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/sched/mm.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 "iostat.h" 21 #include <trace/events/f2fs.h> 22 23 #define on_f2fs_build_free_nids(nm_i) 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_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.", 38 __func__, nid); 39 f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE); 40 return -EFSCORRUPTED; 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 discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 49 struct sysinfo val; 50 unsigned long avail_ram; 51 unsigned long mem_size = 0; 52 bool res = false; 53 54 if (!nm_i) 55 return true; 56 57 si_meminfo(&val); 58 59 /* only uses low memory */ 60 avail_ram = val.totalram - val.totalhigh; 61 62 /* 63 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively 64 */ 65 if (type == FREE_NIDS) { 66 mem_size = (nm_i->nid_cnt[FREE_NID] * 67 sizeof(struct free_nid)) >> PAGE_SHIFT; 68 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 69 } else if (type == NAT_ENTRIES) { 70 mem_size = (nm_i->nat_cnt[TOTAL_NAT] * 71 sizeof(struct nat_entry)) >> PAGE_SHIFT; 72 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 73 if (excess_cached_nats(sbi)) 74 res = false; 75 } else if (type == DIRTY_DENTS) { 76 if (sbi->sb->s_bdi->wb.dirty_exceeded) 77 return false; 78 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS); 79 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 80 } else if (type == INO_ENTRIES) { 81 int i; 82 83 for (i = 0; i < MAX_INO_ENTRY; i++) 84 mem_size += sbi->im[i].ino_num * 85 sizeof(struct ino_entry); 86 mem_size >>= PAGE_SHIFT; 87 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 88 } else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) { 89 enum extent_type etype = type == READ_EXTENT_CACHE ? 90 EX_READ : EX_BLOCK_AGE; 91 struct extent_tree_info *eti = &sbi->extent_tree[etype]; 92 93 mem_size = (atomic_read(&eti->total_ext_tree) * 94 sizeof(struct extent_tree) + 95 atomic_read(&eti->total_ext_node) * 96 sizeof(struct extent_node)) >> PAGE_SHIFT; 97 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 98 } else if (type == DISCARD_CACHE) { 99 mem_size = (atomic_read(&dcc->discard_cmd_cnt) * 100 sizeof(struct discard_cmd)) >> PAGE_SHIFT; 101 res = mem_size < (avail_ram * nm_i->ram_thresh / 100); 102 } else if (type == COMPRESS_PAGE) { 103 #ifdef CONFIG_F2FS_FS_COMPRESSION 104 unsigned long free_ram = val.freeram; 105 106 /* 107 * free memory is lower than watermark or cached page count 108 * exceed threshold, deny caching compress page. 109 */ 110 res = (free_ram > avail_ram * sbi->compress_watermark / 100) && 111 (COMPRESS_MAPPING(sbi)->nrpages < 112 free_ram * sbi->compress_percent / 100); 113 #else 114 res = false; 115 #endif 116 } else { 117 if (!sbi->sb->s_bdi->wb.dirty_exceeded) 118 return true; 119 } 120 return res; 121 } 122 123 static void clear_node_page_dirty(struct page *page) 124 { 125 if (PageDirty(page)) { 126 f2fs_clear_page_cache_dirty_tag(page_folio(page)); 127 clear_page_dirty_for_io(page); 128 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES); 129 } 130 ClearPageUptodate(page); 131 } 132 133 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 134 { 135 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid)); 136 } 137 138 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 139 { 140 struct page *src_page; 141 struct page *dst_page; 142 pgoff_t dst_off; 143 void *src_addr; 144 void *dst_addr; 145 struct f2fs_nm_info *nm_i = NM_I(sbi); 146 147 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid)); 148 149 /* get current nat block page with lock */ 150 src_page = get_current_nat_page(sbi, nid); 151 if (IS_ERR(src_page)) 152 return src_page; 153 dst_page = f2fs_grab_meta_page(sbi, dst_off); 154 f2fs_bug_on(sbi, PageDirty(src_page)); 155 156 src_addr = page_address(src_page); 157 dst_addr = page_address(dst_page); 158 memcpy(dst_addr, src_addr, PAGE_SIZE); 159 set_page_dirty(dst_page); 160 f2fs_put_page(src_page, 1); 161 162 set_to_next_nat(nm_i, nid); 163 164 return dst_page; 165 } 166 167 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi, 168 nid_t nid, bool no_fail) 169 { 170 struct nat_entry *new; 171 172 new = f2fs_kmem_cache_alloc(nat_entry_slab, 173 GFP_F2FS_ZERO, no_fail, sbi); 174 if (new) { 175 nat_set_nid(new, nid); 176 nat_reset_flag(new); 177 } 178 return new; 179 } 180 181 static void __free_nat_entry(struct nat_entry *e) 182 { 183 kmem_cache_free(nat_entry_slab, e); 184 } 185 186 /* must be locked by nat_tree_lock */ 187 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i, 188 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail) 189 { 190 if (no_fail) 191 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne); 192 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne)) 193 return NULL; 194 195 if (raw_ne) 196 node_info_from_raw_nat(&ne->ni, raw_ne); 197 198 spin_lock(&nm_i->nat_list_lock); 199 list_add_tail(&ne->list, &nm_i->nat_entries); 200 spin_unlock(&nm_i->nat_list_lock); 201 202 nm_i->nat_cnt[TOTAL_NAT]++; 203 nm_i->nat_cnt[RECLAIMABLE_NAT]++; 204 return ne; 205 } 206 207 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n) 208 { 209 struct nat_entry *ne; 210 211 ne = radix_tree_lookup(&nm_i->nat_root, n); 212 213 /* for recent accessed nat entry, move it to tail of lru list */ 214 if (ne && !get_nat_flag(ne, IS_DIRTY)) { 215 spin_lock(&nm_i->nat_list_lock); 216 if (!list_empty(&ne->list)) 217 list_move_tail(&ne->list, &nm_i->nat_entries); 218 spin_unlock(&nm_i->nat_list_lock); 219 } 220 221 return ne; 222 } 223 224 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i, 225 nid_t start, unsigned int nr, struct nat_entry **ep) 226 { 227 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr); 228 } 229 230 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e) 231 { 232 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e)); 233 nm_i->nat_cnt[TOTAL_NAT]--; 234 nm_i->nat_cnt[RECLAIMABLE_NAT]--; 235 __free_nat_entry(e); 236 } 237 238 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i, 239 struct nat_entry *ne) 240 { 241 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid); 242 struct nat_entry_set *head; 243 244 head = radix_tree_lookup(&nm_i->nat_set_root, set); 245 if (!head) { 246 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, 247 GFP_NOFS, true, NULL); 248 249 INIT_LIST_HEAD(&head->entry_list); 250 INIT_LIST_HEAD(&head->set_list); 251 head->set = set; 252 head->entry_cnt = 0; 253 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head); 254 } 255 return head; 256 } 257 258 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i, 259 struct nat_entry *ne) 260 { 261 struct nat_entry_set *head; 262 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR; 263 264 if (!new_ne) 265 head = __grab_nat_entry_set(nm_i, ne); 266 267 /* 268 * update entry_cnt in below condition: 269 * 1. update NEW_ADDR to valid block address; 270 * 2. update old block address to new one; 271 */ 272 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) || 273 !get_nat_flag(ne, IS_DIRTY))) 274 head->entry_cnt++; 275 276 set_nat_flag(ne, IS_PREALLOC, new_ne); 277 278 if (get_nat_flag(ne, IS_DIRTY)) 279 goto refresh_list; 280 281 nm_i->nat_cnt[DIRTY_NAT]++; 282 nm_i->nat_cnt[RECLAIMABLE_NAT]--; 283 set_nat_flag(ne, IS_DIRTY, true); 284 refresh_list: 285 spin_lock(&nm_i->nat_list_lock); 286 if (new_ne) 287 list_del_init(&ne->list); 288 else 289 list_move_tail(&ne->list, &head->entry_list); 290 spin_unlock(&nm_i->nat_list_lock); 291 } 292 293 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i, 294 struct nat_entry_set *set, struct nat_entry *ne) 295 { 296 spin_lock(&nm_i->nat_list_lock); 297 list_move_tail(&ne->list, &nm_i->nat_entries); 298 spin_unlock(&nm_i->nat_list_lock); 299 300 set_nat_flag(ne, IS_DIRTY, false); 301 set->entry_cnt--; 302 nm_i->nat_cnt[DIRTY_NAT]--; 303 nm_i->nat_cnt[RECLAIMABLE_NAT]++; 304 } 305 306 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i, 307 nid_t start, unsigned int nr, struct nat_entry_set **ep) 308 { 309 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep, 310 start, nr); 311 } 312 313 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page) 314 { 315 return NODE_MAPPING(sbi) == page->mapping && 316 IS_DNODE(page) && is_cold_node(page); 317 } 318 319 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi) 320 { 321 spin_lock_init(&sbi->fsync_node_lock); 322 INIT_LIST_HEAD(&sbi->fsync_node_list); 323 sbi->fsync_seg_id = 0; 324 sbi->fsync_node_num = 0; 325 } 326 327 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi, 328 struct page *page) 329 { 330 struct fsync_node_entry *fn; 331 unsigned long flags; 332 unsigned int seq_id; 333 334 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, 335 GFP_NOFS, true, NULL); 336 337 get_page(page); 338 fn->page = page; 339 INIT_LIST_HEAD(&fn->list); 340 341 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 342 list_add_tail(&fn->list, &sbi->fsync_node_list); 343 fn->seq_id = sbi->fsync_seg_id++; 344 seq_id = fn->seq_id; 345 sbi->fsync_node_num++; 346 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 347 348 return seq_id; 349 } 350 351 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page) 352 { 353 struct fsync_node_entry *fn; 354 unsigned long flags; 355 356 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 357 list_for_each_entry(fn, &sbi->fsync_node_list, list) { 358 if (fn->page == page) { 359 list_del(&fn->list); 360 sbi->fsync_node_num--; 361 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 362 kmem_cache_free(fsync_node_entry_slab, fn); 363 put_page(page); 364 return; 365 } 366 } 367 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 368 f2fs_bug_on(sbi, 1); 369 } 370 371 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi) 372 { 373 unsigned long flags; 374 375 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 376 sbi->fsync_seg_id = 0; 377 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 378 } 379 380 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid) 381 { 382 struct f2fs_nm_info *nm_i = NM_I(sbi); 383 struct nat_entry *e; 384 bool need = false; 385 386 f2fs_down_read(&nm_i->nat_tree_lock); 387 e = __lookup_nat_cache(nm_i, nid); 388 if (e) { 389 if (!get_nat_flag(e, IS_CHECKPOINTED) && 390 !get_nat_flag(e, HAS_FSYNCED_INODE)) 391 need = true; 392 } 393 f2fs_up_read(&nm_i->nat_tree_lock); 394 return need; 395 } 396 397 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid) 398 { 399 struct f2fs_nm_info *nm_i = NM_I(sbi); 400 struct nat_entry *e; 401 bool is_cp = true; 402 403 f2fs_down_read(&nm_i->nat_tree_lock); 404 e = __lookup_nat_cache(nm_i, nid); 405 if (e && !get_nat_flag(e, IS_CHECKPOINTED)) 406 is_cp = false; 407 f2fs_up_read(&nm_i->nat_tree_lock); 408 return is_cp; 409 } 410 411 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino) 412 { 413 struct f2fs_nm_info *nm_i = NM_I(sbi); 414 struct nat_entry *e; 415 bool need_update = true; 416 417 f2fs_down_read(&nm_i->nat_tree_lock); 418 e = __lookup_nat_cache(nm_i, ino); 419 if (e && get_nat_flag(e, HAS_LAST_FSYNC) && 420 (get_nat_flag(e, IS_CHECKPOINTED) || 421 get_nat_flag(e, HAS_FSYNCED_INODE))) 422 need_update = false; 423 f2fs_up_read(&nm_i->nat_tree_lock); 424 return need_update; 425 } 426 427 /* must be locked by nat_tree_lock */ 428 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid, 429 struct f2fs_nat_entry *ne) 430 { 431 struct f2fs_nm_info *nm_i = NM_I(sbi); 432 struct nat_entry *new, *e; 433 434 /* Let's mitigate lock contention of nat_tree_lock during checkpoint */ 435 if (f2fs_rwsem_is_locked(&sbi->cp_global_sem)) 436 return; 437 438 new = __alloc_nat_entry(sbi, nid, false); 439 if (!new) 440 return; 441 442 f2fs_down_write(&nm_i->nat_tree_lock); 443 e = __lookup_nat_cache(nm_i, nid); 444 if (!e) 445 e = __init_nat_entry(nm_i, new, ne, false); 446 else 447 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) || 448 nat_get_blkaddr(e) != 449 le32_to_cpu(ne->block_addr) || 450 nat_get_version(e) != ne->version); 451 f2fs_up_write(&nm_i->nat_tree_lock); 452 if (e != new) 453 __free_nat_entry(new); 454 } 455 456 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni, 457 block_t new_blkaddr, bool fsync_done) 458 { 459 struct f2fs_nm_info *nm_i = NM_I(sbi); 460 struct nat_entry *e; 461 struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true); 462 463 f2fs_down_write(&nm_i->nat_tree_lock); 464 e = __lookup_nat_cache(nm_i, ni->nid); 465 if (!e) { 466 e = __init_nat_entry(nm_i, new, NULL, true); 467 copy_node_info(&e->ni, ni); 468 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR); 469 } else if (new_blkaddr == NEW_ADDR) { 470 /* 471 * when nid is reallocated, 472 * previous nat entry can be remained in nat cache. 473 * So, reinitialize it with new information. 474 */ 475 copy_node_info(&e->ni, ni); 476 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR); 477 } 478 /* let's free early to reduce memory consumption */ 479 if (e != new) 480 __free_nat_entry(new); 481 482 /* sanity check */ 483 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr); 484 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR && 485 new_blkaddr == NULL_ADDR); 486 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR && 487 new_blkaddr == NEW_ADDR); 488 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) && 489 new_blkaddr == NEW_ADDR); 490 491 /* increment version no as node is removed */ 492 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) { 493 unsigned char version = nat_get_version(e); 494 495 nat_set_version(e, inc_node_version(version)); 496 } 497 498 /* change address */ 499 nat_set_blkaddr(e, new_blkaddr); 500 if (!__is_valid_data_blkaddr(new_blkaddr)) 501 set_nat_flag(e, IS_CHECKPOINTED, false); 502 __set_nat_cache_dirty(nm_i, e); 503 504 /* update fsync_mark if its inode nat entry is still alive */ 505 if (ni->nid != ni->ino) 506 e = __lookup_nat_cache(nm_i, ni->ino); 507 if (e) { 508 if (fsync_done && ni->nid == ni->ino) 509 set_nat_flag(e, HAS_FSYNCED_INODE, true); 510 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done); 511 } 512 f2fs_up_write(&nm_i->nat_tree_lock); 513 } 514 515 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink) 516 { 517 struct f2fs_nm_info *nm_i = NM_I(sbi); 518 int nr = nr_shrink; 519 520 if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock)) 521 return 0; 522 523 spin_lock(&nm_i->nat_list_lock); 524 while (nr_shrink) { 525 struct nat_entry *ne; 526 527 if (list_empty(&nm_i->nat_entries)) 528 break; 529 530 ne = list_first_entry(&nm_i->nat_entries, 531 struct nat_entry, list); 532 list_del(&ne->list); 533 spin_unlock(&nm_i->nat_list_lock); 534 535 __del_from_nat_cache(nm_i, ne); 536 nr_shrink--; 537 538 spin_lock(&nm_i->nat_list_lock); 539 } 540 spin_unlock(&nm_i->nat_list_lock); 541 542 f2fs_up_write(&nm_i->nat_tree_lock); 543 return nr - nr_shrink; 544 } 545 546 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid, 547 struct node_info *ni, bool checkpoint_context) 548 { 549 struct f2fs_nm_info *nm_i = NM_I(sbi); 550 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 551 struct f2fs_journal *journal = curseg->journal; 552 nid_t start_nid = START_NID(nid); 553 struct f2fs_nat_block *nat_blk; 554 struct page *page = NULL; 555 struct f2fs_nat_entry ne; 556 struct nat_entry *e; 557 pgoff_t index; 558 block_t blkaddr; 559 int i; 560 561 ni->flag = 0; 562 ni->nid = nid; 563 retry: 564 /* Check nat cache */ 565 f2fs_down_read(&nm_i->nat_tree_lock); 566 e = __lookup_nat_cache(nm_i, nid); 567 if (e) { 568 ni->ino = nat_get_ino(e); 569 ni->blk_addr = nat_get_blkaddr(e); 570 ni->version = nat_get_version(e); 571 f2fs_up_read(&nm_i->nat_tree_lock); 572 return 0; 573 } 574 575 /* 576 * Check current segment summary by trying to grab journal_rwsem first. 577 * This sem is on the critical path on the checkpoint requiring the above 578 * nat_tree_lock. Therefore, we should retry, if we failed to grab here 579 * while not bothering checkpoint. 580 */ 581 if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) { 582 down_read(&curseg->journal_rwsem); 583 } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) || 584 !down_read_trylock(&curseg->journal_rwsem)) { 585 f2fs_up_read(&nm_i->nat_tree_lock); 586 goto retry; 587 } 588 589 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0); 590 if (i >= 0) { 591 ne = nat_in_journal(journal, i); 592 node_info_from_raw_nat(ni, &ne); 593 } 594 up_read(&curseg->journal_rwsem); 595 if (i >= 0) { 596 f2fs_up_read(&nm_i->nat_tree_lock); 597 goto cache; 598 } 599 600 /* Fill node_info from nat page */ 601 index = current_nat_addr(sbi, nid); 602 f2fs_up_read(&nm_i->nat_tree_lock); 603 604 page = f2fs_get_meta_page(sbi, index); 605 if (IS_ERR(page)) 606 return PTR_ERR(page); 607 608 nat_blk = (struct f2fs_nat_block *)page_address(page); 609 ne = nat_blk->entries[nid - start_nid]; 610 node_info_from_raw_nat(ni, &ne); 611 f2fs_put_page(page, 1); 612 cache: 613 blkaddr = le32_to_cpu(ne.block_addr); 614 if (__is_valid_data_blkaddr(blkaddr) && 615 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) 616 return -EFAULT; 617 618 /* cache nat entry */ 619 cache_nat_entry(sbi, nid, &ne); 620 return 0; 621 } 622 623 /* 624 * readahead MAX_RA_NODE number of node pages. 625 */ 626 static void f2fs_ra_node_pages(struct page *parent, int start, int n) 627 { 628 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 629 struct blk_plug plug; 630 int i, end; 631 nid_t nid; 632 633 blk_start_plug(&plug); 634 635 /* Then, try readahead for siblings of the desired node */ 636 end = start + n; 637 end = min(end, (int)NIDS_PER_BLOCK); 638 for (i = start; i < end; i++) { 639 nid = get_nid(parent, i, false); 640 f2fs_ra_node_page(sbi, nid); 641 } 642 643 blk_finish_plug(&plug); 644 } 645 646 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs) 647 { 648 const long direct_index = ADDRS_PER_INODE(dn->inode); 649 const long direct_blks = ADDRS_PER_BLOCK(dn->inode); 650 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK; 651 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode); 652 int cur_level = dn->cur_level; 653 int max_level = dn->max_level; 654 pgoff_t base = 0; 655 656 if (!dn->max_level) 657 return pgofs + 1; 658 659 while (max_level-- > cur_level) 660 skipped_unit *= NIDS_PER_BLOCK; 661 662 switch (dn->max_level) { 663 case 3: 664 base += 2 * indirect_blks; 665 fallthrough; 666 case 2: 667 base += 2 * direct_blks; 668 fallthrough; 669 case 1: 670 base += direct_index; 671 break; 672 default: 673 f2fs_bug_on(F2FS_I_SB(dn->inode), 1); 674 } 675 676 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base; 677 } 678 679 /* 680 * The maximum depth is four. 681 * Offset[0] will have raw inode offset. 682 */ 683 static int get_node_path(struct inode *inode, long block, 684 int offset[4], unsigned int noffset[4]) 685 { 686 const long direct_index = ADDRS_PER_INODE(inode); 687 const long direct_blks = ADDRS_PER_BLOCK(inode); 688 const long dptrs_per_blk = NIDS_PER_BLOCK; 689 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK; 690 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK; 691 int n = 0; 692 int level = 0; 693 694 noffset[0] = 0; 695 696 if (block < direct_index) { 697 offset[n] = block; 698 goto got; 699 } 700 block -= direct_index; 701 if (block < direct_blks) { 702 offset[n++] = NODE_DIR1_BLOCK; 703 noffset[n] = 1; 704 offset[n] = block; 705 level = 1; 706 goto got; 707 } 708 block -= direct_blks; 709 if (block < direct_blks) { 710 offset[n++] = NODE_DIR2_BLOCK; 711 noffset[n] = 2; 712 offset[n] = block; 713 level = 1; 714 goto got; 715 } 716 block -= direct_blks; 717 if (block < indirect_blks) { 718 offset[n++] = NODE_IND1_BLOCK; 719 noffset[n] = 3; 720 offset[n++] = block / direct_blks; 721 noffset[n] = 4 + offset[n - 1]; 722 offset[n] = block % direct_blks; 723 level = 2; 724 goto got; 725 } 726 block -= indirect_blks; 727 if (block < indirect_blks) { 728 offset[n++] = NODE_IND2_BLOCK; 729 noffset[n] = 4 + dptrs_per_blk; 730 offset[n++] = block / direct_blks; 731 noffset[n] = 5 + dptrs_per_blk + offset[n - 1]; 732 offset[n] = block % direct_blks; 733 level = 2; 734 goto got; 735 } 736 block -= indirect_blks; 737 if (block < dindirect_blks) { 738 offset[n++] = NODE_DIND_BLOCK; 739 noffset[n] = 5 + (dptrs_per_blk * 2); 740 offset[n++] = block / indirect_blks; 741 noffset[n] = 6 + (dptrs_per_blk * 2) + 742 offset[n - 1] * (dptrs_per_blk + 1); 743 offset[n++] = (block / direct_blks) % dptrs_per_blk; 744 noffset[n] = 7 + (dptrs_per_blk * 2) + 745 offset[n - 2] * (dptrs_per_blk + 1) + 746 offset[n - 1]; 747 offset[n] = block % direct_blks; 748 level = 3; 749 goto got; 750 } else { 751 return -E2BIG; 752 } 753 got: 754 return level; 755 } 756 757 /* 758 * Caller should call f2fs_put_dnode(dn). 759 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and 760 * f2fs_unlock_op() only if mode is set with ALLOC_NODE. 761 */ 762 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode) 763 { 764 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 765 struct page *npage[4]; 766 struct page *parent = NULL; 767 int offset[4]; 768 unsigned int noffset[4]; 769 nid_t nids[4]; 770 int level, i = 0; 771 int err = 0; 772 773 level = get_node_path(dn->inode, index, offset, noffset); 774 if (level < 0) 775 return level; 776 777 nids[0] = dn->inode->i_ino; 778 npage[0] = dn->inode_page; 779 780 if (!npage[0]) { 781 npage[0] = f2fs_get_node_page(sbi, nids[0]); 782 if (IS_ERR(npage[0])) 783 return PTR_ERR(npage[0]); 784 } 785 786 /* if inline_data is set, should not report any block indices */ 787 if (f2fs_has_inline_data(dn->inode) && index) { 788 err = -ENOENT; 789 f2fs_put_page(npage[0], 1); 790 goto release_out; 791 } 792 793 parent = npage[0]; 794 if (level != 0) 795 nids[1] = get_nid(parent, offset[0], true); 796 dn->inode_page = npage[0]; 797 dn->inode_page_locked = true; 798 799 /* get indirect or direct nodes */ 800 for (i = 1; i <= level; i++) { 801 bool done = false; 802 803 if (!nids[i] && mode == ALLOC_NODE) { 804 /* alloc new node */ 805 if (!f2fs_alloc_nid(sbi, &(nids[i]))) { 806 err = -ENOSPC; 807 goto release_pages; 808 } 809 810 dn->nid = nids[i]; 811 npage[i] = f2fs_new_node_page(dn, noffset[i]); 812 if (IS_ERR(npage[i])) { 813 f2fs_alloc_nid_failed(sbi, nids[i]); 814 err = PTR_ERR(npage[i]); 815 goto release_pages; 816 } 817 818 set_nid(parent, offset[i - 1], nids[i], i == 1); 819 f2fs_alloc_nid_done(sbi, nids[i]); 820 done = true; 821 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) { 822 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]); 823 if (IS_ERR(npage[i])) { 824 err = PTR_ERR(npage[i]); 825 goto release_pages; 826 } 827 done = true; 828 } 829 if (i == 1) { 830 dn->inode_page_locked = false; 831 unlock_page(parent); 832 } else { 833 f2fs_put_page(parent, 1); 834 } 835 836 if (!done) { 837 npage[i] = f2fs_get_node_page(sbi, nids[i]); 838 if (IS_ERR(npage[i])) { 839 err = PTR_ERR(npage[i]); 840 f2fs_put_page(npage[0], 0); 841 goto release_out; 842 } 843 } 844 if (i < level) { 845 parent = npage[i]; 846 nids[i + 1] = get_nid(parent, offset[i], false); 847 } 848 } 849 dn->nid = nids[level]; 850 dn->ofs_in_node = offset[level]; 851 dn->node_page = npage[level]; 852 dn->data_blkaddr = f2fs_data_blkaddr(dn); 853 854 if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) && 855 f2fs_sb_has_readonly(sbi)) { 856 unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size; 857 unsigned int ofs_in_node = dn->ofs_in_node; 858 pgoff_t fofs = index; 859 unsigned int c_len; 860 block_t blkaddr; 861 862 /* should align fofs and ofs_in_node to cluster_size */ 863 if (fofs % cluster_size) { 864 fofs = round_down(fofs, cluster_size); 865 ofs_in_node = round_down(ofs_in_node, cluster_size); 866 } 867 868 c_len = f2fs_cluster_blocks_are_contiguous(dn, ofs_in_node); 869 if (!c_len) 870 goto out; 871 872 blkaddr = data_blkaddr(dn->inode, dn->node_page, ofs_in_node); 873 if (blkaddr == COMPRESS_ADDR) 874 blkaddr = data_blkaddr(dn->inode, dn->node_page, 875 ofs_in_node + 1); 876 877 f2fs_update_read_extent_tree_range_compressed(dn->inode, 878 fofs, blkaddr, cluster_size, c_len); 879 } 880 out: 881 return 0; 882 883 release_pages: 884 f2fs_put_page(parent, 1); 885 if (i > 1) 886 f2fs_put_page(npage[0], 0); 887 release_out: 888 dn->inode_page = NULL; 889 dn->node_page = NULL; 890 if (err == -ENOENT) { 891 dn->cur_level = i; 892 dn->max_level = level; 893 dn->ofs_in_node = offset[level]; 894 } 895 return err; 896 } 897 898 static int truncate_node(struct dnode_of_data *dn) 899 { 900 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 901 struct node_info ni; 902 int err; 903 pgoff_t index; 904 905 err = f2fs_get_node_info(sbi, dn->nid, &ni, false); 906 if (err) 907 return err; 908 909 if (ni.blk_addr != NEW_ADDR && 910 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC_ENHANCE)) { 911 f2fs_err_ratelimited(sbi, 912 "nat entry is corrupted, run fsck to fix it, ino:%u, " 913 "nid:%u, blkaddr:%u", ni.ino, ni.nid, ni.blk_addr); 914 set_sbi_flag(sbi, SBI_NEED_FSCK); 915 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT); 916 return -EFSCORRUPTED; 917 } 918 919 /* Deallocate node address */ 920 f2fs_invalidate_blocks(sbi, ni.blk_addr, 1); 921 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino); 922 set_node_addr(sbi, &ni, NULL_ADDR, false); 923 924 if (dn->nid == dn->inode->i_ino) { 925 f2fs_remove_orphan_inode(sbi, dn->nid); 926 dec_valid_inode_count(sbi); 927 f2fs_inode_synced(dn->inode); 928 } 929 930 clear_node_page_dirty(dn->node_page); 931 set_sbi_flag(sbi, SBI_IS_DIRTY); 932 933 index = page_folio(dn->node_page)->index; 934 f2fs_put_page(dn->node_page, 1); 935 936 invalidate_mapping_pages(NODE_MAPPING(sbi), 937 index, index); 938 939 dn->node_page = NULL; 940 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr); 941 942 return 0; 943 } 944 945 static int truncate_dnode(struct dnode_of_data *dn) 946 { 947 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 948 struct page *page; 949 int err; 950 951 if (dn->nid == 0) 952 return 1; 953 954 /* get direct node */ 955 page = f2fs_get_node_page(sbi, dn->nid); 956 if (PTR_ERR(page) == -ENOENT) 957 return 1; 958 else if (IS_ERR(page)) 959 return PTR_ERR(page); 960 961 if (IS_INODE(page) || ino_of_node(page) != dn->inode->i_ino) { 962 f2fs_err(sbi, "incorrect node reference, ino: %lu, nid: %u, ino_of_node: %u", 963 dn->inode->i_ino, dn->nid, ino_of_node(page)); 964 set_sbi_flag(sbi, SBI_NEED_FSCK); 965 f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE); 966 f2fs_put_page(page, 1); 967 return -EFSCORRUPTED; 968 } 969 970 /* Make dnode_of_data for parameter */ 971 dn->node_page = page; 972 dn->ofs_in_node = 0; 973 f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode)); 974 err = truncate_node(dn); 975 if (err) { 976 f2fs_put_page(page, 1); 977 return err; 978 } 979 980 return 1; 981 } 982 983 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs, 984 int ofs, int depth) 985 { 986 struct dnode_of_data rdn = *dn; 987 struct page *page; 988 struct f2fs_node *rn; 989 nid_t child_nid; 990 unsigned int child_nofs; 991 int freed = 0; 992 int i, ret; 993 994 if (dn->nid == 0) 995 return NIDS_PER_BLOCK + 1; 996 997 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr); 998 999 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 1000 if (IS_ERR(page)) { 1001 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page)); 1002 return PTR_ERR(page); 1003 } 1004 1005 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK); 1006 1007 rn = F2FS_NODE(page); 1008 if (depth < 3) { 1009 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) { 1010 child_nid = le32_to_cpu(rn->in.nid[i]); 1011 if (child_nid == 0) 1012 continue; 1013 rdn.nid = child_nid; 1014 ret = truncate_dnode(&rdn); 1015 if (ret < 0) 1016 goto out_err; 1017 if (set_nid(page, i, 0, false)) 1018 dn->node_changed = true; 1019 } 1020 } else { 1021 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1; 1022 for (i = ofs; i < NIDS_PER_BLOCK; i++) { 1023 child_nid = le32_to_cpu(rn->in.nid[i]); 1024 if (child_nid == 0) { 1025 child_nofs += NIDS_PER_BLOCK + 1; 1026 continue; 1027 } 1028 rdn.nid = child_nid; 1029 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1); 1030 if (ret == (NIDS_PER_BLOCK + 1)) { 1031 if (set_nid(page, i, 0, false)) 1032 dn->node_changed = true; 1033 child_nofs += ret; 1034 } else if (ret < 0 && ret != -ENOENT) { 1035 goto out_err; 1036 } 1037 } 1038 freed = child_nofs; 1039 } 1040 1041 if (!ofs) { 1042 /* remove current indirect node */ 1043 dn->node_page = page; 1044 ret = truncate_node(dn); 1045 if (ret) 1046 goto out_err; 1047 freed++; 1048 } else { 1049 f2fs_put_page(page, 1); 1050 } 1051 trace_f2fs_truncate_nodes_exit(dn->inode, freed); 1052 return freed; 1053 1054 out_err: 1055 f2fs_put_page(page, 1); 1056 trace_f2fs_truncate_nodes_exit(dn->inode, ret); 1057 return ret; 1058 } 1059 1060 static int truncate_partial_nodes(struct dnode_of_data *dn, 1061 struct f2fs_inode *ri, int *offset, int depth) 1062 { 1063 struct page *pages[2]; 1064 nid_t nid[3]; 1065 nid_t child_nid; 1066 int err = 0; 1067 int i; 1068 int idx = depth - 2; 1069 1070 nid[0] = get_nid(dn->inode_page, offset[0], true); 1071 if (!nid[0]) 1072 return 0; 1073 1074 /* get indirect nodes in the path */ 1075 for (i = 0; i < idx + 1; i++) { 1076 /* reference count'll be increased */ 1077 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]); 1078 if (IS_ERR(pages[i])) { 1079 err = PTR_ERR(pages[i]); 1080 idx = i - 1; 1081 goto fail; 1082 } 1083 nid[i + 1] = get_nid(pages[i], offset[i + 1], false); 1084 } 1085 1086 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK); 1087 1088 /* free direct nodes linked to a partial indirect node */ 1089 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) { 1090 child_nid = get_nid(pages[idx], i, false); 1091 if (!child_nid) 1092 continue; 1093 dn->nid = child_nid; 1094 err = truncate_dnode(dn); 1095 if (err < 0) 1096 goto fail; 1097 if (set_nid(pages[idx], i, 0, false)) 1098 dn->node_changed = true; 1099 } 1100 1101 if (offset[idx + 1] == 0) { 1102 dn->node_page = pages[idx]; 1103 dn->nid = nid[idx]; 1104 err = truncate_node(dn); 1105 if (err) 1106 goto fail; 1107 } else { 1108 f2fs_put_page(pages[idx], 1); 1109 } 1110 offset[idx]++; 1111 offset[idx + 1] = 0; 1112 idx--; 1113 fail: 1114 for (i = idx; i >= 0; i--) 1115 f2fs_put_page(pages[i], 1); 1116 1117 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err); 1118 1119 return err; 1120 } 1121 1122 /* 1123 * All the block addresses of data and nodes should be nullified. 1124 */ 1125 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from) 1126 { 1127 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1128 int err = 0, cont = 1; 1129 int level, offset[4], noffset[4]; 1130 unsigned int nofs = 0; 1131 struct f2fs_inode *ri; 1132 struct dnode_of_data dn; 1133 struct page *page; 1134 1135 trace_f2fs_truncate_inode_blocks_enter(inode, from); 1136 1137 level = get_node_path(inode, from, offset, noffset); 1138 if (level < 0) { 1139 trace_f2fs_truncate_inode_blocks_exit(inode, level); 1140 return level; 1141 } 1142 1143 page = f2fs_get_node_page(sbi, inode->i_ino); 1144 if (IS_ERR(page)) { 1145 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page)); 1146 return PTR_ERR(page); 1147 } 1148 1149 set_new_dnode(&dn, inode, page, NULL, 0); 1150 unlock_page(page); 1151 1152 ri = F2FS_INODE(page); 1153 switch (level) { 1154 case 0: 1155 case 1: 1156 nofs = noffset[1]; 1157 break; 1158 case 2: 1159 nofs = noffset[1]; 1160 if (!offset[level - 1]) 1161 goto skip_partial; 1162 err = truncate_partial_nodes(&dn, ri, offset, level); 1163 if (err < 0 && err != -ENOENT) 1164 goto fail; 1165 nofs += 1 + NIDS_PER_BLOCK; 1166 break; 1167 case 3: 1168 nofs = 5 + 2 * NIDS_PER_BLOCK; 1169 if (!offset[level - 1]) 1170 goto skip_partial; 1171 err = truncate_partial_nodes(&dn, ri, offset, level); 1172 if (err < 0 && err != -ENOENT) 1173 goto fail; 1174 break; 1175 default: 1176 BUG(); 1177 } 1178 1179 skip_partial: 1180 while (cont) { 1181 dn.nid = get_nid(page, offset[0], true); 1182 switch (offset[0]) { 1183 case NODE_DIR1_BLOCK: 1184 case NODE_DIR2_BLOCK: 1185 err = truncate_dnode(&dn); 1186 break; 1187 1188 case NODE_IND1_BLOCK: 1189 case NODE_IND2_BLOCK: 1190 err = truncate_nodes(&dn, nofs, offset[1], 2); 1191 break; 1192 1193 case NODE_DIND_BLOCK: 1194 err = truncate_nodes(&dn, nofs, offset[1], 3); 1195 cont = 0; 1196 break; 1197 1198 default: 1199 BUG(); 1200 } 1201 if (err == -ENOENT) { 1202 set_sbi_flag(F2FS_P_SB(page), SBI_NEED_FSCK); 1203 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 1204 f2fs_err_ratelimited(sbi, 1205 "truncate node fail, ino:%lu, nid:%u, " 1206 "offset[0]:%d, offset[1]:%d, nofs:%d", 1207 inode->i_ino, dn.nid, offset[0], 1208 offset[1], nofs); 1209 err = 0; 1210 } 1211 if (err < 0) 1212 goto fail; 1213 if (offset[1] == 0 && get_nid(page, offset[0], true)) { 1214 lock_page(page); 1215 BUG_ON(page->mapping != NODE_MAPPING(sbi)); 1216 set_nid(page, offset[0], 0, true); 1217 unlock_page(page); 1218 } 1219 offset[1] = 0; 1220 offset[0]++; 1221 nofs += err; 1222 } 1223 fail: 1224 f2fs_put_page(page, 0); 1225 trace_f2fs_truncate_inode_blocks_exit(inode, err); 1226 return err > 0 ? 0 : err; 1227 } 1228 1229 /* caller must lock inode page */ 1230 int f2fs_truncate_xattr_node(struct inode *inode) 1231 { 1232 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1233 nid_t nid = F2FS_I(inode)->i_xattr_nid; 1234 struct dnode_of_data dn; 1235 struct page *npage; 1236 int err; 1237 1238 if (!nid) 1239 return 0; 1240 1241 npage = f2fs_get_node_page(sbi, nid); 1242 if (IS_ERR(npage)) 1243 return PTR_ERR(npage); 1244 1245 set_new_dnode(&dn, inode, NULL, npage, nid); 1246 err = truncate_node(&dn); 1247 if (err) { 1248 f2fs_put_page(npage, 1); 1249 return err; 1250 } 1251 1252 f2fs_i_xnid_write(inode, 0); 1253 1254 return 0; 1255 } 1256 1257 /* 1258 * Caller should grab and release a rwsem by calling f2fs_lock_op() and 1259 * f2fs_unlock_op(). 1260 */ 1261 int f2fs_remove_inode_page(struct inode *inode) 1262 { 1263 struct dnode_of_data dn; 1264 int err; 1265 1266 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1267 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE); 1268 if (err) 1269 return err; 1270 1271 err = f2fs_truncate_xattr_node(inode); 1272 if (err) { 1273 f2fs_put_dnode(&dn); 1274 return err; 1275 } 1276 1277 /* remove potential inline_data blocks */ 1278 if (!IS_DEVICE_ALIASING(inode) && 1279 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1280 S_ISLNK(inode->i_mode))) 1281 f2fs_truncate_data_blocks_range(&dn, 1); 1282 1283 /* 0 is possible, after f2fs_new_inode() has failed */ 1284 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) { 1285 f2fs_put_dnode(&dn); 1286 return -EIO; 1287 } 1288 1289 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) { 1290 f2fs_warn(F2FS_I_SB(inode), 1291 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu", 1292 inode->i_ino, (unsigned long long)inode->i_blocks); 1293 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK); 1294 } 1295 1296 /* will put inode & node pages */ 1297 err = truncate_node(&dn); 1298 if (err) { 1299 f2fs_put_dnode(&dn); 1300 return err; 1301 } 1302 return 0; 1303 } 1304 1305 struct page *f2fs_new_inode_page(struct inode *inode) 1306 { 1307 struct dnode_of_data dn; 1308 1309 /* allocate inode page for new inode */ 1310 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1311 1312 /* caller should f2fs_put_page(page, 1); */ 1313 return f2fs_new_node_page(&dn, 0); 1314 } 1315 1316 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs) 1317 { 1318 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1319 struct node_info new_ni; 1320 struct page *page; 1321 int err; 1322 1323 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1324 return ERR_PTR(-EPERM); 1325 1326 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false); 1327 if (!page) 1328 return ERR_PTR(-ENOMEM); 1329 1330 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs)))) 1331 goto fail; 1332 1333 #ifdef CONFIG_F2FS_CHECK_FS 1334 err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false); 1335 if (err) { 1336 dec_valid_node_count(sbi, dn->inode, !ofs); 1337 goto fail; 1338 } 1339 if (unlikely(new_ni.blk_addr != NULL_ADDR)) { 1340 err = -EFSCORRUPTED; 1341 dec_valid_node_count(sbi, dn->inode, !ofs); 1342 set_sbi_flag(sbi, SBI_NEED_FSCK); 1343 f2fs_warn_ratelimited(sbi, 1344 "f2fs_new_node_page: inconsistent nat entry, " 1345 "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u", 1346 new_ni.ino, new_ni.nid, new_ni.blk_addr, 1347 new_ni.version, new_ni.flag); 1348 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT); 1349 goto fail; 1350 } 1351 #endif 1352 new_ni.nid = dn->nid; 1353 new_ni.ino = dn->inode->i_ino; 1354 new_ni.blk_addr = NULL_ADDR; 1355 new_ni.flag = 0; 1356 new_ni.version = 0; 1357 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 1358 1359 f2fs_wait_on_page_writeback(page, NODE, true, true); 1360 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 1361 set_cold_node(page, S_ISDIR(dn->inode->i_mode)); 1362 if (!PageUptodate(page)) 1363 SetPageUptodate(page); 1364 if (set_page_dirty(page)) 1365 dn->node_changed = true; 1366 1367 if (f2fs_has_xattr_block(ofs)) 1368 f2fs_i_xnid_write(dn->inode, dn->nid); 1369 1370 if (ofs == 0) 1371 inc_valid_inode_count(sbi); 1372 return page; 1373 fail: 1374 clear_node_page_dirty(page); 1375 f2fs_put_page(page, 1); 1376 return ERR_PTR(err); 1377 } 1378 1379 /* 1380 * Caller should do after getting the following values. 1381 * 0: f2fs_put_page(page, 0) 1382 * LOCKED_PAGE or error: f2fs_put_page(page, 1) 1383 */ 1384 static int read_node_page(struct page *page, blk_opf_t op_flags) 1385 { 1386 struct folio *folio = page_folio(page); 1387 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1388 struct node_info ni; 1389 struct f2fs_io_info fio = { 1390 .sbi = sbi, 1391 .type = NODE, 1392 .op = REQ_OP_READ, 1393 .op_flags = op_flags, 1394 .page = page, 1395 .encrypted_page = NULL, 1396 }; 1397 int err; 1398 1399 if (folio_test_uptodate(folio)) { 1400 if (!f2fs_inode_chksum_verify(sbi, page)) { 1401 folio_clear_uptodate(folio); 1402 return -EFSBADCRC; 1403 } 1404 return LOCKED_PAGE; 1405 } 1406 1407 err = f2fs_get_node_info(sbi, folio->index, &ni, false); 1408 if (err) 1409 return err; 1410 1411 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */ 1412 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) { 1413 folio_clear_uptodate(folio); 1414 return -ENOENT; 1415 } 1416 1417 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr; 1418 1419 err = f2fs_submit_page_bio(&fio); 1420 1421 if (!err) 1422 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE); 1423 1424 return err; 1425 } 1426 1427 /* 1428 * Readahead a node page 1429 */ 1430 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 1431 { 1432 struct page *apage; 1433 int err; 1434 1435 if (!nid) 1436 return; 1437 if (f2fs_check_nid_range(sbi, nid)) 1438 return; 1439 1440 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid); 1441 if (apage) 1442 return; 1443 1444 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1445 if (!apage) 1446 return; 1447 1448 err = read_node_page(apage, REQ_RAHEAD); 1449 f2fs_put_page(apage, err ? 1 : 0); 1450 } 1451 1452 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid, 1453 struct page *parent, int start) 1454 { 1455 struct page *page; 1456 int err; 1457 1458 if (!nid) 1459 return ERR_PTR(-ENOENT); 1460 if (f2fs_check_nid_range(sbi, nid)) 1461 return ERR_PTR(-EINVAL); 1462 repeat: 1463 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1464 if (!page) 1465 return ERR_PTR(-ENOMEM); 1466 1467 err = read_node_page(page, 0); 1468 if (err < 0) { 1469 goto out_put_err; 1470 } else if (err == LOCKED_PAGE) { 1471 err = 0; 1472 goto page_hit; 1473 } 1474 1475 if (parent) 1476 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE); 1477 1478 lock_page(page); 1479 1480 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1481 f2fs_put_page(page, 1); 1482 goto repeat; 1483 } 1484 1485 if (unlikely(!PageUptodate(page))) { 1486 err = -EIO; 1487 goto out_err; 1488 } 1489 1490 if (!f2fs_inode_chksum_verify(sbi, page)) { 1491 err = -EFSBADCRC; 1492 goto out_err; 1493 } 1494 page_hit: 1495 if (likely(nid == nid_of_node(page))) 1496 return page; 1497 1498 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]", 1499 nid, nid_of_node(page), ino_of_node(page), 1500 ofs_of_node(page), cpver_of_node(page), 1501 next_blkaddr_of_node(page)); 1502 set_sbi_flag(sbi, SBI_NEED_FSCK); 1503 f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER); 1504 err = -EFSCORRUPTED; 1505 out_err: 1506 ClearPageUptodate(page); 1507 out_put_err: 1508 /* ENOENT comes from read_node_page which is not an error. */ 1509 if (err != -ENOENT) 1510 f2fs_handle_page_eio(sbi, page_folio(page), NODE); 1511 f2fs_put_page(page, 1); 1512 return ERR_PTR(err); 1513 } 1514 1515 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 1516 { 1517 return __get_node_page(sbi, nid, NULL, 0); 1518 } 1519 1520 struct page *f2fs_get_node_page_ra(struct page *parent, int start) 1521 { 1522 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 1523 nid_t nid = get_nid(parent, start, false); 1524 1525 return __get_node_page(sbi, nid, parent, start); 1526 } 1527 1528 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino) 1529 { 1530 struct inode *inode; 1531 struct page *page; 1532 int ret; 1533 1534 /* should flush inline_data before evict_inode */ 1535 inode = ilookup(sbi->sb, ino); 1536 if (!inode) 1537 return; 1538 1539 page = f2fs_pagecache_get_page(inode->i_mapping, 0, 1540 FGP_LOCK|FGP_NOWAIT, 0); 1541 if (!page) 1542 goto iput_out; 1543 1544 if (!PageUptodate(page)) 1545 goto page_out; 1546 1547 if (!PageDirty(page)) 1548 goto page_out; 1549 1550 if (!clear_page_dirty_for_io(page)) 1551 goto page_out; 1552 1553 ret = f2fs_write_inline_data(inode, page_folio(page)); 1554 inode_dec_dirty_pages(inode); 1555 f2fs_remove_dirty_inode(inode); 1556 if (ret) 1557 set_page_dirty(page); 1558 page_out: 1559 f2fs_put_page(page, 1); 1560 iput_out: 1561 iput(inode); 1562 } 1563 1564 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino) 1565 { 1566 pgoff_t index; 1567 struct folio_batch fbatch; 1568 struct page *last_page = NULL; 1569 int nr_folios; 1570 1571 folio_batch_init(&fbatch); 1572 index = 0; 1573 1574 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1575 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1576 &fbatch))) { 1577 int i; 1578 1579 for (i = 0; i < nr_folios; i++) { 1580 struct page *page = &fbatch.folios[i]->page; 1581 1582 if (unlikely(f2fs_cp_error(sbi))) { 1583 f2fs_put_page(last_page, 0); 1584 folio_batch_release(&fbatch); 1585 return ERR_PTR(-EIO); 1586 } 1587 1588 if (!IS_DNODE(page) || !is_cold_node(page)) 1589 continue; 1590 if (ino_of_node(page) != ino) 1591 continue; 1592 1593 lock_page(page); 1594 1595 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1596 continue_unlock: 1597 unlock_page(page); 1598 continue; 1599 } 1600 if (ino_of_node(page) != ino) 1601 goto continue_unlock; 1602 1603 if (!PageDirty(page)) { 1604 /* someone wrote it for us */ 1605 goto continue_unlock; 1606 } 1607 1608 if (last_page) 1609 f2fs_put_page(last_page, 0); 1610 1611 get_page(page); 1612 last_page = page; 1613 unlock_page(page); 1614 } 1615 folio_batch_release(&fbatch); 1616 cond_resched(); 1617 } 1618 return last_page; 1619 } 1620 1621 static int __write_node_page(struct page *page, bool atomic, bool *submitted, 1622 struct writeback_control *wbc, bool do_balance, 1623 enum iostat_type io_type, unsigned int *seq_id) 1624 { 1625 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1626 struct folio *folio = page_folio(page); 1627 nid_t nid; 1628 struct node_info ni; 1629 struct f2fs_io_info fio = { 1630 .sbi = sbi, 1631 .ino = ino_of_node(page), 1632 .type = NODE, 1633 .op = REQ_OP_WRITE, 1634 .op_flags = wbc_to_write_flags(wbc), 1635 .page = page, 1636 .encrypted_page = NULL, 1637 .submitted = 0, 1638 .io_type = io_type, 1639 .io_wbc = wbc, 1640 }; 1641 unsigned int seq; 1642 1643 trace_f2fs_writepage(folio, NODE); 1644 1645 if (unlikely(f2fs_cp_error(sbi))) { 1646 /* keep node pages in remount-ro mode */ 1647 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY) 1648 goto redirty_out; 1649 folio_clear_uptodate(folio); 1650 dec_page_count(sbi, F2FS_DIRTY_NODES); 1651 folio_unlock(folio); 1652 return 0; 1653 } 1654 1655 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1656 goto redirty_out; 1657 1658 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 1659 wbc->sync_mode == WB_SYNC_NONE && 1660 IS_DNODE(page) && is_cold_node(page)) 1661 goto redirty_out; 1662 1663 /* get old block addr of this node page */ 1664 nid = nid_of_node(page); 1665 f2fs_bug_on(sbi, folio->index != nid); 1666 1667 if (f2fs_get_node_info(sbi, nid, &ni, !do_balance)) 1668 goto redirty_out; 1669 1670 if (wbc->for_reclaim) { 1671 if (!f2fs_down_read_trylock(&sbi->node_write)) 1672 goto redirty_out; 1673 } else { 1674 f2fs_down_read(&sbi->node_write); 1675 } 1676 1677 /* This page is already truncated */ 1678 if (unlikely(ni.blk_addr == NULL_ADDR)) { 1679 folio_clear_uptodate(folio); 1680 dec_page_count(sbi, F2FS_DIRTY_NODES); 1681 f2fs_up_read(&sbi->node_write); 1682 folio_unlock(folio); 1683 return 0; 1684 } 1685 1686 if (__is_valid_data_blkaddr(ni.blk_addr) && 1687 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, 1688 DATA_GENERIC_ENHANCE)) { 1689 f2fs_up_read(&sbi->node_write); 1690 goto redirty_out; 1691 } 1692 1693 if (atomic && !test_opt(sbi, NOBARRIER)) 1694 fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 1695 1696 /* should add to global list before clearing PAGECACHE status */ 1697 if (f2fs_in_warm_node_list(sbi, page)) { 1698 seq = f2fs_add_fsync_node_entry(sbi, page); 1699 if (seq_id) 1700 *seq_id = seq; 1701 } 1702 1703 folio_start_writeback(folio); 1704 1705 fio.old_blkaddr = ni.blk_addr; 1706 f2fs_do_write_node_page(nid, &fio); 1707 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page)); 1708 dec_page_count(sbi, F2FS_DIRTY_NODES); 1709 f2fs_up_read(&sbi->node_write); 1710 1711 if (wbc->for_reclaim) { 1712 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE); 1713 submitted = NULL; 1714 } 1715 1716 folio_unlock(folio); 1717 1718 if (unlikely(f2fs_cp_error(sbi))) { 1719 f2fs_submit_merged_write(sbi, NODE); 1720 submitted = NULL; 1721 } 1722 if (submitted) 1723 *submitted = fio.submitted; 1724 1725 if (do_balance) 1726 f2fs_balance_fs(sbi, false); 1727 return 0; 1728 1729 redirty_out: 1730 folio_redirty_for_writepage(wbc, folio); 1731 return AOP_WRITEPAGE_ACTIVATE; 1732 } 1733 1734 int f2fs_move_node_page(struct page *node_page, int gc_type) 1735 { 1736 int err = 0; 1737 1738 if (gc_type == FG_GC) { 1739 struct writeback_control wbc = { 1740 .sync_mode = WB_SYNC_ALL, 1741 .nr_to_write = 1, 1742 .for_reclaim = 0, 1743 }; 1744 1745 f2fs_wait_on_page_writeback(node_page, NODE, true, true); 1746 1747 set_page_dirty(node_page); 1748 1749 if (!clear_page_dirty_for_io(node_page)) { 1750 err = -EAGAIN; 1751 goto out_page; 1752 } 1753 1754 if (__write_node_page(node_page, false, NULL, 1755 &wbc, false, FS_GC_NODE_IO, NULL)) { 1756 err = -EAGAIN; 1757 unlock_page(node_page); 1758 } 1759 goto release_page; 1760 } else { 1761 /* set page dirty and write it */ 1762 if (!folio_test_writeback(page_folio(node_page))) 1763 set_page_dirty(node_page); 1764 } 1765 out_page: 1766 unlock_page(node_page); 1767 release_page: 1768 f2fs_put_page(node_page, 0); 1769 return err; 1770 } 1771 1772 static int f2fs_write_node_page(struct page *page, 1773 struct writeback_control *wbc) 1774 { 1775 return __write_node_page(page, false, NULL, wbc, false, 1776 FS_NODE_IO, NULL); 1777 } 1778 1779 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode, 1780 struct writeback_control *wbc, bool atomic, 1781 unsigned int *seq_id) 1782 { 1783 pgoff_t index; 1784 struct folio_batch fbatch; 1785 int ret = 0; 1786 struct page *last_page = NULL; 1787 bool marked = false; 1788 nid_t ino = inode->i_ino; 1789 int nr_folios; 1790 int nwritten = 0; 1791 1792 if (atomic) { 1793 last_page = last_fsync_dnode(sbi, ino); 1794 if (IS_ERR_OR_NULL(last_page)) 1795 return PTR_ERR_OR_ZERO(last_page); 1796 } 1797 retry: 1798 folio_batch_init(&fbatch); 1799 index = 0; 1800 1801 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1802 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1803 &fbatch))) { 1804 int i; 1805 1806 for (i = 0; i < nr_folios; i++) { 1807 struct page *page = &fbatch.folios[i]->page; 1808 bool submitted = false; 1809 1810 if (unlikely(f2fs_cp_error(sbi))) { 1811 f2fs_put_page(last_page, 0); 1812 folio_batch_release(&fbatch); 1813 ret = -EIO; 1814 goto out; 1815 } 1816 1817 if (!IS_DNODE(page) || !is_cold_node(page)) 1818 continue; 1819 if (ino_of_node(page) != ino) 1820 continue; 1821 1822 lock_page(page); 1823 1824 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1825 continue_unlock: 1826 unlock_page(page); 1827 continue; 1828 } 1829 if (ino_of_node(page) != ino) 1830 goto continue_unlock; 1831 1832 if (!PageDirty(page) && page != last_page) { 1833 /* someone wrote it for us */ 1834 goto continue_unlock; 1835 } 1836 1837 f2fs_wait_on_page_writeback(page, NODE, true, true); 1838 1839 set_fsync_mark(page, 0); 1840 set_dentry_mark(page, 0); 1841 1842 if (!atomic || page == last_page) { 1843 set_fsync_mark(page, 1); 1844 percpu_counter_inc(&sbi->rf_node_block_count); 1845 if (IS_INODE(page)) { 1846 if (is_inode_flag_set(inode, 1847 FI_DIRTY_INODE)) 1848 f2fs_update_inode(inode, page); 1849 set_dentry_mark(page, 1850 f2fs_need_dentry_mark(sbi, ino)); 1851 } 1852 /* may be written by other thread */ 1853 if (!PageDirty(page)) 1854 set_page_dirty(page); 1855 } 1856 1857 if (!clear_page_dirty_for_io(page)) 1858 goto continue_unlock; 1859 1860 ret = __write_node_page(page, atomic && 1861 page == last_page, 1862 &submitted, wbc, true, 1863 FS_NODE_IO, seq_id); 1864 if (ret) { 1865 unlock_page(page); 1866 f2fs_put_page(last_page, 0); 1867 break; 1868 } else if (submitted) { 1869 nwritten++; 1870 } 1871 1872 if (page == last_page) { 1873 f2fs_put_page(page, 0); 1874 marked = true; 1875 break; 1876 } 1877 } 1878 folio_batch_release(&fbatch); 1879 cond_resched(); 1880 1881 if (ret || marked) 1882 break; 1883 } 1884 if (!ret && atomic && !marked) { 1885 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx", 1886 ino, page_folio(last_page)->index); 1887 lock_page(last_page); 1888 f2fs_wait_on_page_writeback(last_page, NODE, true, true); 1889 set_page_dirty(last_page); 1890 unlock_page(last_page); 1891 goto retry; 1892 } 1893 out: 1894 if (nwritten) 1895 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE); 1896 return ret ? -EIO : 0; 1897 } 1898 1899 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data) 1900 { 1901 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1902 bool clean; 1903 1904 if (inode->i_ino != ino) 1905 return 0; 1906 1907 if (!is_inode_flag_set(inode, FI_DIRTY_INODE)) 1908 return 0; 1909 1910 spin_lock(&sbi->inode_lock[DIRTY_META]); 1911 clean = list_empty(&F2FS_I(inode)->gdirty_list); 1912 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1913 1914 if (clean) 1915 return 0; 1916 1917 inode = igrab(inode); 1918 if (!inode) 1919 return 0; 1920 return 1; 1921 } 1922 1923 static bool flush_dirty_inode(struct page *page) 1924 { 1925 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1926 struct inode *inode; 1927 nid_t ino = ino_of_node(page); 1928 1929 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL); 1930 if (!inode) 1931 return false; 1932 1933 f2fs_update_inode(inode, page); 1934 unlock_page(page); 1935 1936 iput(inode); 1937 return true; 1938 } 1939 1940 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi) 1941 { 1942 pgoff_t index = 0; 1943 struct folio_batch fbatch; 1944 int nr_folios; 1945 1946 folio_batch_init(&fbatch); 1947 1948 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1949 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1950 &fbatch))) { 1951 int i; 1952 1953 for (i = 0; i < nr_folios; i++) { 1954 struct page *page = &fbatch.folios[i]->page; 1955 1956 if (!IS_INODE(page)) 1957 continue; 1958 1959 lock_page(page); 1960 1961 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1962 continue_unlock: 1963 unlock_page(page); 1964 continue; 1965 } 1966 1967 if (!PageDirty(page)) { 1968 /* someone wrote it for us */ 1969 goto continue_unlock; 1970 } 1971 1972 /* flush inline_data, if it's async context. */ 1973 if (page_private_inline(page)) { 1974 clear_page_private_inline(page); 1975 unlock_page(page); 1976 flush_inline_data(sbi, ino_of_node(page)); 1977 continue; 1978 } 1979 unlock_page(page); 1980 } 1981 folio_batch_release(&fbatch); 1982 cond_resched(); 1983 } 1984 } 1985 1986 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi, 1987 struct writeback_control *wbc, 1988 bool do_balance, enum iostat_type io_type) 1989 { 1990 pgoff_t index; 1991 struct folio_batch fbatch; 1992 int step = 0; 1993 int nwritten = 0; 1994 int ret = 0; 1995 int nr_folios, done = 0; 1996 1997 folio_batch_init(&fbatch); 1998 1999 next_step: 2000 index = 0; 2001 2002 while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), 2003 &index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 2004 &fbatch))) { 2005 int i; 2006 2007 for (i = 0; i < nr_folios; i++) { 2008 struct page *page = &fbatch.folios[i]->page; 2009 bool submitted = false; 2010 2011 /* give a priority to WB_SYNC threads */ 2012 if (atomic_read(&sbi->wb_sync_req[NODE]) && 2013 wbc->sync_mode == WB_SYNC_NONE) { 2014 done = 1; 2015 break; 2016 } 2017 2018 /* 2019 * flushing sequence with step: 2020 * 0. indirect nodes 2021 * 1. dentry dnodes 2022 * 2. file dnodes 2023 */ 2024 if (step == 0 && IS_DNODE(page)) 2025 continue; 2026 if (step == 1 && (!IS_DNODE(page) || 2027 is_cold_node(page))) 2028 continue; 2029 if (step == 2 && (!IS_DNODE(page) || 2030 !is_cold_node(page))) 2031 continue; 2032 lock_node: 2033 if (wbc->sync_mode == WB_SYNC_ALL) 2034 lock_page(page); 2035 else if (!trylock_page(page)) 2036 continue; 2037 2038 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 2039 continue_unlock: 2040 unlock_page(page); 2041 continue; 2042 } 2043 2044 if (!PageDirty(page)) { 2045 /* someone wrote it for us */ 2046 goto continue_unlock; 2047 } 2048 2049 /* flush inline_data/inode, if it's async context. */ 2050 if (!do_balance) 2051 goto write_node; 2052 2053 /* flush inline_data */ 2054 if (page_private_inline(page)) { 2055 clear_page_private_inline(page); 2056 unlock_page(page); 2057 flush_inline_data(sbi, ino_of_node(page)); 2058 goto lock_node; 2059 } 2060 2061 /* flush dirty inode */ 2062 if (IS_INODE(page) && flush_dirty_inode(page)) 2063 goto lock_node; 2064 write_node: 2065 f2fs_wait_on_page_writeback(page, NODE, true, true); 2066 2067 if (!clear_page_dirty_for_io(page)) 2068 goto continue_unlock; 2069 2070 set_fsync_mark(page, 0); 2071 set_dentry_mark(page, 0); 2072 2073 ret = __write_node_page(page, false, &submitted, 2074 wbc, do_balance, io_type, NULL); 2075 if (ret) 2076 unlock_page(page); 2077 else if (submitted) 2078 nwritten++; 2079 2080 if (--wbc->nr_to_write == 0) 2081 break; 2082 } 2083 folio_batch_release(&fbatch); 2084 cond_resched(); 2085 2086 if (wbc->nr_to_write == 0) { 2087 step = 2; 2088 break; 2089 } 2090 } 2091 2092 if (step < 2) { 2093 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2094 wbc->sync_mode == WB_SYNC_NONE && step == 1) 2095 goto out; 2096 step++; 2097 goto next_step; 2098 } 2099 out: 2100 if (nwritten) 2101 f2fs_submit_merged_write(sbi, NODE); 2102 2103 if (unlikely(f2fs_cp_error(sbi))) 2104 return -EIO; 2105 return ret; 2106 } 2107 2108 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, 2109 unsigned int seq_id) 2110 { 2111 struct fsync_node_entry *fn; 2112 struct page *page; 2113 struct list_head *head = &sbi->fsync_node_list; 2114 unsigned long flags; 2115 unsigned int cur_seq_id = 0; 2116 2117 while (seq_id && cur_seq_id < seq_id) { 2118 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 2119 if (list_empty(head)) { 2120 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2121 break; 2122 } 2123 fn = list_first_entry(head, struct fsync_node_entry, list); 2124 if (fn->seq_id > seq_id) { 2125 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2126 break; 2127 } 2128 cur_seq_id = fn->seq_id; 2129 page = fn->page; 2130 get_page(page); 2131 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2132 2133 f2fs_wait_on_page_writeback(page, NODE, true, false); 2134 2135 put_page(page); 2136 } 2137 2138 return filemap_check_errors(NODE_MAPPING(sbi)); 2139 } 2140 2141 static int f2fs_write_node_pages(struct address_space *mapping, 2142 struct writeback_control *wbc) 2143 { 2144 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2145 struct blk_plug plug; 2146 long diff; 2147 2148 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2149 goto skip_write; 2150 2151 /* balancing f2fs's metadata in background */ 2152 f2fs_balance_fs_bg(sbi, true); 2153 2154 /* collect a number of dirty node pages and write together */ 2155 if (wbc->sync_mode != WB_SYNC_ALL && 2156 get_pages(sbi, F2FS_DIRTY_NODES) < 2157 nr_pages_to_skip(sbi, NODE)) 2158 goto skip_write; 2159 2160 if (wbc->sync_mode == WB_SYNC_ALL) 2161 atomic_inc(&sbi->wb_sync_req[NODE]); 2162 else if (atomic_read(&sbi->wb_sync_req[NODE])) { 2163 /* to avoid potential deadlock */ 2164 if (current->plug) 2165 blk_finish_plug(current->plug); 2166 goto skip_write; 2167 } 2168 2169 trace_f2fs_writepages(mapping->host, wbc, NODE); 2170 2171 diff = nr_pages_to_write(sbi, NODE, wbc); 2172 blk_start_plug(&plug); 2173 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO); 2174 blk_finish_plug(&plug); 2175 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); 2176 2177 if (wbc->sync_mode == WB_SYNC_ALL) 2178 atomic_dec(&sbi->wb_sync_req[NODE]); 2179 return 0; 2180 2181 skip_write: 2182 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES); 2183 trace_f2fs_writepages(mapping->host, wbc, NODE); 2184 return 0; 2185 } 2186 2187 static bool f2fs_dirty_node_folio(struct address_space *mapping, 2188 struct folio *folio) 2189 { 2190 trace_f2fs_set_page_dirty(folio, NODE); 2191 2192 if (!folio_test_uptodate(folio)) 2193 folio_mark_uptodate(folio); 2194 #ifdef CONFIG_F2FS_CHECK_FS 2195 if (IS_INODE(&folio->page)) 2196 f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page); 2197 #endif 2198 if (filemap_dirty_folio(mapping, folio)) { 2199 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES); 2200 set_page_private_reference(&folio->page); 2201 return true; 2202 } 2203 return false; 2204 } 2205 2206 /* 2207 * Structure of the f2fs node operations 2208 */ 2209 const struct address_space_operations f2fs_node_aops = { 2210 .writepage = f2fs_write_node_page, 2211 .writepages = f2fs_write_node_pages, 2212 .dirty_folio = f2fs_dirty_node_folio, 2213 .invalidate_folio = f2fs_invalidate_folio, 2214 .release_folio = f2fs_release_folio, 2215 .migrate_folio = filemap_migrate_folio, 2216 }; 2217 2218 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i, 2219 nid_t n) 2220 { 2221 return radix_tree_lookup(&nm_i->free_nid_root, n); 2222 } 2223 2224 static int __insert_free_nid(struct f2fs_sb_info *sbi, 2225 struct free_nid *i) 2226 { 2227 struct f2fs_nm_info *nm_i = NM_I(sbi); 2228 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i); 2229 2230 if (err) 2231 return err; 2232 2233 nm_i->nid_cnt[FREE_NID]++; 2234 list_add_tail(&i->list, &nm_i->free_nid_list); 2235 return 0; 2236 } 2237 2238 static void __remove_free_nid(struct f2fs_sb_info *sbi, 2239 struct free_nid *i, enum nid_state state) 2240 { 2241 struct f2fs_nm_info *nm_i = NM_I(sbi); 2242 2243 f2fs_bug_on(sbi, state != i->state); 2244 nm_i->nid_cnt[state]--; 2245 if (state == FREE_NID) 2246 list_del(&i->list); 2247 radix_tree_delete(&nm_i->free_nid_root, i->nid); 2248 } 2249 2250 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i, 2251 enum nid_state org_state, enum nid_state dst_state) 2252 { 2253 struct f2fs_nm_info *nm_i = NM_I(sbi); 2254 2255 f2fs_bug_on(sbi, org_state != i->state); 2256 i->state = dst_state; 2257 nm_i->nid_cnt[org_state]--; 2258 nm_i->nid_cnt[dst_state]++; 2259 2260 switch (dst_state) { 2261 case PREALLOC_NID: 2262 list_del(&i->list); 2263 break; 2264 case FREE_NID: 2265 list_add_tail(&i->list, &nm_i->free_nid_list); 2266 break; 2267 default: 2268 BUG_ON(1); 2269 } 2270 } 2271 2272 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi) 2273 { 2274 struct f2fs_nm_info *nm_i = NM_I(sbi); 2275 unsigned int i; 2276 bool ret = true; 2277 2278 f2fs_down_read(&nm_i->nat_tree_lock); 2279 for (i = 0; i < nm_i->nat_blocks; i++) { 2280 if (!test_bit_le(i, nm_i->nat_block_bitmap)) { 2281 ret = false; 2282 break; 2283 } 2284 } 2285 f2fs_up_read(&nm_i->nat_tree_lock); 2286 2287 return ret; 2288 } 2289 2290 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid, 2291 bool set, bool build) 2292 { 2293 struct f2fs_nm_info *nm_i = NM_I(sbi); 2294 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid); 2295 unsigned int nid_ofs = nid - START_NID(nid); 2296 2297 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap)) 2298 return; 2299 2300 if (set) { 2301 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2302 return; 2303 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2304 nm_i->free_nid_count[nat_ofs]++; 2305 } else { 2306 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2307 return; 2308 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2309 if (!build) 2310 nm_i->free_nid_count[nat_ofs]--; 2311 } 2312 } 2313 2314 /* return if the nid is recognized as free */ 2315 static bool add_free_nid(struct f2fs_sb_info *sbi, 2316 nid_t nid, bool build, bool update) 2317 { 2318 struct f2fs_nm_info *nm_i = NM_I(sbi); 2319 struct free_nid *i, *e; 2320 struct nat_entry *ne; 2321 int err = -EINVAL; 2322 bool ret = false; 2323 2324 /* 0 nid should not be used */ 2325 if (unlikely(nid == 0)) 2326 return false; 2327 2328 if (unlikely(f2fs_check_nid_range(sbi, nid))) 2329 return false; 2330 2331 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL); 2332 i->nid = nid; 2333 i->state = FREE_NID; 2334 2335 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 2336 2337 spin_lock(&nm_i->nid_list_lock); 2338 2339 if (build) { 2340 /* 2341 * Thread A Thread B 2342 * - f2fs_create 2343 * - f2fs_new_inode 2344 * - f2fs_alloc_nid 2345 * - __insert_nid_to_list(PREALLOC_NID) 2346 * - f2fs_balance_fs_bg 2347 * - f2fs_build_free_nids 2348 * - __f2fs_build_free_nids 2349 * - scan_nat_page 2350 * - add_free_nid 2351 * - __lookup_nat_cache 2352 * - f2fs_add_link 2353 * - f2fs_init_inode_metadata 2354 * - f2fs_new_inode_page 2355 * - f2fs_new_node_page 2356 * - set_node_addr 2357 * - f2fs_alloc_nid_done 2358 * - __remove_nid_from_list(PREALLOC_NID) 2359 * - __insert_nid_to_list(FREE_NID) 2360 */ 2361 ne = __lookup_nat_cache(nm_i, nid); 2362 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) || 2363 nat_get_blkaddr(ne) != NULL_ADDR)) 2364 goto err_out; 2365 2366 e = __lookup_free_nid_list(nm_i, nid); 2367 if (e) { 2368 if (e->state == FREE_NID) 2369 ret = true; 2370 goto err_out; 2371 } 2372 } 2373 ret = true; 2374 err = __insert_free_nid(sbi, i); 2375 err_out: 2376 if (update) { 2377 update_free_nid_bitmap(sbi, nid, ret, build); 2378 if (!build) 2379 nm_i->available_nids++; 2380 } 2381 spin_unlock(&nm_i->nid_list_lock); 2382 radix_tree_preload_end(); 2383 2384 if (err) 2385 kmem_cache_free(free_nid_slab, i); 2386 return ret; 2387 } 2388 2389 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid) 2390 { 2391 struct f2fs_nm_info *nm_i = NM_I(sbi); 2392 struct free_nid *i; 2393 bool need_free = false; 2394 2395 spin_lock(&nm_i->nid_list_lock); 2396 i = __lookup_free_nid_list(nm_i, nid); 2397 if (i && i->state == FREE_NID) { 2398 __remove_free_nid(sbi, i, FREE_NID); 2399 need_free = true; 2400 } 2401 spin_unlock(&nm_i->nid_list_lock); 2402 2403 if (need_free) 2404 kmem_cache_free(free_nid_slab, i); 2405 } 2406 2407 static int scan_nat_page(struct f2fs_sb_info *sbi, 2408 struct page *nat_page, nid_t start_nid) 2409 { 2410 struct f2fs_nm_info *nm_i = NM_I(sbi); 2411 struct f2fs_nat_block *nat_blk = page_address(nat_page); 2412 block_t blk_addr; 2413 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid); 2414 int i; 2415 2416 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap); 2417 2418 i = start_nid % NAT_ENTRY_PER_BLOCK; 2419 2420 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) { 2421 if (unlikely(start_nid >= nm_i->max_nid)) 2422 break; 2423 2424 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr); 2425 2426 if (blk_addr == NEW_ADDR) 2427 return -EFSCORRUPTED; 2428 2429 if (blk_addr == NULL_ADDR) { 2430 add_free_nid(sbi, start_nid, true, true); 2431 } else { 2432 spin_lock(&NM_I(sbi)->nid_list_lock); 2433 update_free_nid_bitmap(sbi, start_nid, false, true); 2434 spin_unlock(&NM_I(sbi)->nid_list_lock); 2435 } 2436 } 2437 2438 return 0; 2439 } 2440 2441 static void scan_curseg_cache(struct f2fs_sb_info *sbi) 2442 { 2443 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2444 struct f2fs_journal *journal = curseg->journal; 2445 int i; 2446 2447 down_read(&curseg->journal_rwsem); 2448 for (i = 0; i < nats_in_cursum(journal); i++) { 2449 block_t addr; 2450 nid_t nid; 2451 2452 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr); 2453 nid = le32_to_cpu(nid_in_journal(journal, i)); 2454 if (addr == NULL_ADDR) 2455 add_free_nid(sbi, nid, true, false); 2456 else 2457 remove_free_nid(sbi, nid); 2458 } 2459 up_read(&curseg->journal_rwsem); 2460 } 2461 2462 static void scan_free_nid_bits(struct f2fs_sb_info *sbi) 2463 { 2464 struct f2fs_nm_info *nm_i = NM_I(sbi); 2465 unsigned int i, idx; 2466 nid_t nid; 2467 2468 f2fs_down_read(&nm_i->nat_tree_lock); 2469 2470 for (i = 0; i < nm_i->nat_blocks; i++) { 2471 if (!test_bit_le(i, nm_i->nat_block_bitmap)) 2472 continue; 2473 if (!nm_i->free_nid_count[i]) 2474 continue; 2475 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) { 2476 idx = find_next_bit_le(nm_i->free_nid_bitmap[i], 2477 NAT_ENTRY_PER_BLOCK, idx); 2478 if (idx >= NAT_ENTRY_PER_BLOCK) 2479 break; 2480 2481 nid = i * NAT_ENTRY_PER_BLOCK + idx; 2482 add_free_nid(sbi, nid, true, false); 2483 2484 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS) 2485 goto out; 2486 } 2487 } 2488 out: 2489 scan_curseg_cache(sbi); 2490 2491 f2fs_up_read(&nm_i->nat_tree_lock); 2492 } 2493 2494 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi, 2495 bool sync, bool mount) 2496 { 2497 struct f2fs_nm_info *nm_i = NM_I(sbi); 2498 int i = 0, ret; 2499 nid_t nid = nm_i->next_scan_nid; 2500 2501 if (unlikely(nid >= nm_i->max_nid)) 2502 nid = 0; 2503 2504 if (unlikely(nid % NAT_ENTRY_PER_BLOCK)) 2505 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK; 2506 2507 /* Enough entries */ 2508 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2509 return 0; 2510 2511 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS)) 2512 return 0; 2513 2514 if (!mount) { 2515 /* try to find free nids in free_nid_bitmap */ 2516 scan_free_nid_bits(sbi); 2517 2518 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2519 return 0; 2520 } 2521 2522 /* readahead nat pages to be scanned */ 2523 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, 2524 META_NAT, true); 2525 2526 f2fs_down_read(&nm_i->nat_tree_lock); 2527 2528 while (1) { 2529 if (!test_bit_le(NAT_BLOCK_OFFSET(nid), 2530 nm_i->nat_block_bitmap)) { 2531 struct page *page = get_current_nat_page(sbi, nid); 2532 2533 if (IS_ERR(page)) { 2534 ret = PTR_ERR(page); 2535 } else { 2536 ret = scan_nat_page(sbi, page, nid); 2537 f2fs_put_page(page, 1); 2538 } 2539 2540 if (ret) { 2541 f2fs_up_read(&nm_i->nat_tree_lock); 2542 2543 if (ret == -EFSCORRUPTED) { 2544 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it"); 2545 set_sbi_flag(sbi, SBI_NEED_FSCK); 2546 f2fs_handle_error(sbi, 2547 ERROR_INCONSISTENT_NAT); 2548 } 2549 2550 return ret; 2551 } 2552 } 2553 2554 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK)); 2555 if (unlikely(nid >= nm_i->max_nid)) 2556 nid = 0; 2557 2558 if (++i >= FREE_NID_PAGES) 2559 break; 2560 } 2561 2562 /* go to the next free nat pages to find free nids abundantly */ 2563 nm_i->next_scan_nid = nid; 2564 2565 /* find free nids from current sum_pages */ 2566 scan_curseg_cache(sbi); 2567 2568 f2fs_up_read(&nm_i->nat_tree_lock); 2569 2570 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid), 2571 nm_i->ra_nid_pages, META_NAT, false); 2572 2573 return 0; 2574 } 2575 2576 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount) 2577 { 2578 int ret; 2579 2580 mutex_lock(&NM_I(sbi)->build_lock); 2581 ret = __f2fs_build_free_nids(sbi, sync, mount); 2582 mutex_unlock(&NM_I(sbi)->build_lock); 2583 2584 return ret; 2585 } 2586 2587 /* 2588 * If this function returns success, caller can obtain a new nid 2589 * from second parameter of this function. 2590 * The returned nid could be used ino as well as nid when inode is created. 2591 */ 2592 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid) 2593 { 2594 struct f2fs_nm_info *nm_i = NM_I(sbi); 2595 struct free_nid *i = NULL; 2596 retry: 2597 if (time_to_inject(sbi, FAULT_ALLOC_NID)) 2598 return false; 2599 2600 spin_lock(&nm_i->nid_list_lock); 2601 2602 if (unlikely(nm_i->available_nids == 0)) { 2603 spin_unlock(&nm_i->nid_list_lock); 2604 return false; 2605 } 2606 2607 /* We should not use stale free nids created by f2fs_build_free_nids */ 2608 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) { 2609 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list)); 2610 i = list_first_entry(&nm_i->free_nid_list, 2611 struct free_nid, list); 2612 *nid = i->nid; 2613 2614 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID); 2615 nm_i->available_nids--; 2616 2617 update_free_nid_bitmap(sbi, *nid, false, false); 2618 2619 spin_unlock(&nm_i->nid_list_lock); 2620 return true; 2621 } 2622 spin_unlock(&nm_i->nid_list_lock); 2623 2624 /* Let's scan nat pages and its caches to get free nids */ 2625 if (!f2fs_build_free_nids(sbi, true, false)) 2626 goto retry; 2627 return false; 2628 } 2629 2630 /* 2631 * f2fs_alloc_nid() should be called prior to this function. 2632 */ 2633 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 2634 { 2635 struct f2fs_nm_info *nm_i = NM_I(sbi); 2636 struct free_nid *i; 2637 2638 spin_lock(&nm_i->nid_list_lock); 2639 i = __lookup_free_nid_list(nm_i, nid); 2640 f2fs_bug_on(sbi, !i); 2641 __remove_free_nid(sbi, i, PREALLOC_NID); 2642 spin_unlock(&nm_i->nid_list_lock); 2643 2644 kmem_cache_free(free_nid_slab, i); 2645 } 2646 2647 /* 2648 * f2fs_alloc_nid() should be called prior to this function. 2649 */ 2650 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 2651 { 2652 struct f2fs_nm_info *nm_i = NM_I(sbi); 2653 struct free_nid *i; 2654 bool need_free = false; 2655 2656 if (!nid) 2657 return; 2658 2659 spin_lock(&nm_i->nid_list_lock); 2660 i = __lookup_free_nid_list(nm_i, nid); 2661 f2fs_bug_on(sbi, !i); 2662 2663 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) { 2664 __remove_free_nid(sbi, i, PREALLOC_NID); 2665 need_free = true; 2666 } else { 2667 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID); 2668 } 2669 2670 nm_i->available_nids++; 2671 2672 update_free_nid_bitmap(sbi, nid, true, false); 2673 2674 spin_unlock(&nm_i->nid_list_lock); 2675 2676 if (need_free) 2677 kmem_cache_free(free_nid_slab, i); 2678 } 2679 2680 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink) 2681 { 2682 struct f2fs_nm_info *nm_i = NM_I(sbi); 2683 int nr = nr_shrink; 2684 2685 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2686 return 0; 2687 2688 if (!mutex_trylock(&nm_i->build_lock)) 2689 return 0; 2690 2691 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) { 2692 struct free_nid *i, *next; 2693 unsigned int batch = SHRINK_NID_BATCH_SIZE; 2694 2695 spin_lock(&nm_i->nid_list_lock); 2696 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) { 2697 if (!nr_shrink || !batch || 2698 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2699 break; 2700 __remove_free_nid(sbi, i, FREE_NID); 2701 kmem_cache_free(free_nid_slab, i); 2702 nr_shrink--; 2703 batch--; 2704 } 2705 spin_unlock(&nm_i->nid_list_lock); 2706 } 2707 2708 mutex_unlock(&nm_i->build_lock); 2709 2710 return nr - nr_shrink; 2711 } 2712 2713 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page) 2714 { 2715 void *src_addr, *dst_addr; 2716 size_t inline_size; 2717 struct page *ipage; 2718 struct f2fs_inode *ri; 2719 2720 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 2721 if (IS_ERR(ipage)) 2722 return PTR_ERR(ipage); 2723 2724 ri = F2FS_INODE(page); 2725 if (ri->i_inline & F2FS_INLINE_XATTR) { 2726 if (!f2fs_has_inline_xattr(inode)) { 2727 set_inode_flag(inode, FI_INLINE_XATTR); 2728 stat_inc_inline_xattr(inode); 2729 } 2730 } else { 2731 if (f2fs_has_inline_xattr(inode)) { 2732 stat_dec_inline_xattr(inode); 2733 clear_inode_flag(inode, FI_INLINE_XATTR); 2734 } 2735 goto update_inode; 2736 } 2737 2738 dst_addr = inline_xattr_addr(inode, ipage); 2739 src_addr = inline_xattr_addr(inode, page); 2740 inline_size = inline_xattr_size(inode); 2741 2742 f2fs_wait_on_page_writeback(ipage, NODE, true, true); 2743 memcpy(dst_addr, src_addr, inline_size); 2744 update_inode: 2745 f2fs_update_inode(inode, ipage); 2746 f2fs_put_page(ipage, 1); 2747 return 0; 2748 } 2749 2750 int f2fs_recover_xattr_data(struct inode *inode, struct page *page) 2751 { 2752 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2753 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid; 2754 nid_t new_xnid; 2755 struct dnode_of_data dn; 2756 struct node_info ni; 2757 struct page *xpage; 2758 int err; 2759 2760 if (!prev_xnid) 2761 goto recover_xnid; 2762 2763 /* 1: invalidate the previous xattr nid */ 2764 err = f2fs_get_node_info(sbi, prev_xnid, &ni, false); 2765 if (err) 2766 return err; 2767 2768 f2fs_invalidate_blocks(sbi, ni.blk_addr, 1); 2769 dec_valid_node_count(sbi, inode, false); 2770 set_node_addr(sbi, &ni, NULL_ADDR, false); 2771 2772 recover_xnid: 2773 /* 2: update xattr nid in inode */ 2774 if (!f2fs_alloc_nid(sbi, &new_xnid)) 2775 return -ENOSPC; 2776 2777 set_new_dnode(&dn, inode, NULL, NULL, new_xnid); 2778 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET); 2779 if (IS_ERR(xpage)) { 2780 f2fs_alloc_nid_failed(sbi, new_xnid); 2781 return PTR_ERR(xpage); 2782 } 2783 2784 f2fs_alloc_nid_done(sbi, new_xnid); 2785 f2fs_update_inode_page(inode); 2786 2787 /* 3: update and set xattr node page dirty */ 2788 if (page) { 2789 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), 2790 VALID_XATTR_BLOCK_SIZE); 2791 set_page_dirty(xpage); 2792 } 2793 f2fs_put_page(xpage, 1); 2794 2795 return 0; 2796 } 2797 2798 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 2799 { 2800 struct f2fs_inode *src, *dst; 2801 nid_t ino = ino_of_node(page); 2802 struct node_info old_ni, new_ni; 2803 struct page *ipage; 2804 int err; 2805 2806 err = f2fs_get_node_info(sbi, ino, &old_ni, false); 2807 if (err) 2808 return err; 2809 2810 if (unlikely(old_ni.blk_addr != NULL_ADDR)) 2811 return -EINVAL; 2812 retry: 2813 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false); 2814 if (!ipage) { 2815 memalloc_retry_wait(GFP_NOFS); 2816 goto retry; 2817 } 2818 2819 /* Should not use this inode from free nid list */ 2820 remove_free_nid(sbi, ino); 2821 2822 if (!PageUptodate(ipage)) 2823 SetPageUptodate(ipage); 2824 fill_node_footer(ipage, ino, ino, 0, true); 2825 set_cold_node(ipage, false); 2826 2827 src = F2FS_INODE(page); 2828 dst = F2FS_INODE(ipage); 2829 2830 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext)); 2831 dst->i_size = 0; 2832 dst->i_blocks = cpu_to_le64(1); 2833 dst->i_links = cpu_to_le32(1); 2834 dst->i_xattr_nid = 0; 2835 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR); 2836 if (dst->i_inline & F2FS_EXTRA_ATTR) { 2837 dst->i_extra_isize = src->i_extra_isize; 2838 2839 if (f2fs_sb_has_flexible_inline_xattr(sbi) && 2840 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2841 i_inline_xattr_size)) 2842 dst->i_inline_xattr_size = src->i_inline_xattr_size; 2843 2844 if (f2fs_sb_has_project_quota(sbi) && 2845 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2846 i_projid)) 2847 dst->i_projid = src->i_projid; 2848 2849 if (f2fs_sb_has_inode_crtime(sbi) && 2850 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2851 i_crtime_nsec)) { 2852 dst->i_crtime = src->i_crtime; 2853 dst->i_crtime_nsec = src->i_crtime_nsec; 2854 } 2855 } 2856 2857 new_ni = old_ni; 2858 new_ni.ino = ino; 2859 2860 if (unlikely(inc_valid_node_count(sbi, NULL, true))) 2861 WARN_ON(1); 2862 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 2863 inc_valid_inode_count(sbi); 2864 set_page_dirty(ipage); 2865 f2fs_put_page(ipage, 1); 2866 return 0; 2867 } 2868 2869 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi, 2870 unsigned int segno, struct f2fs_summary_block *sum) 2871 { 2872 struct f2fs_node *rn; 2873 struct f2fs_summary *sum_entry; 2874 block_t addr; 2875 int i, idx, last_offset, nrpages; 2876 2877 /* scan the node segment */ 2878 last_offset = BLKS_PER_SEG(sbi); 2879 addr = START_BLOCK(sbi, segno); 2880 sum_entry = &sum->entries[0]; 2881 2882 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) { 2883 nrpages = bio_max_segs(last_offset - i); 2884 2885 /* readahead node pages */ 2886 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true); 2887 2888 for (idx = addr; idx < addr + nrpages; idx++) { 2889 struct page *page = f2fs_get_tmp_page(sbi, idx); 2890 2891 if (IS_ERR(page)) 2892 return PTR_ERR(page); 2893 2894 rn = F2FS_NODE(page); 2895 sum_entry->nid = rn->footer.nid; 2896 sum_entry->version = 0; 2897 sum_entry->ofs_in_node = 0; 2898 sum_entry++; 2899 f2fs_put_page(page, 1); 2900 } 2901 2902 invalidate_mapping_pages(META_MAPPING(sbi), addr, 2903 addr + nrpages); 2904 } 2905 return 0; 2906 } 2907 2908 static void remove_nats_in_journal(struct f2fs_sb_info *sbi) 2909 { 2910 struct f2fs_nm_info *nm_i = NM_I(sbi); 2911 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2912 struct f2fs_journal *journal = curseg->journal; 2913 int i; 2914 2915 down_write(&curseg->journal_rwsem); 2916 for (i = 0; i < nats_in_cursum(journal); i++) { 2917 struct nat_entry *ne; 2918 struct f2fs_nat_entry raw_ne; 2919 nid_t nid = le32_to_cpu(nid_in_journal(journal, i)); 2920 2921 if (f2fs_check_nid_range(sbi, nid)) 2922 continue; 2923 2924 raw_ne = nat_in_journal(journal, i); 2925 2926 ne = __lookup_nat_cache(nm_i, nid); 2927 if (!ne) { 2928 ne = __alloc_nat_entry(sbi, nid, true); 2929 __init_nat_entry(nm_i, ne, &raw_ne, true); 2930 } 2931 2932 /* 2933 * if a free nat in journal has not been used after last 2934 * checkpoint, we should remove it from available nids, 2935 * since later we will add it again. 2936 */ 2937 if (!get_nat_flag(ne, IS_DIRTY) && 2938 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) { 2939 spin_lock(&nm_i->nid_list_lock); 2940 nm_i->available_nids--; 2941 spin_unlock(&nm_i->nid_list_lock); 2942 } 2943 2944 __set_nat_cache_dirty(nm_i, ne); 2945 } 2946 update_nats_in_cursum(journal, -i); 2947 up_write(&curseg->journal_rwsem); 2948 } 2949 2950 static void __adjust_nat_entry_set(struct nat_entry_set *nes, 2951 struct list_head *head, int max) 2952 { 2953 struct nat_entry_set *cur; 2954 2955 if (nes->entry_cnt >= max) 2956 goto add_out; 2957 2958 list_for_each_entry(cur, head, set_list) { 2959 if (cur->entry_cnt >= nes->entry_cnt) { 2960 list_add(&nes->set_list, cur->set_list.prev); 2961 return; 2962 } 2963 } 2964 add_out: 2965 list_add_tail(&nes->set_list, head); 2966 } 2967 2968 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs, 2969 unsigned int valid) 2970 { 2971 if (valid == 0) { 2972 __set_bit_le(nat_ofs, nm_i->empty_nat_bits); 2973 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2974 return; 2975 } 2976 2977 __clear_bit_le(nat_ofs, nm_i->empty_nat_bits); 2978 if (valid == NAT_ENTRY_PER_BLOCK) 2979 __set_bit_le(nat_ofs, nm_i->full_nat_bits); 2980 else 2981 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2982 } 2983 2984 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid, 2985 struct page *page) 2986 { 2987 struct f2fs_nm_info *nm_i = NM_I(sbi); 2988 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK; 2989 struct f2fs_nat_block *nat_blk = page_address(page); 2990 int valid = 0; 2991 int i = 0; 2992 2993 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 2994 return; 2995 2996 if (nat_index == 0) { 2997 valid = 1; 2998 i = 1; 2999 } 3000 for (; i < NAT_ENTRY_PER_BLOCK; i++) { 3001 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR) 3002 valid++; 3003 } 3004 3005 __update_nat_bits(nm_i, nat_index, valid); 3006 } 3007 3008 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi) 3009 { 3010 struct f2fs_nm_info *nm_i = NM_I(sbi); 3011 unsigned int nat_ofs; 3012 3013 f2fs_down_read(&nm_i->nat_tree_lock); 3014 3015 for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) { 3016 unsigned int valid = 0, nid_ofs = 0; 3017 3018 /* handle nid zero due to it should never be used */ 3019 if (unlikely(nat_ofs == 0)) { 3020 valid = 1; 3021 nid_ofs = 1; 3022 } 3023 3024 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) { 3025 if (!test_bit_le(nid_ofs, 3026 nm_i->free_nid_bitmap[nat_ofs])) 3027 valid++; 3028 } 3029 3030 __update_nat_bits(nm_i, nat_ofs, valid); 3031 } 3032 3033 f2fs_up_read(&nm_i->nat_tree_lock); 3034 } 3035 3036 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi, 3037 struct nat_entry_set *set, struct cp_control *cpc) 3038 { 3039 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 3040 struct f2fs_journal *journal = curseg->journal; 3041 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK; 3042 bool to_journal = true; 3043 struct f2fs_nat_block *nat_blk; 3044 struct nat_entry *ne, *cur; 3045 struct page *page = NULL; 3046 3047 /* 3048 * there are two steps to flush nat entries: 3049 * #1, flush nat entries to journal in current hot data summary block. 3050 * #2, flush nat entries to nat page. 3051 */ 3052 if ((cpc->reason & CP_UMOUNT) || 3053 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL)) 3054 to_journal = false; 3055 3056 if (to_journal) { 3057 down_write(&curseg->journal_rwsem); 3058 } else { 3059 page = get_next_nat_page(sbi, start_nid); 3060 if (IS_ERR(page)) 3061 return PTR_ERR(page); 3062 3063 nat_blk = page_address(page); 3064 f2fs_bug_on(sbi, !nat_blk); 3065 } 3066 3067 /* flush dirty nats in nat entry set */ 3068 list_for_each_entry_safe(ne, cur, &set->entry_list, list) { 3069 struct f2fs_nat_entry *raw_ne; 3070 nid_t nid = nat_get_nid(ne); 3071 int offset; 3072 3073 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR); 3074 3075 if (to_journal) { 3076 offset = f2fs_lookup_journal_in_cursum(journal, 3077 NAT_JOURNAL, nid, 1); 3078 f2fs_bug_on(sbi, offset < 0); 3079 raw_ne = &nat_in_journal(journal, offset); 3080 nid_in_journal(journal, offset) = cpu_to_le32(nid); 3081 } else { 3082 raw_ne = &nat_blk->entries[nid - start_nid]; 3083 } 3084 raw_nat_from_node_info(raw_ne, &ne->ni); 3085 nat_reset_flag(ne); 3086 __clear_nat_cache_dirty(NM_I(sbi), set, ne); 3087 if (nat_get_blkaddr(ne) == NULL_ADDR) { 3088 add_free_nid(sbi, nid, false, true); 3089 } else { 3090 spin_lock(&NM_I(sbi)->nid_list_lock); 3091 update_free_nid_bitmap(sbi, nid, false, false); 3092 spin_unlock(&NM_I(sbi)->nid_list_lock); 3093 } 3094 } 3095 3096 if (to_journal) { 3097 up_write(&curseg->journal_rwsem); 3098 } else { 3099 update_nat_bits(sbi, start_nid, page); 3100 f2fs_put_page(page, 1); 3101 } 3102 3103 /* Allow dirty nats by node block allocation in write_begin */ 3104 if (!set->entry_cnt) { 3105 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set); 3106 kmem_cache_free(nat_entry_set_slab, set); 3107 } 3108 return 0; 3109 } 3110 3111 /* 3112 * This function is called during the checkpointing process. 3113 */ 3114 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3115 { 3116 struct f2fs_nm_info *nm_i = NM_I(sbi); 3117 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 3118 struct f2fs_journal *journal = curseg->journal; 3119 struct nat_entry_set *setvec[NAT_VEC_SIZE]; 3120 struct nat_entry_set *set, *tmp; 3121 unsigned int found; 3122 nid_t set_idx = 0; 3123 LIST_HEAD(sets); 3124 int err = 0; 3125 3126 /* 3127 * during unmount, let's flush nat_bits before checking 3128 * nat_cnt[DIRTY_NAT]. 3129 */ 3130 if (cpc->reason & CP_UMOUNT) { 3131 f2fs_down_write(&nm_i->nat_tree_lock); 3132 remove_nats_in_journal(sbi); 3133 f2fs_up_write(&nm_i->nat_tree_lock); 3134 } 3135 3136 if (!nm_i->nat_cnt[DIRTY_NAT]) 3137 return 0; 3138 3139 f2fs_down_write(&nm_i->nat_tree_lock); 3140 3141 /* 3142 * if there are no enough space in journal to store dirty nat 3143 * entries, remove all entries from journal and merge them 3144 * into nat entry set. 3145 */ 3146 if (cpc->reason & CP_UMOUNT || 3147 !__has_cursum_space(journal, 3148 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL)) 3149 remove_nats_in_journal(sbi); 3150 3151 while ((found = __gang_lookup_nat_set(nm_i, 3152 set_idx, NAT_VEC_SIZE, setvec))) { 3153 unsigned idx; 3154 3155 set_idx = setvec[found - 1]->set + 1; 3156 for (idx = 0; idx < found; idx++) 3157 __adjust_nat_entry_set(setvec[idx], &sets, 3158 MAX_NAT_JENTRIES(journal)); 3159 } 3160 3161 /* flush dirty nats in nat entry set */ 3162 list_for_each_entry_safe(set, tmp, &sets, set_list) { 3163 err = __flush_nat_entry_set(sbi, set, cpc); 3164 if (err) 3165 break; 3166 } 3167 3168 f2fs_up_write(&nm_i->nat_tree_lock); 3169 /* Allow dirty nats by node block allocation in write_begin */ 3170 3171 return err; 3172 } 3173 3174 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi) 3175 { 3176 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3177 struct f2fs_nm_info *nm_i = NM_I(sbi); 3178 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE; 3179 unsigned int i; 3180 __u64 cp_ver = cur_cp_version(ckpt); 3181 block_t nat_bits_addr; 3182 3183 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8); 3184 nm_i->nat_bits = f2fs_kvzalloc(sbi, 3185 F2FS_BLK_TO_BYTES(nm_i->nat_bits_blocks), GFP_KERNEL); 3186 if (!nm_i->nat_bits) 3187 return -ENOMEM; 3188 3189 nm_i->full_nat_bits = nm_i->nat_bits + 8; 3190 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes; 3191 3192 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3193 return 0; 3194 3195 nat_bits_addr = __start_cp_addr(sbi) + BLKS_PER_SEG(sbi) - 3196 nm_i->nat_bits_blocks; 3197 for (i = 0; i < nm_i->nat_bits_blocks; i++) { 3198 struct page *page; 3199 3200 page = f2fs_get_meta_page(sbi, nat_bits_addr++); 3201 if (IS_ERR(page)) 3202 return PTR_ERR(page); 3203 3204 memcpy(nm_i->nat_bits + F2FS_BLK_TO_BYTES(i), 3205 page_address(page), F2FS_BLKSIZE); 3206 f2fs_put_page(page, 1); 3207 } 3208 3209 cp_ver |= (cur_cp_crc(ckpt) << 32); 3210 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) { 3211 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG); 3212 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)", 3213 cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits)); 3214 return 0; 3215 } 3216 3217 f2fs_notice(sbi, "Found nat_bits in checkpoint"); 3218 return 0; 3219 } 3220 3221 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi) 3222 { 3223 struct f2fs_nm_info *nm_i = NM_I(sbi); 3224 unsigned int i = 0; 3225 nid_t nid, last_nid; 3226 3227 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3228 return; 3229 3230 for (i = 0; i < nm_i->nat_blocks; i++) { 3231 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i); 3232 if (i >= nm_i->nat_blocks) 3233 break; 3234 3235 __set_bit_le(i, nm_i->nat_block_bitmap); 3236 3237 nid = i * NAT_ENTRY_PER_BLOCK; 3238 last_nid = nid + NAT_ENTRY_PER_BLOCK; 3239 3240 spin_lock(&NM_I(sbi)->nid_list_lock); 3241 for (; nid < last_nid; nid++) 3242 update_free_nid_bitmap(sbi, nid, true, true); 3243 spin_unlock(&NM_I(sbi)->nid_list_lock); 3244 } 3245 3246 for (i = 0; i < nm_i->nat_blocks; i++) { 3247 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i); 3248 if (i >= nm_i->nat_blocks) 3249 break; 3250 3251 __set_bit_le(i, nm_i->nat_block_bitmap); 3252 } 3253 } 3254 3255 static int init_node_manager(struct f2fs_sb_info *sbi) 3256 { 3257 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 3258 struct f2fs_nm_info *nm_i = NM_I(sbi); 3259 unsigned char *version_bitmap; 3260 unsigned int nat_segs; 3261 int err; 3262 3263 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 3264 3265 /* segment_count_nat includes pair segment so divide to 2. */ 3266 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 3267 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 3268 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks; 3269 3270 /* not used nids: 0, node, meta, (and root counted as valid node) */ 3271 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count - 3272 F2FS_RESERVED_NODE_NUM; 3273 nm_i->nid_cnt[FREE_NID] = 0; 3274 nm_i->nid_cnt[PREALLOC_NID] = 0; 3275 nm_i->ram_thresh = DEF_RAM_THRESHOLD; 3276 nm_i->ra_nid_pages = DEF_RA_NID_PAGES; 3277 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD; 3278 nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS; 3279 3280 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC); 3281 INIT_LIST_HEAD(&nm_i->free_nid_list); 3282 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO); 3283 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO); 3284 INIT_LIST_HEAD(&nm_i->nat_entries); 3285 spin_lock_init(&nm_i->nat_list_lock); 3286 3287 mutex_init(&nm_i->build_lock); 3288 spin_lock_init(&nm_i->nid_list_lock); 3289 init_f2fs_rwsem(&nm_i->nat_tree_lock); 3290 3291 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 3292 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 3293 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 3294 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size, 3295 GFP_KERNEL); 3296 if (!nm_i->nat_bitmap) 3297 return -ENOMEM; 3298 3299 err = __get_nat_bitmaps(sbi); 3300 if (err) 3301 return err; 3302 3303 #ifdef CONFIG_F2FS_CHECK_FS 3304 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size, 3305 GFP_KERNEL); 3306 if (!nm_i->nat_bitmap_mir) 3307 return -ENOMEM; 3308 #endif 3309 3310 return 0; 3311 } 3312 3313 static int init_free_nid_cache(struct f2fs_sb_info *sbi) 3314 { 3315 struct f2fs_nm_info *nm_i = NM_I(sbi); 3316 int i; 3317 3318 nm_i->free_nid_bitmap = 3319 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *), 3320 nm_i->nat_blocks), 3321 GFP_KERNEL); 3322 if (!nm_i->free_nid_bitmap) 3323 return -ENOMEM; 3324 3325 for (i = 0; i < nm_i->nat_blocks; i++) { 3326 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi, 3327 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL); 3328 if (!nm_i->free_nid_bitmap[i]) 3329 return -ENOMEM; 3330 } 3331 3332 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8, 3333 GFP_KERNEL); 3334 if (!nm_i->nat_block_bitmap) 3335 return -ENOMEM; 3336 3337 nm_i->free_nid_count = 3338 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short), 3339 nm_i->nat_blocks), 3340 GFP_KERNEL); 3341 if (!nm_i->free_nid_count) 3342 return -ENOMEM; 3343 return 0; 3344 } 3345 3346 int f2fs_build_node_manager(struct f2fs_sb_info *sbi) 3347 { 3348 int err; 3349 3350 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info), 3351 GFP_KERNEL); 3352 if (!sbi->nm_info) 3353 return -ENOMEM; 3354 3355 err = init_node_manager(sbi); 3356 if (err) 3357 return err; 3358 3359 err = init_free_nid_cache(sbi); 3360 if (err) 3361 return err; 3362 3363 /* load free nid status from nat_bits table */ 3364 load_free_nid_bitmap(sbi); 3365 3366 return f2fs_build_free_nids(sbi, true, true); 3367 } 3368 3369 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi) 3370 { 3371 struct f2fs_nm_info *nm_i = NM_I(sbi); 3372 struct free_nid *i, *next_i; 3373 void *vec[NAT_VEC_SIZE]; 3374 struct nat_entry **natvec = (struct nat_entry **)vec; 3375 struct nat_entry_set **setvec = (struct nat_entry_set **)vec; 3376 nid_t nid = 0; 3377 unsigned int found; 3378 3379 if (!nm_i) 3380 return; 3381 3382 /* destroy free nid list */ 3383 spin_lock(&nm_i->nid_list_lock); 3384 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 3385 __remove_free_nid(sbi, i, FREE_NID); 3386 spin_unlock(&nm_i->nid_list_lock); 3387 kmem_cache_free(free_nid_slab, i); 3388 spin_lock(&nm_i->nid_list_lock); 3389 } 3390 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]); 3391 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]); 3392 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list)); 3393 spin_unlock(&nm_i->nid_list_lock); 3394 3395 /* destroy nat cache */ 3396 f2fs_down_write(&nm_i->nat_tree_lock); 3397 while ((found = __gang_lookup_nat_cache(nm_i, 3398 nid, NAT_VEC_SIZE, natvec))) { 3399 unsigned idx; 3400 3401 nid = nat_get_nid(natvec[found - 1]) + 1; 3402 for (idx = 0; idx < found; idx++) { 3403 spin_lock(&nm_i->nat_list_lock); 3404 list_del(&natvec[idx]->list); 3405 spin_unlock(&nm_i->nat_list_lock); 3406 3407 __del_from_nat_cache(nm_i, natvec[idx]); 3408 } 3409 } 3410 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]); 3411 3412 /* destroy nat set cache */ 3413 nid = 0; 3414 memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE); 3415 while ((found = __gang_lookup_nat_set(nm_i, 3416 nid, NAT_VEC_SIZE, setvec))) { 3417 unsigned idx; 3418 3419 nid = setvec[found - 1]->set + 1; 3420 for (idx = 0; idx < found; idx++) { 3421 /* entry_cnt is not zero, when cp_error was occurred */ 3422 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list)); 3423 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set); 3424 kmem_cache_free(nat_entry_set_slab, setvec[idx]); 3425 } 3426 } 3427 f2fs_up_write(&nm_i->nat_tree_lock); 3428 3429 kvfree(nm_i->nat_block_bitmap); 3430 if (nm_i->free_nid_bitmap) { 3431 int i; 3432 3433 for (i = 0; i < nm_i->nat_blocks; i++) 3434 kvfree(nm_i->free_nid_bitmap[i]); 3435 kvfree(nm_i->free_nid_bitmap); 3436 } 3437 kvfree(nm_i->free_nid_count); 3438 3439 kvfree(nm_i->nat_bitmap); 3440 kvfree(nm_i->nat_bits); 3441 #ifdef CONFIG_F2FS_CHECK_FS 3442 kvfree(nm_i->nat_bitmap_mir); 3443 #endif 3444 sbi->nm_info = NULL; 3445 kfree(nm_i); 3446 } 3447 3448 int __init f2fs_create_node_manager_caches(void) 3449 { 3450 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry", 3451 sizeof(struct nat_entry)); 3452 if (!nat_entry_slab) 3453 goto fail; 3454 3455 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid", 3456 sizeof(struct free_nid)); 3457 if (!free_nid_slab) 3458 goto destroy_nat_entry; 3459 3460 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set", 3461 sizeof(struct nat_entry_set)); 3462 if (!nat_entry_set_slab) 3463 goto destroy_free_nid; 3464 3465 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry", 3466 sizeof(struct fsync_node_entry)); 3467 if (!fsync_node_entry_slab) 3468 goto destroy_nat_entry_set; 3469 return 0; 3470 3471 destroy_nat_entry_set: 3472 kmem_cache_destroy(nat_entry_set_slab); 3473 destroy_free_nid: 3474 kmem_cache_destroy(free_nid_slab); 3475 destroy_nat_entry: 3476 kmem_cache_destroy(nat_entry_slab); 3477 fail: 3478 return -ENOMEM; 3479 } 3480 3481 void f2fs_destroy_node_manager_caches(void) 3482 { 3483 kmem_cache_destroy(fsync_node_entry_slab); 3484 kmem_cache_destroy(nat_entry_set_slab); 3485 kmem_cache_destroy(free_nid_slab); 3486 kmem_cache_destroy(nat_entry_slab); 3487 } 3488