fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/cleancache.h>
  41 #include <linux/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_extents.h"       /* Needed for trace points definition */
  48 #include "ext4_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52 #include "fsmap.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct ext4_lazy_init *ext4_li_info;
  58 static struct mutex ext4_li_mtx;
  59 static struct ratelimit_state ext4_mount_msg_ratelimit;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  71 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  72 static int ext4_unfreeze(struct super_block *sb);
  73 static int ext4_freeze(struct super_block *sb);
  74 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  75                        const char *dev_name, void *data);
  76 static inline int ext2_feature_set_ok(struct super_block *sb);
  77 static inline int ext3_feature_set_ok(struct super_block *sb);
  78 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  79 static void ext4_destroy_lazyinit_thread(void);
  80 static void ext4_unregister_li_request(struct super_block *sb);
  81 static void ext4_clear_request_list(void);
  82 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  83                                             unsigned int journal_inum);
  84
  85 /*
  86  * Lock ordering
  87  *
  88  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  89  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  90  *
  91  * page fault path:
  92  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  93  *   page lock -> i_data_sem (rw)
  94  *
  95  * buffered write path:
  96  * sb_start_write -> i_mutex -> mmap_sem
  97  * sb_start_write -> i_mutex -> transaction start -> page lock ->
  98  *   i_data_sem (rw)
  99  *
 100  * truncate:
 101  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 102  *   i_mmap_rwsem (w) -> page lock
 103  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 104  *   transaction start -> i_data_sem (rw)
 105  *
 106  * direct IO:
 107  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
 108  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
 109  *   transaction start -> i_data_sem (rw)
 110  *
 111  * writepages:
 112  * transaction start -> page lock(s) -> i_data_sem (rw)
 113  */
 114
 115 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 116 static struct file_system_type ext2_fs_type = {
 117         .owner          = THIS_MODULE,
 118         .name           = "ext2",
 119         .mount          = ext4_mount,
 120         .kill_sb        = kill_block_super,
 121         .fs_flags       = FS_REQUIRES_DEV,
 122 };
 123 MODULE_ALIAS_FS("ext2");
 124 MODULE_ALIAS("ext2");
 125 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 126 #else
 127 #define IS_EXT2_SB(sb) (0)
 128 #endif
 129
 130
 131 static struct file_system_type ext3_fs_type = {
 132         .owner          = THIS_MODULE,
 133         .name           = "ext3",
 134         .mount          = ext4_mount,
 135         .kill_sb        = kill_block_super,
 136         .fs_flags       = FS_REQUIRES_DEV,
 137 };
 138 MODULE_ALIAS_FS("ext3");
 139 MODULE_ALIAS("ext3");
 140 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 141
 142 static int ext4_verify_csum_type(struct super_block *sb,
 143                                  struct ext4_super_block *es)
 144 {
 145         if (!ext4_has_feature_metadata_csum(sb))
 146                 return 1;
 147
 148         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 149 }
 150
 151 static __le32 ext4_superblock_csum(struct super_block *sb,
 152                                    struct ext4_super_block *es)
 153 {
 154         struct ext4_sb_info *sbi = EXT4_SB(sb);
 155         int offset = offsetof(struct ext4_super_block, s_checksum);
 156         __u32 csum;
 157
 158         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 159
 160         return cpu_to_le32(csum);
 161 }
 162
 163 static int ext4_superblock_csum_verify(struct super_block *sb,
 164                                        struct ext4_super_block *es)
 165 {
 166         if (!ext4_has_metadata_csum(sb))
 167                 return 1;
 168
 169         return es->s_checksum == ext4_superblock_csum(sb, es);
 170 }
 171
 172 void ext4_superblock_csum_set(struct super_block *sb)
 173 {
 174         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 175
 176         if (!ext4_has_metadata_csum(sb))
 177                 return;
 178
 179         es->s_checksum = ext4_superblock_csum(sb, es);
 180 }
 181
 182 void *ext4_kvmalloc(size_t size, gfp_t flags)
 183 {
 184         void *ret;
 185
 186         ret = kmalloc(size, flags | __GFP_NOWARN);
 187         if (!ret)
 188                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 189         return ret;
 190 }
 191
 192 void *ext4_kvzalloc(size_t size, gfp_t flags)
 193 {
 194         void *ret;
 195
 196         ret = kzalloc(size, flags | __GFP_NOWARN);
 197         if (!ret)
 198                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 199         return ret;
 200 }
 201
 202 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 203                                struct ext4_group_desc *bg)
 204 {
 205         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 208 }
 209
 210 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 211                                struct ext4_group_desc *bg)
 212 {
 213         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 216 }
 217
 218 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le32_to_cpu(bg->bg_inode_table_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 224 }
 225
 226 __u32 ext4_free_group_clusters(struct super_block *sb,
 227                                struct ext4_group_desc *bg)
 228 {
 229         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 230                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 231                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 232 }
 233
 234 __u32 ext4_free_inodes_count(struct super_block *sb,
 235                               struct ext4_group_desc *bg)
 236 {
 237         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 238                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 239                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 240 }
 241
 242 __u32 ext4_used_dirs_count(struct super_block *sb,
 243                               struct ext4_group_desc *bg)
 244 {
 245         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 246                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 247                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 248 }
 249
 250 __u32 ext4_itable_unused_count(struct super_block *sb,
 251                               struct ext4_group_desc *bg)
 252 {
 253         return le16_to_cpu(bg->bg_itable_unused_lo) |
 254                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 255                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 256 }
 257
 258 void ext4_block_bitmap_set(struct super_block *sb,
 259                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 260 {
 261         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 264 }
 265
 266 void ext4_inode_bitmap_set(struct super_block *sb,
 267                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 268 {
 269         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 272 }
 273
 274 void ext4_inode_table_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 276 {
 277         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 280 }
 281
 282 void ext4_free_group_clusters_set(struct super_block *sb,
 283                                   struct ext4_group_desc *bg, __u32 count)
 284 {
 285         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 286         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 287                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 288 }
 289
 290 void ext4_free_inodes_set(struct super_block *sb,
 291                           struct ext4_group_desc *bg, __u32 count)
 292 {
 293         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 294         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 295                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 296 }
 297
 298 void ext4_used_dirs_set(struct super_block *sb,
 299                           struct ext4_group_desc *bg, __u32 count)
 300 {
 301         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 302         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 303                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 304 }
 305
 306 void ext4_itable_unused_set(struct super_block *sb,
 307                           struct ext4_group_desc *bg, __u32 count)
 308 {
 309         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 310         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 311                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 312 }
 313
 314
 315 static void __save_error_info(struct super_block *sb, const char *func,
 316                             unsigned int line)
 317 {
 318         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 319
 320         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 321         if (bdev_read_only(sb->s_bdev))
 322                 return;
 323         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 324         es->s_last_error_time = cpu_to_le32(get_seconds());
 325         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 326         es->s_last_error_line = cpu_to_le32(line);
 327         if (!es->s_first_error_time) {
 328                 es->s_first_error_time = es->s_last_error_time;
 329                 strncpy(es->s_first_error_func, func,
 330                         sizeof(es->s_first_error_func));
 331                 es->s_first_error_line = cpu_to_le32(line);
 332                 es->s_first_error_ino = es->s_last_error_ino;
 333                 es->s_first_error_block = es->s_last_error_block;
 334         }
 335         /*
 336          * Start the daily error reporting function if it hasn't been
 337          * started already
 338          */
 339         if (!es->s_error_count)
 340                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 341         le32_add_cpu(&es->s_error_count, 1);
 342 }
 343
 344 static void save_error_info(struct super_block *sb, const char *func,
 345                             unsigned int line)
 346 {
 347         __save_error_info(sb, func, line);
 348         ext4_commit_super(sb, 1);
 349 }
 350
 351 /*
 352  * The del_gendisk() function uninitializes the disk-specific data
 353  * structures, including the bdi structure, without telling anyone
 354  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 355  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 356  * This is a kludge to prevent these oops until we can put in a proper
 357  * hook in del_gendisk() to inform the VFS and file system layers.
 358  */
 359 static int block_device_ejected(struct super_block *sb)
 360 {
 361         struct inode *bd_inode = sb->s_bdev->bd_inode;
 362         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 363
 364         return bdi->dev == NULL;
 365 }
 366
 367 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 368 {
 369         struct super_block              *sb = journal->j_private;
 370         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 371         int                             error = is_journal_aborted(journal);
 372         struct ext4_journal_cb_entry    *jce;
 373
 374         BUG_ON(txn->t_state == T_FINISHED);
 375         spin_lock(&sbi->s_md_lock);
 376         while (!list_empty(&txn->t_private_list)) {
 377                 jce = list_entry(txn->t_private_list.next,
 378                                  struct ext4_journal_cb_entry, jce_list);
 379                 list_del_init(&jce->jce_list);
 380                 spin_unlock(&sbi->s_md_lock);
 381                 jce->jce_func(sb, jce, error);
 382                 spin_lock(&sbi->s_md_lock);
 383         }
 384         spin_unlock(&sbi->s_md_lock);
 385 }
 386
 387 /* Deal with the reporting of failure conditions on a filesystem such as
 388  * inconsistencies detected or read IO failures.
 389  *
 390  * On ext2, we can store the error state of the filesystem in the
 391  * superblock.  That is not possible on ext4, because we may have other
 392  * write ordering constraints on the superblock which prevent us from
 393  * writing it out straight away; and given that the journal is about to
 394  * be aborted, we can't rely on the current, or future, transactions to
 395  * write out the superblock safely.
 396  *
 397  * We'll just use the jbd2_journal_abort() error code to record an error in
 398  * the journal instead.  On recovery, the journal will complain about
 399  * that error until we've noted it down and cleared it.
 400  */
 401
 402 static void ext4_handle_error(struct super_block *sb)
 403 {
 404         if (sb->s_flags & MS_RDONLY)
 405                 return;
 406
 407         if (!test_opt(sb, ERRORS_CONT)) {
 408                 journal_t *journal = EXT4_SB(sb)->s_journal;
 409
 410                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 411                 if (journal)
 412                         jbd2_journal_abort(journal, -EIO);
 413         }
 414         if (test_opt(sb, ERRORS_RO)) {
 415                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 416                 /*
 417                  * Make sure updated value of ->s_mount_flags will be visible
 418                  * before ->s_flags update
 419                  */
 420                 smp_wmb();
 421                 sb->s_flags |= MS_RDONLY;
 422         }
 423         if (test_opt(sb, ERRORS_PANIC)) {
 424                 if (EXT4_SB(sb)->s_journal &&
 425                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 426                         return;
 427                 panic("EXT4-fs (device %s): panic forced after error\n",
 428                         sb->s_id);
 429         }
 430 }
 431
 432 #define ext4_error_ratelimit(sb)                                        \
 433                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 434                              "EXT4-fs error")
 435
 436 void __ext4_error(struct super_block *sb, const char *function,
 437                   unsigned int line, const char *fmt, ...)
 438 {
 439         struct va_format vaf;
 440         va_list args;
 441
 442         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 443                 return;
 444
 445         if (ext4_error_ratelimit(sb)) {
 446                 va_start(args, fmt);
 447                 vaf.fmt = fmt;
 448                 vaf.va = &args;
 449                 printk(KERN_CRIT
 450                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 451                        sb->s_id, function, line, current->comm, &vaf);
 452                 va_end(args);
 453         }
 454         save_error_info(sb, function, line);
 455         ext4_handle_error(sb);
 456 }
 457
 458 void __ext4_error_inode(struct inode *inode, const char *function,
 459                         unsigned int line, ext4_fsblk_t block,
 460                         const char *fmt, ...)
 461 {
 462         va_list args;
 463         struct va_format vaf;
 464         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 465
 466         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 467                 return;
 468
 469         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 470         es->s_last_error_block = cpu_to_le64(block);
 471         if (ext4_error_ratelimit(inode->i_sb)) {
 472                 va_start(args, fmt);
 473                 vaf.fmt = fmt;
 474                 vaf.va = &args;
 475                 if (block)
 476                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 477                                "inode #%lu: block %llu: comm %s: %pV\n",
 478                                inode->i_sb->s_id, function, line, inode->i_ino,
 479                                block, current->comm, &vaf);
 480                 else
 481                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 482                                "inode #%lu: comm %s: %pV\n",
 483                                inode->i_sb->s_id, function, line, inode->i_ino,
 484                                current->comm, &vaf);
 485                 va_end(args);
 486         }
 487         save_error_info(inode->i_sb, function, line);
 488         ext4_handle_error(inode->i_sb);
 489 }
 490
 491 void __ext4_error_file(struct file *file, const char *function,
 492                        unsigned int line, ext4_fsblk_t block,
 493                        const char *fmt, ...)
 494 {
 495         va_list args;
 496         struct va_format vaf;
 497         struct ext4_super_block *es;
 498         struct inode *inode = file_inode(file);
 499         char pathname[80], *path;
 500
 501         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 502                 return;
 503
 504         es = EXT4_SB(inode->i_sb)->s_es;
 505         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 506         if (ext4_error_ratelimit(inode->i_sb)) {
 507                 path = file_path(file, pathname, sizeof(pathname));
 508                 if (IS_ERR(path))
 509                         path = "(unknown)";
 510                 va_start(args, fmt);
 511                 vaf.fmt = fmt;
 512                 vaf.va = &args;
 513                 if (block)
 514                         printk(KERN_CRIT
 515                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 516                                "block %llu: comm %s: path %s: %pV\n",
 517                                inode->i_sb->s_id, function, line, inode->i_ino,
 518                                block, current->comm, path, &vaf);
 519                 else
 520                         printk(KERN_CRIT
 521                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 522                                "comm %s: path %s: %pV\n",
 523                                inode->i_sb->s_id, function, line, inode->i_ino,
 524                                current->comm, path, &vaf);
 525                 va_end(args);
 526         }
 527         save_error_info(inode->i_sb, function, line);
 528         ext4_handle_error(inode->i_sb);
 529 }
 530
 531 const char *ext4_decode_error(struct super_block *sb, int errno,
 532                               char nbuf[16])
 533 {
 534         char *errstr = NULL;
 535
 536         switch (errno) {
 537         case -EFSCORRUPTED:
 538                 errstr = "Corrupt filesystem";
 539                 break;
 540         case -EFSBADCRC:
 541                 errstr = "Filesystem failed CRC";
 542                 break;
 543         case -EIO:
 544                 errstr = "IO failure";
 545                 break;
 546         case -ENOMEM:
 547                 errstr = "Out of memory";
 548                 break;
 549         case -EROFS:
 550                 if (!sb || (EXT4_SB(sb)->s_journal &&
 551                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 552                         errstr = "Journal has aborted";
 553                 else
 554                         errstr = "Readonly filesystem";
 555                 break;
 556         default:
 557                 /* If the caller passed in an extra buffer for unknown
 558                  * errors, textualise them now.  Else we just return
 559                  * NULL. */
 560                 if (nbuf) {
 561                         /* Check for truncated error codes... */
 562                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 563                                 errstr = nbuf;
 564                 }
 565                 break;
 566         }
 567
 568         return errstr;
 569 }
 570
 571 /* __ext4_std_error decodes expected errors from journaling functions
 572  * automatically and invokes the appropriate error response.  */
 573
 574 void __ext4_std_error(struct super_block *sb, const char *function,
 575                       unsigned int line, int errno)
 576 {
 577         char nbuf[16];
 578         const char *errstr;
 579
 580         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 581                 return;
 582
 583         /* Special case: if the error is EROFS, and we're not already
 584          * inside a transaction, then there's really no point in logging
 585          * an error. */
 586         if (errno == -EROFS && journal_current_handle() == NULL &&
 587             (sb->s_flags & MS_RDONLY))
 588                 return;
 589
 590         if (ext4_error_ratelimit(sb)) {
 591                 errstr = ext4_decode_error(sb, errno, nbuf);
 592                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 593                        sb->s_id, function, line, errstr);
 594         }
 595
 596         save_error_info(sb, function, line);
 597         ext4_handle_error(sb);
 598 }
 599
 600 /*
 601  * ext4_abort is a much stronger failure handler than ext4_error.  The
 602  * abort function may be used to deal with unrecoverable failures such
 603  * as journal IO errors or ENOMEM at a critical moment in log management.
 604  *
 605  * We unconditionally force the filesystem into an ABORT|READONLY state,
 606  * unless the error response on the fs has been set to panic in which
 607  * case we take the easy way out and panic immediately.
 608  */
 609
 610 void __ext4_abort(struct super_block *sb, const char *function,
 611                 unsigned int line, const char *fmt, ...)
 612 {
 613         struct va_format vaf;
 614         va_list args;
 615
 616         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 617                 return;
 618
 619         save_error_info(sb, function, line);
 620         va_start(args, fmt);
 621         vaf.fmt = fmt;
 622         vaf.va = &args;
 623         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 624                sb->s_id, function, line, &vaf);
 625         va_end(args);
 626
 627         if ((sb->s_flags & MS_RDONLY) == 0) {
 628                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 629                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 630                 /*
 631                  * Make sure updated value of ->s_mount_flags will be visible
 632                  * before ->s_flags update
 633                  */
 634                 smp_wmb();
 635                 sb->s_flags |= MS_RDONLY;
 636                 if (EXT4_SB(sb)->s_journal)
 637                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 638                 save_error_info(sb, function, line);
 639         }
 640         if (test_opt(sb, ERRORS_PANIC)) {
 641                 if (EXT4_SB(sb)->s_journal &&
 642                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 643                         return;
 644                 panic("EXT4-fs panic from previous error\n");
 645         }
 646 }
 647
 648 void __ext4_msg(struct super_block *sb,
 649                 const char *prefix, const char *fmt, ...)
 650 {
 651         struct va_format vaf;
 652         va_list args;
 653
 654         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 655                 return;
 656
 657         va_start(args, fmt);
 658         vaf.fmt = fmt;
 659         vaf.va = &args;
 660         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 661         va_end(args);
 662 }
 663
 664 #define ext4_warning_ratelimit(sb)                                      \
 665                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 666                              "EXT4-fs warning")
 667
 668 void __ext4_warning(struct super_block *sb, const char *function,
 669                     unsigned int line, const char *fmt, ...)
 670 {
 671         struct va_format vaf;
 672         va_list args;
 673
 674         if (!ext4_warning_ratelimit(sb))
 675                 return;
 676
 677         va_start(args, fmt);
 678         vaf.fmt = fmt;
 679         vaf.va = &args;
 680         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 681                sb->s_id, function, line, &vaf);
 682         va_end(args);
 683 }
 684
 685 void __ext4_warning_inode(const struct inode *inode, const char *function,
 686                           unsigned int line, const char *fmt, ...)
 687 {
 688         struct va_format vaf;
 689         va_list args;
 690
 691         if (!ext4_warning_ratelimit(inode->i_sb))
 692                 return;
 693
 694         va_start(args, fmt);
 695         vaf.fmt = fmt;
 696         vaf.va = &args;
 697         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 698                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 699                function, line, inode->i_ino, current->comm, &vaf);
 700         va_end(args);
 701 }
 702
 703 void __ext4_grp_locked_error(const char *function, unsigned int line,
 704                              struct super_block *sb, ext4_group_t grp,
 705                              unsigned long ino, ext4_fsblk_t block,
 706                              const char *fmt, ...)
 707 __releases(bitlock)
 708 __acquires(bitlock)
 709 {
 710         struct va_format vaf;
 711         va_list args;
 712         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 713
 714         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 715                 return;
 716
 717         es->s_last_error_ino = cpu_to_le32(ino);
 718         es->s_last_error_block = cpu_to_le64(block);
 719         __save_error_info(sb, function, line);
 720
 721         if (ext4_error_ratelimit(sb)) {
 722                 va_start(args, fmt);
 723                 vaf.fmt = fmt;
 724                 vaf.va = &args;
 725                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 726                        sb->s_id, function, line, grp);
 727                 if (ino)
 728                         printk(KERN_CONT "inode %lu: ", ino);
 729                 if (block)
 730                         printk(KERN_CONT "block %llu:",
 731                                (unsigned long long) block);
 732                 printk(KERN_CONT "%pV\n", &vaf);
 733                 va_end(args);
 734         }
 735
 736         if (test_opt(sb, ERRORS_CONT)) {
 737                 ext4_commit_super(sb, 0);
 738                 return;
 739         }
 740
 741         ext4_unlock_group(sb, grp);
 742         ext4_handle_error(sb);
 743         /*
 744          * We only get here in the ERRORS_RO case; relocking the group
 745          * may be dangerous, but nothing bad will happen since the
 746          * filesystem will have already been marked read/only and the
 747          * journal has been aborted.  We return 1 as a hint to callers
 748          * who might what to use the return value from
 749          * ext4_grp_locked_error() to distinguish between the
 750          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 751          * aggressively from the ext4 function in question, with a
 752          * more appropriate error code.
 753          */
 754         ext4_lock_group(sb, grp);
 755         return;
 756 }
 757
 758 void ext4_update_dynamic_rev(struct super_block *sb)
 759 {
 760         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 761
 762         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 763                 return;
 764
 765         ext4_warning(sb,
 766                      "updating to rev %d because of new feature flag, "
 767                      "running e2fsck is recommended",
 768                      EXT4_DYNAMIC_REV);
 769
 770         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 771         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 772         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 773         /* leave es->s_feature_*compat flags alone */
 774         /* es->s_uuid will be set by e2fsck if empty */
 775
 776         /*
 777          * The rest of the superblock fields should be zero, and if not it
 778          * means they are likely already in use, so leave them alone.  We
 779          * can leave it up to e2fsck to clean up any inconsistencies there.
 780          */
 781 }
 782
 783 /*
 784  * Open the external journal device
 785  */
 786 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 787 {
 788         struct block_device *bdev;
 789         char b[BDEVNAME_SIZE];
 790
 791         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 792         if (IS_ERR(bdev))
 793                 goto fail;
 794         return bdev;
 795
 796 fail:
 797         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 798                         __bdevname(dev, b), PTR_ERR(bdev));
 799         return NULL;
 800 }
 801
 802 /*
 803  * Release the journal device
 804  */
 805 static void ext4_blkdev_put(struct block_device *bdev)
 806 {
 807         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 808 }
 809
 810 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 811 {
 812         struct block_device *bdev;
 813         bdev = sbi->journal_bdev;
 814         if (bdev) {
 815                 ext4_blkdev_put(bdev);
 816                 sbi->journal_bdev = NULL;
 817         }
 818 }
 819
 820 static inline struct inode *orphan_list_entry(struct list_head *l)
 821 {
 822         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 823 }
 824
 825 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 826 {
 827         struct list_head *l;
 828
 829         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 830                  le32_to_cpu(sbi->s_es->s_last_orphan));
 831
 832         printk(KERN_ERR "sb_info orphan list:\n");
 833         list_for_each(l, &sbi->s_orphan) {
 834                 struct inode *inode = orphan_list_entry(l);
 835                 printk(KERN_ERR "  "
 836                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 837                        inode->i_sb->s_id, inode->i_ino, inode,
 838                        inode->i_mode, inode->i_nlink,
 839                        NEXT_ORPHAN(inode));
 840         }
 841 }
 842
 843 #ifdef CONFIG_QUOTA
 844 static int ext4_quota_off(struct super_block *sb, int type);
 845
 846 static inline void ext4_quota_off_umount(struct super_block *sb)
 847 {
 848         int type;
 849
 850         if (ext4_has_feature_quota(sb)) {
 851                 dquot_disable(sb, -1,
 852                               DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 853         } else {
 854                 /* Use our quota_off function to clear inode flags etc. */
 855                 for (type = 0; type < EXT4_MAXQUOTAS; type++)
 856                         ext4_quota_off(sb, type);
 857         }
 858 }
 859 #else
 860 static inline void ext4_quota_off_umount(struct super_block *sb)
 861 {
 862 }
 863 #endif
 864
 865 static void ext4_put_super(struct super_block *sb)
 866 {
 867         struct ext4_sb_info *sbi = EXT4_SB(sb);
 868         struct ext4_super_block *es = sbi->s_es;
 869         int aborted = 0;
 870         int i, err;
 871
 872         ext4_unregister_li_request(sb);
 873         ext4_quota_off_umount(sb);
 874
 875         flush_workqueue(sbi->rsv_conversion_wq);
 876         destroy_workqueue(sbi->rsv_conversion_wq);
 877
 878         if (sbi->s_journal) {
 879                 aborted = is_journal_aborted(sbi->s_journal);
 880                 err = jbd2_journal_destroy(sbi->s_journal);
 881                 sbi->s_journal = NULL;
 882                 if ((err < 0) && !aborted)
 883                         ext4_abort(sb, "Couldn't clean up the journal");
 884         }
 885
 886         ext4_unregister_sysfs(sb);
 887         ext4_es_unregister_shrinker(sbi);
 888         del_timer_sync(&sbi->s_err_report);
 889         ext4_release_system_zone(sb);
 890         ext4_mb_release(sb);
 891         ext4_ext_release(sb);
 892
 893         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
 894                 ext4_clear_feature_journal_needs_recovery(sb);
 895                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 896         }
 897         if (!(sb->s_flags & MS_RDONLY))
 898                 ext4_commit_super(sb, 1);
 899
 900         for (i = 0; i < sbi->s_gdb_count; i++)
 901                 brelse(sbi->s_group_desc[i]);
 902         kvfree(sbi->s_group_desc);
 903         kvfree(sbi->s_flex_groups);
 904         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 905         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 906         percpu_counter_destroy(&sbi->s_dirs_counter);
 907         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 908         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 909 #ifdef CONFIG_QUOTA
 910         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 911                 kfree(sbi->s_qf_names[i]);
 912 #endif
 913
 914         /* Debugging code just in case the in-memory inode orphan list
 915          * isn't empty.  The on-disk one can be non-empty if we've
 916          * detected an error and taken the fs readonly, but the
 917          * in-memory list had better be clean by this point. */
 918         if (!list_empty(&sbi->s_orphan))
 919                 dump_orphan_list(sb, sbi);
 920         J_ASSERT(list_empty(&sbi->s_orphan));
 921
 922         sync_blockdev(sb->s_bdev);
 923         invalidate_bdev(sb->s_bdev);
 924         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 925                 /*
 926                  * Invalidate the journal device's buffers.  We don't want them
 927                  * floating about in memory - the physical journal device may
 928                  * hotswapped, and it breaks the `ro-after' testing code.
 929                  */
 930                 sync_blockdev(sbi->journal_bdev);
 931                 invalidate_bdev(sbi->journal_bdev);
 932                 ext4_blkdev_remove(sbi);
 933         }
 934         if (sbi->s_mb_cache) {
 935                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 936                 sbi->s_mb_cache = NULL;
 937         }
 938         if (sbi->s_mmp_tsk)
 939                 kthread_stop(sbi->s_mmp_tsk);
 940         brelse(sbi->s_sbh);
 941         sb->s_fs_info = NULL;
 942         /*
 943          * Now that we are completely done shutting down the
 944          * superblock, we need to actually destroy the kobject.
 945          */
 946         kobject_put(&sbi->s_kobj);
 947         wait_for_completion(&sbi->s_kobj_unregister);
 948         if (sbi->s_chksum_driver)
 949                 crypto_free_shash(sbi->s_chksum_driver);
 950         kfree(sbi->s_blockgroup_lock);
 951         kfree(sbi);
 952 }
 953
 954 static struct kmem_cache *ext4_inode_cachep;
 955
 956 /*
 957  * Called inside transaction, so use GFP_NOFS
 958  */
 959 static struct inode *ext4_alloc_inode(struct super_block *sb)
 960 {
 961         struct ext4_inode_info *ei;
 962
 963         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 964         if (!ei)
 965                 return NULL;
 966
 967         ei->vfs_inode.i_version = 1;
 968         spin_lock_init(&ei->i_raw_lock);
 969         INIT_LIST_HEAD(&ei->i_prealloc_list);
 970         spin_lock_init(&ei->i_prealloc_lock);
 971         ext4_es_init_tree(&ei->i_es_tree);
 972         rwlock_init(&ei->i_es_lock);
 973         INIT_LIST_HEAD(&ei->i_es_list);
 974         ei->i_es_all_nr = 0;
 975         ei->i_es_shk_nr = 0;
 976         ei->i_es_shrink_lblk = 0;
 977         ei->i_reserved_data_blocks = 0;
 978         ei->i_reserved_meta_blocks = 0;
 979         ei->i_allocated_meta_blocks = 0;
 980         ei->i_da_metadata_calc_len = 0;
 981         ei->i_da_metadata_calc_last_lblock = 0;
 982         spin_lock_init(&(ei->i_block_reservation_lock));
 983 #ifdef CONFIG_QUOTA
 984         ei->i_reserved_quota = 0;
 985         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 986 #endif
 987         ei->jinode = NULL;
 988         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 989         spin_lock_init(&ei->i_completed_io_lock);
 990         ei->i_sync_tid = 0;
 991         ei->i_datasync_tid = 0;
 992         atomic_set(&ei->i_unwritten, 0);
 993         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 994         return &ei->vfs_inode;
 995 }
 996
 997 static int ext4_drop_inode(struct inode *inode)
 998 {
 999         int drop = generic_drop_inode(inode);
1000
1001         trace_ext4_drop_inode(inode, drop);
1002         return drop;
1003 }
1004
1005 static void ext4_i_callback(struct rcu_head *head)
1006 {
1007         struct inode *inode = container_of(head, struct inode, i_rcu);
1008         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1009 }
1010
1011 static void ext4_destroy_inode(struct inode *inode)
1012 {
1013         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1014                 ext4_msg(inode->i_sb, KERN_ERR,
1015                          "Inode %lu (%p): orphan list check failed!",
1016                          inode->i_ino, EXT4_I(inode));
1017                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1018                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1019                                 true);
1020                 dump_stack();
1021         }
1022         call_rcu(&inode->i_rcu, ext4_i_callback);
1023 }
1024
1025 static void init_once(void *foo)
1026 {
1027         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1028
1029         INIT_LIST_HEAD(&ei->i_orphan);
1030         init_rwsem(&ei->xattr_sem);
1031         init_rwsem(&ei->i_data_sem);
1032         init_rwsem(&ei->i_mmap_sem);
1033         inode_init_once(&ei->vfs_inode);
1034 }
1035
1036 static int __init init_inodecache(void)
1037 {
1038         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1039                                              sizeof(struct ext4_inode_info),
1040                                              0, (SLAB_RECLAIM_ACCOUNT|
1041                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1042                                              init_once);
1043         if (ext4_inode_cachep == NULL)
1044                 return -ENOMEM;
1045         return 0;
1046 }
1047
1048 static void destroy_inodecache(void)
1049 {
1050         /*
1051          * Make sure all delayed rcu free inodes are flushed before we
1052          * destroy cache.
1053          */
1054         rcu_barrier();
1055         kmem_cache_destroy(ext4_inode_cachep);
1056 }
1057
1058 void ext4_clear_inode(struct inode *inode)
1059 {
1060         invalidate_inode_buffers(inode);
1061         clear_inode(inode);
1062         dquot_drop(inode);
1063         ext4_discard_preallocations(inode);
1064         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1065         if (EXT4_I(inode)->jinode) {
1066                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1067                                                EXT4_I(inode)->jinode);
1068                 jbd2_free_inode(EXT4_I(inode)->jinode);
1069                 EXT4_I(inode)->jinode = NULL;
1070         }
1071 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1072         fscrypt_put_encryption_info(inode, NULL);
1073 #endif
1074 }
1075
1076 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1077                                         u64 ino, u32 generation)
1078 {
1079         struct inode *inode;
1080
1081         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1082                 return ERR_PTR(-ESTALE);
1083         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1084                 return ERR_PTR(-ESTALE);
1085
1086         /* iget isn't really right if the inode is currently unallocated!!
1087          *
1088          * ext4_read_inode will return a bad_inode if the inode had been
1089          * deleted, so we should be safe.
1090          *
1091          * Currently we don't know the generation for parent directory, so
1092          * a generation of 0 means "accept any"
1093          */
1094         inode = ext4_iget_normal(sb, ino);
1095         if (IS_ERR(inode))
1096                 return ERR_CAST(inode);
1097         if (generation && inode->i_generation != generation) {
1098                 iput(inode);
1099                 return ERR_PTR(-ESTALE);
1100         }
1101
1102         return inode;
1103 }
1104
1105 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1106                                         int fh_len, int fh_type)
1107 {
1108         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1109                                     ext4_nfs_get_inode);
1110 }
1111
1112 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1113                                         int fh_len, int fh_type)
1114 {
1115         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1116                                     ext4_nfs_get_inode);
1117 }
1118
1119 /*
1120  * Try to release metadata pages (indirect blocks, directories) which are
1121  * mapped via the block device.  Since these pages could have journal heads
1122  * which would prevent try_to_free_buffers() from freeing them, we must use
1123  * jbd2 layer's try_to_free_buffers() function to release them.
1124  */
1125 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1126                                  gfp_t wait)
1127 {
1128         journal_t *journal = EXT4_SB(sb)->s_journal;
1129
1130         WARN_ON(PageChecked(page));
1131         if (!page_has_buffers(page))
1132                 return 0;
1133         if (journal)
1134                 return jbd2_journal_try_to_free_buffers(journal, page,
1135                                                 wait & ~__GFP_DIRECT_RECLAIM);
1136         return try_to_free_buffers(page);
1137 }
1138
1139 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1140 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1141 {
1142         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1143                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1144 }
1145
1146 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1147                                                         void *fs_data)
1148 {
1149         handle_t *handle = fs_data;
1150         int res, res2, retries = 0;
1151
1152         res = ext4_convert_inline_data(inode);
1153         if (res)
1154                 return res;
1155
1156         /*
1157          * If a journal handle was specified, then the encryption context is
1158          * being set on a new inode via inheritance and is part of a larger
1159          * transaction to create the inode.  Otherwise the encryption context is
1160          * being set on an existing inode in its own transaction.  Only in the
1161          * latter case should the "retry on ENOSPC" logic be used.
1162          */
1163
1164         if (handle) {
1165                 res = ext4_xattr_set_handle(handle, inode,
1166                                             EXT4_XATTR_INDEX_ENCRYPTION,
1167                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1168                                             ctx, len, 0);
1169                 if (!res) {
1170                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1171                         ext4_clear_inode_state(inode,
1172                                         EXT4_STATE_MAY_INLINE_DATA);
1173                         /*
1174                          * Update inode->i_flags - e.g. S_DAX may get disabled
1175                          */
1176                         ext4_set_inode_flags(inode);
1177                 }
1178                 return res;
1179         }
1180
1181 retry:
1182         handle = ext4_journal_start(inode, EXT4_HT_MISC,
1183                         ext4_jbd2_credits_xattr(inode));
1184         if (IS_ERR(handle))
1185                 return PTR_ERR(handle);
1186
1187         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1188                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1189                                     ctx, len, 0);
1190         if (!res) {
1191                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1192                 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1193                 ext4_set_inode_flags(inode);
1194                 res = ext4_mark_inode_dirty(handle, inode);
1195                 if (res)
1196                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1197         }
1198         res2 = ext4_journal_stop(handle);
1199
1200         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1201                 goto retry;
1202         if (!res)
1203                 res = res2;
1204         return res;
1205 }
1206
1207 static int ext4_dummy_context(struct inode *inode)
1208 {
1209         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1210 }
1211
1212 static unsigned ext4_max_namelen(struct inode *inode)
1213 {
1214         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1215                 EXT4_NAME_LEN;
1216 }
1217
1218 static const struct fscrypt_operations ext4_cryptops = {
1219         .key_prefix             = "ext4:",
1220         .get_context            = ext4_get_context,
1221         .set_context            = ext4_set_context,
1222         .dummy_context          = ext4_dummy_context,
1223         .is_encrypted           = ext4_encrypted_inode,
1224         .empty_dir              = ext4_empty_dir,
1225         .max_namelen            = ext4_max_namelen,
1226 };
1227 #else
1228 static const struct fscrypt_operations ext4_cryptops = {
1229         .is_encrypted           = ext4_encrypted_inode,
1230 };
1231 #endif
1232
1233 #ifdef CONFIG_QUOTA
1234 static const char * const quotatypes[] = INITQFNAMES;
1235 #define QTYPE2NAME(t) (quotatypes[t])
1236
1237 static int ext4_write_dquot(struct dquot *dquot);
1238 static int ext4_acquire_dquot(struct dquot *dquot);
1239 static int ext4_release_dquot(struct dquot *dquot);
1240 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1241 static int ext4_write_info(struct super_block *sb, int type);
1242 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1243                          const struct path *path);
1244 static int ext4_quota_on_mount(struct super_block *sb, int type);
1245 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1246                                size_t len, loff_t off);
1247 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1248                                 const char *data, size_t len, loff_t off);
1249 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1250                              unsigned int flags);
1251 static int ext4_enable_quotas(struct super_block *sb);
1252 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1253
1254 static struct dquot **ext4_get_dquots(struct inode *inode)
1255 {
1256         return EXT4_I(inode)->i_dquot;
1257 }
1258
1259 static const struct dquot_operations ext4_quota_operations = {
1260         .get_reserved_space = ext4_get_reserved_space,
1261         .write_dquot    = ext4_write_dquot,
1262         .acquire_dquot  = ext4_acquire_dquot,
1263         .release_dquot  = ext4_release_dquot,
1264         .mark_dirty     = ext4_mark_dquot_dirty,
1265         .write_info     = ext4_write_info,
1266         .alloc_dquot    = dquot_alloc,
1267         .destroy_dquot  = dquot_destroy,
1268         .get_projid     = ext4_get_projid,
1269         .get_next_id    = ext4_get_next_id,
1270 };
1271
1272 static const struct quotactl_ops ext4_qctl_operations = {
1273         .quota_on       = ext4_quota_on,
1274         .quota_off      = ext4_quota_off,
1275         .quota_sync     = dquot_quota_sync,
1276         .get_state      = dquot_get_state,
1277         .set_info       = dquot_set_dqinfo,
1278         .get_dqblk      = dquot_get_dqblk,
1279         .set_dqblk      = dquot_set_dqblk,
1280         .get_nextdqblk  = dquot_get_next_dqblk,
1281 };
1282 #endif
1283
1284 static const struct super_operations ext4_sops = {
1285         .alloc_inode    = ext4_alloc_inode,
1286         .destroy_inode  = ext4_destroy_inode,
1287         .write_inode    = ext4_write_inode,
1288         .dirty_inode    = ext4_dirty_inode,
1289         .drop_inode     = ext4_drop_inode,
1290         .evict_inode    = ext4_evict_inode,
1291         .put_super      = ext4_put_super,
1292         .sync_fs        = ext4_sync_fs,
1293         .freeze_fs      = ext4_freeze,
1294         .unfreeze_fs    = ext4_unfreeze,
1295         .statfs         = ext4_statfs,
1296         .remount_fs     = ext4_remount,
1297         .show_options   = ext4_show_options,
1298 #ifdef CONFIG_QUOTA
1299         .quota_read     = ext4_quota_read,
1300         .quota_write    = ext4_quota_write,
1301         .get_dquots     = ext4_get_dquots,
1302 #endif
1303         .bdev_try_to_free_page = bdev_try_to_free_page,
1304 };
1305
1306 static const struct export_operations ext4_export_ops = {
1307         .fh_to_dentry = ext4_fh_to_dentry,
1308         .fh_to_parent = ext4_fh_to_parent,
1309         .get_parent = ext4_get_parent,
1310 };
1311
1312 enum {
1313         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1314         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1315         Opt_nouid32, Opt_debug, Opt_removed,
1316         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1317         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1318         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1319         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1320         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1321         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1322         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1323         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1324         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1325         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1326         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1327         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1328         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1329         Opt_inode_readahead_blks, Opt_journal_ioprio,
1330         Opt_dioread_nolock, Opt_dioread_lock,
1331         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1332         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1333 };
1334
1335 static const match_table_t tokens = {
1336         {Opt_bsd_df, "bsddf"},
1337         {Opt_minix_df, "minixdf"},
1338         {Opt_grpid, "grpid"},
1339         {Opt_grpid, "bsdgroups"},
1340         {Opt_nogrpid, "nogrpid"},
1341         {Opt_nogrpid, "sysvgroups"},
1342         {Opt_resgid, "resgid=%u"},
1343         {Opt_resuid, "resuid=%u"},
1344         {Opt_sb, "sb=%u"},
1345         {Opt_err_cont, "errors=continue"},
1346         {Opt_err_panic, "errors=panic"},
1347         {Opt_err_ro, "errors=remount-ro"},
1348         {Opt_nouid32, "nouid32"},
1349         {Opt_debug, "debug"},
1350         {Opt_removed, "oldalloc"},
1351         {Opt_removed, "orlov"},
1352         {Opt_user_xattr, "user_xattr"},
1353         {Opt_nouser_xattr, "nouser_xattr"},
1354         {Opt_acl, "acl"},
1355         {Opt_noacl, "noacl"},
1356         {Opt_noload, "norecovery"},
1357         {Opt_noload, "noload"},
1358         {Opt_removed, "nobh"},
1359         {Opt_removed, "bh"},
1360         {Opt_commit, "commit=%u"},
1361         {Opt_min_batch_time, "min_batch_time=%u"},
1362         {Opt_max_batch_time, "max_batch_time=%u"},
1363         {Opt_journal_dev, "journal_dev=%u"},
1364         {Opt_journal_path, "journal_path=%s"},
1365         {Opt_journal_checksum, "journal_checksum"},
1366         {Opt_nojournal_checksum, "nojournal_checksum"},
1367         {Opt_journal_async_commit, "journal_async_commit"},
1368         {Opt_abort, "abort"},
1369         {Opt_data_journal, "data=journal"},
1370         {Opt_data_ordered, "data=ordered"},
1371         {Opt_data_writeback, "data=writeback"},
1372         {Opt_data_err_abort, "data_err=abort"},
1373         {Opt_data_err_ignore, "data_err=ignore"},
1374         {Opt_offusrjquota, "usrjquota="},
1375         {Opt_usrjquota, "usrjquota=%s"},
1376         {Opt_offgrpjquota, "grpjquota="},
1377         {Opt_grpjquota, "grpjquota=%s"},
1378         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1379         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1380         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1381         {Opt_grpquota, "grpquota"},
1382         {Opt_noquota, "noquota"},
1383         {Opt_quota, "quota"},
1384         {Opt_usrquota, "usrquota"},
1385         {Opt_prjquota, "prjquota"},
1386         {Opt_barrier, "barrier=%u"},
1387         {Opt_barrier, "barrier"},
1388         {Opt_nobarrier, "nobarrier"},
1389         {Opt_i_version, "i_version"},
1390         {Opt_dax, "dax"},
1391         {Opt_stripe, "stripe=%u"},
1392         {Opt_delalloc, "delalloc"},
1393         {Opt_lazytime, "lazytime"},
1394         {Opt_nolazytime, "nolazytime"},
1395         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1396         {Opt_nodelalloc, "nodelalloc"},
1397         {Opt_removed, "mblk_io_submit"},
1398         {Opt_removed, "nomblk_io_submit"},
1399         {Opt_block_validity, "block_validity"},
1400         {Opt_noblock_validity, "noblock_validity"},
1401         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1402         {Opt_journal_ioprio, "journal_ioprio=%u"},
1403         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1404         {Opt_auto_da_alloc, "auto_da_alloc"},
1405         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1406         {Opt_dioread_nolock, "dioread_nolock"},
1407         {Opt_dioread_lock, "dioread_lock"},
1408         {Opt_discard, "discard"},
1409         {Opt_nodiscard, "nodiscard"},
1410         {Opt_init_itable, "init_itable=%u"},
1411         {Opt_init_itable, "init_itable"},
1412         {Opt_noinit_itable, "noinit_itable"},
1413         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1414         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1415         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1416         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1417         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1418         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1419         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1420         {Opt_err, NULL},
1421 };
1422
1423 static ext4_fsblk_t get_sb_block(void **data)
1424 {
1425         ext4_fsblk_t    sb_block;
1426         char            *options = (char *) *data;
1427
1428         if (!options || strncmp(options, "sb=", 3) != 0)
1429                 return 1;       /* Default location */
1430
1431         options += 3;
1432         /* TODO: use simple_strtoll with >32bit ext4 */
1433         sb_block = simple_strtoul(options, &options, 0);
1434         if (*options && *options != ',') {
1435                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1436                        (char *) *data);
1437                 return 1;
1438         }
1439         if (*options == ',')
1440                 options++;
1441         *data = (void *) options;
1442
1443         return sb_block;
1444 }
1445
1446 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1447 static const char deprecated_msg[] =
1448         "Mount option \"%s\" will be removed by %s\n"
1449         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1450
1451 #ifdef CONFIG_QUOTA
1452 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1453 {
1454         struct ext4_sb_info *sbi = EXT4_SB(sb);
1455         char *qname;
1456         int ret = -1;
1457
1458         if (sb_any_quota_loaded(sb) &&
1459                 !sbi->s_qf_names[qtype]) {
1460                 ext4_msg(sb, KERN_ERR,
1461                         "Cannot change journaled "
1462                         "quota options when quota turned on");
1463                 return -1;
1464         }
1465         if (ext4_has_feature_quota(sb)) {
1466                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1467                          "ignored when QUOTA feature is enabled");
1468                 return 1;
1469         }
1470         qname = match_strdup(args);
1471         if (!qname) {
1472                 ext4_msg(sb, KERN_ERR,
1473                         "Not enough memory for storing quotafile name");
1474                 return -1;
1475         }
1476         if (sbi->s_qf_names[qtype]) {
1477                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1478                         ret = 1;
1479                 else
1480                         ext4_msg(sb, KERN_ERR,
1481                                  "%s quota file already specified",
1482                                  QTYPE2NAME(qtype));
1483                 goto errout;
1484         }
1485         if (strchr(qname, '/')) {
1486                 ext4_msg(sb, KERN_ERR,
1487                         "quotafile must be on filesystem root");
1488                 goto errout;
1489         }
1490         sbi->s_qf_names[qtype] = qname;
1491         set_opt(sb, QUOTA);
1492         return 1;
1493 errout:
1494         kfree(qname);
1495         return ret;
1496 }
1497
1498 static int clear_qf_name(struct super_block *sb, int qtype)
1499 {
1500
1501         struct ext4_sb_info *sbi = EXT4_SB(sb);
1502
1503         if (sb_any_quota_loaded(sb) &&
1504                 sbi->s_qf_names[qtype]) {
1505                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1506                         " when quota turned on");
1507                 return -1;
1508         }
1509         kfree(sbi->s_qf_names[qtype]);
1510         sbi->s_qf_names[qtype] = NULL;
1511         return 1;
1512 }
1513 #endif
1514
1515 #define MOPT_SET        0x0001
1516 #define MOPT_CLEAR      0x0002
1517 #define MOPT_NOSUPPORT  0x0004
1518 #define MOPT_EXPLICIT   0x0008
1519 #define MOPT_CLEAR_ERR  0x0010
1520 #define MOPT_GTE0       0x0020
1521 #ifdef CONFIG_QUOTA
1522 #define MOPT_Q          0
1523 #define MOPT_QFMT       0x0040
1524 #else
1525 #define MOPT_Q          MOPT_NOSUPPORT
1526 #define MOPT_QFMT       MOPT_NOSUPPORT
1527 #endif
1528 #define MOPT_DATAJ      0x0080
1529 #define MOPT_NO_EXT2    0x0100
1530 #define MOPT_NO_EXT3    0x0200
1531 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1532 #define MOPT_STRING     0x0400
1533
1534 static const struct mount_opts {
1535         int     token;
1536         int     mount_opt;
1537         int     flags;
1538 } ext4_mount_opts[] = {
1539         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1540         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1541         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1542         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1543         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1544         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1545         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1546          MOPT_EXT4_ONLY | MOPT_SET},
1547         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1548          MOPT_EXT4_ONLY | MOPT_CLEAR},
1549         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1550         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1551         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1552          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1553         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1554          MOPT_EXT4_ONLY | MOPT_CLEAR},
1555         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1556          MOPT_EXT4_ONLY | MOPT_CLEAR},
1557         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1558          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1559         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1560                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1561          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1562         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1563         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1564         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1565         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1566         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1567          MOPT_NO_EXT2},
1568         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1569          MOPT_NO_EXT2},
1570         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1571         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1572         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1573         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1574         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1575         {Opt_commit, 0, MOPT_GTE0},
1576         {Opt_max_batch_time, 0, MOPT_GTE0},
1577         {Opt_min_batch_time, 0, MOPT_GTE0},
1578         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1579         {Opt_init_itable, 0, MOPT_GTE0},
1580         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1581         {Opt_stripe, 0, MOPT_GTE0},
1582         {Opt_resuid, 0, MOPT_GTE0},
1583         {Opt_resgid, 0, MOPT_GTE0},
1584         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1585         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1586         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1587         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1588         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1589         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1590          MOPT_NO_EXT2 | MOPT_DATAJ},
1591         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1592         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1593 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1594         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1595         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1596 #else
1597         {Opt_acl, 0, MOPT_NOSUPPORT},
1598         {Opt_noacl, 0, MOPT_NOSUPPORT},
1599 #endif
1600         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1601         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1602         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1603         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1604         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1605                                                         MOPT_SET | MOPT_Q},
1606         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1607                                                         MOPT_SET | MOPT_Q},
1608         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1609                                                         MOPT_SET | MOPT_Q},
1610         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1611                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1612                                                         MOPT_CLEAR | MOPT_Q},
1613         {Opt_usrjquota, 0, MOPT_Q},
1614         {Opt_grpjquota, 0, MOPT_Q},
1615         {Opt_offusrjquota, 0, MOPT_Q},
1616         {Opt_offgrpjquota, 0, MOPT_Q},
1617         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1618         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1619         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1620         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1621         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1622         {Opt_err, 0, 0}
1623 };
1624
1625 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1626                             substring_t *args, unsigned long *journal_devnum,
1627                             unsigned int *journal_ioprio, int is_remount)
1628 {
1629         struct ext4_sb_info *sbi = EXT4_SB(sb);
1630         const struct mount_opts *m;
1631         kuid_t uid;
1632         kgid_t gid;
1633         int arg = 0;
1634
1635 #ifdef CONFIG_QUOTA
1636         if (token == Opt_usrjquota)
1637                 return set_qf_name(sb, USRQUOTA, &args[0]);
1638         else if (token == Opt_grpjquota)
1639                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1640         else if (token == Opt_offusrjquota)
1641                 return clear_qf_name(sb, USRQUOTA);
1642         else if (token == Opt_offgrpjquota)
1643                 return clear_qf_name(sb, GRPQUOTA);
1644 #endif
1645         switch (token) {
1646         case Opt_noacl:
1647         case Opt_nouser_xattr:
1648                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1649                 break;
1650         case Opt_sb:
1651                 return 1;       /* handled by get_sb_block() */
1652         case Opt_removed:
1653                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1654                 return 1;
1655         case Opt_abort:
1656                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1657                 return 1;
1658         case Opt_i_version:
1659                 sb->s_flags |= MS_I_VERSION;
1660                 return 1;
1661         case Opt_lazytime:
1662                 sb->s_flags |= MS_LAZYTIME;
1663                 return 1;
1664         case Opt_nolazytime:
1665                 sb->s_flags &= ~MS_LAZYTIME;
1666                 return 1;
1667         }
1668
1669         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1670                 if (token == m->token)
1671                         break;
1672
1673         if (m->token == Opt_err) {
1674                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1675                          "or missing value", opt);
1676                 return -1;
1677         }
1678
1679         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1680                 ext4_msg(sb, KERN_ERR,
1681                          "Mount option \"%s\" incompatible with ext2", opt);
1682                 return -1;
1683         }
1684         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1685                 ext4_msg(sb, KERN_ERR,
1686                          "Mount option \"%s\" incompatible with ext3", opt);
1687                 return -1;
1688         }
1689
1690         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1691                 return -1;
1692         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1693                 return -1;
1694         if (m->flags & MOPT_EXPLICIT) {
1695                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1696                         set_opt2(sb, EXPLICIT_DELALLOC);
1697                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1698                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1699                 } else
1700                         return -1;
1701         }
1702         if (m->flags & MOPT_CLEAR_ERR)
1703                 clear_opt(sb, ERRORS_MASK);
1704         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1705                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1706                          "options when quota turned on");
1707                 return -1;
1708         }
1709
1710         if (m->flags & MOPT_NOSUPPORT) {
1711                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1712         } else if (token == Opt_commit) {
1713                 if (arg == 0)
1714                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1715                 sbi->s_commit_interval = HZ * arg;
1716         } else if (token == Opt_debug_want_extra_isize) {
1717                 sbi->s_want_extra_isize = arg;
1718         } else if (token == Opt_max_batch_time) {
1719                 sbi->s_max_batch_time = arg;
1720         } else if (token == Opt_min_batch_time) {
1721                 sbi->s_min_batch_time = arg;
1722         } else if (token == Opt_inode_readahead_blks) {
1723                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1724                         ext4_msg(sb, KERN_ERR,
1725                                  "EXT4-fs: inode_readahead_blks must be "
1726                                  "0 or a power of 2 smaller than 2^31");
1727                         return -1;
1728                 }
1729                 sbi->s_inode_readahead_blks = arg;
1730         } else if (token == Opt_init_itable) {
1731                 set_opt(sb, INIT_INODE_TABLE);
1732                 if (!args->from)
1733                         arg = EXT4_DEF_LI_WAIT_MULT;
1734                 sbi->s_li_wait_mult = arg;
1735         } else if (token == Opt_max_dir_size_kb) {
1736                 sbi->s_max_dir_size_kb = arg;
1737         } else if (token == Opt_stripe) {
1738                 sbi->s_stripe = arg;
1739         } else if (token == Opt_resuid) {
1740                 uid = make_kuid(current_user_ns(), arg);
1741                 if (!uid_valid(uid)) {
1742                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1743                         return -1;
1744                 }
1745                 sbi->s_resuid = uid;
1746         } else if (token == Opt_resgid) {
1747                 gid = make_kgid(current_user_ns(), arg);
1748                 if (!gid_valid(gid)) {
1749                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1750                         return -1;
1751                 }
1752                 sbi->s_resgid = gid;
1753         } else if (token == Opt_journal_dev) {
1754                 if (is_remount) {
1755                         ext4_msg(sb, KERN_ERR,
1756                                  "Cannot specify journal on remount");
1757                         return -1;
1758                 }
1759                 *journal_devnum = arg;
1760         } else if (token == Opt_journal_path) {
1761                 char *journal_path;
1762                 struct inode *journal_inode;
1763                 struct path path;
1764                 int error;
1765
1766                 if (is_remount) {
1767                         ext4_msg(sb, KERN_ERR,
1768                                  "Cannot specify journal on remount");
1769                         return -1;
1770                 }
1771                 journal_path = match_strdup(&args[0]);
1772                 if (!journal_path) {
1773                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1774                                 "journal device string");
1775                         return -1;
1776                 }
1777
1778                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1779                 if (error) {
1780                         ext4_msg(sb, KERN_ERR, "error: could not find "
1781                                 "journal device path: error %d", error);
1782                         kfree(journal_path);
1783                         return -1;
1784                 }
1785
1786                 journal_inode = d_inode(path.dentry);
1787                 if (!S_ISBLK(journal_inode->i_mode)) {
1788                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1789                                 "is not a block device", journal_path);
1790                         path_put(&path);
1791                         kfree(journal_path);
1792                         return -1;
1793                 }
1794
1795                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1796                 path_put(&path);
1797                 kfree(journal_path);
1798         } else if (token == Opt_journal_ioprio) {
1799                 if (arg > 7) {
1800                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1801                                  " (must be 0-7)");
1802                         return -1;
1803                 }
1804                 *journal_ioprio =
1805                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1806         } else if (token == Opt_test_dummy_encryption) {
1807 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1808                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1809                 ext4_msg(sb, KERN_WARNING,
1810                          "Test dummy encryption mode enabled");
1811 #else
1812                 ext4_msg(sb, KERN_WARNING,
1813                          "Test dummy encryption mount option ignored");
1814 #endif
1815         } else if (m->flags & MOPT_DATAJ) {
1816                 if (is_remount) {
1817                         if (!sbi->s_journal)
1818                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1819                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1820                                 ext4_msg(sb, KERN_ERR,
1821                                          "Cannot change data mode on remount");
1822                                 return -1;
1823                         }
1824                 } else {
1825                         clear_opt(sb, DATA_FLAGS);
1826                         sbi->s_mount_opt |= m->mount_opt;
1827                 }
1828 #ifdef CONFIG_QUOTA
1829         } else if (m->flags & MOPT_QFMT) {
1830                 if (sb_any_quota_loaded(sb) &&
1831                     sbi->s_jquota_fmt != m->mount_opt) {
1832                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1833                                  "quota options when quota turned on");
1834                         return -1;
1835                 }
1836                 if (ext4_has_feature_quota(sb)) {
1837                         ext4_msg(sb, KERN_INFO,
1838                                  "Quota format mount options ignored "
1839                                  "when QUOTA feature is enabled");
1840                         return 1;
1841                 }
1842                 sbi->s_jquota_fmt = m->mount_opt;
1843 #endif
1844         } else if (token == Opt_dax) {
1845 #ifdef CONFIG_FS_DAX
1846                 ext4_msg(sb, KERN_WARNING,
1847                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1848                         sbi->s_mount_opt |= m->mount_opt;
1849 #else
1850                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1851                 return -1;
1852 #endif
1853         } else if (token == Opt_data_err_abort) {
1854                 sbi->s_mount_opt |= m->mount_opt;
1855         } else if (token == Opt_data_err_ignore) {
1856                 sbi->s_mount_opt &= ~m->mount_opt;
1857         } else {
1858                 if (!args->from)
1859                         arg = 1;
1860                 if (m->flags & MOPT_CLEAR)
1861                         arg = !arg;
1862                 else if (unlikely(!(m->flags & MOPT_SET))) {
1863                         ext4_msg(sb, KERN_WARNING,
1864                                  "buggy handling of option %s", opt);
1865                         WARN_ON(1);
1866                         return -1;
1867                 }
1868                 if (arg != 0)
1869                         sbi->s_mount_opt |= m->mount_opt;
1870                 else
1871                         sbi->s_mount_opt &= ~m->mount_opt;
1872         }
1873         return 1;
1874 }
1875
1876 static int parse_options(char *options, struct super_block *sb,
1877                          unsigned long *journal_devnum,
1878                          unsigned int *journal_ioprio,
1879                          int is_remount)
1880 {
1881         struct ext4_sb_info *sbi = EXT4_SB(sb);
1882         char *p;
1883         substring_t args[MAX_OPT_ARGS];
1884         int token;
1885
1886         if (!options)
1887                 return 1;
1888
1889         while ((p = strsep(&options, ",")) != NULL) {
1890                 if (!*p)
1891                         continue;
1892                 /*
1893                  * Initialize args struct so we know whether arg was
1894                  * found; some options take optional arguments.
1895                  */
1896                 args[0].to = args[0].from = NULL;
1897                 token = match_token(p, tokens, args);
1898                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1899                                      journal_ioprio, is_remount) < 0)
1900                         return 0;
1901         }
1902 #ifdef CONFIG_QUOTA
1903         /*
1904          * We do the test below only for project quotas. 'usrquota' and
1905          * 'grpquota' mount options are allowed even without quota feature
1906          * to support legacy quotas in quota files.
1907          */
1908         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1909                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1910                          "Cannot enable project quota enforcement.");
1911                 return 0;
1912         }
1913         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1914                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1915                         clear_opt(sb, USRQUOTA);
1916
1917                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1918                         clear_opt(sb, GRPQUOTA);
1919
1920                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1921                         ext4_msg(sb, KERN_ERR, "old and new quota "
1922                                         "format mixing");
1923                         return 0;
1924                 }
1925
1926                 if (!sbi->s_jquota_fmt) {
1927                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1928                                         "not specified");
1929                         return 0;
1930                 }
1931         }
1932 #endif
1933         if (test_opt(sb, DIOREAD_NOLOCK)) {
1934                 int blocksize =
1935                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1936
1937                 if (blocksize < PAGE_SIZE) {
1938                         ext4_msg(sb, KERN_ERR, "can't mount with "
1939                                  "dioread_nolock if block size != PAGE_SIZE");
1940                         return 0;
1941                 }
1942         }
1943         return 1;
1944 }
1945
1946 static inline void ext4_show_quota_options(struct seq_file *seq,
1947                                            struct super_block *sb)
1948 {
1949 #if defined(CONFIG_QUOTA)
1950         struct ext4_sb_info *sbi = EXT4_SB(sb);
1951
1952         if (sbi->s_jquota_fmt) {
1953                 char *fmtname = "";
1954
1955                 switch (sbi->s_jquota_fmt) {
1956                 case QFMT_VFS_OLD:
1957                         fmtname = "vfsold";
1958                         break;
1959                 case QFMT_VFS_V0:
1960                         fmtname = "vfsv0";
1961                         break;
1962                 case QFMT_VFS_V1:
1963                         fmtname = "vfsv1";
1964                         break;
1965                 }
1966                 seq_printf(seq, ",jqfmt=%s", fmtname);
1967         }
1968
1969         if (sbi->s_qf_names[USRQUOTA])
1970                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1971
1972         if (sbi->s_qf_names[GRPQUOTA])
1973                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1974 #endif
1975 }
1976
1977 static const char *token2str(int token)
1978 {
1979         const struct match_token *t;
1980
1981         for (t = tokens; t->token != Opt_err; t++)
1982                 if (t->token == token && !strchr(t->pattern, '='))
1983                         break;
1984         return t->pattern;
1985 }
1986
1987 /*
1988  * Show an option if
1989  *  - it's set to a non-default value OR
1990  *  - if the per-sb default is different from the global default
1991  */
1992 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1993                               int nodefs)
1994 {
1995         struct ext4_sb_info *sbi = EXT4_SB(sb);
1996         struct ext4_super_block *es = sbi->s_es;
1997         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1998         const struct mount_opts *m;
1999         char sep = nodefs ? '\n' : ',';
2000
2001 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2002 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2003
2004         if (sbi->s_sb_block != 1)
2005                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2006
2007         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2008                 int want_set = m->flags & MOPT_SET;
2009                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2010                     (m->flags & MOPT_CLEAR_ERR))
2011                         continue;
2012                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2013                         continue; /* skip if same as the default */
2014                 if ((want_set &&
2015                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2016                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2017                         continue; /* select Opt_noFoo vs Opt_Foo */
2018                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2019         }
2020
2021         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2022             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2023                 SEQ_OPTS_PRINT("resuid=%u",
2024                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2025         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2026             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2027                 SEQ_OPTS_PRINT("resgid=%u",
2028                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2029         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2030         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2031                 SEQ_OPTS_PUTS("errors=remount-ro");
2032         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2033                 SEQ_OPTS_PUTS("errors=continue");
2034         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2035                 SEQ_OPTS_PUTS("errors=panic");
2036         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2037                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2038         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2039                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2040         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2041                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2042         if (sb->s_flags & MS_I_VERSION)
2043                 SEQ_OPTS_PUTS("i_version");
2044         if (nodefs || sbi->s_stripe)
2045                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2046         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2047                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2048                         SEQ_OPTS_PUTS("data=journal");
2049                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2050                         SEQ_OPTS_PUTS("data=ordered");
2051                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2052                         SEQ_OPTS_PUTS("data=writeback");
2053         }
2054         if (nodefs ||
2055             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2056                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2057                                sbi->s_inode_readahead_blks);
2058
2059         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2060                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2061                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2062         if (nodefs || sbi->s_max_dir_size_kb)
2063                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2064         if (test_opt(sb, DATA_ERR_ABORT))
2065                 SEQ_OPTS_PUTS("data_err=abort");
2066
2067         ext4_show_quota_options(seq, sb);
2068         return 0;
2069 }
2070
2071 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2072 {
2073         return _ext4_show_options(seq, root->d_sb, 0);
2074 }
2075
2076 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2077 {
2078         struct super_block *sb = seq->private;
2079         int rc;
2080
2081         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
2082         rc = _ext4_show_options(seq, sb, 1);
2083         seq_puts(seq, "\n");
2084         return rc;
2085 }
2086
2087 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2088                             int read_only)
2089 {
2090         struct ext4_sb_info *sbi = EXT4_SB(sb);
2091         int res = 0;
2092
2093         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2094                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2095                          "forcing read-only mode");
2096                 res = MS_RDONLY;
2097         }
2098         if (read_only)
2099                 goto done;
2100         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2101                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2102                          "running e2fsck is recommended");
2103         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2104                 ext4_msg(sb, KERN_WARNING,
2105                          "warning: mounting fs with errors, "
2106                          "running e2fsck is recommended");
2107         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2108                  le16_to_cpu(es->s_mnt_count) >=
2109                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2110                 ext4_msg(sb, KERN_WARNING,
2111                          "warning: maximal mount count reached, "
2112                          "running e2fsck is recommended");
2113         else if (le32_to_cpu(es->s_checkinterval) &&
2114                 (le32_to_cpu(es->s_lastcheck) +
2115                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2116                 ext4_msg(sb, KERN_WARNING,
2117                          "warning: checktime reached, "
2118                          "running e2fsck is recommended");
2119         if (!sbi->s_journal)
2120                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2121         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2122                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2123         le16_add_cpu(&es->s_mnt_count, 1);
2124         es->s_mtime = cpu_to_le32(get_seconds());
2125         ext4_update_dynamic_rev(sb);
2126         if (sbi->s_journal)
2127                 ext4_set_feature_journal_needs_recovery(sb);
2128
2129         ext4_commit_super(sb, 1);
2130 done:
2131         if (test_opt(sb, DEBUG))
2132                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2133                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2134                         sb->s_blocksize,
2135                         sbi->s_groups_count,
2136                         EXT4_BLOCKS_PER_GROUP(sb),
2137                         EXT4_INODES_PER_GROUP(sb),
2138                         sbi->s_mount_opt, sbi->s_mount_opt2);
2139
2140         cleancache_init_fs(sb);
2141         return res;
2142 }
2143
2144 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2145 {
2146         struct ext4_sb_info *sbi = EXT4_SB(sb);
2147         struct flex_groups *new_groups;
2148         int size;
2149
2150         if (!sbi->s_log_groups_per_flex)
2151                 return 0;
2152
2153         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2154         if (size <= sbi->s_flex_groups_allocated)
2155                 return 0;
2156
2157         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2158         new_groups = kvzalloc(size, GFP_KERNEL);
2159         if (!new_groups) {
2160                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2161                          size / (int) sizeof(struct flex_groups));
2162                 return -ENOMEM;
2163         }
2164
2165         if (sbi->s_flex_groups) {
2166                 memcpy(new_groups, sbi->s_flex_groups,
2167                        (sbi->s_flex_groups_allocated *
2168                         sizeof(struct flex_groups)));
2169                 kvfree(sbi->s_flex_groups);
2170         }
2171         sbi->s_flex_groups = new_groups;
2172         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2173         return 0;
2174 }
2175
2176 static int ext4_fill_flex_info(struct super_block *sb)
2177 {
2178         struct ext4_sb_info *sbi = EXT4_SB(sb);
2179         struct ext4_group_desc *gdp = NULL;
2180         ext4_group_t flex_group;
2181         int i, err;
2182
2183         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2184         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2185                 sbi->s_log_groups_per_flex = 0;
2186                 return 1;
2187         }
2188
2189         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2190         if (err)
2191                 goto failed;
2192
2193         for (i = 0; i < sbi->s_groups_count; i++) {
2194                 gdp = ext4_get_group_desc(sb, i, NULL);
2195
2196                 flex_group = ext4_flex_group(sbi, i);
2197                 atomic_add(ext4_free_inodes_count(sb, gdp),
2198                            &sbi->s_flex_groups[flex_group].free_inodes);
2199                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2200                              &sbi->s_flex_groups[flex_group].free_clusters);
2201                 atomic_add(ext4_used_dirs_count(sb, gdp),
2202                            &sbi->s_flex_groups[flex_group].used_dirs);
2203         }
2204
2205         return 1;
2206 failed:
2207         return 0;
2208 }
2209
2210 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2211                                    struct ext4_group_desc *gdp)
2212 {
2213         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2214         __u16 crc = 0;
2215         __le32 le_group = cpu_to_le32(block_group);
2216         struct ext4_sb_info *sbi = EXT4_SB(sb);
2217
2218         if (ext4_has_metadata_csum(sbi->s_sb)) {
2219                 /* Use new metadata_csum algorithm */
2220                 __u32 csum32;
2221                 __u16 dummy_csum = 0;
2222
2223                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2224                                      sizeof(le_group));
2225                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2226                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2227                                      sizeof(dummy_csum));
2228                 offset += sizeof(dummy_csum);
2229                 if (offset < sbi->s_desc_size)
2230                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2231                                              sbi->s_desc_size - offset);
2232
2233                 crc = csum32 & 0xFFFF;
2234                 goto out;
2235         }
2236
2237         /* old crc16 code */
2238         if (!ext4_has_feature_gdt_csum(sb))
2239                 return 0;
2240
2241         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2242         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2243         crc = crc16(crc, (__u8 *)gdp, offset);
2244         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2245         /* for checksum of struct ext4_group_desc do the rest...*/
2246         if (ext4_has_feature_64bit(sb) &&
2247             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2248                 crc = crc16(crc, (__u8 *)gdp + offset,
2249                             le16_to_cpu(sbi->s_es->s_desc_size) -
2250                                 offset);
2251
2252 out:
2253         return cpu_to_le16(crc);
2254 }
2255
2256 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2257                                 struct ext4_group_desc *gdp)
2258 {
2259         if (ext4_has_group_desc_csum(sb) &&
2260             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2261                 return 0;
2262
2263         return 1;
2264 }
2265
2266 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2267                               struct ext4_group_desc *gdp)
2268 {
2269         if (!ext4_has_group_desc_csum(sb))
2270                 return;
2271         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2272 }
2273
2274 /* Called at mount-time, super-block is locked */
2275 static int ext4_check_descriptors(struct super_block *sb,
2276                                   ext4_fsblk_t sb_block,
2277                                   ext4_group_t *first_not_zeroed)
2278 {
2279         struct ext4_sb_info *sbi = EXT4_SB(sb);
2280         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2281         ext4_fsblk_t last_block;
2282         ext4_fsblk_t block_bitmap;
2283         ext4_fsblk_t inode_bitmap;
2284         ext4_fsblk_t inode_table;
2285         int flexbg_flag = 0;
2286         ext4_group_t i, grp = sbi->s_groups_count;
2287
2288         if (ext4_has_feature_flex_bg(sb))
2289                 flexbg_flag = 1;
2290
2291         ext4_debug("Checking group descriptors");
2292
2293         for (i = 0; i < sbi->s_groups_count; i++) {
2294                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2295
2296                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2297                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2298                 else
2299                         last_block = first_block +
2300                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2301
2302                 if ((grp == sbi->s_groups_count) &&
2303                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2304                         grp = i;
2305
2306                 block_bitmap = ext4_block_bitmap(sb, gdp);
2307                 if (block_bitmap == sb_block) {
2308                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2309                                  "Block bitmap for group %u overlaps "
2310                                  "superblock", i);
2311                 }
2312                 if (block_bitmap < first_block || block_bitmap > last_block) {
2313                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2314                                "Block bitmap for group %u not in group "
2315                                "(block %llu)!", i, block_bitmap);
2316                         return 0;
2317                 }
2318                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2319                 if (inode_bitmap == sb_block) {
2320                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2321                                  "Inode bitmap for group %u overlaps "
2322                                  "superblock", i);
2323                 }
2324                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2325                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2326                                "Inode bitmap for group %u not in group "
2327                                "(block %llu)!", i, inode_bitmap);
2328                         return 0;
2329                 }
2330                 inode_table = ext4_inode_table(sb, gdp);
2331                 if (inode_table == sb_block) {
2332                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2333                                  "Inode table for group %u overlaps "
2334                                  "superblock", i);
2335                 }
2336                 if (inode_table < first_block ||
2337                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2338                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2339                                "Inode table for group %u not in group "
2340                                "(block %llu)!", i, inode_table);
2341                         return 0;
2342                 }
2343                 ext4_lock_group(sb, i);
2344                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2345                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2346                                  "Checksum for group %u failed (%u!=%u)",
2347                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2348                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2349                         if (!(sb->s_flags & MS_RDONLY)) {
2350                                 ext4_unlock_group(sb, i);
2351                                 return 0;
2352                         }
2353                 }
2354                 ext4_unlock_group(sb, i);
2355                 if (!flexbg_flag)
2356                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2357         }
2358         if (NULL != first_not_zeroed)
2359                 *first_not_zeroed = grp;
2360         return 1;
2361 }
2362
2363 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2364  * the superblock) which were deleted from all directories, but held open by
2365  * a process at the time of a crash.  We walk the list and try to delete these
2366  * inodes at recovery time (only with a read-write filesystem).
2367  *
2368  * In order to keep the orphan inode chain consistent during traversal (in
2369  * case of crash during recovery), we link each inode into the superblock
2370  * orphan list_head and handle it the same way as an inode deletion during
2371  * normal operation (which journals the operations for us).
2372  *
2373  * We only do an iget() and an iput() on each inode, which is very safe if we
2374  * accidentally point at an in-use or already deleted inode.  The worst that
2375  * can happen in this case is that we get a "bit already cleared" message from
2376  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2377  * e2fsck was run on this filesystem, and it must have already done the orphan
2378  * inode cleanup for us, so we can safely abort without any further action.
2379  */
2380 static void ext4_orphan_cleanup(struct super_block *sb,
2381                                 struct ext4_super_block *es)
2382 {
2383         unsigned int s_flags = sb->s_flags;
2384         int ret, nr_orphans = 0, nr_truncates = 0;
2385 #ifdef CONFIG_QUOTA
2386         int i;
2387 #endif
2388         if (!es->s_last_orphan) {
2389                 jbd_debug(4, "no orphan inodes to clean up\n");
2390                 return;
2391         }
2392
2393         if (bdev_read_only(sb->s_bdev)) {
2394                 ext4_msg(sb, KERN_ERR, "write access "
2395                         "unavailable, skipping orphan cleanup");
2396                 return;
2397         }
2398
2399         /* Check if feature set would not allow a r/w mount */
2400         if (!ext4_feature_set_ok(sb, 0)) {
2401                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2402                          "unknown ROCOMPAT features");
2403                 return;
2404         }
2405
2406         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2407                 /* don't clear list on RO mount w/ errors */
2408                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2409                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2410                                   "clearing orphan list.\n");
2411                         es->s_last_orphan = 0;
2412                 }
2413                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2414                 return;
2415         }
2416
2417         if (s_flags & MS_RDONLY) {
2418                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2419                 sb->s_flags &= ~MS_RDONLY;
2420         }
2421 #ifdef CONFIG_QUOTA
2422         /* Needed for iput() to work correctly and not trash data */
2423         sb->s_flags |= MS_ACTIVE;
2424         /* Turn on quotas so that they are updated correctly */
2425         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2426                 if (EXT4_SB(sb)->s_qf_names[i]) {
2427                         int ret = ext4_quota_on_mount(sb, i);
2428                         if (ret < 0)
2429                                 ext4_msg(sb, KERN_ERR,
2430                                         "Cannot turn on journaled "
2431                                         "quota: error %d", ret);
2432                 }
2433         }
2434 #endif
2435
2436         while (es->s_last_orphan) {
2437                 struct inode *inode;
2438
2439                 /*
2440                  * We may have encountered an error during cleanup; if
2441                  * so, skip the rest.
2442                  */
2443                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2444                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2445                         es->s_last_orphan = 0;
2446                         break;
2447                 }
2448
2449                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2450                 if (IS_ERR(inode)) {
2451                         es->s_last_orphan = 0;
2452                         break;
2453                 }
2454
2455                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2456                 dquot_initialize(inode);
2457                 if (inode->i_nlink) {
2458                         if (test_opt(sb, DEBUG))
2459                                 ext4_msg(sb, KERN_DEBUG,
2460                                         "%s: truncating inode %lu to %lld bytes",
2461                                         __func__, inode->i_ino, inode->i_size);
2462                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2463                                   inode->i_ino, inode->i_size);
2464                         inode_lock(inode);
2465                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2466                         ret = ext4_truncate(inode);
2467                         if (ret)
2468                                 ext4_std_error(inode->i_sb, ret);
2469                         inode_unlock(inode);
2470                         nr_truncates++;
2471                 } else {
2472                         if (test_opt(sb, DEBUG))
2473                                 ext4_msg(sb, KERN_DEBUG,
2474                                         "%s: deleting unreferenced inode %lu",
2475                                         __func__, inode->i_ino);
2476                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2477                                   inode->i_ino);
2478                         nr_orphans++;
2479                 }
2480                 iput(inode);  /* The delete magic happens here! */
2481         }
2482
2483 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2484
2485         if (nr_orphans)
2486                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2487                        PLURAL(nr_orphans));
2488         if (nr_truncates)
2489                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2490                        PLURAL(nr_truncates));
2491 #ifdef CONFIG_QUOTA
2492         /* Turn quotas off */
2493         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2494                 if (sb_dqopt(sb)->files[i])
2495                         dquot_quota_off(sb, i);
2496         }
2497 #endif
2498         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2499 }
2500
2501 /*
2502  * Maximal extent format file size.
2503  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2504  * extent format containers, within a sector_t, and within i_blocks
2505  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2506  * so that won't be a limiting factor.
2507  *
2508  * However there is other limiting factor. We do store extents in the form
2509  * of starting block and length, hence the resulting length of the extent
2510  * covering maximum file size must fit into on-disk format containers as
2511  * well. Given that length is always by 1 unit bigger than max unit (because
2512  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2513  *
2514  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2515  */
2516 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2517 {
2518         loff_t res;
2519         loff_t upper_limit = MAX_LFS_FILESIZE;
2520
2521         /* small i_blocks in vfs inode? */
2522         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2523                 /*
2524                  * CONFIG_LBDAF is not enabled implies the inode
2525                  * i_block represent total blocks in 512 bytes
2526                  * 32 == size of vfs inode i_blocks * 8
2527                  */
2528                 upper_limit = (1LL << 32) - 1;
2529
2530                 /* total blocks in file system block size */
2531                 upper_limit >>= (blkbits - 9);
2532                 upper_limit <<= blkbits;
2533         }
2534
2535         /*
2536          * 32-bit extent-start container, ee_block. We lower the maxbytes
2537          * by one fs block, so ee_len can cover the extent of maximum file
2538          * size
2539          */
2540         res = (1LL << 32) - 1;
2541         res <<= blkbits;
2542
2543         /* Sanity check against vm- & vfs- imposed limits */
2544         if (res > upper_limit)
2545                 res = upper_limit;
2546
2547         return res;
2548 }
2549
2550 /*
2551  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2552  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2553  * We need to be 1 filesystem block less than the 2^48 sector limit.
2554  */
2555 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2556 {
2557         loff_t res = EXT4_NDIR_BLOCKS;
2558         int meta_blocks;
2559         loff_t upper_limit;
2560         /* This is calculated to be the largest file size for a dense, block
2561          * mapped file such that the file's total number of 512-byte sectors,
2562          * including data and all indirect blocks, does not exceed (2^48 - 1).
2563          *
2564          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2565          * number of 512-byte sectors of the file.
2566          */
2567
2568         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2569                 /*
2570                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2571                  * the inode i_block field represents total file blocks in
2572                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2573                  */
2574                 upper_limit = (1LL << 32) - 1;
2575
2576                 /* total blocks in file system block size */
2577                 upper_limit >>= (bits - 9);
2578
2579         } else {
2580                 /*
2581                  * We use 48 bit ext4_inode i_blocks
2582                  * With EXT4_HUGE_FILE_FL set the i_blocks
2583                  * represent total number of blocks in
2584                  * file system block size
2585                  */
2586                 upper_limit = (1LL << 48) - 1;
2587
2588         }
2589
2590         /* indirect blocks */
2591         meta_blocks = 1;
2592         /* double indirect blocks */
2593         meta_blocks += 1 + (1LL << (bits-2));
2594         /* tripple indirect blocks */
2595         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2596
2597         upper_limit -= meta_blocks;
2598         upper_limit <<= bits;
2599
2600         res += 1LL << (bits-2);
2601         res += 1LL << (2*(bits-2));
2602         res += 1LL << (3*(bits-2));
2603         res <<= bits;
2604         if (res > upper_limit)
2605                 res = upper_limit;
2606
2607         if (res > MAX_LFS_FILESIZE)
2608                 res = MAX_LFS_FILESIZE;
2609
2610         return res;
2611 }
2612
2613 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2614                                    ext4_fsblk_t logical_sb_block, int nr)
2615 {
2616         struct ext4_sb_info *sbi = EXT4_SB(sb);
2617         ext4_group_t bg, first_meta_bg;
2618         int has_super = 0;
2619
2620         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2621
2622         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2623                 return logical_sb_block + nr + 1;
2624         bg = sbi->s_desc_per_block * nr;
2625         if (ext4_bg_has_super(sb, bg))
2626                 has_super = 1;
2627
2628         /*
2629          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2630          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2631          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2632          * compensate.
2633          */
2634         if (sb->s_blocksize == 1024 && nr == 0 &&
2635             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2636                 has_super++;
2637
2638         return (has_super + ext4_group_first_block_no(sb, bg));
2639 }
2640
2641 /**
2642  * ext4_get_stripe_size: Get the stripe size.
2643  * @sbi: In memory super block info
2644  *
2645  * If we have specified it via mount option, then
2646  * use the mount option value. If the value specified at mount time is
2647  * greater than the blocks per group use the super block value.
2648  * If the super block value is greater than blocks per group return 0.
2649  * Allocator needs it be less than blocks per group.
2650  *
2651  */
2652 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2653 {
2654         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2655         unsigned long stripe_width =
2656                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2657         int ret;
2658
2659         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2660                 ret = sbi->s_stripe;
2661         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2662                 ret = stripe_width;
2663         else if (stride && stride <= sbi->s_blocks_per_group)
2664                 ret = stride;
2665         else
2666                 ret = 0;
2667
2668         /*
2669          * If the stripe width is 1, this makes no sense and
2670          * we set it to 0 to turn off stripe handling code.
2671          */
2672         if (ret <= 1)
2673                 ret = 0;
2674
2675         return ret;
2676 }
2677
2678 /*
2679  * Check whether this filesystem can be mounted based on
2680  * the features present and the RDONLY/RDWR mount requested.
2681  * Returns 1 if this filesystem can be mounted as requested,
2682  * 0 if it cannot be.
2683  */
2684 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2685 {
2686         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2687                 ext4_msg(sb, KERN_ERR,
2688                         "Couldn't mount because of "
2689                         "unsupported optional features (%x)",
2690                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2691                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2692                 return 0;
2693         }
2694
2695         if (readonly)
2696                 return 1;
2697
2698         if (ext4_has_feature_readonly(sb)) {
2699                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2700                 sb->s_flags |= MS_RDONLY;
2701                 return 1;
2702         }
2703
2704         /* Check that feature set is OK for a read-write mount */
2705         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2706                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2707                          "unsupported optional features (%x)",
2708                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2709                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2710                 return 0;
2711         }
2712         /*
2713          * Large file size enabled file system can only be mounted
2714          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2715          */
2716         if (ext4_has_feature_huge_file(sb)) {
2717                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2718                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2719                                  "cannot be mounted RDWR without "
2720                                  "CONFIG_LBDAF");
2721                         return 0;
2722                 }
2723         }
2724         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2725                 ext4_msg(sb, KERN_ERR,
2726                          "Can't support bigalloc feature without "
2727                          "extents feature\n");
2728                 return 0;
2729         }
2730
2731 #ifndef CONFIG_QUOTA
2732         if (ext4_has_feature_quota(sb) && !readonly) {
2733                 ext4_msg(sb, KERN_ERR,
2734                          "Filesystem with quota feature cannot be mounted RDWR "
2735                          "without CONFIG_QUOTA");
2736                 return 0;
2737         }
2738         if (ext4_has_feature_project(sb) && !readonly) {
2739                 ext4_msg(sb, KERN_ERR,
2740                          "Filesystem with project quota feature cannot be mounted RDWR "
2741                          "without CONFIG_QUOTA");
2742                 return 0;
2743         }
2744 #endif  /* CONFIG_QUOTA */
2745         return 1;
2746 }
2747
2748 /*
2749  * This function is called once a day if we have errors logged
2750  * on the file system
2751  */
2752 static void print_daily_error_info(unsigned long arg)
2753 {
2754         struct super_block *sb = (struct super_block *) arg;
2755         struct ext4_sb_info *sbi;
2756         struct ext4_super_block *es;
2757
2758         sbi = EXT4_SB(sb);
2759         es = sbi->s_es;
2760
2761         if (es->s_error_count)
2762                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2763                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2764                          le32_to_cpu(es->s_error_count));
2765         if (es->s_first_error_time) {
2766                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2767                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2768                        (int) sizeof(es->s_first_error_func),
2769                        es->s_first_error_func,
2770                        le32_to_cpu(es->s_first_error_line));
2771                 if (es->s_first_error_ino)
2772                         printk(KERN_CONT ": inode %u",
2773                                le32_to_cpu(es->s_first_error_ino));
2774                 if (es->s_first_error_block)
2775                         printk(KERN_CONT ": block %llu", (unsigned long long)
2776                                le64_to_cpu(es->s_first_error_block));
2777                 printk(KERN_CONT "\n");
2778         }
2779         if (es->s_last_error_time) {
2780                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2781                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2782                        (int) sizeof(es->s_last_error_func),
2783                        es->s_last_error_func,
2784                        le32_to_cpu(es->s_last_error_line));
2785                 if (es->s_last_error_ino)
2786                         printk(KERN_CONT ": inode %u",
2787                                le32_to_cpu(es->s_last_error_ino));
2788                 if (es->s_last_error_block)
2789                         printk(KERN_CONT ": block %llu", (unsigned long long)
2790                                le64_to_cpu(es->s_last_error_block));
2791                 printk(KERN_CONT "\n");
2792         }
2793         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2794 }
2795
2796 /* Find next suitable group and run ext4_init_inode_table */
2797 static int ext4_run_li_request(struct ext4_li_request *elr)
2798 {
2799         struct ext4_group_desc *gdp = NULL;
2800         ext4_group_t group, ngroups;
2801         struct super_block *sb;
2802         unsigned long timeout = 0;
2803         int ret = 0;
2804
2805         sb = elr->lr_super;
2806         ngroups = EXT4_SB(sb)->s_groups_count;
2807
2808         for (group = elr->lr_next_group; group < ngroups; group++) {
2809                 gdp = ext4_get_group_desc(sb, group, NULL);
2810                 if (!gdp) {
2811                         ret = 1;
2812                         break;
2813                 }
2814
2815                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2816                         break;
2817         }
2818
2819         if (group >= ngroups)
2820                 ret = 1;
2821
2822         if (!ret) {
2823                 timeout = jiffies;
2824                 ret = ext4_init_inode_table(sb, group,
2825                                             elr->lr_timeout ? 0 : 1);
2826                 if (elr->lr_timeout == 0) {
2827                         timeout = (jiffies - timeout) *
2828                                   elr->lr_sbi->s_li_wait_mult;
2829                         elr->lr_timeout = timeout;
2830                 }
2831                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2832                 elr->lr_next_group = group + 1;
2833         }
2834         return ret;
2835 }
2836
2837 /*
2838  * Remove lr_request from the list_request and free the
2839  * request structure. Should be called with li_list_mtx held
2840  */
2841 static void ext4_remove_li_request(struct ext4_li_request *elr)
2842 {
2843         struct ext4_sb_info *sbi;
2844
2845         if (!elr)
2846                 return;
2847
2848         sbi = elr->lr_sbi;
2849
2850         list_del(&elr->lr_request);
2851         sbi->s_li_request = NULL;
2852         kfree(elr);
2853 }
2854
2855 static void ext4_unregister_li_request(struct super_block *sb)
2856 {
2857         mutex_lock(&ext4_li_mtx);
2858         if (!ext4_li_info) {
2859                 mutex_unlock(&ext4_li_mtx);
2860                 return;
2861         }
2862
2863         mutex_lock(&ext4_li_info->li_list_mtx);
2864         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2865         mutex_unlock(&ext4_li_info->li_list_mtx);
2866         mutex_unlock(&ext4_li_mtx);
2867 }
2868
2869 static struct task_struct *ext4_lazyinit_task;
2870
2871 /*
2872  * This is the function where ext4lazyinit thread lives. It walks
2873  * through the request list searching for next scheduled filesystem.
2874  * When such a fs is found, run the lazy initialization request
2875  * (ext4_rn_li_request) and keep track of the time spend in this
2876  * function. Based on that time we compute next schedule time of
2877  * the request. When walking through the list is complete, compute
2878  * next waking time and put itself into sleep.
2879  */
2880 static int ext4_lazyinit_thread(void *arg)
2881 {
2882         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2883         struct list_head *pos, *n;
2884         struct ext4_li_request *elr;
2885         unsigned long next_wakeup, cur;
2886
2887         BUG_ON(NULL == eli);
2888
2889 cont_thread:
2890         while (true) {
2891                 next_wakeup = MAX_JIFFY_OFFSET;
2892
2893                 mutex_lock(&eli->li_list_mtx);
2894                 if (list_empty(&eli->li_request_list)) {
2895                         mutex_unlock(&eli->li_list_mtx);
2896                         goto exit_thread;
2897                 }
2898                 list_for_each_safe(pos, n, &eli->li_request_list) {
2899                         int err = 0;
2900                         int progress = 0;
2901                         elr = list_entry(pos, struct ext4_li_request,
2902                                          lr_request);
2903
2904                         if (time_before(jiffies, elr->lr_next_sched)) {
2905                                 if (time_before(elr->lr_next_sched, next_wakeup))
2906                                         next_wakeup = elr->lr_next_sched;
2907                                 continue;
2908                         }
2909                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2910                                 if (sb_start_write_trylock(elr->lr_super)) {
2911                                         progress = 1;
2912                                         /*
2913                                          * We hold sb->s_umount, sb can not
2914                                          * be removed from the list, it is
2915                                          * now safe to drop li_list_mtx
2916                                          */
2917                                         mutex_unlock(&eli->li_list_mtx);
2918                                         err = ext4_run_li_request(elr);
2919                                         sb_end_write(elr->lr_super);
2920                                         mutex_lock(&eli->li_list_mtx);
2921                                         n = pos->next;
2922                                 }
2923                                 up_read((&elr->lr_super->s_umount));
2924                         }
2925                         /* error, remove the lazy_init job */
2926                         if (err) {
2927                                 ext4_remove_li_request(elr);
2928                                 continue;
2929                         }
2930                         if (!progress) {
2931                                 elr->lr_next_sched = jiffies +
2932                                         (prandom_u32()
2933                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2934                         }
2935                         if (time_before(elr->lr_next_sched, next_wakeup))
2936                                 next_wakeup = elr->lr_next_sched;
2937                 }
2938                 mutex_unlock(&eli->li_list_mtx);
2939
2940                 try_to_freeze();
2941
2942                 cur = jiffies;
2943                 if ((time_after_eq(cur, next_wakeup)) ||
2944                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2945                         cond_resched();
2946                         continue;
2947                 }
2948
2949                 schedule_timeout_interruptible(next_wakeup - cur);
2950
2951                 if (kthread_should_stop()) {
2952                         ext4_clear_request_list();
2953                         goto exit_thread;
2954                 }
2955         }
2956
2957 exit_thread:
2958         /*
2959          * It looks like the request list is empty, but we need
2960          * to check it under the li_list_mtx lock, to prevent any
2961          * additions into it, and of course we should lock ext4_li_mtx
2962          * to atomically free the list and ext4_li_info, because at
2963          * this point another ext4 filesystem could be registering
2964          * new one.
2965          */
2966         mutex_lock(&ext4_li_mtx);
2967         mutex_lock(&eli->li_list_mtx);
2968         if (!list_empty(&eli->li_request_list)) {
2969                 mutex_unlock(&eli->li_list_mtx);
2970                 mutex_unlock(&ext4_li_mtx);
2971                 goto cont_thread;
2972         }
2973         mutex_unlock(&eli->li_list_mtx);
2974         kfree(ext4_li_info);
2975         ext4_li_info = NULL;
2976         mutex_unlock(&ext4_li_mtx);
2977
2978         return 0;
2979 }
2980
2981 static void ext4_clear_request_list(void)
2982 {
2983         struct list_head *pos, *n;
2984         struct ext4_li_request *elr;
2985
2986         mutex_lock(&ext4_li_info->li_list_mtx);
2987         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2988                 elr = list_entry(pos, struct ext4_li_request,
2989                                  lr_request);
2990                 ext4_remove_li_request(elr);
2991         }
2992         mutex_unlock(&ext4_li_info->li_list_mtx);
2993 }
2994
2995 static int ext4_run_lazyinit_thread(void)
2996 {
2997         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2998                                          ext4_li_info, "ext4lazyinit");
2999         if (IS_ERR(ext4_lazyinit_task)) {
3000                 int err = PTR_ERR(ext4_lazyinit_task);
3001                 ext4_clear_request_list();
3002                 kfree(ext4_li_info);
3003                 ext4_li_info = NULL;
3004                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3005                                  "initialization thread\n",
3006                                  err);
3007                 return err;
3008         }
3009         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3010         return 0;
3011 }
3012
3013 /*
3014  * Check whether it make sense to run itable init. thread or not.
3015  * If there is at least one uninitialized inode table, return
3016  * corresponding group number, else the loop goes through all
3017  * groups and return total number of groups.
3018  */
3019 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3020 {
3021         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3022         struct ext4_group_desc *gdp = NULL;
3023
3024         for (group = 0; group < ngroups; group++) {
3025                 gdp = ext4_get_group_desc(sb, group, NULL);
3026                 if (!gdp)
3027                         continue;
3028
3029                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3030                         break;
3031         }
3032
3033         return group;
3034 }
3035
3036 static int ext4_li_info_new(void)
3037 {
3038         struct ext4_lazy_init *eli = NULL;
3039
3040         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3041         if (!eli)
3042                 return -ENOMEM;
3043
3044         INIT_LIST_HEAD(&eli->li_request_list);
3045         mutex_init(&eli->li_list_mtx);
3046
3047         eli->li_state |= EXT4_LAZYINIT_QUIT;
3048
3049         ext4_li_info = eli;
3050
3051         return 0;
3052 }
3053
3054 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3055                                             ext4_group_t start)
3056 {
3057         struct ext4_sb_info *sbi = EXT4_SB(sb);
3058         struct ext4_li_request *elr;
3059
3060         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3061         if (!elr)
3062                 return NULL;
3063
3064         elr->lr_super = sb;
3065         elr->lr_sbi = sbi;
3066         elr->lr_next_group = start;
3067
3068         /*
3069          * Randomize first schedule time of the request to
3070          * spread the inode table initialization requests
3071          * better.
3072          */
3073         elr->lr_next_sched = jiffies + (prandom_u32() %
3074                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3075         return elr;
3076 }
3077
3078 int ext4_register_li_request(struct super_block *sb,
3079                              ext4_group_t first_not_zeroed)
3080 {
3081         struct ext4_sb_info *sbi = EXT4_SB(sb);
3082         struct ext4_li_request *elr = NULL;
3083         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3084         int ret = 0;
3085
3086         mutex_lock(&ext4_li_mtx);
3087         if (sbi->s_li_request != NULL) {
3088                 /*
3089                  * Reset timeout so it can be computed again, because
3090                  * s_li_wait_mult might have changed.
3091                  */
3092                 sbi->s_li_request->lr_timeout = 0;
3093                 goto out;
3094         }
3095
3096         if (first_not_zeroed == ngroups ||
3097             (sb->s_flags & MS_RDONLY) ||
3098             !test_opt(sb, INIT_INODE_TABLE))
3099                 goto out;
3100
3101         elr = ext4_li_request_new(sb, first_not_zeroed);
3102         if (!elr) {
3103                 ret = -ENOMEM;
3104                 goto out;
3105         }
3106
3107         if (NULL == ext4_li_info) {
3108                 ret = ext4_li_info_new();
3109                 if (ret)
3110                         goto out;
3111         }
3112
3113         mutex_lock(&ext4_li_info->li_list_mtx);
3114         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3115         mutex_unlock(&ext4_li_info->li_list_mtx);
3116
3117         sbi->s_li_request = elr;
3118         /*
3119          * set elr to NULL here since it has been inserted to
3120          * the request_list and the removal and free of it is
3121          * handled by ext4_clear_request_list from now on.
3122          */
3123         elr = NULL;
3124
3125         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3126                 ret = ext4_run_lazyinit_thread();
3127                 if (ret)
3128                         goto out;
3129         }
3130 out:
3131         mutex_unlock(&ext4_li_mtx);
3132         if (ret)
3133                 kfree(elr);
3134         return ret;
3135 }
3136
3137 /*
3138  * We do not need to lock anything since this is called on
3139  * module unload.
3140  */
3141 static void ext4_destroy_lazyinit_thread(void)
3142 {
3143         /*
3144          * If thread exited earlier
3145          * there's nothing to be done.
3146          */
3147         if (!ext4_li_info || !ext4_lazyinit_task)
3148                 return;
3149
3150         kthread_stop(ext4_lazyinit_task);
3151 }
3152
3153 static int set_journal_csum_feature_set(struct super_block *sb)
3154 {
3155         int ret = 1;
3156         int compat, incompat;
3157         struct ext4_sb_info *sbi = EXT4_SB(sb);
3158
3159         if (ext4_has_metadata_csum(sb)) {
3160                 /* journal checksum v3 */
3161                 compat = 0;
3162                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3163         } else {
3164                 /* journal checksum v1 */
3165                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3166                 incompat = 0;
3167         }
3168
3169         jbd2_journal_clear_features(sbi->s_journal,
3170                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3171                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3172                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3173         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3174                 ret = jbd2_journal_set_features(sbi->s_journal,
3175                                 compat, 0,
3176                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3177                                 incompat);
3178         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3179                 ret = jbd2_journal_set_features(sbi->s_journal,
3180                                 compat, 0,
3181                                 incompat);
3182                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3183                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3184         } else {
3185                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3186                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3187         }
3188
3189         return ret;
3190 }
3191
3192 /*
3193  * Note: calculating the overhead so we can be compatible with
3194  * historical BSD practice is quite difficult in the face of
3195  * clusters/bigalloc.  This is because multiple metadata blocks from
3196  * different block group can end up in the same allocation cluster.
3197  * Calculating the exact overhead in the face of clustered allocation
3198  * requires either O(all block bitmaps) in memory or O(number of block
3199  * groups**2) in time.  We will still calculate the superblock for
3200  * older file systems --- and if we come across with a bigalloc file
3201  * system with zero in s_overhead_clusters the estimate will be close to
3202  * correct especially for very large cluster sizes --- but for newer
3203  * file systems, it's better to calculate this figure once at mkfs
3204  * time, and store it in the superblock.  If the superblock value is
3205  * present (even for non-bigalloc file systems), we will use it.
3206  */
3207 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3208                           char *buf)
3209 {
3210         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3211         struct ext4_group_desc  *gdp;
3212         ext4_fsblk_t            first_block, last_block, b;
3213         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3214         int                     s, j, count = 0;
3215
3216         if (!ext4_has_feature_bigalloc(sb))
3217                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3218                         sbi->s_itb_per_group + 2);
3219
3220         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3221                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3222         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3223         for (i = 0; i < ngroups; i++) {
3224                 gdp = ext4_get_group_desc(sb, i, NULL);
3225                 b = ext4_block_bitmap(sb, gdp);
3226                 if (b >= first_block && b <= last_block) {
3227                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3228                         count++;
3229                 }
3230                 b = ext4_inode_bitmap(sb, gdp);
3231                 if (b >= first_block && b <= last_block) {
3232                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3233                         count++;
3234                 }
3235                 b = ext4_inode_table(sb, gdp);
3236                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3237                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3238                                 int c = EXT4_B2C(sbi, b - first_block);
3239                                 ext4_set_bit(c, buf);
3240                                 count++;
3241                         }
3242                 if (i != grp)
3243                         continue;
3244                 s = 0;
3245                 if (ext4_bg_has_super(sb, grp)) {
3246                         ext4_set_bit(s++, buf);
3247                         count++;
3248                 }
3249                 j = ext4_bg_num_gdb(sb, grp);
3250                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3251                         ext4_error(sb, "Invalid number of block group "
3252                                    "descriptor blocks: %d", j);
3253                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3254                 }
3255                 count += j;
3256                 for (; j > 0; j--)
3257                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3258         }
3259         if (!count)
3260                 return 0;
3261         return EXT4_CLUSTERS_PER_GROUP(sb) -
3262                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3263 }
3264
3265 /*
3266  * Compute the overhead and stash it in sbi->s_overhead
3267  */
3268 int ext4_calculate_overhead(struct super_block *sb)
3269 {
3270         struct ext4_sb_info *sbi = EXT4_SB(sb);
3271         struct ext4_super_block *es = sbi->s_es;
3272         struct inode *j_inode;
3273         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3274         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3275         ext4_fsblk_t overhead = 0;
3276         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3277
3278         if (!buf)
3279                 return -ENOMEM;
3280
3281         /*
3282          * Compute the overhead (FS structures).  This is constant
3283          * for a given filesystem unless the number of block groups
3284          * changes so we cache the previous value until it does.
3285          */
3286
3287         /*
3288          * All of the blocks before first_data_block are overhead
3289          */
3290         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3291
3292         /*
3293          * Add the overhead found in each block group
3294          */
3295         for (i = 0; i < ngroups; i++) {
3296                 int blks;
3297
3298                 blks = count_overhead(sb, i, buf);
3299                 overhead += blks;
3300                 if (blks)
3301                         memset(buf, 0, PAGE_SIZE);
3302                 cond_resched();
3303         }
3304
3305         /*
3306          * Add the internal journal blocks whether the journal has been
3307          * loaded or not
3308          */
3309         if (sbi->s_journal && !sbi->journal_bdev)
3310                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3311         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3312                 j_inode = ext4_get_journal_inode(sb, j_inum);
3313                 if (j_inode) {
3314                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3315                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3316                         iput(j_inode);
3317                 } else {
3318                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3319                 }
3320         }
3321         sbi->s_overhead = overhead;
3322         smp_wmb();
3323         free_page((unsigned long) buf);
3324         return 0;
3325 }
3326
3327 static void ext4_set_resv_clusters(struct super_block *sb)
3328 {
3329         ext4_fsblk_t resv_clusters;
3330         struct ext4_sb_info *sbi = EXT4_SB(sb);
3331
3332         /*
3333          * There's no need to reserve anything when we aren't using extents.
3334          * The space estimates are exact, there are no unwritten extents,
3335          * hole punching doesn't need new metadata... This is needed especially
3336          * to keep ext2/3 backward compatibility.
3337          */
3338         if (!ext4_has_feature_extents(sb))
3339                 return;
3340         /*
3341          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3342          * This should cover the situations where we can not afford to run
3343          * out of space like for example punch hole, or converting
3344          * unwritten extents in delalloc path. In most cases such
3345          * allocation would require 1, or 2 blocks, higher numbers are
3346          * very rare.
3347          */
3348         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3349                          sbi->s_cluster_bits);
3350
3351         do_div(resv_clusters, 50);
3352         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3353
3354         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3355 }
3356
3357 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3358 {
3359         char *orig_data = kstrdup(data, GFP_KERNEL);
3360         struct buffer_head *bh;
3361         struct ext4_super_block *es = NULL;
3362         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3363         ext4_fsblk_t block;
3364         ext4_fsblk_t sb_block = get_sb_block(&data);
3365         ext4_fsblk_t logical_sb_block;
3366         unsigned long offset = 0;
3367         unsigned long journal_devnum = 0;
3368         unsigned long def_mount_opts;
3369         struct inode *root;
3370         const char *descr;
3371         int ret = -ENOMEM;
3372         int blocksize, clustersize;
3373         unsigned int db_count;
3374         unsigned int i;
3375         int needs_recovery, has_huge_files, has_bigalloc;
3376         __u64 blocks_count;
3377         int err = 0;
3378         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3379         ext4_group_t first_not_zeroed;
3380
3381         if ((data && !orig_data) || !sbi)
3382                 goto out_free_base;
3383
3384         sbi->s_blockgroup_lock =
3385                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3386         if (!sbi->s_blockgroup_lock)
3387                 goto out_free_base;
3388
3389         sb->s_fs_info = sbi;
3390         sbi->s_sb = sb;
3391         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3392         sbi->s_sb_block = sb_block;
3393         if (sb->s_bdev->bd_part)
3394                 sbi->s_sectors_written_start =
3395                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3396
3397         /* Cleanup superblock name */
3398         strreplace(sb->s_id, '/', '!');
3399
3400         /* -EINVAL is default */
3401         ret = -EINVAL;
3402         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3403         if (!blocksize) {
3404                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3405                 goto out_fail;
3406         }
3407
3408         /*
3409          * The ext4 superblock will not be buffer aligned for other than 1kB
3410          * block sizes.  We need to calculate the offset from buffer start.
3411          */
3412         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3413                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3414                 offset = do_div(logical_sb_block, blocksize);
3415         } else {
3416                 logical_sb_block = sb_block;
3417         }
3418
3419         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3420                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3421                 goto out_fail;
3422         }
3423         /*
3424          * Note: s_es must be initialized as soon as possible because
3425          *       some ext4 macro-instructions depend on its value
3426          */
3427         es = (struct ext4_super_block *) (bh->b_data + offset);
3428         sbi->s_es = es;
3429         sb->s_magic = le16_to_cpu(es->s_magic);
3430         if (sb->s_magic != EXT4_SUPER_MAGIC)
3431                 goto cantfind_ext4;
3432         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3433
3434         /* Warn if metadata_csum and gdt_csum are both set. */
3435         if (ext4_has_feature_metadata_csum(sb) &&
3436             ext4_has_feature_gdt_csum(sb))
3437                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3438                              "redundant flags; please run fsck.");
3439
3440         /* Check for a known checksum algorithm */
3441         if (!ext4_verify_csum_type(sb, es)) {
3442                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3443                          "unknown checksum algorithm.");
3444                 silent = 1;
3445                 goto cantfind_ext4;
3446         }
3447
3448         /* Load the checksum driver */
3449         if (ext4_has_feature_metadata_csum(sb)) {
3450                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3451                 if (IS_ERR(sbi->s_chksum_driver)) {
3452                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3453                         ret = PTR_ERR(sbi->s_chksum_driver);
3454                         sbi->s_chksum_driver = NULL;
3455                         goto failed_mount;
3456                 }
3457         }
3458
3459         /* Check superblock checksum */
3460         if (!ext4_superblock_csum_verify(sb, es)) {
3461                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3462                          "invalid superblock checksum.  Run e2fsck?");
3463                 silent = 1;
3464                 ret = -EFSBADCRC;
3465                 goto cantfind_ext4;
3466         }
3467
3468         /* Precompute checksum seed for all metadata */
3469         if (ext4_has_feature_csum_seed(sb))
3470                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3471         else if (ext4_has_metadata_csum(sb))
3472                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3473                                                sizeof(es->s_uuid));
3474
3475         /* Set defaults before we parse the mount options */
3476         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3477         set_opt(sb, INIT_INODE_TABLE);
3478         if (def_mount_opts & EXT4_DEFM_DEBUG)
3479                 set_opt(sb, DEBUG);
3480         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3481                 set_opt(sb, GRPID);
3482         if (def_mount_opts & EXT4_DEFM_UID16)
3483                 set_opt(sb, NO_UID32);
3484         /* xattr user namespace & acls are now defaulted on */
3485         set_opt(sb, XATTR_USER);
3486 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3487         set_opt(sb, POSIX_ACL);
3488 #endif
3489         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3490         if (ext4_has_metadata_csum(sb))
3491                 set_opt(sb, JOURNAL_CHECKSUM);
3492
3493         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3494                 set_opt(sb, JOURNAL_DATA);
3495         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3496                 set_opt(sb, ORDERED_DATA);
3497         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3498                 set_opt(sb, WRITEBACK_DATA);
3499
3500         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3501                 set_opt(sb, ERRORS_PANIC);
3502         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3503                 set_opt(sb, ERRORS_CONT);
3504         else
3505                 set_opt(sb, ERRORS_RO);
3506         /* block_validity enabled by default; disable with noblock_validity */
3507         set_opt(sb, BLOCK_VALIDITY);
3508         if (def_mount_opts & EXT4_DEFM_DISCARD)
3509                 set_opt(sb, DISCARD);
3510
3511         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3512         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3513         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3514         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3515         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3516
3517         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3518                 set_opt(sb, BARRIER);
3519
3520         /*
3521          * enable delayed allocation by default
3522          * Use -o nodelalloc to turn it off
3523          */
3524         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3525             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3526                 set_opt(sb, DELALLOC);
3527
3528         /*
3529          * set default s_li_wait_mult for lazyinit, for the case there is
3530          * no mount option specified.
3531          */
3532         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3533
3534         if (sbi->s_es->s_mount_opts[0]) {
3535                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3536                                               sizeof(sbi->s_es->s_mount_opts),
3537                                               GFP_KERNEL);
3538                 if (!s_mount_opts)
3539                         goto failed_mount;
3540                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3541                                    &journal_ioprio, 0)) {
3542                         ext4_msg(sb, KERN_WARNING,
3543                                  "failed to parse options in superblock: %s",
3544                                  s_mount_opts);
3545                 }
3546                 kfree(s_mount_opts);
3547         }
3548         sbi->s_def_mount_opt = sbi->s_mount_opt;
3549         if (!parse_options((char *) data, sb, &journal_devnum,
3550                            &journal_ioprio, 0))
3551                 goto failed_mount;
3552
3553         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3554                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3555                             "with data=journal disables delayed "
3556                             "allocation and O_DIRECT support!\n");
3557                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3558                         ext4_msg(sb, KERN_ERR, "can't mount with "
3559                                  "both data=journal and delalloc");
3560                         goto failed_mount;
3561                 }
3562                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3563                         ext4_msg(sb, KERN_ERR, "can't mount with "
3564                                  "both data=journal and dioread_nolock");
3565                         goto failed_mount;
3566                 }
3567                 if (test_opt(sb, DAX)) {
3568                         ext4_msg(sb, KERN_ERR, "can't mount with "
3569                                  "both data=journal and dax");
3570                         goto failed_mount;
3571                 }
3572                 if (ext4_has_feature_encrypt(sb)) {
3573                         ext4_msg(sb, KERN_WARNING,
3574                                  "encrypted files will use data=ordered "
3575                                  "instead of data journaling mode");
3576                 }
3577                 if (test_opt(sb, DELALLOC))
3578                         clear_opt(sb, DELALLOC);
3579         } else {
3580                 sb->s_iflags |= SB_I_CGROUPWB;
3581         }
3582
3583         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3584                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3585
3586         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3587             (ext4_has_compat_features(sb) ||
3588              ext4_has_ro_compat_features(sb) ||
3589              ext4_has_incompat_features(sb)))
3590                 ext4_msg(sb, KERN_WARNING,
3591                        "feature flags set on rev 0 fs, "
3592                        "running e2fsck is recommended");
3593
3594         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3595                 set_opt2(sb, HURD_COMPAT);
3596                 if (ext4_has_feature_64bit(sb)) {
3597                         ext4_msg(sb, KERN_ERR,
3598                                  "The Hurd can't support 64-bit file systems");
3599                         goto failed_mount;
3600                 }
3601         }
3602
3603         if (IS_EXT2_SB(sb)) {
3604                 if (ext2_feature_set_ok(sb))
3605                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3606                                  "using the ext4 subsystem");
3607                 else {
3608                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3609                                  "to feature incompatibilities");
3610                         goto failed_mount;
3611                 }
3612         }
3613
3614         if (IS_EXT3_SB(sb)) {
3615                 if (ext3_feature_set_ok(sb))
3616                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3617                                  "using the ext4 subsystem");
3618                 else {
3619                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3620                                  "to feature incompatibilities");
3621                         goto failed_mount;
3622                 }
3623         }
3624
3625         /*
3626          * Check feature flags regardless of the revision level, since we
3627          * previously didn't change the revision level when setting the flags,
3628          * so there is a chance incompat flags are set on a rev 0 filesystem.
3629          */
3630         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3631                 goto failed_mount;
3632
3633         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3634         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3635             blocksize > EXT4_MAX_BLOCK_SIZE) {
3636                 ext4_msg(sb, KERN_ERR,
3637                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3638                          blocksize, le32_to_cpu(es->s_log_block_size));
3639                 goto failed_mount;
3640         }
3641         if (le32_to_cpu(es->s_log_block_size) >
3642             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3643                 ext4_msg(sb, KERN_ERR,
3644                          "Invalid log block size: %u",
3645                          le32_to_cpu(es->s_log_block_size));
3646                 goto failed_mount;
3647         }
3648
3649         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3650                 ext4_msg(sb, KERN_ERR,
3651                          "Number of reserved GDT blocks insanely large: %d",
3652                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3653                 goto failed_mount;
3654         }
3655
3656         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3657                 err = bdev_dax_supported(sb, blocksize);
3658                 if (err)
3659                         goto failed_mount;
3660         }
3661
3662         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3663                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3664                          es->s_encryption_level);
3665                 goto failed_mount;
3666         }
3667
3668         if (sb->s_blocksize != blocksize) {
3669                 /* Validate the filesystem blocksize */
3670                 if (!sb_set_blocksize(sb, blocksize)) {
3671                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3672                                         blocksize);
3673                         goto failed_mount;
3674                 }
3675
3676                 brelse(bh);
3677                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3678                 offset = do_div(logical_sb_block, blocksize);
3679                 bh = sb_bread_unmovable(sb, logical_sb_block);
3680                 if (!bh) {
3681                         ext4_msg(sb, KERN_ERR,
3682                                "Can't read superblock on 2nd try");
3683                         goto failed_mount;
3684                 }
3685                 es = (struct ext4_super_block *)(bh->b_data + offset);
3686                 sbi->s_es = es;
3687                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3688                         ext4_msg(sb, KERN_ERR,
3689                                "Magic mismatch, very weird!");
3690                         goto failed_mount;
3691                 }
3692         }
3693
3694         has_huge_files = ext4_has_feature_huge_file(sb);
3695         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3696                                                       has_huge_files);
3697         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3698
3699         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3700                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3701                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3702         } else {
3703                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3704                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3705                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3706                     (!is_power_of_2(sbi->s_inode_size)) ||
3707                     (sbi->s_inode_size > blocksize)) {
3708                         ext4_msg(sb, KERN_ERR,
3709                                "unsupported inode size: %d",
3710                                sbi->s_inode_size);
3711                         goto failed_mount;
3712                 }
3713                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3714                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3715         }
3716
3717         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3718         if (ext4_has_feature_64bit(sb)) {
3719                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3720                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3721                     !is_power_of_2(sbi->s_desc_size)) {
3722                         ext4_msg(sb, KERN_ERR,
3723                                "unsupported descriptor size %lu",
3724                                sbi->s_desc_size);
3725                         goto failed_mount;
3726                 }
3727         } else
3728                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3729
3730         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3731         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3732
3733         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3734         if (sbi->s_inodes_per_block == 0)
3735                 goto cantfind_ext4;
3736         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3737             sbi->s_inodes_per_group > blocksize * 8) {
3738                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3739                          sbi->s_blocks_per_group);
3740                 goto failed_mount;
3741         }
3742         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3743                                         sbi->s_inodes_per_block;
3744         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3745         sbi->s_sbh = bh;
3746         sbi->s_mount_state = le16_to_cpu(es->s_state);
3747         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3748         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3749
3750         for (i = 0; i < 4; i++)
3751                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3752         sbi->s_def_hash_version = es->s_def_hash_version;
3753         if (ext4_has_feature_dir_index(sb)) {
3754                 i = le32_to_cpu(es->s_flags);
3755                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3756                         sbi->s_hash_unsigned = 3;
3757                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3758 #ifdef __CHAR_UNSIGNED__
3759                         if (!(sb->s_flags & MS_RDONLY))
3760                                 es->s_flags |=
3761                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3762                         sbi->s_hash_unsigned = 3;
3763 #else
3764                         if (!(sb->s_flags & MS_RDONLY))
3765                                 es->s_flags |=
3766                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3767 #endif
3768                 }
3769         }
3770
3771         /* Handle clustersize */
3772         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3773         has_bigalloc = ext4_has_feature_bigalloc(sb);
3774         if (has_bigalloc) {
3775                 if (clustersize < blocksize) {
3776                         ext4_msg(sb, KERN_ERR,
3777                                  "cluster size (%d) smaller than "
3778                                  "block size (%d)", clustersize, blocksize);
3779                         goto failed_mount;
3780                 }
3781                 if (le32_to_cpu(es->s_log_cluster_size) >
3782                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3783                         ext4_msg(sb, KERN_ERR,
3784                                  "Invalid log cluster size: %u",
3785                                  le32_to_cpu(es->s_log_cluster_size));
3786                         goto failed_mount;
3787                 }
3788                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3789                         le32_to_cpu(es->s_log_block_size);
3790                 sbi->s_clusters_per_group =
3791                         le32_to_cpu(es->s_clusters_per_group);
3792                 if (sbi->s_clusters_per_group > blocksize * 8) {
3793                         ext4_msg(sb, KERN_ERR,
3794                                  "#clusters per group too big: %lu",
3795                                  sbi->s_clusters_per_group);
3796                         goto failed_mount;
3797                 }
3798                 if (sbi->s_blocks_per_group !=
3799                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3800                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3801                                  "clusters per group (%lu) inconsistent",
3802                                  sbi->s_blocks_per_group,
3803                                  sbi->s_clusters_per_group);
3804                         goto failed_mount;
3805                 }
3806         } else {
3807                 if (clustersize != blocksize) {
3808                         ext4_warning(sb, "fragment/cluster size (%d) != "
3809                                      "block size (%d)", clustersize,
3810                                      blocksize);
3811                         clustersize = blocksize;
3812                 }
3813                 if (sbi->s_blocks_per_group > blocksize * 8) {
3814                         ext4_msg(sb, KERN_ERR,
3815                                  "#blocks per group too big: %lu",
3816                                  sbi->s_blocks_per_group);
3817                         goto failed_mount;
3818                 }
3819                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3820                 sbi->s_cluster_bits = 0;
3821         }
3822         sbi->s_cluster_ratio = clustersize / blocksize;
3823
3824         /* Do we have standard group size of clustersize * 8 blocks ? */
3825         if (sbi->s_blocks_per_group == clustersize << 3)
3826                 set_opt2(sb, STD_GROUP_SIZE);
3827
3828         /*
3829          * Test whether we have more sectors than will fit in sector_t,
3830          * and whether the max offset is addressable by the page cache.
3831          */
3832         err = generic_check_addressable(sb->s_blocksize_bits,
3833                                         ext4_blocks_count(es));
3834         if (err) {
3835                 ext4_msg(sb, KERN_ERR, "filesystem"
3836                          " too large to mount safely on this system");
3837                 if (sizeof(sector_t) < 8)
3838                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3839                 goto failed_mount;
3840         }
3841
3842         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3843                 goto cantfind_ext4;
3844
3845         /* check blocks count against device size */
3846         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3847         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3848                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3849                        "exceeds size of device (%llu blocks)",
3850                        ext4_blocks_count(es), blocks_count);
3851                 goto failed_mount;
3852         }
3853
3854         /*
3855          * It makes no sense for the first data block to be beyond the end
3856          * of the filesystem.
3857          */
3858         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3859                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3860                          "block %u is beyond end of filesystem (%llu)",
3861                          le32_to_cpu(es->s_first_data_block),
3862                          ext4_blocks_count(es));
3863                 goto failed_mount;
3864         }
3865         blocks_count = (ext4_blocks_count(es) -
3866                         le32_to_cpu(es->s_first_data_block) +
3867                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3868         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3869         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3870                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3871                        "(block count %llu, first data block %u, "
3872                        "blocks per group %lu)", sbi->s_groups_count,
3873                        ext4_blocks_count(es),
3874                        le32_to_cpu(es->s_first_data_block),
3875                        EXT4_BLOCKS_PER_GROUP(sb));
3876                 goto failed_mount;
3877         }
3878         sbi->s_groups_count = blocks_count;
3879         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3880                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3881         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3882                    EXT4_DESC_PER_BLOCK(sb);
3883         if (ext4_has_feature_meta_bg(sb)) {
3884                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3885                         ext4_msg(sb, KERN_WARNING,
3886                                  "first meta block group too large: %u "
3887                                  "(group descriptor block count %u)",
3888                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3889                         goto failed_mount;
3890                 }
3891         }
3892         sbi->s_group_desc = kvmalloc(db_count *
3893                                           sizeof(struct buffer_head *),
3894                                           GFP_KERNEL);
3895         if (sbi->s_group_desc == NULL) {
3896                 ext4_msg(sb, KERN_ERR, "not enough memory");
3897                 ret = -ENOMEM;
3898                 goto failed_mount;
3899         }
3900
3901         bgl_lock_init(sbi->s_blockgroup_lock);
3902
3903         /* Pre-read the descriptors into the buffer cache */
3904         for (i = 0; i < db_count; i++) {
3905                 block = descriptor_loc(sb, logical_sb_block, i);
3906                 sb_breadahead(sb, block);
3907         }
3908
3909         for (i = 0; i < db_count; i++) {
3910                 block = descriptor_loc(sb, logical_sb_block, i);
3911                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3912                 if (!sbi->s_group_desc[i]) {
3913                         ext4_msg(sb, KERN_ERR,
3914                                "can't read group descriptor %d", i);
3915                         db_count = i;
3916                         goto failed_mount2;
3917                 }
3918         }
3919         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3920                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3921                 ret = -EFSCORRUPTED;
3922                 goto failed_mount2;
3923         }
3924
3925         sbi->s_gdb_count = db_count;
3926         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3927         spin_lock_init(&sbi->s_next_gen_lock);
3928
3929         setup_timer(&sbi->s_err_report, print_daily_error_info,
3930                 (unsigned long) sb);
3931
3932         /* Register extent status tree shrinker */
3933         if (ext4_es_register_shrinker(sbi))
3934                 goto failed_mount3;
3935
3936         sbi->s_stripe = ext4_get_stripe_size(sbi);
3937         sbi->s_extent_max_zeroout_kb = 32;
3938
3939         /*
3940          * set up enough so that it can read an inode
3941          */
3942         sb->s_op = &ext4_sops;
3943         sb->s_export_op = &ext4_export_ops;
3944         sb->s_xattr = ext4_xattr_handlers;
3945         sb->s_cop = &ext4_cryptops;
3946 #ifdef CONFIG_QUOTA
3947         sb->dq_op = &ext4_quota_operations;
3948         if (ext4_has_feature_quota(sb))
3949                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3950         else
3951                 sb->s_qcop = &ext4_qctl_operations;
3952         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
3953 #endif
3954         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3955
3956         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3957         mutex_init(&sbi->s_orphan_lock);
3958
3959         sb->s_root = NULL;
3960
3961         needs_recovery = (es->s_last_orphan != 0 ||
3962                           ext4_has_feature_journal_needs_recovery(sb));
3963
3964         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3965                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3966                         goto failed_mount3a;
3967
3968         /*
3969          * The first inode we look at is the journal inode.  Don't try
3970          * root first: it may be modified in the journal!
3971          */
3972         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3973                 err = ext4_load_journal(sb, es, journal_devnum);
3974                 if (err)
3975                         goto failed_mount3a;
3976         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3977                    ext4_has_feature_journal_needs_recovery(sb)) {
3978                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3979                        "suppressed and not mounted read-only");
3980                 goto failed_mount_wq;
3981         } else {
3982                 /* Nojournal mode, all journal mount options are illegal */
3983                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3984                         ext4_msg(sb, KERN_ERR, "can't mount with "
3985                                  "journal_checksum, fs mounted w/o journal");
3986                         goto failed_mount_wq;
3987                 }
3988                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3989                         ext4_msg(sb, KERN_ERR, "can't mount with "
3990                                  "journal_async_commit, fs mounted w/o journal");
3991                         goto failed_mount_wq;
3992                 }
3993                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3994                         ext4_msg(sb, KERN_ERR, "can't mount with "
3995                                  "commit=%lu, fs mounted w/o journal",
3996                                  sbi->s_commit_interval / HZ);
3997                         goto failed_mount_wq;
3998                 }
3999                 if (EXT4_MOUNT_DATA_FLAGS &
4000                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4001                         ext4_msg(sb, KERN_ERR, "can't mount with "
4002                                  "data=, fs mounted w/o journal");
4003                         goto failed_mount_wq;
4004                 }
4005                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4006                 clear_opt(sb, JOURNAL_CHECKSUM);
4007                 clear_opt(sb, DATA_FLAGS);
4008                 sbi->s_journal = NULL;
4009                 needs_recovery = 0;
4010                 goto no_journal;
4011         }
4012
4013         if (ext4_has_feature_64bit(sb) &&
4014             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4015                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4016                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4017                 goto failed_mount_wq;
4018         }
4019
4020         if (!set_journal_csum_feature_set(sb)) {
4021                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4022                          "feature set");
4023                 goto failed_mount_wq;
4024         }
4025
4026         /* We have now updated the journal if required, so we can
4027          * validate the data journaling mode. */
4028         switch (test_opt(sb, DATA_FLAGS)) {
4029         case 0:
4030                 /* No mode set, assume a default based on the journal
4031                  * capabilities: ORDERED_DATA if the journal can
4032                  * cope, else JOURNAL_DATA
4033                  */
4034                 if (jbd2_journal_check_available_features
4035                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4036                         set_opt(sb, ORDERED_DATA);
4037                 else
4038                         set_opt(sb, JOURNAL_DATA);
4039                 break;
4040
4041         case EXT4_MOUNT_ORDERED_DATA:
4042         case EXT4_MOUNT_WRITEBACK_DATA:
4043                 if (!jbd2_journal_check_available_features
4044                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4045                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4046                                "requested data journaling mode");
4047                         goto failed_mount_wq;
4048                 }
4049         default:
4050                 break;
4051         }
4052
4053         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4054             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4055                 ext4_msg(sb, KERN_ERR, "can't mount with "
4056                         "journal_async_commit in data=ordered mode");
4057                 goto failed_mount_wq;
4058         }
4059
4060         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4061
4062         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4063
4064 no_journal:
4065         sbi->s_mb_cache = ext4_xattr_create_cache();
4066         if (!sbi->s_mb_cache) {
4067                 ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4068                 goto failed_mount_wq;
4069         }
4070
4071         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4072             (blocksize != PAGE_SIZE)) {
4073                 ext4_msg(sb, KERN_ERR,
4074                          "Unsupported blocksize for fs encryption");
4075                 goto failed_mount_wq;
4076         }
4077
4078         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
4079             !ext4_has_feature_encrypt(sb)) {
4080                 ext4_set_feature_encrypt(sb);
4081                 ext4_commit_super(sb, 1);
4082         }
4083
4084         /*
4085          * Get the # of file system overhead blocks from the
4086          * superblock if present.
4087          */
4088         if (es->s_overhead_clusters)
4089                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4090         else {
4091                 err = ext4_calculate_overhead(sb);
4092                 if (err)
4093                         goto failed_mount_wq;
4094         }
4095
4096         /*
4097          * The maximum number of concurrent works can be high and
4098          * concurrency isn't really necessary.  Limit it to 1.
4099          */
4100         EXT4_SB(sb)->rsv_conversion_wq =
4101                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4102         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4103                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4104                 ret = -ENOMEM;
4105                 goto failed_mount4;
4106         }
4107
4108         /*
4109          * The jbd2_journal_load will have done any necessary log recovery,
4110          * so we can safely mount the rest of the filesystem now.
4111          */
4112
4113         root = ext4_iget(sb, EXT4_ROOT_INO);
4114         if (IS_ERR(root)) {
4115                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4116                 ret = PTR_ERR(root);
4117                 root = NULL;
4118                 goto failed_mount4;
4119         }
4120         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4121                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4122                 iput(root);
4123                 goto failed_mount4;
4124         }
4125         sb->s_root = d_make_root(root);
4126         if (!sb->s_root) {
4127                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4128                 ret = -ENOMEM;
4129                 goto failed_mount4;
4130         }
4131
4132         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4133                 sb->s_flags |= MS_RDONLY;
4134
4135         /* determine the minimum size of new large inodes, if present */
4136         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4137             sbi->s_want_extra_isize == 0) {
4138                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4139                                                      EXT4_GOOD_OLD_INODE_SIZE;
4140                 if (ext4_has_feature_extra_isize(sb)) {
4141                         if (sbi->s_want_extra_isize <
4142                             le16_to_cpu(es->s_want_extra_isize))
4143                                 sbi->s_want_extra_isize =
4144                                         le16_to_cpu(es->s_want_extra_isize);
4145                         if (sbi->s_want_extra_isize <
4146                             le16_to_cpu(es->s_min_extra_isize))
4147                                 sbi->s_want_extra_isize =
4148                                         le16_to_cpu(es->s_min_extra_isize);
4149                 }
4150         }
4151         /* Check if enough inode space is available */
4152         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4153                                                         sbi->s_inode_size) {
4154                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4155                                                        EXT4_GOOD_OLD_INODE_SIZE;
4156                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4157                          "available");
4158         }
4159
4160         ext4_set_resv_clusters(sb);
4161
4162         err = ext4_setup_system_zone(sb);
4163         if (err) {
4164                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4165                          "zone (%d)", err);
4166                 goto failed_mount4a;
4167         }
4168
4169         ext4_ext_init(sb);
4170         err = ext4_mb_init(sb);
4171         if (err) {
4172                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4173                          err);
4174                 goto failed_mount5;
4175         }
4176
4177         block = ext4_count_free_clusters(sb);
4178         ext4_free_blocks_count_set(sbi->s_es,
4179                                    EXT4_C2B(sbi, block));
4180         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4181                                   GFP_KERNEL);
4182         if (!err) {
4183                 unsigned long freei = ext4_count_free_inodes(sb);
4184                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4185                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4186                                           GFP_KERNEL);
4187         }
4188         if (!err)
4189                 err = percpu_counter_init(&sbi->s_dirs_counter,
4190                                           ext4_count_dirs(sb), GFP_KERNEL);
4191         if (!err)
4192                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4193                                           GFP_KERNEL);
4194         if (!err)
4195                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4196
4197         if (err) {
4198                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4199                 goto failed_mount6;
4200         }
4201
4202         if (ext4_has_feature_flex_bg(sb))
4203                 if (!ext4_fill_flex_info(sb)) {
4204                         ext4_msg(sb, KERN_ERR,
4205                                "unable to initialize "
4206                                "flex_bg meta info!");
4207                         goto failed_mount6;
4208                 }
4209
4210         err = ext4_register_li_request(sb, first_not_zeroed);
4211         if (err)
4212                 goto failed_mount6;
4213
4214         err = ext4_register_sysfs(sb);
4215         if (err)
4216                 goto failed_mount7;
4217
4218 #ifdef CONFIG_QUOTA
4219         /* Enable quota usage during mount. */
4220         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
4221                 err = ext4_enable_quotas(sb);
4222                 if (err)
4223                         goto failed_mount8;
4224         }
4225 #endif  /* CONFIG_QUOTA */
4226
4227         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4228         ext4_orphan_cleanup(sb, es);
4229         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4230         if (needs_recovery) {
4231                 ext4_msg(sb, KERN_INFO, "recovery complete");
4232                 ext4_mark_recovery_complete(sb, es);
4233         }
4234         if (EXT4_SB(sb)->s_journal) {
4235                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4236                         descr = " journalled data mode";
4237                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4238                         descr = " ordered data mode";
4239                 else
4240                         descr = " writeback data mode";
4241         } else
4242                 descr = "out journal";
4243
4244         if (test_opt(sb, DISCARD)) {
4245                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4246                 if (!blk_queue_discard(q))
4247                         ext4_msg(sb, KERN_WARNING,
4248                                  "mounting with \"discard\" option, but "
4249                                  "the device does not support discard");
4250         }
4251
4252         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4253                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4254                          "Opts: %.*s%s%s", descr,
4255                          (int) sizeof(sbi->s_es->s_mount_opts),
4256                          sbi->s_es->s_mount_opts,
4257                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4258
4259         if (es->s_error_count)
4260                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4261
4262         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4263         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4264         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4265         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4266
4267         kfree(orig_data);
4268         return 0;
4269
4270 cantfind_ext4:
4271         if (!silent)
4272                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4273         goto failed_mount;
4274
4275 #ifdef CONFIG_QUOTA
4276 failed_mount8:
4277         ext4_unregister_sysfs(sb);
4278 #endif
4279 failed_mount7:
4280         ext4_unregister_li_request(sb);
4281 failed_mount6:
4282         ext4_mb_release(sb);
4283         if (sbi->s_flex_groups)
4284                 kvfree(sbi->s_flex_groups);
4285         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4286         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4287         percpu_counter_destroy(&sbi->s_dirs_counter);
4288         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4289 failed_mount5:
4290         ext4_ext_release(sb);
4291         ext4_release_system_zone(sb);
4292 failed_mount4a:
4293         dput(sb->s_root);
4294         sb->s_root = NULL;
4295 failed_mount4:
4296         ext4_msg(sb, KERN_ERR, "mount failed");
4297         if (EXT4_SB(sb)->rsv_conversion_wq)
4298                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4299 failed_mount_wq:
4300         if (sbi->s_mb_cache) {
4301                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
4302                 sbi->s_mb_cache = NULL;
4303         }
4304         if (sbi->s_journal) {
4305                 jbd2_journal_destroy(sbi->s_journal);
4306                 sbi->s_journal = NULL;
4307         }
4308 failed_mount3a:
4309         ext4_es_unregister_shrinker(sbi);
4310 failed_mount3:
4311         del_timer_sync(&sbi->s_err_report);
4312         if (sbi->s_mmp_tsk)
4313                 kthread_stop(sbi->s_mmp_tsk);
4314 failed_mount2:
4315         for (i = 0; i < db_count; i++)
4316                 brelse(sbi->s_group_desc[i]);
4317         kvfree(sbi->s_group_desc);
4318 failed_mount:
4319         if (sbi->s_chksum_driver)
4320                 crypto_free_shash(sbi->s_chksum_driver);
4321 #ifdef CONFIG_QUOTA
4322         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4323                 kfree(sbi->s_qf_names[i]);
4324 #endif
4325         ext4_blkdev_remove(sbi);
4326         brelse(bh);
4327 out_fail:
4328         sb->s_fs_info = NULL;
4329         kfree(sbi->s_blockgroup_lock);
4330 out_free_base:
4331         kfree(sbi);
4332         kfree(orig_data);
4333         return err ? err : ret;
4334 }
4335
4336 /*
4337  * Setup any per-fs journal parameters now.  We'll do this both on
4338  * initial mount, once the journal has been initialised but before we've
4339  * done any recovery; and again on any subsequent remount.
4340  */
4341 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4342 {
4343         struct ext4_sb_info *sbi = EXT4_SB(sb);
4344
4345         journal->j_commit_interval = sbi->s_commit_interval;
4346         journal->j_min_batch_time = sbi->s_min_batch_time;
4347         journal->j_max_batch_time = sbi->s_max_batch_time;
4348
4349         write_lock(&journal->j_state_lock);
4350         if (test_opt(sb, BARRIER))
4351                 journal->j_flags |= JBD2_BARRIER;
4352         else
4353                 journal->j_flags &= ~JBD2_BARRIER;
4354         if (test_opt(sb, DATA_ERR_ABORT))
4355                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4356         else
4357                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4358         write_unlock(&journal->j_state_lock);
4359 }
4360
4361 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4362                                              unsigned int journal_inum)
4363 {
4364         struct inode *journal_inode;
4365
4366         /*
4367          * Test for the existence of a valid inode on disk.  Bad things
4368          * happen if we iget() an unused inode, as the subsequent iput()
4369          * will try to delete it.
4370          */
4371         journal_inode = ext4_iget(sb, journal_inum);
4372         if (IS_ERR(journal_inode)) {
4373                 ext4_msg(sb, KERN_ERR, "no journal found");
4374                 return NULL;
4375         }
4376         if (!journal_inode->i_nlink) {
4377                 make_bad_inode(journal_inode);
4378                 iput(journal_inode);
4379                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4380                 return NULL;
4381         }
4382
4383         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4384                   journal_inode, journal_inode->i_size);
4385         if (!S_ISREG(journal_inode->i_mode)) {
4386                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4387                 iput(journal_inode);
4388                 return NULL;
4389         }
4390         return journal_inode;
4391 }
4392
4393 static journal_t *ext4_get_journal(struct super_block *sb,
4394                                    unsigned int journal_inum)
4395 {
4396         struct inode *journal_inode;
4397         journal_t *journal;
4398
4399         BUG_ON(!ext4_has_feature_journal(sb));
4400
4401         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4402         if (!journal_inode)
4403                 return NULL;
4404
4405         journal = jbd2_journal_init_inode(journal_inode);
4406         if (!journal) {
4407                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4408                 iput(journal_inode);
4409                 return NULL;
4410         }
4411         journal->j_private = sb;
4412         ext4_init_journal_params(sb, journal);
4413         return journal;
4414 }
4415
4416 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4417                                        dev_t j_dev)
4418 {
4419         struct buffer_head *bh;
4420         journal_t *journal;
4421         ext4_fsblk_t start;
4422         ext4_fsblk_t len;
4423         int hblock, blocksize;
4424         ext4_fsblk_t sb_block;
4425         unsigned long offset;
4426         struct ext4_super_block *es;
4427         struct block_device *bdev;
4428
4429         BUG_ON(!ext4_has_feature_journal(sb));
4430
4431         bdev = ext4_blkdev_get(j_dev, sb);
4432         if (bdev == NULL)
4433                 return NULL;
4434
4435         blocksize = sb->s_blocksize;
4436         hblock = bdev_logical_block_size(bdev);
4437         if (blocksize < hblock) {
4438                 ext4_msg(sb, KERN_ERR,
4439                         "blocksize too small for journal device");
4440                 goto out_bdev;
4441         }
4442
4443         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4444         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4445         set_blocksize(bdev, blocksize);
4446         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4447                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4448                        "external journal");
4449                 goto out_bdev;
4450         }
4451
4452         es = (struct ext4_super_block *) (bh->b_data + offset);
4453         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4454             !(le32_to_cpu(es->s_feature_incompat) &
4455               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4456                 ext4_msg(sb, KERN_ERR, "external journal has "
4457                                         "bad superblock");
4458                 brelse(bh);
4459                 goto out_bdev;
4460         }
4461
4462         if ((le32_to_cpu(es->s_feature_ro_compat) &
4463              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4464             es->s_checksum != ext4_superblock_csum(sb, es)) {
4465                 ext4_msg(sb, KERN_ERR, "external journal has "
4466                                        "corrupt superblock");
4467                 brelse(bh);
4468                 goto out_bdev;
4469         }
4470
4471         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4472                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4473                 brelse(bh);
4474                 goto out_bdev;
4475         }
4476
4477         len = ext4_blocks_count(es);
4478         start = sb_block + 1;
4479         brelse(bh);     /* we're done with the superblock */
4480
4481         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4482                                         start, len, blocksize);
4483         if (!journal) {
4484                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4485                 goto out_bdev;
4486         }
4487         journal->j_private = sb;
4488         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4489         wait_on_buffer(journal->j_sb_buffer);
4490         if (!buffer_uptodate(journal->j_sb_buffer)) {
4491                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4492                 goto out_journal;
4493         }
4494         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4495                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4496                                         "user (unsupported) - %d",
4497                         be32_to_cpu(journal->j_superblock->s_nr_users));
4498                 goto out_journal;
4499         }
4500         EXT4_SB(sb)->journal_bdev = bdev;
4501         ext4_init_journal_params(sb, journal);
4502         return journal;
4503
4504 out_journal:
4505         jbd2_journal_destroy(journal);
4506 out_bdev:
4507         ext4_blkdev_put(bdev);
4508         return NULL;
4509 }
4510
4511 static int ext4_load_journal(struct super_block *sb,
4512                              struct ext4_super_block *es,
4513                              unsigned long journal_devnum)
4514 {
4515         journal_t *journal;
4516         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4517         dev_t journal_dev;
4518         int err = 0;
4519         int really_read_only;
4520
4521         BUG_ON(!ext4_has_feature_journal(sb));
4522
4523         if (journal_devnum &&
4524             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4525                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4526                         "numbers have changed");
4527                 journal_dev = new_decode_dev(journal_devnum);
4528         } else
4529                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4530
4531         really_read_only = bdev_read_only(sb->s_bdev);
4532
4533         /*
4534          * Are we loading a blank journal or performing recovery after a
4535          * crash?  For recovery, we need to check in advance whether we
4536          * can get read-write access to the device.
4537          */
4538         if (ext4_has_feature_journal_needs_recovery(sb)) {
4539                 if (sb->s_flags & MS_RDONLY) {
4540                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4541                                         "required on readonly filesystem");
4542                         if (really_read_only) {
4543                                 ext4_msg(sb, KERN_ERR, "write access "
4544                                         "unavailable, cannot proceed");
4545                                 return -EROFS;
4546                         }
4547                         ext4_msg(sb, KERN_INFO, "write access will "
4548                                "be enabled during recovery");
4549                 }
4550         }
4551
4552         if (journal_inum && journal_dev) {
4553                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4554                        "and inode journals!");
4555                 return -EINVAL;
4556         }
4557
4558         if (journal_inum) {
4559                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4560                         return -EINVAL;
4561         } else {
4562                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4563                         return -EINVAL;
4564         }
4565
4566         if (!(journal->j_flags & JBD2_BARRIER))
4567                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4568
4569         if (!ext4_has_feature_journal_needs_recovery(sb))
4570                 err = jbd2_journal_wipe(journal, !really_read_only);
4571         if (!err) {
4572                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4573                 if (save)
4574                         memcpy(save, ((char *) es) +
4575                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4576                 err = jbd2_journal_load(journal);
4577                 if (save)
4578                         memcpy(((char *) es) + EXT4_S_ERR_START,
4579                                save, EXT4_S_ERR_LEN);
4580                 kfree(save);
4581         }
4582
4583         if (err) {
4584                 ext4_msg(sb, KERN_ERR, "error loading journal");
4585                 jbd2_journal_destroy(journal);
4586                 return err;
4587         }
4588
4589         EXT4_SB(sb)->s_journal = journal;
4590         ext4_clear_journal_err(sb, es);
4591
4592         if (!really_read_only && journal_devnum &&
4593             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4594                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4595
4596                 /* Make sure we flush the recovery flag to disk. */
4597                 ext4_commit_super(sb, 1);
4598         }
4599
4600         return 0;
4601 }
4602
4603 static int ext4_commit_super(struct super_block *sb, int sync)
4604 {
4605         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4606         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4607         int error = 0;
4608
4609         if (!sbh || block_device_ejected(sb))
4610                 return error;
4611         /*
4612          * If the file system is mounted read-only, don't update the
4613          * superblock write time.  This avoids updating the superblock
4614          * write time when we are mounting the root file system
4615          * read/only but we need to replay the journal; at that point,
4616          * for people who are east of GMT and who make their clock
4617          * tick in localtime for Windows bug-for-bug compatibility,
4618          * the clock is set in the future, and this will cause e2fsck
4619          * to complain and force a full file system check.
4620          */
4621         if (!(sb->s_flags & MS_RDONLY))
4622                 es->s_wtime = cpu_to_le32(get_seconds());
4623         if (sb->s_bdev->bd_part)
4624                 es->s_kbytes_written =
4625                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4626                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4627                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4628         else
4629                 es->s_kbytes_written =
4630                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4631         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4632                 ext4_free_blocks_count_set(es,
4633                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4634                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4635         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4636                 es->s_free_inodes_count =
4637                         cpu_to_le32(percpu_counter_sum_positive(
4638                                 &EXT4_SB(sb)->s_freeinodes_counter));
4639         BUFFER_TRACE(sbh, "marking dirty");
4640         ext4_superblock_csum_set(sb);
4641         if (sync)
4642                 lock_buffer(sbh);
4643         if (buffer_write_io_error(sbh)) {
4644                 /*
4645                  * Oh, dear.  A previous attempt to write the
4646                  * superblock failed.  This could happen because the
4647                  * USB device was yanked out.  Or it could happen to
4648                  * be a transient write error and maybe the block will
4649                  * be remapped.  Nothing we can do but to retry the
4650                  * write and hope for the best.
4651                  */
4652                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4653                        "superblock detected");
4654                 clear_buffer_write_io_error(sbh);
4655                 set_buffer_uptodate(sbh);
4656         }
4657         mark_buffer_dirty(sbh);
4658         if (sync) {
4659                 unlock_buffer(sbh);
4660                 error = __sync_dirty_buffer(sbh,
4661                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4662                 if (error)
4663                         return error;
4664
4665                 error = buffer_write_io_error(sbh);
4666                 if (error) {
4667                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4668                                "superblock");
4669                         clear_buffer_write_io_error(sbh);
4670                         set_buffer_uptodate(sbh);
4671                 }
4672         }
4673         return error;
4674 }
4675
4676 /*
4677  * Have we just finished recovery?  If so, and if we are mounting (or
4678  * remounting) the filesystem readonly, then we will end up with a
4679  * consistent fs on disk.  Record that fact.
4680  */
4681 static void ext4_mark_recovery_complete(struct super_block *sb,
4682                                         struct ext4_super_block *es)
4683 {
4684         journal_t *journal = EXT4_SB(sb)->s_journal;
4685
4686         if (!ext4_has_feature_journal(sb)) {
4687                 BUG_ON(journal != NULL);
4688                 return;
4689         }
4690         jbd2_journal_lock_updates(journal);
4691         if (jbd2_journal_flush(journal) < 0)
4692                 goto out;
4693
4694         if (ext4_has_feature_journal_needs_recovery(sb) &&
4695             sb->s_flags & MS_RDONLY) {
4696                 ext4_clear_feature_journal_needs_recovery(sb);
4697                 ext4_commit_super(sb, 1);
4698         }
4699
4700 out:
4701         jbd2_journal_unlock_updates(journal);
4702 }
4703
4704 /*
4705  * If we are mounting (or read-write remounting) a filesystem whose journal
4706  * has recorded an error from a previous lifetime, move that error to the
4707  * main filesystem now.
4708  */
4709 static void ext4_clear_journal_err(struct super_block *sb,
4710                                    struct ext4_super_block *es)
4711 {
4712         journal_t *journal;
4713         int j_errno;
4714         const char *errstr;
4715
4716         BUG_ON(!ext4_has_feature_journal(sb));
4717
4718         journal = EXT4_SB(sb)->s_journal;
4719
4720         /*
4721          * Now check for any error status which may have been recorded in the
4722          * journal by a prior ext4_error() or ext4_abort()
4723          */
4724
4725         j_errno = jbd2_journal_errno(journal);
4726         if (j_errno) {
4727                 char nbuf[16];
4728
4729                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4730                 ext4_warning(sb, "Filesystem error recorded "
4731                              "from previous mount: %s", errstr);
4732                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4733
4734                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4735                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4736                 ext4_commit_super(sb, 1);
4737
4738                 jbd2_journal_clear_err(journal);
4739                 jbd2_journal_update_sb_errno(journal);
4740         }
4741 }
4742
4743 /*
4744  * Force the running and committing transactions to commit,
4745  * and wait on the commit.
4746  */
4747 int ext4_force_commit(struct super_block *sb)
4748 {
4749         journal_t *journal;
4750
4751         if (sb->s_flags & MS_RDONLY)
4752                 return 0;
4753
4754         journal = EXT4_SB(sb)->s_journal;
4755         return ext4_journal_force_commit(journal);
4756 }
4757
4758 static int ext4_sync_fs(struct super_block *sb, int wait)
4759 {
4760         int ret = 0;
4761         tid_t target;
4762         bool needs_barrier = false;
4763         struct ext4_sb_info *sbi = EXT4_SB(sb);
4764
4765         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
4766                 return 0;
4767
4768         trace_ext4_sync_fs(sb, wait);
4769         flush_workqueue(sbi->rsv_conversion_wq);
4770         /*
4771          * Writeback quota in non-journalled quota case - journalled quota has
4772          * no dirty dquots
4773          */
4774         dquot_writeback_dquots(sb, -1);
4775         /*
4776          * Data writeback is possible w/o journal transaction, so barrier must
4777          * being sent at the end of the function. But we can skip it if
4778          * transaction_commit will do it for us.
4779          */
4780         if (sbi->s_journal) {
4781                 target = jbd2_get_latest_transaction(sbi->s_journal);
4782                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4783                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4784                         needs_barrier = true;
4785
4786                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4787                         if (wait)
4788                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4789                                                            target);
4790                 }
4791         } else if (wait && test_opt(sb, BARRIER))
4792                 needs_barrier = true;
4793         if (needs_barrier) {
4794                 int err;
4795                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4796                 if (!ret)
4797                         ret = err;
4798         }
4799
4800         return ret;
4801 }
4802
4803 /*
4804  * LVM calls this function before a (read-only) snapshot is created.  This
4805  * gives us a chance to flush the journal completely and mark the fs clean.
4806  *
4807  * Note that only this function cannot bring a filesystem to be in a clean
4808  * state independently. It relies on upper layer to stop all data & metadata
4809  * modifications.
4810  */
4811 static int ext4_freeze(struct super_block *sb)
4812 {
4813         int error = 0;
4814         journal_t *journal;
4815
4816         if (sb->s_flags & MS_RDONLY)
4817                 return 0;
4818
4819         journal = EXT4_SB(sb)->s_journal;
4820
4821         if (journal) {
4822                 /* Now we set up the journal barrier. */
4823                 jbd2_journal_lock_updates(journal);
4824
4825                 /*
4826                  * Don't clear the needs_recovery flag if we failed to
4827                  * flush the journal.
4828                  */
4829                 error = jbd2_journal_flush(journal);
4830                 if (error < 0)
4831                         goto out;
4832
4833                 /* Journal blocked and flushed, clear needs_recovery flag. */
4834                 ext4_clear_feature_journal_needs_recovery(sb);
4835         }
4836
4837         error = ext4_commit_super(sb, 1);
4838 out:
4839         if (journal)
4840                 /* we rely on upper layer to stop further updates */
4841                 jbd2_journal_unlock_updates(journal);
4842         return error;
4843 }
4844
4845 /*
4846  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4847  * flag here, even though the filesystem is not technically dirty yet.
4848  */
4849 static int ext4_unfreeze(struct super_block *sb)
4850 {
4851         if ((sb->s_flags & MS_RDONLY) || ext4_forced_shutdown(EXT4_SB(sb)))
4852                 return 0;
4853
4854         if (EXT4_SB(sb)->s_journal) {
4855                 /* Reset the needs_recovery flag before the fs is unlocked. */
4856                 ext4_set_feature_journal_needs_recovery(sb);
4857         }
4858
4859         ext4_commit_super(sb, 1);
4860         return 0;
4861 }
4862
4863 /*
4864  * Structure to save mount options for ext4_remount's benefit
4865  */
4866 struct ext4_mount_options {
4867         unsigned long s_mount_opt;
4868         unsigned long s_mount_opt2;
4869         kuid_t s_resuid;
4870         kgid_t s_resgid;
4871         unsigned long s_commit_interval;
4872         u32 s_min_batch_time, s_max_batch_time;
4873 #ifdef CONFIG_QUOTA
4874         int s_jquota_fmt;
4875         char *s_qf_names[EXT4_MAXQUOTAS];
4876 #endif
4877 };
4878
4879 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4880 {
4881         struct ext4_super_block *es;
4882         struct ext4_sb_info *sbi = EXT4_SB(sb);
4883         unsigned long old_sb_flags;
4884         struct ext4_mount_options old_opts;
4885         int enable_quota = 0;
4886         ext4_group_t g;
4887         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4888         int err = 0;
4889 #ifdef CONFIG_QUOTA
4890         int i, j;
4891 #endif
4892         char *orig_data = kstrdup(data, GFP_KERNEL);
4893
4894         /* Store the original options */
4895         old_sb_flags = sb->s_flags;
4896         old_opts.s_mount_opt = sbi->s_mount_opt;
4897         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4898         old_opts.s_resuid = sbi->s_resuid;
4899         old_opts.s_resgid = sbi->s_resgid;
4900         old_opts.s_commit_interval = sbi->s_commit_interval;
4901         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4902         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4903 #ifdef CONFIG_QUOTA
4904         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4905         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4906                 if (sbi->s_qf_names[i]) {
4907                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4908                                                          GFP_KERNEL);
4909                         if (!old_opts.s_qf_names[i]) {
4910                                 for (j = 0; j < i; j++)
4911                                         kfree(old_opts.s_qf_names[j]);
4912                                 kfree(orig_data);
4913                                 return -ENOMEM;
4914                         }
4915                 } else
4916                         old_opts.s_qf_names[i] = NULL;
4917 #endif
4918         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4919                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4920
4921         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4922                 err = -EINVAL;
4923                 goto restore_opts;
4924         }
4925
4926         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4927             test_opt(sb, JOURNAL_CHECKSUM)) {
4928                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4929                          "during remount not supported; ignoring");
4930                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4931         }
4932
4933         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4934                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4935                         ext4_msg(sb, KERN_ERR, "can't mount with "
4936                                  "both data=journal and delalloc");
4937                         err = -EINVAL;
4938                         goto restore_opts;
4939                 }
4940                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4941                         ext4_msg(sb, KERN_ERR, "can't mount with "
4942                                  "both data=journal and dioread_nolock");
4943                         err = -EINVAL;
4944                         goto restore_opts;
4945                 }
4946                 if (test_opt(sb, DAX)) {
4947                         ext4_msg(sb, KERN_ERR, "can't mount with "
4948                                  "both data=journal and dax");
4949                         err = -EINVAL;
4950                         goto restore_opts;
4951                 }
4952         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
4953                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4954                         ext4_msg(sb, KERN_ERR, "can't mount with "
4955                                 "journal_async_commit in data=ordered mode");
4956                         err = -EINVAL;
4957                         goto restore_opts;
4958                 }
4959         }
4960
4961         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4962                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4963                         "dax flag with busy inodes while remounting");
4964                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4965         }
4966
4967         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4968                 ext4_abort(sb, "Abort forced by user");
4969
4970         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4971                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4972
4973         es = sbi->s_es;
4974
4975         if (sbi->s_journal) {
4976                 ext4_init_journal_params(sb, sbi->s_journal);
4977                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4978         }
4979
4980         if (*flags & MS_LAZYTIME)
4981                 sb->s_flags |= MS_LAZYTIME;
4982
4983         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4984                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4985                         err = -EROFS;
4986                         goto restore_opts;
4987                 }
4988
4989                 if (*flags & MS_RDONLY) {
4990                         err = sync_filesystem(sb);
4991                         if (err < 0)
4992                                 goto restore_opts;
4993                         err = dquot_suspend(sb, -1);
4994                         if (err < 0)
4995                                 goto restore_opts;
4996
4997                         /*
4998                          * First of all, the unconditional stuff we have to do
4999                          * to disable replay of the journal when we next remount
5000                          */
5001                         sb->s_flags |= MS_RDONLY;
5002
5003                         /*
5004                          * OK, test if we are remounting a valid rw partition
5005                          * readonly, and if so set the rdonly flag and then
5006                          * mark the partition as valid again.
5007                          */
5008                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5009                             (sbi->s_mount_state & EXT4_VALID_FS))
5010                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5011
5012                         if (sbi->s_journal)
5013                                 ext4_mark_recovery_complete(sb, es);
5014                 } else {
5015                         /* Make sure we can mount this feature set readwrite */
5016                         if (ext4_has_feature_readonly(sb) ||
5017                             !ext4_feature_set_ok(sb, 0)) {
5018                                 err = -EROFS;
5019                                 goto restore_opts;
5020                         }
5021                         /*
5022                          * Make sure the group descriptor checksums
5023                          * are sane.  If they aren't, refuse to remount r/w.
5024                          */
5025                         for (g = 0; g < sbi->s_groups_count; g++) {
5026                                 struct ext4_group_desc *gdp =
5027                                         ext4_get_group_desc(sb, g, NULL);
5028
5029                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5030                                         ext4_msg(sb, KERN_ERR,
5031                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5032                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5033                                                le16_to_cpu(gdp->bg_checksum));
5034                                         err = -EFSBADCRC;
5035                                         goto restore_opts;
5036                                 }
5037                         }
5038
5039                         /*
5040                          * If we have an unprocessed orphan list hanging
5041                          * around from a previously readonly bdev mount,
5042                          * require a full umount/remount for now.
5043                          */
5044                         if (es->s_last_orphan) {
5045                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5046                                        "remount RDWR because of unprocessed "
5047                                        "orphan inode list.  Please "
5048                                        "umount/remount instead");
5049                                 err = -EINVAL;
5050                                 goto restore_opts;
5051                         }
5052
5053                         /*
5054                          * Mounting a RDONLY partition read-write, so reread
5055                          * and store the current valid flag.  (It may have
5056                          * been changed by e2fsck since we originally mounted
5057                          * the partition.)
5058                          */
5059                         if (sbi->s_journal)
5060                                 ext4_clear_journal_err(sb, es);
5061                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5062                         if (!ext4_setup_super(sb, es, 0))
5063                                 sb->s_flags &= ~MS_RDONLY;
5064                         if (ext4_has_feature_mmp(sb))
5065                                 if (ext4_multi_mount_protect(sb,
5066                                                 le64_to_cpu(es->s_mmp_block))) {
5067                                         err = -EROFS;
5068                                         goto restore_opts;
5069                                 }
5070                         enable_quota = 1;
5071                 }
5072         }
5073
5074         /*
5075          * Reinitialize lazy itable initialization thread based on
5076          * current settings
5077          */
5078         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5079                 ext4_unregister_li_request(sb);
5080         else {
5081                 ext4_group_t first_not_zeroed;
5082                 first_not_zeroed = ext4_has_uninit_itable(sb);
5083                 ext4_register_li_request(sb, first_not_zeroed);
5084         }
5085
5086         ext4_setup_system_zone(sb);
5087         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5088                 ext4_commit_super(sb, 1);
5089
5090 #ifdef CONFIG_QUOTA
5091         /* Release old quota file names */
5092         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5093                 kfree(old_opts.s_qf_names[i]);
5094         if (enable_quota) {
5095                 if (sb_any_quota_suspended(sb))
5096                         dquot_resume(sb, -1);
5097                 else if (ext4_has_feature_quota(sb)) {
5098                         err = ext4_enable_quotas(sb);
5099                         if (err)
5100                                 goto restore_opts;
5101                 }
5102         }
5103 #endif
5104
5105         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5106         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5107         kfree(orig_data);
5108         return 0;
5109
5110 restore_opts:
5111         sb->s_flags = old_sb_flags;
5112         sbi->s_mount_opt = old_opts.s_mount_opt;
5113         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5114         sbi->s_resuid = old_opts.s_resuid;
5115         sbi->s_resgid = old_opts.s_resgid;
5116         sbi->s_commit_interval = old_opts.s_commit_interval;
5117         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5118         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5119 #ifdef CONFIG_QUOTA
5120         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5121         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5122                 kfree(sbi->s_qf_names[i]);
5123                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5124         }
5125 #endif
5126         kfree(orig_data);
5127         return err;
5128 }
5129
5130 #ifdef CONFIG_QUOTA
5131 static int ext4_statfs_project(struct super_block *sb,
5132                                kprojid_t projid, struct kstatfs *buf)
5133 {
5134         struct kqid qid;
5135         struct dquot *dquot;
5136         u64 limit;
5137         u64 curblock;
5138
5139         qid = make_kqid_projid(projid);
5140         dquot = dqget(sb, qid);
5141         if (IS_ERR(dquot))
5142                 return PTR_ERR(dquot);
5143         spin_lock(&dq_data_lock);
5144
5145         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5146                  dquot->dq_dqb.dqb_bsoftlimit :
5147                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5148         if (limit && buf->f_blocks > limit) {
5149                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5150                 buf->f_blocks = limit;
5151                 buf->f_bfree = buf->f_bavail =
5152                         (buf->f_blocks > curblock) ?
5153                          (buf->f_blocks - curblock) : 0;
5154         }
5155
5156         limit = dquot->dq_dqb.dqb_isoftlimit ?
5157                 dquot->dq_dqb.dqb_isoftlimit :
5158                 dquot->dq_dqb.dqb_ihardlimit;
5159         if (limit && buf->f_files > limit) {
5160                 buf->f_files = limit;
5161                 buf->f_ffree =
5162                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5163                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5164         }
5165
5166         spin_unlock(&dq_data_lock);
5167         dqput(dquot);
5168         return 0;
5169 }
5170 #endif
5171
5172 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5173 {
5174         struct super_block *sb = dentry->d_sb;
5175         struct ext4_sb_info *sbi = EXT4_SB(sb);
5176         struct ext4_super_block *es = sbi->s_es;
5177         ext4_fsblk_t overhead = 0, resv_blocks;
5178         u64 fsid;
5179         s64 bfree;
5180         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5181
5182         if (!test_opt(sb, MINIX_DF))
5183                 overhead = sbi->s_overhead;
5184
5185         buf->f_type = EXT4_SUPER_MAGIC;
5186         buf->f_bsize = sb->s_blocksize;
5187         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5188         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5189                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5190         /* prevent underflow in case that few free space is available */
5191         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5192         buf->f_bavail = buf->f_bfree -
5193                         (ext4_r_blocks_count(es) + resv_blocks);
5194         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5195                 buf->f_bavail = 0;
5196         buf->f_files = le32_to_cpu(es->s_inodes_count);
5197         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5198         buf->f_namelen = EXT4_NAME_LEN;
5199         fsid = le64_to_cpup((void *)es->s_uuid) ^
5200                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5201         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5202         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5203
5204 #ifdef CONFIG_QUOTA
5205         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5206             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5207                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5208 #endif
5209         return 0;
5210 }
5211
5212 /* Helper function for writing quotas on sync - we need to start transaction
5213  * before quota file is locked for write. Otherwise the are possible deadlocks:
5214  * Process 1                         Process 2
5215  * ext4_create()                     quota_sync()
5216  *   jbd2_journal_start()                  write_dquot()
5217  *   dquot_initialize()                         down(dqio_mutex)
5218  *     down(dqio_mutex)                    jbd2_journal_start()
5219  *
5220  */
5221
5222 #ifdef CONFIG_QUOTA
5223
5224 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5225 {
5226         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5227 }
5228
5229 static int ext4_write_dquot(struct dquot *dquot)
5230 {
5231         int ret, err;
5232         handle_t *handle;
5233         struct inode *inode;
5234
5235         inode = dquot_to_inode(dquot);
5236         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5237                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5238         if (IS_ERR(handle))
5239                 return PTR_ERR(handle);
5240         ret = dquot_commit(dquot);
5241         err = ext4_journal_stop(handle);
5242         if (!ret)
5243                 ret = err;
5244         return ret;
5245 }
5246
5247 static int ext4_acquire_dquot(struct dquot *dquot)
5248 {
5249         int ret, err;
5250         handle_t *handle;
5251
5252         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5253                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5254         if (IS_ERR(handle))
5255                 return PTR_ERR(handle);
5256         ret = dquot_acquire(dquot);
5257         err = ext4_journal_stop(handle);
5258         if (!ret)
5259                 ret = err;
5260         return ret;
5261 }
5262
5263 static int ext4_release_dquot(struct dquot *dquot)
5264 {
5265         int ret, err;
5266         handle_t *handle;
5267
5268         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5269                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5270         if (IS_ERR(handle)) {
5271                 /* Release dquot anyway to avoid endless cycle in dqput() */
5272                 dquot_release(dquot);
5273                 return PTR_ERR(handle);
5274         }
5275         ret = dquot_release(dquot);
5276         err = ext4_journal_stop(handle);
5277         if (!ret)
5278                 ret = err;
5279         return ret;
5280 }
5281
5282 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5283 {
5284         struct super_block *sb = dquot->dq_sb;
5285         struct ext4_sb_info *sbi = EXT4_SB(sb);
5286
5287         /* Are we journaling quotas? */
5288         if (ext4_has_feature_quota(sb) ||
5289             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5290                 dquot_mark_dquot_dirty(dquot);
5291                 return ext4_write_dquot(dquot);
5292         } else {
5293                 return dquot_mark_dquot_dirty(dquot);
5294         }
5295 }
5296
5297 static int ext4_write_info(struct super_block *sb, int type)
5298 {
5299         int ret, err;
5300         handle_t *handle;
5301
5302         /* Data block + inode block */
5303         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5304         if (IS_ERR(handle))
5305                 return PTR_ERR(handle);
5306         ret = dquot_commit_info(sb, type);
5307         err = ext4_journal_stop(handle);
5308         if (!ret)
5309                 ret = err;
5310         return ret;
5311 }
5312
5313 /*
5314  * Turn on quotas during mount time - we need to find
5315  * the quota file and such...
5316  */
5317 static int ext4_quota_on_mount(struct super_block *sb, int type)
5318 {
5319         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5320                                         EXT4_SB(sb)->s_jquota_fmt, type);
5321 }
5322
5323 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5324 {
5325         struct ext4_inode_info *ei = EXT4_I(inode);
5326
5327         /* The first argument of lockdep_set_subclass has to be
5328          * *exactly* the same as the argument to init_rwsem() --- in
5329          * this case, in init_once() --- or lockdep gets unhappy
5330          * because the name of the lock is set using the
5331          * stringification of the argument to init_rwsem().
5332          */
5333         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5334         lockdep_set_subclass(&ei->i_data_sem, subclass);
5335 }
5336
5337 /*
5338  * Standard function to be called on quota_on
5339  */
5340 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5341                          const struct path *path)
5342 {
5343         int err;
5344
5345         if (!test_opt(sb, QUOTA))
5346                 return -EINVAL;
5347
5348         /* Quotafile not on the same filesystem? */
5349         if (path->dentry->d_sb != sb)
5350                 return -EXDEV;
5351         /* Journaling quota? */
5352         if (EXT4_SB(sb)->s_qf_names[type]) {
5353                 /* Quotafile not in fs root? */
5354                 if (path->dentry->d_parent != sb->s_root)
5355                         ext4_msg(sb, KERN_WARNING,
5356                                 "Quota file not on filesystem root. "
5357                                 "Journaled quota will not work");
5358         }
5359
5360         /*
5361          * When we journal data on quota file, we have to flush journal to see
5362          * all updates to the file when we bypass pagecache...
5363          */
5364         if (EXT4_SB(sb)->s_journal &&
5365             ext4_should_journal_data(d_inode(path->dentry))) {
5366                 /*
5367                  * We don't need to lock updates but journal_flush() could
5368                  * otherwise be livelocked...
5369                  */
5370                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5371                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5372                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5373                 if (err)
5374                         return err;
5375         }
5376
5377         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5378         err = dquot_quota_on(sb, type, format_id, path);
5379         if (err) {
5380                 lockdep_set_quota_inode(path->dentry->d_inode,
5381                                              I_DATA_SEM_NORMAL);
5382         } else {
5383                 struct inode *inode = d_inode(path->dentry);
5384                 handle_t *handle;
5385
5386                 /*
5387                  * Set inode flags to prevent userspace from messing with quota
5388                  * files. If this fails, we return success anyway since quotas
5389                  * are already enabled and this is not a hard failure.
5390                  */
5391                 inode_lock(inode);
5392                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5393                 if (IS_ERR(handle))
5394                         goto unlock_inode;
5395                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5396                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5397                                 S_NOATIME | S_IMMUTABLE);
5398                 ext4_mark_inode_dirty(handle, inode);
5399                 ext4_journal_stop(handle);
5400         unlock_inode:
5401                 inode_unlock(inode);
5402         }
5403         return err;
5404 }
5405
5406 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5407                              unsigned int flags)
5408 {
5409         int err;
5410         struct inode *qf_inode;
5411         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5412                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5413                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5414                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5415         };
5416
5417         BUG_ON(!ext4_has_feature_quota(sb));
5418
5419         if (!qf_inums[type])
5420                 return -EPERM;
5421
5422         qf_inode = ext4_iget(sb, qf_inums[type]);
5423         if (IS_ERR(qf_inode)) {
5424                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5425                 return PTR_ERR(qf_inode);
5426         }
5427
5428         /* Don't account quota for quota files to avoid recursion */
5429         qf_inode->i_flags |= S_NOQUOTA;
5430         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5431         err = dquot_enable(qf_inode, type, format_id, flags);
5432         iput(qf_inode);
5433         if (err)
5434                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5435
5436         return err;
5437 }
5438
5439 /* Enable usage tracking for all quota types. */
5440 static int ext4_enable_quotas(struct super_block *sb)
5441 {
5442         int type, err = 0;
5443         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5444                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5445                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5446                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5447         };
5448         bool quota_mopt[EXT4_MAXQUOTAS] = {
5449                 test_opt(sb, USRQUOTA),
5450                 test_opt(sb, GRPQUOTA),
5451                 test_opt(sb, PRJQUOTA),
5452         };
5453
5454         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5455         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5456                 if (qf_inums[type]) {
5457                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5458                                 DQUOT_USAGE_ENABLED |
5459                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5460                         if (err) {
5461                                 ext4_warning(sb,
5462                                         "Failed to enable quota tracking "
5463                                         "(type=%d, err=%d). Please run "
5464                                         "e2fsck to fix.", type, err);
5465                                 return err;
5466                         }
5467                 }
5468         }
5469         return 0;
5470 }
5471
5472 static int ext4_quota_off(struct super_block *sb, int type)
5473 {
5474         struct inode *inode = sb_dqopt(sb)->files[type];
5475         handle_t *handle;
5476         int err;
5477
5478         /* Force all delayed allocation blocks to be allocated.
5479          * Caller already holds s_umount sem */
5480         if (test_opt(sb, DELALLOC))
5481                 sync_filesystem(sb);
5482
5483         if (!inode || !igrab(inode))
5484                 goto out;
5485
5486         err = dquot_quota_off(sb, type);
5487         if (err)
5488                 goto out_put;
5489
5490         inode_lock(inode);
5491         /*
5492          * Update modification times of quota files when userspace can
5493          * start looking at them. If we fail, we return success anyway since
5494          * this is not a hard failure and quotas are already disabled.
5495          */
5496         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5497         if (IS_ERR(handle))
5498                 goto out_unlock;
5499         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5500         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5501         inode->i_mtime = inode->i_ctime = current_time(inode);
5502         ext4_mark_inode_dirty(handle, inode);
5503         ext4_journal_stop(handle);
5504 out_unlock:
5505         inode_unlock(inode);
5506 out_put:
5507         iput(inode);
5508         return err;
5509 out:
5510         return dquot_quota_off(sb, type);
5511 }
5512
5513 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5514  * acquiring the locks... As quota files are never truncated and quota code
5515  * itself serializes the operations (and no one else should touch the files)
5516  * we don't have to be afraid of races */
5517 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5518                                size_t len, loff_t off)
5519 {
5520         struct inode *inode = sb_dqopt(sb)->files[type];
5521         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5522         int offset = off & (sb->s_blocksize - 1);
5523         int tocopy;
5524         size_t toread;
5525         struct buffer_head *bh;
5526         loff_t i_size = i_size_read(inode);
5527
5528         if (off > i_size)
5529                 return 0;
5530         if (off+len > i_size)
5531                 len = i_size-off;
5532         toread = len;
5533         while (toread > 0) {
5534                 tocopy = sb->s_blocksize - offset < toread ?
5535                                 sb->s_blocksize - offset : toread;
5536                 bh = ext4_bread(NULL, inode, blk, 0);
5537                 if (IS_ERR(bh))
5538                         return PTR_ERR(bh);
5539                 if (!bh)        /* A hole? */
5540                         memset(data, 0, tocopy);
5541                 else
5542                         memcpy(data, bh->b_data+offset, tocopy);
5543                 brelse(bh);
5544                 offset = 0;
5545                 toread -= tocopy;
5546                 data += tocopy;
5547                 blk++;
5548         }
5549         return len;
5550 }
5551
5552 /* Write to quotafile (we know the transaction is already started and has
5553  * enough credits) */
5554 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5555                                 const char *data, size_t len, loff_t off)
5556 {
5557         struct inode *inode = sb_dqopt(sb)->files[type];
5558         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5559         int err, offset = off & (sb->s_blocksize - 1);
5560         int retries = 0;
5561         struct buffer_head *bh;
5562         handle_t *handle = journal_current_handle();
5563
5564         if (EXT4_SB(sb)->s_journal && !handle) {
5565                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5566                         " cancelled because transaction is not started",
5567                         (unsigned long long)off, (unsigned long long)len);
5568                 return -EIO;
5569         }
5570         /*
5571          * Since we account only one data block in transaction credits,
5572          * then it is impossible to cross a block boundary.
5573          */
5574         if (sb->s_blocksize - offset < len) {
5575                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5576                         " cancelled because not block aligned",
5577                         (unsigned long long)off, (unsigned long long)len);
5578                 return -EIO;
5579         }
5580
5581         do {
5582                 bh = ext4_bread(handle, inode, blk,
5583                                 EXT4_GET_BLOCKS_CREATE |
5584                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5585         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5586                  ext4_should_retry_alloc(inode->i_sb, &retries));
5587         if (IS_ERR(bh))
5588                 return PTR_ERR(bh);
5589         if (!bh)
5590                 goto out;
5591         BUFFER_TRACE(bh, "get write access");
5592         err = ext4_journal_get_write_access(handle, bh);
5593         if (err) {
5594                 brelse(bh);
5595                 return err;
5596         }
5597         lock_buffer(bh);
5598         memcpy(bh->b_data+offset, data, len);
5599         flush_dcache_page(bh->b_page);
5600         unlock_buffer(bh);
5601         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5602         brelse(bh);
5603 out:
5604         if (inode->i_size < off + len) {
5605                 i_size_write(inode, off + len);
5606                 EXT4_I(inode)->i_disksize = inode->i_size;
5607                 ext4_mark_inode_dirty(handle, inode);
5608         }
5609         return len;
5610 }
5611
5612 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5613 {
5614         const struct quota_format_ops   *ops;
5615
5616         if (!sb_has_quota_loaded(sb, qid->type))
5617                 return -ESRCH;
5618         ops = sb_dqopt(sb)->ops[qid->type];
5619         if (!ops || !ops->get_next_id)
5620                 return -ENOSYS;
5621         return dquot_get_next_id(sb, qid);
5622 }
5623 #endif
5624
5625 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5626                        const char *dev_name, void *data)
5627 {
5628         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5629 }
5630
5631 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5632 static inline void register_as_ext2(void)
5633 {
5634         int err = register_filesystem(&ext2_fs_type);
5635         if (err)
5636                 printk(KERN_WARNING
5637                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5638 }
5639
5640 static inline void unregister_as_ext2(void)
5641 {
5642         unregister_filesystem(&ext2_fs_type);
5643 }
5644
5645 static inline int ext2_feature_set_ok(struct super_block *sb)
5646 {
5647         if (ext4_has_unknown_ext2_incompat_features(sb))
5648                 return 0;
5649         if (sb->s_flags & MS_RDONLY)
5650                 return 1;
5651         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5652                 return 0;
5653         return 1;
5654 }
5655 #else
5656 static inline void register_as_ext2(void) { }
5657 static inline void unregister_as_ext2(void) { }
5658 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5659 #endif
5660
5661 static inline void register_as_ext3(void)
5662 {
5663         int err = register_filesystem(&ext3_fs_type);
5664         if (err)
5665                 printk(KERN_WARNING
5666                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5667 }
5668
5669 static inline void unregister_as_ext3(void)
5670 {
5671         unregister_filesystem(&ext3_fs_type);
5672 }
5673
5674 static inline int ext3_feature_set_ok(struct super_block *sb)
5675 {
5676         if (ext4_has_unknown_ext3_incompat_features(sb))
5677                 return 0;
5678         if (!ext4_has_feature_journal(sb))
5679                 return 0;
5680         if (sb->s_flags & MS_RDONLY)
5681                 return 1;
5682         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5683                 return 0;
5684         return 1;
5685 }
5686
5687 static struct file_system_type ext4_fs_type = {
5688         .owner          = THIS_MODULE,
5689         .name           = "ext4",
5690         .mount          = ext4_mount,
5691         .kill_sb        = kill_block_super,
5692         .fs_flags       = FS_REQUIRES_DEV,
5693 };
5694 MODULE_ALIAS_FS("ext4");
5695
5696 /* Shared across all ext4 file systems */
5697 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5698
5699 static int __init ext4_init_fs(void)
5700 {
5701         int i, err;
5702
5703         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5704         ext4_li_info = NULL;
5705         mutex_init(&ext4_li_mtx);
5706
5707         /* Build-time check for flags consistency */
5708         ext4_check_flag_values();
5709
5710         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5711                 init_waitqueue_head(&ext4__ioend_wq[i]);
5712
5713         err = ext4_init_es();
5714         if (err)
5715                 return err;
5716
5717         err = ext4_init_pageio();
5718         if (err)
5719                 goto out5;
5720
5721         err = ext4_init_system_zone();
5722         if (err)
5723                 goto out4;
5724
5725         err = ext4_init_sysfs();
5726         if (err)
5727                 goto out3;
5728
5729         err = ext4_init_mballoc();
5730         if (err)
5731                 goto out2;
5732         err = init_inodecache();
5733         if (err)
5734                 goto out1;
5735         register_as_ext3();
5736         register_as_ext2();
5737         err = register_filesystem(&ext4_fs_type);
5738         if (err)
5739                 goto out;
5740
5741         return 0;
5742 out:
5743         unregister_as_ext2();
5744         unregister_as_ext3();
5745         destroy_inodecache();
5746 out1:
5747         ext4_exit_mballoc();
5748 out2:
5749         ext4_exit_sysfs();
5750 out3:
5751         ext4_exit_system_zone();
5752 out4:
5753         ext4_exit_pageio();
5754 out5:
5755         ext4_exit_es();
5756
5757         return err;
5758 }
5759
5760 static void __exit ext4_exit_fs(void)
5761 {
5762         ext4_destroy_lazyinit_thread();
5763         unregister_as_ext2();
5764         unregister_as_ext3();
5765         unregister_filesystem(&ext4_fs_type);
5766         destroy_inodecache();
5767         ext4_exit_mballoc();
5768         ext4_exit_sysfs();
5769         ext4_exit_system_zone();
5770         ext4_exit_pageio();
5771         ext4_exit_es();
5772 }
5773
5774 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5775 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5776 MODULE_LICENSE("GPL");
5777 module_init(ext4_init_fs)
5778 module_exit(ext4_exit_fs)