsem().
5608          */
5609         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5610         lockdep_set_subclass(&ei->i_data_sem, subclass);
5611 }
5612
5613 /*
5614  * Standard function to be called on quota_on
5615  */
5616 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5617                          const struct path *path)
5618 {
5619         int err;
5620
5621         if (!test_opt(sb, QUOTA))
5622                 return -EINVAL;
5623
5624         /* Quotafile not on the same filesystem? */
5625         if (path->dentry->d_sb != sb)
5626                 return -EXDEV;
5627         /* Journaling quota? */
5628         if (EXT4_SB(sb)->s_qf_names[type]) {
5629                 /* Quotafile not in fs root? */
5630                 if (path->dentry->d_parent != sb->s_root)
5631                         ext4_msg(sb, KERN_WARNING,
5632                                 "Quota file not on filesystem root. "
5633                                 "Journaled quota will not work");
5634                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5635         } else {
5636                 /*
5637                  * Clear the flag just in case mount options changed since
5638                  * last time.
5639                  */
5640                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5641         }
5642
5643         /*
5644          * When we journal data on quota file, we have to flush journal to see
5645          * all updates to the file when we bypass pagecache...
5646          */
5647         if (EXT4_SB(sb)->s_journal &&
5648             ext4_should_journal_data(d_inode(path->dentry))) {
5649                 /*
5650                  * We don't need to lock updates but journal_flush() could
5651                  * otherwise be livelocked...
5652                  */
5653                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5654                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5655                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5656                 if (err)
5657                         return err;
5658         }
5659
5660         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5661         err = dquot_quota_on(sb, type, format_id, path);
5662         if (err) {
5663                 lockdep_set_quota_inode(path->dentry->d_inode,
5664                                              I_DATA_SEM_NORMAL);
5665         } else {
5666                 struct inode *inode = d_inode(path->dentry);
5667                 handle_t *handle;
5668
5669                 /*
5670                  * Set inode flags to prevent userspace from messing with quota
5671                  * files. If this fails, we return success anyway since quotas
5672                  * are already enabled and this is not a hard failure.
5673                  */
5674                 inode_lock(inode);
5675                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5676                 if (IS_ERR(handle))
5677                         goto unlock_inode;
5678                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5679                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5680                                 S_NOATIME | S_IMMUTABLE);
5681                 ext4_mark_inode_dirty(handle, inode);
5682                 ext4_journal_stop(handle);
5683         unlock_inode:
5684                 inode_unlock(inode);
5685         }
5686         return err;
5687 }
5688
5689 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5690                              unsigned int flags)
5691 {
5692         int err;
5693         struct inode *qf_inode;
5694         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5695                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5696                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5697                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5698         };
5699
5700         BUG_ON(!ext4_has_feature_quota(sb));
5701
5702         if (!qf_inums[type])
5703                 return -EPERM;
5704
5705         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
5706         if (IS_ERR(qf_inode)) {
5707                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5708                 return PTR_ERR(qf_inode);
5709         }
5710
5711         /* Don't account quota for quota files to avoid recursion */
5712         qf_inode->i_flags |= S_NOQUOTA;
5713         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5714         err = dquot_enable(qf_inode, type, format_id, flags);
5715         if (err)
5716                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5717         iput(qf_inode);
5718
5719         return err;
5720 }
5721
5722 /* Enable usage tracking for all quota types. */
5723 static int ext4_enable_quotas(struct super_block *sb)
5724 {
5725         int type, err = 0;
5726         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5727                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5728                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5729                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5730         };
5731         bool quota_mopt[EXT4_MAXQUOTAS] = {
5732                 test_opt(sb, USRQUOTA),
5733                 test_opt(sb, GRPQUOTA),
5734                 test_opt(sb, PRJQUOTA),
5735         };
5736
5737         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5738         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5739                 if (qf_inums[type]) {
5740                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5741                                 DQUOT_USAGE_ENABLED |
5742                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5743                         if (err) {
5744                                 ext4_warning(sb,
5745                                         "Failed to enable quota tracking "
5746                                         "(type=%d, err=%d). Please run "
5747                                         "e2fsck to fix.", type, err);
5748                                 for (type--; type >= 0; type--)
5749                                         dquot_quota_off(sb, type);
5750
5751                                 return err;
5752                         }
5753                 }
5754         }
5755         return 0;
5756 }
5757
5758 static int ext4_quota_off(struct super_block *sb, int type)
5759 {
5760         struct inode *inode = sb_dqopt(sb)->files[type];
5761         handle_t *handle;
5762         int err;
5763
5764         /* Force all delayed allocation blocks to be allocated.
5765          * Caller already holds s_umount sem */
5766         if (test_opt(sb, DELALLOC))
5767                 sync_filesystem(sb);
5768
5769         if (!inode || !igrab(inode))
5770                 goto out;
5771
5772         err = dquot_quota_off(sb, type);
5773         if (err || ext4_has_feature_quota(sb))
5774                 goto out_put;
5775
5776         inode_lock(inode);
5777         /*
5778          * Update modification times of quota files when userspace can
5779          * start looking at them. If we fail, we return success anyway since
5780          * this is not a hard failure and quotas are already disabled.
5781          */
5782         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5783         if (IS_ERR(handle))
5784                 goto out_unlock;
5785         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5786         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5787         inode->i_mtime = inode->i_ctime = current_time(inode);
5788         ext4_mark_inode_dirty(handle, inode);
5789         ext4_journal_stop(handle);
5790 out_unlock:
5791         inode_unlock(inode);
5792 out_put:
5793         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5794         iput(inode);
5795         return err;
5796 out:
5797         return dquot_quota_off(sb, type);
5798 }
5799
5800 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5801  * acquiring the locks... As quota files are never truncated and quota code
5802  * itself serializes the operations (and no one else should touch the files)
5803  * we don't have to be afraid of races */
5804 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5805                                size_t len, loff_t off)
5806 {
5807         struct inode *inode = sb_dqopt(sb)->files[type];
5808         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5809         int offset = off & (sb->s_blocksize - 1);
5810         int tocopy;
5811         size_t toread;
5812         struct buffer_head *bh;
5813         loff_t i_size = i_size_read(inode);
5814
5815         if (off > i_size)
5816                 return 0;
5817         if (off+len > i_size)
5818                 len = i_size-off;
5819         toread = len;
5820         while (toread > 0) {
5821                 tocopy = sb->s_blocksize - offset < toread ?
5822                                 sb->s_blocksize - offset : toread;
5823                 bh = ext4_bread(NULL, inode, blk, 0);
5824                 if (IS_ERR(bh))
5825                         return PTR_ERR(bh);
5826                 if (!bh)        /* A hole? */
5827                         memset(data, 0, tocopy);
5828                 else
5829                         memcpy(data, bh->b_data+offset, tocopy);
5830                 brelse(bh);
5831                 offset = 0;
5832                 toread -= tocopy;
5833                 data += tocopy;
5834                 blk++;
5835         }
5836         return len;
5837 }
5838
5839 /* Write to quotafile (we know the transaction is already started and has
5840  * enough credits) */
5841 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5842                                 const char *data, size_t len, loff_t off)
5843 {
5844         struct inode *inode = sb_dqopt(sb)->files[type];
5845         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5846         int err, offset = off & (sb->s_blocksize - 1);
5847         int retries = 0;
5848         struct buffer_head *bh;
5849         handle_t *handle = journal_current_handle();
5850
5851         if (EXT4_SB(sb)->s_journal && !handle) {
5852                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5853                         " cancelled because transaction is not started",
5854                         (unsigned long long)off, (unsigned long long)len);
5855                 return -EIO;
5856         }
5857         /*
5858          * Since we account only one data block in transaction credits,
5859          * then it is impossible to cross a block boundary.
5860          */
5861         if (sb->s_blocksize - offset < len) {
5862                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5863                         " cancelled because not block aligned",
5864                         (unsigned long long)off, (unsigned long long)len);
5865                 return -EIO;
5866         }
5867
5868         do {
5869                 bh = ext4_bread(handle, inode, blk,
5870                                 EXT4_GET_BLOCKS_CREATE |
5871                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5872         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5873                  ext4_should_retry_alloc(inode->i_sb, &retries));
5874         if (IS_ERR(bh))
5875                 return PTR_ERR(bh);
5876         if (!bh)
5877                 goto out;
5878         BUFFER_TRACE(bh, "get write access");
5879         err = ext4_journal_get_write_access(handle, bh);
5880         if (err) {
5881                 brelse(bh);
5882                 return err;
5883         }
5884         lock_buffer(bh);
5885         memcpy(bh->b_data+offset, data, len);
5886         flush_dcache_page(bh->b_page);
5887         unlock_buffer(bh);
5888         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5889         brelse(bh);
5890 out:
5891         if (inode->i_size < off + len) {
5892                 i_size_write(inode, off + len);
5893                 EXT4_I(inode)->i_disksize = inode->i_size;
5894                 ext4_mark_inode_dirty(handle, inode);
5895         }
5896         return len;
5897 }
5898
5899 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5900 {
5901         const struct quota_format_ops   *ops;
5902
5903         if (!sb_has_quota_loaded(sb, qid->type))
5904                 return -ESRCH;
5905         ops = sb_dqopt(sb)->ops[qid->type];
5906         if (!ops || !ops->get_next_id)
5907                 return -ENOSYS;
5908         return dquot_get_next_id(sb, qid);
5909 }
5910 #endif
5911
5912 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5913                        const char *dev_name, void *data)
5914 {
5915         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5916 }
5917
5918 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5919 static inline void register_as_ext2(void)
5920 {
5921         int err = register_filesystem(&ext2_fs_type);
5922         if (err)
5923                 printk(KERN_WARNING
5924                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5925 }
5926
5927 static inline void unregister_as_ext2(void)
5928 {
5929         unregister_filesystem(&ext2_fs_type);
5930 }
5931
5932 static inline int ext2_feature_set_ok(struct super_block *sb)
5933 {
5934         if (ext4_has_unknown_ext2_incompat_features(sb))
5935                 return 0;
5936         if (sb_rdonly(sb))
5937                 return 1;
5938         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5939                 return 0;
5940         return 1;
5941 }
5942 #else
5943 static inline void register_as_ext2(void) { }
5944 static inline void unregister_as_ext2(void) { }
5945 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5946 #endif
5947
5948 static inline void register_as_ext3(void)
5949 {
5950         int err = register_filesystem(&ext3_fs_type);
5951         if (err)
5952                 printk(KERN_WARNING
5953                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5954 }
5955
5956 static inline void unregister_as_ext3(void)
5957 {
5958         unregister_filesystem(&ext3_fs_type);
5959 }
5960
5961 static inline int ext3_feature_set_ok(struct super_block *sb)
5962 {
5963         if (ext4_has_unknown_ext3_incompat_features(sb))
5964                 return 0;
5965         if (!ext4_has_feature_journal(sb))
5966                 return 0;
5967         if (sb_rdonly(sb))
5968                 return 1;
5969         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5970                 return 0;
5971         return 1;
5972 }
5973
5974 static struct file_system_type ext4_fs_type = {
5975         .owner          = THIS_MODULE,
5976         .name           = "ext4",
5977         .mount          = ext4_mount,
5978         .kill_sb        = kill_block_super,
5979         .fs_flags       = FS_REQUIRES_DEV,
5980 };
5981 MODULE_ALIAS_FS("ext4");
5982
5983 /* Shared across all ext4 file systems */
5984 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5985
5986 static int __init ext4_init_fs(void)
5987 {
5988         int i, err;
5989
5990         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5991         ext4_li_info = NULL;
5992         mutex_init(&ext4_li_mtx);
5993
5994         /* Build-time check for flags consistency */
5995         ext4_check_flag_values();
5996
5997         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5998                 init_waitqueue_head(&ext4__ioend_wq[i]);
5999
6000         err = ext4_init_es();
6001         if (err)
6002                 return err;
6003
6004         err = ext4_init_pending();
6005         if (err)
6006                 goto out6;
6007
6008         err = ext4_init_pageio();
6009         if (err)
6010                 goto out5;
6011
6012         err = ext4_init_system_zone();
6013         if (err)
6014                 goto out4;
6015
6016         err = ext4_init_sysfs();
6017         if (err)
6018                 goto out3;
6019
6020         err = ext4_init_mballoc();
6021         if (err)
6022                 goto out2;
6023         err = init_inodecache();
6024         if (err)
6025                 goto out1;
6026         register_as_ext3();
6027         register_as_ext2();
6028         err = register_filesystem(&ext4_fs_type);
6029         if (err)
6030                 goto out;
6031
6032         return 0;
6033 out:
6034         unregister_as_ext2();
6035         unregister_as_ext3();
6036         destroy_inodecache();
6037 out1:
6038         ext4_exit_mballoc();
6039 out2:
6040         ext4_exit_sysfs();
6041 out3:
6042         ext4_exit_system_zone();
6043 out4:
6044         ext4_exit_pageio();
6045 out5:
6046         ext4_exit_pending();
6047 out6:
6048         ext4_exit_es();
6049
6050         return err;
6051 }
6052
6053 static void __exit ext4_exit_fs(void)
6054 {
6055         ext4_destroy_lazyinit_thread();
6056         unregister_as_ext2();
6057         unregister_as_ext3();
6058         unregister_filesystem(&ext4_fs_type);
6059         destroy_inodecache();
6060         ext4_exit_mballoc();
6061         ext4_exit_sysfs();
6062         ext4_exit_system_zone();
6063         ext4_exit_pageio();
6064         ext4_exit_es();
6065         ext4_exit_pending();
6066 }
6067
6068 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6069 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6070 MODULE_LICENSE("GPL");
6071 MODULE_SOFTDEP("pre: crc32c");
6072 module_init(ext4_init_fs)
6073 module_exit(ext4_exit_fs)