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(nmi) mutex_is_locked(&(nm_i)->build_lock) 24 25 static struct kmem_cache *nat_entry_slab; 26 static struct kmem_cache *free_nid_slab; 27 static struct kmem_cache *nat_entry_set_slab; 28 static struct kmem_cache *fsync_node_entry_slab; 29 30 /* 31 * Check whether the given nid is within node id range. 32 */ 33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid) 34 { 35 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) { 36 set_sbi_flag(sbi, SBI_NEED_FSCK); 37 f2fs_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); 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->nid = nid; 562 retry: 563 /* Check nat cache */ 564 f2fs_down_read(&nm_i->nat_tree_lock); 565 e = __lookup_nat_cache(nm_i, nid); 566 if (e) { 567 ni->ino = nat_get_ino(e); 568 ni->blk_addr = nat_get_blkaddr(e); 569 ni->version = nat_get_version(e); 570 f2fs_up_read(&nm_i->nat_tree_lock); 571 return 0; 572 } 573 574 /* 575 * Check current segment summary by trying to grab journal_rwsem first. 576 * This sem is on the critical path on the checkpoint requiring the above 577 * nat_tree_lock. Therefore, we should retry, if we failed to grab here 578 * while not bothering checkpoint. 579 */ 580 if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) { 581 down_read(&curseg->journal_rwsem); 582 } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) || 583 !down_read_trylock(&curseg->journal_rwsem)) { 584 f2fs_up_read(&nm_i->nat_tree_lock); 585 goto retry; 586 } 587 588 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0); 589 if (i >= 0) { 590 ne = nat_in_journal(journal, i); 591 node_info_from_raw_nat(ni, &ne); 592 } 593 up_read(&curseg->journal_rwsem); 594 if (i >= 0) { 595 f2fs_up_read(&nm_i->nat_tree_lock); 596 goto cache; 597 } 598 599 /* Fill node_info from nat page */ 600 index = current_nat_addr(sbi, nid); 601 f2fs_up_read(&nm_i->nat_tree_lock); 602 603 page = f2fs_get_meta_page(sbi, index); 604 if (IS_ERR(page)) 605 return PTR_ERR(page); 606 607 nat_blk = (struct f2fs_nat_block *)page_address(page); 608 ne = nat_blk->entries[nid - start_nid]; 609 node_info_from_raw_nat(ni, &ne); 610 f2fs_put_page(page, 1); 611 cache: 612 blkaddr = le32_to_cpu(ne.block_addr); 613 if (__is_valid_data_blkaddr(blkaddr) && 614 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) 615 return -EFAULT; 616 617 /* cache nat entry */ 618 cache_nat_entry(sbi, nid, &ne); 619 return 0; 620 } 621 622 /* 623 * readahead MAX_RA_NODE number of node pages. 624 */ 625 static void f2fs_ra_node_pages(struct page *parent, int start, int n) 626 { 627 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 628 struct blk_plug plug; 629 int i, end; 630 nid_t nid; 631 632 blk_start_plug(&plug); 633 634 /* Then, try readahead for siblings of the desired node */ 635 end = start + n; 636 end = min(end, NIDS_PER_BLOCK); 637 for (i = start; i < end; i++) { 638 nid = get_nid(parent, i, false); 639 f2fs_ra_node_page(sbi, nid); 640 } 641 642 blk_finish_plug(&plug); 643 } 644 645 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs) 646 { 647 const long direct_index = ADDRS_PER_INODE(dn->inode); 648 const long direct_blks = ADDRS_PER_BLOCK(dn->inode); 649 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK; 650 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode); 651 int cur_level = dn->cur_level; 652 int max_level = dn->max_level; 653 pgoff_t base = 0; 654 655 if (!dn->max_level) 656 return pgofs + 1; 657 658 while (max_level-- > cur_level) 659 skipped_unit *= NIDS_PER_BLOCK; 660 661 switch (dn->max_level) { 662 case 3: 663 base += 2 * indirect_blks; 664 fallthrough; 665 case 2: 666 base += 2 * direct_blks; 667 fallthrough; 668 case 1: 669 base += direct_index; 670 break; 671 default: 672 f2fs_bug_on(F2FS_I_SB(dn->inode), 1); 673 } 674 675 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base; 676 } 677 678 /* 679 * The maximum depth is four. 680 * Offset[0] will have raw inode offset. 681 */ 682 static int get_node_path(struct inode *inode, long block, 683 int offset[4], unsigned int noffset[4]) 684 { 685 const long direct_index = ADDRS_PER_INODE(inode); 686 const long direct_blks = ADDRS_PER_BLOCK(inode); 687 const long dptrs_per_blk = NIDS_PER_BLOCK; 688 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK; 689 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK; 690 int n = 0; 691 int level = 0; 692 693 noffset[0] = 0; 694 695 if (block < direct_index) { 696 offset[n] = block; 697 goto got; 698 } 699 block -= direct_index; 700 if (block < direct_blks) { 701 offset[n++] = NODE_DIR1_BLOCK; 702 noffset[n] = 1; 703 offset[n] = block; 704 level = 1; 705 goto got; 706 } 707 block -= direct_blks; 708 if (block < direct_blks) { 709 offset[n++] = NODE_DIR2_BLOCK; 710 noffset[n] = 2; 711 offset[n] = block; 712 level = 1; 713 goto got; 714 } 715 block -= direct_blks; 716 if (block < indirect_blks) { 717 offset[n++] = NODE_IND1_BLOCK; 718 noffset[n] = 3; 719 offset[n++] = block / direct_blks; 720 noffset[n] = 4 + offset[n - 1]; 721 offset[n] = block % direct_blks; 722 level = 2; 723 goto got; 724 } 725 block -= indirect_blks; 726 if (block < indirect_blks) { 727 offset[n++] = NODE_IND2_BLOCK; 728 noffset[n] = 4 + dptrs_per_blk; 729 offset[n++] = block / direct_blks; 730 noffset[n] = 5 + dptrs_per_blk + offset[n - 1]; 731 offset[n] = block % direct_blks; 732 level = 2; 733 goto got; 734 } 735 block -= indirect_blks; 736 if (block < dindirect_blks) { 737 offset[n++] = NODE_DIND_BLOCK; 738 noffset[n] = 5 + (dptrs_per_blk * 2); 739 offset[n++] = block / indirect_blks; 740 noffset[n] = 6 + (dptrs_per_blk * 2) + 741 offset[n - 1] * (dptrs_per_blk + 1); 742 offset[n++] = (block / direct_blks) % dptrs_per_blk; 743 noffset[n] = 7 + (dptrs_per_blk * 2) + 744 offset[n - 2] * (dptrs_per_blk + 1) + 745 offset[n - 1]; 746 offset[n] = block % direct_blks; 747 level = 3; 748 goto got; 749 } else { 750 return -E2BIG; 751 } 752 got: 753 return level; 754 } 755 756 /* 757 * Caller should call f2fs_put_dnode(dn). 758 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and 759 * f2fs_unlock_op() only if mode is set with ALLOC_NODE. 760 */ 761 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode) 762 { 763 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 764 struct page *npage[4]; 765 struct page *parent = NULL; 766 int offset[4]; 767 unsigned int noffset[4]; 768 nid_t nids[4]; 769 int level, i = 0; 770 int err = 0; 771 772 level = get_node_path(dn->inode, index, offset, noffset); 773 if (level < 0) 774 return level; 775 776 nids[0] = dn->inode->i_ino; 777 npage[0] = dn->inode_page; 778 779 if (!npage[0]) { 780 npage[0] = f2fs_get_node_page(sbi, nids[0]); 781 if (IS_ERR(npage[0])) 782 return PTR_ERR(npage[0]); 783 } 784 785 /* if inline_data is set, should not report any block indices */ 786 if (f2fs_has_inline_data(dn->inode) && index) { 787 err = -ENOENT; 788 f2fs_put_page(npage[0], 1); 789 goto release_out; 790 } 791 792 parent = npage[0]; 793 if (level != 0) 794 nids[1] = get_nid(parent, offset[0], true); 795 dn->inode_page = npage[0]; 796 dn->inode_page_locked = true; 797 798 /* get indirect or direct nodes */ 799 for (i = 1; i <= level; i++) { 800 bool done = false; 801 802 if (!nids[i] && mode == ALLOC_NODE) { 803 /* alloc new node */ 804 if (!f2fs_alloc_nid(sbi, &(nids[i]))) { 805 err = -ENOSPC; 806 goto release_pages; 807 } 808 809 dn->nid = nids[i]; 810 npage[i] = f2fs_new_node_page(dn, noffset[i]); 811 if (IS_ERR(npage[i])) { 812 f2fs_alloc_nid_failed(sbi, nids[i]); 813 err = PTR_ERR(npage[i]); 814 goto release_pages; 815 } 816 817 set_nid(parent, offset[i - 1], nids[i], i == 1); 818 f2fs_alloc_nid_done(sbi, nids[i]); 819 done = true; 820 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) { 821 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]); 822 if (IS_ERR(npage[i])) { 823 err = PTR_ERR(npage[i]); 824 goto release_pages; 825 } 826 done = true; 827 } 828 if (i == 1) { 829 dn->inode_page_locked = false; 830 unlock_page(parent); 831 } else { 832 f2fs_put_page(parent, 1); 833 } 834 835 if (!done) { 836 npage[i] = f2fs_get_node_page(sbi, nids[i]); 837 if (IS_ERR(npage[i])) { 838 err = PTR_ERR(npage[i]); 839 f2fs_put_page(npage[0], 0); 840 goto release_out; 841 } 842 } 843 if (i < level) { 844 parent = npage[i]; 845 nids[i + 1] = get_nid(parent, offset[i], false); 846 } 847 } 848 dn->nid = nids[level]; 849 dn->ofs_in_node = offset[level]; 850 dn->node_page = npage[level]; 851 dn->data_blkaddr = f2fs_data_blkaddr(dn); 852 853 if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) && 854 f2fs_sb_has_readonly(sbi)) { 855 unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn); 856 block_t blkaddr; 857 858 if (!c_len) 859 goto out; 860 861 blkaddr = f2fs_data_blkaddr(dn); 862 if (blkaddr == COMPRESS_ADDR) 863 blkaddr = data_blkaddr(dn->inode, dn->node_page, 864 dn->ofs_in_node + 1); 865 866 f2fs_update_read_extent_tree_range_compressed(dn->inode, 867 index, blkaddr, 868 F2FS_I(dn->inode)->i_cluster_size, 869 c_len); 870 } 871 out: 872 return 0; 873 874 release_pages: 875 f2fs_put_page(parent, 1); 876 if (i > 1) 877 f2fs_put_page(npage[0], 0); 878 release_out: 879 dn->inode_page = NULL; 880 dn->node_page = NULL; 881 if (err == -ENOENT) { 882 dn->cur_level = i; 883 dn->max_level = level; 884 dn->ofs_in_node = offset[level]; 885 } 886 return err; 887 } 888 889 static int truncate_node(struct dnode_of_data *dn) 890 { 891 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 892 struct node_info ni; 893 int err; 894 pgoff_t index; 895 896 err = f2fs_get_node_info(sbi, dn->nid, &ni, false); 897 if (err) 898 return err; 899 900 /* Deallocate node address */ 901 f2fs_invalidate_blocks(sbi, ni.blk_addr); 902 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino); 903 set_node_addr(sbi, &ni, NULL_ADDR, false); 904 905 if (dn->nid == dn->inode->i_ino) { 906 f2fs_remove_orphan_inode(sbi, dn->nid); 907 dec_valid_inode_count(sbi); 908 f2fs_inode_synced(dn->inode); 909 } 910 911 clear_node_page_dirty(dn->node_page); 912 set_sbi_flag(sbi, SBI_IS_DIRTY); 913 914 index = dn->node_page->index; 915 f2fs_put_page(dn->node_page, 1); 916 917 invalidate_mapping_pages(NODE_MAPPING(sbi), 918 index, index); 919 920 dn->node_page = NULL; 921 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr); 922 923 return 0; 924 } 925 926 static int truncate_dnode(struct dnode_of_data *dn) 927 { 928 struct page *page; 929 int err; 930 931 if (dn->nid == 0) 932 return 1; 933 934 /* get direct node */ 935 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 936 if (PTR_ERR(page) == -ENOENT) 937 return 1; 938 else if (IS_ERR(page)) 939 return PTR_ERR(page); 940 941 /* Make dnode_of_data for parameter */ 942 dn->node_page = page; 943 dn->ofs_in_node = 0; 944 f2fs_truncate_data_blocks(dn); 945 err = truncate_node(dn); 946 if (err) 947 return err; 948 949 return 1; 950 } 951 952 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs, 953 int ofs, int depth) 954 { 955 struct dnode_of_data rdn = *dn; 956 struct page *page; 957 struct f2fs_node *rn; 958 nid_t child_nid; 959 unsigned int child_nofs; 960 int freed = 0; 961 int i, ret; 962 963 if (dn->nid == 0) 964 return NIDS_PER_BLOCK + 1; 965 966 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr); 967 968 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 969 if (IS_ERR(page)) { 970 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page)); 971 return PTR_ERR(page); 972 } 973 974 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK); 975 976 rn = F2FS_NODE(page); 977 if (depth < 3) { 978 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) { 979 child_nid = le32_to_cpu(rn->in.nid[i]); 980 if (child_nid == 0) 981 continue; 982 rdn.nid = child_nid; 983 ret = truncate_dnode(&rdn); 984 if (ret < 0) 985 goto out_err; 986 if (set_nid(page, i, 0, false)) 987 dn->node_changed = true; 988 } 989 } else { 990 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1; 991 for (i = ofs; i < NIDS_PER_BLOCK; i++) { 992 child_nid = le32_to_cpu(rn->in.nid[i]); 993 if (child_nid == 0) { 994 child_nofs += NIDS_PER_BLOCK + 1; 995 continue; 996 } 997 rdn.nid = child_nid; 998 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1); 999 if (ret == (NIDS_PER_BLOCK + 1)) { 1000 if (set_nid(page, i, 0, false)) 1001 dn->node_changed = true; 1002 child_nofs += ret; 1003 } else if (ret < 0 && ret != -ENOENT) { 1004 goto out_err; 1005 } 1006 } 1007 freed = child_nofs; 1008 } 1009 1010 if (!ofs) { 1011 /* remove current indirect node */ 1012 dn->node_page = page; 1013 ret = truncate_node(dn); 1014 if (ret) 1015 goto out_err; 1016 freed++; 1017 } else { 1018 f2fs_put_page(page, 1); 1019 } 1020 trace_f2fs_truncate_nodes_exit(dn->inode, freed); 1021 return freed; 1022 1023 out_err: 1024 f2fs_put_page(page, 1); 1025 trace_f2fs_truncate_nodes_exit(dn->inode, ret); 1026 return ret; 1027 } 1028 1029 static int truncate_partial_nodes(struct dnode_of_data *dn, 1030 struct f2fs_inode *ri, int *offset, int depth) 1031 { 1032 struct page *pages[2]; 1033 nid_t nid[3]; 1034 nid_t child_nid; 1035 int err = 0; 1036 int i; 1037 int idx = depth - 2; 1038 1039 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1040 if (!nid[0]) 1041 return 0; 1042 1043 /* get indirect nodes in the path */ 1044 for (i = 0; i < idx + 1; i++) { 1045 /* reference count'll be increased */ 1046 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]); 1047 if (IS_ERR(pages[i])) { 1048 err = PTR_ERR(pages[i]); 1049 idx = i - 1; 1050 goto fail; 1051 } 1052 nid[i + 1] = get_nid(pages[i], offset[i + 1], false); 1053 } 1054 1055 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK); 1056 1057 /* free direct nodes linked to a partial indirect node */ 1058 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) { 1059 child_nid = get_nid(pages[idx], i, false); 1060 if (!child_nid) 1061 continue; 1062 dn->nid = child_nid; 1063 err = truncate_dnode(dn); 1064 if (err < 0) 1065 goto fail; 1066 if (set_nid(pages[idx], i, 0, false)) 1067 dn->node_changed = true; 1068 } 1069 1070 if (offset[idx + 1] == 0) { 1071 dn->node_page = pages[idx]; 1072 dn->nid = nid[idx]; 1073 err = truncate_node(dn); 1074 if (err) 1075 goto fail; 1076 } else { 1077 f2fs_put_page(pages[idx], 1); 1078 } 1079 offset[idx]++; 1080 offset[idx + 1] = 0; 1081 idx--; 1082 fail: 1083 for (i = idx; i >= 0; i--) 1084 f2fs_put_page(pages[i], 1); 1085 1086 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err); 1087 1088 return err; 1089 } 1090 1091 /* 1092 * All the block addresses of data and nodes should be nullified. 1093 */ 1094 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from) 1095 { 1096 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1097 int err = 0, cont = 1; 1098 int level, offset[4], noffset[4]; 1099 unsigned int nofs = 0; 1100 struct f2fs_inode *ri; 1101 struct dnode_of_data dn; 1102 struct page *page; 1103 1104 trace_f2fs_truncate_inode_blocks_enter(inode, from); 1105 1106 level = get_node_path(inode, from, offset, noffset); 1107 if (level < 0) { 1108 trace_f2fs_truncate_inode_blocks_exit(inode, level); 1109 return level; 1110 } 1111 1112 page = f2fs_get_node_page(sbi, inode->i_ino); 1113 if (IS_ERR(page)) { 1114 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page)); 1115 return PTR_ERR(page); 1116 } 1117 1118 set_new_dnode(&dn, inode, page, NULL, 0); 1119 unlock_page(page); 1120 1121 ri = F2FS_INODE(page); 1122 switch (level) { 1123 case 0: 1124 case 1: 1125 nofs = noffset[1]; 1126 break; 1127 case 2: 1128 nofs = noffset[1]; 1129 if (!offset[level - 1]) 1130 goto skip_partial; 1131 err = truncate_partial_nodes(&dn, ri, offset, level); 1132 if (err < 0 && err != -ENOENT) 1133 goto fail; 1134 nofs += 1 + NIDS_PER_BLOCK; 1135 break; 1136 case 3: 1137 nofs = 5 + 2 * NIDS_PER_BLOCK; 1138 if (!offset[level - 1]) 1139 goto skip_partial; 1140 err = truncate_partial_nodes(&dn, ri, offset, level); 1141 if (err < 0 && err != -ENOENT) 1142 goto fail; 1143 break; 1144 default: 1145 BUG(); 1146 } 1147 1148 skip_partial: 1149 while (cont) { 1150 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1151 switch (offset[0]) { 1152 case NODE_DIR1_BLOCK: 1153 case NODE_DIR2_BLOCK: 1154 err = truncate_dnode(&dn); 1155 break; 1156 1157 case NODE_IND1_BLOCK: 1158 case NODE_IND2_BLOCK: 1159 err = truncate_nodes(&dn, nofs, offset[1], 2); 1160 break; 1161 1162 case NODE_DIND_BLOCK: 1163 err = truncate_nodes(&dn, nofs, offset[1], 3); 1164 cont = 0; 1165 break; 1166 1167 default: 1168 BUG(); 1169 } 1170 if (err < 0 && err != -ENOENT) 1171 goto fail; 1172 if (offset[1] == 0 && 1173 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) { 1174 lock_page(page); 1175 BUG_ON(page->mapping != NODE_MAPPING(sbi)); 1176 f2fs_wait_on_page_writeback(page, NODE, true, true); 1177 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0; 1178 set_page_dirty(page); 1179 unlock_page(page); 1180 } 1181 offset[1] = 0; 1182 offset[0]++; 1183 nofs += err; 1184 } 1185 fail: 1186 f2fs_put_page(page, 0); 1187 trace_f2fs_truncate_inode_blocks_exit(inode, err); 1188 return err > 0 ? 0 : err; 1189 } 1190 1191 /* caller must lock inode page */ 1192 int f2fs_truncate_xattr_node(struct inode *inode) 1193 { 1194 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1195 nid_t nid = F2FS_I(inode)->i_xattr_nid; 1196 struct dnode_of_data dn; 1197 struct page *npage; 1198 int err; 1199 1200 if (!nid) 1201 return 0; 1202 1203 npage = f2fs_get_node_page(sbi, nid); 1204 if (IS_ERR(npage)) 1205 return PTR_ERR(npage); 1206 1207 set_new_dnode(&dn, inode, NULL, npage, nid); 1208 err = truncate_node(&dn); 1209 if (err) { 1210 f2fs_put_page(npage, 1); 1211 return err; 1212 } 1213 1214 f2fs_i_xnid_write(inode, 0); 1215 1216 return 0; 1217 } 1218 1219 /* 1220 * Caller should grab and release a rwsem by calling f2fs_lock_op() and 1221 * f2fs_unlock_op(). 1222 */ 1223 int f2fs_remove_inode_page(struct inode *inode) 1224 { 1225 struct dnode_of_data dn; 1226 int err; 1227 1228 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1229 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE); 1230 if (err) 1231 return err; 1232 1233 err = f2fs_truncate_xattr_node(inode); 1234 if (err) { 1235 f2fs_put_dnode(&dn); 1236 return err; 1237 } 1238 1239 /* remove potential inline_data blocks */ 1240 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1241 S_ISLNK(inode->i_mode)) 1242 f2fs_truncate_data_blocks_range(&dn, 1); 1243 1244 /* 0 is possible, after f2fs_new_inode() has failed */ 1245 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) { 1246 f2fs_put_dnode(&dn); 1247 return -EIO; 1248 } 1249 1250 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) { 1251 f2fs_warn(F2FS_I_SB(inode), 1252 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu", 1253 inode->i_ino, (unsigned long long)inode->i_blocks); 1254 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK); 1255 } 1256 1257 /* will put inode & node pages */ 1258 err = truncate_node(&dn); 1259 if (err) { 1260 f2fs_put_dnode(&dn); 1261 return err; 1262 } 1263 return 0; 1264 } 1265 1266 struct page *f2fs_new_inode_page(struct inode *inode) 1267 { 1268 struct dnode_of_data dn; 1269 1270 /* allocate inode page for new inode */ 1271 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1272 1273 /* caller should f2fs_put_page(page, 1); */ 1274 return f2fs_new_node_page(&dn, 0); 1275 } 1276 1277 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs) 1278 { 1279 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1280 struct node_info new_ni; 1281 struct page *page; 1282 int err; 1283 1284 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1285 return ERR_PTR(-EPERM); 1286 1287 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false); 1288 if (!page) 1289 return ERR_PTR(-ENOMEM); 1290 1291 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs)))) 1292 goto fail; 1293 1294 #ifdef CONFIG_F2FS_CHECK_FS 1295 err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false); 1296 if (err) { 1297 dec_valid_node_count(sbi, dn->inode, !ofs); 1298 goto fail; 1299 } 1300 if (unlikely(new_ni.blk_addr != NULL_ADDR)) { 1301 err = -EFSCORRUPTED; 1302 set_sbi_flag(sbi, SBI_NEED_FSCK); 1303 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 1304 goto fail; 1305 } 1306 #endif 1307 new_ni.nid = dn->nid; 1308 new_ni.ino = dn->inode->i_ino; 1309 new_ni.blk_addr = NULL_ADDR; 1310 new_ni.flag = 0; 1311 new_ni.version = 0; 1312 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 1313 1314 f2fs_wait_on_page_writeback(page, NODE, true, true); 1315 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 1316 set_cold_node(page, S_ISDIR(dn->inode->i_mode)); 1317 if (!PageUptodate(page)) 1318 SetPageUptodate(page); 1319 if (set_page_dirty(page)) 1320 dn->node_changed = true; 1321 1322 if (f2fs_has_xattr_block(ofs)) 1323 f2fs_i_xnid_write(dn->inode, dn->nid); 1324 1325 if (ofs == 0) 1326 inc_valid_inode_count(sbi); 1327 return page; 1328 1329 fail: 1330 clear_node_page_dirty(page); 1331 f2fs_put_page(page, 1); 1332 return ERR_PTR(err); 1333 } 1334 1335 /* 1336 * Caller should do after getting the following values. 1337 * 0: f2fs_put_page(page, 0) 1338 * LOCKED_PAGE or error: f2fs_put_page(page, 1) 1339 */ 1340 static int read_node_page(struct page *page, blk_opf_t op_flags) 1341 { 1342 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1343 struct node_info ni; 1344 struct f2fs_io_info fio = { 1345 .sbi = sbi, 1346 .type = NODE, 1347 .op = REQ_OP_READ, 1348 .op_flags = op_flags, 1349 .page = page, 1350 .encrypted_page = NULL, 1351 }; 1352 int err; 1353 1354 if (PageUptodate(page)) { 1355 if (!f2fs_inode_chksum_verify(sbi, page)) { 1356 ClearPageUptodate(page); 1357 return -EFSBADCRC; 1358 } 1359 return LOCKED_PAGE; 1360 } 1361 1362 err = f2fs_get_node_info(sbi, page->index, &ni, false); 1363 if (err) 1364 return err; 1365 1366 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */ 1367 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) { 1368 ClearPageUptodate(page); 1369 return -ENOENT; 1370 } 1371 1372 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr; 1373 1374 err = f2fs_submit_page_bio(&fio); 1375 1376 if (!err) 1377 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE); 1378 1379 return err; 1380 } 1381 1382 /* 1383 * Readahead a node page 1384 */ 1385 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 1386 { 1387 struct page *apage; 1388 int err; 1389 1390 if (!nid) 1391 return; 1392 if (f2fs_check_nid_range(sbi, nid)) 1393 return; 1394 1395 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid); 1396 if (apage) 1397 return; 1398 1399 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1400 if (!apage) 1401 return; 1402 1403 err = read_node_page(apage, REQ_RAHEAD); 1404 f2fs_put_page(apage, err ? 1 : 0); 1405 } 1406 1407 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid, 1408 struct page *parent, int start) 1409 { 1410 struct page *page; 1411 int err; 1412 1413 if (!nid) 1414 return ERR_PTR(-ENOENT); 1415 if (f2fs_check_nid_range(sbi, nid)) 1416 return ERR_PTR(-EINVAL); 1417 repeat: 1418 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1419 if (!page) 1420 return ERR_PTR(-ENOMEM); 1421 1422 err = read_node_page(page, 0); 1423 if (err < 0) { 1424 goto out_put_err; 1425 } else if (err == LOCKED_PAGE) { 1426 err = 0; 1427 goto page_hit; 1428 } 1429 1430 if (parent) 1431 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE); 1432 1433 lock_page(page); 1434 1435 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1436 f2fs_put_page(page, 1); 1437 goto repeat; 1438 } 1439 1440 if (unlikely(!PageUptodate(page))) { 1441 err = -EIO; 1442 goto out_err; 1443 } 1444 1445 if (!f2fs_inode_chksum_verify(sbi, page)) { 1446 err = -EFSBADCRC; 1447 goto out_err; 1448 } 1449 page_hit: 1450 if (likely(nid == nid_of_node(page))) 1451 return page; 1452 1453 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]", 1454 nid, nid_of_node(page), ino_of_node(page), 1455 ofs_of_node(page), cpver_of_node(page), 1456 next_blkaddr_of_node(page)); 1457 set_sbi_flag(sbi, SBI_NEED_FSCK); 1458 err = -EINVAL; 1459 out_err: 1460 ClearPageUptodate(page); 1461 out_put_err: 1462 /* ENOENT comes from read_node_page which is not an error. */ 1463 if (err != -ENOENT) 1464 f2fs_handle_page_eio(sbi, page->index, NODE); 1465 f2fs_put_page(page, 1); 1466 return ERR_PTR(err); 1467 } 1468 1469 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 1470 { 1471 return __get_node_page(sbi, nid, NULL, 0); 1472 } 1473 1474 struct page *f2fs_get_node_page_ra(struct page *parent, int start) 1475 { 1476 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 1477 nid_t nid = get_nid(parent, start, false); 1478 1479 return __get_node_page(sbi, nid, parent, start); 1480 } 1481 1482 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino) 1483 { 1484 struct inode *inode; 1485 struct page *page; 1486 int ret; 1487 1488 /* should flush inline_data before evict_inode */ 1489 inode = ilookup(sbi->sb, ino); 1490 if (!inode) 1491 return; 1492 1493 page = f2fs_pagecache_get_page(inode->i_mapping, 0, 1494 FGP_LOCK|FGP_NOWAIT, 0); 1495 if (!page) 1496 goto iput_out; 1497 1498 if (!PageUptodate(page)) 1499 goto page_out; 1500 1501 if (!PageDirty(page)) 1502 goto page_out; 1503 1504 if (!clear_page_dirty_for_io(page)) 1505 goto page_out; 1506 1507 ret = f2fs_write_inline_data(inode, page); 1508 inode_dec_dirty_pages(inode); 1509 f2fs_remove_dirty_inode(inode); 1510 if (ret) 1511 set_page_dirty(page); 1512 page_out: 1513 f2fs_put_page(page, 1); 1514 iput_out: 1515 iput(inode); 1516 } 1517 1518 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino) 1519 { 1520 pgoff_t index; 1521 struct folio_batch fbatch; 1522 struct page *last_page = NULL; 1523 int nr_folios; 1524 1525 folio_batch_init(&fbatch); 1526 index = 0; 1527 1528 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1529 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1530 &fbatch))) { 1531 int i; 1532 1533 for (i = 0; i < nr_folios; i++) { 1534 struct page *page = &fbatch.folios[i]->page; 1535 1536 if (unlikely(f2fs_cp_error(sbi))) { 1537 f2fs_put_page(last_page, 0); 1538 folio_batch_release(&fbatch); 1539 return ERR_PTR(-EIO); 1540 } 1541 1542 if (!IS_DNODE(page) || !is_cold_node(page)) 1543 continue; 1544 if (ino_of_node(page) != ino) 1545 continue; 1546 1547 lock_page(page); 1548 1549 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1550 continue_unlock: 1551 unlock_page(page); 1552 continue; 1553 } 1554 if (ino_of_node(page) != ino) 1555 goto continue_unlock; 1556 1557 if (!PageDirty(page)) { 1558 /* someone wrote it for us */ 1559 goto continue_unlock; 1560 } 1561 1562 if (last_page) 1563 f2fs_put_page(last_page, 0); 1564 1565 get_page(page); 1566 last_page = page; 1567 unlock_page(page); 1568 } 1569 folio_batch_release(&fbatch); 1570 cond_resched(); 1571 } 1572 return last_page; 1573 } 1574 1575 static int __write_node_page(struct page *page, bool atomic, bool *submitted, 1576 struct writeback_control *wbc, bool do_balance, 1577 enum iostat_type io_type, unsigned int *seq_id) 1578 { 1579 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1580 nid_t nid; 1581 struct node_info ni; 1582 struct f2fs_io_info fio = { 1583 .sbi = sbi, 1584 .ino = ino_of_node(page), 1585 .type = NODE, 1586 .op = REQ_OP_WRITE, 1587 .op_flags = wbc_to_write_flags(wbc), 1588 .page = page, 1589 .encrypted_page = NULL, 1590 .submitted = 0, 1591 .io_type = io_type, 1592 .io_wbc = wbc, 1593 }; 1594 unsigned int seq; 1595 1596 trace_f2fs_writepage(page, NODE); 1597 1598 if (unlikely(f2fs_cp_error(sbi))) { 1599 /* keep node pages in remount-ro mode */ 1600 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY) 1601 goto redirty_out; 1602 ClearPageUptodate(page); 1603 dec_page_count(sbi, F2FS_DIRTY_NODES); 1604 unlock_page(page); 1605 return 0; 1606 } 1607 1608 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1609 goto redirty_out; 1610 1611 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 1612 wbc->sync_mode == WB_SYNC_NONE && 1613 IS_DNODE(page) && is_cold_node(page)) 1614 goto redirty_out; 1615 1616 /* get old block addr of this node page */ 1617 nid = nid_of_node(page); 1618 f2fs_bug_on(sbi, page->index != nid); 1619 1620 if (f2fs_get_node_info(sbi, nid, &ni, !do_balance)) 1621 goto redirty_out; 1622 1623 if (wbc->for_reclaim) { 1624 if (!f2fs_down_read_trylock(&sbi->node_write)) 1625 goto redirty_out; 1626 } else { 1627 f2fs_down_read(&sbi->node_write); 1628 } 1629 1630 /* This page is already truncated */ 1631 if (unlikely(ni.blk_addr == NULL_ADDR)) { 1632 ClearPageUptodate(page); 1633 dec_page_count(sbi, F2FS_DIRTY_NODES); 1634 f2fs_up_read(&sbi->node_write); 1635 unlock_page(page); 1636 return 0; 1637 } 1638 1639 if (__is_valid_data_blkaddr(ni.blk_addr) && 1640 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, 1641 DATA_GENERIC_ENHANCE)) { 1642 f2fs_up_read(&sbi->node_write); 1643 goto redirty_out; 1644 } 1645 1646 if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi)) 1647 fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 1648 1649 /* should add to global list before clearing PAGECACHE status */ 1650 if (f2fs_in_warm_node_list(sbi, page)) { 1651 seq = f2fs_add_fsync_node_entry(sbi, page); 1652 if (seq_id) 1653 *seq_id = seq; 1654 } 1655 1656 set_page_writeback(page); 1657 1658 fio.old_blkaddr = ni.blk_addr; 1659 f2fs_do_write_node_page(nid, &fio); 1660 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page)); 1661 dec_page_count(sbi, F2FS_DIRTY_NODES); 1662 f2fs_up_read(&sbi->node_write); 1663 1664 if (wbc->for_reclaim) { 1665 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE); 1666 submitted = NULL; 1667 } 1668 1669 unlock_page(page); 1670 1671 if (unlikely(f2fs_cp_error(sbi))) { 1672 f2fs_submit_merged_write(sbi, NODE); 1673 submitted = NULL; 1674 } 1675 if (submitted) 1676 *submitted = fio.submitted; 1677 1678 if (do_balance) 1679 f2fs_balance_fs(sbi, false); 1680 return 0; 1681 1682 redirty_out: 1683 redirty_page_for_writepage(wbc, page); 1684 return AOP_WRITEPAGE_ACTIVATE; 1685 } 1686 1687 int f2fs_move_node_page(struct page *node_page, int gc_type) 1688 { 1689 int err = 0; 1690 1691 if (gc_type == FG_GC) { 1692 struct writeback_control wbc = { 1693 .sync_mode = WB_SYNC_ALL, 1694 .nr_to_write = 1, 1695 .for_reclaim = 0, 1696 }; 1697 1698 f2fs_wait_on_page_writeback(node_page, NODE, true, true); 1699 1700 set_page_dirty(node_page); 1701 1702 if (!clear_page_dirty_for_io(node_page)) { 1703 err = -EAGAIN; 1704 goto out_page; 1705 } 1706 1707 if (__write_node_page(node_page, false, NULL, 1708 &wbc, false, FS_GC_NODE_IO, NULL)) { 1709 err = -EAGAIN; 1710 unlock_page(node_page); 1711 } 1712 goto release_page; 1713 } else { 1714 /* set page dirty and write it */ 1715 if (!PageWriteback(node_page)) 1716 set_page_dirty(node_page); 1717 } 1718 out_page: 1719 unlock_page(node_page); 1720 release_page: 1721 f2fs_put_page(node_page, 0); 1722 return err; 1723 } 1724 1725 static int f2fs_write_node_page(struct page *page, 1726 struct writeback_control *wbc) 1727 { 1728 return __write_node_page(page, false, NULL, wbc, false, 1729 FS_NODE_IO, NULL); 1730 } 1731 1732 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode, 1733 struct writeback_control *wbc, bool atomic, 1734 unsigned int *seq_id) 1735 { 1736 pgoff_t index; 1737 struct folio_batch fbatch; 1738 int ret = 0; 1739 struct page *last_page = NULL; 1740 bool marked = false; 1741 nid_t ino = inode->i_ino; 1742 int nr_folios; 1743 int nwritten = 0; 1744 1745 if (atomic) { 1746 last_page = last_fsync_dnode(sbi, ino); 1747 if (IS_ERR_OR_NULL(last_page)) 1748 return PTR_ERR_OR_ZERO(last_page); 1749 } 1750 retry: 1751 folio_batch_init(&fbatch); 1752 index = 0; 1753 1754 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1755 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1756 &fbatch))) { 1757 int i; 1758 1759 for (i = 0; i < nr_folios; i++) { 1760 struct page *page = &fbatch.folios[i]->page; 1761 bool submitted = false; 1762 1763 if (unlikely(f2fs_cp_error(sbi))) { 1764 f2fs_put_page(last_page, 0); 1765 folio_batch_release(&fbatch); 1766 ret = -EIO; 1767 goto out; 1768 } 1769 1770 if (!IS_DNODE(page) || !is_cold_node(page)) 1771 continue; 1772 if (ino_of_node(page) != ino) 1773 continue; 1774 1775 lock_page(page); 1776 1777 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1778 continue_unlock: 1779 unlock_page(page); 1780 continue; 1781 } 1782 if (ino_of_node(page) != ino) 1783 goto continue_unlock; 1784 1785 if (!PageDirty(page) && page != last_page) { 1786 /* someone wrote it for us */ 1787 goto continue_unlock; 1788 } 1789 1790 f2fs_wait_on_page_writeback(page, NODE, true, true); 1791 1792 set_fsync_mark(page, 0); 1793 set_dentry_mark(page, 0); 1794 1795 if (!atomic || page == last_page) { 1796 set_fsync_mark(page, 1); 1797 percpu_counter_inc(&sbi->rf_node_block_count); 1798 if (IS_INODE(page)) { 1799 if (is_inode_flag_set(inode, 1800 FI_DIRTY_INODE)) 1801 f2fs_update_inode(inode, page); 1802 set_dentry_mark(page, 1803 f2fs_need_dentry_mark(sbi, ino)); 1804 } 1805 /* may be written by other thread */ 1806 if (!PageDirty(page)) 1807 set_page_dirty(page); 1808 } 1809 1810 if (!clear_page_dirty_for_io(page)) 1811 goto continue_unlock; 1812 1813 ret = __write_node_page(page, atomic && 1814 page == last_page, 1815 &submitted, wbc, true, 1816 FS_NODE_IO, seq_id); 1817 if (ret) { 1818 unlock_page(page); 1819 f2fs_put_page(last_page, 0); 1820 break; 1821 } else if (submitted) { 1822 nwritten++; 1823 } 1824 1825 if (page == last_page) { 1826 f2fs_put_page(page, 0); 1827 marked = true; 1828 break; 1829 } 1830 } 1831 folio_batch_release(&fbatch); 1832 cond_resched(); 1833 1834 if (ret || marked) 1835 break; 1836 } 1837 if (!ret && atomic && !marked) { 1838 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx", 1839 ino, last_page->index); 1840 lock_page(last_page); 1841 f2fs_wait_on_page_writeback(last_page, NODE, true, true); 1842 set_page_dirty(last_page); 1843 unlock_page(last_page); 1844 goto retry; 1845 } 1846 out: 1847 if (nwritten) 1848 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE); 1849 return ret ? -EIO : 0; 1850 } 1851 1852 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data) 1853 { 1854 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1855 bool clean; 1856 1857 if (inode->i_ino != ino) 1858 return 0; 1859 1860 if (!is_inode_flag_set(inode, FI_DIRTY_INODE)) 1861 return 0; 1862 1863 spin_lock(&sbi->inode_lock[DIRTY_META]); 1864 clean = list_empty(&F2FS_I(inode)->gdirty_list); 1865 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1866 1867 if (clean) 1868 return 0; 1869 1870 inode = igrab(inode); 1871 if (!inode) 1872 return 0; 1873 return 1; 1874 } 1875 1876 static bool flush_dirty_inode(struct page *page) 1877 { 1878 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1879 struct inode *inode; 1880 nid_t ino = ino_of_node(page); 1881 1882 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL); 1883 if (!inode) 1884 return false; 1885 1886 f2fs_update_inode(inode, page); 1887 unlock_page(page); 1888 1889 iput(inode); 1890 return true; 1891 } 1892 1893 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi) 1894 { 1895 pgoff_t index = 0; 1896 struct folio_batch fbatch; 1897 int nr_folios; 1898 1899 folio_batch_init(&fbatch); 1900 1901 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1902 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1903 &fbatch))) { 1904 int i; 1905 1906 for (i = 0; i < nr_folios; i++) { 1907 struct page *page = &fbatch.folios[i]->page; 1908 1909 if (!IS_DNODE(page)) 1910 continue; 1911 1912 lock_page(page); 1913 1914 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1915 continue_unlock: 1916 unlock_page(page); 1917 continue; 1918 } 1919 1920 if (!PageDirty(page)) { 1921 /* someone wrote it for us */ 1922 goto continue_unlock; 1923 } 1924 1925 /* flush inline_data, if it's async context. */ 1926 if (page_private_inline(page)) { 1927 clear_page_private_inline(page); 1928 unlock_page(page); 1929 flush_inline_data(sbi, ino_of_node(page)); 1930 continue; 1931 } 1932 unlock_page(page); 1933 } 1934 folio_batch_release(&fbatch); 1935 cond_resched(); 1936 } 1937 } 1938 1939 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi, 1940 struct writeback_control *wbc, 1941 bool do_balance, enum iostat_type io_type) 1942 { 1943 pgoff_t index; 1944 struct folio_batch fbatch; 1945 int step = 0; 1946 int nwritten = 0; 1947 int ret = 0; 1948 int nr_folios, done = 0; 1949 1950 folio_batch_init(&fbatch); 1951 1952 next_step: 1953 index = 0; 1954 1955 while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), 1956 &index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1957 &fbatch))) { 1958 int i; 1959 1960 for (i = 0; i < nr_folios; i++) { 1961 struct page *page = &fbatch.folios[i]->page; 1962 bool submitted = false; 1963 1964 /* give a priority to WB_SYNC threads */ 1965 if (atomic_read(&sbi->wb_sync_req[NODE]) && 1966 wbc->sync_mode == WB_SYNC_NONE) { 1967 done = 1; 1968 break; 1969 } 1970 1971 /* 1972 * flushing sequence with step: 1973 * 0. indirect nodes 1974 * 1. dentry dnodes 1975 * 2. file dnodes 1976 */ 1977 if (step == 0 && IS_DNODE(page)) 1978 continue; 1979 if (step == 1 && (!IS_DNODE(page) || 1980 is_cold_node(page))) 1981 continue; 1982 if (step == 2 && (!IS_DNODE(page) || 1983 !is_cold_node(page))) 1984 continue; 1985 lock_node: 1986 if (wbc->sync_mode == WB_SYNC_ALL) 1987 lock_page(page); 1988 else if (!trylock_page(page)) 1989 continue; 1990 1991 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1992 continue_unlock: 1993 unlock_page(page); 1994 continue; 1995 } 1996 1997 if (!PageDirty(page)) { 1998 /* someone wrote it for us */ 1999 goto continue_unlock; 2000 } 2001 2002 /* flush inline_data/inode, if it's async context. */ 2003 if (!do_balance) 2004 goto write_node; 2005 2006 /* flush inline_data */ 2007 if (page_private_inline(page)) { 2008 clear_page_private_inline(page); 2009 unlock_page(page); 2010 flush_inline_data(sbi, ino_of_node(page)); 2011 goto lock_node; 2012 } 2013 2014 /* flush dirty inode */ 2015 if (IS_INODE(page) && flush_dirty_inode(page)) 2016 goto lock_node; 2017 write_node: 2018 f2fs_wait_on_page_writeback(page, NODE, true, true); 2019 2020 if (!clear_page_dirty_for_io(page)) 2021 goto continue_unlock; 2022 2023 set_fsync_mark(page, 0); 2024 set_dentry_mark(page, 0); 2025 2026 ret = __write_node_page(page, false, &submitted, 2027 wbc, do_balance, io_type, NULL); 2028 if (ret) 2029 unlock_page(page); 2030 else if (submitted) 2031 nwritten++; 2032 2033 if (--wbc->nr_to_write == 0) 2034 break; 2035 } 2036 folio_batch_release(&fbatch); 2037 cond_resched(); 2038 2039 if (wbc->nr_to_write == 0) { 2040 step = 2; 2041 break; 2042 } 2043 } 2044 2045 if (step < 2) { 2046 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2047 wbc->sync_mode == WB_SYNC_NONE && step == 1) 2048 goto out; 2049 step++; 2050 goto next_step; 2051 } 2052 out: 2053 if (nwritten) 2054 f2fs_submit_merged_write(sbi, NODE); 2055 2056 if (unlikely(f2fs_cp_error(sbi))) 2057 return -EIO; 2058 return ret; 2059 } 2060 2061 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, 2062 unsigned int seq_id) 2063 { 2064 struct fsync_node_entry *fn; 2065 struct page *page; 2066 struct list_head *head = &sbi->fsync_node_list; 2067 unsigned long flags; 2068 unsigned int cur_seq_id = 0; 2069 2070 while (seq_id && cur_seq_id < seq_id) { 2071 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 2072 if (list_empty(head)) { 2073 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2074 break; 2075 } 2076 fn = list_first_entry(head, struct fsync_node_entry, list); 2077 if (fn->seq_id > seq_id) { 2078 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2079 break; 2080 } 2081 cur_seq_id = fn->seq_id; 2082 page = fn->page; 2083 get_page(page); 2084 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2085 2086 f2fs_wait_on_page_writeback(page, NODE, true, false); 2087 2088 put_page(page); 2089 } 2090 2091 return filemap_check_errors(NODE_MAPPING(sbi)); 2092 } 2093 2094 static int f2fs_write_node_pages(struct address_space *mapping, 2095 struct writeback_control *wbc) 2096 { 2097 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2098 struct blk_plug plug; 2099 long diff; 2100 2101 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2102 goto skip_write; 2103 2104 /* balancing f2fs's metadata in background */ 2105 f2fs_balance_fs_bg(sbi, true); 2106 2107 /* collect a number of dirty node pages and write together */ 2108 if (wbc->sync_mode != WB_SYNC_ALL && 2109 get_pages(sbi, F2FS_DIRTY_NODES) < 2110 nr_pages_to_skip(sbi, NODE)) 2111 goto skip_write; 2112 2113 if (wbc->sync_mode == WB_SYNC_ALL) 2114 atomic_inc(&sbi->wb_sync_req[NODE]); 2115 else if (atomic_read(&sbi->wb_sync_req[NODE])) { 2116 /* to avoid potential deadlock */ 2117 if (current->plug) 2118 blk_finish_plug(current->plug); 2119 goto skip_write; 2120 } 2121 2122 trace_f2fs_writepages(mapping->host, wbc, NODE); 2123 2124 diff = nr_pages_to_write(sbi, NODE, wbc); 2125 blk_start_plug(&plug); 2126 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO); 2127 blk_finish_plug(&plug); 2128 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); 2129 2130 if (wbc->sync_mode == WB_SYNC_ALL) 2131 atomic_dec(&sbi->wb_sync_req[NODE]); 2132 return 0; 2133 2134 skip_write: 2135 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES); 2136 trace_f2fs_writepages(mapping->host, wbc, NODE); 2137 return 0; 2138 } 2139 2140 static bool f2fs_dirty_node_folio(struct address_space *mapping, 2141 struct folio *folio) 2142 { 2143 trace_f2fs_set_page_dirty(&folio->page, NODE); 2144 2145 if (!folio_test_uptodate(folio)) 2146 folio_mark_uptodate(folio); 2147 #ifdef CONFIG_F2FS_CHECK_FS 2148 if (IS_INODE(&folio->page)) 2149 f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page); 2150 #endif 2151 if (filemap_dirty_folio(mapping, folio)) { 2152 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES); 2153 set_page_private_reference(&folio->page); 2154 return true; 2155 } 2156 return false; 2157 } 2158 2159 /* 2160 * Structure of the f2fs node operations 2161 */ 2162 const struct address_space_operations f2fs_node_aops = { 2163 .writepage = f2fs_write_node_page, 2164 .writepages = f2fs_write_node_pages, 2165 .dirty_folio = f2fs_dirty_node_folio, 2166 .invalidate_folio = f2fs_invalidate_folio, 2167 .release_folio = f2fs_release_folio, 2168 .migrate_folio = filemap_migrate_folio, 2169 }; 2170 2171 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i, 2172 nid_t n) 2173 { 2174 return radix_tree_lookup(&nm_i->free_nid_root, n); 2175 } 2176 2177 static int __insert_free_nid(struct f2fs_sb_info *sbi, 2178 struct free_nid *i) 2179 { 2180 struct f2fs_nm_info *nm_i = NM_I(sbi); 2181 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i); 2182 2183 if (err) 2184 return err; 2185 2186 nm_i->nid_cnt[FREE_NID]++; 2187 list_add_tail(&i->list, &nm_i->free_nid_list); 2188 return 0; 2189 } 2190 2191 static void __remove_free_nid(struct f2fs_sb_info *sbi, 2192 struct free_nid *i, enum nid_state state) 2193 { 2194 struct f2fs_nm_info *nm_i = NM_I(sbi); 2195 2196 f2fs_bug_on(sbi, state != i->state); 2197 nm_i->nid_cnt[state]--; 2198 if (state == FREE_NID) 2199 list_del(&i->list); 2200 radix_tree_delete(&nm_i->free_nid_root, i->nid); 2201 } 2202 2203 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i, 2204 enum nid_state org_state, enum nid_state dst_state) 2205 { 2206 struct f2fs_nm_info *nm_i = NM_I(sbi); 2207 2208 f2fs_bug_on(sbi, org_state != i->state); 2209 i->state = dst_state; 2210 nm_i->nid_cnt[org_state]--; 2211 nm_i->nid_cnt[dst_state]++; 2212 2213 switch (dst_state) { 2214 case PREALLOC_NID: 2215 list_del(&i->list); 2216 break; 2217 case FREE_NID: 2218 list_add_tail(&i->list, &nm_i->free_nid_list); 2219 break; 2220 default: 2221 BUG_ON(1); 2222 } 2223 } 2224 2225 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi) 2226 { 2227 struct f2fs_nm_info *nm_i = NM_I(sbi); 2228 unsigned int i; 2229 bool ret = true; 2230 2231 f2fs_down_read(&nm_i->nat_tree_lock); 2232 for (i = 0; i < nm_i->nat_blocks; i++) { 2233 if (!test_bit_le(i, nm_i->nat_block_bitmap)) { 2234 ret = false; 2235 break; 2236 } 2237 } 2238 f2fs_up_read(&nm_i->nat_tree_lock); 2239 2240 return ret; 2241 } 2242 2243 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid, 2244 bool set, bool build) 2245 { 2246 struct f2fs_nm_info *nm_i = NM_I(sbi); 2247 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid); 2248 unsigned int nid_ofs = nid - START_NID(nid); 2249 2250 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap)) 2251 return; 2252 2253 if (set) { 2254 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2255 return; 2256 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2257 nm_i->free_nid_count[nat_ofs]++; 2258 } else { 2259 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2260 return; 2261 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2262 if (!build) 2263 nm_i->free_nid_count[nat_ofs]--; 2264 } 2265 } 2266 2267 /* return if the nid is recognized as free */ 2268 static bool add_free_nid(struct f2fs_sb_info *sbi, 2269 nid_t nid, bool build, bool update) 2270 { 2271 struct f2fs_nm_info *nm_i = NM_I(sbi); 2272 struct free_nid *i, *e; 2273 struct nat_entry *ne; 2274 int err = -EINVAL; 2275 bool ret = false; 2276 2277 /* 0 nid should not be used */ 2278 if (unlikely(nid == 0)) 2279 return false; 2280 2281 if (unlikely(f2fs_check_nid_range(sbi, nid))) 2282 return false; 2283 2284 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL); 2285 i->nid = nid; 2286 i->state = FREE_NID; 2287 2288 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 2289 2290 spin_lock(&nm_i->nid_list_lock); 2291 2292 if (build) { 2293 /* 2294 * Thread A Thread B 2295 * - f2fs_create 2296 * - f2fs_new_inode 2297 * - f2fs_alloc_nid 2298 * - __insert_nid_to_list(PREALLOC_NID) 2299 * - f2fs_balance_fs_bg 2300 * - f2fs_build_free_nids 2301 * - __f2fs_build_free_nids 2302 * - scan_nat_page 2303 * - add_free_nid 2304 * - __lookup_nat_cache 2305 * - f2fs_add_link 2306 * - f2fs_init_inode_metadata 2307 * - f2fs_new_inode_page 2308 * - f2fs_new_node_page 2309 * - set_node_addr 2310 * - f2fs_alloc_nid_done 2311 * - __remove_nid_from_list(PREALLOC_NID) 2312 * - __insert_nid_to_list(FREE_NID) 2313 */ 2314 ne = __lookup_nat_cache(nm_i, nid); 2315 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) || 2316 nat_get_blkaddr(ne) != NULL_ADDR)) 2317 goto err_out; 2318 2319 e = __lookup_free_nid_list(nm_i, nid); 2320 if (e) { 2321 if (e->state == FREE_NID) 2322 ret = true; 2323 goto err_out; 2324 } 2325 } 2326 ret = true; 2327 err = __insert_free_nid(sbi, i); 2328 err_out: 2329 if (update) { 2330 update_free_nid_bitmap(sbi, nid, ret, build); 2331 if (!build) 2332 nm_i->available_nids++; 2333 } 2334 spin_unlock(&nm_i->nid_list_lock); 2335 radix_tree_preload_end(); 2336 2337 if (err) 2338 kmem_cache_free(free_nid_slab, i); 2339 return ret; 2340 } 2341 2342 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid) 2343 { 2344 struct f2fs_nm_info *nm_i = NM_I(sbi); 2345 struct free_nid *i; 2346 bool need_free = false; 2347 2348 spin_lock(&nm_i->nid_list_lock); 2349 i = __lookup_free_nid_list(nm_i, nid); 2350 if (i && i->state == FREE_NID) { 2351 __remove_free_nid(sbi, i, FREE_NID); 2352 need_free = true; 2353 } 2354 spin_unlock(&nm_i->nid_list_lock); 2355 2356 if (need_free) 2357 kmem_cache_free(free_nid_slab, i); 2358 } 2359 2360 static int scan_nat_page(struct f2fs_sb_info *sbi, 2361 struct page *nat_page, nid_t start_nid) 2362 { 2363 struct f2fs_nm_info *nm_i = NM_I(sbi); 2364 struct f2fs_nat_block *nat_blk = page_address(nat_page); 2365 block_t blk_addr; 2366 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid); 2367 int i; 2368 2369 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap); 2370 2371 i = start_nid % NAT_ENTRY_PER_BLOCK; 2372 2373 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) { 2374 if (unlikely(start_nid >= nm_i->max_nid)) 2375 break; 2376 2377 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr); 2378 2379 if (blk_addr == NEW_ADDR) 2380 return -EINVAL; 2381 2382 if (blk_addr == NULL_ADDR) { 2383 add_free_nid(sbi, start_nid, true, true); 2384 } else { 2385 spin_lock(&NM_I(sbi)->nid_list_lock); 2386 update_free_nid_bitmap(sbi, start_nid, false, true); 2387 spin_unlock(&NM_I(sbi)->nid_list_lock); 2388 } 2389 } 2390 2391 return 0; 2392 } 2393 2394 static void scan_curseg_cache(struct f2fs_sb_info *sbi) 2395 { 2396 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2397 struct f2fs_journal *journal = curseg->journal; 2398 int i; 2399 2400 down_read(&curseg->journal_rwsem); 2401 for (i = 0; i < nats_in_cursum(journal); i++) { 2402 block_t addr; 2403 nid_t nid; 2404 2405 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr); 2406 nid = le32_to_cpu(nid_in_journal(journal, i)); 2407 if (addr == NULL_ADDR) 2408 add_free_nid(sbi, nid, true, false); 2409 else 2410 remove_free_nid(sbi, nid); 2411 } 2412 up_read(&curseg->journal_rwsem); 2413 } 2414 2415 static void scan_free_nid_bits(struct f2fs_sb_info *sbi) 2416 { 2417 struct f2fs_nm_info *nm_i = NM_I(sbi); 2418 unsigned int i, idx; 2419 nid_t nid; 2420 2421 f2fs_down_read(&nm_i->nat_tree_lock); 2422 2423 for (i = 0; i < nm_i->nat_blocks; i++) { 2424 if (!test_bit_le(i, nm_i->nat_block_bitmap)) 2425 continue; 2426 if (!nm_i->free_nid_count[i]) 2427 continue; 2428 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) { 2429 idx = find_next_bit_le(nm_i->free_nid_bitmap[i], 2430 NAT_ENTRY_PER_BLOCK, idx); 2431 if (idx >= NAT_ENTRY_PER_BLOCK) 2432 break; 2433 2434 nid = i * NAT_ENTRY_PER_BLOCK + idx; 2435 add_free_nid(sbi, nid, true, false); 2436 2437 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS) 2438 goto out; 2439 } 2440 } 2441 out: 2442 scan_curseg_cache(sbi); 2443 2444 f2fs_up_read(&nm_i->nat_tree_lock); 2445 } 2446 2447 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi, 2448 bool sync, bool mount) 2449 { 2450 struct f2fs_nm_info *nm_i = NM_I(sbi); 2451 int i = 0, ret; 2452 nid_t nid = nm_i->next_scan_nid; 2453 2454 if (unlikely(nid >= nm_i->max_nid)) 2455 nid = 0; 2456 2457 if (unlikely(nid % NAT_ENTRY_PER_BLOCK)) 2458 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK; 2459 2460 /* Enough entries */ 2461 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2462 return 0; 2463 2464 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS)) 2465 return 0; 2466 2467 if (!mount) { 2468 /* try to find free nids in free_nid_bitmap */ 2469 scan_free_nid_bits(sbi); 2470 2471 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2472 return 0; 2473 } 2474 2475 /* readahead nat pages to be scanned */ 2476 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, 2477 META_NAT, true); 2478 2479 f2fs_down_read(&nm_i->nat_tree_lock); 2480 2481 while (1) { 2482 if (!test_bit_le(NAT_BLOCK_OFFSET(nid), 2483 nm_i->nat_block_bitmap)) { 2484 struct page *page = get_current_nat_page(sbi, nid); 2485 2486 if (IS_ERR(page)) { 2487 ret = PTR_ERR(page); 2488 } else { 2489 ret = scan_nat_page(sbi, page, nid); 2490 f2fs_put_page(page, 1); 2491 } 2492 2493 if (ret) { 2494 f2fs_up_read(&nm_i->nat_tree_lock); 2495 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it"); 2496 return ret; 2497 } 2498 } 2499 2500 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK)); 2501 if (unlikely(nid >= nm_i->max_nid)) 2502 nid = 0; 2503 2504 if (++i >= FREE_NID_PAGES) 2505 break; 2506 } 2507 2508 /* go to the next free nat pages to find free nids abundantly */ 2509 nm_i->next_scan_nid = nid; 2510 2511 /* find free nids from current sum_pages */ 2512 scan_curseg_cache(sbi); 2513 2514 f2fs_up_read(&nm_i->nat_tree_lock); 2515 2516 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid), 2517 nm_i->ra_nid_pages, META_NAT, false); 2518 2519 return 0; 2520 } 2521 2522 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount) 2523 { 2524 int ret; 2525 2526 mutex_lock(&NM_I(sbi)->build_lock); 2527 ret = __f2fs_build_free_nids(sbi, sync, mount); 2528 mutex_unlock(&NM_I(sbi)->build_lock); 2529 2530 return ret; 2531 } 2532 2533 /* 2534 * If this function returns success, caller can obtain a new nid 2535 * from second parameter of this function. 2536 * The returned nid could be used ino as well as nid when inode is created. 2537 */ 2538 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid) 2539 { 2540 struct f2fs_nm_info *nm_i = NM_I(sbi); 2541 struct free_nid *i = NULL; 2542 retry: 2543 if (time_to_inject(sbi, FAULT_ALLOC_NID)) 2544 return false; 2545 2546 spin_lock(&nm_i->nid_list_lock); 2547 2548 if (unlikely(nm_i->available_nids == 0)) { 2549 spin_unlock(&nm_i->nid_list_lock); 2550 return false; 2551 } 2552 2553 /* We should not use stale free nids created by f2fs_build_free_nids */ 2554 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) { 2555 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list)); 2556 i = list_first_entry(&nm_i->free_nid_list, 2557 struct free_nid, list); 2558 *nid = i->nid; 2559 2560 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID); 2561 nm_i->available_nids--; 2562 2563 update_free_nid_bitmap(sbi, *nid, false, false); 2564 2565 spin_unlock(&nm_i->nid_list_lock); 2566 return true; 2567 } 2568 spin_unlock(&nm_i->nid_list_lock); 2569 2570 /* Let's scan nat pages and its caches to get free nids */ 2571 if (!f2fs_build_free_nids(sbi, true, false)) 2572 goto retry; 2573 return false; 2574 } 2575 2576 /* 2577 * f2fs_alloc_nid() should be called prior to this function. 2578 */ 2579 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 2580 { 2581 struct f2fs_nm_info *nm_i = NM_I(sbi); 2582 struct free_nid *i; 2583 2584 spin_lock(&nm_i->nid_list_lock); 2585 i = __lookup_free_nid_list(nm_i, nid); 2586 f2fs_bug_on(sbi, !i); 2587 __remove_free_nid(sbi, i, PREALLOC_NID); 2588 spin_unlock(&nm_i->nid_list_lock); 2589 2590 kmem_cache_free(free_nid_slab, i); 2591 } 2592 2593 /* 2594 * f2fs_alloc_nid() should be called prior to this function. 2595 */ 2596 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 2597 { 2598 struct f2fs_nm_info *nm_i = NM_I(sbi); 2599 struct free_nid *i; 2600 bool need_free = false; 2601 2602 if (!nid) 2603 return; 2604 2605 spin_lock(&nm_i->nid_list_lock); 2606 i = __lookup_free_nid_list(nm_i, nid); 2607 f2fs_bug_on(sbi, !i); 2608 2609 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) { 2610 __remove_free_nid(sbi, i, PREALLOC_NID); 2611 need_free = true; 2612 } else { 2613 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID); 2614 } 2615 2616 nm_i->available_nids++; 2617 2618 update_free_nid_bitmap(sbi, nid, true, false); 2619 2620 spin_unlock(&nm_i->nid_list_lock); 2621 2622 if (need_free) 2623 kmem_cache_free(free_nid_slab, i); 2624 } 2625 2626 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink) 2627 { 2628 struct f2fs_nm_info *nm_i = NM_I(sbi); 2629 int nr = nr_shrink; 2630 2631 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2632 return 0; 2633 2634 if (!mutex_trylock(&nm_i->build_lock)) 2635 return 0; 2636 2637 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) { 2638 struct free_nid *i, *next; 2639 unsigned int batch = SHRINK_NID_BATCH_SIZE; 2640 2641 spin_lock(&nm_i->nid_list_lock); 2642 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) { 2643 if (!nr_shrink || !batch || 2644 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2645 break; 2646 __remove_free_nid(sbi, i, FREE_NID); 2647 kmem_cache_free(free_nid_slab, i); 2648 nr_shrink--; 2649 batch--; 2650 } 2651 spin_unlock(&nm_i->nid_list_lock); 2652 } 2653 2654 mutex_unlock(&nm_i->build_lock); 2655 2656 return nr - nr_shrink; 2657 } 2658 2659 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page) 2660 { 2661 void *src_addr, *dst_addr; 2662 size_t inline_size; 2663 struct page *ipage; 2664 struct f2fs_inode *ri; 2665 2666 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 2667 if (IS_ERR(ipage)) 2668 return PTR_ERR(ipage); 2669 2670 ri = F2FS_INODE(page); 2671 if (ri->i_inline & F2FS_INLINE_XATTR) { 2672 if (!f2fs_has_inline_xattr(inode)) { 2673 set_inode_flag(inode, FI_INLINE_XATTR); 2674 stat_inc_inline_xattr(inode); 2675 } 2676 } else { 2677 if (f2fs_has_inline_xattr(inode)) { 2678 stat_dec_inline_xattr(inode); 2679 clear_inode_flag(inode, FI_INLINE_XATTR); 2680 } 2681 goto update_inode; 2682 } 2683 2684 dst_addr = inline_xattr_addr(inode, ipage); 2685 src_addr = inline_xattr_addr(inode, page); 2686 inline_size = inline_xattr_size(inode); 2687 2688 f2fs_wait_on_page_writeback(ipage, NODE, true, true); 2689 memcpy(dst_addr, src_addr, inline_size); 2690 update_inode: 2691 f2fs_update_inode(inode, ipage); 2692 f2fs_put_page(ipage, 1); 2693 return 0; 2694 } 2695 2696 int f2fs_recover_xattr_data(struct inode *inode, struct page *page) 2697 { 2698 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2699 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid; 2700 nid_t new_xnid; 2701 struct dnode_of_data dn; 2702 struct node_info ni; 2703 struct page *xpage; 2704 int err; 2705 2706 if (!prev_xnid) 2707 goto recover_xnid; 2708 2709 /* 1: invalidate the previous xattr nid */ 2710 err = f2fs_get_node_info(sbi, prev_xnid, &ni, false); 2711 if (err) 2712 return err; 2713 2714 f2fs_invalidate_blocks(sbi, ni.blk_addr); 2715 dec_valid_node_count(sbi, inode, false); 2716 set_node_addr(sbi, &ni, NULL_ADDR, false); 2717 2718 recover_xnid: 2719 /* 2: update xattr nid in inode */ 2720 if (!f2fs_alloc_nid(sbi, &new_xnid)) 2721 return -ENOSPC; 2722 2723 set_new_dnode(&dn, inode, NULL, NULL, new_xnid); 2724 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET); 2725 if (IS_ERR(xpage)) { 2726 f2fs_alloc_nid_failed(sbi, new_xnid); 2727 return PTR_ERR(xpage); 2728 } 2729 2730 f2fs_alloc_nid_done(sbi, new_xnid); 2731 f2fs_update_inode_page(inode); 2732 2733 /* 3: update and set xattr node page dirty */ 2734 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE); 2735 2736 set_page_dirty(xpage); 2737 f2fs_put_page(xpage, 1); 2738 2739 return 0; 2740 } 2741 2742 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 2743 { 2744 struct f2fs_inode *src, *dst; 2745 nid_t ino = ino_of_node(page); 2746 struct node_info old_ni, new_ni; 2747 struct page *ipage; 2748 int err; 2749 2750 err = f2fs_get_node_info(sbi, ino, &old_ni, false); 2751 if (err) 2752 return err; 2753 2754 if (unlikely(old_ni.blk_addr != NULL_ADDR)) 2755 return -EINVAL; 2756 retry: 2757 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false); 2758 if (!ipage) { 2759 memalloc_retry_wait(GFP_NOFS); 2760 goto retry; 2761 } 2762 2763 /* Should not use this inode from free nid list */ 2764 remove_free_nid(sbi, ino); 2765 2766 if (!PageUptodate(ipage)) 2767 SetPageUptodate(ipage); 2768 fill_node_footer(ipage, ino, ino, 0, true); 2769 set_cold_node(ipage, false); 2770 2771 src = F2FS_INODE(page); 2772 dst = F2FS_INODE(ipage); 2773 2774 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext)); 2775 dst->i_size = 0; 2776 dst->i_blocks = cpu_to_le64(1); 2777 dst->i_links = cpu_to_le32(1); 2778 dst->i_xattr_nid = 0; 2779 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR); 2780 if (dst->i_inline & F2FS_EXTRA_ATTR) { 2781 dst->i_extra_isize = src->i_extra_isize; 2782 2783 if (f2fs_sb_has_flexible_inline_xattr(sbi) && 2784 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2785 i_inline_xattr_size)) 2786 dst->i_inline_xattr_size = src->i_inline_xattr_size; 2787 2788 if (f2fs_sb_has_project_quota(sbi) && 2789 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2790 i_projid)) 2791 dst->i_projid = src->i_projid; 2792 2793 if (f2fs_sb_has_inode_crtime(sbi) && 2794 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2795 i_crtime_nsec)) { 2796 dst->i_crtime = src->i_crtime; 2797 dst->i_crtime_nsec = src->i_crtime_nsec; 2798 } 2799 } 2800 2801 new_ni = old_ni; 2802 new_ni.ino = ino; 2803 2804 if (unlikely(inc_valid_node_count(sbi, NULL, true))) 2805 WARN_ON(1); 2806 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 2807 inc_valid_inode_count(sbi); 2808 set_page_dirty(ipage); 2809 f2fs_put_page(ipage, 1); 2810 return 0; 2811 } 2812 2813 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi, 2814 unsigned int segno, struct f2fs_summary_block *sum) 2815 { 2816 struct f2fs_node *rn; 2817 struct f2fs_summary *sum_entry; 2818 block_t addr; 2819 int i, idx, last_offset, nrpages; 2820 2821 /* scan the node segment */ 2822 last_offset = sbi->blocks_per_seg; 2823 addr = START_BLOCK(sbi, segno); 2824 sum_entry = &sum->entries[0]; 2825 2826 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) { 2827 nrpages = bio_max_segs(last_offset - i); 2828 2829 /* readahead node pages */ 2830 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true); 2831 2832 for (idx = addr; idx < addr + nrpages; idx++) { 2833 struct page *page = f2fs_get_tmp_page(sbi, idx); 2834 2835 if (IS_ERR(page)) 2836 return PTR_ERR(page); 2837 2838 rn = F2FS_NODE(page); 2839 sum_entry->nid = rn->footer.nid; 2840 sum_entry->version = 0; 2841 sum_entry->ofs_in_node = 0; 2842 sum_entry++; 2843 f2fs_put_page(page, 1); 2844 } 2845 2846 invalidate_mapping_pages(META_MAPPING(sbi), addr, 2847 addr + nrpages); 2848 } 2849 return 0; 2850 } 2851 2852 static void remove_nats_in_journal(struct f2fs_sb_info *sbi) 2853 { 2854 struct f2fs_nm_info *nm_i = NM_I(sbi); 2855 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2856 struct f2fs_journal *journal = curseg->journal; 2857 int i; 2858 2859 down_write(&curseg->journal_rwsem); 2860 for (i = 0; i < nats_in_cursum(journal); i++) { 2861 struct nat_entry *ne; 2862 struct f2fs_nat_entry raw_ne; 2863 nid_t nid = le32_to_cpu(nid_in_journal(journal, i)); 2864 2865 if (f2fs_check_nid_range(sbi, nid)) 2866 continue; 2867 2868 raw_ne = nat_in_journal(journal, i); 2869 2870 ne = __lookup_nat_cache(nm_i, nid); 2871 if (!ne) { 2872 ne = __alloc_nat_entry(sbi, nid, true); 2873 __init_nat_entry(nm_i, ne, &raw_ne, true); 2874 } 2875 2876 /* 2877 * if a free nat in journal has not been used after last 2878 * checkpoint, we should remove it from available nids, 2879 * since later we will add it again. 2880 */ 2881 if (!get_nat_flag(ne, IS_DIRTY) && 2882 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) { 2883 spin_lock(&nm_i->nid_list_lock); 2884 nm_i->available_nids--; 2885 spin_unlock(&nm_i->nid_list_lock); 2886 } 2887 2888 __set_nat_cache_dirty(nm_i, ne); 2889 } 2890 update_nats_in_cursum(journal, -i); 2891 up_write(&curseg->journal_rwsem); 2892 } 2893 2894 static void __adjust_nat_entry_set(struct nat_entry_set *nes, 2895 struct list_head *head, int max) 2896 { 2897 struct nat_entry_set *cur; 2898 2899 if (nes->entry_cnt >= max) 2900 goto add_out; 2901 2902 list_for_each_entry(cur, head, set_list) { 2903 if (cur->entry_cnt >= nes->entry_cnt) { 2904 list_add(&nes->set_list, cur->set_list.prev); 2905 return; 2906 } 2907 } 2908 add_out: 2909 list_add_tail(&nes->set_list, head); 2910 } 2911 2912 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs, 2913 unsigned int valid) 2914 { 2915 if (valid == 0) { 2916 __set_bit_le(nat_ofs, nm_i->empty_nat_bits); 2917 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2918 return; 2919 } 2920 2921 __clear_bit_le(nat_ofs, nm_i->empty_nat_bits); 2922 if (valid == NAT_ENTRY_PER_BLOCK) 2923 __set_bit_le(nat_ofs, nm_i->full_nat_bits); 2924 else 2925 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2926 } 2927 2928 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid, 2929 struct page *page) 2930 { 2931 struct f2fs_nm_info *nm_i = NM_I(sbi); 2932 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK; 2933 struct f2fs_nat_block *nat_blk = page_address(page); 2934 int valid = 0; 2935 int i = 0; 2936 2937 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 2938 return; 2939 2940 if (nat_index == 0) { 2941 valid = 1; 2942 i = 1; 2943 } 2944 for (; i < NAT_ENTRY_PER_BLOCK; i++) { 2945 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR) 2946 valid++; 2947 } 2948 2949 __update_nat_bits(nm_i, nat_index, valid); 2950 } 2951 2952 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi) 2953 { 2954 struct f2fs_nm_info *nm_i = NM_I(sbi); 2955 unsigned int nat_ofs; 2956 2957 f2fs_down_read(&nm_i->nat_tree_lock); 2958 2959 for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) { 2960 unsigned int valid = 0, nid_ofs = 0; 2961 2962 /* handle nid zero due to it should never be used */ 2963 if (unlikely(nat_ofs == 0)) { 2964 valid = 1; 2965 nid_ofs = 1; 2966 } 2967 2968 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) { 2969 if (!test_bit_le(nid_ofs, 2970 nm_i->free_nid_bitmap[nat_ofs])) 2971 valid++; 2972 } 2973 2974 __update_nat_bits(nm_i, nat_ofs, valid); 2975 } 2976 2977 f2fs_up_read(&nm_i->nat_tree_lock); 2978 } 2979 2980 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi, 2981 struct nat_entry_set *set, struct cp_control *cpc) 2982 { 2983 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2984 struct f2fs_journal *journal = curseg->journal; 2985 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK; 2986 bool to_journal = true; 2987 struct f2fs_nat_block *nat_blk; 2988 struct nat_entry *ne, *cur; 2989 struct page *page = NULL; 2990 2991 /* 2992 * there are two steps to flush nat entries: 2993 * #1, flush nat entries to journal in current hot data summary block. 2994 * #2, flush nat entries to nat page. 2995 */ 2996 if ((cpc->reason & CP_UMOUNT) || 2997 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL)) 2998 to_journal = false; 2999 3000 if (to_journal) { 3001 down_write(&curseg->journal_rwsem); 3002 } else { 3003 page = get_next_nat_page(sbi, start_nid); 3004 if (IS_ERR(page)) 3005 return PTR_ERR(page); 3006 3007 nat_blk = page_address(page); 3008 f2fs_bug_on(sbi, !nat_blk); 3009 } 3010 3011 /* flush dirty nats in nat entry set */ 3012 list_for_each_entry_safe(ne, cur, &set->entry_list, list) { 3013 struct f2fs_nat_entry *raw_ne; 3014 nid_t nid = nat_get_nid(ne); 3015 int offset; 3016 3017 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR); 3018 3019 if (to_journal) { 3020 offset = f2fs_lookup_journal_in_cursum(journal, 3021 NAT_JOURNAL, nid, 1); 3022 f2fs_bug_on(sbi, offset < 0); 3023 raw_ne = &nat_in_journal(journal, offset); 3024 nid_in_journal(journal, offset) = cpu_to_le32(nid); 3025 } else { 3026 raw_ne = &nat_blk->entries[nid - start_nid]; 3027 } 3028 raw_nat_from_node_info(raw_ne, &ne->ni); 3029 nat_reset_flag(ne); 3030 __clear_nat_cache_dirty(NM_I(sbi), set, ne); 3031 if (nat_get_blkaddr(ne) == NULL_ADDR) { 3032 add_free_nid(sbi, nid, false, true); 3033 } else { 3034 spin_lock(&NM_I(sbi)->nid_list_lock); 3035 update_free_nid_bitmap(sbi, nid, false, false); 3036 spin_unlock(&NM_I(sbi)->nid_list_lock); 3037 } 3038 } 3039 3040 if (to_journal) { 3041 up_write(&curseg->journal_rwsem); 3042 } else { 3043 update_nat_bits(sbi, start_nid, page); 3044 f2fs_put_page(page, 1); 3045 } 3046 3047 /* Allow dirty nats by node block allocation in write_begin */ 3048 if (!set->entry_cnt) { 3049 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set); 3050 kmem_cache_free(nat_entry_set_slab, set); 3051 } 3052 return 0; 3053 } 3054 3055 /* 3056 * This function is called during the checkpointing process. 3057 */ 3058 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3059 { 3060 struct f2fs_nm_info *nm_i = NM_I(sbi); 3061 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 3062 struct f2fs_journal *journal = curseg->journal; 3063 struct nat_entry_set *setvec[SETVEC_SIZE]; 3064 struct nat_entry_set *set, *tmp; 3065 unsigned int found; 3066 nid_t set_idx = 0; 3067 LIST_HEAD(sets); 3068 int err = 0; 3069 3070 /* 3071 * during unmount, let's flush nat_bits before checking 3072 * nat_cnt[DIRTY_NAT]. 3073 */ 3074 if (cpc->reason & CP_UMOUNT) { 3075 f2fs_down_write(&nm_i->nat_tree_lock); 3076 remove_nats_in_journal(sbi); 3077 f2fs_up_write(&nm_i->nat_tree_lock); 3078 } 3079 3080 if (!nm_i->nat_cnt[DIRTY_NAT]) 3081 return 0; 3082 3083 f2fs_down_write(&nm_i->nat_tree_lock); 3084 3085 /* 3086 * if there are no enough space in journal to store dirty nat 3087 * entries, remove all entries from journal and merge them 3088 * into nat entry set. 3089 */ 3090 if (cpc->reason & CP_UMOUNT || 3091 !__has_cursum_space(journal, 3092 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL)) 3093 remove_nats_in_journal(sbi); 3094 3095 while ((found = __gang_lookup_nat_set(nm_i, 3096 set_idx, SETVEC_SIZE, setvec))) { 3097 unsigned idx; 3098 3099 set_idx = setvec[found - 1]->set + 1; 3100 for (idx = 0; idx < found; idx++) 3101 __adjust_nat_entry_set(setvec[idx], &sets, 3102 MAX_NAT_JENTRIES(journal)); 3103 } 3104 3105 /* flush dirty nats in nat entry set */ 3106 list_for_each_entry_safe(set, tmp, &sets, set_list) { 3107 err = __flush_nat_entry_set(sbi, set, cpc); 3108 if (err) 3109 break; 3110 } 3111 3112 f2fs_up_write(&nm_i->nat_tree_lock); 3113 /* Allow dirty nats by node block allocation in write_begin */ 3114 3115 return err; 3116 } 3117 3118 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi) 3119 { 3120 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3121 struct f2fs_nm_info *nm_i = NM_I(sbi); 3122 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE; 3123 unsigned int i; 3124 __u64 cp_ver = cur_cp_version(ckpt); 3125 block_t nat_bits_addr; 3126 3127 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8); 3128 nm_i->nat_bits = f2fs_kvzalloc(sbi, 3129 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL); 3130 if (!nm_i->nat_bits) 3131 return -ENOMEM; 3132 3133 nm_i->full_nat_bits = nm_i->nat_bits + 8; 3134 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes; 3135 3136 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3137 return 0; 3138 3139 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg - 3140 nm_i->nat_bits_blocks; 3141 for (i = 0; i < nm_i->nat_bits_blocks; i++) { 3142 struct page *page; 3143 3144 page = f2fs_get_meta_page(sbi, nat_bits_addr++); 3145 if (IS_ERR(page)) 3146 return PTR_ERR(page); 3147 3148 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS), 3149 page_address(page), F2FS_BLKSIZE); 3150 f2fs_put_page(page, 1); 3151 } 3152 3153 cp_ver |= (cur_cp_crc(ckpt) << 32); 3154 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) { 3155 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG); 3156 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)", 3157 cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits)); 3158 return 0; 3159 } 3160 3161 f2fs_notice(sbi, "Found nat_bits in checkpoint"); 3162 return 0; 3163 } 3164 3165 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi) 3166 { 3167 struct f2fs_nm_info *nm_i = NM_I(sbi); 3168 unsigned int i = 0; 3169 nid_t nid, last_nid; 3170 3171 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3172 return; 3173 3174 for (i = 0; i < nm_i->nat_blocks; i++) { 3175 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i); 3176 if (i >= nm_i->nat_blocks) 3177 break; 3178 3179 __set_bit_le(i, nm_i->nat_block_bitmap); 3180 3181 nid = i * NAT_ENTRY_PER_BLOCK; 3182 last_nid = nid + NAT_ENTRY_PER_BLOCK; 3183 3184 spin_lock(&NM_I(sbi)->nid_list_lock); 3185 for (; nid < last_nid; nid++) 3186 update_free_nid_bitmap(sbi, nid, true, true); 3187 spin_unlock(&NM_I(sbi)->nid_list_lock); 3188 } 3189 3190 for (i = 0; i < nm_i->nat_blocks; i++) { 3191 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i); 3192 if (i >= nm_i->nat_blocks) 3193 break; 3194 3195 __set_bit_le(i, nm_i->nat_block_bitmap); 3196 } 3197 } 3198 3199 static int init_node_manager(struct f2fs_sb_info *sbi) 3200 { 3201 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 3202 struct f2fs_nm_info *nm_i = NM_I(sbi); 3203 unsigned char *version_bitmap; 3204 unsigned int nat_segs; 3205 int err; 3206 3207 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 3208 3209 /* segment_count_nat includes pair segment so divide to 2. */ 3210 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 3211 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 3212 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks; 3213 3214 /* not used nids: 0, node, meta, (and root counted as valid node) */ 3215 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count - 3216 F2FS_RESERVED_NODE_NUM; 3217 nm_i->nid_cnt[FREE_NID] = 0; 3218 nm_i->nid_cnt[PREALLOC_NID] = 0; 3219 nm_i->ram_thresh = DEF_RAM_THRESHOLD; 3220 nm_i->ra_nid_pages = DEF_RA_NID_PAGES; 3221 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD; 3222 nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS; 3223 3224 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC); 3225 INIT_LIST_HEAD(&nm_i->free_nid_list); 3226 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO); 3227 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO); 3228 INIT_LIST_HEAD(&nm_i->nat_entries); 3229 spin_lock_init(&nm_i->nat_list_lock); 3230 3231 mutex_init(&nm_i->build_lock); 3232 spin_lock_init(&nm_i->nid_list_lock); 3233 init_f2fs_rwsem(&nm_i->nat_tree_lock); 3234 3235 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 3236 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 3237 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 3238 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size, 3239 GFP_KERNEL); 3240 if (!nm_i->nat_bitmap) 3241 return -ENOMEM; 3242 3243 err = __get_nat_bitmaps(sbi); 3244 if (err) 3245 return err; 3246 3247 #ifdef CONFIG_F2FS_CHECK_FS 3248 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size, 3249 GFP_KERNEL); 3250 if (!nm_i->nat_bitmap_mir) 3251 return -ENOMEM; 3252 #endif 3253 3254 return 0; 3255 } 3256 3257 static int init_free_nid_cache(struct f2fs_sb_info *sbi) 3258 { 3259 struct f2fs_nm_info *nm_i = NM_I(sbi); 3260 int i; 3261 3262 nm_i->free_nid_bitmap = 3263 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *), 3264 nm_i->nat_blocks), 3265 GFP_KERNEL); 3266 if (!nm_i->free_nid_bitmap) 3267 return -ENOMEM; 3268 3269 for (i = 0; i < nm_i->nat_blocks; i++) { 3270 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi, 3271 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL); 3272 if (!nm_i->free_nid_bitmap[i]) 3273 return -ENOMEM; 3274 } 3275 3276 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8, 3277 GFP_KERNEL); 3278 if (!nm_i->nat_block_bitmap) 3279 return -ENOMEM; 3280 3281 nm_i->free_nid_count = 3282 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short), 3283 nm_i->nat_blocks), 3284 GFP_KERNEL); 3285 if (!nm_i->free_nid_count) 3286 return -ENOMEM; 3287 return 0; 3288 } 3289 3290 int f2fs_build_node_manager(struct f2fs_sb_info *sbi) 3291 { 3292 int err; 3293 3294 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info), 3295 GFP_KERNEL); 3296 if (!sbi->nm_info) 3297 return -ENOMEM; 3298 3299 err = init_node_manager(sbi); 3300 if (err) 3301 return err; 3302 3303 err = init_free_nid_cache(sbi); 3304 if (err) 3305 return err; 3306 3307 /* load free nid status from nat_bits table */ 3308 load_free_nid_bitmap(sbi); 3309 3310 return f2fs_build_free_nids(sbi, true, true); 3311 } 3312 3313 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi) 3314 { 3315 struct f2fs_nm_info *nm_i = NM_I(sbi); 3316 struct free_nid *i, *next_i; 3317 struct nat_entry *natvec[NATVEC_SIZE]; 3318 struct nat_entry_set *setvec[SETVEC_SIZE]; 3319 nid_t nid = 0; 3320 unsigned int found; 3321 3322 if (!nm_i) 3323 return; 3324 3325 /* destroy free nid list */ 3326 spin_lock(&nm_i->nid_list_lock); 3327 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 3328 __remove_free_nid(sbi, i, FREE_NID); 3329 spin_unlock(&nm_i->nid_list_lock); 3330 kmem_cache_free(free_nid_slab, i); 3331 spin_lock(&nm_i->nid_list_lock); 3332 } 3333 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]); 3334 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]); 3335 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list)); 3336 spin_unlock(&nm_i->nid_list_lock); 3337 3338 /* destroy nat cache */ 3339 f2fs_down_write(&nm_i->nat_tree_lock); 3340 while ((found = __gang_lookup_nat_cache(nm_i, 3341 nid, NATVEC_SIZE, natvec))) { 3342 unsigned idx; 3343 3344 nid = nat_get_nid(natvec[found - 1]) + 1; 3345 for (idx = 0; idx < found; idx++) { 3346 spin_lock(&nm_i->nat_list_lock); 3347 list_del(&natvec[idx]->list); 3348 spin_unlock(&nm_i->nat_list_lock); 3349 3350 __del_from_nat_cache(nm_i, natvec[idx]); 3351 } 3352 } 3353 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]); 3354 3355 /* destroy nat set cache */ 3356 nid = 0; 3357 while ((found = __gang_lookup_nat_set(nm_i, 3358 nid, SETVEC_SIZE, setvec))) { 3359 unsigned idx; 3360 3361 nid = setvec[found - 1]->set + 1; 3362 for (idx = 0; idx < found; idx++) { 3363 /* entry_cnt is not zero, when cp_error was occurred */ 3364 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list)); 3365 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set); 3366 kmem_cache_free(nat_entry_set_slab, setvec[idx]); 3367 } 3368 } 3369 f2fs_up_write(&nm_i->nat_tree_lock); 3370 3371 kvfree(nm_i->nat_block_bitmap); 3372 if (nm_i->free_nid_bitmap) { 3373 int i; 3374 3375 for (i = 0; i < nm_i->nat_blocks; i++) 3376 kvfree(nm_i->free_nid_bitmap[i]); 3377 kvfree(nm_i->free_nid_bitmap); 3378 } 3379 kvfree(nm_i->free_nid_count); 3380 3381 kvfree(nm_i->nat_bitmap); 3382 kvfree(nm_i->nat_bits); 3383 #ifdef CONFIG_F2FS_CHECK_FS 3384 kvfree(nm_i->nat_bitmap_mir); 3385 #endif 3386 sbi->nm_info = NULL; 3387 kfree(nm_i); 3388 } 3389 3390 int __init f2fs_create_node_manager_caches(void) 3391 { 3392 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry", 3393 sizeof(struct nat_entry)); 3394 if (!nat_entry_slab) 3395 goto fail; 3396 3397 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid", 3398 sizeof(struct free_nid)); 3399 if (!free_nid_slab) 3400 goto destroy_nat_entry; 3401 3402 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set", 3403 sizeof(struct nat_entry_set)); 3404 if (!nat_entry_set_slab) 3405 goto destroy_free_nid; 3406 3407 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry", 3408 sizeof(struct fsync_node_entry)); 3409 if (!fsync_node_entry_slab) 3410 goto destroy_nat_entry_set; 3411 return 0; 3412 3413 destroy_nat_entry_set: 3414 kmem_cache_destroy(nat_entry_set_slab); 3415 destroy_free_nid: 3416 kmem_cache_destroy(free_nid_slab); 3417 destroy_nat_entry: 3418 kmem_cache_destroy(nat_entry_slab); 3419 fail: 3420 return -ENOMEM; 3421 } 3422 3423 void f2fs_destroy_node_manager_caches(void) 3424 { 3425 kmem_cache_destroy(fsync_node_entry_slab); 3426 kmem_cache_destroy(nat_entry_set_slab); 3427 kmem_cache_destroy(free_nid_slab); 3428 kmem_cache_destroy(nat_entry_slab); 3429 } 3430