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 if (!level) { 1140 level = -EFSCORRUPTED; 1141 f2fs_err(sbi, "%s: inode ino=%lx has corrupted node block, from:%lu addrs:%u", 1142 __func__, inode->i_ino, 1143 from, ADDRS_PER_INODE(inode)); 1144 set_sbi_flag(sbi, SBI_NEED_FSCK); 1145 } 1146 trace_f2fs_truncate_inode_blocks_exit(inode, level); 1147 return level; 1148 } 1149 1150 page = f2fs_get_node_page(sbi, inode->i_ino); 1151 if (IS_ERR(page)) { 1152 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page)); 1153 return PTR_ERR(page); 1154 } 1155 1156 set_new_dnode(&dn, inode, page, NULL, 0); 1157 unlock_page(page); 1158 1159 ri = F2FS_INODE(page); 1160 switch (level) { 1161 case 0: 1162 case 1: 1163 nofs = noffset[1]; 1164 break; 1165 case 2: 1166 nofs = noffset[1]; 1167 if (!offset[level - 1]) 1168 goto skip_partial; 1169 err = truncate_partial_nodes(&dn, ri, offset, level); 1170 if (err < 0 && err != -ENOENT) 1171 goto fail; 1172 nofs += 1 + NIDS_PER_BLOCK; 1173 break; 1174 case 3: 1175 nofs = 5 + 2 * NIDS_PER_BLOCK; 1176 if (!offset[level - 1]) 1177 goto skip_partial; 1178 err = truncate_partial_nodes(&dn, ri, offset, level); 1179 if (err < 0 && err != -ENOENT) 1180 goto fail; 1181 break; 1182 default: 1183 BUG(); 1184 } 1185 1186 skip_partial: 1187 while (cont) { 1188 dn.nid = get_nid(page, offset[0], true); 1189 switch (offset[0]) { 1190 case NODE_DIR1_BLOCK: 1191 case NODE_DIR2_BLOCK: 1192 err = truncate_dnode(&dn); 1193 break; 1194 1195 case NODE_IND1_BLOCK: 1196 case NODE_IND2_BLOCK: 1197 err = truncate_nodes(&dn, nofs, offset[1], 2); 1198 break; 1199 1200 case NODE_DIND_BLOCK: 1201 err = truncate_nodes(&dn, nofs, offset[1], 3); 1202 cont = 0; 1203 break; 1204 1205 default: 1206 BUG(); 1207 } 1208 if (err == -ENOENT) { 1209 set_sbi_flag(F2FS_P_SB(page), SBI_NEED_FSCK); 1210 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 1211 f2fs_err_ratelimited(sbi, 1212 "truncate node fail, ino:%lu, nid:%u, " 1213 "offset[0]:%d, offset[1]:%d, nofs:%d", 1214 inode->i_ino, dn.nid, offset[0], 1215 offset[1], nofs); 1216 err = 0; 1217 } 1218 if (err < 0) 1219 goto fail; 1220 if (offset[1] == 0 && get_nid(page, offset[0], true)) { 1221 lock_page(page); 1222 BUG_ON(page->mapping != NODE_MAPPING(sbi)); 1223 set_nid(page, offset[0], 0, true); 1224 unlock_page(page); 1225 } 1226 offset[1] = 0; 1227 offset[0]++; 1228 nofs += err; 1229 } 1230 fail: 1231 f2fs_put_page(page, 0); 1232 trace_f2fs_truncate_inode_blocks_exit(inode, err); 1233 return err > 0 ? 0 : err; 1234 } 1235 1236 /* caller must lock inode page */ 1237 int f2fs_truncate_xattr_node(struct inode *inode) 1238 { 1239 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1240 nid_t nid = F2FS_I(inode)->i_xattr_nid; 1241 struct dnode_of_data dn; 1242 struct page *npage; 1243 int err; 1244 1245 if (!nid) 1246 return 0; 1247 1248 npage = f2fs_get_node_page(sbi, nid); 1249 if (IS_ERR(npage)) 1250 return PTR_ERR(npage); 1251 1252 set_new_dnode(&dn, inode, NULL, npage, nid); 1253 err = truncate_node(&dn); 1254 if (err) { 1255 f2fs_put_page(npage, 1); 1256 return err; 1257 } 1258 1259 f2fs_i_xnid_write(inode, 0); 1260 1261 return 0; 1262 } 1263 1264 /* 1265 * Caller should grab and release a rwsem by calling f2fs_lock_op() and 1266 * f2fs_unlock_op(). 1267 */ 1268 int f2fs_remove_inode_page(struct inode *inode) 1269 { 1270 struct dnode_of_data dn; 1271 int err; 1272 1273 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1274 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE); 1275 if (err) 1276 return err; 1277 1278 err = f2fs_truncate_xattr_node(inode); 1279 if (err) { 1280 f2fs_put_dnode(&dn); 1281 return err; 1282 } 1283 1284 /* remove potential inline_data blocks */ 1285 if (!IS_DEVICE_ALIASING(inode) && 1286 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1287 S_ISLNK(inode->i_mode))) 1288 f2fs_truncate_data_blocks_range(&dn, 1); 1289 1290 /* 0 is possible, after f2fs_new_inode() has failed */ 1291 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) { 1292 f2fs_put_dnode(&dn); 1293 return -EIO; 1294 } 1295 1296 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) { 1297 f2fs_warn(F2FS_I_SB(inode), 1298 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu", 1299 inode->i_ino, (unsigned long long)inode->i_blocks); 1300 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK); 1301 } 1302 1303 /* will put inode & node pages */ 1304 err = truncate_node(&dn); 1305 if (err) { 1306 f2fs_put_dnode(&dn); 1307 return err; 1308 } 1309 return 0; 1310 } 1311 1312 struct page *f2fs_new_inode_page(struct inode *inode) 1313 { 1314 struct dnode_of_data dn; 1315 1316 /* allocate inode page for new inode */ 1317 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1318 1319 /* caller should f2fs_put_page(page, 1); */ 1320 return f2fs_new_node_page(&dn, 0); 1321 } 1322 1323 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs) 1324 { 1325 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1326 struct node_info new_ni; 1327 struct page *page; 1328 int err; 1329 1330 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1331 return ERR_PTR(-EPERM); 1332 1333 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false); 1334 if (!page) 1335 return ERR_PTR(-ENOMEM); 1336 1337 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs)))) 1338 goto fail; 1339 1340 #ifdef CONFIG_F2FS_CHECK_FS 1341 err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false); 1342 if (err) { 1343 dec_valid_node_count(sbi, dn->inode, !ofs); 1344 goto fail; 1345 } 1346 if (unlikely(new_ni.blk_addr != NULL_ADDR)) { 1347 err = -EFSCORRUPTED; 1348 dec_valid_node_count(sbi, dn->inode, !ofs); 1349 set_sbi_flag(sbi, SBI_NEED_FSCK); 1350 f2fs_warn_ratelimited(sbi, 1351 "f2fs_new_node_page: inconsistent nat entry, " 1352 "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u", 1353 new_ni.ino, new_ni.nid, new_ni.blk_addr, 1354 new_ni.version, new_ni.flag); 1355 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT); 1356 goto fail; 1357 } 1358 #endif 1359 new_ni.nid = dn->nid; 1360 new_ni.ino = dn->inode->i_ino; 1361 new_ni.blk_addr = NULL_ADDR; 1362 new_ni.flag = 0; 1363 new_ni.version = 0; 1364 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 1365 1366 f2fs_wait_on_page_writeback(page, NODE, true, true); 1367 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 1368 set_cold_node(page, S_ISDIR(dn->inode->i_mode)); 1369 if (!PageUptodate(page)) 1370 SetPageUptodate(page); 1371 if (set_page_dirty(page)) 1372 dn->node_changed = true; 1373 1374 if (f2fs_has_xattr_block(ofs)) 1375 f2fs_i_xnid_write(dn->inode, dn->nid); 1376 1377 if (ofs == 0) 1378 inc_valid_inode_count(sbi); 1379 return page; 1380 fail: 1381 clear_node_page_dirty(page); 1382 f2fs_put_page(page, 1); 1383 return ERR_PTR(err); 1384 } 1385 1386 /* 1387 * Caller should do after getting the following values. 1388 * 0: f2fs_put_page(page, 0) 1389 * LOCKED_PAGE or error: f2fs_put_page(page, 1) 1390 */ 1391 static int read_node_page(struct page *page, blk_opf_t op_flags) 1392 { 1393 struct folio *folio = page_folio(page); 1394 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1395 struct node_info ni; 1396 struct f2fs_io_info fio = { 1397 .sbi = sbi, 1398 .type = NODE, 1399 .op = REQ_OP_READ, 1400 .op_flags = op_flags, 1401 .page = page, 1402 .encrypted_page = NULL, 1403 }; 1404 int err; 1405 1406 if (folio_test_uptodate(folio)) { 1407 if (!f2fs_inode_chksum_verify(sbi, page)) { 1408 folio_clear_uptodate(folio); 1409 return -EFSBADCRC; 1410 } 1411 return LOCKED_PAGE; 1412 } 1413 1414 err = f2fs_get_node_info(sbi, folio->index, &ni, false); 1415 if (err) 1416 return err; 1417 1418 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */ 1419 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) { 1420 folio_clear_uptodate(folio); 1421 return -ENOENT; 1422 } 1423 1424 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr; 1425 1426 err = f2fs_submit_page_bio(&fio); 1427 1428 if (!err) 1429 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE); 1430 1431 return err; 1432 } 1433 1434 /* 1435 * Readahead a node page 1436 */ 1437 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 1438 { 1439 struct page *apage; 1440 int err; 1441 1442 if (!nid) 1443 return; 1444 if (f2fs_check_nid_range(sbi, nid)) 1445 return; 1446 1447 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid); 1448 if (apage) 1449 return; 1450 1451 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1452 if (!apage) 1453 return; 1454 1455 err = read_node_page(apage, REQ_RAHEAD); 1456 f2fs_put_page(apage, err ? 1 : 0); 1457 } 1458 1459 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid, 1460 struct page *parent, int start) 1461 { 1462 struct page *page; 1463 int err; 1464 1465 if (!nid) 1466 return ERR_PTR(-ENOENT); 1467 if (f2fs_check_nid_range(sbi, nid)) 1468 return ERR_PTR(-EINVAL); 1469 repeat: 1470 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1471 if (!page) 1472 return ERR_PTR(-ENOMEM); 1473 1474 err = read_node_page(page, 0); 1475 if (err < 0) { 1476 goto out_put_err; 1477 } else if (err == LOCKED_PAGE) { 1478 err = 0; 1479 goto page_hit; 1480 } 1481 1482 if (parent) 1483 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE); 1484 1485 lock_page(page); 1486 1487 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1488 f2fs_put_page(page, 1); 1489 goto repeat; 1490 } 1491 1492 if (unlikely(!PageUptodate(page))) { 1493 err = -EIO; 1494 goto out_err; 1495 } 1496 1497 if (!f2fs_inode_chksum_verify(sbi, page)) { 1498 err = -EFSBADCRC; 1499 goto out_err; 1500 } 1501 page_hit: 1502 if (likely(nid == nid_of_node(page))) 1503 return page; 1504 1505 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]", 1506 nid, nid_of_node(page), ino_of_node(page), 1507 ofs_of_node(page), cpver_of_node(page), 1508 next_blkaddr_of_node(page)); 1509 set_sbi_flag(sbi, SBI_NEED_FSCK); 1510 f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER); 1511 err = -EFSCORRUPTED; 1512 out_err: 1513 ClearPageUptodate(page); 1514 out_put_err: 1515 /* ENOENT comes from read_node_page which is not an error. */ 1516 if (err != -ENOENT) 1517 f2fs_handle_page_eio(sbi, page_folio(page), NODE); 1518 f2fs_put_page(page, 1); 1519 return ERR_PTR(err); 1520 } 1521 1522 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 1523 { 1524 return __get_node_page(sbi, nid, NULL, 0); 1525 } 1526 1527 struct page *f2fs_get_node_page_ra(struct page *parent, int start) 1528 { 1529 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 1530 nid_t nid = get_nid(parent, start, false); 1531 1532 return __get_node_page(sbi, nid, parent, start); 1533 } 1534 1535 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino) 1536 { 1537 struct inode *inode; 1538 struct page *page; 1539 int ret; 1540 1541 /* should flush inline_data before evict_inode */ 1542 inode = ilookup(sbi->sb, ino); 1543 if (!inode) 1544 return; 1545 1546 page = f2fs_pagecache_get_page(inode->i_mapping, 0, 1547 FGP_LOCK|FGP_NOWAIT, 0); 1548 if (!page) 1549 goto iput_out; 1550 1551 if (!PageUptodate(page)) 1552 goto page_out; 1553 1554 if (!PageDirty(page)) 1555 goto page_out; 1556 1557 if (!clear_page_dirty_for_io(page)) 1558 goto page_out; 1559 1560 ret = f2fs_write_inline_data(inode, page_folio(page)); 1561 inode_dec_dirty_pages(inode); 1562 f2fs_remove_dirty_inode(inode); 1563 if (ret) 1564 set_page_dirty(page); 1565 page_out: 1566 f2fs_put_page(page, 1); 1567 iput_out: 1568 iput(inode); 1569 } 1570 1571 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino) 1572 { 1573 pgoff_t index; 1574 struct folio_batch fbatch; 1575 struct page *last_page = NULL; 1576 int nr_folios; 1577 1578 folio_batch_init(&fbatch); 1579 index = 0; 1580 1581 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1582 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1583 &fbatch))) { 1584 int i; 1585 1586 for (i = 0; i < nr_folios; i++) { 1587 struct page *page = &fbatch.folios[i]->page; 1588 1589 if (unlikely(f2fs_cp_error(sbi))) { 1590 f2fs_put_page(last_page, 0); 1591 folio_batch_release(&fbatch); 1592 return ERR_PTR(-EIO); 1593 } 1594 1595 if (!IS_DNODE(page) || !is_cold_node(page)) 1596 continue; 1597 if (ino_of_node(page) != ino) 1598 continue; 1599 1600 lock_page(page); 1601 1602 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1603 continue_unlock: 1604 unlock_page(page); 1605 continue; 1606 } 1607 if (ino_of_node(page) != ino) 1608 goto continue_unlock; 1609 1610 if (!PageDirty(page)) { 1611 /* someone wrote it for us */ 1612 goto continue_unlock; 1613 } 1614 1615 if (last_page) 1616 f2fs_put_page(last_page, 0); 1617 1618 get_page(page); 1619 last_page = page; 1620 unlock_page(page); 1621 } 1622 folio_batch_release(&fbatch); 1623 cond_resched(); 1624 } 1625 return last_page; 1626 } 1627 1628 static int __write_node_page(struct page *page, bool atomic, bool *submitted, 1629 struct writeback_control *wbc, bool do_balance, 1630 enum iostat_type io_type, unsigned int *seq_id) 1631 { 1632 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1633 struct folio *folio = page_folio(page); 1634 nid_t nid; 1635 struct node_info ni; 1636 struct f2fs_io_info fio = { 1637 .sbi = sbi, 1638 .ino = ino_of_node(page), 1639 .type = NODE, 1640 .op = REQ_OP_WRITE, 1641 .op_flags = wbc_to_write_flags(wbc), 1642 .page = page, 1643 .encrypted_page = NULL, 1644 .submitted = 0, 1645 .io_type = io_type, 1646 .io_wbc = wbc, 1647 }; 1648 unsigned int seq; 1649 1650 trace_f2fs_writepage(folio, NODE); 1651 1652 if (unlikely(f2fs_cp_error(sbi))) { 1653 /* keep node pages in remount-ro mode */ 1654 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY) 1655 goto redirty_out; 1656 folio_clear_uptodate(folio); 1657 dec_page_count(sbi, F2FS_DIRTY_NODES); 1658 folio_unlock(folio); 1659 return 0; 1660 } 1661 1662 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1663 goto redirty_out; 1664 1665 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 1666 wbc->sync_mode == WB_SYNC_NONE && 1667 IS_DNODE(page) && is_cold_node(page)) 1668 goto redirty_out; 1669 1670 /* get old block addr of this node page */ 1671 nid = nid_of_node(page); 1672 f2fs_bug_on(sbi, folio->index != nid); 1673 1674 if (f2fs_get_node_info(sbi, nid, &ni, !do_balance)) 1675 goto redirty_out; 1676 1677 if (wbc->for_reclaim) { 1678 if (!f2fs_down_read_trylock(&sbi->node_write)) 1679 goto redirty_out; 1680 } else { 1681 f2fs_down_read(&sbi->node_write); 1682 } 1683 1684 /* This page is already truncated */ 1685 if (unlikely(ni.blk_addr == NULL_ADDR)) { 1686 folio_clear_uptodate(folio); 1687 dec_page_count(sbi, F2FS_DIRTY_NODES); 1688 f2fs_up_read(&sbi->node_write); 1689 folio_unlock(folio); 1690 return 0; 1691 } 1692 1693 if (__is_valid_data_blkaddr(ni.blk_addr) && 1694 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, 1695 DATA_GENERIC_ENHANCE)) { 1696 f2fs_up_read(&sbi->node_write); 1697 goto redirty_out; 1698 } 1699 1700 if (atomic && !test_opt(sbi, NOBARRIER)) 1701 fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 1702 1703 /* should add to global list before clearing PAGECACHE status */ 1704 if (f2fs_in_warm_node_list(sbi, page)) { 1705 seq = f2fs_add_fsync_node_entry(sbi, page); 1706 if (seq_id) 1707 *seq_id = seq; 1708 } 1709 1710 folio_start_writeback(folio); 1711 1712 fio.old_blkaddr = ni.blk_addr; 1713 f2fs_do_write_node_page(nid, &fio); 1714 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page)); 1715 dec_page_count(sbi, F2FS_DIRTY_NODES); 1716 f2fs_up_read(&sbi->node_write); 1717 1718 if (wbc->for_reclaim) { 1719 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE); 1720 submitted = NULL; 1721 } 1722 1723 folio_unlock(folio); 1724 1725 if (unlikely(f2fs_cp_error(sbi))) { 1726 f2fs_submit_merged_write(sbi, NODE); 1727 submitted = NULL; 1728 } 1729 if (submitted) 1730 *submitted = fio.submitted; 1731 1732 if (do_balance) 1733 f2fs_balance_fs(sbi, false); 1734 return 0; 1735 1736 redirty_out: 1737 folio_redirty_for_writepage(wbc, folio); 1738 return AOP_WRITEPAGE_ACTIVATE; 1739 } 1740 1741 int f2fs_move_node_page(struct page *node_page, int gc_type) 1742 { 1743 int err = 0; 1744 1745 if (gc_type == FG_GC) { 1746 struct writeback_control wbc = { 1747 .sync_mode = WB_SYNC_ALL, 1748 .nr_to_write = 1, 1749 .for_reclaim = 0, 1750 }; 1751 1752 f2fs_wait_on_page_writeback(node_page, NODE, true, true); 1753 1754 set_page_dirty(node_page); 1755 1756 if (!clear_page_dirty_for_io(node_page)) { 1757 err = -EAGAIN; 1758 goto out_page; 1759 } 1760 1761 if (__write_node_page(node_page, false, NULL, 1762 &wbc, false, FS_GC_NODE_IO, NULL)) { 1763 err = -EAGAIN; 1764 unlock_page(node_page); 1765 } 1766 goto release_page; 1767 } else { 1768 /* set page dirty and write it */ 1769 if (!folio_test_writeback(page_folio(node_page))) 1770 set_page_dirty(node_page); 1771 } 1772 out_page: 1773 unlock_page(node_page); 1774 release_page: 1775 f2fs_put_page(node_page, 0); 1776 return err; 1777 } 1778 1779 static int f2fs_write_node_page(struct page *page, 1780 struct writeback_control *wbc) 1781 { 1782 return __write_node_page(page, false, NULL, wbc, false, 1783 FS_NODE_IO, NULL); 1784 } 1785 1786 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode, 1787 struct writeback_control *wbc, bool atomic, 1788 unsigned int *seq_id) 1789 { 1790 pgoff_t index; 1791 struct folio_batch fbatch; 1792 int ret = 0; 1793 struct page *last_page = NULL; 1794 bool marked = false; 1795 nid_t ino = inode->i_ino; 1796 int nr_folios; 1797 int nwritten = 0; 1798 1799 if (atomic) { 1800 last_page = last_fsync_dnode(sbi, ino); 1801 if (IS_ERR_OR_NULL(last_page)) 1802 return PTR_ERR_OR_ZERO(last_page); 1803 } 1804 retry: 1805 folio_batch_init(&fbatch); 1806 index = 0; 1807 1808 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1809 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1810 &fbatch))) { 1811 int i; 1812 1813 for (i = 0; i < nr_folios; i++) { 1814 struct folio *folio = fbatch.folios[i]; 1815 bool submitted = false; 1816 1817 if (unlikely(f2fs_cp_error(sbi))) { 1818 f2fs_put_page(last_page, 0); 1819 folio_batch_release(&fbatch); 1820 ret = -EIO; 1821 goto out; 1822 } 1823 1824 if (!IS_DNODE(&folio->page) || !is_cold_node(&folio->page)) 1825 continue; 1826 if (ino_of_node(&folio->page) != ino) 1827 continue; 1828 1829 folio_lock(folio); 1830 1831 if (unlikely(folio->mapping != NODE_MAPPING(sbi))) { 1832 continue_unlock: 1833 folio_unlock(folio); 1834 continue; 1835 } 1836 if (ino_of_node(&folio->page) != ino) 1837 goto continue_unlock; 1838 1839 if (!folio_test_dirty(folio) && &folio->page != last_page) { 1840 /* someone wrote it for us */ 1841 goto continue_unlock; 1842 } 1843 1844 f2fs_folio_wait_writeback(folio, NODE, true, true); 1845 1846 set_fsync_mark(&folio->page, 0); 1847 set_dentry_mark(&folio->page, 0); 1848 1849 if (!atomic || &folio->page == last_page) { 1850 set_fsync_mark(&folio->page, 1); 1851 percpu_counter_inc(&sbi->rf_node_block_count); 1852 if (IS_INODE(&folio->page)) { 1853 if (is_inode_flag_set(inode, 1854 FI_DIRTY_INODE)) 1855 f2fs_update_inode(inode, &folio->page); 1856 set_dentry_mark(&folio->page, 1857 f2fs_need_dentry_mark(sbi, ino)); 1858 } 1859 /* may be written by other thread */ 1860 if (!folio_test_dirty(folio)) 1861 folio_mark_dirty(folio); 1862 } 1863 1864 if (!folio_clear_dirty_for_io(folio)) 1865 goto continue_unlock; 1866 1867 ret = __write_node_page(&folio->page, atomic && 1868 &folio->page == last_page, 1869 &submitted, wbc, true, 1870 FS_NODE_IO, seq_id); 1871 if (ret) { 1872 folio_unlock(folio); 1873 f2fs_put_page(last_page, 0); 1874 break; 1875 } else if (submitted) { 1876 nwritten++; 1877 } 1878 1879 if (&folio->page == last_page) { 1880 f2fs_folio_put(folio, false); 1881 marked = true; 1882 break; 1883 } 1884 } 1885 folio_batch_release(&fbatch); 1886 cond_resched(); 1887 1888 if (ret || marked) 1889 break; 1890 } 1891 if (!ret && atomic && !marked) { 1892 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx", 1893 ino, page_folio(last_page)->index); 1894 lock_page(last_page); 1895 f2fs_wait_on_page_writeback(last_page, NODE, true, true); 1896 set_page_dirty(last_page); 1897 unlock_page(last_page); 1898 goto retry; 1899 } 1900 out: 1901 if (nwritten) 1902 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE); 1903 return ret ? -EIO : 0; 1904 } 1905 1906 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data) 1907 { 1908 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1909 bool clean; 1910 1911 if (inode->i_ino != ino) 1912 return 0; 1913 1914 if (!is_inode_flag_set(inode, FI_DIRTY_INODE)) 1915 return 0; 1916 1917 spin_lock(&sbi->inode_lock[DIRTY_META]); 1918 clean = list_empty(&F2FS_I(inode)->gdirty_list); 1919 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1920 1921 if (clean) 1922 return 0; 1923 1924 inode = igrab(inode); 1925 if (!inode) 1926 return 0; 1927 return 1; 1928 } 1929 1930 static bool flush_dirty_inode(struct folio *folio) 1931 { 1932 struct f2fs_sb_info *sbi = F2FS_F_SB(folio); 1933 struct inode *inode; 1934 nid_t ino = ino_of_node(&folio->page); 1935 1936 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL); 1937 if (!inode) 1938 return false; 1939 1940 f2fs_update_inode(inode, &folio->page); 1941 folio_unlock(folio); 1942 1943 iput(inode); 1944 return true; 1945 } 1946 1947 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi) 1948 { 1949 pgoff_t index = 0; 1950 struct folio_batch fbatch; 1951 int nr_folios; 1952 1953 folio_batch_init(&fbatch); 1954 1955 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1956 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1957 &fbatch))) { 1958 int i; 1959 1960 for (i = 0; i < nr_folios; i++) { 1961 struct folio *folio = fbatch.folios[i]; 1962 1963 if (!IS_INODE(&folio->page)) 1964 continue; 1965 1966 folio_lock(folio); 1967 1968 if (unlikely(folio->mapping != NODE_MAPPING(sbi))) 1969 goto unlock; 1970 if (!folio_test_dirty(folio)) 1971 goto unlock; 1972 1973 /* flush inline_data, if it's async context. */ 1974 if (page_private_inline(&folio->page)) { 1975 clear_page_private_inline(&folio->page); 1976 folio_unlock(folio); 1977 flush_inline_data(sbi, ino_of_node(&folio->page)); 1978 continue; 1979 } 1980 unlock: 1981 folio_unlock(folio); 1982 } 1983 folio_batch_release(&fbatch); 1984 cond_resched(); 1985 } 1986 } 1987 1988 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi, 1989 struct writeback_control *wbc, 1990 bool do_balance, enum iostat_type io_type) 1991 { 1992 pgoff_t index; 1993 struct folio_batch fbatch; 1994 int step = 0; 1995 int nwritten = 0; 1996 int ret = 0; 1997 int nr_folios, done = 0; 1998 1999 folio_batch_init(&fbatch); 2000 2001 next_step: 2002 index = 0; 2003 2004 while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), 2005 &index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 2006 &fbatch))) { 2007 int i; 2008 2009 for (i = 0; i < nr_folios; i++) { 2010 struct folio *folio = fbatch.folios[i]; 2011 bool submitted = false; 2012 2013 /* give a priority to WB_SYNC threads */ 2014 if (atomic_read(&sbi->wb_sync_req[NODE]) && 2015 wbc->sync_mode == WB_SYNC_NONE) { 2016 done = 1; 2017 break; 2018 } 2019 2020 /* 2021 * flushing sequence with step: 2022 * 0. indirect nodes 2023 * 1. dentry dnodes 2024 * 2. file dnodes 2025 */ 2026 if (step == 0 && IS_DNODE(&folio->page)) 2027 continue; 2028 if (step == 1 && (!IS_DNODE(&folio->page) || 2029 is_cold_node(&folio->page))) 2030 continue; 2031 if (step == 2 && (!IS_DNODE(&folio->page) || 2032 !is_cold_node(&folio->page))) 2033 continue; 2034 lock_node: 2035 if (wbc->sync_mode == WB_SYNC_ALL) 2036 folio_lock(folio); 2037 else if (!folio_trylock(folio)) 2038 continue; 2039 2040 if (unlikely(folio->mapping != NODE_MAPPING(sbi))) { 2041 continue_unlock: 2042 folio_unlock(folio); 2043 continue; 2044 } 2045 2046 if (!folio_test_dirty(folio)) { 2047 /* someone wrote it for us */ 2048 goto continue_unlock; 2049 } 2050 2051 /* flush inline_data/inode, if it's async context. */ 2052 if (!do_balance) 2053 goto write_node; 2054 2055 /* flush inline_data */ 2056 if (page_private_inline(&folio->page)) { 2057 clear_page_private_inline(&folio->page); 2058 folio_unlock(folio); 2059 flush_inline_data(sbi, ino_of_node(&folio->page)); 2060 goto lock_node; 2061 } 2062 2063 /* flush dirty inode */ 2064 if (IS_INODE(&folio->page) && flush_dirty_inode(folio)) 2065 goto lock_node; 2066 write_node: 2067 f2fs_folio_wait_writeback(folio, NODE, true, true); 2068 2069 if (!folio_clear_dirty_for_io(folio)) 2070 goto continue_unlock; 2071 2072 set_fsync_mark(&folio->page, 0); 2073 set_dentry_mark(&folio->page, 0); 2074 2075 ret = __write_node_page(&folio->page, false, &submitted, 2076 wbc, do_balance, io_type, NULL); 2077 if (ret) 2078 folio_unlock(folio); 2079 else if (submitted) 2080 nwritten++; 2081 2082 if (--wbc->nr_to_write == 0) 2083 break; 2084 } 2085 folio_batch_release(&fbatch); 2086 cond_resched(); 2087 2088 if (wbc->nr_to_write == 0) { 2089 step = 2; 2090 break; 2091 } 2092 } 2093 2094 if (step < 2) { 2095 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2096 wbc->sync_mode == WB_SYNC_NONE && step == 1) 2097 goto out; 2098 step++; 2099 goto next_step; 2100 } 2101 out: 2102 if (nwritten) 2103 f2fs_submit_merged_write(sbi, NODE); 2104 2105 if (unlikely(f2fs_cp_error(sbi))) 2106 return -EIO; 2107 return ret; 2108 } 2109 2110 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, 2111 unsigned int seq_id) 2112 { 2113 struct fsync_node_entry *fn; 2114 struct page *page; 2115 struct list_head *head = &sbi->fsync_node_list; 2116 unsigned long flags; 2117 unsigned int cur_seq_id = 0; 2118 2119 while (seq_id && cur_seq_id < seq_id) { 2120 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 2121 if (list_empty(head)) { 2122 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2123 break; 2124 } 2125 fn = list_first_entry(head, struct fsync_node_entry, list); 2126 if (fn->seq_id > seq_id) { 2127 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2128 break; 2129 } 2130 cur_seq_id = fn->seq_id; 2131 page = fn->page; 2132 get_page(page); 2133 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2134 2135 f2fs_wait_on_page_writeback(page, NODE, true, false); 2136 2137 put_page(page); 2138 } 2139 2140 return filemap_check_errors(NODE_MAPPING(sbi)); 2141 } 2142 2143 static int f2fs_write_node_pages(struct address_space *mapping, 2144 struct writeback_control *wbc) 2145 { 2146 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2147 struct blk_plug plug; 2148 long diff; 2149 2150 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2151 goto skip_write; 2152 2153 /* balancing f2fs's metadata in background */ 2154 f2fs_balance_fs_bg(sbi, true); 2155 2156 /* collect a number of dirty node pages and write together */ 2157 if (wbc->sync_mode != WB_SYNC_ALL && 2158 get_pages(sbi, F2FS_DIRTY_NODES) < 2159 nr_pages_to_skip(sbi, NODE)) 2160 goto skip_write; 2161 2162 if (wbc->sync_mode == WB_SYNC_ALL) 2163 atomic_inc(&sbi->wb_sync_req[NODE]); 2164 else if (atomic_read(&sbi->wb_sync_req[NODE])) { 2165 /* to avoid potential deadlock */ 2166 if (current->plug) 2167 blk_finish_plug(current->plug); 2168 goto skip_write; 2169 } 2170 2171 trace_f2fs_writepages(mapping->host, wbc, NODE); 2172 2173 diff = nr_pages_to_write(sbi, NODE, wbc); 2174 blk_start_plug(&plug); 2175 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO); 2176 blk_finish_plug(&plug); 2177 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); 2178 2179 if (wbc->sync_mode == WB_SYNC_ALL) 2180 atomic_dec(&sbi->wb_sync_req[NODE]); 2181 return 0; 2182 2183 skip_write: 2184 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES); 2185 trace_f2fs_writepages(mapping->host, wbc, NODE); 2186 return 0; 2187 } 2188 2189 static bool f2fs_dirty_node_folio(struct address_space *mapping, 2190 struct folio *folio) 2191 { 2192 trace_f2fs_set_page_dirty(folio, NODE); 2193 2194 if (!folio_test_uptodate(folio)) 2195 folio_mark_uptodate(folio); 2196 #ifdef CONFIG_F2FS_CHECK_FS 2197 if (IS_INODE(&folio->page)) 2198 f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page); 2199 #endif 2200 if (filemap_dirty_folio(mapping, folio)) { 2201 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES); 2202 set_page_private_reference(&folio->page); 2203 return true; 2204 } 2205 return false; 2206 } 2207 2208 /* 2209 * Structure of the f2fs node operations 2210 */ 2211 const struct address_space_operations f2fs_node_aops = { 2212 .writepage = f2fs_write_node_page, 2213 .writepages = f2fs_write_node_pages, 2214 .dirty_folio = f2fs_dirty_node_folio, 2215 .invalidate_folio = f2fs_invalidate_folio, 2216 .release_folio = f2fs_release_folio, 2217 .migrate_folio = filemap_migrate_folio, 2218 }; 2219 2220 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i, 2221 nid_t n) 2222 { 2223 return radix_tree_lookup(&nm_i->free_nid_root, n); 2224 } 2225 2226 static int __insert_free_nid(struct f2fs_sb_info *sbi, 2227 struct free_nid *i) 2228 { 2229 struct f2fs_nm_info *nm_i = NM_I(sbi); 2230 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i); 2231 2232 if (err) 2233 return err; 2234 2235 nm_i->nid_cnt[FREE_NID]++; 2236 list_add_tail(&i->list, &nm_i->free_nid_list); 2237 return 0; 2238 } 2239 2240 static void __remove_free_nid(struct f2fs_sb_info *sbi, 2241 struct free_nid *i, enum nid_state state) 2242 { 2243 struct f2fs_nm_info *nm_i = NM_I(sbi); 2244 2245 f2fs_bug_on(sbi, state != i->state); 2246 nm_i->nid_cnt[state]--; 2247 if (state == FREE_NID) 2248 list_del(&i->list); 2249 radix_tree_delete(&nm_i->free_nid_root, i->nid); 2250 } 2251 2252 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i, 2253 enum nid_state org_state, enum nid_state dst_state) 2254 { 2255 struct f2fs_nm_info *nm_i = NM_I(sbi); 2256 2257 f2fs_bug_on(sbi, org_state != i->state); 2258 i->state = dst_state; 2259 nm_i->nid_cnt[org_state]--; 2260 nm_i->nid_cnt[dst_state]++; 2261 2262 switch (dst_state) { 2263 case PREALLOC_NID: 2264 list_del(&i->list); 2265 break; 2266 case FREE_NID: 2267 list_add_tail(&i->list, &nm_i->free_nid_list); 2268 break; 2269 default: 2270 BUG_ON(1); 2271 } 2272 } 2273 2274 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi) 2275 { 2276 struct f2fs_nm_info *nm_i = NM_I(sbi); 2277 unsigned int i; 2278 bool ret = true; 2279 2280 f2fs_down_read(&nm_i->nat_tree_lock); 2281 for (i = 0; i < nm_i->nat_blocks; i++) { 2282 if (!test_bit_le(i, nm_i->nat_block_bitmap)) { 2283 ret = false; 2284 break; 2285 } 2286 } 2287 f2fs_up_read(&nm_i->nat_tree_lock); 2288 2289 return ret; 2290 } 2291 2292 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid, 2293 bool set, bool build) 2294 { 2295 struct f2fs_nm_info *nm_i = NM_I(sbi); 2296 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid); 2297 unsigned int nid_ofs = nid - START_NID(nid); 2298 2299 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap)) 2300 return; 2301 2302 if (set) { 2303 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2304 return; 2305 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2306 nm_i->free_nid_count[nat_ofs]++; 2307 } else { 2308 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2309 return; 2310 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2311 if (!build) 2312 nm_i->free_nid_count[nat_ofs]--; 2313 } 2314 } 2315 2316 /* return if the nid is recognized as free */ 2317 static bool add_free_nid(struct f2fs_sb_info *sbi, 2318 nid_t nid, bool build, bool update) 2319 { 2320 struct f2fs_nm_info *nm_i = NM_I(sbi); 2321 struct free_nid *i, *e; 2322 struct nat_entry *ne; 2323 int err = -EINVAL; 2324 bool ret = false; 2325 2326 /* 0 nid should not be used */ 2327 if (unlikely(nid == 0)) 2328 return false; 2329 2330 if (unlikely(f2fs_check_nid_range(sbi, nid))) 2331 return false; 2332 2333 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL); 2334 i->nid = nid; 2335 i->state = FREE_NID; 2336 2337 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 2338 2339 spin_lock(&nm_i->nid_list_lock); 2340 2341 if (build) { 2342 /* 2343 * Thread A Thread B 2344 * - f2fs_create 2345 * - f2fs_new_inode 2346 * - f2fs_alloc_nid 2347 * - __insert_nid_to_list(PREALLOC_NID) 2348 * - f2fs_balance_fs_bg 2349 * - f2fs_build_free_nids 2350 * - __f2fs_build_free_nids 2351 * - scan_nat_page 2352 * - add_free_nid 2353 * - __lookup_nat_cache 2354 * - f2fs_add_link 2355 * - f2fs_init_inode_metadata 2356 * - f2fs_new_inode_page 2357 * - f2fs_new_node_page 2358 * - set_node_addr 2359 * - f2fs_alloc_nid_done 2360 * - __remove_nid_from_list(PREALLOC_NID) 2361 * - __insert_nid_to_list(FREE_NID) 2362 */ 2363 ne = __lookup_nat_cache(nm_i, nid); 2364 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) || 2365 nat_get_blkaddr(ne) != NULL_ADDR)) 2366 goto err_out; 2367 2368 e = __lookup_free_nid_list(nm_i, nid); 2369 if (e) { 2370 if (e->state == FREE_NID) 2371 ret = true; 2372 goto err_out; 2373 } 2374 } 2375 ret = true; 2376 err = __insert_free_nid(sbi, i); 2377 err_out: 2378 if (update) { 2379 update_free_nid_bitmap(sbi, nid, ret, build); 2380 if (!build) 2381 nm_i->available_nids++; 2382 } 2383 spin_unlock(&nm_i->nid_list_lock); 2384 radix_tree_preload_end(); 2385 2386 if (err) 2387 kmem_cache_free(free_nid_slab, i); 2388 return ret; 2389 } 2390 2391 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid) 2392 { 2393 struct f2fs_nm_info *nm_i = NM_I(sbi); 2394 struct free_nid *i; 2395 bool need_free = false; 2396 2397 spin_lock(&nm_i->nid_list_lock); 2398 i = __lookup_free_nid_list(nm_i, nid); 2399 if (i && i->state == FREE_NID) { 2400 __remove_free_nid(sbi, i, FREE_NID); 2401 need_free = true; 2402 } 2403 spin_unlock(&nm_i->nid_list_lock); 2404 2405 if (need_free) 2406 kmem_cache_free(free_nid_slab, i); 2407 } 2408 2409 static int scan_nat_page(struct f2fs_sb_info *sbi, 2410 struct page *nat_page, nid_t start_nid) 2411 { 2412 struct f2fs_nm_info *nm_i = NM_I(sbi); 2413 struct f2fs_nat_block *nat_blk = page_address(nat_page); 2414 block_t blk_addr; 2415 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid); 2416 int i; 2417 2418 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap); 2419 2420 i = start_nid % NAT_ENTRY_PER_BLOCK; 2421 2422 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) { 2423 if (unlikely(start_nid >= nm_i->max_nid)) 2424 break; 2425 2426 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr); 2427 2428 if (blk_addr == NEW_ADDR) 2429 return -EFSCORRUPTED; 2430 2431 if (blk_addr == NULL_ADDR) { 2432 add_free_nid(sbi, start_nid, true, true); 2433 } else { 2434 spin_lock(&NM_I(sbi)->nid_list_lock); 2435 update_free_nid_bitmap(sbi, start_nid, false, true); 2436 spin_unlock(&NM_I(sbi)->nid_list_lock); 2437 } 2438 } 2439 2440 return 0; 2441 } 2442 2443 static void scan_curseg_cache(struct f2fs_sb_info *sbi) 2444 { 2445 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2446 struct f2fs_journal *journal = curseg->journal; 2447 int i; 2448 2449 down_read(&curseg->journal_rwsem); 2450 for (i = 0; i < nats_in_cursum(journal); i++) { 2451 block_t addr; 2452 nid_t nid; 2453 2454 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr); 2455 nid = le32_to_cpu(nid_in_journal(journal, i)); 2456 if (addr == NULL_ADDR) 2457 add_free_nid(sbi, nid, true, false); 2458 else 2459 remove_free_nid(sbi, nid); 2460 } 2461 up_read(&curseg->journal_rwsem); 2462 } 2463 2464 static void scan_free_nid_bits(struct f2fs_sb_info *sbi) 2465 { 2466 struct f2fs_nm_info *nm_i = NM_I(sbi); 2467 unsigned int i, idx; 2468 nid_t nid; 2469 2470 f2fs_down_read(&nm_i->nat_tree_lock); 2471 2472 for (i = 0; i < nm_i->nat_blocks; i++) { 2473 if (!test_bit_le(i, nm_i->nat_block_bitmap)) 2474 continue; 2475 if (!nm_i->free_nid_count[i]) 2476 continue; 2477 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) { 2478 idx = find_next_bit_le(nm_i->free_nid_bitmap[i], 2479 NAT_ENTRY_PER_BLOCK, idx); 2480 if (idx >= NAT_ENTRY_PER_BLOCK) 2481 break; 2482 2483 nid = i * NAT_ENTRY_PER_BLOCK + idx; 2484 add_free_nid(sbi, nid, true, false); 2485 2486 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS) 2487 goto out; 2488 } 2489 } 2490 out: 2491 scan_curseg_cache(sbi); 2492 2493 f2fs_up_read(&nm_i->nat_tree_lock); 2494 } 2495 2496 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi, 2497 bool sync, bool mount) 2498 { 2499 struct f2fs_nm_info *nm_i = NM_I(sbi); 2500 int i = 0, ret; 2501 nid_t nid = nm_i->next_scan_nid; 2502 2503 if (unlikely(nid >= nm_i->max_nid)) 2504 nid = 0; 2505 2506 if (unlikely(nid % NAT_ENTRY_PER_BLOCK)) 2507 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK; 2508 2509 /* Enough entries */ 2510 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2511 return 0; 2512 2513 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS)) 2514 return 0; 2515 2516 if (!mount) { 2517 /* try to find free nids in free_nid_bitmap */ 2518 scan_free_nid_bits(sbi); 2519 2520 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2521 return 0; 2522 } 2523 2524 /* readahead nat pages to be scanned */ 2525 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, 2526 META_NAT, true); 2527 2528 f2fs_down_read(&nm_i->nat_tree_lock); 2529 2530 while (1) { 2531 if (!test_bit_le(NAT_BLOCK_OFFSET(nid), 2532 nm_i->nat_block_bitmap)) { 2533 struct page *page = get_current_nat_page(sbi, nid); 2534 2535 if (IS_ERR(page)) { 2536 ret = PTR_ERR(page); 2537 } else { 2538 ret = scan_nat_page(sbi, page, nid); 2539 f2fs_put_page(page, 1); 2540 } 2541 2542 if (ret) { 2543 f2fs_up_read(&nm_i->nat_tree_lock); 2544 2545 if (ret == -EFSCORRUPTED) { 2546 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it"); 2547 set_sbi_flag(sbi, SBI_NEED_FSCK); 2548 f2fs_handle_error(sbi, 2549 ERROR_INCONSISTENT_NAT); 2550 } 2551 2552 return ret; 2553 } 2554 } 2555 2556 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK)); 2557 if (unlikely(nid >= nm_i->max_nid)) 2558 nid = 0; 2559 2560 if (++i >= FREE_NID_PAGES) 2561 break; 2562 } 2563 2564 /* go to the next free nat pages to find free nids abundantly */ 2565 nm_i->next_scan_nid = nid; 2566 2567 /* find free nids from current sum_pages */ 2568 scan_curseg_cache(sbi); 2569 2570 f2fs_up_read(&nm_i->nat_tree_lock); 2571 2572 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid), 2573 nm_i->ra_nid_pages, META_NAT, false); 2574 2575 return 0; 2576 } 2577 2578 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount) 2579 { 2580 int ret; 2581 2582 mutex_lock(&NM_I(sbi)->build_lock); 2583 ret = __f2fs_build_free_nids(sbi, sync, mount); 2584 mutex_unlock(&NM_I(sbi)->build_lock); 2585 2586 return ret; 2587 } 2588 2589 /* 2590 * If this function returns success, caller can obtain a new nid 2591 * from second parameter of this function. 2592 * The returned nid could be used ino as well as nid when inode is created. 2593 */ 2594 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid) 2595 { 2596 struct f2fs_nm_info *nm_i = NM_I(sbi); 2597 struct free_nid *i = NULL; 2598 retry: 2599 if (time_to_inject(sbi, FAULT_ALLOC_NID)) 2600 return false; 2601 2602 spin_lock(&nm_i->nid_list_lock); 2603 2604 if (unlikely(nm_i->available_nids == 0)) { 2605 spin_unlock(&nm_i->nid_list_lock); 2606 return false; 2607 } 2608 2609 /* We should not use stale free nids created by f2fs_build_free_nids */ 2610 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) { 2611 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list)); 2612 i = list_first_entry(&nm_i->free_nid_list, 2613 struct free_nid, list); 2614 *nid = i->nid; 2615 2616 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID); 2617 nm_i->available_nids--; 2618 2619 update_free_nid_bitmap(sbi, *nid, false, false); 2620 2621 spin_unlock(&nm_i->nid_list_lock); 2622 return true; 2623 } 2624 spin_unlock(&nm_i->nid_list_lock); 2625 2626 /* Let's scan nat pages and its caches to get free nids */ 2627 if (!f2fs_build_free_nids(sbi, true, false)) 2628 goto retry; 2629 return false; 2630 } 2631 2632 /* 2633 * f2fs_alloc_nid() should be called prior to this function. 2634 */ 2635 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 2636 { 2637 struct f2fs_nm_info *nm_i = NM_I(sbi); 2638 struct free_nid *i; 2639 2640 spin_lock(&nm_i->nid_list_lock); 2641 i = __lookup_free_nid_list(nm_i, nid); 2642 f2fs_bug_on(sbi, !i); 2643 __remove_free_nid(sbi, i, PREALLOC_NID); 2644 spin_unlock(&nm_i->nid_list_lock); 2645 2646 kmem_cache_free(free_nid_slab, i); 2647 } 2648 2649 /* 2650 * f2fs_alloc_nid() should be called prior to this function. 2651 */ 2652 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 2653 { 2654 struct f2fs_nm_info *nm_i = NM_I(sbi); 2655 struct free_nid *i; 2656 bool need_free = false; 2657 2658 if (!nid) 2659 return; 2660 2661 spin_lock(&nm_i->nid_list_lock); 2662 i = __lookup_free_nid_list(nm_i, nid); 2663 f2fs_bug_on(sbi, !i); 2664 2665 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) { 2666 __remove_free_nid(sbi, i, PREALLOC_NID); 2667 need_free = true; 2668 } else { 2669 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID); 2670 } 2671 2672 nm_i->available_nids++; 2673 2674 update_free_nid_bitmap(sbi, nid, true, false); 2675 2676 spin_unlock(&nm_i->nid_list_lock); 2677 2678 if (need_free) 2679 kmem_cache_free(free_nid_slab, i); 2680 } 2681 2682 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink) 2683 { 2684 struct f2fs_nm_info *nm_i = NM_I(sbi); 2685 int nr = nr_shrink; 2686 2687 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2688 return 0; 2689 2690 if (!mutex_trylock(&nm_i->build_lock)) 2691 return 0; 2692 2693 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) { 2694 struct free_nid *i, *next; 2695 unsigned int batch = SHRINK_NID_BATCH_SIZE; 2696 2697 spin_lock(&nm_i->nid_list_lock); 2698 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) { 2699 if (!nr_shrink || !batch || 2700 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2701 break; 2702 __remove_free_nid(sbi, i, FREE_NID); 2703 kmem_cache_free(free_nid_slab, i); 2704 nr_shrink--; 2705 batch--; 2706 } 2707 spin_unlock(&nm_i->nid_list_lock); 2708 } 2709 2710 mutex_unlock(&nm_i->build_lock); 2711 2712 return nr - nr_shrink; 2713 } 2714 2715 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page) 2716 { 2717 void *src_addr, *dst_addr; 2718 size_t inline_size; 2719 struct page *ipage; 2720 struct f2fs_inode *ri; 2721 2722 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 2723 if (IS_ERR(ipage)) 2724 return PTR_ERR(ipage); 2725 2726 ri = F2FS_INODE(page); 2727 if (ri->i_inline & F2FS_INLINE_XATTR) { 2728 if (!f2fs_has_inline_xattr(inode)) { 2729 set_inode_flag(inode, FI_INLINE_XATTR); 2730 stat_inc_inline_xattr(inode); 2731 } 2732 } else { 2733 if (f2fs_has_inline_xattr(inode)) { 2734 stat_dec_inline_xattr(inode); 2735 clear_inode_flag(inode, FI_INLINE_XATTR); 2736 } 2737 goto update_inode; 2738 } 2739 2740 dst_addr = inline_xattr_addr(inode, ipage); 2741 src_addr = inline_xattr_addr(inode, page); 2742 inline_size = inline_xattr_size(inode); 2743 2744 f2fs_wait_on_page_writeback(ipage, NODE, true, true); 2745 memcpy(dst_addr, src_addr, inline_size); 2746 update_inode: 2747 f2fs_update_inode(inode, ipage); 2748 f2fs_put_page(ipage, 1); 2749 return 0; 2750 } 2751 2752 int f2fs_recover_xattr_data(struct inode *inode, struct page *page) 2753 { 2754 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2755 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid; 2756 nid_t new_xnid; 2757 struct dnode_of_data dn; 2758 struct node_info ni; 2759 struct page *xpage; 2760 int err; 2761 2762 if (!prev_xnid) 2763 goto recover_xnid; 2764 2765 /* 1: invalidate the previous xattr nid */ 2766 err = f2fs_get_node_info(sbi, prev_xnid, &ni, false); 2767 if (err) 2768 return err; 2769 2770 f2fs_invalidate_blocks(sbi, ni.blk_addr, 1); 2771 dec_valid_node_count(sbi, inode, false); 2772 set_node_addr(sbi, &ni, NULL_ADDR, false); 2773 2774 recover_xnid: 2775 /* 2: update xattr nid in inode */ 2776 if (!f2fs_alloc_nid(sbi, &new_xnid)) 2777 return -ENOSPC; 2778 2779 set_new_dnode(&dn, inode, NULL, NULL, new_xnid); 2780 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET); 2781 if (IS_ERR(xpage)) { 2782 f2fs_alloc_nid_failed(sbi, new_xnid); 2783 return PTR_ERR(xpage); 2784 } 2785 2786 f2fs_alloc_nid_done(sbi, new_xnid); 2787 f2fs_update_inode_page(inode); 2788 2789 /* 3: update and set xattr node page dirty */ 2790 if (page) { 2791 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), 2792 VALID_XATTR_BLOCK_SIZE); 2793 set_page_dirty(xpage); 2794 } 2795 f2fs_put_page(xpage, 1); 2796 2797 return 0; 2798 } 2799 2800 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 2801 { 2802 struct f2fs_inode *src, *dst; 2803 nid_t ino = ino_of_node(page); 2804 struct node_info old_ni, new_ni; 2805 struct page *ipage; 2806 int err; 2807 2808 err = f2fs_get_node_info(sbi, ino, &old_ni, false); 2809 if (err) 2810 return err; 2811 2812 if (unlikely(old_ni.blk_addr != NULL_ADDR)) 2813 return -EINVAL; 2814 retry: 2815 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false); 2816 if (!ipage) { 2817 memalloc_retry_wait(GFP_NOFS); 2818 goto retry; 2819 } 2820 2821 /* Should not use this inode from free nid list */ 2822 remove_free_nid(sbi, ino); 2823 2824 if (!PageUptodate(ipage)) 2825 SetPageUptodate(ipage); 2826 fill_node_footer(ipage, ino, ino, 0, true); 2827 set_cold_node(ipage, false); 2828 2829 src = F2FS_INODE(page); 2830 dst = F2FS_INODE(ipage); 2831 2832 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext)); 2833 dst->i_size = 0; 2834 dst->i_blocks = cpu_to_le64(1); 2835 dst->i_links = cpu_to_le32(1); 2836 dst->i_xattr_nid = 0; 2837 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR); 2838 if (dst->i_inline & F2FS_EXTRA_ATTR) { 2839 dst->i_extra_isize = src->i_extra_isize; 2840 2841 if (f2fs_sb_has_flexible_inline_xattr(sbi) && 2842 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2843 i_inline_xattr_size)) 2844 dst->i_inline_xattr_size = src->i_inline_xattr_size; 2845 2846 if (f2fs_sb_has_project_quota(sbi) && 2847 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2848 i_projid)) 2849 dst->i_projid = src->i_projid; 2850 2851 if (f2fs_sb_has_inode_crtime(sbi) && 2852 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2853 i_crtime_nsec)) { 2854 dst->i_crtime = src->i_crtime; 2855 dst->i_crtime_nsec = src->i_crtime_nsec; 2856 } 2857 } 2858 2859 new_ni = old_ni; 2860 new_ni.ino = ino; 2861 2862 if (unlikely(inc_valid_node_count(sbi, NULL, true))) 2863 WARN_ON(1); 2864 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 2865 inc_valid_inode_count(sbi); 2866 set_page_dirty(ipage); 2867 f2fs_put_page(ipage, 1); 2868 return 0; 2869 } 2870 2871 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi, 2872 unsigned int segno, struct f2fs_summary_block *sum) 2873 { 2874 struct f2fs_node *rn; 2875 struct f2fs_summary *sum_entry; 2876 block_t addr; 2877 int i, idx, last_offset, nrpages; 2878 2879 /* scan the node segment */ 2880 last_offset = BLKS_PER_SEG(sbi); 2881 addr = START_BLOCK(sbi, segno); 2882 sum_entry = &sum->entries[0]; 2883 2884 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) { 2885 nrpages = bio_max_segs(last_offset - i); 2886 2887 /* readahead node pages */ 2888 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true); 2889 2890 for (idx = addr; idx < addr + nrpages; idx++) { 2891 struct page *page = f2fs_get_tmp_page(sbi, idx); 2892 2893 if (IS_ERR(page)) 2894 return PTR_ERR(page); 2895 2896 rn = F2FS_NODE(page); 2897 sum_entry->nid = rn->footer.nid; 2898 sum_entry->version = 0; 2899 sum_entry->ofs_in_node = 0; 2900 sum_entry++; 2901 f2fs_put_page(page, 1); 2902 } 2903 2904 invalidate_mapping_pages(META_MAPPING(sbi), addr, 2905 addr + nrpages); 2906 } 2907 return 0; 2908 } 2909 2910 static void remove_nats_in_journal(struct f2fs_sb_info *sbi) 2911 { 2912 struct f2fs_nm_info *nm_i = NM_I(sbi); 2913 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2914 struct f2fs_journal *journal = curseg->journal; 2915 int i; 2916 2917 down_write(&curseg->journal_rwsem); 2918 for (i = 0; i < nats_in_cursum(journal); i++) { 2919 struct nat_entry *ne; 2920 struct f2fs_nat_entry raw_ne; 2921 nid_t nid = le32_to_cpu(nid_in_journal(journal, i)); 2922 2923 if (f2fs_check_nid_range(sbi, nid)) 2924 continue; 2925 2926 raw_ne = nat_in_journal(journal, i); 2927 2928 ne = __lookup_nat_cache(nm_i, nid); 2929 if (!ne) { 2930 ne = __alloc_nat_entry(sbi, nid, true); 2931 __init_nat_entry(nm_i, ne, &raw_ne, true); 2932 } 2933 2934 /* 2935 * if a free nat in journal has not been used after last 2936 * checkpoint, we should remove it from available nids, 2937 * since later we will add it again. 2938 */ 2939 if (!get_nat_flag(ne, IS_DIRTY) && 2940 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) { 2941 spin_lock(&nm_i->nid_list_lock); 2942 nm_i->available_nids--; 2943 spin_unlock(&nm_i->nid_list_lock); 2944 } 2945 2946 __set_nat_cache_dirty(nm_i, ne); 2947 } 2948 update_nats_in_cursum(journal, -i); 2949 up_write(&curseg->journal_rwsem); 2950 } 2951 2952 static void __adjust_nat_entry_set(struct nat_entry_set *nes, 2953 struct list_head *head, int max) 2954 { 2955 struct nat_entry_set *cur; 2956 2957 if (nes->entry_cnt >= max) 2958 goto add_out; 2959 2960 list_for_each_entry(cur, head, set_list) { 2961 if (cur->entry_cnt >= nes->entry_cnt) { 2962 list_add(&nes->set_list, cur->set_list.prev); 2963 return; 2964 } 2965 } 2966 add_out: 2967 list_add_tail(&nes->set_list, head); 2968 } 2969 2970 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs, 2971 unsigned int valid) 2972 { 2973 if (valid == 0) { 2974 __set_bit_le(nat_ofs, nm_i->empty_nat_bits); 2975 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2976 return; 2977 } 2978 2979 __clear_bit_le(nat_ofs, nm_i->empty_nat_bits); 2980 if (valid == NAT_ENTRY_PER_BLOCK) 2981 __set_bit_le(nat_ofs, nm_i->full_nat_bits); 2982 else 2983 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2984 } 2985 2986 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid, 2987 struct page *page) 2988 { 2989 struct f2fs_nm_info *nm_i = NM_I(sbi); 2990 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK; 2991 struct f2fs_nat_block *nat_blk = page_address(page); 2992 int valid = 0; 2993 int i = 0; 2994 2995 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 2996 return; 2997 2998 if (nat_index == 0) { 2999 valid = 1; 3000 i = 1; 3001 } 3002 for (; i < NAT_ENTRY_PER_BLOCK; i++) { 3003 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR) 3004 valid++; 3005 } 3006 3007 __update_nat_bits(nm_i, nat_index, valid); 3008 } 3009 3010 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi) 3011 { 3012 struct f2fs_nm_info *nm_i = NM_I(sbi); 3013 unsigned int nat_ofs; 3014 3015 f2fs_down_read(&nm_i->nat_tree_lock); 3016 3017 for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) { 3018 unsigned int valid = 0, nid_ofs = 0; 3019 3020 /* handle nid zero due to it should never be used */ 3021 if (unlikely(nat_ofs == 0)) { 3022 valid = 1; 3023 nid_ofs = 1; 3024 } 3025 3026 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) { 3027 if (!test_bit_le(nid_ofs, 3028 nm_i->free_nid_bitmap[nat_ofs])) 3029 valid++; 3030 } 3031 3032 __update_nat_bits(nm_i, nat_ofs, valid); 3033 } 3034 3035 f2fs_up_read(&nm_i->nat_tree_lock); 3036 } 3037 3038 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi, 3039 struct nat_entry_set *set, struct cp_control *cpc) 3040 { 3041 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 3042 struct f2fs_journal *journal = curseg->journal; 3043 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK; 3044 bool to_journal = true; 3045 struct f2fs_nat_block *nat_blk; 3046 struct nat_entry *ne, *cur; 3047 struct page *page = NULL; 3048 3049 /* 3050 * there are two steps to flush nat entries: 3051 * #1, flush nat entries to journal in current hot data summary block. 3052 * #2, flush nat entries to nat page. 3053 */ 3054 if ((cpc->reason & CP_UMOUNT) || 3055 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL)) 3056 to_journal = false; 3057 3058 if (to_journal) { 3059 down_write(&curseg->journal_rwsem); 3060 } else { 3061 page = get_next_nat_page(sbi, start_nid); 3062 if (IS_ERR(page)) 3063 return PTR_ERR(page); 3064 3065 nat_blk = page_address(page); 3066 f2fs_bug_on(sbi, !nat_blk); 3067 } 3068 3069 /* flush dirty nats in nat entry set */ 3070 list_for_each_entry_safe(ne, cur, &set->entry_list, list) { 3071 struct f2fs_nat_entry *raw_ne; 3072 nid_t nid = nat_get_nid(ne); 3073 int offset; 3074 3075 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR); 3076 3077 if (to_journal) { 3078 offset = f2fs_lookup_journal_in_cursum(journal, 3079 NAT_JOURNAL, nid, 1); 3080 f2fs_bug_on(sbi, offset < 0); 3081 raw_ne = &nat_in_journal(journal, offset); 3082 nid_in_journal(journal, offset) = cpu_to_le32(nid); 3083 } else { 3084 raw_ne = &nat_blk->entries[nid - start_nid]; 3085 } 3086 raw_nat_from_node_info(raw_ne, &ne->ni); 3087 nat_reset_flag(ne); 3088 __clear_nat_cache_dirty(NM_I(sbi), set, ne); 3089 if (nat_get_blkaddr(ne) == NULL_ADDR) { 3090 add_free_nid(sbi, nid, false, true); 3091 } else { 3092 spin_lock(&NM_I(sbi)->nid_list_lock); 3093 update_free_nid_bitmap(sbi, nid, false, false); 3094 spin_unlock(&NM_I(sbi)->nid_list_lock); 3095 } 3096 } 3097 3098 if (to_journal) { 3099 up_write(&curseg->journal_rwsem); 3100 } else { 3101 update_nat_bits(sbi, start_nid, page); 3102 f2fs_put_page(page, 1); 3103 } 3104 3105 /* Allow dirty nats by node block allocation in write_begin */ 3106 if (!set->entry_cnt) { 3107 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set); 3108 kmem_cache_free(nat_entry_set_slab, set); 3109 } 3110 return 0; 3111 } 3112 3113 /* 3114 * This function is called during the checkpointing process. 3115 */ 3116 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3117 { 3118 struct f2fs_nm_info *nm_i = NM_I(sbi); 3119 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 3120 struct f2fs_journal *journal = curseg->journal; 3121 struct nat_entry_set *setvec[NAT_VEC_SIZE]; 3122 struct nat_entry_set *set, *tmp; 3123 unsigned int found; 3124 nid_t set_idx = 0; 3125 LIST_HEAD(sets); 3126 int err = 0; 3127 3128 /* 3129 * during unmount, let's flush nat_bits before checking 3130 * nat_cnt[DIRTY_NAT]. 3131 */ 3132 if (cpc->reason & CP_UMOUNT) { 3133 f2fs_down_write(&nm_i->nat_tree_lock); 3134 remove_nats_in_journal(sbi); 3135 f2fs_up_write(&nm_i->nat_tree_lock); 3136 } 3137 3138 if (!nm_i->nat_cnt[DIRTY_NAT]) 3139 return 0; 3140 3141 f2fs_down_write(&nm_i->nat_tree_lock); 3142 3143 /* 3144 * if there are no enough space in journal to store dirty nat 3145 * entries, remove all entries from journal and merge them 3146 * into nat entry set. 3147 */ 3148 if (cpc->reason & CP_UMOUNT || 3149 !__has_cursum_space(journal, 3150 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL)) 3151 remove_nats_in_journal(sbi); 3152 3153 while ((found = __gang_lookup_nat_set(nm_i, 3154 set_idx, NAT_VEC_SIZE, setvec))) { 3155 unsigned idx; 3156 3157 set_idx = setvec[found - 1]->set + 1; 3158 for (idx = 0; idx < found; idx++) 3159 __adjust_nat_entry_set(setvec[idx], &sets, 3160 MAX_NAT_JENTRIES(journal)); 3161 } 3162 3163 /* flush dirty nats in nat entry set */ 3164 list_for_each_entry_safe(set, tmp, &sets, set_list) { 3165 err = __flush_nat_entry_set(sbi, set, cpc); 3166 if (err) 3167 break; 3168 } 3169 3170 f2fs_up_write(&nm_i->nat_tree_lock); 3171 /* Allow dirty nats by node block allocation in write_begin */ 3172 3173 return err; 3174 } 3175 3176 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi) 3177 { 3178 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3179 struct f2fs_nm_info *nm_i = NM_I(sbi); 3180 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE; 3181 unsigned int i; 3182 __u64 cp_ver = cur_cp_version(ckpt); 3183 block_t nat_bits_addr; 3184 3185 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8); 3186 nm_i->nat_bits = f2fs_kvzalloc(sbi, 3187 F2FS_BLK_TO_BYTES(nm_i->nat_bits_blocks), GFP_KERNEL); 3188 if (!nm_i->nat_bits) 3189 return -ENOMEM; 3190 3191 nm_i->full_nat_bits = nm_i->nat_bits + 8; 3192 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes; 3193 3194 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3195 return 0; 3196 3197 nat_bits_addr = __start_cp_addr(sbi) + BLKS_PER_SEG(sbi) - 3198 nm_i->nat_bits_blocks; 3199 for (i = 0; i < nm_i->nat_bits_blocks; i++) { 3200 struct page *page; 3201 3202 page = f2fs_get_meta_page(sbi, nat_bits_addr++); 3203 if (IS_ERR(page)) 3204 return PTR_ERR(page); 3205 3206 memcpy(nm_i->nat_bits + F2FS_BLK_TO_BYTES(i), 3207 page_address(page), F2FS_BLKSIZE); 3208 f2fs_put_page(page, 1); 3209 } 3210 3211 cp_ver |= (cur_cp_crc(ckpt) << 32); 3212 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) { 3213 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG); 3214 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)", 3215 cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits)); 3216 return 0; 3217 } 3218 3219 f2fs_notice(sbi, "Found nat_bits in checkpoint"); 3220 return 0; 3221 } 3222 3223 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi) 3224 { 3225 struct f2fs_nm_info *nm_i = NM_I(sbi); 3226 unsigned int i = 0; 3227 nid_t nid, last_nid; 3228 3229 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3230 return; 3231 3232 for (i = 0; i < nm_i->nat_blocks; i++) { 3233 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i); 3234 if (i >= nm_i->nat_blocks) 3235 break; 3236 3237 __set_bit_le(i, nm_i->nat_block_bitmap); 3238 3239 nid = i * NAT_ENTRY_PER_BLOCK; 3240 last_nid = nid + NAT_ENTRY_PER_BLOCK; 3241 3242 spin_lock(&NM_I(sbi)->nid_list_lock); 3243 for (; nid < last_nid; nid++) 3244 update_free_nid_bitmap(sbi, nid, true, true); 3245 spin_unlock(&NM_I(sbi)->nid_list_lock); 3246 } 3247 3248 for (i = 0; i < nm_i->nat_blocks; i++) { 3249 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i); 3250 if (i >= nm_i->nat_blocks) 3251 break; 3252 3253 __set_bit_le(i, nm_i->nat_block_bitmap); 3254 } 3255 } 3256 3257 static int init_node_manager(struct f2fs_sb_info *sbi) 3258 { 3259 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 3260 struct f2fs_nm_info *nm_i = NM_I(sbi); 3261 unsigned char *version_bitmap; 3262 unsigned int nat_segs; 3263 int err; 3264 3265 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 3266 3267 /* segment_count_nat includes pair segment so divide to 2. */ 3268 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 3269 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 3270 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks; 3271 3272 /* not used nids: 0, node, meta, (and root counted as valid node) */ 3273 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count - 3274 F2FS_RESERVED_NODE_NUM; 3275 nm_i->nid_cnt[FREE_NID] = 0; 3276 nm_i->nid_cnt[PREALLOC_NID] = 0; 3277 nm_i->ram_thresh = DEF_RAM_THRESHOLD; 3278 nm_i->ra_nid_pages = DEF_RA_NID_PAGES; 3279 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD; 3280 nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS; 3281 3282 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC); 3283 INIT_LIST_HEAD(&nm_i->free_nid_list); 3284 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO); 3285 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO); 3286 INIT_LIST_HEAD(&nm_i->nat_entries); 3287 spin_lock_init(&nm_i->nat_list_lock); 3288 3289 mutex_init(&nm_i->build_lock); 3290 spin_lock_init(&nm_i->nid_list_lock); 3291 init_f2fs_rwsem(&nm_i->nat_tree_lock); 3292 3293 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 3294 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 3295 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 3296 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size, 3297 GFP_KERNEL); 3298 if (!nm_i->nat_bitmap) 3299 return -ENOMEM; 3300 3301 err = __get_nat_bitmaps(sbi); 3302 if (err) 3303 return err; 3304 3305 #ifdef CONFIG_F2FS_CHECK_FS 3306 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size, 3307 GFP_KERNEL); 3308 if (!nm_i->nat_bitmap_mir) 3309 return -ENOMEM; 3310 #endif 3311 3312 return 0; 3313 } 3314 3315 static int init_free_nid_cache(struct f2fs_sb_info *sbi) 3316 { 3317 struct f2fs_nm_info *nm_i = NM_I(sbi); 3318 int i; 3319 3320 nm_i->free_nid_bitmap = 3321 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *), 3322 nm_i->nat_blocks), 3323 GFP_KERNEL); 3324 if (!nm_i->free_nid_bitmap) 3325 return -ENOMEM; 3326 3327 for (i = 0; i < nm_i->nat_blocks; i++) { 3328 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi, 3329 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL); 3330 if (!nm_i->free_nid_bitmap[i]) 3331 return -ENOMEM; 3332 } 3333 3334 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8, 3335 GFP_KERNEL); 3336 if (!nm_i->nat_block_bitmap) 3337 return -ENOMEM; 3338 3339 nm_i->free_nid_count = 3340 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short), 3341 nm_i->nat_blocks), 3342 GFP_KERNEL); 3343 if (!nm_i->free_nid_count) 3344 return -ENOMEM; 3345 return 0; 3346 } 3347 3348 int f2fs_build_node_manager(struct f2fs_sb_info *sbi) 3349 { 3350 int err; 3351 3352 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info), 3353 GFP_KERNEL); 3354 if (!sbi->nm_info) 3355 return -ENOMEM; 3356 3357 err = init_node_manager(sbi); 3358 if (err) 3359 return err; 3360 3361 err = init_free_nid_cache(sbi); 3362 if (err) 3363 return err; 3364 3365 /* load free nid status from nat_bits table */ 3366 load_free_nid_bitmap(sbi); 3367 3368 return f2fs_build_free_nids(sbi, true, true); 3369 } 3370 3371 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi) 3372 { 3373 struct f2fs_nm_info *nm_i = NM_I(sbi); 3374 struct free_nid *i, *next_i; 3375 void *vec[NAT_VEC_SIZE]; 3376 struct nat_entry **natvec = (struct nat_entry **)vec; 3377 struct nat_entry_set **setvec = (struct nat_entry_set **)vec; 3378 nid_t nid = 0; 3379 unsigned int found; 3380 3381 if (!nm_i) 3382 return; 3383 3384 /* destroy free nid list */ 3385 spin_lock(&nm_i->nid_list_lock); 3386 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 3387 __remove_free_nid(sbi, i, FREE_NID); 3388 spin_unlock(&nm_i->nid_list_lock); 3389 kmem_cache_free(free_nid_slab, i); 3390 spin_lock(&nm_i->nid_list_lock); 3391 } 3392 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]); 3393 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]); 3394 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list)); 3395 spin_unlock(&nm_i->nid_list_lock); 3396 3397 /* destroy nat cache */ 3398 f2fs_down_write(&nm_i->nat_tree_lock); 3399 while ((found = __gang_lookup_nat_cache(nm_i, 3400 nid, NAT_VEC_SIZE, natvec))) { 3401 unsigned idx; 3402 3403 nid = nat_get_nid(natvec[found - 1]) + 1; 3404 for (idx = 0; idx < found; idx++) { 3405 spin_lock(&nm_i->nat_list_lock); 3406 list_del(&natvec[idx]->list); 3407 spin_unlock(&nm_i->nat_list_lock); 3408 3409 __del_from_nat_cache(nm_i, natvec[idx]); 3410 } 3411 } 3412 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]); 3413 3414 /* destroy nat set cache */ 3415 nid = 0; 3416 memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE); 3417 while ((found = __gang_lookup_nat_set(nm_i, 3418 nid, NAT_VEC_SIZE, setvec))) { 3419 unsigned idx; 3420 3421 nid = setvec[found - 1]->set + 1; 3422 for (idx = 0; idx < found; idx++) { 3423 /* entry_cnt is not zero, when cp_error was occurred */ 3424 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list)); 3425 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set); 3426 kmem_cache_free(nat_entry_set_slab, setvec[idx]); 3427 } 3428 } 3429 f2fs_up_write(&nm_i->nat_tree_lock); 3430 3431 kvfree(nm_i->nat_block_bitmap); 3432 if (nm_i->free_nid_bitmap) { 3433 int i; 3434 3435 for (i = 0; i < nm_i->nat_blocks; i++) 3436 kvfree(nm_i->free_nid_bitmap[i]); 3437 kvfree(nm_i->free_nid_bitmap); 3438 } 3439 kvfree(nm_i->free_nid_count); 3440 3441 kvfree(nm_i->nat_bitmap); 3442 kvfree(nm_i->nat_bits); 3443 #ifdef CONFIG_F2FS_CHECK_FS 3444 kvfree(nm_i->nat_bitmap_mir); 3445 #endif 3446 sbi->nm_info = NULL; 3447 kfree(nm_i); 3448 } 3449 3450 int __init f2fs_create_node_manager_caches(void) 3451 { 3452 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry", 3453 sizeof(struct nat_entry)); 3454 if (!nat_entry_slab) 3455 goto fail; 3456 3457 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid", 3458 sizeof(struct free_nid)); 3459 if (!free_nid_slab) 3460 goto destroy_nat_entry; 3461 3462 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set", 3463 sizeof(struct nat_entry_set)); 3464 if (!nat_entry_set_slab) 3465 goto destroy_free_nid; 3466 3467 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry", 3468 sizeof(struct fsync_node_entry)); 3469 if (!fsync_node_entry_slab) 3470 goto destroy_nat_entry_set; 3471 return 0; 3472 3473 destroy_nat_entry_set: 3474 kmem_cache_destroy(nat_entry_set_slab); 3475 destroy_free_nid: 3476 kmem_cache_destroy(free_nid_slab); 3477 destroy_nat_entry: 3478 kmem_cache_destroy(nat_entry_slab); 3479 fail: 3480 return -ENOMEM; 3481 } 3482 3483 void f2fs_destroy_node_manager_caches(void) 3484 { 3485 kmem_cache_destroy(fsync_node_entry_slab); 3486 kmem_cache_destroy(nat_entry_set_slab); 3487 kmem_cache_destroy(free_nid_slab); 3488 kmem_cache_destroy(nat_entry_slab); 3489 } 3490