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 #include "trace.h" 34 35 #define CREATE_TRACE_POINTS 36 #include <trace/events/f2fs.h> 37 38 static struct proc_dir_entry *f2fs_proc_root; 39 static struct kmem_cache *f2fs_inode_cachep; 40 static struct kset *f2fs_kset; 41 42 enum { 43 Opt_gc_background, 44 Opt_disable_roll_forward, 45 Opt_discard, 46 Opt_noheap, 47 Opt_user_xattr, 48 Opt_nouser_xattr, 49 Opt_acl, 50 Opt_noacl, 51 Opt_active_logs, 52 Opt_disable_ext_identify, 53 Opt_inline_xattr, 54 Opt_inline_data, 55 Opt_inline_dentry, 56 Opt_flush_merge, 57 Opt_nobarrier, 58 Opt_fastboot, 59 Opt_err, 60 }; 61 62 static match_table_t f2fs_tokens = { 63 {Opt_gc_background, "background_gc=%s"}, 64 {Opt_disable_roll_forward, "disable_roll_forward"}, 65 {Opt_discard, "discard"}, 66 {Opt_noheap, "no_heap"}, 67 {Opt_user_xattr, "user_xattr"}, 68 {Opt_nouser_xattr, "nouser_xattr"}, 69 {Opt_acl, "acl"}, 70 {Opt_noacl, "noacl"}, 71 {Opt_active_logs, "active_logs=%u"}, 72 {Opt_disable_ext_identify, "disable_ext_identify"}, 73 {Opt_inline_xattr, "inline_xattr"}, 74 {Opt_inline_data, "inline_data"}, 75 {Opt_inline_dentry, "inline_dentry"}, 76 {Opt_flush_merge, "flush_merge"}, 77 {Opt_nobarrier, "nobarrier"}, 78 {Opt_fastboot, "fastboot"}, 79 {Opt_err, NULL}, 80 }; 81 82 /* Sysfs support for f2fs */ 83 enum { 84 GC_THREAD, /* struct f2fs_gc_thread */ 85 SM_INFO, /* struct f2fs_sm_info */ 86 NM_INFO, /* struct f2fs_nm_info */ 87 F2FS_SBI, /* struct f2fs_sb_info */ 88 }; 89 90 struct f2fs_attr { 91 struct attribute attr; 92 ssize_t (*show)(struct f2fs_attr *, struct f2fs_sb_info *, char *); 93 ssize_t (*store)(struct f2fs_attr *, struct f2fs_sb_info *, 94 const char *, size_t); 95 int struct_type; 96 int offset; 97 }; 98 99 static unsigned char *__struct_ptr(struct f2fs_sb_info *sbi, int struct_type) 100 { 101 if (struct_type == GC_THREAD) 102 return (unsigned char *)sbi->gc_thread; 103 else if (struct_type == SM_INFO) 104 return (unsigned char *)SM_I(sbi); 105 else if (struct_type == NM_INFO) 106 return (unsigned char *)NM_I(sbi); 107 else if (struct_type == F2FS_SBI) 108 return (unsigned char *)sbi; 109 return NULL; 110 } 111 112 static ssize_t f2fs_sbi_show(struct f2fs_attr *a, 113 struct f2fs_sb_info *sbi, char *buf) 114 { 115 unsigned char *ptr = NULL; 116 unsigned int *ui; 117 118 ptr = __struct_ptr(sbi, a->struct_type); 119 if (!ptr) 120 return -EINVAL; 121 122 ui = (unsigned int *)(ptr + a->offset); 123 124 return snprintf(buf, PAGE_SIZE, "%u\n", *ui); 125 } 126 127 static ssize_t f2fs_sbi_store(struct f2fs_attr *a, 128 struct f2fs_sb_info *sbi, 129 const char *buf, size_t count) 130 { 131 unsigned char *ptr; 132 unsigned long t; 133 unsigned int *ui; 134 ssize_t ret; 135 136 ptr = __struct_ptr(sbi, a->struct_type); 137 if (!ptr) 138 return -EINVAL; 139 140 ui = (unsigned int *)(ptr + a->offset); 141 142 ret = kstrtoul(skip_spaces(buf), 0, &t); 143 if (ret < 0) 144 return ret; 145 *ui = t; 146 return count; 147 } 148 149 static ssize_t f2fs_attr_show(struct kobject *kobj, 150 struct attribute *attr, char *buf) 151 { 152 struct f2fs_sb_info *sbi = container_of(kobj, struct f2fs_sb_info, 153 s_kobj); 154 struct f2fs_attr *a = container_of(attr, struct f2fs_attr, attr); 155 156 return a->show ? a->show(a, sbi, buf) : 0; 157 } 158 159 static ssize_t f2fs_attr_store(struct kobject *kobj, struct attribute *attr, 160 const char *buf, size_t len) 161 { 162 struct f2fs_sb_info *sbi = container_of(kobj, struct f2fs_sb_info, 163 s_kobj); 164 struct f2fs_attr *a = container_of(attr, struct f2fs_attr, attr); 165 166 return a->store ? a->store(a, sbi, buf, len) : 0; 167 } 168 169 static void f2fs_sb_release(struct kobject *kobj) 170 { 171 struct f2fs_sb_info *sbi = container_of(kobj, struct f2fs_sb_info, 172 s_kobj); 173 complete(&sbi->s_kobj_unregister); 174 } 175 176 #define F2FS_ATTR_OFFSET(_struct_type, _name, _mode, _show, _store, _offset) \ 177 static struct f2fs_attr f2fs_attr_##_name = { \ 178 .attr = {.name = __stringify(_name), .mode = _mode }, \ 179 .show = _show, \ 180 .store = _store, \ 181 .struct_type = _struct_type, \ 182 .offset = _offset \ 183 } 184 185 #define F2FS_RW_ATTR(struct_type, struct_name, name, elname) \ 186 F2FS_ATTR_OFFSET(struct_type, name, 0644, \ 187 f2fs_sbi_show, f2fs_sbi_store, \ 188 offsetof(struct struct_name, elname)) 189 190 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_min_sleep_time, min_sleep_time); 191 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_max_sleep_time, max_sleep_time); 192 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_no_gc_sleep_time, no_gc_sleep_time); 193 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_idle, gc_idle); 194 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, reclaim_segments, rec_prefree_segments); 195 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, max_small_discards, max_discards); 196 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, ipu_policy, ipu_policy); 197 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, min_ipu_util, min_ipu_util); 198 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, min_fsync_blocks, min_fsync_blocks); 199 F2FS_RW_ATTR(NM_INFO, f2fs_nm_info, ram_thresh, ram_thresh); 200 F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, max_victim_search, max_victim_search); 201 F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, dir_level, dir_level); 202 203 #define ATTR_LIST(name) (&f2fs_attr_##name.attr) 204 static struct attribute *f2fs_attrs[] = { 205 ATTR_LIST(gc_min_sleep_time), 206 ATTR_LIST(gc_max_sleep_time), 207 ATTR_LIST(gc_no_gc_sleep_time), 208 ATTR_LIST(gc_idle), 209 ATTR_LIST(reclaim_segments), 210 ATTR_LIST(max_small_discards), 211 ATTR_LIST(ipu_policy), 212 ATTR_LIST(min_ipu_util), 213 ATTR_LIST(min_fsync_blocks), 214 ATTR_LIST(max_victim_search), 215 ATTR_LIST(dir_level), 216 ATTR_LIST(ram_thresh), 217 NULL, 218 }; 219 220 static const struct sysfs_ops f2fs_attr_ops = { 221 .show = f2fs_attr_show, 222 .store = f2fs_attr_store, 223 }; 224 225 static struct kobj_type f2fs_ktype = { 226 .default_attrs = f2fs_attrs, 227 .sysfs_ops = &f2fs_attr_ops, 228 .release = f2fs_sb_release, 229 }; 230 231 void f2fs_msg(struct super_block *sb, const char *level, const char *fmt, ...) 232 { 233 struct va_format vaf; 234 va_list args; 235 236 va_start(args, fmt); 237 vaf.fmt = fmt; 238 vaf.va = &args; 239 printk("%sF2FS-fs (%s): %pV\n", level, sb->s_id, &vaf); 240 va_end(args); 241 } 242 243 static void init_once(void *foo) 244 { 245 struct f2fs_inode_info *fi = (struct f2fs_inode_info *) foo; 246 247 inode_init_once(&fi->vfs_inode); 248 } 249 250 static int parse_options(struct super_block *sb, char *options) 251 { 252 struct f2fs_sb_info *sbi = F2FS_SB(sb); 253 substring_t args[MAX_OPT_ARGS]; 254 char *p, *name; 255 int arg = 0; 256 257 if (!options) 258 return 0; 259 260 while ((p = strsep(&options, ",")) != NULL) { 261 int token; 262 if (!*p) 263 continue; 264 /* 265 * Initialize args struct so we know whether arg was 266 * found; some options take optional arguments. 267 */ 268 args[0].to = args[0].from = NULL; 269 token = match_token(p, f2fs_tokens, args); 270 271 switch (token) { 272 case Opt_gc_background: 273 name = match_strdup(&args[0]); 274 275 if (!name) 276 return -ENOMEM; 277 if (strlen(name) == 2 && !strncmp(name, "on", 2)) 278 set_opt(sbi, BG_GC); 279 else if (strlen(name) == 3 && !strncmp(name, "off", 3)) 280 clear_opt(sbi, BG_GC); 281 else { 282 kfree(name); 283 return -EINVAL; 284 } 285 kfree(name); 286 break; 287 case Opt_disable_roll_forward: 288 set_opt(sbi, DISABLE_ROLL_FORWARD); 289 break; 290 case Opt_discard: 291 set_opt(sbi, DISCARD); 292 break; 293 case Opt_noheap: 294 set_opt(sbi, NOHEAP); 295 break; 296 #ifdef CONFIG_F2FS_FS_XATTR 297 case Opt_user_xattr: 298 set_opt(sbi, XATTR_USER); 299 break; 300 case Opt_nouser_xattr: 301 clear_opt(sbi, XATTR_USER); 302 break; 303 case Opt_inline_xattr: 304 set_opt(sbi, INLINE_XATTR); 305 break; 306 #else 307 case Opt_user_xattr: 308 f2fs_msg(sb, KERN_INFO, 309 "user_xattr options not supported"); 310 break; 311 case Opt_nouser_xattr: 312 f2fs_msg(sb, KERN_INFO, 313 "nouser_xattr options not supported"); 314 break; 315 case Opt_inline_xattr: 316 f2fs_msg(sb, KERN_INFO, 317 "inline_xattr options not supported"); 318 break; 319 #endif 320 #ifdef CONFIG_F2FS_FS_POSIX_ACL 321 case Opt_acl: 322 set_opt(sbi, POSIX_ACL); 323 break; 324 case Opt_noacl: 325 clear_opt(sbi, POSIX_ACL); 326 break; 327 #else 328 case Opt_acl: 329 f2fs_msg(sb, KERN_INFO, "acl options not supported"); 330 break; 331 case Opt_noacl: 332 f2fs_msg(sb, KERN_INFO, "noacl options not supported"); 333 break; 334 #endif 335 case Opt_active_logs: 336 if (args->from && match_int(args, &arg)) 337 return -EINVAL; 338 if (arg != 2 && arg != 4 && arg != NR_CURSEG_TYPE) 339 return -EINVAL; 340 sbi->active_logs = arg; 341 break; 342 case Opt_disable_ext_identify: 343 set_opt(sbi, DISABLE_EXT_IDENTIFY); 344 break; 345 case Opt_inline_data: 346 set_opt(sbi, INLINE_DATA); 347 break; 348 case Opt_inline_dentry: 349 set_opt(sbi, INLINE_DENTRY); 350 break; 351 case Opt_flush_merge: 352 set_opt(sbi, FLUSH_MERGE); 353 break; 354 case Opt_nobarrier: 355 set_opt(sbi, NOBARRIER); 356 break; 357 case Opt_fastboot: 358 set_opt(sbi, FASTBOOT); 359 break; 360 default: 361 f2fs_msg(sb, KERN_ERR, 362 "Unrecognized mount option \"%s\" or missing value", 363 p); 364 return -EINVAL; 365 } 366 } 367 return 0; 368 } 369 370 static struct inode *f2fs_alloc_inode(struct super_block *sb) 371 { 372 struct f2fs_inode_info *fi; 373 374 fi = kmem_cache_alloc(f2fs_inode_cachep, GFP_F2FS_ZERO); 375 if (!fi) 376 return NULL; 377 378 init_once((void *) fi); 379 380 /* Initialize f2fs-specific inode info */ 381 fi->vfs_inode.i_version = 1; 382 atomic_set(&fi->dirty_pages, 0); 383 fi->i_current_depth = 1; 384 fi->i_advise = 0; 385 rwlock_init(&fi->ext.ext_lock); 386 init_rwsem(&fi->i_sem); 387 INIT_RADIX_TREE(&fi->inmem_root, GFP_NOFS); 388 INIT_LIST_HEAD(&fi->inmem_pages); 389 mutex_init(&fi->inmem_lock); 390 391 set_inode_flag(fi, FI_NEW_INODE); 392 393 if (test_opt(F2FS_SB(sb), INLINE_XATTR)) 394 set_inode_flag(fi, FI_INLINE_XATTR); 395 396 /* Will be used by directory only */ 397 fi->i_dir_level = F2FS_SB(sb)->dir_level; 398 399 return &fi->vfs_inode; 400 } 401 402 static int f2fs_drop_inode(struct inode *inode) 403 { 404 /* 405 * This is to avoid a deadlock condition like below. 406 * writeback_single_inode(inode) 407 * - f2fs_write_data_page 408 * - f2fs_gc -> iput -> evict 409 * - inode_wait_for_writeback(inode) 410 */ 411 if (!inode_unhashed(inode) && inode->i_state & I_SYNC) 412 return 0; 413 return generic_drop_inode(inode); 414 } 415 416 /* 417 * f2fs_dirty_inode() is called from __mark_inode_dirty() 418 * 419 * We should call set_dirty_inode to write the dirty inode through write_inode. 420 */ 421 static void f2fs_dirty_inode(struct inode *inode, int flags) 422 { 423 set_inode_flag(F2FS_I(inode), FI_DIRTY_INODE); 424 } 425 426 static void f2fs_i_callback(struct rcu_head *head) 427 { 428 struct inode *inode = container_of(head, struct inode, i_rcu); 429 kmem_cache_free(f2fs_inode_cachep, F2FS_I(inode)); 430 } 431 432 static void f2fs_destroy_inode(struct inode *inode) 433 { 434 call_rcu(&inode->i_rcu, f2fs_i_callback); 435 } 436 437 static void f2fs_put_super(struct super_block *sb) 438 { 439 struct f2fs_sb_info *sbi = F2FS_SB(sb); 440 441 if (sbi->s_proc) { 442 remove_proc_entry("segment_info", sbi->s_proc); 443 remove_proc_entry(sb->s_id, f2fs_proc_root); 444 } 445 kobject_del(&sbi->s_kobj); 446 447 f2fs_destroy_stats(sbi); 448 stop_gc_thread(sbi); 449 450 /* 451 * We don't need to do checkpoint when superblock is clean. 452 * But, the previous checkpoint was not done by umount, it needs to do 453 * clean checkpoint again. 454 */ 455 if (sbi->s_dirty || 456 !is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG)) { 457 struct cp_control cpc = { 458 .reason = CP_UMOUNT, 459 }; 460 write_checkpoint(sbi, &cpc); 461 } 462 463 /* 464 * normally superblock is clean, so we need to release this. 465 * In addition, EIO will skip do checkpoint, we need this as well. 466 */ 467 release_dirty_inode(sbi); 468 release_discard_addrs(sbi); 469 470 iput(sbi->node_inode); 471 iput(sbi->meta_inode); 472 473 /* destroy f2fs internal modules */ 474 destroy_node_manager(sbi); 475 destroy_segment_manager(sbi); 476 477 kfree(sbi->ckpt); 478 kobject_put(&sbi->s_kobj); 479 wait_for_completion(&sbi->s_kobj_unregister); 480 481 sb->s_fs_info = NULL; 482 brelse(sbi->raw_super_buf); 483 kfree(sbi); 484 } 485 486 int f2fs_sync_fs(struct super_block *sb, int sync) 487 { 488 struct f2fs_sb_info *sbi = F2FS_SB(sb); 489 490 trace_f2fs_sync_fs(sb, sync); 491 492 if (sync) { 493 struct cp_control cpc; 494 495 cpc.reason = (test_opt(sbi, FASTBOOT) || sbi->s_closing) ? 496 CP_UMOUNT : CP_SYNC; 497 mutex_lock(&sbi->gc_mutex); 498 write_checkpoint(sbi, &cpc); 499 mutex_unlock(&sbi->gc_mutex); 500 } else { 501 f2fs_balance_fs(sbi); 502 } 503 f2fs_trace_ios(NULL, NULL, 1); 504 505 return 0; 506 } 507 508 static int f2fs_freeze(struct super_block *sb) 509 { 510 int err; 511 512 if (f2fs_readonly(sb)) 513 return 0; 514 515 err = f2fs_sync_fs(sb, 1); 516 return err; 517 } 518 519 static int f2fs_unfreeze(struct super_block *sb) 520 { 521 return 0; 522 } 523 524 static int f2fs_statfs(struct dentry *dentry, struct kstatfs *buf) 525 { 526 struct super_block *sb = dentry->d_sb; 527 struct f2fs_sb_info *sbi = F2FS_SB(sb); 528 u64 id = huge_encode_dev(sb->s_bdev->bd_dev); 529 block_t total_count, user_block_count, start_count, ovp_count; 530 531 total_count = le64_to_cpu(sbi->raw_super->block_count); 532 user_block_count = sbi->user_block_count; 533 start_count = le32_to_cpu(sbi->raw_super->segment0_blkaddr); 534 ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg; 535 buf->f_type = F2FS_SUPER_MAGIC; 536 buf->f_bsize = sbi->blocksize; 537 538 buf->f_blocks = total_count - start_count; 539 buf->f_bfree = buf->f_blocks - valid_user_blocks(sbi) - ovp_count; 540 buf->f_bavail = user_block_count - valid_user_blocks(sbi); 541 542 buf->f_files = sbi->total_node_count - F2FS_RESERVED_NODE_NUM; 543 buf->f_ffree = buf->f_files - valid_inode_count(sbi); 544 545 buf->f_namelen = F2FS_NAME_LEN; 546 buf->f_fsid.val[0] = (u32)id; 547 buf->f_fsid.val[1] = (u32)(id >> 32); 548 549 return 0; 550 } 551 552 static int f2fs_show_options(struct seq_file *seq, struct dentry *root) 553 { 554 struct f2fs_sb_info *sbi = F2FS_SB(root->d_sb); 555 556 if (!f2fs_readonly(sbi->sb) && test_opt(sbi, BG_GC)) 557 seq_printf(seq, ",background_gc=%s", "on"); 558 else 559 seq_printf(seq, ",background_gc=%s", "off"); 560 if (test_opt(sbi, DISABLE_ROLL_FORWARD)) 561 seq_puts(seq, ",disable_roll_forward"); 562 if (test_opt(sbi, DISCARD)) 563 seq_puts(seq, ",discard"); 564 if (test_opt(sbi, NOHEAP)) 565 seq_puts(seq, ",no_heap_alloc"); 566 #ifdef CONFIG_F2FS_FS_XATTR 567 if (test_opt(sbi, XATTR_USER)) 568 seq_puts(seq, ",user_xattr"); 569 else 570 seq_puts(seq, ",nouser_xattr"); 571 if (test_opt(sbi, INLINE_XATTR)) 572 seq_puts(seq, ",inline_xattr"); 573 #endif 574 #ifdef CONFIG_F2FS_FS_POSIX_ACL 575 if (test_opt(sbi, POSIX_ACL)) 576 seq_puts(seq, ",acl"); 577 else 578 seq_puts(seq, ",noacl"); 579 #endif 580 if (test_opt(sbi, DISABLE_EXT_IDENTIFY)) 581 seq_puts(seq, ",disable_ext_identify"); 582 if (test_opt(sbi, INLINE_DATA)) 583 seq_puts(seq, ",inline_data"); 584 if (test_opt(sbi, INLINE_DENTRY)) 585 seq_puts(seq, ",inline_dentry"); 586 if (!f2fs_readonly(sbi->sb) && test_opt(sbi, FLUSH_MERGE)) 587 seq_puts(seq, ",flush_merge"); 588 if (test_opt(sbi, NOBARRIER)) 589 seq_puts(seq, ",nobarrier"); 590 if (test_opt(sbi, FASTBOOT)) 591 seq_puts(seq, ",fastboot"); 592 seq_printf(seq, ",active_logs=%u", sbi->active_logs); 593 594 return 0; 595 } 596 597 static int segment_info_seq_show(struct seq_file *seq, void *offset) 598 { 599 struct super_block *sb = seq->private; 600 struct f2fs_sb_info *sbi = F2FS_SB(sb); 601 unsigned int total_segs = 602 le32_to_cpu(sbi->raw_super->segment_count_main); 603 int i; 604 605 seq_puts(seq, "format: segment_type|valid_blocks\n" 606 "segment_type(0:HD, 1:WD, 2:CD, 3:HN, 4:WN, 5:CN)\n"); 607 608 for (i = 0; i < total_segs; i++) { 609 struct seg_entry *se = get_seg_entry(sbi, i); 610 611 if ((i % 10) == 0) 612 seq_printf(seq, "%-5d", i); 613 seq_printf(seq, "%d|%-3u", se->type, 614 get_valid_blocks(sbi, i, 1)); 615 if ((i % 10) == 9 || i == (total_segs - 1)) 616 seq_putc(seq, '\n'); 617 else 618 seq_putc(seq, ' '); 619 } 620 621 return 0; 622 } 623 624 static int segment_info_open_fs(struct inode *inode, struct file *file) 625 { 626 return single_open(file, segment_info_seq_show, PDE_DATA(inode)); 627 } 628 629 static const struct file_operations f2fs_seq_segment_info_fops = { 630 .owner = THIS_MODULE, 631 .open = segment_info_open_fs, 632 .read = seq_read, 633 .llseek = seq_lseek, 634 .release = single_release, 635 }; 636 637 static int f2fs_remount(struct super_block *sb, int *flags, char *data) 638 { 639 struct f2fs_sb_info *sbi = F2FS_SB(sb); 640 struct f2fs_mount_info org_mount_opt; 641 int err, active_logs; 642 bool need_restart_gc = false; 643 bool need_stop_gc = false; 644 645 sync_filesystem(sb); 646 647 /* 648 * Save the old mount options in case we 649 * need to restore them. 650 */ 651 org_mount_opt = sbi->mount_opt; 652 active_logs = sbi->active_logs; 653 654 sbi->mount_opt.opt = 0; 655 sbi->active_logs = NR_CURSEG_TYPE; 656 657 /* parse mount options */ 658 err = parse_options(sb, data); 659 if (err) 660 goto restore_opts; 661 662 /* 663 * Previous and new state of filesystem is RO, 664 * so skip checking GC and FLUSH_MERGE conditions. 665 */ 666 if (f2fs_readonly(sb) && (*flags & MS_RDONLY)) 667 goto skip; 668 669 /* 670 * We stop the GC thread if FS is mounted as RO 671 * or if background_gc = off is passed in mount 672 * option. Also sync the filesystem. 673 */ 674 if ((*flags & MS_RDONLY) || !test_opt(sbi, BG_GC)) { 675 if (sbi->gc_thread) { 676 stop_gc_thread(sbi); 677 f2fs_sync_fs(sb, 1); 678 need_restart_gc = true; 679 } 680 } else if (!sbi->gc_thread) { 681 err = start_gc_thread(sbi); 682 if (err) 683 goto restore_opts; 684 need_stop_gc = true; 685 } 686 687 /* 688 * We stop issue flush thread if FS is mounted as RO 689 * or if flush_merge is not passed in mount option. 690 */ 691 if ((*flags & MS_RDONLY) || !test_opt(sbi, FLUSH_MERGE)) { 692 destroy_flush_cmd_control(sbi); 693 } else if (!SM_I(sbi)->cmd_control_info) { 694 err = create_flush_cmd_control(sbi); 695 if (err) 696 goto restore_gc; 697 } 698 skip: 699 /* Update the POSIXACL Flag */ 700 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) | 701 (test_opt(sbi, POSIX_ACL) ? MS_POSIXACL : 0); 702 return 0; 703 restore_gc: 704 if (need_restart_gc) { 705 if (start_gc_thread(sbi)) 706 f2fs_msg(sbi->sb, KERN_WARNING, 707 "background gc thread has stopped"); 708 } else if (need_stop_gc) { 709 stop_gc_thread(sbi); 710 } 711 restore_opts: 712 sbi->mount_opt = org_mount_opt; 713 sbi->active_logs = active_logs; 714 return err; 715 } 716 717 static struct super_operations f2fs_sops = { 718 .alloc_inode = f2fs_alloc_inode, 719 .drop_inode = f2fs_drop_inode, 720 .destroy_inode = f2fs_destroy_inode, 721 .write_inode = f2fs_write_inode, 722 .dirty_inode = f2fs_dirty_inode, 723 .show_options = f2fs_show_options, 724 .evict_inode = f2fs_evict_inode, 725 .put_super = f2fs_put_super, 726 .sync_fs = f2fs_sync_fs, 727 .freeze_fs = f2fs_freeze, 728 .unfreeze_fs = f2fs_unfreeze, 729 .statfs = f2fs_statfs, 730 .remount_fs = f2fs_remount, 731 }; 732 733 static struct inode *f2fs_nfs_get_inode(struct super_block *sb, 734 u64 ino, u32 generation) 735 { 736 struct f2fs_sb_info *sbi = F2FS_SB(sb); 737 struct inode *inode; 738 739 if (check_nid_range(sbi, ino)) 740 return ERR_PTR(-ESTALE); 741 742 /* 743 * f2fs_iget isn't quite right if the inode is currently unallocated! 744 * However f2fs_iget currently does appropriate checks to handle stale 745 * inodes so everything is OK. 746 */ 747 inode = f2fs_iget(sb, ino); 748 if (IS_ERR(inode)) 749 return ERR_CAST(inode); 750 if (unlikely(generation && inode->i_generation != generation)) { 751 /* we didn't find the right inode.. */ 752 iput(inode); 753 return ERR_PTR(-ESTALE); 754 } 755 return inode; 756 } 757 758 static struct dentry *f2fs_fh_to_dentry(struct super_block *sb, struct fid *fid, 759 int fh_len, int fh_type) 760 { 761 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 762 f2fs_nfs_get_inode); 763 } 764 765 static struct dentry *f2fs_fh_to_parent(struct super_block *sb, struct fid *fid, 766 int fh_len, int fh_type) 767 { 768 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 769 f2fs_nfs_get_inode); 770 } 771 772 static const struct export_operations f2fs_export_ops = { 773 .fh_to_dentry = f2fs_fh_to_dentry, 774 .fh_to_parent = f2fs_fh_to_parent, 775 .get_parent = f2fs_get_parent, 776 }; 777 778 static loff_t max_file_size(unsigned bits) 779 { 780 loff_t result = (DEF_ADDRS_PER_INODE - F2FS_INLINE_XATTR_ADDRS); 781 loff_t leaf_count = ADDRS_PER_BLOCK; 782 783 /* two direct node blocks */ 784 result += (leaf_count * 2); 785 786 /* two indirect node blocks */ 787 leaf_count *= NIDS_PER_BLOCK; 788 result += (leaf_count * 2); 789 790 /* one double indirect node block */ 791 leaf_count *= NIDS_PER_BLOCK; 792 result += leaf_count; 793 794 result <<= bits; 795 return result; 796 } 797 798 static int sanity_check_raw_super(struct super_block *sb, 799 struct f2fs_super_block *raw_super) 800 { 801 unsigned int blocksize; 802 803 if (F2FS_SUPER_MAGIC != le32_to_cpu(raw_super->magic)) { 804 f2fs_msg(sb, KERN_INFO, 805 "Magic Mismatch, valid(0x%x) - read(0x%x)", 806 F2FS_SUPER_MAGIC, le32_to_cpu(raw_super->magic)); 807 return 1; 808 } 809 810 /* Currently, support only 4KB page cache size */ 811 if (F2FS_BLKSIZE != PAGE_CACHE_SIZE) { 812 f2fs_msg(sb, KERN_INFO, 813 "Invalid page_cache_size (%lu), supports only 4KB\n", 814 PAGE_CACHE_SIZE); 815 return 1; 816 } 817 818 /* Currently, support only 4KB block size */ 819 blocksize = 1 << le32_to_cpu(raw_super->log_blocksize); 820 if (blocksize != F2FS_BLKSIZE) { 821 f2fs_msg(sb, KERN_INFO, 822 "Invalid blocksize (%u), supports only 4KB\n", 823 blocksize); 824 return 1; 825 } 826 827 /* Currently, support 512/1024/2048/4096 bytes sector size */ 828 if (le32_to_cpu(raw_super->log_sectorsize) > 829 F2FS_MAX_LOG_SECTOR_SIZE || 830 le32_to_cpu(raw_super->log_sectorsize) < 831 F2FS_MIN_LOG_SECTOR_SIZE) { 832 f2fs_msg(sb, KERN_INFO, "Invalid log sectorsize (%u)", 833 le32_to_cpu(raw_super->log_sectorsize)); 834 return 1; 835 } 836 if (le32_to_cpu(raw_super->log_sectors_per_block) + 837 le32_to_cpu(raw_super->log_sectorsize) != 838 F2FS_MAX_LOG_SECTOR_SIZE) { 839 f2fs_msg(sb, KERN_INFO, 840 "Invalid log sectors per block(%u) log sectorsize(%u)", 841 le32_to_cpu(raw_super->log_sectors_per_block), 842 le32_to_cpu(raw_super->log_sectorsize)); 843 return 1; 844 } 845 return 0; 846 } 847 848 static int sanity_check_ckpt(struct f2fs_sb_info *sbi) 849 { 850 unsigned int total, fsmeta; 851 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 852 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 853 854 total = le32_to_cpu(raw_super->segment_count); 855 fsmeta = le32_to_cpu(raw_super->segment_count_ckpt); 856 fsmeta += le32_to_cpu(raw_super->segment_count_sit); 857 fsmeta += le32_to_cpu(raw_super->segment_count_nat); 858 fsmeta += le32_to_cpu(ckpt->rsvd_segment_count); 859 fsmeta += le32_to_cpu(raw_super->segment_count_ssa); 860 861 if (unlikely(fsmeta >= total)) 862 return 1; 863 864 if (unlikely(f2fs_cp_error(sbi))) { 865 f2fs_msg(sbi->sb, KERN_ERR, "A bug case: need to run fsck"); 866 return 1; 867 } 868 return 0; 869 } 870 871 static void init_sb_info(struct f2fs_sb_info *sbi) 872 { 873 struct f2fs_super_block *raw_super = sbi->raw_super; 874 int i; 875 876 sbi->log_sectors_per_block = 877 le32_to_cpu(raw_super->log_sectors_per_block); 878 sbi->log_blocksize = le32_to_cpu(raw_super->log_blocksize); 879 sbi->blocksize = 1 << sbi->log_blocksize; 880 sbi->log_blocks_per_seg = le32_to_cpu(raw_super->log_blocks_per_seg); 881 sbi->blocks_per_seg = 1 << sbi->log_blocks_per_seg; 882 sbi->segs_per_sec = le32_to_cpu(raw_super->segs_per_sec); 883 sbi->secs_per_zone = le32_to_cpu(raw_super->secs_per_zone); 884 sbi->total_sections = le32_to_cpu(raw_super->section_count); 885 sbi->total_node_count = 886 (le32_to_cpu(raw_super->segment_count_nat) / 2) 887 * sbi->blocks_per_seg * NAT_ENTRY_PER_BLOCK; 888 sbi->root_ino_num = le32_to_cpu(raw_super->root_ino); 889 sbi->node_ino_num = le32_to_cpu(raw_super->node_ino); 890 sbi->meta_ino_num = le32_to_cpu(raw_super->meta_ino); 891 sbi->cur_victim_sec = NULL_SECNO; 892 sbi->max_victim_search = DEF_MAX_VICTIM_SEARCH; 893 894 for (i = 0; i < NR_COUNT_TYPE; i++) 895 atomic_set(&sbi->nr_pages[i], 0); 896 897 sbi->dir_level = DEF_DIR_LEVEL; 898 sbi->need_fsck = false; 899 } 900 901 /* 902 * Read f2fs raw super block. 903 * Because we have two copies of super block, so read the first one at first, 904 * if the first one is invalid, move to read the second one. 905 */ 906 static int read_raw_super_block(struct super_block *sb, 907 struct f2fs_super_block **raw_super, 908 struct buffer_head **raw_super_buf) 909 { 910 int block = 0; 911 912 retry: 913 *raw_super_buf = sb_bread(sb, block); 914 if (!*raw_super_buf) { 915 f2fs_msg(sb, KERN_ERR, "Unable to read %dth superblock", 916 block + 1); 917 if (block == 0) { 918 block++; 919 goto retry; 920 } else { 921 return -EIO; 922 } 923 } 924 925 *raw_super = (struct f2fs_super_block *) 926 ((char *)(*raw_super_buf)->b_data + F2FS_SUPER_OFFSET); 927 928 /* sanity checking of raw super */ 929 if (sanity_check_raw_super(sb, *raw_super)) { 930 brelse(*raw_super_buf); 931 f2fs_msg(sb, KERN_ERR, 932 "Can't find valid F2FS filesystem in %dth superblock", 933 block + 1); 934 if (block == 0) { 935 block++; 936 goto retry; 937 } else { 938 return -EINVAL; 939 } 940 } 941 942 return 0; 943 } 944 945 static int f2fs_fill_super(struct super_block *sb, void *data, int silent) 946 { 947 struct f2fs_sb_info *sbi; 948 struct f2fs_super_block *raw_super = NULL; 949 struct buffer_head *raw_super_buf; 950 struct inode *root; 951 long err = -EINVAL; 952 bool retry = true; 953 int i; 954 955 try_onemore: 956 /* allocate memory for f2fs-specific super block info */ 957 sbi = kzalloc(sizeof(struct f2fs_sb_info), GFP_KERNEL); 958 if (!sbi) 959 return -ENOMEM; 960 961 /* set a block size */ 962 if (unlikely(!sb_set_blocksize(sb, F2FS_BLKSIZE))) { 963 f2fs_msg(sb, KERN_ERR, "unable to set blocksize"); 964 goto free_sbi; 965 } 966 967 err = read_raw_super_block(sb, &raw_super, &raw_super_buf); 968 if (err) 969 goto free_sbi; 970 971 sb->s_fs_info = sbi; 972 /* init some FS parameters */ 973 sbi->active_logs = NR_CURSEG_TYPE; 974 975 set_opt(sbi, BG_GC); 976 977 #ifdef CONFIG_F2FS_FS_XATTR 978 set_opt(sbi, XATTR_USER); 979 #endif 980 #ifdef CONFIG_F2FS_FS_POSIX_ACL 981 set_opt(sbi, POSIX_ACL); 982 #endif 983 /* parse mount options */ 984 err = parse_options(sb, (char *)data); 985 if (err) 986 goto free_sb_buf; 987 988 sb->s_maxbytes = max_file_size(le32_to_cpu(raw_super->log_blocksize)); 989 sb->s_max_links = F2FS_LINK_MAX; 990 get_random_bytes(&sbi->s_next_generation, sizeof(u32)); 991 992 sb->s_op = &f2fs_sops; 993 sb->s_xattr = f2fs_xattr_handlers; 994 sb->s_export_op = &f2fs_export_ops; 995 sb->s_magic = F2FS_SUPER_MAGIC; 996 sb->s_time_gran = 1; 997 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) | 998 (test_opt(sbi, POSIX_ACL) ? MS_POSIXACL : 0); 999 memcpy(sb->s_uuid, raw_super->uuid, sizeof(raw_super->uuid)); 1000 1001 /* init f2fs-specific super block info */ 1002 sbi->sb = sb; 1003 sbi->raw_super = raw_super; 1004 sbi->raw_super_buf = raw_super_buf; 1005 mutex_init(&sbi->gc_mutex); 1006 mutex_init(&sbi->writepages); 1007 mutex_init(&sbi->cp_mutex); 1008 init_rwsem(&sbi->node_write); 1009 sbi->por_doing = false; 1010 spin_lock_init(&sbi->stat_lock); 1011 1012 init_rwsem(&sbi->read_io.io_rwsem); 1013 sbi->read_io.sbi = sbi; 1014 sbi->read_io.bio = NULL; 1015 for (i = 0; i < NR_PAGE_TYPE; i++) { 1016 init_rwsem(&sbi->write_io[i].io_rwsem); 1017 sbi->write_io[i].sbi = sbi; 1018 sbi->write_io[i].bio = NULL; 1019 } 1020 1021 init_rwsem(&sbi->cp_rwsem); 1022 init_waitqueue_head(&sbi->cp_wait); 1023 init_sb_info(sbi); 1024 1025 /* get an inode for meta space */ 1026 sbi->meta_inode = f2fs_iget(sb, F2FS_META_INO(sbi)); 1027 if (IS_ERR(sbi->meta_inode)) { 1028 f2fs_msg(sb, KERN_ERR, "Failed to read F2FS meta data inode"); 1029 err = PTR_ERR(sbi->meta_inode); 1030 goto free_sb_buf; 1031 } 1032 1033 err = get_valid_checkpoint(sbi); 1034 if (err) { 1035 f2fs_msg(sb, KERN_ERR, "Failed to get valid F2FS checkpoint"); 1036 goto free_meta_inode; 1037 } 1038 1039 /* sanity checking of checkpoint */ 1040 err = -EINVAL; 1041 if (sanity_check_ckpt(sbi)) { 1042 f2fs_msg(sb, KERN_ERR, "Invalid F2FS checkpoint"); 1043 goto free_cp; 1044 } 1045 1046 sbi->total_valid_node_count = 1047 le32_to_cpu(sbi->ckpt->valid_node_count); 1048 sbi->total_valid_inode_count = 1049 le32_to_cpu(sbi->ckpt->valid_inode_count); 1050 sbi->user_block_count = le64_to_cpu(sbi->ckpt->user_block_count); 1051 sbi->total_valid_block_count = 1052 le64_to_cpu(sbi->ckpt->valid_block_count); 1053 sbi->last_valid_block_count = sbi->total_valid_block_count; 1054 sbi->alloc_valid_block_count = 0; 1055 INIT_LIST_HEAD(&sbi->dir_inode_list); 1056 spin_lock_init(&sbi->dir_inode_lock); 1057 1058 init_ino_entry_info(sbi); 1059 1060 /* setup f2fs internal modules */ 1061 err = build_segment_manager(sbi); 1062 if (err) { 1063 f2fs_msg(sb, KERN_ERR, 1064 "Failed to initialize F2FS segment manager"); 1065 goto free_sm; 1066 } 1067 err = build_node_manager(sbi); 1068 if (err) { 1069 f2fs_msg(sb, KERN_ERR, 1070 "Failed to initialize F2FS node manager"); 1071 goto free_nm; 1072 } 1073 1074 build_gc_manager(sbi); 1075 1076 /* get an inode for node space */ 1077 sbi->node_inode = f2fs_iget(sb, F2FS_NODE_INO(sbi)); 1078 if (IS_ERR(sbi->node_inode)) { 1079 f2fs_msg(sb, KERN_ERR, "Failed to read node inode"); 1080 err = PTR_ERR(sbi->node_inode); 1081 goto free_nm; 1082 } 1083 1084 /* if there are nt orphan nodes free them */ 1085 recover_orphan_inodes(sbi); 1086 1087 /* read root inode and dentry */ 1088 root = f2fs_iget(sb, F2FS_ROOT_INO(sbi)); 1089 if (IS_ERR(root)) { 1090 f2fs_msg(sb, KERN_ERR, "Failed to read root inode"); 1091 err = PTR_ERR(root); 1092 goto free_node_inode; 1093 } 1094 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 1095 iput(root); 1096 err = -EINVAL; 1097 goto free_node_inode; 1098 } 1099 1100 sb->s_root = d_make_root(root); /* allocate root dentry */ 1101 if (!sb->s_root) { 1102 err = -ENOMEM; 1103 goto free_root_inode; 1104 } 1105 1106 err = f2fs_build_stats(sbi); 1107 if (err) 1108 goto free_root_inode; 1109 1110 if (f2fs_proc_root) 1111 sbi->s_proc = proc_mkdir(sb->s_id, f2fs_proc_root); 1112 1113 if (sbi->s_proc) 1114 proc_create_data("segment_info", S_IRUGO, sbi->s_proc, 1115 &f2fs_seq_segment_info_fops, sb); 1116 1117 if (test_opt(sbi, DISCARD)) { 1118 struct request_queue *q = bdev_get_queue(sb->s_bdev); 1119 if (!blk_queue_discard(q)) 1120 f2fs_msg(sb, KERN_WARNING, 1121 "mounting with \"discard\" option, but " 1122 "the device does not support discard"); 1123 } 1124 1125 sbi->s_kobj.kset = f2fs_kset; 1126 init_completion(&sbi->s_kobj_unregister); 1127 err = kobject_init_and_add(&sbi->s_kobj, &f2fs_ktype, NULL, 1128 "%s", sb->s_id); 1129 if (err) 1130 goto free_proc; 1131 1132 if (!retry) 1133 sbi->need_fsck = true; 1134 1135 /* recover fsynced data */ 1136 if (!test_opt(sbi, DISABLE_ROLL_FORWARD)) { 1137 err = recover_fsync_data(sbi); 1138 if (err) { 1139 f2fs_msg(sb, KERN_ERR, 1140 "Cannot recover all fsync data errno=%ld", err); 1141 goto free_kobj; 1142 } 1143 } 1144 1145 /* 1146 * If filesystem is not mounted as read-only then 1147 * do start the gc_thread. 1148 */ 1149 if (test_opt(sbi, BG_GC) && !f2fs_readonly(sb)) { 1150 /* After POR, we can run background GC thread.*/ 1151 err = start_gc_thread(sbi); 1152 if (err) 1153 goto free_kobj; 1154 } 1155 return 0; 1156 1157 free_kobj: 1158 kobject_del(&sbi->s_kobj); 1159 free_proc: 1160 if (sbi->s_proc) { 1161 remove_proc_entry("segment_info", sbi->s_proc); 1162 remove_proc_entry(sb->s_id, f2fs_proc_root); 1163 } 1164 f2fs_destroy_stats(sbi); 1165 free_root_inode: 1166 dput(sb->s_root); 1167 sb->s_root = NULL; 1168 free_node_inode: 1169 iput(sbi->node_inode); 1170 free_nm: 1171 destroy_node_manager(sbi); 1172 free_sm: 1173 destroy_segment_manager(sbi); 1174 free_cp: 1175 kfree(sbi->ckpt); 1176 free_meta_inode: 1177 make_bad_inode(sbi->meta_inode); 1178 iput(sbi->meta_inode); 1179 free_sb_buf: 1180 brelse(raw_super_buf); 1181 free_sbi: 1182 kfree(sbi); 1183 1184 /* give only one another chance */ 1185 if (retry) { 1186 retry = 0; 1187 shrink_dcache_sb(sb); 1188 goto try_onemore; 1189 } 1190 return err; 1191 } 1192 1193 static struct dentry *f2fs_mount(struct file_system_type *fs_type, int flags, 1194 const char *dev_name, void *data) 1195 { 1196 return mount_bdev(fs_type, flags, dev_name, data, f2fs_fill_super); 1197 } 1198 1199 static void kill_f2fs_super(struct super_block *sb) 1200 { 1201 if (sb->s_root) 1202 F2FS_SB(sb)->s_closing = true; 1203 kill_block_super(sb); 1204 } 1205 1206 static struct file_system_type f2fs_fs_type = { 1207 .owner = THIS_MODULE, 1208 .name = "f2fs", 1209 .mount = f2fs_mount, 1210 .kill_sb = kill_f2fs_super, 1211 .fs_flags = FS_REQUIRES_DEV, 1212 }; 1213 MODULE_ALIAS_FS("f2fs"); 1214 1215 static int __init init_inodecache(void) 1216 { 1217 f2fs_inode_cachep = f2fs_kmem_cache_create("f2fs_inode_cache", 1218 sizeof(struct f2fs_inode_info)); 1219 if (!f2fs_inode_cachep) 1220 return -ENOMEM; 1221 return 0; 1222 } 1223 1224 static void destroy_inodecache(void) 1225 { 1226 /* 1227 * Make sure all delayed rcu free inodes are flushed before we 1228 * destroy cache. 1229 */ 1230 rcu_barrier(); 1231 kmem_cache_destroy(f2fs_inode_cachep); 1232 } 1233 1234 static int __init init_f2fs_fs(void) 1235 { 1236 int err; 1237 1238 f2fs_build_trace_ios(); 1239 1240 err = init_inodecache(); 1241 if (err) 1242 goto fail; 1243 err = create_node_manager_caches(); 1244 if (err) 1245 goto free_inodecache; 1246 err = create_segment_manager_caches(); 1247 if (err) 1248 goto free_node_manager_caches; 1249 err = create_checkpoint_caches(); 1250 if (err) 1251 goto free_segment_manager_caches; 1252 f2fs_kset = kset_create_and_add("f2fs", NULL, fs_kobj); 1253 if (!f2fs_kset) { 1254 err = -ENOMEM; 1255 goto free_checkpoint_caches; 1256 } 1257 err = register_filesystem(&f2fs_fs_type); 1258 if (err) 1259 goto free_kset; 1260 f2fs_create_root_stats(); 1261 f2fs_proc_root = proc_mkdir("fs/f2fs", NULL); 1262 return 0; 1263 1264 free_kset: 1265 kset_unregister(f2fs_kset); 1266 free_checkpoint_caches: 1267 destroy_checkpoint_caches(); 1268 free_segment_manager_caches: 1269 destroy_segment_manager_caches(); 1270 free_node_manager_caches: 1271 destroy_node_manager_caches(); 1272 free_inodecache: 1273 destroy_inodecache(); 1274 fail: 1275 return err; 1276 } 1277 1278 static void __exit exit_f2fs_fs(void) 1279 { 1280 remove_proc_entry("fs/f2fs", NULL); 1281 f2fs_destroy_root_stats(); 1282 unregister_filesystem(&f2fs_fs_type); 1283 destroy_checkpoint_caches(); 1284 destroy_segment_manager_caches(); 1285 destroy_node_manager_caches(); 1286 destroy_inodecache(); 1287 kset_unregister(f2fs_kset); 1288 f2fs_destroy_trace_ios(); 1289 } 1290 1291 module_init(init_f2fs_fs) 1292 module_exit(exit_f2fs_fs) 1293 1294 MODULE_AUTHOR("Samsung Electronics's Praesto Team"); 1295 MODULE_DESCRIPTION("Flash Friendly File System"); 1296 MODULE_LICENSE("GPL"); 1297