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%, 50% 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) { 89 struct extent_tree_info *eti = &sbi->extent_tree[EX_READ]; 90 91 mem_size = (atomic_read(&eti->total_ext_tree) * 92 sizeof(struct extent_tree) + 93 atomic_read(&eti->total_ext_node) * 94 sizeof(struct extent_node)) >> PAGE_SHIFT; 95 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 96 } else if (type == DISCARD_CACHE) { 97 mem_size = (atomic_read(&dcc->discard_cmd_cnt) * 98 sizeof(struct discard_cmd)) >> PAGE_SHIFT; 99 res = mem_size < (avail_ram * nm_i->ram_thresh / 100); 100 } else if (type == COMPRESS_PAGE) { 101 #ifdef CONFIG_F2FS_FS_COMPRESSION 102 unsigned long free_ram = val.freeram; 103 104 /* 105 * free memory is lower than watermark or cached page count 106 * exceed threshold, deny caching compress page. 107 */ 108 res = (free_ram > avail_ram * sbi->compress_watermark / 100) && 109 (COMPRESS_MAPPING(sbi)->nrpages < 110 free_ram * sbi->compress_percent / 100); 111 #else 112 res = false; 113 #endif 114 } else { 115 if (!sbi->sb->s_bdi->wb.dirty_exceeded) 116 return true; 117 } 118 return res; 119 } 120 121 static void clear_node_page_dirty(struct page *page) 122 { 123 if (PageDirty(page)) { 124 f2fs_clear_page_cache_dirty_tag(page); 125 clear_page_dirty_for_io(page); 126 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES); 127 } 128 ClearPageUptodate(page); 129 } 130 131 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 132 { 133 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid)); 134 } 135 136 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 137 { 138 struct page *src_page; 139 struct page *dst_page; 140 pgoff_t dst_off; 141 void *src_addr; 142 void *dst_addr; 143 struct f2fs_nm_info *nm_i = NM_I(sbi); 144 145 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid)); 146 147 /* get current nat block page with lock */ 148 src_page = get_current_nat_page(sbi, nid); 149 if (IS_ERR(src_page)) 150 return src_page; 151 dst_page = f2fs_grab_meta_page(sbi, dst_off); 152 f2fs_bug_on(sbi, PageDirty(src_page)); 153 154 src_addr = page_address(src_page); 155 dst_addr = page_address(dst_page); 156 memcpy(dst_addr, src_addr, PAGE_SIZE); 157 set_page_dirty(dst_page); 158 f2fs_put_page(src_page, 1); 159 160 set_to_next_nat(nm_i, nid); 161 162 return dst_page; 163 } 164 165 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi, 166 nid_t nid, bool no_fail) 167 { 168 struct nat_entry *new; 169 170 new = f2fs_kmem_cache_alloc(nat_entry_slab, 171 GFP_F2FS_ZERO, no_fail, sbi); 172 if (new) { 173 nat_set_nid(new, nid); 174 nat_reset_flag(new); 175 } 176 return new; 177 } 178 179 static void __free_nat_entry(struct nat_entry *e) 180 { 181 kmem_cache_free(nat_entry_slab, e); 182 } 183 184 /* must be locked by nat_tree_lock */ 185 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i, 186 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail) 187 { 188 if (no_fail) 189 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne); 190 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne)) 191 return NULL; 192 193 if (raw_ne) 194 node_info_from_raw_nat(&ne->ni, raw_ne); 195 196 spin_lock(&nm_i->nat_list_lock); 197 list_add_tail(&ne->list, &nm_i->nat_entries); 198 spin_unlock(&nm_i->nat_list_lock); 199 200 nm_i->nat_cnt[TOTAL_NAT]++; 201 nm_i->nat_cnt[RECLAIMABLE_NAT]++; 202 return ne; 203 } 204 205 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n) 206 { 207 struct nat_entry *ne; 208 209 ne = radix_tree_lookup(&nm_i->nat_root, n); 210 211 /* for recent accessed nat entry, move it to tail of lru list */ 212 if (ne && !get_nat_flag(ne, IS_DIRTY)) { 213 spin_lock(&nm_i->nat_list_lock); 214 if (!list_empty(&ne->list)) 215 list_move_tail(&ne->list, &nm_i->nat_entries); 216 spin_unlock(&nm_i->nat_list_lock); 217 } 218 219 return ne; 220 } 221 222 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i, 223 nid_t start, unsigned int nr, struct nat_entry **ep) 224 { 225 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr); 226 } 227 228 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e) 229 { 230 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e)); 231 nm_i->nat_cnt[TOTAL_NAT]--; 232 nm_i->nat_cnt[RECLAIMABLE_NAT]--; 233 __free_nat_entry(e); 234 } 235 236 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i, 237 struct nat_entry *ne) 238 { 239 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid); 240 struct nat_entry_set *head; 241 242 head = radix_tree_lookup(&nm_i->nat_set_root, set); 243 if (!head) { 244 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, 245 GFP_NOFS, true, NULL); 246 247 INIT_LIST_HEAD(&head->entry_list); 248 INIT_LIST_HEAD(&head->set_list); 249 head->set = set; 250 head->entry_cnt = 0; 251 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head); 252 } 253 return head; 254 } 255 256 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i, 257 struct nat_entry *ne) 258 { 259 struct nat_entry_set *head; 260 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR; 261 262 if (!new_ne) 263 head = __grab_nat_entry_set(nm_i, ne); 264 265 /* 266 * update entry_cnt in below condition: 267 * 1. update NEW_ADDR to valid block address; 268 * 2. update old block address to new one; 269 */ 270 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) || 271 !get_nat_flag(ne, IS_DIRTY))) 272 head->entry_cnt++; 273 274 set_nat_flag(ne, IS_PREALLOC, new_ne); 275 276 if (get_nat_flag(ne, IS_DIRTY)) 277 goto refresh_list; 278 279 nm_i->nat_cnt[DIRTY_NAT]++; 280 nm_i->nat_cnt[RECLAIMABLE_NAT]--; 281 set_nat_flag(ne, IS_DIRTY, true); 282 refresh_list: 283 spin_lock(&nm_i->nat_list_lock); 284 if (new_ne) 285 list_del_init(&ne->list); 286 else 287 list_move_tail(&ne->list, &head->entry_list); 288 spin_unlock(&nm_i->nat_list_lock); 289 } 290 291 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i, 292 struct nat_entry_set *set, struct nat_entry *ne) 293 { 294 spin_lock(&nm_i->nat_list_lock); 295 list_move_tail(&ne->list, &nm_i->nat_entries); 296 spin_unlock(&nm_i->nat_list_lock); 297 298 set_nat_flag(ne, IS_DIRTY, false); 299 set->entry_cnt--; 300 nm_i->nat_cnt[DIRTY_NAT]--; 301 nm_i->nat_cnt[RECLAIMABLE_NAT]++; 302 } 303 304 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i, 305 nid_t start, unsigned int nr, struct nat_entry_set **ep) 306 { 307 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep, 308 start, nr); 309 } 310 311 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page) 312 { 313 return NODE_MAPPING(sbi) == page->mapping && 314 IS_DNODE(page) && is_cold_node(page); 315 } 316 317 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi) 318 { 319 spin_lock_init(&sbi->fsync_node_lock); 320 INIT_LIST_HEAD(&sbi->fsync_node_list); 321 sbi->fsync_seg_id = 0; 322 sbi->fsync_node_num = 0; 323 } 324 325 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi, 326 struct page *page) 327 { 328 struct fsync_node_entry *fn; 329 unsigned long flags; 330 unsigned int seq_id; 331 332 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, 333 GFP_NOFS, true, NULL); 334 335 get_page(page); 336 fn->page = page; 337 INIT_LIST_HEAD(&fn->list); 338 339 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 340 list_add_tail(&fn->list, &sbi->fsync_node_list); 341 fn->seq_id = sbi->fsync_seg_id++; 342 seq_id = fn->seq_id; 343 sbi->fsync_node_num++; 344 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 345 346 return seq_id; 347 } 348 349 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page) 350 { 351 struct fsync_node_entry *fn; 352 unsigned long flags; 353 354 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 355 list_for_each_entry(fn, &sbi->fsync_node_list, list) { 356 if (fn->page == page) { 357 list_del(&fn->list); 358 sbi->fsync_node_num--; 359 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 360 kmem_cache_free(fsync_node_entry_slab, fn); 361 put_page(page); 362 return; 363 } 364 } 365 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 366 f2fs_bug_on(sbi, 1); 367 } 368 369 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi) 370 { 371 unsigned long flags; 372 373 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 374 sbi->fsync_seg_id = 0; 375 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 376 } 377 378 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid) 379 { 380 struct f2fs_nm_info *nm_i = NM_I(sbi); 381 struct nat_entry *e; 382 bool need = false; 383 384 f2fs_down_read(&nm_i->nat_tree_lock); 385 e = __lookup_nat_cache(nm_i, nid); 386 if (e) { 387 if (!get_nat_flag(e, IS_CHECKPOINTED) && 388 !get_nat_flag(e, HAS_FSYNCED_INODE)) 389 need = true; 390 } 391 f2fs_up_read(&nm_i->nat_tree_lock); 392 return need; 393 } 394 395 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid) 396 { 397 struct f2fs_nm_info *nm_i = NM_I(sbi); 398 struct nat_entry *e; 399 bool is_cp = true; 400 401 f2fs_down_read(&nm_i->nat_tree_lock); 402 e = __lookup_nat_cache(nm_i, nid); 403 if (e && !get_nat_flag(e, IS_CHECKPOINTED)) 404 is_cp = false; 405 f2fs_up_read(&nm_i->nat_tree_lock); 406 return is_cp; 407 } 408 409 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino) 410 { 411 struct f2fs_nm_info *nm_i = NM_I(sbi); 412 struct nat_entry *e; 413 bool need_update = true; 414 415 f2fs_down_read(&nm_i->nat_tree_lock); 416 e = __lookup_nat_cache(nm_i, ino); 417 if (e && get_nat_flag(e, HAS_LAST_FSYNC) && 418 (get_nat_flag(e, IS_CHECKPOINTED) || 419 get_nat_flag(e, HAS_FSYNCED_INODE))) 420 need_update = false; 421 f2fs_up_read(&nm_i->nat_tree_lock); 422 return need_update; 423 } 424 425 /* must be locked by nat_tree_lock */ 426 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid, 427 struct f2fs_nat_entry *ne) 428 { 429 struct f2fs_nm_info *nm_i = NM_I(sbi); 430 struct nat_entry *new, *e; 431 432 /* Let's mitigate lock contention of nat_tree_lock during checkpoint */ 433 if (f2fs_rwsem_is_locked(&sbi->cp_global_sem)) 434 return; 435 436 new = __alloc_nat_entry(sbi, nid, false); 437 if (!new) 438 return; 439 440 f2fs_down_write(&nm_i->nat_tree_lock); 441 e = __lookup_nat_cache(nm_i, nid); 442 if (!e) 443 e = __init_nat_entry(nm_i, new, ne, false); 444 else 445 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) || 446 nat_get_blkaddr(e) != 447 le32_to_cpu(ne->block_addr) || 448 nat_get_version(e) != ne->version); 449 f2fs_up_write(&nm_i->nat_tree_lock); 450 if (e != new) 451 __free_nat_entry(new); 452 } 453 454 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni, 455 block_t new_blkaddr, bool fsync_done) 456 { 457 struct f2fs_nm_info *nm_i = NM_I(sbi); 458 struct nat_entry *e; 459 struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true); 460 461 f2fs_down_write(&nm_i->nat_tree_lock); 462 e = __lookup_nat_cache(nm_i, ni->nid); 463 if (!e) { 464 e = __init_nat_entry(nm_i, new, NULL, true); 465 copy_node_info(&e->ni, ni); 466 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR); 467 } else if (new_blkaddr == NEW_ADDR) { 468 /* 469 * when nid is reallocated, 470 * previous nat entry can be remained in nat cache. 471 * So, reinitialize it with new information. 472 */ 473 copy_node_info(&e->ni, ni); 474 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR); 475 } 476 /* let's free early to reduce memory consumption */ 477 if (e != new) 478 __free_nat_entry(new); 479 480 /* sanity check */ 481 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr); 482 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR && 483 new_blkaddr == NULL_ADDR); 484 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR && 485 new_blkaddr == NEW_ADDR); 486 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) && 487 new_blkaddr == NEW_ADDR); 488 489 /* increment version no as node is removed */ 490 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) { 491 unsigned char version = nat_get_version(e); 492 493 nat_set_version(e, inc_node_version(version)); 494 } 495 496 /* change address */ 497 nat_set_blkaddr(e, new_blkaddr); 498 if (!__is_valid_data_blkaddr(new_blkaddr)) 499 set_nat_flag(e, IS_CHECKPOINTED, false); 500 __set_nat_cache_dirty(nm_i, e); 501 502 /* update fsync_mark if its inode nat entry is still alive */ 503 if (ni->nid != ni->ino) 504 e = __lookup_nat_cache(nm_i, ni->ino); 505 if (e) { 506 if (fsync_done && ni->nid == ni->ino) 507 set_nat_flag(e, HAS_FSYNCED_INODE, true); 508 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done); 509 } 510 f2fs_up_write(&nm_i->nat_tree_lock); 511 } 512 513 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink) 514 { 515 struct f2fs_nm_info *nm_i = NM_I(sbi); 516 int nr = nr_shrink; 517 518 if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock)) 519 return 0; 520 521 spin_lock(&nm_i->nat_list_lock); 522 while (nr_shrink) { 523 struct nat_entry *ne; 524 525 if (list_empty(&nm_i->nat_entries)) 526 break; 527 528 ne = list_first_entry(&nm_i->nat_entries, 529 struct nat_entry, list); 530 list_del(&ne->list); 531 spin_unlock(&nm_i->nat_list_lock); 532 533 __del_from_nat_cache(nm_i, ne); 534 nr_shrink--; 535 536 spin_lock(&nm_i->nat_list_lock); 537 } 538 spin_unlock(&nm_i->nat_list_lock); 539 540 f2fs_up_write(&nm_i->nat_tree_lock); 541 return nr - nr_shrink; 542 } 543 544 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid, 545 struct node_info *ni, bool checkpoint_context) 546 { 547 struct f2fs_nm_info *nm_i = NM_I(sbi); 548 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 549 struct f2fs_journal *journal = curseg->journal; 550 nid_t start_nid = START_NID(nid); 551 struct f2fs_nat_block *nat_blk; 552 struct page *page = NULL; 553 struct f2fs_nat_entry ne; 554 struct nat_entry *e; 555 pgoff_t index; 556 block_t blkaddr; 557 int i; 558 559 ni->nid = nid; 560 retry: 561 /* Check nat cache */ 562 f2fs_down_read(&nm_i->nat_tree_lock); 563 e = __lookup_nat_cache(nm_i, nid); 564 if (e) { 565 ni->ino = nat_get_ino(e); 566 ni->blk_addr = nat_get_blkaddr(e); 567 ni->version = nat_get_version(e); 568 f2fs_up_read(&nm_i->nat_tree_lock); 569 return 0; 570 } 571 572 /* 573 * Check current segment summary by trying to grab journal_rwsem first. 574 * This sem is on the critical path on the checkpoint requiring the above 575 * nat_tree_lock. Therefore, we should retry, if we failed to grab here 576 * while not bothering checkpoint. 577 */ 578 if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) { 579 down_read(&curseg->journal_rwsem); 580 } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) || 581 !down_read_trylock(&curseg->journal_rwsem)) { 582 f2fs_up_read(&nm_i->nat_tree_lock); 583 goto retry; 584 } 585 586 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0); 587 if (i >= 0) { 588 ne = nat_in_journal(journal, i); 589 node_info_from_raw_nat(ni, &ne); 590 } 591 up_read(&curseg->journal_rwsem); 592 if (i >= 0) { 593 f2fs_up_read(&nm_i->nat_tree_lock); 594 goto cache; 595 } 596 597 /* Fill node_info from nat page */ 598 index = current_nat_addr(sbi, nid); 599 f2fs_up_read(&nm_i->nat_tree_lock); 600 601 page = f2fs_get_meta_page(sbi, index); 602 if (IS_ERR(page)) 603 return PTR_ERR(page); 604 605 nat_blk = (struct f2fs_nat_block *)page_address(page); 606 ne = nat_blk->entries[nid - start_nid]; 607 node_info_from_raw_nat(ni, &ne); 608 f2fs_put_page(page, 1); 609 cache: 610 blkaddr = le32_to_cpu(ne.block_addr); 611 if (__is_valid_data_blkaddr(blkaddr) && 612 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) 613 return -EFAULT; 614 615 /* cache nat entry */ 616 cache_nat_entry(sbi, nid, &ne); 617 return 0; 618 } 619 620 /* 621 * readahead MAX_RA_NODE number of node pages. 622 */ 623 static void f2fs_ra_node_pages(struct page *parent, int start, int n) 624 { 625 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 626 struct blk_plug plug; 627 int i, end; 628 nid_t nid; 629 630 blk_start_plug(&plug); 631 632 /* Then, try readahead for siblings of the desired node */ 633 end = start + n; 634 end = min(end, NIDS_PER_BLOCK); 635 for (i = start; i < end; i++) { 636 nid = get_nid(parent, i, false); 637 f2fs_ra_node_page(sbi, nid); 638 } 639 640 blk_finish_plug(&plug); 641 } 642 643 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs) 644 { 645 const long direct_index = ADDRS_PER_INODE(dn->inode); 646 const long direct_blks = ADDRS_PER_BLOCK(dn->inode); 647 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK; 648 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode); 649 int cur_level = dn->cur_level; 650 int max_level = dn->max_level; 651 pgoff_t base = 0; 652 653 if (!dn->max_level) 654 return pgofs + 1; 655 656 while (max_level-- > cur_level) 657 skipped_unit *= NIDS_PER_BLOCK; 658 659 switch (dn->max_level) { 660 case 3: 661 base += 2 * indirect_blks; 662 fallthrough; 663 case 2: 664 base += 2 * direct_blks; 665 fallthrough; 666 case 1: 667 base += direct_index; 668 break; 669 default: 670 f2fs_bug_on(F2FS_I_SB(dn->inode), 1); 671 } 672 673 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base; 674 } 675 676 /* 677 * The maximum depth is four. 678 * Offset[0] will have raw inode offset. 679 */ 680 static int get_node_path(struct inode *inode, long block, 681 int offset[4], unsigned int noffset[4]) 682 { 683 const long direct_index = ADDRS_PER_INODE(inode); 684 const long direct_blks = ADDRS_PER_BLOCK(inode); 685 const long dptrs_per_blk = NIDS_PER_BLOCK; 686 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK; 687 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK; 688 int n = 0; 689 int level = 0; 690 691 noffset[0] = 0; 692 693 if (block < direct_index) { 694 offset[n] = block; 695 goto got; 696 } 697 block -= direct_index; 698 if (block < direct_blks) { 699 offset[n++] = NODE_DIR1_BLOCK; 700 noffset[n] = 1; 701 offset[n] = block; 702 level = 1; 703 goto got; 704 } 705 block -= direct_blks; 706 if (block < direct_blks) { 707 offset[n++] = NODE_DIR2_BLOCK; 708 noffset[n] = 2; 709 offset[n] = block; 710 level = 1; 711 goto got; 712 } 713 block -= direct_blks; 714 if (block < indirect_blks) { 715 offset[n++] = NODE_IND1_BLOCK; 716 noffset[n] = 3; 717 offset[n++] = block / direct_blks; 718 noffset[n] = 4 + offset[n - 1]; 719 offset[n] = block % direct_blks; 720 level = 2; 721 goto got; 722 } 723 block -= indirect_blks; 724 if (block < indirect_blks) { 725 offset[n++] = NODE_IND2_BLOCK; 726 noffset[n] = 4 + dptrs_per_blk; 727 offset[n++] = block / direct_blks; 728 noffset[n] = 5 + dptrs_per_blk + offset[n - 1]; 729 offset[n] = block % direct_blks; 730 level = 2; 731 goto got; 732 } 733 block -= indirect_blks; 734 if (block < dindirect_blks) { 735 offset[n++] = NODE_DIND_BLOCK; 736 noffset[n] = 5 + (dptrs_per_blk * 2); 737 offset[n++] = block / indirect_blks; 738 noffset[n] = 6 + (dptrs_per_blk * 2) + 739 offset[n - 1] * (dptrs_per_blk + 1); 740 offset[n++] = (block / direct_blks) % dptrs_per_blk; 741 noffset[n] = 7 + (dptrs_per_blk * 2) + 742 offset[n - 2] * (dptrs_per_blk + 1) + 743 offset[n - 1]; 744 offset[n] = block % direct_blks; 745 level = 3; 746 goto got; 747 } else { 748 return -E2BIG; 749 } 750 got: 751 return level; 752 } 753 754 /* 755 * Caller should call f2fs_put_dnode(dn). 756 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and 757 * f2fs_unlock_op() only if mode is set with ALLOC_NODE. 758 */ 759 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode) 760 { 761 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 762 struct page *npage[4]; 763 struct page *parent = NULL; 764 int offset[4]; 765 unsigned int noffset[4]; 766 nid_t nids[4]; 767 int level, i = 0; 768 int err = 0; 769 770 level = get_node_path(dn->inode, index, offset, noffset); 771 if (level < 0) 772 return level; 773 774 nids[0] = dn->inode->i_ino; 775 npage[0] = dn->inode_page; 776 777 if (!npage[0]) { 778 npage[0] = f2fs_get_node_page(sbi, nids[0]); 779 if (IS_ERR(npage[0])) 780 return PTR_ERR(npage[0]); 781 } 782 783 /* if inline_data is set, should not report any block indices */ 784 if (f2fs_has_inline_data(dn->inode) && index) { 785 err = -ENOENT; 786 f2fs_put_page(npage[0], 1); 787 goto release_out; 788 } 789 790 parent = npage[0]; 791 if (level != 0) 792 nids[1] = get_nid(parent, offset[0], true); 793 dn->inode_page = npage[0]; 794 dn->inode_page_locked = true; 795 796 /* get indirect or direct nodes */ 797 for (i = 1; i <= level; i++) { 798 bool done = false; 799 800 if (!nids[i] && mode == ALLOC_NODE) { 801 /* alloc new node */ 802 if (!f2fs_alloc_nid(sbi, &(nids[i]))) { 803 err = -ENOSPC; 804 goto release_pages; 805 } 806 807 dn->nid = nids[i]; 808 npage[i] = f2fs_new_node_page(dn, noffset[i]); 809 if (IS_ERR(npage[i])) { 810 f2fs_alloc_nid_failed(sbi, nids[i]); 811 err = PTR_ERR(npage[i]); 812 goto release_pages; 813 } 814 815 set_nid(parent, offset[i - 1], nids[i], i == 1); 816 f2fs_alloc_nid_done(sbi, nids[i]); 817 done = true; 818 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) { 819 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]); 820 if (IS_ERR(npage[i])) { 821 err = PTR_ERR(npage[i]); 822 goto release_pages; 823 } 824 done = true; 825 } 826 if (i == 1) { 827 dn->inode_page_locked = false; 828 unlock_page(parent); 829 } else { 830 f2fs_put_page(parent, 1); 831 } 832 833 if (!done) { 834 npage[i] = f2fs_get_node_page(sbi, nids[i]); 835 if (IS_ERR(npage[i])) { 836 err = PTR_ERR(npage[i]); 837 f2fs_put_page(npage[0], 0); 838 goto release_out; 839 } 840 } 841 if (i < level) { 842 parent = npage[i]; 843 nids[i + 1] = get_nid(parent, offset[i], false); 844 } 845 } 846 dn->nid = nids[level]; 847 dn->ofs_in_node = offset[level]; 848 dn->node_page = npage[level]; 849 dn->data_blkaddr = f2fs_data_blkaddr(dn); 850 851 if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) && 852 f2fs_sb_has_readonly(sbi)) { 853 unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn); 854 block_t blkaddr; 855 856 if (!c_len) 857 goto out; 858 859 blkaddr = f2fs_data_blkaddr(dn); 860 if (blkaddr == COMPRESS_ADDR) 861 blkaddr = data_blkaddr(dn->inode, dn->node_page, 862 dn->ofs_in_node + 1); 863 864 f2fs_update_read_extent_tree_range_compressed(dn->inode, 865 index, blkaddr, 866 F2FS_I(dn->inode)->i_cluster_size, 867 c_len); 868 } 869 out: 870 return 0; 871 872 release_pages: 873 f2fs_put_page(parent, 1); 874 if (i > 1) 875 f2fs_put_page(npage[0], 0); 876 release_out: 877 dn->inode_page = NULL; 878 dn->node_page = NULL; 879 if (err == -ENOENT) { 880 dn->cur_level = i; 881 dn->max_level = level; 882 dn->ofs_in_node = offset[level]; 883 } 884 return err; 885 } 886 887 static int truncate_node(struct dnode_of_data *dn) 888 { 889 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 890 struct node_info ni; 891 int err; 892 pgoff_t index; 893 894 err = f2fs_get_node_info(sbi, dn->nid, &ni, false); 895 if (err) 896 return err; 897 898 /* Deallocate node address */ 899 f2fs_invalidate_blocks(sbi, ni.blk_addr); 900 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino); 901 set_node_addr(sbi, &ni, NULL_ADDR, false); 902 903 if (dn->nid == dn->inode->i_ino) { 904 f2fs_remove_orphan_inode(sbi, dn->nid); 905 dec_valid_inode_count(sbi); 906 f2fs_inode_synced(dn->inode); 907 } 908 909 clear_node_page_dirty(dn->node_page); 910 set_sbi_flag(sbi, SBI_IS_DIRTY); 911 912 index = dn->node_page->index; 913 f2fs_put_page(dn->node_page, 1); 914 915 invalidate_mapping_pages(NODE_MAPPING(sbi), 916 index, index); 917 918 dn->node_page = NULL; 919 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr); 920 921 return 0; 922 } 923 924 static int truncate_dnode(struct dnode_of_data *dn) 925 { 926 struct page *page; 927 int err; 928 929 if (dn->nid == 0) 930 return 1; 931 932 /* get direct node */ 933 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 934 if (PTR_ERR(page) == -ENOENT) 935 return 1; 936 else if (IS_ERR(page)) 937 return PTR_ERR(page); 938 939 /* Make dnode_of_data for parameter */ 940 dn->node_page = page; 941 dn->ofs_in_node = 0; 942 f2fs_truncate_data_blocks(dn); 943 err = truncate_node(dn); 944 if (err) 945 return err; 946 947 return 1; 948 } 949 950 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs, 951 int ofs, int depth) 952 { 953 struct dnode_of_data rdn = *dn; 954 struct page *page; 955 struct f2fs_node *rn; 956 nid_t child_nid; 957 unsigned int child_nofs; 958 int freed = 0; 959 int i, ret; 960 961 if (dn->nid == 0) 962 return NIDS_PER_BLOCK + 1; 963 964 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr); 965 966 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 967 if (IS_ERR(page)) { 968 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page)); 969 return PTR_ERR(page); 970 } 971 972 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK); 973 974 rn = F2FS_NODE(page); 975 if (depth < 3) { 976 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) { 977 child_nid = le32_to_cpu(rn->in.nid[i]); 978 if (child_nid == 0) 979 continue; 980 rdn.nid = child_nid; 981 ret = truncate_dnode(&rdn); 982 if (ret < 0) 983 goto out_err; 984 if (set_nid(page, i, 0, false)) 985 dn->node_changed = true; 986 } 987 } else { 988 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1; 989 for (i = ofs; i < NIDS_PER_BLOCK; i++) { 990 child_nid = le32_to_cpu(rn->in.nid[i]); 991 if (child_nid == 0) { 992 child_nofs += NIDS_PER_BLOCK + 1; 993 continue; 994 } 995 rdn.nid = child_nid; 996 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1); 997 if (ret == (NIDS_PER_BLOCK + 1)) { 998 if (set_nid(page, i, 0, false)) 999 dn->node_changed = true; 1000 child_nofs += ret; 1001 } else if (ret < 0 && ret != -ENOENT) { 1002 goto out_err; 1003 } 1004 } 1005 freed = child_nofs; 1006 } 1007 1008 if (!ofs) { 1009 /* remove current indirect node */ 1010 dn->node_page = page; 1011 ret = truncate_node(dn); 1012 if (ret) 1013 goto out_err; 1014 freed++; 1015 } else { 1016 f2fs_put_page(page, 1); 1017 } 1018 trace_f2fs_truncate_nodes_exit(dn->inode, freed); 1019 return freed; 1020 1021 out_err: 1022 f2fs_put_page(page, 1); 1023 trace_f2fs_truncate_nodes_exit(dn->inode, ret); 1024 return ret; 1025 } 1026 1027 static int truncate_partial_nodes(struct dnode_of_data *dn, 1028 struct f2fs_inode *ri, int *offset, int depth) 1029 { 1030 struct page *pages[2]; 1031 nid_t nid[3]; 1032 nid_t child_nid; 1033 int err = 0; 1034 int i; 1035 int idx = depth - 2; 1036 1037 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1038 if (!nid[0]) 1039 return 0; 1040 1041 /* get indirect nodes in the path */ 1042 for (i = 0; i < idx + 1; i++) { 1043 /* reference count'll be increased */ 1044 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]); 1045 if (IS_ERR(pages[i])) { 1046 err = PTR_ERR(pages[i]); 1047 idx = i - 1; 1048 goto fail; 1049 } 1050 nid[i + 1] = get_nid(pages[i], offset[i + 1], false); 1051 } 1052 1053 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK); 1054 1055 /* free direct nodes linked to a partial indirect node */ 1056 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) { 1057 child_nid = get_nid(pages[idx], i, false); 1058 if (!child_nid) 1059 continue; 1060 dn->nid = child_nid; 1061 err = truncate_dnode(dn); 1062 if (err < 0) 1063 goto fail; 1064 if (set_nid(pages[idx], i, 0, false)) 1065 dn->node_changed = true; 1066 } 1067 1068 if (offset[idx + 1] == 0) { 1069 dn->node_page = pages[idx]; 1070 dn->nid = nid[idx]; 1071 err = truncate_node(dn); 1072 if (err) 1073 goto fail; 1074 } else { 1075 f2fs_put_page(pages[idx], 1); 1076 } 1077 offset[idx]++; 1078 offset[idx + 1] = 0; 1079 idx--; 1080 fail: 1081 for (i = idx; i >= 0; i--) 1082 f2fs_put_page(pages[i], 1); 1083 1084 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err); 1085 1086 return err; 1087 } 1088 1089 /* 1090 * All the block addresses of data and nodes should be nullified. 1091 */ 1092 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from) 1093 { 1094 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1095 int err = 0, cont = 1; 1096 int level, offset[4], noffset[4]; 1097 unsigned int nofs = 0; 1098 struct f2fs_inode *ri; 1099 struct dnode_of_data dn; 1100 struct page *page; 1101 1102 trace_f2fs_truncate_inode_blocks_enter(inode, from); 1103 1104 level = get_node_path(inode, from, offset, noffset); 1105 if (level < 0) { 1106 trace_f2fs_truncate_inode_blocks_exit(inode, level); 1107 return level; 1108 } 1109 1110 page = f2fs_get_node_page(sbi, inode->i_ino); 1111 if (IS_ERR(page)) { 1112 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page)); 1113 return PTR_ERR(page); 1114 } 1115 1116 set_new_dnode(&dn, inode, page, NULL, 0); 1117 unlock_page(page); 1118 1119 ri = F2FS_INODE(page); 1120 switch (level) { 1121 case 0: 1122 case 1: 1123 nofs = noffset[1]; 1124 break; 1125 case 2: 1126 nofs = noffset[1]; 1127 if (!offset[level - 1]) 1128 goto skip_partial; 1129 err = truncate_partial_nodes(&dn, ri, offset, level); 1130 if (err < 0 && err != -ENOENT) 1131 goto fail; 1132 nofs += 1 + NIDS_PER_BLOCK; 1133 break; 1134 case 3: 1135 nofs = 5 + 2 * NIDS_PER_BLOCK; 1136 if (!offset[level - 1]) 1137 goto skip_partial; 1138 err = truncate_partial_nodes(&dn, ri, offset, level); 1139 if (err < 0 && err != -ENOENT) 1140 goto fail; 1141 break; 1142 default: 1143 BUG(); 1144 } 1145 1146 skip_partial: 1147 while (cont) { 1148 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1149 switch (offset[0]) { 1150 case NODE_DIR1_BLOCK: 1151 case NODE_DIR2_BLOCK: 1152 err = truncate_dnode(&dn); 1153 break; 1154 1155 case NODE_IND1_BLOCK: 1156 case NODE_IND2_BLOCK: 1157 err = truncate_nodes(&dn, nofs, offset[1], 2); 1158 break; 1159 1160 case NODE_DIND_BLOCK: 1161 err = truncate_nodes(&dn, nofs, offset[1], 3); 1162 cont = 0; 1163 break; 1164 1165 default: 1166 BUG(); 1167 } 1168 if (err < 0 && err != -ENOENT) 1169 goto fail; 1170 if (offset[1] == 0 && 1171 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) { 1172 lock_page(page); 1173 BUG_ON(page->mapping != NODE_MAPPING(sbi)); 1174 f2fs_wait_on_page_writeback(page, NODE, true, true); 1175 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0; 1176 set_page_dirty(page); 1177 unlock_page(page); 1178 } 1179 offset[1] = 0; 1180 offset[0]++; 1181 nofs += err; 1182 } 1183 fail: 1184 f2fs_put_page(page, 0); 1185 trace_f2fs_truncate_inode_blocks_exit(inode, err); 1186 return err > 0 ? 0 : err; 1187 } 1188 1189 /* caller must lock inode page */ 1190 int f2fs_truncate_xattr_node(struct inode *inode) 1191 { 1192 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1193 nid_t nid = F2FS_I(inode)->i_xattr_nid; 1194 struct dnode_of_data dn; 1195 struct page *npage; 1196 int err; 1197 1198 if (!nid) 1199 return 0; 1200 1201 npage = f2fs_get_node_page(sbi, nid); 1202 if (IS_ERR(npage)) 1203 return PTR_ERR(npage); 1204 1205 set_new_dnode(&dn, inode, NULL, npage, nid); 1206 err = truncate_node(&dn); 1207 if (err) { 1208 f2fs_put_page(npage, 1); 1209 return err; 1210 } 1211 1212 f2fs_i_xnid_write(inode, 0); 1213 1214 return 0; 1215 } 1216 1217 /* 1218 * Caller should grab and release a rwsem by calling f2fs_lock_op() and 1219 * f2fs_unlock_op(). 1220 */ 1221 int f2fs_remove_inode_page(struct inode *inode) 1222 { 1223 struct dnode_of_data dn; 1224 int err; 1225 1226 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1227 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE); 1228 if (err) 1229 return err; 1230 1231 err = f2fs_truncate_xattr_node(inode); 1232 if (err) { 1233 f2fs_put_dnode(&dn); 1234 return err; 1235 } 1236 1237 /* remove potential inline_data blocks */ 1238 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1239 S_ISLNK(inode->i_mode)) 1240 f2fs_truncate_data_blocks_range(&dn, 1); 1241 1242 /* 0 is possible, after f2fs_new_inode() has failed */ 1243 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) { 1244 f2fs_put_dnode(&dn); 1245 return -EIO; 1246 } 1247 1248 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) { 1249 f2fs_warn(F2FS_I_SB(inode), 1250 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu", 1251 inode->i_ino, (unsigned long long)inode->i_blocks); 1252 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK); 1253 } 1254 1255 /* will put inode & node pages */ 1256 err = truncate_node(&dn); 1257 if (err) { 1258 f2fs_put_dnode(&dn); 1259 return err; 1260 } 1261 return 0; 1262 } 1263 1264 struct page *f2fs_new_inode_page(struct inode *inode) 1265 { 1266 struct dnode_of_data dn; 1267 1268 /* allocate inode page for new inode */ 1269 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1270 1271 /* caller should f2fs_put_page(page, 1); */ 1272 return f2fs_new_node_page(&dn, 0); 1273 } 1274 1275 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs) 1276 { 1277 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1278 struct node_info new_ni; 1279 struct page *page; 1280 int err; 1281 1282 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1283 return ERR_PTR(-EPERM); 1284 1285 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false); 1286 if (!page) 1287 return ERR_PTR(-ENOMEM); 1288 1289 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs)))) 1290 goto fail; 1291 1292 #ifdef CONFIG_F2FS_CHECK_FS 1293 err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false); 1294 if (err) { 1295 dec_valid_node_count(sbi, dn->inode, !ofs); 1296 goto fail; 1297 } 1298 if (unlikely(new_ni.blk_addr != NULL_ADDR)) { 1299 err = -EFSCORRUPTED; 1300 set_sbi_flag(sbi, SBI_NEED_FSCK); 1301 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 1302 goto fail; 1303 } 1304 #endif 1305 new_ni.nid = dn->nid; 1306 new_ni.ino = dn->inode->i_ino; 1307 new_ni.blk_addr = NULL_ADDR; 1308 new_ni.flag = 0; 1309 new_ni.version = 0; 1310 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 1311 1312 f2fs_wait_on_page_writeback(page, NODE, true, true); 1313 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 1314 set_cold_node(page, S_ISDIR(dn->inode->i_mode)); 1315 if (!PageUptodate(page)) 1316 SetPageUptodate(page); 1317 if (set_page_dirty(page)) 1318 dn->node_changed = true; 1319 1320 if (f2fs_has_xattr_block(ofs)) 1321 f2fs_i_xnid_write(dn->inode, dn->nid); 1322 1323 if (ofs == 0) 1324 inc_valid_inode_count(sbi); 1325 return page; 1326 1327 fail: 1328 clear_node_page_dirty(page); 1329 f2fs_put_page(page, 1); 1330 return ERR_PTR(err); 1331 } 1332 1333 /* 1334 * Caller should do after getting the following values. 1335 * 0: f2fs_put_page(page, 0) 1336 * LOCKED_PAGE or error: f2fs_put_page(page, 1) 1337 */ 1338 static int read_node_page(struct page *page, blk_opf_t op_flags) 1339 { 1340 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1341 struct node_info ni; 1342 struct f2fs_io_info fio = { 1343 .sbi = sbi, 1344 .type = NODE, 1345 .op = REQ_OP_READ, 1346 .op_flags = op_flags, 1347 .page = page, 1348 .encrypted_page = NULL, 1349 }; 1350 int err; 1351 1352 if (PageUptodate(page)) { 1353 if (!f2fs_inode_chksum_verify(sbi, page)) { 1354 ClearPageUptodate(page); 1355 return -EFSBADCRC; 1356 } 1357 return LOCKED_PAGE; 1358 } 1359 1360 err = f2fs_get_node_info(sbi, page->index, &ni, false); 1361 if (err) 1362 return err; 1363 1364 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */ 1365 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) { 1366 ClearPageUptodate(page); 1367 return -ENOENT; 1368 } 1369 1370 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr; 1371 1372 err = f2fs_submit_page_bio(&fio); 1373 1374 if (!err) 1375 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE); 1376 1377 return err; 1378 } 1379 1380 /* 1381 * Readahead a node page 1382 */ 1383 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 1384 { 1385 struct page *apage; 1386 int err; 1387 1388 if (!nid) 1389 return; 1390 if (f2fs_check_nid_range(sbi, nid)) 1391 return; 1392 1393 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid); 1394 if (apage) 1395 return; 1396 1397 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1398 if (!apage) 1399 return; 1400 1401 err = read_node_page(apage, REQ_RAHEAD); 1402 f2fs_put_page(apage, err ? 1 : 0); 1403 } 1404 1405 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid, 1406 struct page *parent, int start) 1407 { 1408 struct page *page; 1409 int err; 1410 1411 if (!nid) 1412 return ERR_PTR(-ENOENT); 1413 if (f2fs_check_nid_range(sbi, nid)) 1414 return ERR_PTR(-EINVAL); 1415 repeat: 1416 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1417 if (!page) 1418 return ERR_PTR(-ENOMEM); 1419 1420 err = read_node_page(page, 0); 1421 if (err < 0) { 1422 goto out_put_err; 1423 } else if (err == LOCKED_PAGE) { 1424 err = 0; 1425 goto page_hit; 1426 } 1427 1428 if (parent) 1429 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE); 1430 1431 lock_page(page); 1432 1433 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1434 f2fs_put_page(page, 1); 1435 goto repeat; 1436 } 1437 1438 if (unlikely(!PageUptodate(page))) { 1439 err = -EIO; 1440 goto out_err; 1441 } 1442 1443 if (!f2fs_inode_chksum_verify(sbi, page)) { 1444 err = -EFSBADCRC; 1445 goto out_err; 1446 } 1447 page_hit: 1448 if (likely(nid == nid_of_node(page))) 1449 return page; 1450 1451 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]", 1452 nid, nid_of_node(page), ino_of_node(page), 1453 ofs_of_node(page), cpver_of_node(page), 1454 next_blkaddr_of_node(page)); 1455 set_sbi_flag(sbi, SBI_NEED_FSCK); 1456 err = -EINVAL; 1457 out_err: 1458 ClearPageUptodate(page); 1459 out_put_err: 1460 /* ENOENT comes from read_node_page which is not an error. */ 1461 if (err != -ENOENT) 1462 f2fs_handle_page_eio(sbi, page->index, NODE); 1463 f2fs_put_page(page, 1); 1464 return ERR_PTR(err); 1465 } 1466 1467 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 1468 { 1469 return __get_node_page(sbi, nid, NULL, 0); 1470 } 1471 1472 struct page *f2fs_get_node_page_ra(struct page *parent, int start) 1473 { 1474 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 1475 nid_t nid = get_nid(parent, start, false); 1476 1477 return __get_node_page(sbi, nid, parent, start); 1478 } 1479 1480 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino) 1481 { 1482 struct inode *inode; 1483 struct page *page; 1484 int ret; 1485 1486 /* should flush inline_data before evict_inode */ 1487 inode = ilookup(sbi->sb, ino); 1488 if (!inode) 1489 return; 1490 1491 page = f2fs_pagecache_get_page(inode->i_mapping, 0, 1492 FGP_LOCK|FGP_NOWAIT, 0); 1493 if (!page) 1494 goto iput_out; 1495 1496 if (!PageUptodate(page)) 1497 goto page_out; 1498 1499 if (!PageDirty(page)) 1500 goto page_out; 1501 1502 if (!clear_page_dirty_for_io(page)) 1503 goto page_out; 1504 1505 ret = f2fs_write_inline_data(inode, page); 1506 inode_dec_dirty_pages(inode); 1507 f2fs_remove_dirty_inode(inode); 1508 if (ret) 1509 set_page_dirty(page); 1510 page_out: 1511 f2fs_put_page(page, 1); 1512 iput_out: 1513 iput(inode); 1514 } 1515 1516 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino) 1517 { 1518 pgoff_t index; 1519 struct pagevec pvec; 1520 struct page *last_page = NULL; 1521 int nr_pages; 1522 1523 pagevec_init(&pvec); 1524 index = 0; 1525 1526 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1527 PAGECACHE_TAG_DIRTY))) { 1528 int i; 1529 1530 for (i = 0; i < nr_pages; i++) { 1531 struct page *page = pvec.pages[i]; 1532 1533 if (unlikely(f2fs_cp_error(sbi))) { 1534 f2fs_put_page(last_page, 0); 1535 pagevec_release(&pvec); 1536 return ERR_PTR(-EIO); 1537 } 1538 1539 if (!IS_DNODE(page) || !is_cold_node(page)) 1540 continue; 1541 if (ino_of_node(page) != ino) 1542 continue; 1543 1544 lock_page(page); 1545 1546 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1547 continue_unlock: 1548 unlock_page(page); 1549 continue; 1550 } 1551 if (ino_of_node(page) != ino) 1552 goto continue_unlock; 1553 1554 if (!PageDirty(page)) { 1555 /* someone wrote it for us */ 1556 goto continue_unlock; 1557 } 1558 1559 if (last_page) 1560 f2fs_put_page(last_page, 0); 1561 1562 get_page(page); 1563 last_page = page; 1564 unlock_page(page); 1565 } 1566 pagevec_release(&pvec); 1567 cond_resched(); 1568 } 1569 return last_page; 1570 } 1571 1572 static int __write_node_page(struct page *page, bool atomic, bool *submitted, 1573 struct writeback_control *wbc, bool do_balance, 1574 enum iostat_type io_type, unsigned int *seq_id) 1575 { 1576 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1577 nid_t nid; 1578 struct node_info ni; 1579 struct f2fs_io_info fio = { 1580 .sbi = sbi, 1581 .ino = ino_of_node(page), 1582 .type = NODE, 1583 .op = REQ_OP_WRITE, 1584 .op_flags = wbc_to_write_flags(wbc), 1585 .page = page, 1586 .encrypted_page = NULL, 1587 .submitted = false, 1588 .io_type = io_type, 1589 .io_wbc = wbc, 1590 }; 1591 unsigned int seq; 1592 1593 trace_f2fs_writepage(page, NODE); 1594 1595 if (unlikely(f2fs_cp_error(sbi))) { 1596 ClearPageUptodate(page); 1597 dec_page_count(sbi, F2FS_DIRTY_NODES); 1598 unlock_page(page); 1599 return 0; 1600 } 1601 1602 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1603 goto redirty_out; 1604 1605 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 1606 wbc->sync_mode == WB_SYNC_NONE && 1607 IS_DNODE(page) && is_cold_node(page)) 1608 goto redirty_out; 1609 1610 /* get old block addr of this node page */ 1611 nid = nid_of_node(page); 1612 f2fs_bug_on(sbi, page->index != nid); 1613 1614 if (f2fs_get_node_info(sbi, nid, &ni, !do_balance)) 1615 goto redirty_out; 1616 1617 if (wbc->for_reclaim) { 1618 if (!f2fs_down_read_trylock(&sbi->node_write)) 1619 goto redirty_out; 1620 } else { 1621 f2fs_down_read(&sbi->node_write); 1622 } 1623 1624 /* This page is already truncated */ 1625 if (unlikely(ni.blk_addr == NULL_ADDR)) { 1626 ClearPageUptodate(page); 1627 dec_page_count(sbi, F2FS_DIRTY_NODES); 1628 f2fs_up_read(&sbi->node_write); 1629 unlock_page(page); 1630 return 0; 1631 } 1632 1633 if (__is_valid_data_blkaddr(ni.blk_addr) && 1634 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, 1635 DATA_GENERIC_ENHANCE)) { 1636 f2fs_up_read(&sbi->node_write); 1637 goto redirty_out; 1638 } 1639 1640 if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi)) 1641 fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 1642 1643 /* should add to global list before clearing PAGECACHE status */ 1644 if (f2fs_in_warm_node_list(sbi, page)) { 1645 seq = f2fs_add_fsync_node_entry(sbi, page); 1646 if (seq_id) 1647 *seq_id = seq; 1648 } 1649 1650 set_page_writeback(page); 1651 ClearPageError(page); 1652 1653 fio.old_blkaddr = ni.blk_addr; 1654 f2fs_do_write_node_page(nid, &fio); 1655 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page)); 1656 dec_page_count(sbi, F2FS_DIRTY_NODES); 1657 f2fs_up_read(&sbi->node_write); 1658 1659 if (wbc->for_reclaim) { 1660 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE); 1661 submitted = NULL; 1662 } 1663 1664 unlock_page(page); 1665 1666 if (unlikely(f2fs_cp_error(sbi))) { 1667 f2fs_submit_merged_write(sbi, NODE); 1668 submitted = NULL; 1669 } 1670 if (submitted) 1671 *submitted = fio.submitted; 1672 1673 if (do_balance) 1674 f2fs_balance_fs(sbi, false); 1675 return 0; 1676 1677 redirty_out: 1678 redirty_page_for_writepage(wbc, page); 1679 return AOP_WRITEPAGE_ACTIVATE; 1680 } 1681 1682 int f2fs_move_node_page(struct page *node_page, int gc_type) 1683 { 1684 int err = 0; 1685 1686 if (gc_type == FG_GC) { 1687 struct writeback_control wbc = { 1688 .sync_mode = WB_SYNC_ALL, 1689 .nr_to_write = 1, 1690 .for_reclaim = 0, 1691 }; 1692 1693 f2fs_wait_on_page_writeback(node_page, NODE, true, true); 1694 1695 set_page_dirty(node_page); 1696 1697 if (!clear_page_dirty_for_io(node_page)) { 1698 err = -EAGAIN; 1699 goto out_page; 1700 } 1701 1702 if (__write_node_page(node_page, false, NULL, 1703 &wbc, false, FS_GC_NODE_IO, NULL)) { 1704 err = -EAGAIN; 1705 unlock_page(node_page); 1706 } 1707 goto release_page; 1708 } else { 1709 /* set page dirty and write it */ 1710 if (!PageWriteback(node_page)) 1711 set_page_dirty(node_page); 1712 } 1713 out_page: 1714 unlock_page(node_page); 1715 release_page: 1716 f2fs_put_page(node_page, 0); 1717 return err; 1718 } 1719 1720 static int f2fs_write_node_page(struct page *page, 1721 struct writeback_control *wbc) 1722 { 1723 return __write_node_page(page, false, NULL, wbc, false, 1724 FS_NODE_IO, NULL); 1725 } 1726 1727 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode, 1728 struct writeback_control *wbc, bool atomic, 1729 unsigned int *seq_id) 1730 { 1731 pgoff_t index; 1732 struct pagevec pvec; 1733 int ret = 0; 1734 struct page *last_page = NULL; 1735 bool marked = false; 1736 nid_t ino = inode->i_ino; 1737 int nr_pages; 1738 int nwritten = 0; 1739 1740 if (atomic) { 1741 last_page = last_fsync_dnode(sbi, ino); 1742 if (IS_ERR_OR_NULL(last_page)) 1743 return PTR_ERR_OR_ZERO(last_page); 1744 } 1745 retry: 1746 pagevec_init(&pvec); 1747 index = 0; 1748 1749 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1750 PAGECACHE_TAG_DIRTY))) { 1751 int i; 1752 1753 for (i = 0; i < nr_pages; i++) { 1754 struct page *page = pvec.pages[i]; 1755 bool submitted = false; 1756 1757 if (unlikely(f2fs_cp_error(sbi))) { 1758 f2fs_put_page(last_page, 0); 1759 pagevec_release(&pvec); 1760 ret = -EIO; 1761 goto out; 1762 } 1763 1764 if (!IS_DNODE(page) || !is_cold_node(page)) 1765 continue; 1766 if (ino_of_node(page) != ino) 1767 continue; 1768 1769 lock_page(page); 1770 1771 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1772 continue_unlock: 1773 unlock_page(page); 1774 continue; 1775 } 1776 if (ino_of_node(page) != ino) 1777 goto continue_unlock; 1778 1779 if (!PageDirty(page) && page != last_page) { 1780 /* someone wrote it for us */ 1781 goto continue_unlock; 1782 } 1783 1784 f2fs_wait_on_page_writeback(page, NODE, true, true); 1785 1786 set_fsync_mark(page, 0); 1787 set_dentry_mark(page, 0); 1788 1789 if (!atomic || page == last_page) { 1790 set_fsync_mark(page, 1); 1791 percpu_counter_inc(&sbi->rf_node_block_count); 1792 if (IS_INODE(page)) { 1793 if (is_inode_flag_set(inode, 1794 FI_DIRTY_INODE)) 1795 f2fs_update_inode(inode, page); 1796 set_dentry_mark(page, 1797 f2fs_need_dentry_mark(sbi, ino)); 1798 } 1799 /* may be written by other thread */ 1800 if (!PageDirty(page)) 1801 set_page_dirty(page); 1802 } 1803 1804 if (!clear_page_dirty_for_io(page)) 1805 goto continue_unlock; 1806 1807 ret = __write_node_page(page, atomic && 1808 page == last_page, 1809 &submitted, wbc, true, 1810 FS_NODE_IO, seq_id); 1811 if (ret) { 1812 unlock_page(page); 1813 f2fs_put_page(last_page, 0); 1814 break; 1815 } else if (submitted) { 1816 nwritten++; 1817 } 1818 1819 if (page == last_page) { 1820 f2fs_put_page(page, 0); 1821 marked = true; 1822 break; 1823 } 1824 } 1825 pagevec_release(&pvec); 1826 cond_resched(); 1827 1828 if (ret || marked) 1829 break; 1830 } 1831 if (!ret && atomic && !marked) { 1832 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx", 1833 ino, last_page->index); 1834 lock_page(last_page); 1835 f2fs_wait_on_page_writeback(last_page, NODE, true, true); 1836 set_page_dirty(last_page); 1837 unlock_page(last_page); 1838 goto retry; 1839 } 1840 out: 1841 if (nwritten) 1842 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE); 1843 return ret ? -EIO : 0; 1844 } 1845 1846 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data) 1847 { 1848 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1849 bool clean; 1850 1851 if (inode->i_ino != ino) 1852 return 0; 1853 1854 if (!is_inode_flag_set(inode, FI_DIRTY_INODE)) 1855 return 0; 1856 1857 spin_lock(&sbi->inode_lock[DIRTY_META]); 1858 clean = list_empty(&F2FS_I(inode)->gdirty_list); 1859 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1860 1861 if (clean) 1862 return 0; 1863 1864 inode = igrab(inode); 1865 if (!inode) 1866 return 0; 1867 return 1; 1868 } 1869 1870 static bool flush_dirty_inode(struct page *page) 1871 { 1872 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1873 struct inode *inode; 1874 nid_t ino = ino_of_node(page); 1875 1876 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL); 1877 if (!inode) 1878 return false; 1879 1880 f2fs_update_inode(inode, page); 1881 unlock_page(page); 1882 1883 iput(inode); 1884 return true; 1885 } 1886 1887 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi) 1888 { 1889 pgoff_t index = 0; 1890 struct pagevec pvec; 1891 int nr_pages; 1892 1893 pagevec_init(&pvec); 1894 1895 while ((nr_pages = pagevec_lookup_tag(&pvec, 1896 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) { 1897 int i; 1898 1899 for (i = 0; i < nr_pages; i++) { 1900 struct page *page = pvec.pages[i]; 1901 1902 if (!IS_DNODE(page)) 1903 continue; 1904 1905 lock_page(page); 1906 1907 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1908 continue_unlock: 1909 unlock_page(page); 1910 continue; 1911 } 1912 1913 if (!PageDirty(page)) { 1914 /* someone wrote it for us */ 1915 goto continue_unlock; 1916 } 1917 1918 /* flush inline_data, if it's async context. */ 1919 if (page_private_inline(page)) { 1920 clear_page_private_inline(page); 1921 unlock_page(page); 1922 flush_inline_data(sbi, ino_of_node(page)); 1923 continue; 1924 } 1925 unlock_page(page); 1926 } 1927 pagevec_release(&pvec); 1928 cond_resched(); 1929 } 1930 } 1931 1932 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi, 1933 struct writeback_control *wbc, 1934 bool do_balance, enum iostat_type io_type) 1935 { 1936 pgoff_t index; 1937 struct pagevec pvec; 1938 int step = 0; 1939 int nwritten = 0; 1940 int ret = 0; 1941 int nr_pages, done = 0; 1942 1943 pagevec_init(&pvec); 1944 1945 next_step: 1946 index = 0; 1947 1948 while (!done && (nr_pages = pagevec_lookup_tag(&pvec, 1949 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) { 1950 int i; 1951 1952 for (i = 0; i < nr_pages; i++) { 1953 struct page *page = pvec.pages[i]; 1954 bool submitted = false; 1955 1956 /* give a priority to WB_SYNC threads */ 1957 if (atomic_read(&sbi->wb_sync_req[NODE]) && 1958 wbc->sync_mode == WB_SYNC_NONE) { 1959 done = 1; 1960 break; 1961 } 1962 1963 /* 1964 * flushing sequence with step: 1965 * 0. indirect nodes 1966 * 1. dentry dnodes 1967 * 2. file dnodes 1968 */ 1969 if (step == 0 && IS_DNODE(page)) 1970 continue; 1971 if (step == 1 && (!IS_DNODE(page) || 1972 is_cold_node(page))) 1973 continue; 1974 if (step == 2 && (!IS_DNODE(page) || 1975 !is_cold_node(page))) 1976 continue; 1977 lock_node: 1978 if (wbc->sync_mode == WB_SYNC_ALL) 1979 lock_page(page); 1980 else if (!trylock_page(page)) 1981 continue; 1982 1983 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1984 continue_unlock: 1985 unlock_page(page); 1986 continue; 1987 } 1988 1989 if (!PageDirty(page)) { 1990 /* someone wrote it for us */ 1991 goto continue_unlock; 1992 } 1993 1994 /* flush inline_data/inode, if it's async context. */ 1995 if (!do_balance) 1996 goto write_node; 1997 1998 /* flush inline_data */ 1999 if (page_private_inline(page)) { 2000 clear_page_private_inline(page); 2001 unlock_page(page); 2002 flush_inline_data(sbi, ino_of_node(page)); 2003 goto lock_node; 2004 } 2005 2006 /* flush dirty inode */ 2007 if (IS_INODE(page) && flush_dirty_inode(page)) 2008 goto lock_node; 2009 write_node: 2010 f2fs_wait_on_page_writeback(page, NODE, true, true); 2011 2012 if (!clear_page_dirty_for_io(page)) 2013 goto continue_unlock; 2014 2015 set_fsync_mark(page, 0); 2016 set_dentry_mark(page, 0); 2017 2018 ret = __write_node_page(page, false, &submitted, 2019 wbc, do_balance, io_type, NULL); 2020 if (ret) 2021 unlock_page(page); 2022 else if (submitted) 2023 nwritten++; 2024 2025 if (--wbc->nr_to_write == 0) 2026 break; 2027 } 2028 pagevec_release(&pvec); 2029 cond_resched(); 2030 2031 if (wbc->nr_to_write == 0) { 2032 step = 2; 2033 break; 2034 } 2035 } 2036 2037 if (step < 2) { 2038 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2039 wbc->sync_mode == WB_SYNC_NONE && step == 1) 2040 goto out; 2041 step++; 2042 goto next_step; 2043 } 2044 out: 2045 if (nwritten) 2046 f2fs_submit_merged_write(sbi, NODE); 2047 2048 if (unlikely(f2fs_cp_error(sbi))) 2049 return -EIO; 2050 return ret; 2051 } 2052 2053 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, 2054 unsigned int seq_id) 2055 { 2056 struct fsync_node_entry *fn; 2057 struct page *page; 2058 struct list_head *head = &sbi->fsync_node_list; 2059 unsigned long flags; 2060 unsigned int cur_seq_id = 0; 2061 int ret2, ret = 0; 2062 2063 while (seq_id && cur_seq_id < seq_id) { 2064 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 2065 if (list_empty(head)) { 2066 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2067 break; 2068 } 2069 fn = list_first_entry(head, struct fsync_node_entry, list); 2070 if (fn->seq_id > seq_id) { 2071 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2072 break; 2073 } 2074 cur_seq_id = fn->seq_id; 2075 page = fn->page; 2076 get_page(page); 2077 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2078 2079 f2fs_wait_on_page_writeback(page, NODE, true, false); 2080 if (TestClearPageError(page)) 2081 ret = -EIO; 2082 2083 put_page(page); 2084 2085 if (ret) 2086 break; 2087 } 2088 2089 ret2 = filemap_check_errors(NODE_MAPPING(sbi)); 2090 if (!ret) 2091 ret = ret2; 2092 2093 return ret; 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 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID); 2547 return false; 2548 } 2549 2550 spin_lock(&nm_i->nid_list_lock); 2551 2552 if (unlikely(nm_i->available_nids == 0)) { 2553 spin_unlock(&nm_i->nid_list_lock); 2554 return false; 2555 } 2556 2557 /* We should not use stale free nids created by f2fs_build_free_nids */ 2558 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) { 2559 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list)); 2560 i = list_first_entry(&nm_i->free_nid_list, 2561 struct free_nid, list); 2562 *nid = i->nid; 2563 2564 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID); 2565 nm_i->available_nids--; 2566 2567 update_free_nid_bitmap(sbi, *nid, false, false); 2568 2569 spin_unlock(&nm_i->nid_list_lock); 2570 return true; 2571 } 2572 spin_unlock(&nm_i->nid_list_lock); 2573 2574 /* Let's scan nat pages and its caches to get free nids */ 2575 if (!f2fs_build_free_nids(sbi, true, false)) 2576 goto retry; 2577 return false; 2578 } 2579 2580 /* 2581 * f2fs_alloc_nid() should be called prior to this function. 2582 */ 2583 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 2584 { 2585 struct f2fs_nm_info *nm_i = NM_I(sbi); 2586 struct free_nid *i; 2587 2588 spin_lock(&nm_i->nid_list_lock); 2589 i = __lookup_free_nid_list(nm_i, nid); 2590 f2fs_bug_on(sbi, !i); 2591 __remove_free_nid(sbi, i, PREALLOC_NID); 2592 spin_unlock(&nm_i->nid_list_lock); 2593 2594 kmem_cache_free(free_nid_slab, i); 2595 } 2596 2597 /* 2598 * f2fs_alloc_nid() should be called prior to this function. 2599 */ 2600 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 2601 { 2602 struct f2fs_nm_info *nm_i = NM_I(sbi); 2603 struct free_nid *i; 2604 bool need_free = false; 2605 2606 if (!nid) 2607 return; 2608 2609 spin_lock(&nm_i->nid_list_lock); 2610 i = __lookup_free_nid_list(nm_i, nid); 2611 f2fs_bug_on(sbi, !i); 2612 2613 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) { 2614 __remove_free_nid(sbi, i, PREALLOC_NID); 2615 need_free = true; 2616 } else { 2617 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID); 2618 } 2619 2620 nm_i->available_nids++; 2621 2622 update_free_nid_bitmap(sbi, nid, true, false); 2623 2624 spin_unlock(&nm_i->nid_list_lock); 2625 2626 if (need_free) 2627 kmem_cache_free(free_nid_slab, i); 2628 } 2629 2630 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink) 2631 { 2632 struct f2fs_nm_info *nm_i = NM_I(sbi); 2633 int nr = nr_shrink; 2634 2635 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2636 return 0; 2637 2638 if (!mutex_trylock(&nm_i->build_lock)) 2639 return 0; 2640 2641 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) { 2642 struct free_nid *i, *next; 2643 unsigned int batch = SHRINK_NID_BATCH_SIZE; 2644 2645 spin_lock(&nm_i->nid_list_lock); 2646 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) { 2647 if (!nr_shrink || !batch || 2648 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2649 break; 2650 __remove_free_nid(sbi, i, FREE_NID); 2651 kmem_cache_free(free_nid_slab, i); 2652 nr_shrink--; 2653 batch--; 2654 } 2655 spin_unlock(&nm_i->nid_list_lock); 2656 } 2657 2658 mutex_unlock(&nm_i->build_lock); 2659 2660 return nr - nr_shrink; 2661 } 2662 2663 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page) 2664 { 2665 void *src_addr, *dst_addr; 2666 size_t inline_size; 2667 struct page *ipage; 2668 struct f2fs_inode *ri; 2669 2670 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 2671 if (IS_ERR(ipage)) 2672 return PTR_ERR(ipage); 2673 2674 ri = F2FS_INODE(page); 2675 if (ri->i_inline & F2FS_INLINE_XATTR) { 2676 if (!f2fs_has_inline_xattr(inode)) { 2677 set_inode_flag(inode, FI_INLINE_XATTR); 2678 stat_inc_inline_xattr(inode); 2679 } 2680 } else { 2681 if (f2fs_has_inline_xattr(inode)) { 2682 stat_dec_inline_xattr(inode); 2683 clear_inode_flag(inode, FI_INLINE_XATTR); 2684 } 2685 goto update_inode; 2686 } 2687 2688 dst_addr = inline_xattr_addr(inode, ipage); 2689 src_addr = inline_xattr_addr(inode, page); 2690 inline_size = inline_xattr_size(inode); 2691 2692 f2fs_wait_on_page_writeback(ipage, NODE, true, true); 2693 memcpy(dst_addr, src_addr, inline_size); 2694 update_inode: 2695 f2fs_update_inode(inode, ipage); 2696 f2fs_put_page(ipage, 1); 2697 return 0; 2698 } 2699 2700 int f2fs_recover_xattr_data(struct inode *inode, struct page *page) 2701 { 2702 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2703 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid; 2704 nid_t new_xnid; 2705 struct dnode_of_data dn; 2706 struct node_info ni; 2707 struct page *xpage; 2708 int err; 2709 2710 if (!prev_xnid) 2711 goto recover_xnid; 2712 2713 /* 1: invalidate the previous xattr nid */ 2714 err = f2fs_get_node_info(sbi, prev_xnid, &ni, false); 2715 if (err) 2716 return err; 2717 2718 f2fs_invalidate_blocks(sbi, ni.blk_addr); 2719 dec_valid_node_count(sbi, inode, false); 2720 set_node_addr(sbi, &ni, NULL_ADDR, false); 2721 2722 recover_xnid: 2723 /* 2: update xattr nid in inode */ 2724 if (!f2fs_alloc_nid(sbi, &new_xnid)) 2725 return -ENOSPC; 2726 2727 set_new_dnode(&dn, inode, NULL, NULL, new_xnid); 2728 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET); 2729 if (IS_ERR(xpage)) { 2730 f2fs_alloc_nid_failed(sbi, new_xnid); 2731 return PTR_ERR(xpage); 2732 } 2733 2734 f2fs_alloc_nid_done(sbi, new_xnid); 2735 f2fs_update_inode_page(inode); 2736 2737 /* 3: update and set xattr node page dirty */ 2738 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE); 2739 2740 set_page_dirty(xpage); 2741 f2fs_put_page(xpage, 1); 2742 2743 return 0; 2744 } 2745 2746 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 2747 { 2748 struct f2fs_inode *src, *dst; 2749 nid_t ino = ino_of_node(page); 2750 struct node_info old_ni, new_ni; 2751 struct page *ipage; 2752 int err; 2753 2754 err = f2fs_get_node_info(sbi, ino, &old_ni, false); 2755 if (err) 2756 return err; 2757 2758 if (unlikely(old_ni.blk_addr != NULL_ADDR)) 2759 return -EINVAL; 2760 retry: 2761 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false); 2762 if (!ipage) { 2763 memalloc_retry_wait(GFP_NOFS); 2764 goto retry; 2765 } 2766 2767 /* Should not use this inode from free nid list */ 2768 remove_free_nid(sbi, ino); 2769 2770 if (!PageUptodate(ipage)) 2771 SetPageUptodate(ipage); 2772 fill_node_footer(ipage, ino, ino, 0, true); 2773 set_cold_node(ipage, false); 2774 2775 src = F2FS_INODE(page); 2776 dst = F2FS_INODE(ipage); 2777 2778 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext)); 2779 dst->i_size = 0; 2780 dst->i_blocks = cpu_to_le64(1); 2781 dst->i_links = cpu_to_le32(1); 2782 dst->i_xattr_nid = 0; 2783 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR); 2784 if (dst->i_inline & F2FS_EXTRA_ATTR) { 2785 dst->i_extra_isize = src->i_extra_isize; 2786 2787 if (f2fs_sb_has_flexible_inline_xattr(sbi) && 2788 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2789 i_inline_xattr_size)) 2790 dst->i_inline_xattr_size = src->i_inline_xattr_size; 2791 2792 if (f2fs_sb_has_project_quota(sbi) && 2793 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2794 i_projid)) 2795 dst->i_projid = src->i_projid; 2796 2797 if (f2fs_sb_has_inode_crtime(sbi) && 2798 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2799 i_crtime_nsec)) { 2800 dst->i_crtime = src->i_crtime; 2801 dst->i_crtime_nsec = src->i_crtime_nsec; 2802 } 2803 } 2804 2805 new_ni = old_ni; 2806 new_ni.ino = ino; 2807 2808 if (unlikely(inc_valid_node_count(sbi, NULL, true))) 2809 WARN_ON(1); 2810 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 2811 inc_valid_inode_count(sbi); 2812 set_page_dirty(ipage); 2813 f2fs_put_page(ipage, 1); 2814 return 0; 2815 } 2816 2817 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi, 2818 unsigned int segno, struct f2fs_summary_block *sum) 2819 { 2820 struct f2fs_node *rn; 2821 struct f2fs_summary *sum_entry; 2822 block_t addr; 2823 int i, idx, last_offset, nrpages; 2824 2825 /* scan the node segment */ 2826 last_offset = sbi->blocks_per_seg; 2827 addr = START_BLOCK(sbi, segno); 2828 sum_entry = &sum->entries[0]; 2829 2830 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) { 2831 nrpages = bio_max_segs(last_offset - i); 2832 2833 /* readahead node pages */ 2834 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true); 2835 2836 for (idx = addr; idx < addr + nrpages; idx++) { 2837 struct page *page = f2fs_get_tmp_page(sbi, idx); 2838 2839 if (IS_ERR(page)) 2840 return PTR_ERR(page); 2841 2842 rn = F2FS_NODE(page); 2843 sum_entry->nid = rn->footer.nid; 2844 sum_entry->version = 0; 2845 sum_entry->ofs_in_node = 0; 2846 sum_entry++; 2847 f2fs_put_page(page, 1); 2848 } 2849 2850 invalidate_mapping_pages(META_MAPPING(sbi), addr, 2851 addr + nrpages); 2852 } 2853 return 0; 2854 } 2855 2856 static void remove_nats_in_journal(struct f2fs_sb_info *sbi) 2857 { 2858 struct f2fs_nm_info *nm_i = NM_I(sbi); 2859 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2860 struct f2fs_journal *journal = curseg->journal; 2861 int i; 2862 2863 down_write(&curseg->journal_rwsem); 2864 for (i = 0; i < nats_in_cursum(journal); i++) { 2865 struct nat_entry *ne; 2866 struct f2fs_nat_entry raw_ne; 2867 nid_t nid = le32_to_cpu(nid_in_journal(journal, i)); 2868 2869 if (f2fs_check_nid_range(sbi, nid)) 2870 continue; 2871 2872 raw_ne = nat_in_journal(journal, i); 2873 2874 ne = __lookup_nat_cache(nm_i, nid); 2875 if (!ne) { 2876 ne = __alloc_nat_entry(sbi, nid, true); 2877 __init_nat_entry(nm_i, ne, &raw_ne, true); 2878 } 2879 2880 /* 2881 * if a free nat in journal has not been used after last 2882 * checkpoint, we should remove it from available nids, 2883 * since later we will add it again. 2884 */ 2885 if (!get_nat_flag(ne, IS_DIRTY) && 2886 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) { 2887 spin_lock(&nm_i->nid_list_lock); 2888 nm_i->available_nids--; 2889 spin_unlock(&nm_i->nid_list_lock); 2890 } 2891 2892 __set_nat_cache_dirty(nm_i, ne); 2893 } 2894 update_nats_in_cursum(journal, -i); 2895 up_write(&curseg->journal_rwsem); 2896 } 2897 2898 static void __adjust_nat_entry_set(struct nat_entry_set *nes, 2899 struct list_head *head, int max) 2900 { 2901 struct nat_entry_set *cur; 2902 2903 if (nes->entry_cnt >= max) 2904 goto add_out; 2905 2906 list_for_each_entry(cur, head, set_list) { 2907 if (cur->entry_cnt >= nes->entry_cnt) { 2908 list_add(&nes->set_list, cur->set_list.prev); 2909 return; 2910 } 2911 } 2912 add_out: 2913 list_add_tail(&nes->set_list, head); 2914 } 2915 2916 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs, 2917 unsigned int valid) 2918 { 2919 if (valid == 0) { 2920 __set_bit_le(nat_ofs, nm_i->empty_nat_bits); 2921 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2922 return; 2923 } 2924 2925 __clear_bit_le(nat_ofs, nm_i->empty_nat_bits); 2926 if (valid == NAT_ENTRY_PER_BLOCK) 2927 __set_bit_le(nat_ofs, nm_i->full_nat_bits); 2928 else 2929 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2930 } 2931 2932 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid, 2933 struct page *page) 2934 { 2935 struct f2fs_nm_info *nm_i = NM_I(sbi); 2936 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK; 2937 struct f2fs_nat_block *nat_blk = page_address(page); 2938 int valid = 0; 2939 int i = 0; 2940 2941 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 2942 return; 2943 2944 if (nat_index == 0) { 2945 valid = 1; 2946 i = 1; 2947 } 2948 for (; i < NAT_ENTRY_PER_BLOCK; i++) { 2949 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR) 2950 valid++; 2951 } 2952 2953 __update_nat_bits(nm_i, nat_index, valid); 2954 } 2955 2956 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi) 2957 { 2958 struct f2fs_nm_info *nm_i = NM_I(sbi); 2959 unsigned int nat_ofs; 2960 2961 f2fs_down_read(&nm_i->nat_tree_lock); 2962 2963 for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) { 2964 unsigned int valid = 0, nid_ofs = 0; 2965 2966 /* handle nid zero due to it should never be used */ 2967 if (unlikely(nat_ofs == 0)) { 2968 valid = 1; 2969 nid_ofs = 1; 2970 } 2971 2972 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) { 2973 if (!test_bit_le(nid_ofs, 2974 nm_i->free_nid_bitmap[nat_ofs])) 2975 valid++; 2976 } 2977 2978 __update_nat_bits(nm_i, nat_ofs, valid); 2979 } 2980 2981 f2fs_up_read(&nm_i->nat_tree_lock); 2982 } 2983 2984 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi, 2985 struct nat_entry_set *set, struct cp_control *cpc) 2986 { 2987 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2988 struct f2fs_journal *journal = curseg->journal; 2989 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK; 2990 bool to_journal = true; 2991 struct f2fs_nat_block *nat_blk; 2992 struct nat_entry *ne, *cur; 2993 struct page *page = NULL; 2994 2995 /* 2996 * there are two steps to flush nat entries: 2997 * #1, flush nat entries to journal in current hot data summary block. 2998 * #2, flush nat entries to nat page. 2999 */ 3000 if ((cpc->reason & CP_UMOUNT) || 3001 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL)) 3002 to_journal = false; 3003 3004 if (to_journal) { 3005 down_write(&curseg->journal_rwsem); 3006 } else { 3007 page = get_next_nat_page(sbi, start_nid); 3008 if (IS_ERR(page)) 3009 return PTR_ERR(page); 3010 3011 nat_blk = page_address(page); 3012 f2fs_bug_on(sbi, !nat_blk); 3013 } 3014 3015 /* flush dirty nats in nat entry set */ 3016 list_for_each_entry_safe(ne, cur, &set->entry_list, list) { 3017 struct f2fs_nat_entry *raw_ne; 3018 nid_t nid = nat_get_nid(ne); 3019 int offset; 3020 3021 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR); 3022 3023 if (to_journal) { 3024 offset = f2fs_lookup_journal_in_cursum(journal, 3025 NAT_JOURNAL, nid, 1); 3026 f2fs_bug_on(sbi, offset < 0); 3027 raw_ne = &nat_in_journal(journal, offset); 3028 nid_in_journal(journal, offset) = cpu_to_le32(nid); 3029 } else { 3030 raw_ne = &nat_blk->entries[nid - start_nid]; 3031 } 3032 raw_nat_from_node_info(raw_ne, &ne->ni); 3033 nat_reset_flag(ne); 3034 __clear_nat_cache_dirty(NM_I(sbi), set, ne); 3035 if (nat_get_blkaddr(ne) == NULL_ADDR) { 3036 add_free_nid(sbi, nid, false, true); 3037 } else { 3038 spin_lock(&NM_I(sbi)->nid_list_lock); 3039 update_free_nid_bitmap(sbi, nid, false, false); 3040 spin_unlock(&NM_I(sbi)->nid_list_lock); 3041 } 3042 } 3043 3044 if (to_journal) { 3045 up_write(&curseg->journal_rwsem); 3046 } else { 3047 update_nat_bits(sbi, start_nid, page); 3048 f2fs_put_page(page, 1); 3049 } 3050 3051 /* Allow dirty nats by node block allocation in write_begin */ 3052 if (!set->entry_cnt) { 3053 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set); 3054 kmem_cache_free(nat_entry_set_slab, set); 3055 } 3056 return 0; 3057 } 3058 3059 /* 3060 * This function is called during the checkpointing process. 3061 */ 3062 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3063 { 3064 struct f2fs_nm_info *nm_i = NM_I(sbi); 3065 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 3066 struct f2fs_journal *journal = curseg->journal; 3067 struct nat_entry_set *setvec[SETVEC_SIZE]; 3068 struct nat_entry_set *set, *tmp; 3069 unsigned int found; 3070 nid_t set_idx = 0; 3071 LIST_HEAD(sets); 3072 int err = 0; 3073 3074 /* 3075 * during unmount, let's flush nat_bits before checking 3076 * nat_cnt[DIRTY_NAT]. 3077 */ 3078 if (cpc->reason & CP_UMOUNT) { 3079 f2fs_down_write(&nm_i->nat_tree_lock); 3080 remove_nats_in_journal(sbi); 3081 f2fs_up_write(&nm_i->nat_tree_lock); 3082 } 3083 3084 if (!nm_i->nat_cnt[DIRTY_NAT]) 3085 return 0; 3086 3087 f2fs_down_write(&nm_i->nat_tree_lock); 3088 3089 /* 3090 * if there are no enough space in journal to store dirty nat 3091 * entries, remove all entries from journal and merge them 3092 * into nat entry set. 3093 */ 3094 if (cpc->reason & CP_UMOUNT || 3095 !__has_cursum_space(journal, 3096 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL)) 3097 remove_nats_in_journal(sbi); 3098 3099 while ((found = __gang_lookup_nat_set(nm_i, 3100 set_idx, SETVEC_SIZE, setvec))) { 3101 unsigned idx; 3102 3103 set_idx = setvec[found - 1]->set + 1; 3104 for (idx = 0; idx < found; idx++) 3105 __adjust_nat_entry_set(setvec[idx], &sets, 3106 MAX_NAT_JENTRIES(journal)); 3107 } 3108 3109 /* flush dirty nats in nat entry set */ 3110 list_for_each_entry_safe(set, tmp, &sets, set_list) { 3111 err = __flush_nat_entry_set(sbi, set, cpc); 3112 if (err) 3113 break; 3114 } 3115 3116 f2fs_up_write(&nm_i->nat_tree_lock); 3117 /* Allow dirty nats by node block allocation in write_begin */ 3118 3119 return err; 3120 } 3121 3122 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi) 3123 { 3124 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3125 struct f2fs_nm_info *nm_i = NM_I(sbi); 3126 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE; 3127 unsigned int i; 3128 __u64 cp_ver = cur_cp_version(ckpt); 3129 block_t nat_bits_addr; 3130 3131 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8); 3132 nm_i->nat_bits = f2fs_kvzalloc(sbi, 3133 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL); 3134 if (!nm_i->nat_bits) 3135 return -ENOMEM; 3136 3137 nm_i->full_nat_bits = nm_i->nat_bits + 8; 3138 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes; 3139 3140 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3141 return 0; 3142 3143 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg - 3144 nm_i->nat_bits_blocks; 3145 for (i = 0; i < nm_i->nat_bits_blocks; i++) { 3146 struct page *page; 3147 3148 page = f2fs_get_meta_page(sbi, nat_bits_addr++); 3149 if (IS_ERR(page)) 3150 return PTR_ERR(page); 3151 3152 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS), 3153 page_address(page), F2FS_BLKSIZE); 3154 f2fs_put_page(page, 1); 3155 } 3156 3157 cp_ver |= (cur_cp_crc(ckpt) << 32); 3158 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) { 3159 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG); 3160 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)", 3161 cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits)); 3162 return 0; 3163 } 3164 3165 f2fs_notice(sbi, "Found nat_bits in checkpoint"); 3166 return 0; 3167 } 3168 3169 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi) 3170 { 3171 struct f2fs_nm_info *nm_i = NM_I(sbi); 3172 unsigned int i = 0; 3173 nid_t nid, last_nid; 3174 3175 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3176 return; 3177 3178 for (i = 0; i < nm_i->nat_blocks; i++) { 3179 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i); 3180 if (i >= nm_i->nat_blocks) 3181 break; 3182 3183 __set_bit_le(i, nm_i->nat_block_bitmap); 3184 3185 nid = i * NAT_ENTRY_PER_BLOCK; 3186 last_nid = nid + NAT_ENTRY_PER_BLOCK; 3187 3188 spin_lock(&NM_I(sbi)->nid_list_lock); 3189 for (; nid < last_nid; nid++) 3190 update_free_nid_bitmap(sbi, nid, true, true); 3191 spin_unlock(&NM_I(sbi)->nid_list_lock); 3192 } 3193 3194 for (i = 0; i < nm_i->nat_blocks; i++) { 3195 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i); 3196 if (i >= nm_i->nat_blocks) 3197 break; 3198 3199 __set_bit_le(i, nm_i->nat_block_bitmap); 3200 } 3201 } 3202 3203 static int init_node_manager(struct f2fs_sb_info *sbi) 3204 { 3205 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 3206 struct f2fs_nm_info *nm_i = NM_I(sbi); 3207 unsigned char *version_bitmap; 3208 unsigned int nat_segs; 3209 int err; 3210 3211 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 3212 3213 /* segment_count_nat includes pair segment so divide to 2. */ 3214 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 3215 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 3216 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks; 3217 3218 /* not used nids: 0, node, meta, (and root counted as valid node) */ 3219 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count - 3220 F2FS_RESERVED_NODE_NUM; 3221 nm_i->nid_cnt[FREE_NID] = 0; 3222 nm_i->nid_cnt[PREALLOC_NID] = 0; 3223 nm_i->ram_thresh = DEF_RAM_THRESHOLD; 3224 nm_i->ra_nid_pages = DEF_RA_NID_PAGES; 3225 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD; 3226 nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS; 3227 3228 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC); 3229 INIT_LIST_HEAD(&nm_i->free_nid_list); 3230 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO); 3231 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO); 3232 INIT_LIST_HEAD(&nm_i->nat_entries); 3233 spin_lock_init(&nm_i->nat_list_lock); 3234 3235 mutex_init(&nm_i->build_lock); 3236 spin_lock_init(&nm_i->nid_list_lock); 3237 init_f2fs_rwsem(&nm_i->nat_tree_lock); 3238 3239 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 3240 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 3241 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 3242 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size, 3243 GFP_KERNEL); 3244 if (!nm_i->nat_bitmap) 3245 return -ENOMEM; 3246 3247 err = __get_nat_bitmaps(sbi); 3248 if (err) 3249 return err; 3250 3251 #ifdef CONFIG_F2FS_CHECK_FS 3252 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size, 3253 GFP_KERNEL); 3254 if (!nm_i->nat_bitmap_mir) 3255 return -ENOMEM; 3256 #endif 3257 3258 return 0; 3259 } 3260 3261 static int init_free_nid_cache(struct f2fs_sb_info *sbi) 3262 { 3263 struct f2fs_nm_info *nm_i = NM_I(sbi); 3264 int i; 3265 3266 nm_i->free_nid_bitmap = 3267 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *), 3268 nm_i->nat_blocks), 3269 GFP_KERNEL); 3270 if (!nm_i->free_nid_bitmap) 3271 return -ENOMEM; 3272 3273 for (i = 0; i < nm_i->nat_blocks; i++) { 3274 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi, 3275 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL); 3276 if (!nm_i->free_nid_bitmap[i]) 3277 return -ENOMEM; 3278 } 3279 3280 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8, 3281 GFP_KERNEL); 3282 if (!nm_i->nat_block_bitmap) 3283 return -ENOMEM; 3284 3285 nm_i->free_nid_count = 3286 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short), 3287 nm_i->nat_blocks), 3288 GFP_KERNEL); 3289 if (!nm_i->free_nid_count) 3290 return -ENOMEM; 3291 return 0; 3292 } 3293 3294 int f2fs_build_node_manager(struct f2fs_sb_info *sbi) 3295 { 3296 int err; 3297 3298 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info), 3299 GFP_KERNEL); 3300 if (!sbi->nm_info) 3301 return -ENOMEM; 3302 3303 err = init_node_manager(sbi); 3304 if (err) 3305 return err; 3306 3307 err = init_free_nid_cache(sbi); 3308 if (err) 3309 return err; 3310 3311 /* load free nid status from nat_bits table */ 3312 load_free_nid_bitmap(sbi); 3313 3314 return f2fs_build_free_nids(sbi, true, true); 3315 } 3316 3317 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi) 3318 { 3319 struct f2fs_nm_info *nm_i = NM_I(sbi); 3320 struct free_nid *i, *next_i; 3321 struct nat_entry *natvec[NATVEC_SIZE]; 3322 struct nat_entry_set *setvec[SETVEC_SIZE]; 3323 nid_t nid = 0; 3324 unsigned int found; 3325 3326 if (!nm_i) 3327 return; 3328 3329 /* destroy free nid list */ 3330 spin_lock(&nm_i->nid_list_lock); 3331 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 3332 __remove_free_nid(sbi, i, FREE_NID); 3333 spin_unlock(&nm_i->nid_list_lock); 3334 kmem_cache_free(free_nid_slab, i); 3335 spin_lock(&nm_i->nid_list_lock); 3336 } 3337 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]); 3338 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]); 3339 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list)); 3340 spin_unlock(&nm_i->nid_list_lock); 3341 3342 /* destroy nat cache */ 3343 f2fs_down_write(&nm_i->nat_tree_lock); 3344 while ((found = __gang_lookup_nat_cache(nm_i, 3345 nid, NATVEC_SIZE, natvec))) { 3346 unsigned idx; 3347 3348 nid = nat_get_nid(natvec[found - 1]) + 1; 3349 for (idx = 0; idx < found; idx++) { 3350 spin_lock(&nm_i->nat_list_lock); 3351 list_del(&natvec[idx]->list); 3352 spin_unlock(&nm_i->nat_list_lock); 3353 3354 __del_from_nat_cache(nm_i, natvec[idx]); 3355 } 3356 } 3357 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]); 3358 3359 /* destroy nat set cache */ 3360 nid = 0; 3361 while ((found = __gang_lookup_nat_set(nm_i, 3362 nid, SETVEC_SIZE, setvec))) { 3363 unsigned idx; 3364 3365 nid = setvec[found - 1]->set + 1; 3366 for (idx = 0; idx < found; idx++) { 3367 /* entry_cnt is not zero, when cp_error was occurred */ 3368 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list)); 3369 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set); 3370 kmem_cache_free(nat_entry_set_slab, setvec[idx]); 3371 } 3372 } 3373 f2fs_up_write(&nm_i->nat_tree_lock); 3374 3375 kvfree(nm_i->nat_block_bitmap); 3376 if (nm_i->free_nid_bitmap) { 3377 int i; 3378 3379 for (i = 0; i < nm_i->nat_blocks; i++) 3380 kvfree(nm_i->free_nid_bitmap[i]); 3381 kvfree(nm_i->free_nid_bitmap); 3382 } 3383 kvfree(nm_i->free_nid_count); 3384 3385 kvfree(nm_i->nat_bitmap); 3386 kvfree(nm_i->nat_bits); 3387 #ifdef CONFIG_F2FS_CHECK_FS 3388 kvfree(nm_i->nat_bitmap_mir); 3389 #endif 3390 sbi->nm_info = NULL; 3391 kfree(nm_i); 3392 } 3393 3394 int __init f2fs_create_node_manager_caches(void) 3395 { 3396 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry", 3397 sizeof(struct nat_entry)); 3398 if (!nat_entry_slab) 3399 goto fail; 3400 3401 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid", 3402 sizeof(struct free_nid)); 3403 if (!free_nid_slab) 3404 goto destroy_nat_entry; 3405 3406 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set", 3407 sizeof(struct nat_entry_set)); 3408 if (!nat_entry_set_slab) 3409 goto destroy_free_nid; 3410 3411 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry", 3412 sizeof(struct fsync_node_entry)); 3413 if (!fsync_node_entry_slab) 3414 goto destroy_nat_entry_set; 3415 return 0; 3416 3417 destroy_nat_entry_set: 3418 kmem_cache_destroy(nat_entry_set_slab); 3419 destroy_free_nid: 3420 kmem_cache_destroy(free_nid_slab); 3421 destroy_nat_entry: 3422 kmem_cache_destroy(nat_entry_slab); 3423 fail: 3424 return -ENOMEM; 3425 } 3426 3427 void f2fs_destroy_node_manager_caches(void) 3428 { 3429 kmem_cache_destroy(fsync_node_entry_slab); 3430 kmem_cache_destroy(nat_entry_set_slab); 3431 kmem_cache_destroy(free_nid_slab); 3432 kmem_cache_destroy(nat_entry_slab); 3433 } 3434