fs/f2fs/super.c

   1 /*
   2  * fs/f2fs/super.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/module.h>
  12 #include <linux/init.h>
  13 #include <linux/fs.h>
  14 #include <linux/statfs.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/kthread.h>
  18 #include <linux/parser.h>
  19 #include <linux/mount.h>
  20 #include <linux/seq_file.h>
  21 #include <linux/proc_fs.h>
  22 #include <linux/random.h>
  23 #include <linux/exportfs.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/f2fs_fs.h>
  26 #include <linux/sysfs.h>
  27
  28 #include "f2fs.h"
  29 #include "node.h"
  30 #include "segment.h"
  31 #include "xattr.h"
  32 #include "gc.h"
  33
  34 #define CREATE_TRACE_POINTS
  35 #include <trace/events/f2fs.h>
  36
  37 static struct proc_dir_entry *f2fs_proc_root;
  38 static struct kmem_cache *f2fs_inode_cachep;
  39 static struct kset *f2fs_kset;
  40
  41 enum {
  42         Opt_gc_background,
  43         Opt_disable_roll_forward,
  44         Opt_discard,
  45         Opt_noheap,
  46         Opt_user_xattr,
  47         Opt_nouser_xattr,
  48         Opt_acl,
  49         Opt_noacl,
  50         Opt_active_logs,
  51         Opt_disable_ext_identify,
  52         Opt_inline_xattr,
  53         Opt_inline_data,
  54         Opt_err,
  55 };
  56
  57 static match_table_t f2fs_tokens = {
  58         {Opt_gc_background, "background_gc=%s"},
  59         {Opt_disable_roll_forward, "disable_roll_forward"},
  60         {Opt_discard, "discard"},
  61         {Opt_noheap, "no_heap"},
  62         {Opt_user_xattr, "user_xattr"},
  63         {Opt_nouser_xattr, "nouser_xattr"},
  64         {Opt_acl, "acl"},
  65         {Opt_noacl, "noacl"},
  66         {Opt_active_logs, "active_logs=%u"},
  67         {Opt_disable_ext_identify, "disable_ext_identify"},
  68         {Opt_inline_xattr, "inline_xattr"},
  69         {Opt_inline_data, "inline_data"},
  70         {Opt_err, NULL},
  71 };
  72
  73 /* Sysfs support for f2fs */
  74 enum {
  75         GC_THREAD,      /* struct f2fs_gc_thread */
  76         SM_INFO,        /* struct f2fs_sm_info */
  77 };
  78
  79 struct f2fs_attr {
  80         struct attribute attr;
  81         ssize_t (*show)(struct f2fs_attr *, struct f2fs_sb_info *, char *);
  82         ssize_t (*store)(struct f2fs_attr *, struct f2fs_sb_info *,
  83                          const char *, size_t);
  84         int struct_type;
  85         int offset;
  86 };
  87
  88 static unsigned char *__struct_ptr(struct f2fs_sb_info *sbi, int struct_type)
  89 {
  90         if (struct_type == GC_THREAD)
  91                 return (unsigned char *)sbi->gc_thread;
  92         else if (struct_type == SM_INFO)
  93                 return (unsigned char *)SM_I(sbi);
  94         return NULL;
  95 }
  96
  97 static ssize_t f2fs_sbi_show(struct f2fs_attr *a,
  98                         struct f2fs_sb_info *sbi, char *buf)
  99 {
 100         unsigned char *ptr = NULL;
 101         unsigned int *ui;
 102
 103         ptr = __struct_ptr(sbi, a->struct_type);
 104         if (!ptr)
 105                 return -EINVAL;
 106
 107         ui = (unsigned int *)(ptr + a->offset);
 108
 109         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
 110 }
 111
 112 static ssize_t f2fs_sbi_store(struct f2fs_attr *a,
 113                         struct f2fs_sb_info *sbi,
 114                         const char *buf, size_t count)
 115 {
 116         unsigned char *ptr;
 117         unsigned long t;
 118         unsigned int *ui;
 119         ssize_t ret;
 120
 121         ptr = __struct_ptr(sbi, a->struct_type);
 122         if (!ptr)
 123                 return -EINVAL;
 124
 125         ui = (unsigned int *)(ptr + a->offset);
 126
 127         ret = kstrtoul(skip_spaces(buf), 0, &t);
 128         if (ret < 0)
 129                 return ret;
 130         *ui = t;
 131         return count;
 132 }
 133
 134 static ssize_t f2fs_attr_show(struct kobject *kobj,
 135                                 struct attribute *attr, char *buf)
 136 {
 137         struct f2fs_sb_info *sbi = container_of(kobj, struct f2fs_sb_info,
 138                                                                 s_kobj);
 139         struct f2fs_attr *a = container_of(attr, struct f2fs_attr, attr);
 140
 141         return a->show ? a->show(a, sbi, buf) : 0;
 142 }
 143
 144 static ssize_t f2fs_attr_store(struct kobject *kobj, struct attribute *attr,
 145                                                 const char *buf, size_t len)
 146 {
 147         struct f2fs_sb_info *sbi = container_of(kobj, struct f2fs_sb_info,
 148                                                                         s_kobj);
 149         struct f2fs_attr *a = container_of(attr, struct f2fs_attr, attr);
 150
 151         return a->store ? a->store(a, sbi, buf, len) : 0;
 152 }
 153
 154 static void f2fs_sb_release(struct kobject *kobj)
 155 {
 156         struct f2fs_sb_info *sbi = container_of(kobj, struct f2fs_sb_info,
 157                                                                 s_kobj);
 158         complete(&sbi->s_kobj_unregister);
 159 }
 160
 161 #define F2FS_ATTR_OFFSET(_struct_type, _name, _mode, _show, _store, _offset) \
 162 static struct f2fs_attr f2fs_attr_##_name = {                   \
 163         .attr = {.name = __stringify(_name), .mode = _mode },   \
 164         .show   = _show,                                        \
 165         .store  = _store,                                       \
 166         .struct_type = _struct_type,                            \
 167         .offset = _offset                                       \
 168 }
 169
 170 #define F2FS_RW_ATTR(struct_type, struct_name, name, elname)    \
 171         F2FS_ATTR_OFFSET(struct_type, name, 0644,               \
 172                 f2fs_sbi_show, f2fs_sbi_store,                  \
 173                 offsetof(struct struct_name, elname))
 174
 175 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_min_sleep_time, min_sleep_time);
 176 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_max_sleep_time, max_sleep_time);
 177 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_no_gc_sleep_time, no_gc_sleep_time);
 178 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_idle, gc_idle);
 179 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, reclaim_segments, rec_prefree_segments);
 180 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, max_small_discards, max_discards);
 181
 182 #define ATTR_LIST(name) (&f2fs_attr_##name.attr)
 183 static struct attribute *f2fs_attrs[] = {
 184         ATTR_LIST(gc_min_sleep_time),
 185         ATTR_LIST(gc_max_sleep_time),
 186         ATTR_LIST(gc_no_gc_sleep_time),
 187         ATTR_LIST(gc_idle),
 188         ATTR_LIST(reclaim_segments),
 189         ATTR_LIST(max_small_discards),
 190         NULL,
 191 };
 192
 193 static const struct sysfs_ops f2fs_attr_ops = {
 194         .show   = f2fs_attr_show,
 195         .store  = f2fs_attr_store,
 196 };
 197
 198 static struct kobj_type f2fs_ktype = {
 199         .default_attrs  = f2fs_attrs,
 200         .sysfs_ops      = &f2fs_attr_ops,
 201         .release        = f2fs_sb_release,
 202 };
 203
 204 void f2fs_msg(struct super_block *sb, const char *level, const char *fmt, ...)
 205 {
 206         struct va_format vaf;
 207         va_list args;
 208
 209         va_start(args, fmt);
 210         vaf.fmt = fmt;
 211         vaf.va = &args;
 212         printk("%sF2FS-fs (%s): %pV\n", level, sb->s_id, &vaf);
 213         va_end(args);
 214 }
 215
 216 static void init_once(void *foo)
 217 {
 218         struct f2fs_inode_info *fi = (struct f2fs_inode_info *) foo;
 219
 220         inode_init_once(&fi->vfs_inode);
 221 }
 222
 223 static int parse_options(struct super_block *sb, char *options)
 224 {
 225         struct f2fs_sb_info *sbi = F2FS_SB(sb);
 226         substring_t args[MAX_OPT_ARGS];
 227         char *p, *name;
 228         int arg = 0;
 229
 230         if (!options)
 231                 return 0;
 232
 233         while ((p = strsep(&options, ",")) != NULL) {
 234                 int token;
 235                 if (!*p)
 236                         continue;
 237                 /*
 238                  * Initialize args struct so we know whether arg was
 239                  * found; some options take optional arguments.
 240                  */
 241                 args[0].to = args[0].from = NULL;
 242                 token = match_token(p, f2fs_tokens, args);
 243
 244                 switch (token) {
 245                 case Opt_gc_background:
 246                         name = match_strdup(&args[0]);
 247
 248                         if (!name)
 249                                 return -ENOMEM;
 250                         if (!strncmp(name, "on", 2))
 251                                 set_opt(sbi, BG_GC);
 252                         else if (!strncmp(name, "off", 3))
 253                                 clear_opt(sbi, BG_GC);
 254                         else {
 255                                 kfree(name);
 256                                 return -EINVAL;
 257                         }
 258                         kfree(name);
 259                         break;
 260                 case Opt_disable_roll_forward:
 261                         set_opt(sbi, DISABLE_ROLL_FORWARD);
 262                         break;
 263                 case Opt_discard:
 264                         set_opt(sbi, DISCARD);
 265                         break;
 266                 case Opt_noheap:
 267                         set_opt(sbi, NOHEAP);
 268                         break;
 269 #ifdef CONFIG_F2FS_FS_XATTR
 270                 case Opt_user_xattr:
 271                         set_opt(sbi, XATTR_USER);
 272                         break;
 273                 case Opt_nouser_xattr:
 274                         clear_opt(sbi, XATTR_USER);
 275                         break;
 276                 case Opt_inline_xattr:
 277                         set_opt(sbi, INLINE_XATTR);
 278                         break;
 279 #else
 280                 case Opt_user_xattr:
 281                         f2fs_msg(sb, KERN_INFO,
 282                                 "user_xattr options not supported");
 283                         break;
 284                 case Opt_nouser_xattr:
 285                         f2fs_msg(sb, KERN_INFO,
 286                                 "nouser_xattr options not supported");
 287                         break;
 288                 case Opt_inline_xattr:
 289                         f2fs_msg(sb, KERN_INFO,
 290                                 "inline_xattr options not supported");
 291                         break;
 292 #endif
 293 #ifdef CONFIG_F2FS_FS_POSIX_ACL
 294                 case Opt_acl:
 295                         set_opt(sbi, POSIX_ACL);
 296                         break;
 297                 case Opt_noacl:
 298                         clear_opt(sbi, POSIX_ACL);
 299                         break;
 300 #else
 301                 case Opt_acl:
 302                         f2fs_msg(sb, KERN_INFO, "acl options not supported");
 303                         break;
 304                 case Opt_noacl:
 305                         f2fs_msg(sb, KERN_INFO, "noacl options not supported");
 306                         break;
 307 #endif
 308                 case Opt_active_logs:
 309                         if (args->from && match_int(args, &arg))
 310                                 return -EINVAL;
 311                         if (arg != 2 && arg != 4 && arg != NR_CURSEG_TYPE)
 312                                 return -EINVAL;
 313                         sbi->active_logs = arg;
 314                         break;
 315                 case Opt_disable_ext_identify:
 316                         set_opt(sbi, DISABLE_EXT_IDENTIFY);
 317                         break;
 318                 case Opt_inline_data:
 319                         set_opt(sbi, INLINE_DATA);
 320                         break;
 321                 default:
 322                         f2fs_msg(sb, KERN_ERR,
 323                                 "Unrecognized mount option \"%s\" or missing value",
 324                                 p);
 325                         return -EINVAL;
 326                 }
 327         }
 328         return 0;
 329 }
 330
 331 static struct inode *f2fs_alloc_inode(struct super_block *sb)
 332 {
 333         struct f2fs_inode_info *fi;
 334
 335         fi = kmem_cache_alloc(f2fs_inode_cachep, GFP_NOFS | __GFP_ZERO);
 336         if (!fi)
 337                 return NULL;
 338
 339         init_once((void *) fi);
 340
 341         /* Initialize f2fs-specific inode info */
 342         fi->vfs_inode.i_version = 1;
 343         atomic_set(&fi->dirty_dents, 0);
 344         fi->i_current_depth = 1;
 345         fi->i_advise = 0;
 346         rwlock_init(&fi->ext.ext_lock);
 347
 348         set_inode_flag(fi, FI_NEW_INODE);
 349
 350         if (test_opt(F2FS_SB(sb), INLINE_XATTR))
 351                 set_inode_flag(fi, FI_INLINE_XATTR);
 352
 353         return &fi->vfs_inode;
 354 }
 355
 356 static int f2fs_drop_inode(struct inode *inode)
 357 {
 358         /*
 359          * This is to avoid a deadlock condition like below.
 360          * writeback_single_inode(inode)
 361          *  - f2fs_write_data_page
 362          *    - f2fs_gc -> iput -> evict
 363          *       - inode_wait_for_writeback(inode)
 364          */
 365         if (!inode_unhashed(inode) && inode->i_state & I_SYNC)
 366                 return 0;
 367         return generic_drop_inode(inode);
 368 }
 369
 370 /*
 371  * f2fs_dirty_inode() is called from __mark_inode_dirty()
 372  *
 373  * We should call set_dirty_inode to write the dirty inode through write_inode.
 374  */
 375 static void f2fs_dirty_inode(struct inode *inode, int flags)
 376 {
 377         set_inode_flag(F2FS_I(inode), FI_DIRTY_INODE);
 378 }
 379
 380 static void f2fs_i_callback(struct rcu_head *head)
 381 {
 382         struct inode *inode = container_of(head, struct inode, i_rcu);
 383         kmem_cache_free(f2fs_inode_cachep, F2FS_I(inode));
 384 }
 385
 386 static void f2fs_destroy_inode(struct inode *inode)
 387 {
 388         call_rcu(&inode->i_rcu, f2fs_i_callback);
 389 }
 390
 391 static void f2fs_put_super(struct super_block *sb)
 392 {
 393         struct f2fs_sb_info *sbi = F2FS_SB(sb);
 394
 395         if (sbi->s_proc) {
 396                 remove_proc_entry("segment_info", sbi->s_proc);
 397                 remove_proc_entry(sb->s_id, f2fs_proc_root);
 398         }
 399         kobject_del(&sbi->s_kobj);
 400
 401         f2fs_destroy_stats(sbi);
 402         stop_gc_thread(sbi);
 403
 404         /* We don't need to do checkpoint when it's clean */
 405         if (sbi->s_dirty && get_pages(sbi, F2FS_DIRTY_NODES))
 406                 write_checkpoint(sbi, true);
 407
 408         iput(sbi->node_inode);
 409         iput(sbi->meta_inode);
 410
 411         /* destroy f2fs internal modules */
 412         destroy_node_manager(sbi);
 413         destroy_segment_manager(sbi);
 414
 415         kfree(sbi->ckpt);
 416         kobject_put(&sbi->s_kobj);
 417         wait_for_completion(&sbi->s_kobj_unregister);
 418
 419         sb->s_fs_info = NULL;
 420         brelse(sbi->raw_super_buf);
 421         kfree(sbi);
 422 }
 423
 424 int f2fs_sync_fs(struct super_block *sb, int sync)
 425 {
 426         struct f2fs_sb_info *sbi = F2FS_SB(sb);
 427
 428         trace_f2fs_sync_fs(sb, sync);
 429
 430         if (!sbi->s_dirty && !get_pages(sbi, F2FS_DIRTY_NODES))
 431                 return 0;
 432
 433         if (sync) {
 434                 mutex_lock(&sbi->gc_mutex);
 435                 write_checkpoint(sbi, false);
 436                 mutex_unlock(&sbi->gc_mutex);
 437         } else {
 438                 f2fs_balance_fs(sbi);
 439         }
 440
 441         return 0;
 442 }
 443
 444 static int f2fs_freeze(struct super_block *sb)
 445 {
 446         int err;
 447
 448         if (f2fs_readonly(sb))
 449                 return 0;
 450
 451         err = f2fs_sync_fs(sb, 1);
 452         return err;
 453 }
 454
 455 static int f2fs_unfreeze(struct super_block *sb)
 456 {
 457         return 0;
 458 }
 459
 460 static int f2fs_statfs(struct dentry *dentry, struct kstatfs *buf)
 461 {
 462         struct super_block *sb = dentry->d_sb;
 463         struct f2fs_sb_info *sbi = F2FS_SB(sb);
 464         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 465         block_t total_count, user_block_count, start_count, ovp_count;
 466
 467         total_count = le64_to_cpu(sbi->raw_super->block_count);
 468         user_block_count = sbi->user_block_count;
 469         start_count = le32_to_cpu(sbi->raw_super->segment0_blkaddr);
 470         ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg;
 471         buf->f_type = F2FS_SUPER_MAGIC;
 472         buf->f_bsize = sbi->blocksize;
 473
 474         buf->f_blocks = total_count - start_count;
 475         buf->f_bfree = buf->f_blocks - valid_user_blocks(sbi) - ovp_count;
 476         buf->f_bavail = user_block_count - valid_user_blocks(sbi);
 477
 478         buf->f_files = sbi->total_node_count;
 479         buf->f_ffree = sbi->total_node_count - valid_inode_count(sbi);
 480
 481         buf->f_namelen = F2FS_NAME_LEN;
 482         buf->f_fsid.val[0] = (u32)id;
 483         buf->f_fsid.val[1] = (u32)(id >> 32);
 484
 485         return 0;
 486 }
 487
 488 static int f2fs_show_options(struct seq_file *seq, struct dentry *root)
 489 {
 490         struct f2fs_sb_info *sbi = F2FS_SB(root->d_sb);
 491
 492         if (!(root->d_sb->s_flags & MS_RDONLY) && test_opt(sbi, BG_GC))
 493                 seq_printf(seq, ",background_gc=%s", "on");
 494         else
 495                 seq_printf(seq, ",background_gc=%s", "off");
 496         if (test_opt(sbi, DISABLE_ROLL_FORWARD))
 497                 seq_puts(seq, ",disable_roll_forward");
 498         if (test_opt(sbi, DISCARD))
 499                 seq_puts(seq, ",discard");
 500         if (test_opt(sbi, NOHEAP))
 501                 seq_puts(seq, ",no_heap_alloc");
 502 #ifdef CONFIG_F2FS_FS_XATTR
 503         if (test_opt(sbi, XATTR_USER))
 504                 seq_puts(seq, ",user_xattr");
 505         else
 506                 seq_puts(seq, ",nouser_xattr");
 507         if (test_opt(sbi, INLINE_XATTR))
 508                 seq_puts(seq, ",inline_xattr");
 509 #endif
 510 #ifdef CONFIG_F2FS_FS_POSIX_ACL
 511         if (test_opt(sbi, POSIX_ACL))
 512                 seq_puts(seq, ",acl");
 513         else
 514                 seq_puts(seq, ",noacl");
 515 #endif
 516         if (test_opt(sbi, DISABLE_EXT_IDENTIFY))
 517                 seq_puts(seq, ",disable_ext_identify");
 518         if (test_opt(sbi, INLINE_DATA))
 519                 seq_puts(seq, ",inline_data");
 520         seq_printf(seq, ",active_logs=%u", sbi->active_logs);
 521
 522         return 0;
 523 }
 524
 525 static int segment_info_seq_show(struct seq_file *seq, void *offset)
 526 {
 527         struct super_block *sb = seq->private;
 528         struct f2fs_sb_info *sbi = F2FS_SB(sb);
 529         unsigned int total_segs = le32_to_cpu(sbi->raw_super->segment_count_main);
 530         int i;
 531
 532         for (i = 0; i < total_segs; i++) {
 533                 seq_printf(seq, "%u", get_valid_blocks(sbi, i, 1));
 534                 if (i != 0 && (i % 10) == 0)
 535                         seq_puts(seq, "\n");
 536                 else
 537                         seq_puts(seq, " ");
 538         }
 539         return 0;
 540 }
 541
 542 static int segment_info_open_fs(struct inode *inode, struct file *file)
 543 {
 544         return single_open(file, segment_info_seq_show, PDE_DATA(inode));
 545 }
 546
 547 static const struct file_operations f2fs_seq_segment_info_fops = {
 548         .owner = THIS_MODULE,
 549         .open = segment_info_open_fs,
 550         .read = seq_read,
 551         .llseek = seq_lseek,
 552         .release = single_release,
 553 };
 554
 555 static int f2fs_remount(struct super_block *sb, int *flags, char *data)
 556 {
 557         struct f2fs_sb_info *sbi = F2FS_SB(sb);
 558         struct f2fs_mount_info org_mount_opt;
 559         int err, active_logs;
 560
 561         /*
 562          * Save the old mount options in case we
 563          * need to restore them.
 564          */
 565         org_mount_opt = sbi->mount_opt;
 566         active_logs = sbi->active_logs;
 567
 568         /* parse mount options */
 569         err = parse_options(sb, data);
 570         if (err)
 571                 goto restore_opts;
 572
 573         /*
 574          * Previous and new state of filesystem is RO,
 575          * so no point in checking GC conditions.
 576          */
 577         if ((sb->s_flags & MS_RDONLY) && (*flags & MS_RDONLY))
 578                 goto skip;
 579
 580         /*
 581          * We stop the GC thread if FS is mounted as RO
 582          * or if background_gc = off is passed in mount
 583          * option. Also sync the filesystem.
 584          */
 585         if ((*flags & MS_RDONLY) || !test_opt(sbi, BG_GC)) {
 586                 if (sbi->gc_thread) {
 587                         stop_gc_thread(sbi);
 588                         f2fs_sync_fs(sb, 1);
 589                 }
 590         } else if (test_opt(sbi, BG_GC) && !sbi->gc_thread) {
 591                 err = start_gc_thread(sbi);
 592                 if (err)
 593                         goto restore_opts;
 594         }
 595 skip:
 596         /* Update the POSIXACL Flag */
 597          sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 598                 (test_opt(sbi, POSIX_ACL) ? MS_POSIXACL : 0);
 599         return 0;
 600
 601 restore_opts:
 602         sbi->mount_opt = org_mount_opt;
 603         sbi->active_logs = active_logs;
 604         return err;
 605 }
 606
 607 static struct super_operations f2fs_sops = {
 608         .alloc_inode    = f2fs_alloc_inode,
 609         .drop_inode     = f2fs_drop_inode,
 610         .destroy_inode  = f2fs_destroy_inode,
 611         .write_inode    = f2fs_write_inode,
 612         .dirty_inode    = f2fs_dirty_inode,
 613         .show_options   = f2fs_show_options,
 614         .evict_inode    = f2fs_evict_inode,
 615         .put_super      = f2fs_put_super,
 616         .sync_fs        = f2fs_sync_fs,
 617         .freeze_fs      = f2fs_freeze,
 618         .unfreeze_fs    = f2fs_unfreeze,
 619         .statfs         = f2fs_statfs,
 620         .remount_fs     = f2fs_remount,
 621 };
 622
 623 static struct inode *f2fs_nfs_get_inode(struct super_block *sb,
 624                 u64 ino, u32 generation)
 625 {
 626         struct f2fs_sb_info *sbi = F2FS_SB(sb);
 627         struct inode *inode;
 628
 629         if (ino < F2FS_ROOT_INO(sbi))
 630                 return ERR_PTR(-ESTALE);
 631
 632         /*
 633          * f2fs_iget isn't quite right if the inode is currently unallocated!
 634          * However f2fs_iget currently does appropriate checks to handle stale
 635          * inodes so everything is OK.
 636          */
 637         inode = f2fs_iget(sb, ino);
 638         if (IS_ERR(inode))
 639                 return ERR_CAST(inode);
 640         if (generation && inode->i_generation != generation) {
 641                 /* we didn't find the right inode.. */
 642                 iput(inode);
 643                 return ERR_PTR(-ESTALE);
 644         }
 645         return inode;
 646 }
 647
 648 static struct dentry *f2fs_fh_to_dentry(struct super_block *sb, struct fid *fid,
 649                 int fh_len, int fh_type)
 650 {
 651         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 652                                     f2fs_nfs_get_inode);
 653 }
 654
 655 static struct dentry *f2fs_fh_to_parent(struct super_block *sb, struct fid *fid,
 656                 int fh_len, int fh_type)
 657 {
 658         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 659                                     f2fs_nfs_get_inode);
 660 }
 661
 662 static const struct export_operations f2fs_export_ops = {
 663         .fh_to_dentry = f2fs_fh_to_dentry,
 664         .fh_to_parent = f2fs_fh_to_parent,
 665         .get_parent = f2fs_get_parent,
 666 };
 667
 668 static loff_t max_file_size(unsigned bits)
 669 {
 670         loff_t result = (DEF_ADDRS_PER_INODE - F2FS_INLINE_XATTR_ADDRS);
 671         loff_t leaf_count = ADDRS_PER_BLOCK;
 672
 673         /* two direct node blocks */
 674         result += (leaf_count * 2);
 675
 676         /* two indirect node blocks */
 677         leaf_count *= NIDS_PER_BLOCK;
 678         result += (leaf_count * 2);
 679
 680         /* one double indirect node block */
 681         leaf_count *= NIDS_PER_BLOCK;
 682         result += leaf_count;
 683
 684         result <<= bits;
 685         return result;
 686 }
 687
 688 static int sanity_check_raw_super(struct super_block *sb,
 689                         struct f2fs_super_block *raw_super)
 690 {
 691         unsigned int blocksize;
 692
 693         if (F2FS_SUPER_MAGIC != le32_to_cpu(raw_super->magic)) {
 694                 f2fs_msg(sb, KERN_INFO,
 695                         "Magic Mismatch, valid(0x%x) - read(0x%x)",
 696                         F2FS_SUPER_MAGIC, le32_to_cpu(raw_super->magic));
 697                 return 1;
 698         }
 699
 700         /* Currently, support only 4KB page cache size */
 701         if (F2FS_BLKSIZE != PAGE_CACHE_SIZE) {
 702                 f2fs_msg(sb, KERN_INFO,
 703                         "Invalid page_cache_size (%lu), supports only 4KB\n",
 704                         PAGE_CACHE_SIZE);
 705                 return 1;
 706         }
 707
 708         /* Currently, support only 4KB block size */
 709         blocksize = 1 << le32_to_cpu(raw_super->log_blocksize);
 710         if (blocksize != F2FS_BLKSIZE) {
 711                 f2fs_msg(sb, KERN_INFO,
 712                         "Invalid blocksize (%u), supports only 4KB\n",
 713                         blocksize);
 714                 return 1;
 715         }
 716
 717         if (le32_to_cpu(raw_super->log_sectorsize) !=
 718                                         F2FS_LOG_SECTOR_SIZE) {
 719                 f2fs_msg(sb, KERN_INFO, "Invalid log sectorsize");
 720                 return 1;
 721         }
 722         if (le32_to_cpu(raw_super->log_sectors_per_block) !=
 723                                         F2FS_LOG_SECTORS_PER_BLOCK) {
 724                 f2fs_msg(sb, KERN_INFO, "Invalid log sectors per block");
 725                 return 1;
 726         }
 727         return 0;
 728 }
 729
 730 static int sanity_check_ckpt(struct f2fs_sb_info *sbi)
 731 {
 732         unsigned int total, fsmeta;
 733         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
 734         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 735
 736         total = le32_to_cpu(raw_super->segment_count);
 737         fsmeta = le32_to_cpu(raw_super->segment_count_ckpt);
 738         fsmeta += le32_to_cpu(raw_super->segment_count_sit);
 739         fsmeta += le32_to_cpu(raw_super->segment_count_nat);
 740         fsmeta += le32_to_cpu(ckpt->rsvd_segment_count);
 741         fsmeta += le32_to_cpu(raw_super->segment_count_ssa);
 742
 743         if (fsmeta >= total)
 744                 return 1;
 745
 746         if (is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) {
 747                 f2fs_msg(sbi->sb, KERN_ERR, "A bug case: need to run fsck");
 748                 return 1;
 749         }
 750         return 0;
 751 }
 752
 753 static void init_sb_info(struct f2fs_sb_info *sbi)
 754 {
 755         struct f2fs_super_block *raw_super = sbi->raw_super;
 756         int i;
 757
 758         sbi->log_sectors_per_block =
 759                 le32_to_cpu(raw_super->log_sectors_per_block);
 760         sbi->log_blocksize = le32_to_cpu(raw_super->log_blocksize);
 761         sbi->blocksize = 1 << sbi->log_blocksize;
 762         sbi->log_blocks_per_seg = le32_to_cpu(raw_super->log_blocks_per_seg);
 763         sbi->blocks_per_seg = 1 << sbi->log_blocks_per_seg;
 764         sbi->segs_per_sec = le32_to_cpu(raw_super->segs_per_sec);
 765         sbi->secs_per_zone = le32_to_cpu(raw_super->secs_per_zone);
 766         sbi->total_sections = le32_to_cpu(raw_super->section_count);
 767         sbi->total_node_count =
 768                 (le32_to_cpu(raw_super->segment_count_nat) / 2)
 769                         * sbi->blocks_per_seg * NAT_ENTRY_PER_BLOCK;
 770         sbi->root_ino_num = le32_to_cpu(raw_super->root_ino);
 771         sbi->node_ino_num = le32_to_cpu(raw_super->node_ino);
 772         sbi->meta_ino_num = le32_to_cpu(raw_super->meta_ino);
 773         sbi->cur_victim_sec = NULL_SECNO;
 774
 775         for (i = 0; i < NR_COUNT_TYPE; i++)
 776                 atomic_set(&sbi->nr_pages[i], 0);
 777 }
 778
 779 /*
 780  * Read f2fs raw super block.
 781  * Because we have two copies of super block, so read the first one at first,
 782  * if the first one is invalid, move to read the second one.
 783  */
 784 static int read_raw_super_block(struct super_block *sb,
 785                         struct f2fs_super_block **raw_super,
 786                         struct buffer_head **raw_super_buf)
 787 {
 788         int block = 0;
 789
 790 retry:
 791         *raw_super_buf = sb_bread(sb, block);
 792         if (!*raw_super_buf) {
 793                 f2fs_msg(sb, KERN_ERR, "Unable to read %dth superblock",
 794                                 block + 1);
 795                 if (block == 0) {
 796                         block++;
 797                         goto retry;
 798                 } else {
 799                         return -EIO;
 800                 }
 801         }
 802
 803         *raw_super = (struct f2fs_super_block *)
 804                 ((char *)(*raw_super_buf)->b_data + F2FS_SUPER_OFFSET);
 805
 806         /* sanity checking of raw super */
 807         if (sanity_check_raw_super(sb, *raw_super)) {
 808                 brelse(*raw_super_buf);
 809                 f2fs_msg(sb, KERN_ERR, "Can't find a valid F2FS filesystem "
 810                                 "in %dth superblock", block + 1);
 811                 if(block == 0) {
 812                         block++;
 813                         goto retry;
 814                 } else {
 815                         return -EINVAL;
 816                 }
 817         }
 818
 819         return 0;
 820 }
 821
 822 static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
 823 {
 824         struct f2fs_sb_info *sbi;
 825         struct f2fs_super_block *raw_super;
 826         struct buffer_head *raw_super_buf;
 827         struct inode *root;
 828         long err = -EINVAL;
 829         int i;
 830
 831         /* allocate memory for f2fs-specific super block info */
 832         sbi = kzalloc(sizeof(struct f2fs_sb_info), GFP_KERNEL);
 833         if (!sbi)
 834                 return -ENOMEM;
 835
 836         /* set a block size */
 837         if (!sb_set_blocksize(sb, F2FS_BLKSIZE)) {
 838                 f2fs_msg(sb, KERN_ERR, "unable to set blocksize");
 839                 goto free_sbi;
 840         }
 841
 842         err = read_raw_super_block(sb, &raw_super, &raw_super_buf);
 843         if (err)
 844                 goto free_sbi;
 845
 846         sb->s_fs_info = sbi;
 847         /* init some FS parameters */
 848         sbi->active_logs = NR_CURSEG_TYPE;
 849
 850         set_opt(sbi, BG_GC);
 851
 852 #ifdef CONFIG_F2FS_FS_XATTR
 853         set_opt(sbi, XATTR_USER);
 854 #endif
 855 #ifdef CONFIG_F2FS_FS_POSIX_ACL
 856         set_opt(sbi, POSIX_ACL);
 857 #endif
 858         /* parse mount options */
 859         err = parse_options(sb, (char *)data);
 860         if (err)
 861                 goto free_sb_buf;
 862
 863         sb->s_maxbytes = max_file_size(le32_to_cpu(raw_super->log_blocksize));
 864         sb->s_max_links = F2FS_LINK_MAX;
 865         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
 866
 867         sb->s_op = &f2fs_sops;
 868         sb->s_xattr = f2fs_xattr_handlers;
 869         sb->s_export_op = &f2fs_export_ops;
 870         sb->s_magic = F2FS_SUPER_MAGIC;
 871         sb->s_time_gran = 1;
 872         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 873                 (test_opt(sbi, POSIX_ACL) ? MS_POSIXACL : 0);
 874         memcpy(sb->s_uuid, raw_super->uuid, sizeof(raw_super->uuid));
 875
 876         /* init f2fs-specific super block info */
 877         sbi->sb = sb;
 878         sbi->raw_super = raw_super;
 879         sbi->raw_super_buf = raw_super_buf;
 880         mutex_init(&sbi->gc_mutex);
 881         mutex_init(&sbi->writepages);
 882         mutex_init(&sbi->cp_mutex);
 883         mutex_init(&sbi->node_write);
 884         sbi->por_doing = false;
 885         spin_lock_init(&sbi->stat_lock);
 886
 887         mutex_init(&sbi->read_io.io_mutex);
 888         for (i = 0; i < NR_PAGE_TYPE; i++)
 889                 mutex_init(&sbi->write_io[i].io_mutex);
 890
 891         init_rwsem(&sbi->cp_rwsem);
 892         init_waitqueue_head(&sbi->cp_wait);
 893         init_sb_info(sbi);
 894
 895         /* get an inode for meta space */
 896         sbi->meta_inode = f2fs_iget(sb, F2FS_META_INO(sbi));
 897         if (IS_ERR(sbi->meta_inode)) {
 898                 f2fs_msg(sb, KERN_ERR, "Failed to read F2FS meta data inode");
 899                 err = PTR_ERR(sbi->meta_inode);
 900                 goto free_sb_buf;
 901         }
 902
 903         err = get_valid_checkpoint(sbi);
 904         if (err) {
 905                 f2fs_msg(sb, KERN_ERR, "Failed to get valid F2FS checkpoint");
 906                 goto free_meta_inode;
 907         }
 908
 909         /* sanity checking of checkpoint */
 910         err = -EINVAL;
 911         if (sanity_check_ckpt(sbi)) {
 912                 f2fs_msg(sb, KERN_ERR, "Invalid F2FS checkpoint");
 913                 goto free_cp;
 914         }
 915
 916         sbi->total_valid_node_count =
 917                                 le32_to_cpu(sbi->ckpt->valid_node_count);
 918         sbi->total_valid_inode_count =
 919                                 le32_to_cpu(sbi->ckpt->valid_inode_count);
 920         sbi->user_block_count = le64_to_cpu(sbi->ckpt->user_block_count);
 921         sbi->total_valid_block_count =
 922                                 le64_to_cpu(sbi->ckpt->valid_block_count);
 923         sbi->last_valid_block_count = sbi->total_valid_block_count;
 924         sbi->alloc_valid_block_count = 0;
 925         INIT_LIST_HEAD(&sbi->dir_inode_list);
 926         spin_lock_init(&sbi->dir_inode_lock);
 927
 928         init_orphan_info(sbi);
 929
 930         /* setup f2fs internal modules */
 931         err = build_segment_manager(sbi);
 932         if (err) {
 933                 f2fs_msg(sb, KERN_ERR,
 934                         "Failed to initialize F2FS segment manager");
 935                 goto free_sm;
 936         }
 937         err = build_node_manager(sbi);
 938         if (err) {
 939                 f2fs_msg(sb, KERN_ERR,
 940                         "Failed to initialize F2FS node manager");
 941                 goto free_nm;
 942         }
 943
 944         build_gc_manager(sbi);
 945
 946         /* get an inode for node space */
 947         sbi->node_inode = f2fs_iget(sb, F2FS_NODE_INO(sbi));
 948         if (IS_ERR(sbi->node_inode)) {
 949                 f2fs_msg(sb, KERN_ERR, "Failed to read node inode");
 950                 err = PTR_ERR(sbi->node_inode);
 951                 goto free_nm;
 952         }
 953
 954         /* if there are nt orphan nodes free them */
 955         err = -EINVAL;
 956         if (recover_orphan_inodes(sbi))
 957                 goto free_node_inode;
 958
 959         /* read root inode and dentry */
 960         root = f2fs_iget(sb, F2FS_ROOT_INO(sbi));
 961         if (IS_ERR(root)) {
 962                 f2fs_msg(sb, KERN_ERR, "Failed to read root inode");
 963                 err = PTR_ERR(root);
 964                 goto free_node_inode;
 965         }
 966         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size)
 967                 goto free_root_inode;
 968
 969         sb->s_root = d_make_root(root); /* allocate root dentry */
 970         if (!sb->s_root) {
 971                 err = -ENOMEM;
 972                 goto free_root_inode;
 973         }
 974
 975         /* recover fsynced data */
 976         if (!test_opt(sbi, DISABLE_ROLL_FORWARD)) {
 977                 err = recover_fsync_data(sbi);
 978                 if (err)
 979                         f2fs_msg(sb, KERN_ERR,
 980                                 "Cannot recover all fsync data errno=%ld", err);
 981         }
 982
 983         /*
 984          * If filesystem is not mounted as read-only then
 985          * do start the gc_thread.
 986          */
 987         if (!(sb->s_flags & MS_RDONLY)) {
 988                 /* After POR, we can run background GC thread.*/
 989                 err = start_gc_thread(sbi);
 990                 if (err)
 991                         goto free_gc;
 992         }
 993
 994         err = f2fs_build_stats(sbi);
 995         if (err)
 996                 goto free_gc;
 997
 998         if (f2fs_proc_root)
 999                 sbi->s_proc = proc_mkdir(sb->s_id, f2fs_proc_root);
1000
1001         if (sbi->s_proc)
1002                 proc_create_data("segment_info", S_IRUGO, sbi->s_proc,
1003                                  &f2fs_seq_segment_info_fops, sb);
1004
1005         if (test_opt(sbi, DISCARD)) {
1006                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1007                 if (!blk_queue_discard(q))
1008                         f2fs_msg(sb, KERN_WARNING,
1009                                         "mounting with \"discard\" option, but "
1010                                         "the device does not support discard");
1011         }
1012
1013         sbi->s_kobj.kset = f2fs_kset;
1014         init_completion(&sbi->s_kobj_unregister);
1015         err = kobject_init_and_add(&sbi->s_kobj, &f2fs_ktype, NULL,
1016                                                         "%s", sb->s_id);
1017         if (err)
1018                 goto fail;
1019
1020         return 0;
1021 fail:
1022         if (sbi->s_proc) {
1023                 remove_proc_entry("segment_info", sbi->s_proc);
1024                 remove_proc_entry(sb->s_id, f2fs_proc_root);
1025         }
1026         f2fs_destroy_stats(sbi);
1027 free_gc:
1028         stop_gc_thread(sbi);
1029 free_root_inode:
1030         dput(sb->s_root);
1031         sb->s_root = NULL;
1032 free_node_inode:
1033         iput(sbi->node_inode);
1034 free_nm:
1035         destroy_node_manager(sbi);
1036 free_sm:
1037         destroy_segment_manager(sbi);
1038 free_cp:
1039         kfree(sbi->ckpt);
1040 free_meta_inode:
1041         make_bad_inode(sbi->meta_inode);
1042         iput(sbi->meta_inode);
1043 free_sb_buf:
1044         brelse(raw_super_buf);
1045 free_sbi:
1046         kfree(sbi);
1047         return err;
1048 }
1049
1050 static struct dentry *f2fs_mount(struct file_system_type *fs_type, int flags,
1051                         const char *dev_name, void *data)
1052 {
1053         return mount_bdev(fs_type, flags, dev_name, data, f2fs_fill_super);
1054 }
1055
1056 static struct file_system_type f2fs_fs_type = {
1057         .owner          = THIS_MODULE,
1058         .name           = "f2fs",
1059         .mount          = f2fs_mount,
1060         .kill_sb        = kill_block_super,
1061         .fs_flags       = FS_REQUIRES_DEV,
1062 };
1063 MODULE_ALIAS_FS("f2fs");
1064
1065 static int __init init_inodecache(void)
1066 {
1067         f2fs_inode_cachep = f2fs_kmem_cache_create("f2fs_inode_cache",
1068                         sizeof(struct f2fs_inode_info), NULL);
1069         if (f2fs_inode_cachep == NULL)
1070                 return -ENOMEM;
1071         return 0;
1072 }
1073
1074 static void destroy_inodecache(void)
1075 {
1076         /*
1077          * Make sure all delayed rcu free inodes are flushed before we
1078          * destroy cache.
1079          */
1080         rcu_barrier();
1081         kmem_cache_destroy(f2fs_inode_cachep);
1082 }
1083
1084 static int __init init_f2fs_fs(void)
1085 {
1086         int err;
1087
1088         err = init_inodecache();
1089         if (err)
1090                 goto fail;
1091         err = create_node_manager_caches();
1092         if (err)
1093                 goto free_inodecache;
1094         err = create_segment_manager_caches();
1095         if (err)
1096                 goto free_node_manager_caches;
1097         err = create_gc_caches();
1098         if (err)
1099                 goto free_segment_manager_caches;
1100         err = create_checkpoint_caches();
1101         if (err)
1102                 goto free_gc_caches;
1103         f2fs_kset = kset_create_and_add("f2fs", NULL, fs_kobj);
1104         if (!f2fs_kset) {
1105                 err = -ENOMEM;
1106                 goto free_checkpoint_caches;
1107         }
1108         err = register_filesystem(&f2fs_fs_type);
1109         if (err)
1110                 goto free_kset;
1111         f2fs_create_root_stats();
1112         f2fs_proc_root = proc_mkdir("fs/f2fs", NULL);
1113         return 0;
1114
1115 free_kset:
1116         kset_unregister(f2fs_kset);
1117 free_checkpoint_caches:
1118         destroy_checkpoint_caches();
1119 free_gc_caches:
1120         destroy_gc_caches();
1121 free_segment_manager_caches:
1122         destroy_segment_manager_caches();
1123 free_node_manager_caches:
1124         destroy_node_manager_caches();
1125 free_inodecache:
1126         destroy_inodecache();
1127 fail:
1128         return err;
1129 }
1130
1131 static void __exit exit_f2fs_fs(void)
1132 {
1133         remove_proc_entry("fs/f2fs", NULL);
1134         f2fs_destroy_root_stats();
1135         unregister_filesystem(&f2fs_fs_type);
1136         destroy_checkpoint_caches();
1137         destroy_gc_caches();
1138         destroy_node_manager_caches();
1139         destroy_inodecache();
1140         kset_unregister(f2fs_kset);
1141 }
1142
1143 module_init(init_f2fs_fs)
1144 module_exit(exit_f2fs_fs)
1145
1146 MODULE_AUTHOR("Samsung Electronics's Praesto Team");
1147 MODULE_DESCRIPTION("Flash Friendly File System");
1148 MODULE_LICENSE("GPL");