1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/node.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/f2fs_fs.h> 10 #include <linux/mpage.h> 11 #include <linux/backing-dev.h> 12 #include <linux/blkdev.h> 13 #include <linux/pagevec.h> 14 #include <linux/swap.h> 15 16 #include "f2fs.h" 17 #include "node.h" 18 #include "segment.h" 19 #include "xattr.h" 20 #include "trace.h" 21 #include <trace/events/f2fs.h> 22 23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock) 24 25 static struct kmem_cache *nat_entry_slab; 26 static struct kmem_cache *free_nid_slab; 27 static struct kmem_cache *nat_entry_set_slab; 28 static struct kmem_cache *fsync_node_entry_slab; 29 30 /* 31 * Check whether the given nid is within node id range. 32 */ 33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid) 34 { 35 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) { 36 set_sbi_flag(sbi, SBI_NEED_FSCK); 37 f2fs_msg(sbi->sb, KERN_WARNING, 38 "%s: out-of-range nid=%x, run fsck to fix.", 39 __func__, nid); 40 return -EINVAL; 41 } 42 return 0; 43 } 44 45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type) 46 { 47 struct f2fs_nm_info *nm_i = NM_I(sbi); 48 struct sysinfo val; 49 unsigned long avail_ram; 50 unsigned long mem_size = 0; 51 bool res = false; 52 53 si_meminfo(&val); 54 55 /* only uses low memory */ 56 avail_ram = val.totalram - val.totalhigh; 57 58 /* 59 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively 60 */ 61 if (type == FREE_NIDS) { 62 mem_size = (nm_i->nid_cnt[FREE_NID] * 63 sizeof(struct free_nid)) >> PAGE_SHIFT; 64 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 65 } else if (type == NAT_ENTRIES) { 66 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >> 67 PAGE_SHIFT; 68 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 69 if (excess_cached_nats(sbi)) 70 res = false; 71 } else if (type == DIRTY_DENTS) { 72 if (sbi->sb->s_bdi->wb.dirty_exceeded) 73 return false; 74 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS); 75 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 76 } else if (type == INO_ENTRIES) { 77 int i; 78 79 for (i = 0; i < MAX_INO_ENTRY; i++) 80 mem_size += sbi->im[i].ino_num * 81 sizeof(struct ino_entry); 82 mem_size >>= PAGE_SHIFT; 83 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 84 } else if (type == EXTENT_CACHE) { 85 mem_size = (atomic_read(&sbi->total_ext_tree) * 86 sizeof(struct extent_tree) + 87 atomic_read(&sbi->total_ext_node) * 88 sizeof(struct extent_node)) >> PAGE_SHIFT; 89 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 90 } else if (type == INMEM_PAGES) { 91 /* it allows 20% / total_ram for inmemory pages */ 92 mem_size = get_pages(sbi, F2FS_INMEM_PAGES); 93 res = mem_size < (val.totalram / 5); 94 } else { 95 if (!sbi->sb->s_bdi->wb.dirty_exceeded) 96 return true; 97 } 98 return res; 99 } 100 101 static void clear_node_page_dirty(struct page *page) 102 { 103 if (PageDirty(page)) { 104 f2fs_clear_page_cache_dirty_tag(page); 105 clear_page_dirty_for_io(page); 106 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES); 107 } 108 ClearPageUptodate(page); 109 } 110 111 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 112 { 113 return f2fs_get_meta_page_nofail(sbi, current_nat_addr(sbi, nid)); 114 } 115 116 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 117 { 118 struct page *src_page; 119 struct page *dst_page; 120 pgoff_t dst_off; 121 void *src_addr; 122 void *dst_addr; 123 struct f2fs_nm_info *nm_i = NM_I(sbi); 124 125 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid)); 126 127 /* get current nat block page with lock */ 128 src_page = get_current_nat_page(sbi, nid); 129 if (IS_ERR(src_page)) 130 return src_page; 131 dst_page = f2fs_grab_meta_page(sbi, dst_off); 132 f2fs_bug_on(sbi, PageDirty(src_page)); 133 134 src_addr = page_address(src_page); 135 dst_addr = page_address(dst_page); 136 memcpy(dst_addr, src_addr, PAGE_SIZE); 137 set_page_dirty(dst_page); 138 f2fs_put_page(src_page, 1); 139 140 set_to_next_nat(nm_i, nid); 141 142 return dst_page; 143 } 144 145 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail) 146 { 147 struct nat_entry *new; 148 149 if (no_fail) 150 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO); 151 else 152 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO); 153 if (new) { 154 nat_set_nid(new, nid); 155 nat_reset_flag(new); 156 } 157 return new; 158 } 159 160 static void __free_nat_entry(struct nat_entry *e) 161 { 162 kmem_cache_free(nat_entry_slab, e); 163 } 164 165 /* must be locked by nat_tree_lock */ 166 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i, 167 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail) 168 { 169 if (no_fail) 170 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne); 171 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne)) 172 return NULL; 173 174 if (raw_ne) 175 node_info_from_raw_nat(&ne->ni, raw_ne); 176 177 spin_lock(&nm_i->nat_list_lock); 178 list_add_tail(&ne->list, &nm_i->nat_entries); 179 spin_unlock(&nm_i->nat_list_lock); 180 181 nm_i->nat_cnt++; 182 return ne; 183 } 184 185 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n) 186 { 187 struct nat_entry *ne; 188 189 ne = radix_tree_lookup(&nm_i->nat_root, n); 190 191 /* for recent accessed nat entry, move it to tail of lru list */ 192 if (ne && !get_nat_flag(ne, IS_DIRTY)) { 193 spin_lock(&nm_i->nat_list_lock); 194 if (!list_empty(&ne->list)) 195 list_move_tail(&ne->list, &nm_i->nat_entries); 196 spin_unlock(&nm_i->nat_list_lock); 197 } 198 199 return ne; 200 } 201 202 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i, 203 nid_t start, unsigned int nr, struct nat_entry **ep) 204 { 205 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr); 206 } 207 208 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e) 209 { 210 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e)); 211 nm_i->nat_cnt--; 212 __free_nat_entry(e); 213 } 214 215 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i, 216 struct nat_entry *ne) 217 { 218 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid); 219 struct nat_entry_set *head; 220 221 head = radix_tree_lookup(&nm_i->nat_set_root, set); 222 if (!head) { 223 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS); 224 225 INIT_LIST_HEAD(&head->entry_list); 226 INIT_LIST_HEAD(&head->set_list); 227 head->set = set; 228 head->entry_cnt = 0; 229 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head); 230 } 231 return head; 232 } 233 234 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i, 235 struct nat_entry *ne) 236 { 237 struct nat_entry_set *head; 238 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR; 239 240 if (!new_ne) 241 head = __grab_nat_entry_set(nm_i, ne); 242 243 /* 244 * update entry_cnt in below condition: 245 * 1. update NEW_ADDR to valid block address; 246 * 2. update old block address to new one; 247 */ 248 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) || 249 !get_nat_flag(ne, IS_DIRTY))) 250 head->entry_cnt++; 251 252 set_nat_flag(ne, IS_PREALLOC, new_ne); 253 254 if (get_nat_flag(ne, IS_DIRTY)) 255 goto refresh_list; 256 257 nm_i->dirty_nat_cnt++; 258 set_nat_flag(ne, IS_DIRTY, true); 259 refresh_list: 260 spin_lock(&nm_i->nat_list_lock); 261 if (new_ne) 262 list_del_init(&ne->list); 263 else 264 list_move_tail(&ne->list, &head->entry_list); 265 spin_unlock(&nm_i->nat_list_lock); 266 } 267 268 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i, 269 struct nat_entry_set *set, struct nat_entry *ne) 270 { 271 spin_lock(&nm_i->nat_list_lock); 272 list_move_tail(&ne->list, &nm_i->nat_entries); 273 spin_unlock(&nm_i->nat_list_lock); 274 275 set_nat_flag(ne, IS_DIRTY, false); 276 set->entry_cnt--; 277 nm_i->dirty_nat_cnt--; 278 } 279 280 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i, 281 nid_t start, unsigned int nr, struct nat_entry_set **ep) 282 { 283 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep, 284 start, nr); 285 } 286 287 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page) 288 { 289 return NODE_MAPPING(sbi) == page->mapping && 290 IS_DNODE(page) && is_cold_node(page); 291 } 292 293 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi) 294 { 295 spin_lock_init(&sbi->fsync_node_lock); 296 INIT_LIST_HEAD(&sbi->fsync_node_list); 297 sbi->fsync_seg_id = 0; 298 sbi->fsync_node_num = 0; 299 } 300 301 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi, 302 struct page *page) 303 { 304 struct fsync_node_entry *fn; 305 unsigned long flags; 306 unsigned int seq_id; 307 308 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS); 309 310 get_page(page); 311 fn->page = page; 312 INIT_LIST_HEAD(&fn->list); 313 314 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 315 list_add_tail(&fn->list, &sbi->fsync_node_list); 316 fn->seq_id = sbi->fsync_seg_id++; 317 seq_id = fn->seq_id; 318 sbi->fsync_node_num++; 319 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 320 321 return seq_id; 322 } 323 324 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page) 325 { 326 struct fsync_node_entry *fn; 327 unsigned long flags; 328 329 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 330 list_for_each_entry(fn, &sbi->fsync_node_list, list) { 331 if (fn->page == page) { 332 list_del(&fn->list); 333 sbi->fsync_node_num--; 334 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 335 kmem_cache_free(fsync_node_entry_slab, fn); 336 put_page(page); 337 return; 338 } 339 } 340 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 341 f2fs_bug_on(sbi, 1); 342 } 343 344 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi) 345 { 346 unsigned long flags; 347 348 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 349 sbi->fsync_seg_id = 0; 350 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 351 } 352 353 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid) 354 { 355 struct f2fs_nm_info *nm_i = NM_I(sbi); 356 struct nat_entry *e; 357 bool need = false; 358 359 down_read(&nm_i->nat_tree_lock); 360 e = __lookup_nat_cache(nm_i, nid); 361 if (e) { 362 if (!get_nat_flag(e, IS_CHECKPOINTED) && 363 !get_nat_flag(e, HAS_FSYNCED_INODE)) 364 need = true; 365 } 366 up_read(&nm_i->nat_tree_lock); 367 return need; 368 } 369 370 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid) 371 { 372 struct f2fs_nm_info *nm_i = NM_I(sbi); 373 struct nat_entry *e; 374 bool is_cp = true; 375 376 down_read(&nm_i->nat_tree_lock); 377 e = __lookup_nat_cache(nm_i, nid); 378 if (e && !get_nat_flag(e, IS_CHECKPOINTED)) 379 is_cp = false; 380 up_read(&nm_i->nat_tree_lock); 381 return is_cp; 382 } 383 384 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino) 385 { 386 struct f2fs_nm_info *nm_i = NM_I(sbi); 387 struct nat_entry *e; 388 bool need_update = true; 389 390 down_read(&nm_i->nat_tree_lock); 391 e = __lookup_nat_cache(nm_i, ino); 392 if (e && get_nat_flag(e, HAS_LAST_FSYNC) && 393 (get_nat_flag(e, IS_CHECKPOINTED) || 394 get_nat_flag(e, HAS_FSYNCED_INODE))) 395 need_update = false; 396 up_read(&nm_i->nat_tree_lock); 397 return need_update; 398 } 399 400 /* must be locked by nat_tree_lock */ 401 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid, 402 struct f2fs_nat_entry *ne) 403 { 404 struct f2fs_nm_info *nm_i = NM_I(sbi); 405 struct nat_entry *new, *e; 406 407 new = __alloc_nat_entry(nid, false); 408 if (!new) 409 return; 410 411 down_write(&nm_i->nat_tree_lock); 412 e = __lookup_nat_cache(nm_i, nid); 413 if (!e) 414 e = __init_nat_entry(nm_i, new, ne, false); 415 else 416 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) || 417 nat_get_blkaddr(e) != 418 le32_to_cpu(ne->block_addr) || 419 nat_get_version(e) != ne->version); 420 up_write(&nm_i->nat_tree_lock); 421 if (e != new) 422 __free_nat_entry(new); 423 } 424 425 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni, 426 block_t new_blkaddr, bool fsync_done) 427 { 428 struct f2fs_nm_info *nm_i = NM_I(sbi); 429 struct nat_entry *e; 430 struct nat_entry *new = __alloc_nat_entry(ni->nid, true); 431 432 down_write(&nm_i->nat_tree_lock); 433 e = __lookup_nat_cache(nm_i, ni->nid); 434 if (!e) { 435 e = __init_nat_entry(nm_i, new, NULL, true); 436 copy_node_info(&e->ni, ni); 437 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR); 438 } else if (new_blkaddr == NEW_ADDR) { 439 /* 440 * when nid is reallocated, 441 * previous nat entry can be remained in nat cache. 442 * So, reinitialize it with new information. 443 */ 444 copy_node_info(&e->ni, ni); 445 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR); 446 } 447 /* let's free early to reduce memory consumption */ 448 if (e != new) 449 __free_nat_entry(new); 450 451 /* sanity check */ 452 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr); 453 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR && 454 new_blkaddr == NULL_ADDR); 455 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR && 456 new_blkaddr == NEW_ADDR); 457 f2fs_bug_on(sbi, is_valid_data_blkaddr(sbi, nat_get_blkaddr(e)) && 458 new_blkaddr == NEW_ADDR); 459 460 /* increment version no as node is removed */ 461 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) { 462 unsigned char version = nat_get_version(e); 463 nat_set_version(e, inc_node_version(version)); 464 } 465 466 /* change address */ 467 nat_set_blkaddr(e, new_blkaddr); 468 if (!is_valid_data_blkaddr(sbi, new_blkaddr)) 469 set_nat_flag(e, IS_CHECKPOINTED, false); 470 __set_nat_cache_dirty(nm_i, e); 471 472 /* update fsync_mark if its inode nat entry is still alive */ 473 if (ni->nid != ni->ino) 474 e = __lookup_nat_cache(nm_i, ni->ino); 475 if (e) { 476 if (fsync_done && ni->nid == ni->ino) 477 set_nat_flag(e, HAS_FSYNCED_INODE, true); 478 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done); 479 } 480 up_write(&nm_i->nat_tree_lock); 481 } 482 483 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink) 484 { 485 struct f2fs_nm_info *nm_i = NM_I(sbi); 486 int nr = nr_shrink; 487 488 if (!down_write_trylock(&nm_i->nat_tree_lock)) 489 return 0; 490 491 spin_lock(&nm_i->nat_list_lock); 492 while (nr_shrink) { 493 struct nat_entry *ne; 494 495 if (list_empty(&nm_i->nat_entries)) 496 break; 497 498 ne = list_first_entry(&nm_i->nat_entries, 499 struct nat_entry, list); 500 list_del(&ne->list); 501 spin_unlock(&nm_i->nat_list_lock); 502 503 __del_from_nat_cache(nm_i, ne); 504 nr_shrink--; 505 506 spin_lock(&nm_i->nat_list_lock); 507 } 508 spin_unlock(&nm_i->nat_list_lock); 509 510 up_write(&nm_i->nat_tree_lock); 511 return nr - nr_shrink; 512 } 513 514 /* 515 * This function always returns success 516 */ 517 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid, 518 struct node_info *ni) 519 { 520 struct f2fs_nm_info *nm_i = NM_I(sbi); 521 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 522 struct f2fs_journal *journal = curseg->journal; 523 nid_t start_nid = START_NID(nid); 524 struct f2fs_nat_block *nat_blk; 525 struct page *page = NULL; 526 struct f2fs_nat_entry ne; 527 struct nat_entry *e; 528 pgoff_t index; 529 int i; 530 531 ni->nid = nid; 532 533 /* Check nat cache */ 534 down_read(&nm_i->nat_tree_lock); 535 e = __lookup_nat_cache(nm_i, nid); 536 if (e) { 537 ni->ino = nat_get_ino(e); 538 ni->blk_addr = nat_get_blkaddr(e); 539 ni->version = nat_get_version(e); 540 up_read(&nm_i->nat_tree_lock); 541 return 0; 542 } 543 544 memset(&ne, 0, sizeof(struct f2fs_nat_entry)); 545 546 /* Check current segment summary */ 547 down_read(&curseg->journal_rwsem); 548 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0); 549 if (i >= 0) { 550 ne = nat_in_journal(journal, i); 551 node_info_from_raw_nat(ni, &ne); 552 } 553 up_read(&curseg->journal_rwsem); 554 if (i >= 0) { 555 up_read(&nm_i->nat_tree_lock); 556 goto cache; 557 } 558 559 /* Fill node_info from nat page */ 560 index = current_nat_addr(sbi, nid); 561 up_read(&nm_i->nat_tree_lock); 562 563 page = f2fs_get_meta_page(sbi, index); 564 if (IS_ERR(page)) 565 return PTR_ERR(page); 566 567 nat_blk = (struct f2fs_nat_block *)page_address(page); 568 ne = nat_blk->entries[nid - start_nid]; 569 node_info_from_raw_nat(ni, &ne); 570 f2fs_put_page(page, 1); 571 cache: 572 /* cache nat entry */ 573 cache_nat_entry(sbi, nid, &ne); 574 return 0; 575 } 576 577 /* 578 * readahead MAX_RA_NODE number of node pages. 579 */ 580 static void f2fs_ra_node_pages(struct page *parent, int start, int n) 581 { 582 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 583 struct blk_plug plug; 584 int i, end; 585 nid_t nid; 586 587 blk_start_plug(&plug); 588 589 /* Then, try readahead for siblings of the desired node */ 590 end = start + n; 591 end = min(end, NIDS_PER_BLOCK); 592 for (i = start; i < end; i++) { 593 nid = get_nid(parent, i, false); 594 f2fs_ra_node_page(sbi, nid); 595 } 596 597 blk_finish_plug(&plug); 598 } 599 600 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs) 601 { 602 const long direct_index = ADDRS_PER_INODE(dn->inode); 603 const long direct_blks = ADDRS_PER_BLOCK; 604 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK; 605 unsigned int skipped_unit = ADDRS_PER_BLOCK; 606 int cur_level = dn->cur_level; 607 int max_level = dn->max_level; 608 pgoff_t base = 0; 609 610 if (!dn->max_level) 611 return pgofs + 1; 612 613 while (max_level-- > cur_level) 614 skipped_unit *= NIDS_PER_BLOCK; 615 616 switch (dn->max_level) { 617 case 3: 618 base += 2 * indirect_blks; 619 /* fall through */ 620 case 2: 621 base += 2 * direct_blks; 622 /* fall through */ 623 case 1: 624 base += direct_index; 625 break; 626 default: 627 f2fs_bug_on(F2FS_I_SB(dn->inode), 1); 628 } 629 630 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base; 631 } 632 633 /* 634 * The maximum depth is four. 635 * Offset[0] will have raw inode offset. 636 */ 637 static int get_node_path(struct inode *inode, long block, 638 int offset[4], unsigned int noffset[4]) 639 { 640 const long direct_index = ADDRS_PER_INODE(inode); 641 const long direct_blks = ADDRS_PER_BLOCK; 642 const long dptrs_per_blk = NIDS_PER_BLOCK; 643 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK; 644 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK; 645 int n = 0; 646 int level = 0; 647 648 noffset[0] = 0; 649 650 if (block < direct_index) { 651 offset[n] = block; 652 goto got; 653 } 654 block -= direct_index; 655 if (block < direct_blks) { 656 offset[n++] = NODE_DIR1_BLOCK; 657 noffset[n] = 1; 658 offset[n] = block; 659 level = 1; 660 goto got; 661 } 662 block -= direct_blks; 663 if (block < direct_blks) { 664 offset[n++] = NODE_DIR2_BLOCK; 665 noffset[n] = 2; 666 offset[n] = block; 667 level = 1; 668 goto got; 669 } 670 block -= direct_blks; 671 if (block < indirect_blks) { 672 offset[n++] = NODE_IND1_BLOCK; 673 noffset[n] = 3; 674 offset[n++] = block / direct_blks; 675 noffset[n] = 4 + offset[n - 1]; 676 offset[n] = block % direct_blks; 677 level = 2; 678 goto got; 679 } 680 block -= indirect_blks; 681 if (block < indirect_blks) { 682 offset[n++] = NODE_IND2_BLOCK; 683 noffset[n] = 4 + dptrs_per_blk; 684 offset[n++] = block / direct_blks; 685 noffset[n] = 5 + dptrs_per_blk + offset[n - 1]; 686 offset[n] = block % direct_blks; 687 level = 2; 688 goto got; 689 } 690 block -= indirect_blks; 691 if (block < dindirect_blks) { 692 offset[n++] = NODE_DIND_BLOCK; 693 noffset[n] = 5 + (dptrs_per_blk * 2); 694 offset[n++] = block / indirect_blks; 695 noffset[n] = 6 + (dptrs_per_blk * 2) + 696 offset[n - 1] * (dptrs_per_blk + 1); 697 offset[n++] = (block / direct_blks) % dptrs_per_blk; 698 noffset[n] = 7 + (dptrs_per_blk * 2) + 699 offset[n - 2] * (dptrs_per_blk + 1) + 700 offset[n - 1]; 701 offset[n] = block % direct_blks; 702 level = 3; 703 goto got; 704 } else { 705 return -E2BIG; 706 } 707 got: 708 return level; 709 } 710 711 /* 712 * Caller should call f2fs_put_dnode(dn). 713 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and 714 * f2fs_unlock_op() only if ro is not set RDONLY_NODE. 715 * In the case of RDONLY_NODE, we don't need to care about mutex. 716 */ 717 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode) 718 { 719 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 720 struct page *npage[4]; 721 struct page *parent = NULL; 722 int offset[4]; 723 unsigned int noffset[4]; 724 nid_t nids[4]; 725 int level, i = 0; 726 int err = 0; 727 728 level = get_node_path(dn->inode, index, offset, noffset); 729 if (level < 0) 730 return level; 731 732 nids[0] = dn->inode->i_ino; 733 npage[0] = dn->inode_page; 734 735 if (!npage[0]) { 736 npage[0] = f2fs_get_node_page(sbi, nids[0]); 737 if (IS_ERR(npage[0])) 738 return PTR_ERR(npage[0]); 739 } 740 741 /* if inline_data is set, should not report any block indices */ 742 if (f2fs_has_inline_data(dn->inode) && index) { 743 err = -ENOENT; 744 f2fs_put_page(npage[0], 1); 745 goto release_out; 746 } 747 748 parent = npage[0]; 749 if (level != 0) 750 nids[1] = get_nid(parent, offset[0], true); 751 dn->inode_page = npage[0]; 752 dn->inode_page_locked = true; 753 754 /* get indirect or direct nodes */ 755 for (i = 1; i <= level; i++) { 756 bool done = false; 757 758 if (!nids[i] && mode == ALLOC_NODE) { 759 /* alloc new node */ 760 if (!f2fs_alloc_nid(sbi, &(nids[i]))) { 761 err = -ENOSPC; 762 goto release_pages; 763 } 764 765 dn->nid = nids[i]; 766 npage[i] = f2fs_new_node_page(dn, noffset[i]); 767 if (IS_ERR(npage[i])) { 768 f2fs_alloc_nid_failed(sbi, nids[i]); 769 err = PTR_ERR(npage[i]); 770 goto release_pages; 771 } 772 773 set_nid(parent, offset[i - 1], nids[i], i == 1); 774 f2fs_alloc_nid_done(sbi, nids[i]); 775 done = true; 776 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) { 777 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]); 778 if (IS_ERR(npage[i])) { 779 err = PTR_ERR(npage[i]); 780 goto release_pages; 781 } 782 done = true; 783 } 784 if (i == 1) { 785 dn->inode_page_locked = false; 786 unlock_page(parent); 787 } else { 788 f2fs_put_page(parent, 1); 789 } 790 791 if (!done) { 792 npage[i] = f2fs_get_node_page(sbi, nids[i]); 793 if (IS_ERR(npage[i])) { 794 err = PTR_ERR(npage[i]); 795 f2fs_put_page(npage[0], 0); 796 goto release_out; 797 } 798 } 799 if (i < level) { 800 parent = npage[i]; 801 nids[i + 1] = get_nid(parent, offset[i], false); 802 } 803 } 804 dn->nid = nids[level]; 805 dn->ofs_in_node = offset[level]; 806 dn->node_page = npage[level]; 807 dn->data_blkaddr = datablock_addr(dn->inode, 808 dn->node_page, dn->ofs_in_node); 809 return 0; 810 811 release_pages: 812 f2fs_put_page(parent, 1); 813 if (i > 1) 814 f2fs_put_page(npage[0], 0); 815 release_out: 816 dn->inode_page = NULL; 817 dn->node_page = NULL; 818 if (err == -ENOENT) { 819 dn->cur_level = i; 820 dn->max_level = level; 821 dn->ofs_in_node = offset[level]; 822 } 823 return err; 824 } 825 826 static int truncate_node(struct dnode_of_data *dn) 827 { 828 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 829 struct node_info ni; 830 int err; 831 pgoff_t index; 832 833 err = f2fs_get_node_info(sbi, dn->nid, &ni); 834 if (err) 835 return err; 836 837 /* Deallocate node address */ 838 f2fs_invalidate_blocks(sbi, ni.blk_addr); 839 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino); 840 set_node_addr(sbi, &ni, NULL_ADDR, false); 841 842 if (dn->nid == dn->inode->i_ino) { 843 f2fs_remove_orphan_inode(sbi, dn->nid); 844 dec_valid_inode_count(sbi); 845 f2fs_inode_synced(dn->inode); 846 } 847 848 clear_node_page_dirty(dn->node_page); 849 set_sbi_flag(sbi, SBI_IS_DIRTY); 850 851 index = dn->node_page->index; 852 f2fs_put_page(dn->node_page, 1); 853 854 invalidate_mapping_pages(NODE_MAPPING(sbi), 855 index, index); 856 857 dn->node_page = NULL; 858 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr); 859 860 return 0; 861 } 862 863 static int truncate_dnode(struct dnode_of_data *dn) 864 { 865 struct page *page; 866 int err; 867 868 if (dn->nid == 0) 869 return 1; 870 871 /* get direct node */ 872 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 873 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT) 874 return 1; 875 else if (IS_ERR(page)) 876 return PTR_ERR(page); 877 878 /* Make dnode_of_data for parameter */ 879 dn->node_page = page; 880 dn->ofs_in_node = 0; 881 f2fs_truncate_data_blocks(dn); 882 err = truncate_node(dn); 883 if (err) 884 return err; 885 886 return 1; 887 } 888 889 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs, 890 int ofs, int depth) 891 { 892 struct dnode_of_data rdn = *dn; 893 struct page *page; 894 struct f2fs_node *rn; 895 nid_t child_nid; 896 unsigned int child_nofs; 897 int freed = 0; 898 int i, ret; 899 900 if (dn->nid == 0) 901 return NIDS_PER_BLOCK + 1; 902 903 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr); 904 905 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 906 if (IS_ERR(page)) { 907 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page)); 908 return PTR_ERR(page); 909 } 910 911 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK); 912 913 rn = F2FS_NODE(page); 914 if (depth < 3) { 915 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) { 916 child_nid = le32_to_cpu(rn->in.nid[i]); 917 if (child_nid == 0) 918 continue; 919 rdn.nid = child_nid; 920 ret = truncate_dnode(&rdn); 921 if (ret < 0) 922 goto out_err; 923 if (set_nid(page, i, 0, false)) 924 dn->node_changed = true; 925 } 926 } else { 927 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1; 928 for (i = ofs; i < NIDS_PER_BLOCK; i++) { 929 child_nid = le32_to_cpu(rn->in.nid[i]); 930 if (child_nid == 0) { 931 child_nofs += NIDS_PER_BLOCK + 1; 932 continue; 933 } 934 rdn.nid = child_nid; 935 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1); 936 if (ret == (NIDS_PER_BLOCK + 1)) { 937 if (set_nid(page, i, 0, false)) 938 dn->node_changed = true; 939 child_nofs += ret; 940 } else if (ret < 0 && ret != -ENOENT) { 941 goto out_err; 942 } 943 } 944 freed = child_nofs; 945 } 946 947 if (!ofs) { 948 /* remove current indirect node */ 949 dn->node_page = page; 950 ret = truncate_node(dn); 951 if (ret) 952 goto out_err; 953 freed++; 954 } else { 955 f2fs_put_page(page, 1); 956 } 957 trace_f2fs_truncate_nodes_exit(dn->inode, freed); 958 return freed; 959 960 out_err: 961 f2fs_put_page(page, 1); 962 trace_f2fs_truncate_nodes_exit(dn->inode, ret); 963 return ret; 964 } 965 966 static int truncate_partial_nodes(struct dnode_of_data *dn, 967 struct f2fs_inode *ri, int *offset, int depth) 968 { 969 struct page *pages[2]; 970 nid_t nid[3]; 971 nid_t child_nid; 972 int err = 0; 973 int i; 974 int idx = depth - 2; 975 976 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 977 if (!nid[0]) 978 return 0; 979 980 /* get indirect nodes in the path */ 981 for (i = 0; i < idx + 1; i++) { 982 /* reference count'll be increased */ 983 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]); 984 if (IS_ERR(pages[i])) { 985 err = PTR_ERR(pages[i]); 986 idx = i - 1; 987 goto fail; 988 } 989 nid[i + 1] = get_nid(pages[i], offset[i + 1], false); 990 } 991 992 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK); 993 994 /* free direct nodes linked to a partial indirect node */ 995 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) { 996 child_nid = get_nid(pages[idx], i, false); 997 if (!child_nid) 998 continue; 999 dn->nid = child_nid; 1000 err = truncate_dnode(dn); 1001 if (err < 0) 1002 goto fail; 1003 if (set_nid(pages[idx], i, 0, false)) 1004 dn->node_changed = true; 1005 } 1006 1007 if (offset[idx + 1] == 0) { 1008 dn->node_page = pages[idx]; 1009 dn->nid = nid[idx]; 1010 err = truncate_node(dn); 1011 if (err) 1012 goto fail; 1013 } else { 1014 f2fs_put_page(pages[idx], 1); 1015 } 1016 offset[idx]++; 1017 offset[idx + 1] = 0; 1018 idx--; 1019 fail: 1020 for (i = idx; i >= 0; i--) 1021 f2fs_put_page(pages[i], 1); 1022 1023 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err); 1024 1025 return err; 1026 } 1027 1028 /* 1029 * All the block addresses of data and nodes should be nullified. 1030 */ 1031 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from) 1032 { 1033 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1034 int err = 0, cont = 1; 1035 int level, offset[4], noffset[4]; 1036 unsigned int nofs = 0; 1037 struct f2fs_inode *ri; 1038 struct dnode_of_data dn; 1039 struct page *page; 1040 1041 trace_f2fs_truncate_inode_blocks_enter(inode, from); 1042 1043 level = get_node_path(inode, from, offset, noffset); 1044 if (level < 0) 1045 return level; 1046 1047 page = f2fs_get_node_page(sbi, inode->i_ino); 1048 if (IS_ERR(page)) { 1049 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page)); 1050 return PTR_ERR(page); 1051 } 1052 1053 set_new_dnode(&dn, inode, page, NULL, 0); 1054 unlock_page(page); 1055 1056 ri = F2FS_INODE(page); 1057 switch (level) { 1058 case 0: 1059 case 1: 1060 nofs = noffset[1]; 1061 break; 1062 case 2: 1063 nofs = noffset[1]; 1064 if (!offset[level - 1]) 1065 goto skip_partial; 1066 err = truncate_partial_nodes(&dn, ri, offset, level); 1067 if (err < 0 && err != -ENOENT) 1068 goto fail; 1069 nofs += 1 + NIDS_PER_BLOCK; 1070 break; 1071 case 3: 1072 nofs = 5 + 2 * NIDS_PER_BLOCK; 1073 if (!offset[level - 1]) 1074 goto skip_partial; 1075 err = truncate_partial_nodes(&dn, ri, offset, level); 1076 if (err < 0 && err != -ENOENT) 1077 goto fail; 1078 break; 1079 default: 1080 BUG(); 1081 } 1082 1083 skip_partial: 1084 while (cont) { 1085 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1086 switch (offset[0]) { 1087 case NODE_DIR1_BLOCK: 1088 case NODE_DIR2_BLOCK: 1089 err = truncate_dnode(&dn); 1090 break; 1091 1092 case NODE_IND1_BLOCK: 1093 case NODE_IND2_BLOCK: 1094 err = truncate_nodes(&dn, nofs, offset[1], 2); 1095 break; 1096 1097 case NODE_DIND_BLOCK: 1098 err = truncate_nodes(&dn, nofs, offset[1], 3); 1099 cont = 0; 1100 break; 1101 1102 default: 1103 BUG(); 1104 } 1105 if (err < 0 && err != -ENOENT) 1106 goto fail; 1107 if (offset[1] == 0 && 1108 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) { 1109 lock_page(page); 1110 BUG_ON(page->mapping != NODE_MAPPING(sbi)); 1111 f2fs_wait_on_page_writeback(page, NODE, true, true); 1112 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0; 1113 set_page_dirty(page); 1114 unlock_page(page); 1115 } 1116 offset[1] = 0; 1117 offset[0]++; 1118 nofs += err; 1119 } 1120 fail: 1121 f2fs_put_page(page, 0); 1122 trace_f2fs_truncate_inode_blocks_exit(inode, err); 1123 return err > 0 ? 0 : err; 1124 } 1125 1126 /* caller must lock inode page */ 1127 int f2fs_truncate_xattr_node(struct inode *inode) 1128 { 1129 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1130 nid_t nid = F2FS_I(inode)->i_xattr_nid; 1131 struct dnode_of_data dn; 1132 struct page *npage; 1133 int err; 1134 1135 if (!nid) 1136 return 0; 1137 1138 npage = f2fs_get_node_page(sbi, nid); 1139 if (IS_ERR(npage)) 1140 return PTR_ERR(npage); 1141 1142 set_new_dnode(&dn, inode, NULL, npage, nid); 1143 err = truncate_node(&dn); 1144 if (err) { 1145 f2fs_put_page(npage, 1); 1146 return err; 1147 } 1148 1149 f2fs_i_xnid_write(inode, 0); 1150 1151 return 0; 1152 } 1153 1154 /* 1155 * Caller should grab and release a rwsem by calling f2fs_lock_op() and 1156 * f2fs_unlock_op(). 1157 */ 1158 int f2fs_remove_inode_page(struct inode *inode) 1159 { 1160 struct dnode_of_data dn; 1161 int err; 1162 1163 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1164 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE); 1165 if (err) 1166 return err; 1167 1168 err = f2fs_truncate_xattr_node(inode); 1169 if (err) { 1170 f2fs_put_dnode(&dn); 1171 return err; 1172 } 1173 1174 /* remove potential inline_data blocks */ 1175 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1176 S_ISLNK(inode->i_mode)) 1177 f2fs_truncate_data_blocks_range(&dn, 1); 1178 1179 /* 0 is possible, after f2fs_new_inode() has failed */ 1180 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) { 1181 f2fs_put_dnode(&dn); 1182 return -EIO; 1183 } 1184 f2fs_bug_on(F2FS_I_SB(inode), 1185 inode->i_blocks != 0 && inode->i_blocks != 8); 1186 1187 /* will put inode & node pages */ 1188 err = truncate_node(&dn); 1189 if (err) { 1190 f2fs_put_dnode(&dn); 1191 return err; 1192 } 1193 return 0; 1194 } 1195 1196 struct page *f2fs_new_inode_page(struct inode *inode) 1197 { 1198 struct dnode_of_data dn; 1199 1200 /* allocate inode page for new inode */ 1201 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1202 1203 /* caller should f2fs_put_page(page, 1); */ 1204 return f2fs_new_node_page(&dn, 0); 1205 } 1206 1207 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs) 1208 { 1209 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1210 struct node_info new_ni; 1211 struct page *page; 1212 int err; 1213 1214 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1215 return ERR_PTR(-EPERM); 1216 1217 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false); 1218 if (!page) 1219 return ERR_PTR(-ENOMEM); 1220 1221 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs)))) 1222 goto fail; 1223 1224 #ifdef CONFIG_F2FS_CHECK_FS 1225 err = f2fs_get_node_info(sbi, dn->nid, &new_ni); 1226 if (err) { 1227 dec_valid_node_count(sbi, dn->inode, !ofs); 1228 goto fail; 1229 } 1230 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR); 1231 #endif 1232 new_ni.nid = dn->nid; 1233 new_ni.ino = dn->inode->i_ino; 1234 new_ni.blk_addr = NULL_ADDR; 1235 new_ni.flag = 0; 1236 new_ni.version = 0; 1237 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 1238 1239 f2fs_wait_on_page_writeback(page, NODE, true, true); 1240 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 1241 set_cold_node(page, S_ISDIR(dn->inode->i_mode)); 1242 if (!PageUptodate(page)) 1243 SetPageUptodate(page); 1244 if (set_page_dirty(page)) 1245 dn->node_changed = true; 1246 1247 if (f2fs_has_xattr_block(ofs)) 1248 f2fs_i_xnid_write(dn->inode, dn->nid); 1249 1250 if (ofs == 0) 1251 inc_valid_inode_count(sbi); 1252 return page; 1253 1254 fail: 1255 clear_node_page_dirty(page); 1256 f2fs_put_page(page, 1); 1257 return ERR_PTR(err); 1258 } 1259 1260 /* 1261 * Caller should do after getting the following values. 1262 * 0: f2fs_put_page(page, 0) 1263 * LOCKED_PAGE or error: f2fs_put_page(page, 1) 1264 */ 1265 static int read_node_page(struct page *page, int op_flags) 1266 { 1267 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1268 struct node_info ni; 1269 struct f2fs_io_info fio = { 1270 .sbi = sbi, 1271 .type = NODE, 1272 .op = REQ_OP_READ, 1273 .op_flags = op_flags, 1274 .page = page, 1275 .encrypted_page = NULL, 1276 }; 1277 int err; 1278 1279 if (PageUptodate(page)) { 1280 #ifdef CONFIG_F2FS_CHECK_FS 1281 f2fs_bug_on(sbi, !f2fs_inode_chksum_verify(sbi, page)); 1282 #endif 1283 return LOCKED_PAGE; 1284 } 1285 1286 err = f2fs_get_node_info(sbi, page->index, &ni); 1287 if (err) 1288 return err; 1289 1290 if (unlikely(ni.blk_addr == NULL_ADDR) || 1291 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) { 1292 ClearPageUptodate(page); 1293 return -ENOENT; 1294 } 1295 1296 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr; 1297 return f2fs_submit_page_bio(&fio); 1298 } 1299 1300 /* 1301 * Readahead a node page 1302 */ 1303 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 1304 { 1305 struct page *apage; 1306 int err; 1307 1308 if (!nid) 1309 return; 1310 if (f2fs_check_nid_range(sbi, nid)) 1311 return; 1312 1313 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid); 1314 if (apage) 1315 return; 1316 1317 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1318 if (!apage) 1319 return; 1320 1321 err = read_node_page(apage, REQ_RAHEAD); 1322 f2fs_put_page(apage, err ? 1 : 0); 1323 } 1324 1325 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid, 1326 struct page *parent, int start) 1327 { 1328 struct page *page; 1329 int err; 1330 1331 if (!nid) 1332 return ERR_PTR(-ENOENT); 1333 if (f2fs_check_nid_range(sbi, nid)) 1334 return ERR_PTR(-EINVAL); 1335 repeat: 1336 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1337 if (!page) 1338 return ERR_PTR(-ENOMEM); 1339 1340 err = read_node_page(page, 0); 1341 if (err < 0) { 1342 f2fs_put_page(page, 1); 1343 return ERR_PTR(err); 1344 } else if (err == LOCKED_PAGE) { 1345 err = 0; 1346 goto page_hit; 1347 } 1348 1349 if (parent) 1350 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE); 1351 1352 lock_page(page); 1353 1354 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1355 f2fs_put_page(page, 1); 1356 goto repeat; 1357 } 1358 1359 if (unlikely(!PageUptodate(page))) { 1360 err = -EIO; 1361 goto out_err; 1362 } 1363 1364 if (!f2fs_inode_chksum_verify(sbi, page)) { 1365 err = -EBADMSG; 1366 goto out_err; 1367 } 1368 page_hit: 1369 if(unlikely(nid != nid_of_node(page))) { 1370 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, " 1371 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]", 1372 nid, nid_of_node(page), ino_of_node(page), 1373 ofs_of_node(page), cpver_of_node(page), 1374 next_blkaddr_of_node(page)); 1375 err = -EINVAL; 1376 out_err: 1377 ClearPageUptodate(page); 1378 f2fs_put_page(page, 1); 1379 return ERR_PTR(err); 1380 } 1381 return page; 1382 } 1383 1384 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 1385 { 1386 return __get_node_page(sbi, nid, NULL, 0); 1387 } 1388 1389 struct page *f2fs_get_node_page_ra(struct page *parent, int start) 1390 { 1391 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 1392 nid_t nid = get_nid(parent, start, false); 1393 1394 return __get_node_page(sbi, nid, parent, start); 1395 } 1396 1397 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino) 1398 { 1399 struct inode *inode; 1400 struct page *page; 1401 int ret; 1402 1403 /* should flush inline_data before evict_inode */ 1404 inode = ilookup(sbi->sb, ino); 1405 if (!inode) 1406 return; 1407 1408 page = f2fs_pagecache_get_page(inode->i_mapping, 0, 1409 FGP_LOCK|FGP_NOWAIT, 0); 1410 if (!page) 1411 goto iput_out; 1412 1413 if (!PageUptodate(page)) 1414 goto page_out; 1415 1416 if (!PageDirty(page)) 1417 goto page_out; 1418 1419 if (!clear_page_dirty_for_io(page)) 1420 goto page_out; 1421 1422 ret = f2fs_write_inline_data(inode, page); 1423 inode_dec_dirty_pages(inode); 1424 f2fs_remove_dirty_inode(inode); 1425 if (ret) 1426 set_page_dirty(page); 1427 page_out: 1428 f2fs_put_page(page, 1); 1429 iput_out: 1430 iput(inode); 1431 } 1432 1433 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino) 1434 { 1435 pgoff_t index; 1436 struct pagevec pvec; 1437 struct page *last_page = NULL; 1438 int nr_pages; 1439 1440 pagevec_init(&pvec); 1441 index = 0; 1442 1443 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1444 PAGECACHE_TAG_DIRTY))) { 1445 int i; 1446 1447 for (i = 0; i < nr_pages; i++) { 1448 struct page *page = pvec.pages[i]; 1449 1450 if (unlikely(f2fs_cp_error(sbi))) { 1451 f2fs_put_page(last_page, 0); 1452 pagevec_release(&pvec); 1453 return ERR_PTR(-EIO); 1454 } 1455 1456 if (!IS_DNODE(page) || !is_cold_node(page)) 1457 continue; 1458 if (ino_of_node(page) != ino) 1459 continue; 1460 1461 lock_page(page); 1462 1463 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1464 continue_unlock: 1465 unlock_page(page); 1466 continue; 1467 } 1468 if (ino_of_node(page) != ino) 1469 goto continue_unlock; 1470 1471 if (!PageDirty(page)) { 1472 /* someone wrote it for us */ 1473 goto continue_unlock; 1474 } 1475 1476 if (last_page) 1477 f2fs_put_page(last_page, 0); 1478 1479 get_page(page); 1480 last_page = page; 1481 unlock_page(page); 1482 } 1483 pagevec_release(&pvec); 1484 cond_resched(); 1485 } 1486 return last_page; 1487 } 1488 1489 static int __write_node_page(struct page *page, bool atomic, bool *submitted, 1490 struct writeback_control *wbc, bool do_balance, 1491 enum iostat_type io_type, unsigned int *seq_id) 1492 { 1493 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1494 nid_t nid; 1495 struct node_info ni; 1496 struct f2fs_io_info fio = { 1497 .sbi = sbi, 1498 .ino = ino_of_node(page), 1499 .type = NODE, 1500 .op = REQ_OP_WRITE, 1501 .op_flags = wbc_to_write_flags(wbc), 1502 .page = page, 1503 .encrypted_page = NULL, 1504 .submitted = false, 1505 .io_type = io_type, 1506 .io_wbc = wbc, 1507 }; 1508 unsigned int seq; 1509 1510 trace_f2fs_writepage(page, NODE); 1511 1512 if (unlikely(f2fs_cp_error(sbi))) 1513 goto redirty_out; 1514 1515 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1516 goto redirty_out; 1517 1518 if (wbc->sync_mode == WB_SYNC_NONE && 1519 IS_DNODE(page) && is_cold_node(page)) 1520 goto redirty_out; 1521 1522 /* get old block addr of this node page */ 1523 nid = nid_of_node(page); 1524 f2fs_bug_on(sbi, page->index != nid); 1525 1526 if (f2fs_get_node_info(sbi, nid, &ni)) 1527 goto redirty_out; 1528 1529 if (wbc->for_reclaim) { 1530 if (!down_read_trylock(&sbi->node_write)) 1531 goto redirty_out; 1532 } else { 1533 down_read(&sbi->node_write); 1534 } 1535 1536 /* This page is already truncated */ 1537 if (unlikely(ni.blk_addr == NULL_ADDR)) { 1538 ClearPageUptodate(page); 1539 dec_page_count(sbi, F2FS_DIRTY_NODES); 1540 up_read(&sbi->node_write); 1541 unlock_page(page); 1542 return 0; 1543 } 1544 1545 if (__is_valid_data_blkaddr(ni.blk_addr) && 1546 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC)) { 1547 up_read(&sbi->node_write); 1548 goto redirty_out; 1549 } 1550 1551 if (atomic && !test_opt(sbi, NOBARRIER)) 1552 fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 1553 1554 set_page_writeback(page); 1555 ClearPageError(page); 1556 1557 if (f2fs_in_warm_node_list(sbi, page)) { 1558 seq = f2fs_add_fsync_node_entry(sbi, page); 1559 if (seq_id) 1560 *seq_id = seq; 1561 } 1562 1563 fio.old_blkaddr = ni.blk_addr; 1564 f2fs_do_write_node_page(nid, &fio); 1565 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page)); 1566 dec_page_count(sbi, F2FS_DIRTY_NODES); 1567 up_read(&sbi->node_write); 1568 1569 if (wbc->for_reclaim) { 1570 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE); 1571 submitted = NULL; 1572 } 1573 1574 unlock_page(page); 1575 1576 if (unlikely(f2fs_cp_error(sbi))) { 1577 f2fs_submit_merged_write(sbi, NODE); 1578 submitted = NULL; 1579 } 1580 if (submitted) 1581 *submitted = fio.submitted; 1582 1583 if (do_balance) 1584 f2fs_balance_fs(sbi, false); 1585 return 0; 1586 1587 redirty_out: 1588 redirty_page_for_writepage(wbc, page); 1589 return AOP_WRITEPAGE_ACTIVATE; 1590 } 1591 1592 int f2fs_move_node_page(struct page *node_page, int gc_type) 1593 { 1594 int err = 0; 1595 1596 if (gc_type == FG_GC) { 1597 struct writeback_control wbc = { 1598 .sync_mode = WB_SYNC_ALL, 1599 .nr_to_write = 1, 1600 .for_reclaim = 0, 1601 }; 1602 1603 f2fs_wait_on_page_writeback(node_page, NODE, true, true); 1604 1605 set_page_dirty(node_page); 1606 1607 if (!clear_page_dirty_for_io(node_page)) { 1608 err = -EAGAIN; 1609 goto out_page; 1610 } 1611 1612 if (__write_node_page(node_page, false, NULL, 1613 &wbc, false, FS_GC_NODE_IO, NULL)) { 1614 err = -EAGAIN; 1615 unlock_page(node_page); 1616 } 1617 goto release_page; 1618 } else { 1619 /* set page dirty and write it */ 1620 if (!PageWriteback(node_page)) 1621 set_page_dirty(node_page); 1622 } 1623 out_page: 1624 unlock_page(node_page); 1625 release_page: 1626 f2fs_put_page(node_page, 0); 1627 return err; 1628 } 1629 1630 static int f2fs_write_node_page(struct page *page, 1631 struct writeback_control *wbc) 1632 { 1633 return __write_node_page(page, false, NULL, wbc, false, 1634 FS_NODE_IO, NULL); 1635 } 1636 1637 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode, 1638 struct writeback_control *wbc, bool atomic, 1639 unsigned int *seq_id) 1640 { 1641 pgoff_t index; 1642 struct pagevec pvec; 1643 int ret = 0; 1644 struct page *last_page = NULL; 1645 bool marked = false; 1646 nid_t ino = inode->i_ino; 1647 int nr_pages; 1648 int nwritten = 0; 1649 1650 if (atomic) { 1651 last_page = last_fsync_dnode(sbi, ino); 1652 if (IS_ERR_OR_NULL(last_page)) 1653 return PTR_ERR_OR_ZERO(last_page); 1654 } 1655 retry: 1656 pagevec_init(&pvec); 1657 index = 0; 1658 1659 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1660 PAGECACHE_TAG_DIRTY))) { 1661 int i; 1662 1663 for (i = 0; i < nr_pages; i++) { 1664 struct page *page = pvec.pages[i]; 1665 bool submitted = false; 1666 1667 if (unlikely(f2fs_cp_error(sbi))) { 1668 f2fs_put_page(last_page, 0); 1669 pagevec_release(&pvec); 1670 ret = -EIO; 1671 goto out; 1672 } 1673 1674 if (!IS_DNODE(page) || !is_cold_node(page)) 1675 continue; 1676 if (ino_of_node(page) != ino) 1677 continue; 1678 1679 lock_page(page); 1680 1681 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1682 continue_unlock: 1683 unlock_page(page); 1684 continue; 1685 } 1686 if (ino_of_node(page) != ino) 1687 goto continue_unlock; 1688 1689 if (!PageDirty(page) && page != last_page) { 1690 /* someone wrote it for us */ 1691 goto continue_unlock; 1692 } 1693 1694 f2fs_wait_on_page_writeback(page, NODE, true, true); 1695 1696 set_fsync_mark(page, 0); 1697 set_dentry_mark(page, 0); 1698 1699 if (!atomic || page == last_page) { 1700 set_fsync_mark(page, 1); 1701 if (IS_INODE(page)) { 1702 if (is_inode_flag_set(inode, 1703 FI_DIRTY_INODE)) 1704 f2fs_update_inode(inode, page); 1705 set_dentry_mark(page, 1706 f2fs_need_dentry_mark(sbi, ino)); 1707 } 1708 /* may be written by other thread */ 1709 if (!PageDirty(page)) 1710 set_page_dirty(page); 1711 } 1712 1713 if (!clear_page_dirty_for_io(page)) 1714 goto continue_unlock; 1715 1716 ret = __write_node_page(page, atomic && 1717 page == last_page, 1718 &submitted, wbc, true, 1719 FS_NODE_IO, seq_id); 1720 if (ret) { 1721 unlock_page(page); 1722 f2fs_put_page(last_page, 0); 1723 break; 1724 } else if (submitted) { 1725 nwritten++; 1726 } 1727 1728 if (page == last_page) { 1729 f2fs_put_page(page, 0); 1730 marked = true; 1731 break; 1732 } 1733 } 1734 pagevec_release(&pvec); 1735 cond_resched(); 1736 1737 if (ret || marked) 1738 break; 1739 } 1740 if (!ret && atomic && !marked) { 1741 f2fs_msg(sbi->sb, KERN_DEBUG, 1742 "Retry to write fsync mark: ino=%u, idx=%lx", 1743 ino, last_page->index); 1744 lock_page(last_page); 1745 f2fs_wait_on_page_writeback(last_page, NODE, true, true); 1746 set_page_dirty(last_page); 1747 unlock_page(last_page); 1748 goto retry; 1749 } 1750 out: 1751 if (nwritten) 1752 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE); 1753 return ret ? -EIO: 0; 1754 } 1755 1756 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi, 1757 struct writeback_control *wbc, 1758 bool do_balance, enum iostat_type io_type) 1759 { 1760 pgoff_t index; 1761 struct pagevec pvec; 1762 int step = 0; 1763 int nwritten = 0; 1764 int ret = 0; 1765 int nr_pages, done = 0; 1766 1767 pagevec_init(&pvec); 1768 1769 next_step: 1770 index = 0; 1771 1772 while (!done && (nr_pages = pagevec_lookup_tag(&pvec, 1773 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) { 1774 int i; 1775 1776 for (i = 0; i < nr_pages; i++) { 1777 struct page *page = pvec.pages[i]; 1778 bool submitted = false; 1779 1780 /* give a priority to WB_SYNC threads */ 1781 if (atomic_read(&sbi->wb_sync_req[NODE]) && 1782 wbc->sync_mode == WB_SYNC_NONE) { 1783 done = 1; 1784 break; 1785 } 1786 1787 /* 1788 * flushing sequence with step: 1789 * 0. indirect nodes 1790 * 1. dentry dnodes 1791 * 2. file dnodes 1792 */ 1793 if (step == 0 && IS_DNODE(page)) 1794 continue; 1795 if (step == 1 && (!IS_DNODE(page) || 1796 is_cold_node(page))) 1797 continue; 1798 if (step == 2 && (!IS_DNODE(page) || 1799 !is_cold_node(page))) 1800 continue; 1801 lock_node: 1802 if (wbc->sync_mode == WB_SYNC_ALL) 1803 lock_page(page); 1804 else if (!trylock_page(page)) 1805 continue; 1806 1807 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1808 continue_unlock: 1809 unlock_page(page); 1810 continue; 1811 } 1812 1813 if (!PageDirty(page)) { 1814 /* someone wrote it for us */ 1815 goto continue_unlock; 1816 } 1817 1818 /* flush inline_data */ 1819 if (is_inline_node(page)) { 1820 clear_inline_node(page); 1821 unlock_page(page); 1822 flush_inline_data(sbi, ino_of_node(page)); 1823 goto lock_node; 1824 } 1825 1826 f2fs_wait_on_page_writeback(page, NODE, true, true); 1827 1828 if (!clear_page_dirty_for_io(page)) 1829 goto continue_unlock; 1830 1831 set_fsync_mark(page, 0); 1832 set_dentry_mark(page, 0); 1833 1834 ret = __write_node_page(page, false, &submitted, 1835 wbc, do_balance, io_type, NULL); 1836 if (ret) 1837 unlock_page(page); 1838 else if (submitted) 1839 nwritten++; 1840 1841 if (--wbc->nr_to_write == 0) 1842 break; 1843 } 1844 pagevec_release(&pvec); 1845 cond_resched(); 1846 1847 if (wbc->nr_to_write == 0) { 1848 step = 2; 1849 break; 1850 } 1851 } 1852 1853 if (step < 2) { 1854 if (wbc->sync_mode == WB_SYNC_NONE && step == 1) 1855 goto out; 1856 step++; 1857 goto next_step; 1858 } 1859 out: 1860 if (nwritten) 1861 f2fs_submit_merged_write(sbi, NODE); 1862 1863 if (unlikely(f2fs_cp_error(sbi))) 1864 return -EIO; 1865 return ret; 1866 } 1867 1868 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, 1869 unsigned int seq_id) 1870 { 1871 struct fsync_node_entry *fn; 1872 struct page *page; 1873 struct list_head *head = &sbi->fsync_node_list; 1874 unsigned long flags; 1875 unsigned int cur_seq_id = 0; 1876 int ret2, ret = 0; 1877 1878 while (seq_id && cur_seq_id < seq_id) { 1879 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 1880 if (list_empty(head)) { 1881 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 1882 break; 1883 } 1884 fn = list_first_entry(head, struct fsync_node_entry, list); 1885 if (fn->seq_id > seq_id) { 1886 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 1887 break; 1888 } 1889 cur_seq_id = fn->seq_id; 1890 page = fn->page; 1891 get_page(page); 1892 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 1893 1894 f2fs_wait_on_page_writeback(page, NODE, true, false); 1895 if (TestClearPageError(page)) 1896 ret = -EIO; 1897 1898 put_page(page); 1899 1900 if (ret) 1901 break; 1902 } 1903 1904 ret2 = filemap_check_errors(NODE_MAPPING(sbi)); 1905 if (!ret) 1906 ret = ret2; 1907 1908 return ret; 1909 } 1910 1911 static int f2fs_write_node_pages(struct address_space *mapping, 1912 struct writeback_control *wbc) 1913 { 1914 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 1915 struct blk_plug plug; 1916 long diff; 1917 1918 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1919 goto skip_write; 1920 1921 /* balancing f2fs's metadata in background */ 1922 f2fs_balance_fs_bg(sbi); 1923 1924 /* collect a number of dirty node pages and write together */ 1925 if (wbc->sync_mode != WB_SYNC_ALL && 1926 get_pages(sbi, F2FS_DIRTY_NODES) < 1927 nr_pages_to_skip(sbi, NODE)) 1928 goto skip_write; 1929 1930 if (wbc->sync_mode == WB_SYNC_ALL) 1931 atomic_inc(&sbi->wb_sync_req[NODE]); 1932 else if (atomic_read(&sbi->wb_sync_req[NODE])) 1933 goto skip_write; 1934 1935 trace_f2fs_writepages(mapping->host, wbc, NODE); 1936 1937 diff = nr_pages_to_write(sbi, NODE, wbc); 1938 blk_start_plug(&plug); 1939 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO); 1940 blk_finish_plug(&plug); 1941 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); 1942 1943 if (wbc->sync_mode == WB_SYNC_ALL) 1944 atomic_dec(&sbi->wb_sync_req[NODE]); 1945 return 0; 1946 1947 skip_write: 1948 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES); 1949 trace_f2fs_writepages(mapping->host, wbc, NODE); 1950 return 0; 1951 } 1952 1953 static int f2fs_set_node_page_dirty(struct page *page) 1954 { 1955 trace_f2fs_set_page_dirty(page, NODE); 1956 1957 if (!PageUptodate(page)) 1958 SetPageUptodate(page); 1959 #ifdef CONFIG_F2FS_CHECK_FS 1960 if (IS_INODE(page)) 1961 f2fs_inode_chksum_set(F2FS_P_SB(page), page); 1962 #endif 1963 if (!PageDirty(page)) { 1964 __set_page_dirty_nobuffers(page); 1965 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES); 1966 f2fs_set_page_private(page, 0); 1967 f2fs_trace_pid(page); 1968 return 1; 1969 } 1970 return 0; 1971 } 1972 1973 /* 1974 * Structure of the f2fs node operations 1975 */ 1976 const struct address_space_operations f2fs_node_aops = { 1977 .writepage = f2fs_write_node_page, 1978 .writepages = f2fs_write_node_pages, 1979 .set_page_dirty = f2fs_set_node_page_dirty, 1980 .invalidatepage = f2fs_invalidate_page, 1981 .releasepage = f2fs_release_page, 1982 #ifdef CONFIG_MIGRATION 1983 .migratepage = f2fs_migrate_page, 1984 #endif 1985 }; 1986 1987 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i, 1988 nid_t n) 1989 { 1990 return radix_tree_lookup(&nm_i->free_nid_root, n); 1991 } 1992 1993 static int __insert_free_nid(struct f2fs_sb_info *sbi, 1994 struct free_nid *i, enum nid_state state) 1995 { 1996 struct f2fs_nm_info *nm_i = NM_I(sbi); 1997 1998 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i); 1999 if (err) 2000 return err; 2001 2002 f2fs_bug_on(sbi, state != i->state); 2003 nm_i->nid_cnt[state]++; 2004 if (state == FREE_NID) 2005 list_add_tail(&i->list, &nm_i->free_nid_list); 2006 return 0; 2007 } 2008 2009 static void __remove_free_nid(struct f2fs_sb_info *sbi, 2010 struct free_nid *i, enum nid_state state) 2011 { 2012 struct f2fs_nm_info *nm_i = NM_I(sbi); 2013 2014 f2fs_bug_on(sbi, state != i->state); 2015 nm_i->nid_cnt[state]--; 2016 if (state == FREE_NID) 2017 list_del(&i->list); 2018 radix_tree_delete(&nm_i->free_nid_root, i->nid); 2019 } 2020 2021 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i, 2022 enum nid_state org_state, enum nid_state dst_state) 2023 { 2024 struct f2fs_nm_info *nm_i = NM_I(sbi); 2025 2026 f2fs_bug_on(sbi, org_state != i->state); 2027 i->state = dst_state; 2028 nm_i->nid_cnt[org_state]--; 2029 nm_i->nid_cnt[dst_state]++; 2030 2031 switch (dst_state) { 2032 case PREALLOC_NID: 2033 list_del(&i->list); 2034 break; 2035 case FREE_NID: 2036 list_add_tail(&i->list, &nm_i->free_nid_list); 2037 break; 2038 default: 2039 BUG_ON(1); 2040 } 2041 } 2042 2043 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid, 2044 bool set, bool build) 2045 { 2046 struct f2fs_nm_info *nm_i = NM_I(sbi); 2047 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid); 2048 unsigned int nid_ofs = nid - START_NID(nid); 2049 2050 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap)) 2051 return; 2052 2053 if (set) { 2054 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2055 return; 2056 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2057 nm_i->free_nid_count[nat_ofs]++; 2058 } else { 2059 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2060 return; 2061 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2062 if (!build) 2063 nm_i->free_nid_count[nat_ofs]--; 2064 } 2065 } 2066 2067 /* return if the nid is recognized as free */ 2068 static bool add_free_nid(struct f2fs_sb_info *sbi, 2069 nid_t nid, bool build, bool update) 2070 { 2071 struct f2fs_nm_info *nm_i = NM_I(sbi); 2072 struct free_nid *i, *e; 2073 struct nat_entry *ne; 2074 int err = -EINVAL; 2075 bool ret = false; 2076 2077 /* 0 nid should not be used */ 2078 if (unlikely(nid == 0)) 2079 return false; 2080 2081 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS); 2082 i->nid = nid; 2083 i->state = FREE_NID; 2084 2085 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 2086 2087 spin_lock(&nm_i->nid_list_lock); 2088 2089 if (build) { 2090 /* 2091 * Thread A Thread B 2092 * - f2fs_create 2093 * - f2fs_new_inode 2094 * - f2fs_alloc_nid 2095 * - __insert_nid_to_list(PREALLOC_NID) 2096 * - f2fs_balance_fs_bg 2097 * - f2fs_build_free_nids 2098 * - __f2fs_build_free_nids 2099 * - scan_nat_page 2100 * - add_free_nid 2101 * - __lookup_nat_cache 2102 * - f2fs_add_link 2103 * - f2fs_init_inode_metadata 2104 * - f2fs_new_inode_page 2105 * - f2fs_new_node_page 2106 * - set_node_addr 2107 * - f2fs_alloc_nid_done 2108 * - __remove_nid_from_list(PREALLOC_NID) 2109 * - __insert_nid_to_list(FREE_NID) 2110 */ 2111 ne = __lookup_nat_cache(nm_i, nid); 2112 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) || 2113 nat_get_blkaddr(ne) != NULL_ADDR)) 2114 goto err_out; 2115 2116 e = __lookup_free_nid_list(nm_i, nid); 2117 if (e) { 2118 if (e->state == FREE_NID) 2119 ret = true; 2120 goto err_out; 2121 } 2122 } 2123 ret = true; 2124 err = __insert_free_nid(sbi, i, FREE_NID); 2125 err_out: 2126 if (update) { 2127 update_free_nid_bitmap(sbi, nid, ret, build); 2128 if (!build) 2129 nm_i->available_nids++; 2130 } 2131 spin_unlock(&nm_i->nid_list_lock); 2132 radix_tree_preload_end(); 2133 2134 if (err) 2135 kmem_cache_free(free_nid_slab, i); 2136 return ret; 2137 } 2138 2139 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid) 2140 { 2141 struct f2fs_nm_info *nm_i = NM_I(sbi); 2142 struct free_nid *i; 2143 bool need_free = false; 2144 2145 spin_lock(&nm_i->nid_list_lock); 2146 i = __lookup_free_nid_list(nm_i, nid); 2147 if (i && i->state == FREE_NID) { 2148 __remove_free_nid(sbi, i, FREE_NID); 2149 need_free = true; 2150 } 2151 spin_unlock(&nm_i->nid_list_lock); 2152 2153 if (need_free) 2154 kmem_cache_free(free_nid_slab, i); 2155 } 2156 2157 static int scan_nat_page(struct f2fs_sb_info *sbi, 2158 struct page *nat_page, nid_t start_nid) 2159 { 2160 struct f2fs_nm_info *nm_i = NM_I(sbi); 2161 struct f2fs_nat_block *nat_blk = page_address(nat_page); 2162 block_t blk_addr; 2163 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid); 2164 int i; 2165 2166 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap); 2167 2168 i = start_nid % NAT_ENTRY_PER_BLOCK; 2169 2170 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) { 2171 if (unlikely(start_nid >= nm_i->max_nid)) 2172 break; 2173 2174 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr); 2175 2176 if (blk_addr == NEW_ADDR) 2177 return -EINVAL; 2178 2179 if (blk_addr == NULL_ADDR) { 2180 add_free_nid(sbi, start_nid, true, true); 2181 } else { 2182 spin_lock(&NM_I(sbi)->nid_list_lock); 2183 update_free_nid_bitmap(sbi, start_nid, false, true); 2184 spin_unlock(&NM_I(sbi)->nid_list_lock); 2185 } 2186 } 2187 2188 return 0; 2189 } 2190 2191 static void scan_curseg_cache(struct f2fs_sb_info *sbi) 2192 { 2193 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2194 struct f2fs_journal *journal = curseg->journal; 2195 int i; 2196 2197 down_read(&curseg->journal_rwsem); 2198 for (i = 0; i < nats_in_cursum(journal); i++) { 2199 block_t addr; 2200 nid_t nid; 2201 2202 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr); 2203 nid = le32_to_cpu(nid_in_journal(journal, i)); 2204 if (addr == NULL_ADDR) 2205 add_free_nid(sbi, nid, true, false); 2206 else 2207 remove_free_nid(sbi, nid); 2208 } 2209 up_read(&curseg->journal_rwsem); 2210 } 2211 2212 static void scan_free_nid_bits(struct f2fs_sb_info *sbi) 2213 { 2214 struct f2fs_nm_info *nm_i = NM_I(sbi); 2215 unsigned int i, idx; 2216 nid_t nid; 2217 2218 down_read(&nm_i->nat_tree_lock); 2219 2220 for (i = 0; i < nm_i->nat_blocks; i++) { 2221 if (!test_bit_le(i, nm_i->nat_block_bitmap)) 2222 continue; 2223 if (!nm_i->free_nid_count[i]) 2224 continue; 2225 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) { 2226 idx = find_next_bit_le(nm_i->free_nid_bitmap[i], 2227 NAT_ENTRY_PER_BLOCK, idx); 2228 if (idx >= NAT_ENTRY_PER_BLOCK) 2229 break; 2230 2231 nid = i * NAT_ENTRY_PER_BLOCK + idx; 2232 add_free_nid(sbi, nid, true, false); 2233 2234 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS) 2235 goto out; 2236 } 2237 } 2238 out: 2239 scan_curseg_cache(sbi); 2240 2241 up_read(&nm_i->nat_tree_lock); 2242 } 2243 2244 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi, 2245 bool sync, bool mount) 2246 { 2247 struct f2fs_nm_info *nm_i = NM_I(sbi); 2248 int i = 0, ret; 2249 nid_t nid = nm_i->next_scan_nid; 2250 2251 if (unlikely(nid >= nm_i->max_nid)) 2252 nid = 0; 2253 2254 /* Enough entries */ 2255 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2256 return 0; 2257 2258 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS)) 2259 return 0; 2260 2261 if (!mount) { 2262 /* try to find free nids in free_nid_bitmap */ 2263 scan_free_nid_bits(sbi); 2264 2265 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2266 return 0; 2267 } 2268 2269 /* readahead nat pages to be scanned */ 2270 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, 2271 META_NAT, true); 2272 2273 down_read(&nm_i->nat_tree_lock); 2274 2275 while (1) { 2276 if (!test_bit_le(NAT_BLOCK_OFFSET(nid), 2277 nm_i->nat_block_bitmap)) { 2278 struct page *page = get_current_nat_page(sbi, nid); 2279 2280 if (IS_ERR(page)) { 2281 ret = PTR_ERR(page); 2282 } else { 2283 ret = scan_nat_page(sbi, page, nid); 2284 f2fs_put_page(page, 1); 2285 } 2286 2287 if (ret) { 2288 up_read(&nm_i->nat_tree_lock); 2289 f2fs_bug_on(sbi, !mount); 2290 f2fs_msg(sbi->sb, KERN_ERR, 2291 "NAT is corrupt, run fsck to fix it"); 2292 return ret; 2293 } 2294 } 2295 2296 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK)); 2297 if (unlikely(nid >= nm_i->max_nid)) 2298 nid = 0; 2299 2300 if (++i >= FREE_NID_PAGES) 2301 break; 2302 } 2303 2304 /* go to the next free nat pages to find free nids abundantly */ 2305 nm_i->next_scan_nid = nid; 2306 2307 /* find free nids from current sum_pages */ 2308 scan_curseg_cache(sbi); 2309 2310 up_read(&nm_i->nat_tree_lock); 2311 2312 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid), 2313 nm_i->ra_nid_pages, META_NAT, false); 2314 2315 return 0; 2316 } 2317 2318 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount) 2319 { 2320 int ret; 2321 2322 mutex_lock(&NM_I(sbi)->build_lock); 2323 ret = __f2fs_build_free_nids(sbi, sync, mount); 2324 mutex_unlock(&NM_I(sbi)->build_lock); 2325 2326 return ret; 2327 } 2328 2329 /* 2330 * If this function returns success, caller can obtain a new nid 2331 * from second parameter of this function. 2332 * The returned nid could be used ino as well as nid when inode is created. 2333 */ 2334 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid) 2335 { 2336 struct f2fs_nm_info *nm_i = NM_I(sbi); 2337 struct free_nid *i = NULL; 2338 retry: 2339 if (time_to_inject(sbi, FAULT_ALLOC_NID)) { 2340 f2fs_show_injection_info(FAULT_ALLOC_NID); 2341 return false; 2342 } 2343 2344 spin_lock(&nm_i->nid_list_lock); 2345 2346 if (unlikely(nm_i->available_nids == 0)) { 2347 spin_unlock(&nm_i->nid_list_lock); 2348 return false; 2349 } 2350 2351 /* We should not use stale free nids created by f2fs_build_free_nids */ 2352 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) { 2353 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list)); 2354 i = list_first_entry(&nm_i->free_nid_list, 2355 struct free_nid, list); 2356 *nid = i->nid; 2357 2358 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID); 2359 nm_i->available_nids--; 2360 2361 update_free_nid_bitmap(sbi, *nid, false, false); 2362 2363 spin_unlock(&nm_i->nid_list_lock); 2364 return true; 2365 } 2366 spin_unlock(&nm_i->nid_list_lock); 2367 2368 /* Let's scan nat pages and its caches to get free nids */ 2369 if (!f2fs_build_free_nids(sbi, true, false)) 2370 goto retry; 2371 return false; 2372 } 2373 2374 /* 2375 * f2fs_alloc_nid() should be called prior to this function. 2376 */ 2377 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 2378 { 2379 struct f2fs_nm_info *nm_i = NM_I(sbi); 2380 struct free_nid *i; 2381 2382 spin_lock(&nm_i->nid_list_lock); 2383 i = __lookup_free_nid_list(nm_i, nid); 2384 f2fs_bug_on(sbi, !i); 2385 __remove_free_nid(sbi, i, PREALLOC_NID); 2386 spin_unlock(&nm_i->nid_list_lock); 2387 2388 kmem_cache_free(free_nid_slab, i); 2389 } 2390 2391 /* 2392 * f2fs_alloc_nid() should be called prior to this function. 2393 */ 2394 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 2395 { 2396 struct f2fs_nm_info *nm_i = NM_I(sbi); 2397 struct free_nid *i; 2398 bool need_free = false; 2399 2400 if (!nid) 2401 return; 2402 2403 spin_lock(&nm_i->nid_list_lock); 2404 i = __lookup_free_nid_list(nm_i, nid); 2405 f2fs_bug_on(sbi, !i); 2406 2407 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) { 2408 __remove_free_nid(sbi, i, PREALLOC_NID); 2409 need_free = true; 2410 } else { 2411 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID); 2412 } 2413 2414 nm_i->available_nids++; 2415 2416 update_free_nid_bitmap(sbi, nid, true, false); 2417 2418 spin_unlock(&nm_i->nid_list_lock); 2419 2420 if (need_free) 2421 kmem_cache_free(free_nid_slab, i); 2422 } 2423 2424 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink) 2425 { 2426 struct f2fs_nm_info *nm_i = NM_I(sbi); 2427 struct free_nid *i, *next; 2428 int nr = nr_shrink; 2429 2430 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2431 return 0; 2432 2433 if (!mutex_trylock(&nm_i->build_lock)) 2434 return 0; 2435 2436 spin_lock(&nm_i->nid_list_lock); 2437 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) { 2438 if (nr_shrink <= 0 || 2439 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2440 break; 2441 2442 __remove_free_nid(sbi, i, FREE_NID); 2443 kmem_cache_free(free_nid_slab, i); 2444 nr_shrink--; 2445 } 2446 spin_unlock(&nm_i->nid_list_lock); 2447 mutex_unlock(&nm_i->build_lock); 2448 2449 return nr - nr_shrink; 2450 } 2451 2452 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page) 2453 { 2454 void *src_addr, *dst_addr; 2455 size_t inline_size; 2456 struct page *ipage; 2457 struct f2fs_inode *ri; 2458 2459 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 2460 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage)); 2461 2462 ri = F2FS_INODE(page); 2463 if (ri->i_inline & F2FS_INLINE_XATTR) { 2464 set_inode_flag(inode, FI_INLINE_XATTR); 2465 } else { 2466 clear_inode_flag(inode, FI_INLINE_XATTR); 2467 goto update_inode; 2468 } 2469 2470 dst_addr = inline_xattr_addr(inode, ipage); 2471 src_addr = inline_xattr_addr(inode, page); 2472 inline_size = inline_xattr_size(inode); 2473 2474 f2fs_wait_on_page_writeback(ipage, NODE, true, true); 2475 memcpy(dst_addr, src_addr, inline_size); 2476 update_inode: 2477 f2fs_update_inode(inode, ipage); 2478 f2fs_put_page(ipage, 1); 2479 } 2480 2481 int f2fs_recover_xattr_data(struct inode *inode, struct page *page) 2482 { 2483 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2484 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid; 2485 nid_t new_xnid; 2486 struct dnode_of_data dn; 2487 struct node_info ni; 2488 struct page *xpage; 2489 int err; 2490 2491 if (!prev_xnid) 2492 goto recover_xnid; 2493 2494 /* 1: invalidate the previous xattr nid */ 2495 err = f2fs_get_node_info(sbi, prev_xnid, &ni); 2496 if (err) 2497 return err; 2498 2499 f2fs_invalidate_blocks(sbi, ni.blk_addr); 2500 dec_valid_node_count(sbi, inode, false); 2501 set_node_addr(sbi, &ni, NULL_ADDR, false); 2502 2503 recover_xnid: 2504 /* 2: update xattr nid in inode */ 2505 if (!f2fs_alloc_nid(sbi, &new_xnid)) 2506 return -ENOSPC; 2507 2508 set_new_dnode(&dn, inode, NULL, NULL, new_xnid); 2509 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET); 2510 if (IS_ERR(xpage)) { 2511 f2fs_alloc_nid_failed(sbi, new_xnid); 2512 return PTR_ERR(xpage); 2513 } 2514 2515 f2fs_alloc_nid_done(sbi, new_xnid); 2516 f2fs_update_inode_page(inode); 2517 2518 /* 3: update and set xattr node page dirty */ 2519 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE); 2520 2521 set_page_dirty(xpage); 2522 f2fs_put_page(xpage, 1); 2523 2524 return 0; 2525 } 2526 2527 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 2528 { 2529 struct f2fs_inode *src, *dst; 2530 nid_t ino = ino_of_node(page); 2531 struct node_info old_ni, new_ni; 2532 struct page *ipage; 2533 int err; 2534 2535 err = f2fs_get_node_info(sbi, ino, &old_ni); 2536 if (err) 2537 return err; 2538 2539 if (unlikely(old_ni.blk_addr != NULL_ADDR)) 2540 return -EINVAL; 2541 retry: 2542 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false); 2543 if (!ipage) { 2544 congestion_wait(BLK_RW_ASYNC, HZ/50); 2545 goto retry; 2546 } 2547 2548 /* Should not use this inode from free nid list */ 2549 remove_free_nid(sbi, ino); 2550 2551 if (!PageUptodate(ipage)) 2552 SetPageUptodate(ipage); 2553 fill_node_footer(ipage, ino, ino, 0, true); 2554 set_cold_node(ipage, false); 2555 2556 src = F2FS_INODE(page); 2557 dst = F2FS_INODE(ipage); 2558 2559 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src); 2560 dst->i_size = 0; 2561 dst->i_blocks = cpu_to_le64(1); 2562 dst->i_links = cpu_to_le32(1); 2563 dst->i_xattr_nid = 0; 2564 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR); 2565 if (dst->i_inline & F2FS_EXTRA_ATTR) { 2566 dst->i_extra_isize = src->i_extra_isize; 2567 2568 if (f2fs_sb_has_flexible_inline_xattr(sbi) && 2569 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2570 i_inline_xattr_size)) 2571 dst->i_inline_xattr_size = src->i_inline_xattr_size; 2572 2573 if (f2fs_sb_has_project_quota(sbi) && 2574 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2575 i_projid)) 2576 dst->i_projid = src->i_projid; 2577 2578 if (f2fs_sb_has_inode_crtime(sbi) && 2579 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2580 i_crtime_nsec)) { 2581 dst->i_crtime = src->i_crtime; 2582 dst->i_crtime_nsec = src->i_crtime_nsec; 2583 } 2584 } 2585 2586 new_ni = old_ni; 2587 new_ni.ino = ino; 2588 2589 if (unlikely(inc_valid_node_count(sbi, NULL, true))) 2590 WARN_ON(1); 2591 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 2592 inc_valid_inode_count(sbi); 2593 set_page_dirty(ipage); 2594 f2fs_put_page(ipage, 1); 2595 return 0; 2596 } 2597 2598 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi, 2599 unsigned int segno, struct f2fs_summary_block *sum) 2600 { 2601 struct f2fs_node *rn; 2602 struct f2fs_summary *sum_entry; 2603 block_t addr; 2604 int i, idx, last_offset, nrpages; 2605 2606 /* scan the node segment */ 2607 last_offset = sbi->blocks_per_seg; 2608 addr = START_BLOCK(sbi, segno); 2609 sum_entry = &sum->entries[0]; 2610 2611 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) { 2612 nrpages = min(last_offset - i, BIO_MAX_PAGES); 2613 2614 /* readahead node pages */ 2615 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true); 2616 2617 for (idx = addr; idx < addr + nrpages; idx++) { 2618 struct page *page = f2fs_get_tmp_page(sbi, idx); 2619 2620 if (IS_ERR(page)) 2621 return PTR_ERR(page); 2622 2623 rn = F2FS_NODE(page); 2624 sum_entry->nid = rn->footer.nid; 2625 sum_entry->version = 0; 2626 sum_entry->ofs_in_node = 0; 2627 sum_entry++; 2628 f2fs_put_page(page, 1); 2629 } 2630 2631 invalidate_mapping_pages(META_MAPPING(sbi), addr, 2632 addr + nrpages); 2633 } 2634 return 0; 2635 } 2636 2637 static void remove_nats_in_journal(struct f2fs_sb_info *sbi) 2638 { 2639 struct f2fs_nm_info *nm_i = NM_I(sbi); 2640 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2641 struct f2fs_journal *journal = curseg->journal; 2642 int i; 2643 2644 down_write(&curseg->journal_rwsem); 2645 for (i = 0; i < nats_in_cursum(journal); i++) { 2646 struct nat_entry *ne; 2647 struct f2fs_nat_entry raw_ne; 2648 nid_t nid = le32_to_cpu(nid_in_journal(journal, i)); 2649 2650 raw_ne = nat_in_journal(journal, i); 2651 2652 ne = __lookup_nat_cache(nm_i, nid); 2653 if (!ne) { 2654 ne = __alloc_nat_entry(nid, true); 2655 __init_nat_entry(nm_i, ne, &raw_ne, true); 2656 } 2657 2658 /* 2659 * if a free nat in journal has not been used after last 2660 * checkpoint, we should remove it from available nids, 2661 * since later we will add it again. 2662 */ 2663 if (!get_nat_flag(ne, IS_DIRTY) && 2664 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) { 2665 spin_lock(&nm_i->nid_list_lock); 2666 nm_i->available_nids--; 2667 spin_unlock(&nm_i->nid_list_lock); 2668 } 2669 2670 __set_nat_cache_dirty(nm_i, ne); 2671 } 2672 update_nats_in_cursum(journal, -i); 2673 up_write(&curseg->journal_rwsem); 2674 } 2675 2676 static void __adjust_nat_entry_set(struct nat_entry_set *nes, 2677 struct list_head *head, int max) 2678 { 2679 struct nat_entry_set *cur; 2680 2681 if (nes->entry_cnt >= max) 2682 goto add_out; 2683 2684 list_for_each_entry(cur, head, set_list) { 2685 if (cur->entry_cnt >= nes->entry_cnt) { 2686 list_add(&nes->set_list, cur->set_list.prev); 2687 return; 2688 } 2689 } 2690 add_out: 2691 list_add_tail(&nes->set_list, head); 2692 } 2693 2694 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid, 2695 struct page *page) 2696 { 2697 struct f2fs_nm_info *nm_i = NM_I(sbi); 2698 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK; 2699 struct f2fs_nat_block *nat_blk = page_address(page); 2700 int valid = 0; 2701 int i = 0; 2702 2703 if (!enabled_nat_bits(sbi, NULL)) 2704 return; 2705 2706 if (nat_index == 0) { 2707 valid = 1; 2708 i = 1; 2709 } 2710 for (; i < NAT_ENTRY_PER_BLOCK; i++) { 2711 if (nat_blk->entries[i].block_addr != NULL_ADDR) 2712 valid++; 2713 } 2714 if (valid == 0) { 2715 __set_bit_le(nat_index, nm_i->empty_nat_bits); 2716 __clear_bit_le(nat_index, nm_i->full_nat_bits); 2717 return; 2718 } 2719 2720 __clear_bit_le(nat_index, nm_i->empty_nat_bits); 2721 if (valid == NAT_ENTRY_PER_BLOCK) 2722 __set_bit_le(nat_index, nm_i->full_nat_bits); 2723 else 2724 __clear_bit_le(nat_index, nm_i->full_nat_bits); 2725 } 2726 2727 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi, 2728 struct nat_entry_set *set, struct cp_control *cpc) 2729 { 2730 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2731 struct f2fs_journal *journal = curseg->journal; 2732 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK; 2733 bool to_journal = true; 2734 struct f2fs_nat_block *nat_blk; 2735 struct nat_entry *ne, *cur; 2736 struct page *page = NULL; 2737 2738 /* 2739 * there are two steps to flush nat entries: 2740 * #1, flush nat entries to journal in current hot data summary block. 2741 * #2, flush nat entries to nat page. 2742 */ 2743 if (enabled_nat_bits(sbi, cpc) || 2744 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL)) 2745 to_journal = false; 2746 2747 if (to_journal) { 2748 down_write(&curseg->journal_rwsem); 2749 } else { 2750 page = get_next_nat_page(sbi, start_nid); 2751 if (IS_ERR(page)) 2752 return PTR_ERR(page); 2753 2754 nat_blk = page_address(page); 2755 f2fs_bug_on(sbi, !nat_blk); 2756 } 2757 2758 /* flush dirty nats in nat entry set */ 2759 list_for_each_entry_safe(ne, cur, &set->entry_list, list) { 2760 struct f2fs_nat_entry *raw_ne; 2761 nid_t nid = nat_get_nid(ne); 2762 int offset; 2763 2764 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR); 2765 2766 if (to_journal) { 2767 offset = f2fs_lookup_journal_in_cursum(journal, 2768 NAT_JOURNAL, nid, 1); 2769 f2fs_bug_on(sbi, offset < 0); 2770 raw_ne = &nat_in_journal(journal, offset); 2771 nid_in_journal(journal, offset) = cpu_to_le32(nid); 2772 } else { 2773 raw_ne = &nat_blk->entries[nid - start_nid]; 2774 } 2775 raw_nat_from_node_info(raw_ne, &ne->ni); 2776 nat_reset_flag(ne); 2777 __clear_nat_cache_dirty(NM_I(sbi), set, ne); 2778 if (nat_get_blkaddr(ne) == NULL_ADDR) { 2779 add_free_nid(sbi, nid, false, true); 2780 } else { 2781 spin_lock(&NM_I(sbi)->nid_list_lock); 2782 update_free_nid_bitmap(sbi, nid, false, false); 2783 spin_unlock(&NM_I(sbi)->nid_list_lock); 2784 } 2785 } 2786 2787 if (to_journal) { 2788 up_write(&curseg->journal_rwsem); 2789 } else { 2790 __update_nat_bits(sbi, start_nid, page); 2791 f2fs_put_page(page, 1); 2792 } 2793 2794 /* Allow dirty nats by node block allocation in write_begin */ 2795 if (!set->entry_cnt) { 2796 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set); 2797 kmem_cache_free(nat_entry_set_slab, set); 2798 } 2799 return 0; 2800 } 2801 2802 /* 2803 * This function is called during the checkpointing process. 2804 */ 2805 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 2806 { 2807 struct f2fs_nm_info *nm_i = NM_I(sbi); 2808 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2809 struct f2fs_journal *journal = curseg->journal; 2810 struct nat_entry_set *setvec[SETVEC_SIZE]; 2811 struct nat_entry_set *set, *tmp; 2812 unsigned int found; 2813 nid_t set_idx = 0; 2814 LIST_HEAD(sets); 2815 int err = 0; 2816 2817 /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */ 2818 if (enabled_nat_bits(sbi, cpc)) { 2819 down_write(&nm_i->nat_tree_lock); 2820 remove_nats_in_journal(sbi); 2821 up_write(&nm_i->nat_tree_lock); 2822 } 2823 2824 if (!nm_i->dirty_nat_cnt) 2825 return 0; 2826 2827 down_write(&nm_i->nat_tree_lock); 2828 2829 /* 2830 * if there are no enough space in journal to store dirty nat 2831 * entries, remove all entries from journal and merge them 2832 * into nat entry set. 2833 */ 2834 if (enabled_nat_bits(sbi, cpc) || 2835 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL)) 2836 remove_nats_in_journal(sbi); 2837 2838 while ((found = __gang_lookup_nat_set(nm_i, 2839 set_idx, SETVEC_SIZE, setvec))) { 2840 unsigned idx; 2841 set_idx = setvec[found - 1]->set + 1; 2842 for (idx = 0; idx < found; idx++) 2843 __adjust_nat_entry_set(setvec[idx], &sets, 2844 MAX_NAT_JENTRIES(journal)); 2845 } 2846 2847 /* flush dirty nats in nat entry set */ 2848 list_for_each_entry_safe(set, tmp, &sets, set_list) { 2849 err = __flush_nat_entry_set(sbi, set, cpc); 2850 if (err) 2851 break; 2852 } 2853 2854 up_write(&nm_i->nat_tree_lock); 2855 /* Allow dirty nats by node block allocation in write_begin */ 2856 2857 return err; 2858 } 2859 2860 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi) 2861 { 2862 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2863 struct f2fs_nm_info *nm_i = NM_I(sbi); 2864 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE; 2865 unsigned int i; 2866 __u64 cp_ver = cur_cp_version(ckpt); 2867 block_t nat_bits_addr; 2868 2869 if (!enabled_nat_bits(sbi, NULL)) 2870 return 0; 2871 2872 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8); 2873 nm_i->nat_bits = f2fs_kzalloc(sbi, 2874 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL); 2875 if (!nm_i->nat_bits) 2876 return -ENOMEM; 2877 2878 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg - 2879 nm_i->nat_bits_blocks; 2880 for (i = 0; i < nm_i->nat_bits_blocks; i++) { 2881 struct page *page; 2882 2883 page = f2fs_get_meta_page(sbi, nat_bits_addr++); 2884 if (IS_ERR(page)) 2885 return PTR_ERR(page); 2886 2887 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS), 2888 page_address(page), F2FS_BLKSIZE); 2889 f2fs_put_page(page, 1); 2890 } 2891 2892 cp_ver |= (cur_cp_crc(ckpt) << 32); 2893 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) { 2894 disable_nat_bits(sbi, true); 2895 return 0; 2896 } 2897 2898 nm_i->full_nat_bits = nm_i->nat_bits + 8; 2899 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes; 2900 2901 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint"); 2902 return 0; 2903 } 2904 2905 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi) 2906 { 2907 struct f2fs_nm_info *nm_i = NM_I(sbi); 2908 unsigned int i = 0; 2909 nid_t nid, last_nid; 2910 2911 if (!enabled_nat_bits(sbi, NULL)) 2912 return; 2913 2914 for (i = 0; i < nm_i->nat_blocks; i++) { 2915 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i); 2916 if (i >= nm_i->nat_blocks) 2917 break; 2918 2919 __set_bit_le(i, nm_i->nat_block_bitmap); 2920 2921 nid = i * NAT_ENTRY_PER_BLOCK; 2922 last_nid = nid + NAT_ENTRY_PER_BLOCK; 2923 2924 spin_lock(&NM_I(sbi)->nid_list_lock); 2925 for (; nid < last_nid; nid++) 2926 update_free_nid_bitmap(sbi, nid, true, true); 2927 spin_unlock(&NM_I(sbi)->nid_list_lock); 2928 } 2929 2930 for (i = 0; i < nm_i->nat_blocks; i++) { 2931 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i); 2932 if (i >= nm_i->nat_blocks) 2933 break; 2934 2935 __set_bit_le(i, nm_i->nat_block_bitmap); 2936 } 2937 } 2938 2939 static int init_node_manager(struct f2fs_sb_info *sbi) 2940 { 2941 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 2942 struct f2fs_nm_info *nm_i = NM_I(sbi); 2943 unsigned char *version_bitmap; 2944 unsigned int nat_segs; 2945 int err; 2946 2947 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 2948 2949 /* segment_count_nat includes pair segment so divide to 2. */ 2950 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 2951 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 2952 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks; 2953 2954 /* not used nids: 0, node, meta, (and root counted as valid node) */ 2955 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count - 2956 sbi->nquota_files - F2FS_RESERVED_NODE_NUM; 2957 nm_i->nid_cnt[FREE_NID] = 0; 2958 nm_i->nid_cnt[PREALLOC_NID] = 0; 2959 nm_i->nat_cnt = 0; 2960 nm_i->ram_thresh = DEF_RAM_THRESHOLD; 2961 nm_i->ra_nid_pages = DEF_RA_NID_PAGES; 2962 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD; 2963 2964 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC); 2965 INIT_LIST_HEAD(&nm_i->free_nid_list); 2966 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO); 2967 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO); 2968 INIT_LIST_HEAD(&nm_i->nat_entries); 2969 spin_lock_init(&nm_i->nat_list_lock); 2970 2971 mutex_init(&nm_i->build_lock); 2972 spin_lock_init(&nm_i->nid_list_lock); 2973 init_rwsem(&nm_i->nat_tree_lock); 2974 2975 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 2976 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 2977 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 2978 if (!version_bitmap) 2979 return -EFAULT; 2980 2981 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size, 2982 GFP_KERNEL); 2983 if (!nm_i->nat_bitmap) 2984 return -ENOMEM; 2985 2986 err = __get_nat_bitmaps(sbi); 2987 if (err) 2988 return err; 2989 2990 #ifdef CONFIG_F2FS_CHECK_FS 2991 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size, 2992 GFP_KERNEL); 2993 if (!nm_i->nat_bitmap_mir) 2994 return -ENOMEM; 2995 #endif 2996 2997 return 0; 2998 } 2999 3000 static int init_free_nid_cache(struct f2fs_sb_info *sbi) 3001 { 3002 struct f2fs_nm_info *nm_i = NM_I(sbi); 3003 int i; 3004 3005 nm_i->free_nid_bitmap = 3006 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *), 3007 nm_i->nat_blocks), 3008 GFP_KERNEL); 3009 if (!nm_i->free_nid_bitmap) 3010 return -ENOMEM; 3011 3012 for (i = 0; i < nm_i->nat_blocks; i++) { 3013 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi, 3014 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL); 3015 if (!nm_i->free_nid_bitmap[i]) 3016 return -ENOMEM; 3017 } 3018 3019 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8, 3020 GFP_KERNEL); 3021 if (!nm_i->nat_block_bitmap) 3022 return -ENOMEM; 3023 3024 nm_i->free_nid_count = 3025 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short), 3026 nm_i->nat_blocks), 3027 GFP_KERNEL); 3028 if (!nm_i->free_nid_count) 3029 return -ENOMEM; 3030 return 0; 3031 } 3032 3033 int f2fs_build_node_manager(struct f2fs_sb_info *sbi) 3034 { 3035 int err; 3036 3037 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info), 3038 GFP_KERNEL); 3039 if (!sbi->nm_info) 3040 return -ENOMEM; 3041 3042 err = init_node_manager(sbi); 3043 if (err) 3044 return err; 3045 3046 err = init_free_nid_cache(sbi); 3047 if (err) 3048 return err; 3049 3050 /* load free nid status from nat_bits table */ 3051 load_free_nid_bitmap(sbi); 3052 3053 return f2fs_build_free_nids(sbi, true, true); 3054 } 3055 3056 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi) 3057 { 3058 struct f2fs_nm_info *nm_i = NM_I(sbi); 3059 struct free_nid *i, *next_i; 3060 struct nat_entry *natvec[NATVEC_SIZE]; 3061 struct nat_entry_set *setvec[SETVEC_SIZE]; 3062 nid_t nid = 0; 3063 unsigned int found; 3064 3065 if (!nm_i) 3066 return; 3067 3068 /* destroy free nid list */ 3069 spin_lock(&nm_i->nid_list_lock); 3070 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 3071 __remove_free_nid(sbi, i, FREE_NID); 3072 spin_unlock(&nm_i->nid_list_lock); 3073 kmem_cache_free(free_nid_slab, i); 3074 spin_lock(&nm_i->nid_list_lock); 3075 } 3076 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]); 3077 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]); 3078 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list)); 3079 spin_unlock(&nm_i->nid_list_lock); 3080 3081 /* destroy nat cache */ 3082 down_write(&nm_i->nat_tree_lock); 3083 while ((found = __gang_lookup_nat_cache(nm_i, 3084 nid, NATVEC_SIZE, natvec))) { 3085 unsigned idx; 3086 3087 nid = nat_get_nid(natvec[found - 1]) + 1; 3088 for (idx = 0; idx < found; idx++) { 3089 spin_lock(&nm_i->nat_list_lock); 3090 list_del(&natvec[idx]->list); 3091 spin_unlock(&nm_i->nat_list_lock); 3092 3093 __del_from_nat_cache(nm_i, natvec[idx]); 3094 } 3095 } 3096 f2fs_bug_on(sbi, nm_i->nat_cnt); 3097 3098 /* destroy nat set cache */ 3099 nid = 0; 3100 while ((found = __gang_lookup_nat_set(nm_i, 3101 nid, SETVEC_SIZE, setvec))) { 3102 unsigned idx; 3103 3104 nid = setvec[found - 1]->set + 1; 3105 for (idx = 0; idx < found; idx++) { 3106 /* entry_cnt is not zero, when cp_error was occurred */ 3107 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list)); 3108 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set); 3109 kmem_cache_free(nat_entry_set_slab, setvec[idx]); 3110 } 3111 } 3112 up_write(&nm_i->nat_tree_lock); 3113 3114 kvfree(nm_i->nat_block_bitmap); 3115 if (nm_i->free_nid_bitmap) { 3116 int i; 3117 3118 for (i = 0; i < nm_i->nat_blocks; i++) 3119 kvfree(nm_i->free_nid_bitmap[i]); 3120 kvfree(nm_i->free_nid_bitmap); 3121 } 3122 kvfree(nm_i->free_nid_count); 3123 3124 kvfree(nm_i->nat_bitmap); 3125 kvfree(nm_i->nat_bits); 3126 #ifdef CONFIG_F2FS_CHECK_FS 3127 kvfree(nm_i->nat_bitmap_mir); 3128 #endif 3129 sbi->nm_info = NULL; 3130 kvfree(nm_i); 3131 } 3132 3133 int __init f2fs_create_node_manager_caches(void) 3134 { 3135 nat_entry_slab = f2fs_kmem_cache_create("nat_entry", 3136 sizeof(struct nat_entry)); 3137 if (!nat_entry_slab) 3138 goto fail; 3139 3140 free_nid_slab = f2fs_kmem_cache_create("free_nid", 3141 sizeof(struct free_nid)); 3142 if (!free_nid_slab) 3143 goto destroy_nat_entry; 3144 3145 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set", 3146 sizeof(struct nat_entry_set)); 3147 if (!nat_entry_set_slab) 3148 goto destroy_free_nid; 3149 3150 fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry", 3151 sizeof(struct fsync_node_entry)); 3152 if (!fsync_node_entry_slab) 3153 goto destroy_nat_entry_set; 3154 return 0; 3155 3156 destroy_nat_entry_set: 3157 kmem_cache_destroy(nat_entry_set_slab); 3158 destroy_free_nid: 3159 kmem_cache_destroy(free_nid_slab); 3160 destroy_nat_entry: 3161 kmem_cache_destroy(nat_entry_slab); 3162 fail: 3163 return -ENOMEM; 3164 } 3165 3166 void f2fs_destroy_node_manager_caches(void) 3167 { 3168 kmem_cache_destroy(fsync_node_entry_slab); 3169 kmem_cache_destroy(nat_entry_set_slab); 3170 kmem_cache_destroy(free_nid_slab); 3171 kmem_cache_destroy(nat_entry_slab); 3172 } 3173