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 f2fs_put_page(page, 1); 948 return err; 949 } 950 951 return 1; 952 } 953 954 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs, 955 int ofs, int depth) 956 { 957 struct dnode_of_data rdn = *dn; 958 struct page *page; 959 struct f2fs_node *rn; 960 nid_t child_nid; 961 unsigned int child_nofs; 962 int freed = 0; 963 int i, ret; 964 965 if (dn->nid == 0) 966 return NIDS_PER_BLOCK + 1; 967 968 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr); 969 970 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 971 if (IS_ERR(page)) { 972 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page)); 973 return PTR_ERR(page); 974 } 975 976 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK); 977 978 rn = F2FS_NODE(page); 979 if (depth < 3) { 980 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) { 981 child_nid = le32_to_cpu(rn->in.nid[i]); 982 if (child_nid == 0) 983 continue; 984 rdn.nid = child_nid; 985 ret = truncate_dnode(&rdn); 986 if (ret < 0) 987 goto out_err; 988 if (set_nid(page, i, 0, false)) 989 dn->node_changed = true; 990 } 991 } else { 992 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1; 993 for (i = ofs; i < NIDS_PER_BLOCK; i++) { 994 child_nid = le32_to_cpu(rn->in.nid[i]); 995 if (child_nid == 0) { 996 child_nofs += NIDS_PER_BLOCK + 1; 997 continue; 998 } 999 rdn.nid = child_nid; 1000 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1); 1001 if (ret == (NIDS_PER_BLOCK + 1)) { 1002 if (set_nid(page, i, 0, false)) 1003 dn->node_changed = true; 1004 child_nofs += ret; 1005 } else if (ret < 0 && ret != -ENOENT) { 1006 goto out_err; 1007 } 1008 } 1009 freed = child_nofs; 1010 } 1011 1012 if (!ofs) { 1013 /* remove current indirect node */ 1014 dn->node_page = page; 1015 ret = truncate_node(dn); 1016 if (ret) 1017 goto out_err; 1018 freed++; 1019 } else { 1020 f2fs_put_page(page, 1); 1021 } 1022 trace_f2fs_truncate_nodes_exit(dn->inode, freed); 1023 return freed; 1024 1025 out_err: 1026 f2fs_put_page(page, 1); 1027 trace_f2fs_truncate_nodes_exit(dn->inode, ret); 1028 return ret; 1029 } 1030 1031 static int truncate_partial_nodes(struct dnode_of_data *dn, 1032 struct f2fs_inode *ri, int *offset, int depth) 1033 { 1034 struct page *pages[2]; 1035 nid_t nid[3]; 1036 nid_t child_nid; 1037 int err = 0; 1038 int i; 1039 int idx = depth - 2; 1040 1041 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1042 if (!nid[0]) 1043 return 0; 1044 1045 /* get indirect nodes in the path */ 1046 for (i = 0; i < idx + 1; i++) { 1047 /* reference count'll be increased */ 1048 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]); 1049 if (IS_ERR(pages[i])) { 1050 err = PTR_ERR(pages[i]); 1051 idx = i - 1; 1052 goto fail; 1053 } 1054 nid[i + 1] = get_nid(pages[i], offset[i + 1], false); 1055 } 1056 1057 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK); 1058 1059 /* free direct nodes linked to a partial indirect node */ 1060 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) { 1061 child_nid = get_nid(pages[idx], i, false); 1062 if (!child_nid) 1063 continue; 1064 dn->nid = child_nid; 1065 err = truncate_dnode(dn); 1066 if (err < 0) 1067 goto fail; 1068 if (set_nid(pages[idx], i, 0, false)) 1069 dn->node_changed = true; 1070 } 1071 1072 if (offset[idx + 1] == 0) { 1073 dn->node_page = pages[idx]; 1074 dn->nid = nid[idx]; 1075 err = truncate_node(dn); 1076 if (err) 1077 goto fail; 1078 } else { 1079 f2fs_put_page(pages[idx], 1); 1080 } 1081 offset[idx]++; 1082 offset[idx + 1] = 0; 1083 idx--; 1084 fail: 1085 for (i = idx; i >= 0; i--) 1086 f2fs_put_page(pages[i], 1); 1087 1088 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err); 1089 1090 return err; 1091 } 1092 1093 /* 1094 * All the block addresses of data and nodes should be nullified. 1095 */ 1096 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from) 1097 { 1098 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1099 int err = 0, cont = 1; 1100 int level, offset[4], noffset[4]; 1101 unsigned int nofs = 0; 1102 struct f2fs_inode *ri; 1103 struct dnode_of_data dn; 1104 struct page *page; 1105 1106 trace_f2fs_truncate_inode_blocks_enter(inode, from); 1107 1108 level = get_node_path(inode, from, offset, noffset); 1109 if (level < 0) { 1110 trace_f2fs_truncate_inode_blocks_exit(inode, level); 1111 return level; 1112 } 1113 1114 page = f2fs_get_node_page(sbi, inode->i_ino); 1115 if (IS_ERR(page)) { 1116 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page)); 1117 return PTR_ERR(page); 1118 } 1119 1120 set_new_dnode(&dn, inode, page, NULL, 0); 1121 unlock_page(page); 1122 1123 ri = F2FS_INODE(page); 1124 switch (level) { 1125 case 0: 1126 case 1: 1127 nofs = noffset[1]; 1128 break; 1129 case 2: 1130 nofs = noffset[1]; 1131 if (!offset[level - 1]) 1132 goto skip_partial; 1133 err = truncate_partial_nodes(&dn, ri, offset, level); 1134 if (err < 0 && err != -ENOENT) 1135 goto fail; 1136 nofs += 1 + NIDS_PER_BLOCK; 1137 break; 1138 case 3: 1139 nofs = 5 + 2 * NIDS_PER_BLOCK; 1140 if (!offset[level - 1]) 1141 goto skip_partial; 1142 err = truncate_partial_nodes(&dn, ri, offset, level); 1143 if (err < 0 && err != -ENOENT) 1144 goto fail; 1145 break; 1146 default: 1147 BUG(); 1148 } 1149 1150 skip_partial: 1151 while (cont) { 1152 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1153 switch (offset[0]) { 1154 case NODE_DIR1_BLOCK: 1155 case NODE_DIR2_BLOCK: 1156 err = truncate_dnode(&dn); 1157 break; 1158 1159 case NODE_IND1_BLOCK: 1160 case NODE_IND2_BLOCK: 1161 err = truncate_nodes(&dn, nofs, offset[1], 2); 1162 break; 1163 1164 case NODE_DIND_BLOCK: 1165 err = truncate_nodes(&dn, nofs, offset[1], 3); 1166 cont = 0; 1167 break; 1168 1169 default: 1170 BUG(); 1171 } 1172 if (err < 0 && err != -ENOENT) 1173 goto fail; 1174 if (offset[1] == 0 && 1175 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) { 1176 lock_page(page); 1177 BUG_ON(page->mapping != NODE_MAPPING(sbi)); 1178 f2fs_wait_on_page_writeback(page, NODE, true, true); 1179 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0; 1180 set_page_dirty(page); 1181 unlock_page(page); 1182 } 1183 offset[1] = 0; 1184 offset[0]++; 1185 nofs += err; 1186 } 1187 fail: 1188 f2fs_put_page(page, 0); 1189 trace_f2fs_truncate_inode_blocks_exit(inode, err); 1190 return err > 0 ? 0 : err; 1191 } 1192 1193 /* caller must lock inode page */ 1194 int f2fs_truncate_xattr_node(struct inode *inode) 1195 { 1196 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1197 nid_t nid = F2FS_I(inode)->i_xattr_nid; 1198 struct dnode_of_data dn; 1199 struct page *npage; 1200 int err; 1201 1202 if (!nid) 1203 return 0; 1204 1205 npage = f2fs_get_node_page(sbi, nid); 1206 if (IS_ERR(npage)) 1207 return PTR_ERR(npage); 1208 1209 set_new_dnode(&dn, inode, NULL, npage, nid); 1210 err = truncate_node(&dn); 1211 if (err) { 1212 f2fs_put_page(npage, 1); 1213 return err; 1214 } 1215 1216 f2fs_i_xnid_write(inode, 0); 1217 1218 return 0; 1219 } 1220 1221 /* 1222 * Caller should grab and release a rwsem by calling f2fs_lock_op() and 1223 * f2fs_unlock_op(). 1224 */ 1225 int f2fs_remove_inode_page(struct inode *inode) 1226 { 1227 struct dnode_of_data dn; 1228 int err; 1229 1230 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1231 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE); 1232 if (err) 1233 return err; 1234 1235 err = f2fs_truncate_xattr_node(inode); 1236 if (err) { 1237 f2fs_put_dnode(&dn); 1238 return err; 1239 } 1240 1241 /* remove potential inline_data blocks */ 1242 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1243 S_ISLNK(inode->i_mode)) 1244 f2fs_truncate_data_blocks_range(&dn, 1); 1245 1246 /* 0 is possible, after f2fs_new_inode() has failed */ 1247 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) { 1248 f2fs_put_dnode(&dn); 1249 return -EIO; 1250 } 1251 1252 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) { 1253 f2fs_warn(F2FS_I_SB(inode), 1254 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu", 1255 inode->i_ino, (unsigned long long)inode->i_blocks); 1256 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK); 1257 } 1258 1259 /* will put inode & node pages */ 1260 err = truncate_node(&dn); 1261 if (err) { 1262 f2fs_put_dnode(&dn); 1263 return err; 1264 } 1265 return 0; 1266 } 1267 1268 struct page *f2fs_new_inode_page(struct inode *inode) 1269 { 1270 struct dnode_of_data dn; 1271 1272 /* allocate inode page for new inode */ 1273 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1274 1275 /* caller should f2fs_put_page(page, 1); */ 1276 return f2fs_new_node_page(&dn, 0); 1277 } 1278 1279 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs) 1280 { 1281 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1282 struct node_info new_ni; 1283 struct page *page; 1284 int err; 1285 1286 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1287 return ERR_PTR(-EPERM); 1288 1289 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false); 1290 if (!page) 1291 return ERR_PTR(-ENOMEM); 1292 1293 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs)))) 1294 goto fail; 1295 1296 #ifdef CONFIG_F2FS_CHECK_FS 1297 err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false); 1298 if (err) { 1299 dec_valid_node_count(sbi, dn->inode, !ofs); 1300 goto fail; 1301 } 1302 if (unlikely(new_ni.blk_addr != NULL_ADDR)) { 1303 err = -EFSCORRUPTED; 1304 set_sbi_flag(sbi, SBI_NEED_FSCK); 1305 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 1306 goto fail; 1307 } 1308 #endif 1309 new_ni.nid = dn->nid; 1310 new_ni.ino = dn->inode->i_ino; 1311 new_ni.blk_addr = NULL_ADDR; 1312 new_ni.flag = 0; 1313 new_ni.version = 0; 1314 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 1315 1316 f2fs_wait_on_page_writeback(page, NODE, true, true); 1317 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 1318 set_cold_node(page, S_ISDIR(dn->inode->i_mode)); 1319 if (!PageUptodate(page)) 1320 SetPageUptodate(page); 1321 if (set_page_dirty(page)) 1322 dn->node_changed = true; 1323 1324 if (f2fs_has_xattr_block(ofs)) 1325 f2fs_i_xnid_write(dn->inode, dn->nid); 1326 1327 if (ofs == 0) 1328 inc_valid_inode_count(sbi); 1329 return page; 1330 1331 fail: 1332 clear_node_page_dirty(page); 1333 f2fs_put_page(page, 1); 1334 return ERR_PTR(err); 1335 } 1336 1337 /* 1338 * Caller should do after getting the following values. 1339 * 0: f2fs_put_page(page, 0) 1340 * LOCKED_PAGE or error: f2fs_put_page(page, 1) 1341 */ 1342 static int read_node_page(struct page *page, blk_opf_t op_flags) 1343 { 1344 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1345 struct node_info ni; 1346 struct f2fs_io_info fio = { 1347 .sbi = sbi, 1348 .type = NODE, 1349 .op = REQ_OP_READ, 1350 .op_flags = op_flags, 1351 .page = page, 1352 .encrypted_page = NULL, 1353 }; 1354 int err; 1355 1356 if (PageUptodate(page)) { 1357 if (!f2fs_inode_chksum_verify(sbi, page)) { 1358 ClearPageUptodate(page); 1359 return -EFSBADCRC; 1360 } 1361 return LOCKED_PAGE; 1362 } 1363 1364 err = f2fs_get_node_info(sbi, page->index, &ni, false); 1365 if (err) 1366 return err; 1367 1368 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */ 1369 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) { 1370 ClearPageUptodate(page); 1371 return -ENOENT; 1372 } 1373 1374 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr; 1375 1376 err = f2fs_submit_page_bio(&fio); 1377 1378 if (!err) 1379 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE); 1380 1381 return err; 1382 } 1383 1384 /* 1385 * Readahead a node page 1386 */ 1387 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 1388 { 1389 struct page *apage; 1390 int err; 1391 1392 if (!nid) 1393 return; 1394 if (f2fs_check_nid_range(sbi, nid)) 1395 return; 1396 1397 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid); 1398 if (apage) 1399 return; 1400 1401 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1402 if (!apage) 1403 return; 1404 1405 err = read_node_page(apage, REQ_RAHEAD); 1406 f2fs_put_page(apage, err ? 1 : 0); 1407 } 1408 1409 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid, 1410 struct page *parent, int start) 1411 { 1412 struct page *page; 1413 int err; 1414 1415 if (!nid) 1416 return ERR_PTR(-ENOENT); 1417 if (f2fs_check_nid_range(sbi, nid)) 1418 return ERR_PTR(-EINVAL); 1419 repeat: 1420 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1421 if (!page) 1422 return ERR_PTR(-ENOMEM); 1423 1424 err = read_node_page(page, 0); 1425 if (err < 0) { 1426 goto out_put_err; 1427 } else if (err == LOCKED_PAGE) { 1428 err = 0; 1429 goto page_hit; 1430 } 1431 1432 if (parent) 1433 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE); 1434 1435 lock_page(page); 1436 1437 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1438 f2fs_put_page(page, 1); 1439 goto repeat; 1440 } 1441 1442 if (unlikely(!PageUptodate(page))) { 1443 err = -EIO; 1444 goto out_err; 1445 } 1446 1447 if (!f2fs_inode_chksum_verify(sbi, page)) { 1448 err = -EFSBADCRC; 1449 goto out_err; 1450 } 1451 page_hit: 1452 if (likely(nid == nid_of_node(page))) 1453 return page; 1454 1455 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]", 1456 nid, nid_of_node(page), ino_of_node(page), 1457 ofs_of_node(page), cpver_of_node(page), 1458 next_blkaddr_of_node(page)); 1459 set_sbi_flag(sbi, SBI_NEED_FSCK); 1460 err = -EINVAL; 1461 out_err: 1462 ClearPageUptodate(page); 1463 out_put_err: 1464 /* ENOENT comes from read_node_page which is not an error. */ 1465 if (err != -ENOENT) 1466 f2fs_handle_page_eio(sbi, page->index, NODE); 1467 f2fs_put_page(page, 1); 1468 return ERR_PTR(err); 1469 } 1470 1471 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 1472 { 1473 return __get_node_page(sbi, nid, NULL, 0); 1474 } 1475 1476 struct page *f2fs_get_node_page_ra(struct page *parent, int start) 1477 { 1478 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 1479 nid_t nid = get_nid(parent, start, false); 1480 1481 return __get_node_page(sbi, nid, parent, start); 1482 } 1483 1484 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino) 1485 { 1486 struct inode *inode; 1487 struct page *page; 1488 int ret; 1489 1490 /* should flush inline_data before evict_inode */ 1491 inode = ilookup(sbi->sb, ino); 1492 if (!inode) 1493 return; 1494 1495 page = f2fs_pagecache_get_page(inode->i_mapping, 0, 1496 FGP_LOCK|FGP_NOWAIT, 0); 1497 if (!page) 1498 goto iput_out; 1499 1500 if (!PageUptodate(page)) 1501 goto page_out; 1502 1503 if (!PageDirty(page)) 1504 goto page_out; 1505 1506 if (!clear_page_dirty_for_io(page)) 1507 goto page_out; 1508 1509 ret = f2fs_write_inline_data(inode, page); 1510 inode_dec_dirty_pages(inode); 1511 f2fs_remove_dirty_inode(inode); 1512 if (ret) 1513 set_page_dirty(page); 1514 page_out: 1515 f2fs_put_page(page, 1); 1516 iput_out: 1517 iput(inode); 1518 } 1519 1520 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino) 1521 { 1522 pgoff_t index; 1523 struct folio_batch fbatch; 1524 struct page *last_page = NULL; 1525 int nr_folios; 1526 1527 folio_batch_init(&fbatch); 1528 index = 0; 1529 1530 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1531 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1532 &fbatch))) { 1533 int i; 1534 1535 for (i = 0; i < nr_folios; i++) { 1536 struct page *page = &fbatch.folios[i]->page; 1537 1538 if (unlikely(f2fs_cp_error(sbi))) { 1539 f2fs_put_page(last_page, 0); 1540 folio_batch_release(&fbatch); 1541 return ERR_PTR(-EIO); 1542 } 1543 1544 if (!IS_DNODE(page) || !is_cold_node(page)) 1545 continue; 1546 if (ino_of_node(page) != ino) 1547 continue; 1548 1549 lock_page(page); 1550 1551 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1552 continue_unlock: 1553 unlock_page(page); 1554 continue; 1555 } 1556 if (ino_of_node(page) != ino) 1557 goto continue_unlock; 1558 1559 if (!PageDirty(page)) { 1560 /* someone wrote it for us */ 1561 goto continue_unlock; 1562 } 1563 1564 if (last_page) 1565 f2fs_put_page(last_page, 0); 1566 1567 get_page(page); 1568 last_page = page; 1569 unlock_page(page); 1570 } 1571 folio_batch_release(&fbatch); 1572 cond_resched(); 1573 } 1574 return last_page; 1575 } 1576 1577 static int __write_node_page(struct page *page, bool atomic, bool *submitted, 1578 struct writeback_control *wbc, bool do_balance, 1579 enum iostat_type io_type, unsigned int *seq_id) 1580 { 1581 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1582 nid_t nid; 1583 struct node_info ni; 1584 struct f2fs_io_info fio = { 1585 .sbi = sbi, 1586 .ino = ino_of_node(page), 1587 .type = NODE, 1588 .op = REQ_OP_WRITE, 1589 .op_flags = wbc_to_write_flags(wbc), 1590 .page = page, 1591 .encrypted_page = NULL, 1592 .submitted = 0, 1593 .io_type = io_type, 1594 .io_wbc = wbc, 1595 }; 1596 unsigned int seq; 1597 1598 trace_f2fs_writepage(page, NODE); 1599 1600 if (unlikely(f2fs_cp_error(sbi))) { 1601 /* keep node pages in remount-ro mode */ 1602 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY) 1603 goto redirty_out; 1604 ClearPageUptodate(page); 1605 dec_page_count(sbi, F2FS_DIRTY_NODES); 1606 unlock_page(page); 1607 return 0; 1608 } 1609 1610 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1611 goto redirty_out; 1612 1613 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 1614 wbc->sync_mode == WB_SYNC_NONE && 1615 IS_DNODE(page) && is_cold_node(page)) 1616 goto redirty_out; 1617 1618 /* get old block addr of this node page */ 1619 nid = nid_of_node(page); 1620 f2fs_bug_on(sbi, page->index != nid); 1621 1622 if (f2fs_get_node_info(sbi, nid, &ni, !do_balance)) 1623 goto redirty_out; 1624 1625 if (wbc->for_reclaim) { 1626 if (!f2fs_down_read_trylock(&sbi->node_write)) 1627 goto redirty_out; 1628 } else { 1629 f2fs_down_read(&sbi->node_write); 1630 } 1631 1632 /* This page is already truncated */ 1633 if (unlikely(ni.blk_addr == NULL_ADDR)) { 1634 ClearPageUptodate(page); 1635 dec_page_count(sbi, F2FS_DIRTY_NODES); 1636 f2fs_up_read(&sbi->node_write); 1637 unlock_page(page); 1638 return 0; 1639 } 1640 1641 if (__is_valid_data_blkaddr(ni.blk_addr) && 1642 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, 1643 DATA_GENERIC_ENHANCE)) { 1644 f2fs_up_read(&sbi->node_write); 1645 goto redirty_out; 1646 } 1647 1648 if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi)) 1649 fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 1650 1651 /* should add to global list before clearing PAGECACHE status */ 1652 if (f2fs_in_warm_node_list(sbi, page)) { 1653 seq = f2fs_add_fsync_node_entry(sbi, page); 1654 if (seq_id) 1655 *seq_id = seq; 1656 } 1657 1658 set_page_writeback(page); 1659 1660 fio.old_blkaddr = ni.blk_addr; 1661 f2fs_do_write_node_page(nid, &fio); 1662 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page)); 1663 dec_page_count(sbi, F2FS_DIRTY_NODES); 1664 f2fs_up_read(&sbi->node_write); 1665 1666 if (wbc->for_reclaim) { 1667 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE); 1668 submitted = NULL; 1669 } 1670 1671 unlock_page(page); 1672 1673 if (unlikely(f2fs_cp_error(sbi))) { 1674 f2fs_submit_merged_write(sbi, NODE); 1675 submitted = NULL; 1676 } 1677 if (submitted) 1678 *submitted = fio.submitted; 1679 1680 if (do_balance) 1681 f2fs_balance_fs(sbi, false); 1682 return 0; 1683 1684 redirty_out: 1685 redirty_page_for_writepage(wbc, page); 1686 return AOP_WRITEPAGE_ACTIVATE; 1687 } 1688 1689 int f2fs_move_node_page(struct page *node_page, int gc_type) 1690 { 1691 int err = 0; 1692 1693 if (gc_type == FG_GC) { 1694 struct writeback_control wbc = { 1695 .sync_mode = WB_SYNC_ALL, 1696 .nr_to_write = 1, 1697 .for_reclaim = 0, 1698 }; 1699 1700 f2fs_wait_on_page_writeback(node_page, NODE, true, true); 1701 1702 set_page_dirty(node_page); 1703 1704 if (!clear_page_dirty_for_io(node_page)) { 1705 err = -EAGAIN; 1706 goto out_page; 1707 } 1708 1709 if (__write_node_page(node_page, false, NULL, 1710 &wbc, false, FS_GC_NODE_IO, NULL)) { 1711 err = -EAGAIN; 1712 unlock_page(node_page); 1713 } 1714 goto release_page; 1715 } else { 1716 /* set page dirty and write it */ 1717 if (!PageWriteback(node_page)) 1718 set_page_dirty(node_page); 1719 } 1720 out_page: 1721 unlock_page(node_page); 1722 release_page: 1723 f2fs_put_page(node_page, 0); 1724 return err; 1725 } 1726 1727 static int f2fs_write_node_page(struct page *page, 1728 struct writeback_control *wbc) 1729 { 1730 return __write_node_page(page, false, NULL, wbc, false, 1731 FS_NODE_IO, NULL); 1732 } 1733 1734 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode, 1735 struct writeback_control *wbc, bool atomic, 1736 unsigned int *seq_id) 1737 { 1738 pgoff_t index; 1739 struct folio_batch fbatch; 1740 int ret = 0; 1741 struct page *last_page = NULL; 1742 bool marked = false; 1743 nid_t ino = inode->i_ino; 1744 int nr_folios; 1745 int nwritten = 0; 1746 1747 if (atomic) { 1748 last_page = last_fsync_dnode(sbi, ino); 1749 if (IS_ERR_OR_NULL(last_page)) 1750 return PTR_ERR_OR_ZERO(last_page); 1751 } 1752 retry: 1753 folio_batch_init(&fbatch); 1754 index = 0; 1755 1756 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1757 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1758 &fbatch))) { 1759 int i; 1760 1761 for (i = 0; i < nr_folios; i++) { 1762 struct page *page = &fbatch.folios[i]->page; 1763 bool submitted = false; 1764 1765 if (unlikely(f2fs_cp_error(sbi))) { 1766 f2fs_put_page(last_page, 0); 1767 folio_batch_release(&fbatch); 1768 ret = -EIO; 1769 goto out; 1770 } 1771 1772 if (!IS_DNODE(page) || !is_cold_node(page)) 1773 continue; 1774 if (ino_of_node(page) != ino) 1775 continue; 1776 1777 lock_page(page); 1778 1779 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1780 continue_unlock: 1781 unlock_page(page); 1782 continue; 1783 } 1784 if (ino_of_node(page) != ino) 1785 goto continue_unlock; 1786 1787 if (!PageDirty(page) && page != last_page) { 1788 /* someone wrote it for us */ 1789 goto continue_unlock; 1790 } 1791 1792 f2fs_wait_on_page_writeback(page, NODE, true, true); 1793 1794 set_fsync_mark(page, 0); 1795 set_dentry_mark(page, 0); 1796 1797 if (!atomic || page == last_page) { 1798 set_fsync_mark(page, 1); 1799 percpu_counter_inc(&sbi->rf_node_block_count); 1800 if (IS_INODE(page)) { 1801 if (is_inode_flag_set(inode, 1802 FI_DIRTY_INODE)) 1803 f2fs_update_inode(inode, page); 1804 set_dentry_mark(page, 1805 f2fs_need_dentry_mark(sbi, ino)); 1806 } 1807 /* may be written by other thread */ 1808 if (!PageDirty(page)) 1809 set_page_dirty(page); 1810 } 1811 1812 if (!clear_page_dirty_for_io(page)) 1813 goto continue_unlock; 1814 1815 ret = __write_node_page(page, atomic && 1816 page == last_page, 1817 &submitted, wbc, true, 1818 FS_NODE_IO, seq_id); 1819 if (ret) { 1820 unlock_page(page); 1821 f2fs_put_page(last_page, 0); 1822 break; 1823 } else if (submitted) { 1824 nwritten++; 1825 } 1826 1827 if (page == last_page) { 1828 f2fs_put_page(page, 0); 1829 marked = true; 1830 break; 1831 } 1832 } 1833 folio_batch_release(&fbatch); 1834 cond_resched(); 1835 1836 if (ret || marked) 1837 break; 1838 } 1839 if (!ret && atomic && !marked) { 1840 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx", 1841 ino, last_page->index); 1842 lock_page(last_page); 1843 f2fs_wait_on_page_writeback(last_page, NODE, true, true); 1844 set_page_dirty(last_page); 1845 unlock_page(last_page); 1846 goto retry; 1847 } 1848 out: 1849 if (nwritten) 1850 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE); 1851 return ret ? -EIO : 0; 1852 } 1853 1854 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data) 1855 { 1856 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1857 bool clean; 1858 1859 if (inode->i_ino != ino) 1860 return 0; 1861 1862 if (!is_inode_flag_set(inode, FI_DIRTY_INODE)) 1863 return 0; 1864 1865 spin_lock(&sbi->inode_lock[DIRTY_META]); 1866 clean = list_empty(&F2FS_I(inode)->gdirty_list); 1867 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1868 1869 if (clean) 1870 return 0; 1871 1872 inode = igrab(inode); 1873 if (!inode) 1874 return 0; 1875 return 1; 1876 } 1877 1878 static bool flush_dirty_inode(struct page *page) 1879 { 1880 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1881 struct inode *inode; 1882 nid_t ino = ino_of_node(page); 1883 1884 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL); 1885 if (!inode) 1886 return false; 1887 1888 f2fs_update_inode(inode, page); 1889 unlock_page(page); 1890 1891 iput(inode); 1892 return true; 1893 } 1894 1895 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi) 1896 { 1897 pgoff_t index = 0; 1898 struct folio_batch fbatch; 1899 int nr_folios; 1900 1901 folio_batch_init(&fbatch); 1902 1903 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index, 1904 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1905 &fbatch))) { 1906 int i; 1907 1908 for (i = 0; i < nr_folios; i++) { 1909 struct page *page = &fbatch.folios[i]->page; 1910 1911 if (!IS_DNODE(page)) 1912 continue; 1913 1914 lock_page(page); 1915 1916 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1917 continue_unlock: 1918 unlock_page(page); 1919 continue; 1920 } 1921 1922 if (!PageDirty(page)) { 1923 /* someone wrote it for us */ 1924 goto continue_unlock; 1925 } 1926 1927 /* flush inline_data, if it's async context. */ 1928 if (page_private_inline(page)) { 1929 clear_page_private_inline(page); 1930 unlock_page(page); 1931 flush_inline_data(sbi, ino_of_node(page)); 1932 continue; 1933 } 1934 unlock_page(page); 1935 } 1936 folio_batch_release(&fbatch); 1937 cond_resched(); 1938 } 1939 } 1940 1941 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi, 1942 struct writeback_control *wbc, 1943 bool do_balance, enum iostat_type io_type) 1944 { 1945 pgoff_t index; 1946 struct folio_batch fbatch; 1947 int step = 0; 1948 int nwritten = 0; 1949 int ret = 0; 1950 int nr_folios, done = 0; 1951 1952 folio_batch_init(&fbatch); 1953 1954 next_step: 1955 index = 0; 1956 1957 while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), 1958 &index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY, 1959 &fbatch))) { 1960 int i; 1961 1962 for (i = 0; i < nr_folios; i++) { 1963 struct page *page = &fbatch.folios[i]->page; 1964 bool submitted = false; 1965 1966 /* give a priority to WB_SYNC threads */ 1967 if (atomic_read(&sbi->wb_sync_req[NODE]) && 1968 wbc->sync_mode == WB_SYNC_NONE) { 1969 done = 1; 1970 break; 1971 } 1972 1973 /* 1974 * flushing sequence with step: 1975 * 0. indirect nodes 1976 * 1. dentry dnodes 1977 * 2. file dnodes 1978 */ 1979 if (step == 0 && IS_DNODE(page)) 1980 continue; 1981 if (step == 1 && (!IS_DNODE(page) || 1982 is_cold_node(page))) 1983 continue; 1984 if (step == 2 && (!IS_DNODE(page) || 1985 !is_cold_node(page))) 1986 continue; 1987 lock_node: 1988 if (wbc->sync_mode == WB_SYNC_ALL) 1989 lock_page(page); 1990 else if (!trylock_page(page)) 1991 continue; 1992 1993 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1994 continue_unlock: 1995 unlock_page(page); 1996 continue; 1997 } 1998 1999 if (!PageDirty(page)) { 2000 /* someone wrote it for us */ 2001 goto continue_unlock; 2002 } 2003 2004 /* flush inline_data/inode, if it's async context. */ 2005 if (!do_balance) 2006 goto write_node; 2007 2008 /* flush inline_data */ 2009 if (page_private_inline(page)) { 2010 clear_page_private_inline(page); 2011 unlock_page(page); 2012 flush_inline_data(sbi, ino_of_node(page)); 2013 goto lock_node; 2014 } 2015 2016 /* flush dirty inode */ 2017 if (IS_INODE(page) && flush_dirty_inode(page)) 2018 goto lock_node; 2019 write_node: 2020 f2fs_wait_on_page_writeback(page, NODE, true, true); 2021 2022 if (!clear_page_dirty_for_io(page)) 2023 goto continue_unlock; 2024 2025 set_fsync_mark(page, 0); 2026 set_dentry_mark(page, 0); 2027 2028 ret = __write_node_page(page, false, &submitted, 2029 wbc, do_balance, io_type, NULL); 2030 if (ret) 2031 unlock_page(page); 2032 else if (submitted) 2033 nwritten++; 2034 2035 if (--wbc->nr_to_write == 0) 2036 break; 2037 } 2038 folio_batch_release(&fbatch); 2039 cond_resched(); 2040 2041 if (wbc->nr_to_write == 0) { 2042 step = 2; 2043 break; 2044 } 2045 } 2046 2047 if (step < 2) { 2048 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2049 wbc->sync_mode == WB_SYNC_NONE && step == 1) 2050 goto out; 2051 step++; 2052 goto next_step; 2053 } 2054 out: 2055 if (nwritten) 2056 f2fs_submit_merged_write(sbi, NODE); 2057 2058 if (unlikely(f2fs_cp_error(sbi))) 2059 return -EIO; 2060 return ret; 2061 } 2062 2063 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, 2064 unsigned int seq_id) 2065 { 2066 struct fsync_node_entry *fn; 2067 struct page *page; 2068 struct list_head *head = &sbi->fsync_node_list; 2069 unsigned long flags; 2070 unsigned int cur_seq_id = 0; 2071 2072 while (seq_id && cur_seq_id < seq_id) { 2073 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 2074 if (list_empty(head)) { 2075 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2076 break; 2077 } 2078 fn = list_first_entry(head, struct fsync_node_entry, list); 2079 if (fn->seq_id > seq_id) { 2080 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2081 break; 2082 } 2083 cur_seq_id = fn->seq_id; 2084 page = fn->page; 2085 get_page(page); 2086 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2087 2088 f2fs_wait_on_page_writeback(page, NODE, true, false); 2089 2090 put_page(page); 2091 } 2092 2093 return filemap_check_errors(NODE_MAPPING(sbi)); 2094 } 2095 2096 static int f2fs_write_node_pages(struct address_space *mapping, 2097 struct writeback_control *wbc) 2098 { 2099 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2100 struct blk_plug plug; 2101 long diff; 2102 2103 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2104 goto skip_write; 2105 2106 /* balancing f2fs's metadata in background */ 2107 f2fs_balance_fs_bg(sbi, true); 2108 2109 /* collect a number of dirty node pages and write together */ 2110 if (wbc->sync_mode != WB_SYNC_ALL && 2111 get_pages(sbi, F2FS_DIRTY_NODES) < 2112 nr_pages_to_skip(sbi, NODE)) 2113 goto skip_write; 2114 2115 if (wbc->sync_mode == WB_SYNC_ALL) 2116 atomic_inc(&sbi->wb_sync_req[NODE]); 2117 else if (atomic_read(&sbi->wb_sync_req[NODE])) { 2118 /* to avoid potential deadlock */ 2119 if (current->plug) 2120 blk_finish_plug(current->plug); 2121 goto skip_write; 2122 } 2123 2124 trace_f2fs_writepages(mapping->host, wbc, NODE); 2125 2126 diff = nr_pages_to_write(sbi, NODE, wbc); 2127 blk_start_plug(&plug); 2128 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO); 2129 blk_finish_plug(&plug); 2130 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); 2131 2132 if (wbc->sync_mode == WB_SYNC_ALL) 2133 atomic_dec(&sbi->wb_sync_req[NODE]); 2134 return 0; 2135 2136 skip_write: 2137 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES); 2138 trace_f2fs_writepages(mapping->host, wbc, NODE); 2139 return 0; 2140 } 2141 2142 static bool f2fs_dirty_node_folio(struct address_space *mapping, 2143 struct folio *folio) 2144 { 2145 trace_f2fs_set_page_dirty(&folio->page, NODE); 2146 2147 if (!folio_test_uptodate(folio)) 2148 folio_mark_uptodate(folio); 2149 #ifdef CONFIG_F2FS_CHECK_FS 2150 if (IS_INODE(&folio->page)) 2151 f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page); 2152 #endif 2153 if (filemap_dirty_folio(mapping, folio)) { 2154 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES); 2155 set_page_private_reference(&folio->page); 2156 return true; 2157 } 2158 return false; 2159 } 2160 2161 /* 2162 * Structure of the f2fs node operations 2163 */ 2164 const struct address_space_operations f2fs_node_aops = { 2165 .writepage = f2fs_write_node_page, 2166 .writepages = f2fs_write_node_pages, 2167 .dirty_folio = f2fs_dirty_node_folio, 2168 .invalidate_folio = f2fs_invalidate_folio, 2169 .release_folio = f2fs_release_folio, 2170 .migrate_folio = filemap_migrate_folio, 2171 }; 2172 2173 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i, 2174 nid_t n) 2175 { 2176 return radix_tree_lookup(&nm_i->free_nid_root, n); 2177 } 2178 2179 static int __insert_free_nid(struct f2fs_sb_info *sbi, 2180 struct free_nid *i) 2181 { 2182 struct f2fs_nm_info *nm_i = NM_I(sbi); 2183 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i); 2184 2185 if (err) 2186 return err; 2187 2188 nm_i->nid_cnt[FREE_NID]++; 2189 list_add_tail(&i->list, &nm_i->free_nid_list); 2190 return 0; 2191 } 2192 2193 static void __remove_free_nid(struct f2fs_sb_info *sbi, 2194 struct free_nid *i, enum nid_state state) 2195 { 2196 struct f2fs_nm_info *nm_i = NM_I(sbi); 2197 2198 f2fs_bug_on(sbi, state != i->state); 2199 nm_i->nid_cnt[state]--; 2200 if (state == FREE_NID) 2201 list_del(&i->list); 2202 radix_tree_delete(&nm_i->free_nid_root, i->nid); 2203 } 2204 2205 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i, 2206 enum nid_state org_state, enum nid_state dst_state) 2207 { 2208 struct f2fs_nm_info *nm_i = NM_I(sbi); 2209 2210 f2fs_bug_on(sbi, org_state != i->state); 2211 i->state = dst_state; 2212 nm_i->nid_cnt[org_state]--; 2213 nm_i->nid_cnt[dst_state]++; 2214 2215 switch (dst_state) { 2216 case PREALLOC_NID: 2217 list_del(&i->list); 2218 break; 2219 case FREE_NID: 2220 list_add_tail(&i->list, &nm_i->free_nid_list); 2221 break; 2222 default: 2223 BUG_ON(1); 2224 } 2225 } 2226 2227 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi) 2228 { 2229 struct f2fs_nm_info *nm_i = NM_I(sbi); 2230 unsigned int i; 2231 bool ret = true; 2232 2233 f2fs_down_read(&nm_i->nat_tree_lock); 2234 for (i = 0; i < nm_i->nat_blocks; i++) { 2235 if (!test_bit_le(i, nm_i->nat_block_bitmap)) { 2236 ret = false; 2237 break; 2238 } 2239 } 2240 f2fs_up_read(&nm_i->nat_tree_lock); 2241 2242 return ret; 2243 } 2244 2245 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid, 2246 bool set, bool build) 2247 { 2248 struct f2fs_nm_info *nm_i = NM_I(sbi); 2249 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid); 2250 unsigned int nid_ofs = nid - START_NID(nid); 2251 2252 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap)) 2253 return; 2254 2255 if (set) { 2256 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2257 return; 2258 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2259 nm_i->free_nid_count[nat_ofs]++; 2260 } else { 2261 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2262 return; 2263 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2264 if (!build) 2265 nm_i->free_nid_count[nat_ofs]--; 2266 } 2267 } 2268 2269 /* return if the nid is recognized as free */ 2270 static bool add_free_nid(struct f2fs_sb_info *sbi, 2271 nid_t nid, bool build, bool update) 2272 { 2273 struct f2fs_nm_info *nm_i = NM_I(sbi); 2274 struct free_nid *i, *e; 2275 struct nat_entry *ne; 2276 int err = -EINVAL; 2277 bool ret = false; 2278 2279 /* 0 nid should not be used */ 2280 if (unlikely(nid == 0)) 2281 return false; 2282 2283 if (unlikely(f2fs_check_nid_range(sbi, nid))) 2284 return false; 2285 2286 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL); 2287 i->nid = nid; 2288 i->state = FREE_NID; 2289 2290 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 2291 2292 spin_lock(&nm_i->nid_list_lock); 2293 2294 if (build) { 2295 /* 2296 * Thread A Thread B 2297 * - f2fs_create 2298 * - f2fs_new_inode 2299 * - f2fs_alloc_nid 2300 * - __insert_nid_to_list(PREALLOC_NID) 2301 * - f2fs_balance_fs_bg 2302 * - f2fs_build_free_nids 2303 * - __f2fs_build_free_nids 2304 * - scan_nat_page 2305 * - add_free_nid 2306 * - __lookup_nat_cache 2307 * - f2fs_add_link 2308 * - f2fs_init_inode_metadata 2309 * - f2fs_new_inode_page 2310 * - f2fs_new_node_page 2311 * - set_node_addr 2312 * - f2fs_alloc_nid_done 2313 * - __remove_nid_from_list(PREALLOC_NID) 2314 * - __insert_nid_to_list(FREE_NID) 2315 */ 2316 ne = __lookup_nat_cache(nm_i, nid); 2317 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) || 2318 nat_get_blkaddr(ne) != NULL_ADDR)) 2319 goto err_out; 2320 2321 e = __lookup_free_nid_list(nm_i, nid); 2322 if (e) { 2323 if (e->state == FREE_NID) 2324 ret = true; 2325 goto err_out; 2326 } 2327 } 2328 ret = true; 2329 err = __insert_free_nid(sbi, i); 2330 err_out: 2331 if (update) { 2332 update_free_nid_bitmap(sbi, nid, ret, build); 2333 if (!build) 2334 nm_i->available_nids++; 2335 } 2336 spin_unlock(&nm_i->nid_list_lock); 2337 radix_tree_preload_end(); 2338 2339 if (err) 2340 kmem_cache_free(free_nid_slab, i); 2341 return ret; 2342 } 2343 2344 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid) 2345 { 2346 struct f2fs_nm_info *nm_i = NM_I(sbi); 2347 struct free_nid *i; 2348 bool need_free = false; 2349 2350 spin_lock(&nm_i->nid_list_lock); 2351 i = __lookup_free_nid_list(nm_i, nid); 2352 if (i && i->state == FREE_NID) { 2353 __remove_free_nid(sbi, i, FREE_NID); 2354 need_free = true; 2355 } 2356 spin_unlock(&nm_i->nid_list_lock); 2357 2358 if (need_free) 2359 kmem_cache_free(free_nid_slab, i); 2360 } 2361 2362 static int scan_nat_page(struct f2fs_sb_info *sbi, 2363 struct page *nat_page, nid_t start_nid) 2364 { 2365 struct f2fs_nm_info *nm_i = NM_I(sbi); 2366 struct f2fs_nat_block *nat_blk = page_address(nat_page); 2367 block_t blk_addr; 2368 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid); 2369 int i; 2370 2371 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap); 2372 2373 i = start_nid % NAT_ENTRY_PER_BLOCK; 2374 2375 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) { 2376 if (unlikely(start_nid >= nm_i->max_nid)) 2377 break; 2378 2379 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr); 2380 2381 if (blk_addr == NEW_ADDR) 2382 return -EINVAL; 2383 2384 if (blk_addr == NULL_ADDR) { 2385 add_free_nid(sbi, start_nid, true, true); 2386 } else { 2387 spin_lock(&NM_I(sbi)->nid_list_lock); 2388 update_free_nid_bitmap(sbi, start_nid, false, true); 2389 spin_unlock(&NM_I(sbi)->nid_list_lock); 2390 } 2391 } 2392 2393 return 0; 2394 } 2395 2396 static void scan_curseg_cache(struct f2fs_sb_info *sbi) 2397 { 2398 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2399 struct f2fs_journal *journal = curseg->journal; 2400 int i; 2401 2402 down_read(&curseg->journal_rwsem); 2403 for (i = 0; i < nats_in_cursum(journal); i++) { 2404 block_t addr; 2405 nid_t nid; 2406 2407 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr); 2408 nid = le32_to_cpu(nid_in_journal(journal, i)); 2409 if (addr == NULL_ADDR) 2410 add_free_nid(sbi, nid, true, false); 2411 else 2412 remove_free_nid(sbi, nid); 2413 } 2414 up_read(&curseg->journal_rwsem); 2415 } 2416 2417 static void scan_free_nid_bits(struct f2fs_sb_info *sbi) 2418 { 2419 struct f2fs_nm_info *nm_i = NM_I(sbi); 2420 unsigned int i, idx; 2421 nid_t nid; 2422 2423 f2fs_down_read(&nm_i->nat_tree_lock); 2424 2425 for (i = 0; i < nm_i->nat_blocks; i++) { 2426 if (!test_bit_le(i, nm_i->nat_block_bitmap)) 2427 continue; 2428 if (!nm_i->free_nid_count[i]) 2429 continue; 2430 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) { 2431 idx = find_next_bit_le(nm_i->free_nid_bitmap[i], 2432 NAT_ENTRY_PER_BLOCK, idx); 2433 if (idx >= NAT_ENTRY_PER_BLOCK) 2434 break; 2435 2436 nid = i * NAT_ENTRY_PER_BLOCK + idx; 2437 add_free_nid(sbi, nid, true, false); 2438 2439 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS) 2440 goto out; 2441 } 2442 } 2443 out: 2444 scan_curseg_cache(sbi); 2445 2446 f2fs_up_read(&nm_i->nat_tree_lock); 2447 } 2448 2449 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi, 2450 bool sync, bool mount) 2451 { 2452 struct f2fs_nm_info *nm_i = NM_I(sbi); 2453 int i = 0, ret; 2454 nid_t nid = nm_i->next_scan_nid; 2455 2456 if (unlikely(nid >= nm_i->max_nid)) 2457 nid = 0; 2458 2459 if (unlikely(nid % NAT_ENTRY_PER_BLOCK)) 2460 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK; 2461 2462 /* Enough entries */ 2463 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2464 return 0; 2465 2466 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS)) 2467 return 0; 2468 2469 if (!mount) { 2470 /* try to find free nids in free_nid_bitmap */ 2471 scan_free_nid_bits(sbi); 2472 2473 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2474 return 0; 2475 } 2476 2477 /* readahead nat pages to be scanned */ 2478 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, 2479 META_NAT, true); 2480 2481 f2fs_down_read(&nm_i->nat_tree_lock); 2482 2483 while (1) { 2484 if (!test_bit_le(NAT_BLOCK_OFFSET(nid), 2485 nm_i->nat_block_bitmap)) { 2486 struct page *page = get_current_nat_page(sbi, nid); 2487 2488 if (IS_ERR(page)) { 2489 ret = PTR_ERR(page); 2490 } else { 2491 ret = scan_nat_page(sbi, page, nid); 2492 f2fs_put_page(page, 1); 2493 } 2494 2495 if (ret) { 2496 f2fs_up_read(&nm_i->nat_tree_lock); 2497 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it"); 2498 return ret; 2499 } 2500 } 2501 2502 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK)); 2503 if (unlikely(nid >= nm_i->max_nid)) 2504 nid = 0; 2505 2506 if (++i >= FREE_NID_PAGES) 2507 break; 2508 } 2509 2510 /* go to the next free nat pages to find free nids abundantly */ 2511 nm_i->next_scan_nid = nid; 2512 2513 /* find free nids from current sum_pages */ 2514 scan_curseg_cache(sbi); 2515 2516 f2fs_up_read(&nm_i->nat_tree_lock); 2517 2518 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid), 2519 nm_i->ra_nid_pages, META_NAT, false); 2520 2521 return 0; 2522 } 2523 2524 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount) 2525 { 2526 int ret; 2527 2528 mutex_lock(&NM_I(sbi)->build_lock); 2529 ret = __f2fs_build_free_nids(sbi, sync, mount); 2530 mutex_unlock(&NM_I(sbi)->build_lock); 2531 2532 return ret; 2533 } 2534 2535 /* 2536 * If this function returns success, caller can obtain a new nid 2537 * from second parameter of this function. 2538 * The returned nid could be used ino as well as nid when inode is created. 2539 */ 2540 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid) 2541 { 2542 struct f2fs_nm_info *nm_i = NM_I(sbi); 2543 struct free_nid *i = NULL; 2544 retry: 2545 if (time_to_inject(sbi, FAULT_ALLOC_NID)) 2546 return false; 2547 2548 spin_lock(&nm_i->nid_list_lock); 2549 2550 if (unlikely(nm_i->available_nids == 0)) { 2551 spin_unlock(&nm_i->nid_list_lock); 2552 return false; 2553 } 2554 2555 /* We should not use stale free nids created by f2fs_build_free_nids */ 2556 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) { 2557 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list)); 2558 i = list_first_entry(&nm_i->free_nid_list, 2559 struct free_nid, list); 2560 *nid = i->nid; 2561 2562 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID); 2563 nm_i->available_nids--; 2564 2565 update_free_nid_bitmap(sbi, *nid, false, false); 2566 2567 spin_unlock(&nm_i->nid_list_lock); 2568 return true; 2569 } 2570 spin_unlock(&nm_i->nid_list_lock); 2571 2572 /* Let's scan nat pages and its caches to get free nids */ 2573 if (!f2fs_build_free_nids(sbi, true, false)) 2574 goto retry; 2575 return false; 2576 } 2577 2578 /* 2579 * f2fs_alloc_nid() should be called prior to this function. 2580 */ 2581 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 2582 { 2583 struct f2fs_nm_info *nm_i = NM_I(sbi); 2584 struct free_nid *i; 2585 2586 spin_lock(&nm_i->nid_list_lock); 2587 i = __lookup_free_nid_list(nm_i, nid); 2588 f2fs_bug_on(sbi, !i); 2589 __remove_free_nid(sbi, i, PREALLOC_NID); 2590 spin_unlock(&nm_i->nid_list_lock); 2591 2592 kmem_cache_free(free_nid_slab, i); 2593 } 2594 2595 /* 2596 * f2fs_alloc_nid() should be called prior to this function. 2597 */ 2598 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 2599 { 2600 struct f2fs_nm_info *nm_i = NM_I(sbi); 2601 struct free_nid *i; 2602 bool need_free = false; 2603 2604 if (!nid) 2605 return; 2606 2607 spin_lock(&nm_i->nid_list_lock); 2608 i = __lookup_free_nid_list(nm_i, nid); 2609 f2fs_bug_on(sbi, !i); 2610 2611 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) { 2612 __remove_free_nid(sbi, i, PREALLOC_NID); 2613 need_free = true; 2614 } else { 2615 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID); 2616 } 2617 2618 nm_i->available_nids++; 2619 2620 update_free_nid_bitmap(sbi, nid, true, false); 2621 2622 spin_unlock(&nm_i->nid_list_lock); 2623 2624 if (need_free) 2625 kmem_cache_free(free_nid_slab, i); 2626 } 2627 2628 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink) 2629 { 2630 struct f2fs_nm_info *nm_i = NM_I(sbi); 2631 int nr = nr_shrink; 2632 2633 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2634 return 0; 2635 2636 if (!mutex_trylock(&nm_i->build_lock)) 2637 return 0; 2638 2639 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) { 2640 struct free_nid *i, *next; 2641 unsigned int batch = SHRINK_NID_BATCH_SIZE; 2642 2643 spin_lock(&nm_i->nid_list_lock); 2644 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) { 2645 if (!nr_shrink || !batch || 2646 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2647 break; 2648 __remove_free_nid(sbi, i, FREE_NID); 2649 kmem_cache_free(free_nid_slab, i); 2650 nr_shrink--; 2651 batch--; 2652 } 2653 spin_unlock(&nm_i->nid_list_lock); 2654 } 2655 2656 mutex_unlock(&nm_i->build_lock); 2657 2658 return nr - nr_shrink; 2659 } 2660 2661 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page) 2662 { 2663 void *src_addr, *dst_addr; 2664 size_t inline_size; 2665 struct page *ipage; 2666 struct f2fs_inode *ri; 2667 2668 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 2669 if (IS_ERR(ipage)) 2670 return PTR_ERR(ipage); 2671 2672 ri = F2FS_INODE(page); 2673 if (ri->i_inline & F2FS_INLINE_XATTR) { 2674 if (!f2fs_has_inline_xattr(inode)) { 2675 set_inode_flag(inode, FI_INLINE_XATTR); 2676 stat_inc_inline_xattr(inode); 2677 } 2678 } else { 2679 if (f2fs_has_inline_xattr(inode)) { 2680 stat_dec_inline_xattr(inode); 2681 clear_inode_flag(inode, FI_INLINE_XATTR); 2682 } 2683 goto update_inode; 2684 } 2685 2686 dst_addr = inline_xattr_addr(inode, ipage); 2687 src_addr = inline_xattr_addr(inode, page); 2688 inline_size = inline_xattr_size(inode); 2689 2690 f2fs_wait_on_page_writeback(ipage, NODE, true, true); 2691 memcpy(dst_addr, src_addr, inline_size); 2692 update_inode: 2693 f2fs_update_inode(inode, ipage); 2694 f2fs_put_page(ipage, 1); 2695 return 0; 2696 } 2697 2698 int f2fs_recover_xattr_data(struct inode *inode, struct page *page) 2699 { 2700 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2701 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid; 2702 nid_t new_xnid; 2703 struct dnode_of_data dn; 2704 struct node_info ni; 2705 struct page *xpage; 2706 int err; 2707 2708 if (!prev_xnid) 2709 goto recover_xnid; 2710 2711 /* 1: invalidate the previous xattr nid */ 2712 err = f2fs_get_node_info(sbi, prev_xnid, &ni, false); 2713 if (err) 2714 return err; 2715 2716 f2fs_invalidate_blocks(sbi, ni.blk_addr); 2717 dec_valid_node_count(sbi, inode, false); 2718 set_node_addr(sbi, &ni, NULL_ADDR, false); 2719 2720 recover_xnid: 2721 /* 2: update xattr nid in inode */ 2722 if (!f2fs_alloc_nid(sbi, &new_xnid)) 2723 return -ENOSPC; 2724 2725 set_new_dnode(&dn, inode, NULL, NULL, new_xnid); 2726 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET); 2727 if (IS_ERR(xpage)) { 2728 f2fs_alloc_nid_failed(sbi, new_xnid); 2729 return PTR_ERR(xpage); 2730 } 2731 2732 f2fs_alloc_nid_done(sbi, new_xnid); 2733 f2fs_update_inode_page(inode); 2734 2735 /* 3: update and set xattr node page dirty */ 2736 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE); 2737 2738 set_page_dirty(xpage); 2739 f2fs_put_page(xpage, 1); 2740 2741 return 0; 2742 } 2743 2744 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 2745 { 2746 struct f2fs_inode *src, *dst; 2747 nid_t ino = ino_of_node(page); 2748 struct node_info old_ni, new_ni; 2749 struct page *ipage; 2750 int err; 2751 2752 err = f2fs_get_node_info(sbi, ino, &old_ni, false); 2753 if (err) 2754 return err; 2755 2756 if (unlikely(old_ni.blk_addr != NULL_ADDR)) 2757 return -EINVAL; 2758 retry: 2759 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false); 2760 if (!ipage) { 2761 memalloc_retry_wait(GFP_NOFS); 2762 goto retry; 2763 } 2764 2765 /* Should not use this inode from free nid list */ 2766 remove_free_nid(sbi, ino); 2767 2768 if (!PageUptodate(ipage)) 2769 SetPageUptodate(ipage); 2770 fill_node_footer(ipage, ino, ino, 0, true); 2771 set_cold_node(ipage, false); 2772 2773 src = F2FS_INODE(page); 2774 dst = F2FS_INODE(ipage); 2775 2776 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext)); 2777 dst->i_size = 0; 2778 dst->i_blocks = cpu_to_le64(1); 2779 dst->i_links = cpu_to_le32(1); 2780 dst->i_xattr_nid = 0; 2781 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR); 2782 if (dst->i_inline & F2FS_EXTRA_ATTR) { 2783 dst->i_extra_isize = src->i_extra_isize; 2784 2785 if (f2fs_sb_has_flexible_inline_xattr(sbi) && 2786 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2787 i_inline_xattr_size)) 2788 dst->i_inline_xattr_size = src->i_inline_xattr_size; 2789 2790 if (f2fs_sb_has_project_quota(sbi) && 2791 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2792 i_projid)) 2793 dst->i_projid = src->i_projid; 2794 2795 if (f2fs_sb_has_inode_crtime(sbi) && 2796 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2797 i_crtime_nsec)) { 2798 dst->i_crtime = src->i_crtime; 2799 dst->i_crtime_nsec = src->i_crtime_nsec; 2800 } 2801 } 2802 2803 new_ni = old_ni; 2804 new_ni.ino = ino; 2805 2806 if (unlikely(inc_valid_node_count(sbi, NULL, true))) 2807 WARN_ON(1); 2808 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 2809 inc_valid_inode_count(sbi); 2810 set_page_dirty(ipage); 2811 f2fs_put_page(ipage, 1); 2812 return 0; 2813 } 2814 2815 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi, 2816 unsigned int segno, struct f2fs_summary_block *sum) 2817 { 2818 struct f2fs_node *rn; 2819 struct f2fs_summary *sum_entry; 2820 block_t addr; 2821 int i, idx, last_offset, nrpages; 2822 2823 /* scan the node segment */ 2824 last_offset = sbi->blocks_per_seg; 2825 addr = START_BLOCK(sbi, segno); 2826 sum_entry = &sum->entries[0]; 2827 2828 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) { 2829 nrpages = bio_max_segs(last_offset - i); 2830 2831 /* readahead node pages */ 2832 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true); 2833 2834 for (idx = addr; idx < addr + nrpages; idx++) { 2835 struct page *page = f2fs_get_tmp_page(sbi, idx); 2836 2837 if (IS_ERR(page)) 2838 return PTR_ERR(page); 2839 2840 rn = F2FS_NODE(page); 2841 sum_entry->nid = rn->footer.nid; 2842 sum_entry->version = 0; 2843 sum_entry->ofs_in_node = 0; 2844 sum_entry++; 2845 f2fs_put_page(page, 1); 2846 } 2847 2848 invalidate_mapping_pages(META_MAPPING(sbi), addr, 2849 addr + nrpages); 2850 } 2851 return 0; 2852 } 2853 2854 static void remove_nats_in_journal(struct f2fs_sb_info *sbi) 2855 { 2856 struct f2fs_nm_info *nm_i = NM_I(sbi); 2857 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2858 struct f2fs_journal *journal = curseg->journal; 2859 int i; 2860 2861 down_write(&curseg->journal_rwsem); 2862 for (i = 0; i < nats_in_cursum(journal); i++) { 2863 struct nat_entry *ne; 2864 struct f2fs_nat_entry raw_ne; 2865 nid_t nid = le32_to_cpu(nid_in_journal(journal, i)); 2866 2867 if (f2fs_check_nid_range(sbi, nid)) 2868 continue; 2869 2870 raw_ne = nat_in_journal(journal, i); 2871 2872 ne = __lookup_nat_cache(nm_i, nid); 2873 if (!ne) { 2874 ne = __alloc_nat_entry(sbi, nid, true); 2875 __init_nat_entry(nm_i, ne, &raw_ne, true); 2876 } 2877 2878 /* 2879 * if a free nat in journal has not been used after last 2880 * checkpoint, we should remove it from available nids, 2881 * since later we will add it again. 2882 */ 2883 if (!get_nat_flag(ne, IS_DIRTY) && 2884 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) { 2885 spin_lock(&nm_i->nid_list_lock); 2886 nm_i->available_nids--; 2887 spin_unlock(&nm_i->nid_list_lock); 2888 } 2889 2890 __set_nat_cache_dirty(nm_i, ne); 2891 } 2892 update_nats_in_cursum(journal, -i); 2893 up_write(&curseg->journal_rwsem); 2894 } 2895 2896 static void __adjust_nat_entry_set(struct nat_entry_set *nes, 2897 struct list_head *head, int max) 2898 { 2899 struct nat_entry_set *cur; 2900 2901 if (nes->entry_cnt >= max) 2902 goto add_out; 2903 2904 list_for_each_entry(cur, head, set_list) { 2905 if (cur->entry_cnt >= nes->entry_cnt) { 2906 list_add(&nes->set_list, cur->set_list.prev); 2907 return; 2908 } 2909 } 2910 add_out: 2911 list_add_tail(&nes->set_list, head); 2912 } 2913 2914 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs, 2915 unsigned int valid) 2916 { 2917 if (valid == 0) { 2918 __set_bit_le(nat_ofs, nm_i->empty_nat_bits); 2919 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2920 return; 2921 } 2922 2923 __clear_bit_le(nat_ofs, nm_i->empty_nat_bits); 2924 if (valid == NAT_ENTRY_PER_BLOCK) 2925 __set_bit_le(nat_ofs, nm_i->full_nat_bits); 2926 else 2927 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2928 } 2929 2930 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid, 2931 struct page *page) 2932 { 2933 struct f2fs_nm_info *nm_i = NM_I(sbi); 2934 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK; 2935 struct f2fs_nat_block *nat_blk = page_address(page); 2936 int valid = 0; 2937 int i = 0; 2938 2939 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 2940 return; 2941 2942 if (nat_index == 0) { 2943 valid = 1; 2944 i = 1; 2945 } 2946 for (; i < NAT_ENTRY_PER_BLOCK; i++) { 2947 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR) 2948 valid++; 2949 } 2950 2951 __update_nat_bits(nm_i, nat_index, valid); 2952 } 2953 2954 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi) 2955 { 2956 struct f2fs_nm_info *nm_i = NM_I(sbi); 2957 unsigned int nat_ofs; 2958 2959 f2fs_down_read(&nm_i->nat_tree_lock); 2960 2961 for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) { 2962 unsigned int valid = 0, nid_ofs = 0; 2963 2964 /* handle nid zero due to it should never be used */ 2965 if (unlikely(nat_ofs == 0)) { 2966 valid = 1; 2967 nid_ofs = 1; 2968 } 2969 2970 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) { 2971 if (!test_bit_le(nid_ofs, 2972 nm_i->free_nid_bitmap[nat_ofs])) 2973 valid++; 2974 } 2975 2976 __update_nat_bits(nm_i, nat_ofs, valid); 2977 } 2978 2979 f2fs_up_read(&nm_i->nat_tree_lock); 2980 } 2981 2982 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi, 2983 struct nat_entry_set *set, struct cp_control *cpc) 2984 { 2985 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2986 struct f2fs_journal *journal = curseg->journal; 2987 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK; 2988 bool to_journal = true; 2989 struct f2fs_nat_block *nat_blk; 2990 struct nat_entry *ne, *cur; 2991 struct page *page = NULL; 2992 2993 /* 2994 * there are two steps to flush nat entries: 2995 * #1, flush nat entries to journal in current hot data summary block. 2996 * #2, flush nat entries to nat page. 2997 */ 2998 if ((cpc->reason & CP_UMOUNT) || 2999 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL)) 3000 to_journal = false; 3001 3002 if (to_journal) { 3003 down_write(&curseg->journal_rwsem); 3004 } else { 3005 page = get_next_nat_page(sbi, start_nid); 3006 if (IS_ERR(page)) 3007 return PTR_ERR(page); 3008 3009 nat_blk = page_address(page); 3010 f2fs_bug_on(sbi, !nat_blk); 3011 } 3012 3013 /* flush dirty nats in nat entry set */ 3014 list_for_each_entry_safe(ne, cur, &set->entry_list, list) { 3015 struct f2fs_nat_entry *raw_ne; 3016 nid_t nid = nat_get_nid(ne); 3017 int offset; 3018 3019 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR); 3020 3021 if (to_journal) { 3022 offset = f2fs_lookup_journal_in_cursum(journal, 3023 NAT_JOURNAL, nid, 1); 3024 f2fs_bug_on(sbi, offset < 0); 3025 raw_ne = &nat_in_journal(journal, offset); 3026 nid_in_journal(journal, offset) = cpu_to_le32(nid); 3027 } else { 3028 raw_ne = &nat_blk->entries[nid - start_nid]; 3029 } 3030 raw_nat_from_node_info(raw_ne, &ne->ni); 3031 nat_reset_flag(ne); 3032 __clear_nat_cache_dirty(NM_I(sbi), set, ne); 3033 if (nat_get_blkaddr(ne) == NULL_ADDR) { 3034 add_free_nid(sbi, nid, false, true); 3035 } else { 3036 spin_lock(&NM_I(sbi)->nid_list_lock); 3037 update_free_nid_bitmap(sbi, nid, false, false); 3038 spin_unlock(&NM_I(sbi)->nid_list_lock); 3039 } 3040 } 3041 3042 if (to_journal) { 3043 up_write(&curseg->journal_rwsem); 3044 } else { 3045 update_nat_bits(sbi, start_nid, page); 3046 f2fs_put_page(page, 1); 3047 } 3048 3049 /* Allow dirty nats by node block allocation in write_begin */ 3050 if (!set->entry_cnt) { 3051 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set); 3052 kmem_cache_free(nat_entry_set_slab, set); 3053 } 3054 return 0; 3055 } 3056 3057 /* 3058 * This function is called during the checkpointing process. 3059 */ 3060 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3061 { 3062 struct f2fs_nm_info *nm_i = NM_I(sbi); 3063 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 3064 struct f2fs_journal *journal = curseg->journal; 3065 struct nat_entry_set *setvec[SETVEC_SIZE]; 3066 struct nat_entry_set *set, *tmp; 3067 unsigned int found; 3068 nid_t set_idx = 0; 3069 LIST_HEAD(sets); 3070 int err = 0; 3071 3072 /* 3073 * during unmount, let's flush nat_bits before checking 3074 * nat_cnt[DIRTY_NAT]. 3075 */ 3076 if (cpc->reason & CP_UMOUNT) { 3077 f2fs_down_write(&nm_i->nat_tree_lock); 3078 remove_nats_in_journal(sbi); 3079 f2fs_up_write(&nm_i->nat_tree_lock); 3080 } 3081 3082 if (!nm_i->nat_cnt[DIRTY_NAT]) 3083 return 0; 3084 3085 f2fs_down_write(&nm_i->nat_tree_lock); 3086 3087 /* 3088 * if there are no enough space in journal to store dirty nat 3089 * entries, remove all entries from journal and merge them 3090 * into nat entry set. 3091 */ 3092 if (cpc->reason & CP_UMOUNT || 3093 !__has_cursum_space(journal, 3094 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL)) 3095 remove_nats_in_journal(sbi); 3096 3097 while ((found = __gang_lookup_nat_set(nm_i, 3098 set_idx, SETVEC_SIZE, setvec))) { 3099 unsigned idx; 3100 3101 set_idx = setvec[found - 1]->set + 1; 3102 for (idx = 0; idx < found; idx++) 3103 __adjust_nat_entry_set(setvec[idx], &sets, 3104 MAX_NAT_JENTRIES(journal)); 3105 } 3106 3107 /* flush dirty nats in nat entry set */ 3108 list_for_each_entry_safe(set, tmp, &sets, set_list) { 3109 err = __flush_nat_entry_set(sbi, set, cpc); 3110 if (err) 3111 break; 3112 } 3113 3114 f2fs_up_write(&nm_i->nat_tree_lock); 3115 /* Allow dirty nats by node block allocation in write_begin */ 3116 3117 return err; 3118 } 3119 3120 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi) 3121 { 3122 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3123 struct f2fs_nm_info *nm_i = NM_I(sbi); 3124 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE; 3125 unsigned int i; 3126 __u64 cp_ver = cur_cp_version(ckpt); 3127 block_t nat_bits_addr; 3128 3129 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8); 3130 nm_i->nat_bits = f2fs_kvzalloc(sbi, 3131 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL); 3132 if (!nm_i->nat_bits) 3133 return -ENOMEM; 3134 3135 nm_i->full_nat_bits = nm_i->nat_bits + 8; 3136 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes; 3137 3138 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3139 return 0; 3140 3141 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg - 3142 nm_i->nat_bits_blocks; 3143 for (i = 0; i < nm_i->nat_bits_blocks; i++) { 3144 struct page *page; 3145 3146 page = f2fs_get_meta_page(sbi, nat_bits_addr++); 3147 if (IS_ERR(page)) 3148 return PTR_ERR(page); 3149 3150 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS), 3151 page_address(page), F2FS_BLKSIZE); 3152 f2fs_put_page(page, 1); 3153 } 3154 3155 cp_ver |= (cur_cp_crc(ckpt) << 32); 3156 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) { 3157 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG); 3158 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)", 3159 cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits)); 3160 return 0; 3161 } 3162 3163 f2fs_notice(sbi, "Found nat_bits in checkpoint"); 3164 return 0; 3165 } 3166 3167 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi) 3168 { 3169 struct f2fs_nm_info *nm_i = NM_I(sbi); 3170 unsigned int i = 0; 3171 nid_t nid, last_nid; 3172 3173 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3174 return; 3175 3176 for (i = 0; i < nm_i->nat_blocks; i++) { 3177 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i); 3178 if (i >= nm_i->nat_blocks) 3179 break; 3180 3181 __set_bit_le(i, nm_i->nat_block_bitmap); 3182 3183 nid = i * NAT_ENTRY_PER_BLOCK; 3184 last_nid = nid + NAT_ENTRY_PER_BLOCK; 3185 3186 spin_lock(&NM_I(sbi)->nid_list_lock); 3187 for (; nid < last_nid; nid++) 3188 update_free_nid_bitmap(sbi, nid, true, true); 3189 spin_unlock(&NM_I(sbi)->nid_list_lock); 3190 } 3191 3192 for (i = 0; i < nm_i->nat_blocks; i++) { 3193 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i); 3194 if (i >= nm_i->nat_blocks) 3195 break; 3196 3197 __set_bit_le(i, nm_i->nat_block_bitmap); 3198 } 3199 } 3200 3201 static int init_node_manager(struct f2fs_sb_info *sbi) 3202 { 3203 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 3204 struct f2fs_nm_info *nm_i = NM_I(sbi); 3205 unsigned char *version_bitmap; 3206 unsigned int nat_segs; 3207 int err; 3208 3209 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 3210 3211 /* segment_count_nat includes pair segment so divide to 2. */ 3212 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 3213 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 3214 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks; 3215 3216 /* not used nids: 0, node, meta, (and root counted as valid node) */ 3217 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count - 3218 F2FS_RESERVED_NODE_NUM; 3219 nm_i->nid_cnt[FREE_NID] = 0; 3220 nm_i->nid_cnt[PREALLOC_NID] = 0; 3221 nm_i->ram_thresh = DEF_RAM_THRESHOLD; 3222 nm_i->ra_nid_pages = DEF_RA_NID_PAGES; 3223 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD; 3224 nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS; 3225 3226 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC); 3227 INIT_LIST_HEAD(&nm_i->free_nid_list); 3228 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO); 3229 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO); 3230 INIT_LIST_HEAD(&nm_i->nat_entries); 3231 spin_lock_init(&nm_i->nat_list_lock); 3232 3233 mutex_init(&nm_i->build_lock); 3234 spin_lock_init(&nm_i->nid_list_lock); 3235 init_f2fs_rwsem(&nm_i->nat_tree_lock); 3236 3237 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 3238 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 3239 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 3240 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size, 3241 GFP_KERNEL); 3242 if (!nm_i->nat_bitmap) 3243 return -ENOMEM; 3244 3245 err = __get_nat_bitmaps(sbi); 3246 if (err) 3247 return err; 3248 3249 #ifdef CONFIG_F2FS_CHECK_FS 3250 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size, 3251 GFP_KERNEL); 3252 if (!nm_i->nat_bitmap_mir) 3253 return -ENOMEM; 3254 #endif 3255 3256 return 0; 3257 } 3258 3259 static int init_free_nid_cache(struct f2fs_sb_info *sbi) 3260 { 3261 struct f2fs_nm_info *nm_i = NM_I(sbi); 3262 int i; 3263 3264 nm_i->free_nid_bitmap = 3265 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *), 3266 nm_i->nat_blocks), 3267 GFP_KERNEL); 3268 if (!nm_i->free_nid_bitmap) 3269 return -ENOMEM; 3270 3271 for (i = 0; i < nm_i->nat_blocks; i++) { 3272 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi, 3273 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL); 3274 if (!nm_i->free_nid_bitmap[i]) 3275 return -ENOMEM; 3276 } 3277 3278 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8, 3279 GFP_KERNEL); 3280 if (!nm_i->nat_block_bitmap) 3281 return -ENOMEM; 3282 3283 nm_i->free_nid_count = 3284 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short), 3285 nm_i->nat_blocks), 3286 GFP_KERNEL); 3287 if (!nm_i->free_nid_count) 3288 return -ENOMEM; 3289 return 0; 3290 } 3291 3292 int f2fs_build_node_manager(struct f2fs_sb_info *sbi) 3293 { 3294 int err; 3295 3296 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info), 3297 GFP_KERNEL); 3298 if (!sbi->nm_info) 3299 return -ENOMEM; 3300 3301 err = init_node_manager(sbi); 3302 if (err) 3303 return err; 3304 3305 err = init_free_nid_cache(sbi); 3306 if (err) 3307 return err; 3308 3309 /* load free nid status from nat_bits table */ 3310 load_free_nid_bitmap(sbi); 3311 3312 return f2fs_build_free_nids(sbi, true, true); 3313 } 3314 3315 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi) 3316 { 3317 struct f2fs_nm_info *nm_i = NM_I(sbi); 3318 struct free_nid *i, *next_i; 3319 struct nat_entry *natvec[NATVEC_SIZE]; 3320 struct nat_entry_set *setvec[SETVEC_SIZE]; 3321 nid_t nid = 0; 3322 unsigned int found; 3323 3324 if (!nm_i) 3325 return; 3326 3327 /* destroy free nid list */ 3328 spin_lock(&nm_i->nid_list_lock); 3329 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 3330 __remove_free_nid(sbi, i, FREE_NID); 3331 spin_unlock(&nm_i->nid_list_lock); 3332 kmem_cache_free(free_nid_slab, i); 3333 spin_lock(&nm_i->nid_list_lock); 3334 } 3335 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]); 3336 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]); 3337 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list)); 3338 spin_unlock(&nm_i->nid_list_lock); 3339 3340 /* destroy nat cache */ 3341 f2fs_down_write(&nm_i->nat_tree_lock); 3342 while ((found = __gang_lookup_nat_cache(nm_i, 3343 nid, NATVEC_SIZE, natvec))) { 3344 unsigned idx; 3345 3346 nid = nat_get_nid(natvec[found - 1]) + 1; 3347 for (idx = 0; idx < found; idx++) { 3348 spin_lock(&nm_i->nat_list_lock); 3349 list_del(&natvec[idx]->list); 3350 spin_unlock(&nm_i->nat_list_lock); 3351 3352 __del_from_nat_cache(nm_i, natvec[idx]); 3353 } 3354 } 3355 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]); 3356 3357 /* destroy nat set cache */ 3358 nid = 0; 3359 while ((found = __gang_lookup_nat_set(nm_i, 3360 nid, SETVEC_SIZE, setvec))) { 3361 unsigned idx; 3362 3363 nid = setvec[found - 1]->set + 1; 3364 for (idx = 0; idx < found; idx++) { 3365 /* entry_cnt is not zero, when cp_error was occurred */ 3366 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list)); 3367 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set); 3368 kmem_cache_free(nat_entry_set_slab, setvec[idx]); 3369 } 3370 } 3371 f2fs_up_write(&nm_i->nat_tree_lock); 3372 3373 kvfree(nm_i->nat_block_bitmap); 3374 if (nm_i->free_nid_bitmap) { 3375 int i; 3376 3377 for (i = 0; i < nm_i->nat_blocks; i++) 3378 kvfree(nm_i->free_nid_bitmap[i]); 3379 kvfree(nm_i->free_nid_bitmap); 3380 } 3381 kvfree(nm_i->free_nid_count); 3382 3383 kvfree(nm_i->nat_bitmap); 3384 kvfree(nm_i->nat_bits); 3385 #ifdef CONFIG_F2FS_CHECK_FS 3386 kvfree(nm_i->nat_bitmap_mir); 3387 #endif 3388 sbi->nm_info = NULL; 3389 kfree(nm_i); 3390 } 3391 3392 int __init f2fs_create_node_manager_caches(void) 3393 { 3394 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry", 3395 sizeof(struct nat_entry)); 3396 if (!nat_entry_slab) 3397 goto fail; 3398 3399 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid", 3400 sizeof(struct free_nid)); 3401 if (!free_nid_slab) 3402 goto destroy_nat_entry; 3403 3404 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set", 3405 sizeof(struct nat_entry_set)); 3406 if (!nat_entry_set_slab) 3407 goto destroy_free_nid; 3408 3409 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry", 3410 sizeof(struct fsync_node_entry)); 3411 if (!fsync_node_entry_slab) 3412 goto destroy_nat_entry_set; 3413 return 0; 3414 3415 destroy_nat_entry_set: 3416 kmem_cache_destroy(nat_entry_set_slab); 3417 destroy_free_nid: 3418 kmem_cache_destroy(free_nid_slab); 3419 destroy_nat_entry: 3420 kmem_cache_destroy(nat_entry_slab); 3421 fail: 3422 return -ENOMEM; 3423 } 3424 3425 void f2fs_destroy_node_manager_caches(void) 3426 { 3427 kmem_cache_destroy(fsync_node_entry_slab); 3428 kmem_cache_destroy(nat_entry_set_slab); 3429 kmem_cache_destroy(free_nid_slab); 3430 kmem_cache_destroy(nat_entry_slab); 3431 } 3432