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/backing-dev.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 return -EFSCORRUPTED; 40 } 41 return 0; 42 } 43 44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type) 45 { 46 struct f2fs_nm_info *nm_i = NM_I(sbi); 47 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 48 struct sysinfo val; 49 unsigned long avail_ram; 50 unsigned long mem_size = 0; 51 bool res = false; 52 53 if (!nm_i) 54 return true; 55 56 si_meminfo(&val); 57 58 /* only uses low memory */ 59 avail_ram = val.totalram - val.totalhigh; 60 61 /* 62 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively 63 */ 64 if (type == FREE_NIDS) { 65 mem_size = (nm_i->nid_cnt[FREE_NID] * 66 sizeof(struct free_nid)) >> PAGE_SHIFT; 67 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 68 } else if (type == NAT_ENTRIES) { 69 mem_size = (nm_i->nat_cnt[TOTAL_NAT] * 70 sizeof(struct nat_entry)) >> PAGE_SHIFT; 71 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 72 if (excess_cached_nats(sbi)) 73 res = false; 74 } else if (type == DIRTY_DENTS) { 75 if (sbi->sb->s_bdi->wb.dirty_exceeded) 76 return false; 77 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS); 78 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 79 } else if (type == INO_ENTRIES) { 80 int i; 81 82 for (i = 0; i < MAX_INO_ENTRY; i++) 83 mem_size += sbi->im[i].ino_num * 84 sizeof(struct ino_entry); 85 mem_size >>= PAGE_SHIFT; 86 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 87 } else if (type == EXTENT_CACHE) { 88 mem_size = (atomic_read(&sbi->total_ext_tree) * 89 sizeof(struct extent_tree) + 90 atomic_read(&sbi->total_ext_node) * 91 sizeof(struct extent_node)) >> PAGE_SHIFT; 92 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 93 } else if (type == INMEM_PAGES) { 94 /* it allows 20% / total_ram for inmemory pages */ 95 mem_size = get_pages(sbi, F2FS_INMEM_PAGES); 96 res = mem_size < (val.totalram / 5); 97 } else if (type == DISCARD_CACHE) { 98 mem_size = (atomic_read(&dcc->discard_cmd_cnt) * 99 sizeof(struct discard_cmd)) >> PAGE_SHIFT; 100 res = mem_size < (avail_ram * nm_i->ram_thresh / 100); 101 } else if (type == COMPRESS_PAGE) { 102 #ifdef CONFIG_F2FS_FS_COMPRESSION 103 unsigned long free_ram = val.freeram; 104 105 /* 106 * free memory is lower than watermark or cached page count 107 * exceed threshold, deny caching compress page. 108 */ 109 res = (free_ram > avail_ram * sbi->compress_watermark / 100) && 110 (COMPRESS_MAPPING(sbi)->nrpages < 111 free_ram * sbi->compress_percent / 100); 112 #else 113 res = false; 114 #endif 115 } else { 116 if (!sbi->sb->s_bdi->wb.dirty_exceeded) 117 return true; 118 } 119 return res; 120 } 121 122 static void clear_node_page_dirty(struct page *page) 123 { 124 if (PageDirty(page)) { 125 f2fs_clear_page_cache_dirty_tag(page); 126 clear_page_dirty_for_io(page); 127 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES); 128 } 129 ClearPageUptodate(page); 130 } 131 132 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 133 { 134 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid)); 135 } 136 137 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 138 { 139 struct page *src_page; 140 struct page *dst_page; 141 pgoff_t dst_off; 142 void *src_addr; 143 void *dst_addr; 144 struct f2fs_nm_info *nm_i = NM_I(sbi); 145 146 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid)); 147 148 /* get current nat block page with lock */ 149 src_page = get_current_nat_page(sbi, nid); 150 if (IS_ERR(src_page)) 151 return src_page; 152 dst_page = f2fs_grab_meta_page(sbi, dst_off); 153 f2fs_bug_on(sbi, PageDirty(src_page)); 154 155 src_addr = page_address(src_page); 156 dst_addr = page_address(dst_page); 157 memcpy(dst_addr, src_addr, PAGE_SIZE); 158 set_page_dirty(dst_page); 159 f2fs_put_page(src_page, 1); 160 161 set_to_next_nat(nm_i, nid); 162 163 return dst_page; 164 } 165 166 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi, 167 nid_t nid, bool no_fail) 168 { 169 struct nat_entry *new; 170 171 new = f2fs_kmem_cache_alloc(nat_entry_slab, 172 GFP_F2FS_ZERO, no_fail, sbi); 173 if (new) { 174 nat_set_nid(new, nid); 175 nat_reset_flag(new); 176 } 177 return new; 178 } 179 180 static void __free_nat_entry(struct nat_entry *e) 181 { 182 kmem_cache_free(nat_entry_slab, e); 183 } 184 185 /* must be locked by nat_tree_lock */ 186 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i, 187 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail) 188 { 189 if (no_fail) 190 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne); 191 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne)) 192 return NULL; 193 194 if (raw_ne) 195 node_info_from_raw_nat(&ne->ni, raw_ne); 196 197 spin_lock(&nm_i->nat_list_lock); 198 list_add_tail(&ne->list, &nm_i->nat_entries); 199 spin_unlock(&nm_i->nat_list_lock); 200 201 nm_i->nat_cnt[TOTAL_NAT]++; 202 nm_i->nat_cnt[RECLAIMABLE_NAT]++; 203 return ne; 204 } 205 206 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n) 207 { 208 struct nat_entry *ne; 209 210 ne = radix_tree_lookup(&nm_i->nat_root, n); 211 212 /* for recent accessed nat entry, move it to tail of lru list */ 213 if (ne && !get_nat_flag(ne, IS_DIRTY)) { 214 spin_lock(&nm_i->nat_list_lock); 215 if (!list_empty(&ne->list)) 216 list_move_tail(&ne->list, &nm_i->nat_entries); 217 spin_unlock(&nm_i->nat_list_lock); 218 } 219 220 return ne; 221 } 222 223 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i, 224 nid_t start, unsigned int nr, struct nat_entry **ep) 225 { 226 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr); 227 } 228 229 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e) 230 { 231 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e)); 232 nm_i->nat_cnt[TOTAL_NAT]--; 233 nm_i->nat_cnt[RECLAIMABLE_NAT]--; 234 __free_nat_entry(e); 235 } 236 237 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i, 238 struct nat_entry *ne) 239 { 240 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid); 241 struct nat_entry_set *head; 242 243 head = radix_tree_lookup(&nm_i->nat_set_root, set); 244 if (!head) { 245 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, 246 GFP_NOFS, true, NULL); 247 248 INIT_LIST_HEAD(&head->entry_list); 249 INIT_LIST_HEAD(&head->set_list); 250 head->set = set; 251 head->entry_cnt = 0; 252 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head); 253 } 254 return head; 255 } 256 257 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i, 258 struct nat_entry *ne) 259 { 260 struct nat_entry_set *head; 261 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR; 262 263 if (!new_ne) 264 head = __grab_nat_entry_set(nm_i, ne); 265 266 /* 267 * update entry_cnt in below condition: 268 * 1. update NEW_ADDR to valid block address; 269 * 2. update old block address to new one; 270 */ 271 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) || 272 !get_nat_flag(ne, IS_DIRTY))) 273 head->entry_cnt++; 274 275 set_nat_flag(ne, IS_PREALLOC, new_ne); 276 277 if (get_nat_flag(ne, IS_DIRTY)) 278 goto refresh_list; 279 280 nm_i->nat_cnt[DIRTY_NAT]++; 281 nm_i->nat_cnt[RECLAIMABLE_NAT]--; 282 set_nat_flag(ne, IS_DIRTY, true); 283 refresh_list: 284 spin_lock(&nm_i->nat_list_lock); 285 if (new_ne) 286 list_del_init(&ne->list); 287 else 288 list_move_tail(&ne->list, &head->entry_list); 289 spin_unlock(&nm_i->nat_list_lock); 290 } 291 292 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i, 293 struct nat_entry_set *set, struct nat_entry *ne) 294 { 295 spin_lock(&nm_i->nat_list_lock); 296 list_move_tail(&ne->list, &nm_i->nat_entries); 297 spin_unlock(&nm_i->nat_list_lock); 298 299 set_nat_flag(ne, IS_DIRTY, false); 300 set->entry_cnt--; 301 nm_i->nat_cnt[DIRTY_NAT]--; 302 nm_i->nat_cnt[RECLAIMABLE_NAT]++; 303 } 304 305 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i, 306 nid_t start, unsigned int nr, struct nat_entry_set **ep) 307 { 308 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep, 309 start, nr); 310 } 311 312 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page) 313 { 314 return NODE_MAPPING(sbi) == page->mapping && 315 IS_DNODE(page) && is_cold_node(page); 316 } 317 318 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi) 319 { 320 spin_lock_init(&sbi->fsync_node_lock); 321 INIT_LIST_HEAD(&sbi->fsync_node_list); 322 sbi->fsync_seg_id = 0; 323 sbi->fsync_node_num = 0; 324 } 325 326 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi, 327 struct page *page) 328 { 329 struct fsync_node_entry *fn; 330 unsigned long flags; 331 unsigned int seq_id; 332 333 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, 334 GFP_NOFS, true, NULL); 335 336 get_page(page); 337 fn->page = page; 338 INIT_LIST_HEAD(&fn->list); 339 340 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 341 list_add_tail(&fn->list, &sbi->fsync_node_list); 342 fn->seq_id = sbi->fsync_seg_id++; 343 seq_id = fn->seq_id; 344 sbi->fsync_node_num++; 345 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 346 347 return seq_id; 348 } 349 350 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page) 351 { 352 struct fsync_node_entry *fn; 353 unsigned long flags; 354 355 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 356 list_for_each_entry(fn, &sbi->fsync_node_list, list) { 357 if (fn->page == page) { 358 list_del(&fn->list); 359 sbi->fsync_node_num--; 360 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 361 kmem_cache_free(fsync_node_entry_slab, fn); 362 put_page(page); 363 return; 364 } 365 } 366 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 367 f2fs_bug_on(sbi, 1); 368 } 369 370 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi) 371 { 372 unsigned long flags; 373 374 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 375 sbi->fsync_seg_id = 0; 376 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 377 } 378 379 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid) 380 { 381 struct f2fs_nm_info *nm_i = NM_I(sbi); 382 struct nat_entry *e; 383 bool need = false; 384 385 down_read(&nm_i->nat_tree_lock); 386 e = __lookup_nat_cache(nm_i, nid); 387 if (e) { 388 if (!get_nat_flag(e, IS_CHECKPOINTED) && 389 !get_nat_flag(e, HAS_FSYNCED_INODE)) 390 need = true; 391 } 392 up_read(&nm_i->nat_tree_lock); 393 return need; 394 } 395 396 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid) 397 { 398 struct f2fs_nm_info *nm_i = NM_I(sbi); 399 struct nat_entry *e; 400 bool is_cp = true; 401 402 down_read(&nm_i->nat_tree_lock); 403 e = __lookup_nat_cache(nm_i, nid); 404 if (e && !get_nat_flag(e, IS_CHECKPOINTED)) 405 is_cp = false; 406 up_read(&nm_i->nat_tree_lock); 407 return is_cp; 408 } 409 410 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino) 411 { 412 struct f2fs_nm_info *nm_i = NM_I(sbi); 413 struct nat_entry *e; 414 bool need_update = true; 415 416 down_read(&nm_i->nat_tree_lock); 417 e = __lookup_nat_cache(nm_i, ino); 418 if (e && get_nat_flag(e, HAS_LAST_FSYNC) && 419 (get_nat_flag(e, IS_CHECKPOINTED) || 420 get_nat_flag(e, HAS_FSYNCED_INODE))) 421 need_update = false; 422 up_read(&nm_i->nat_tree_lock); 423 return need_update; 424 } 425 426 /* must be locked by nat_tree_lock */ 427 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid, 428 struct f2fs_nat_entry *ne) 429 { 430 struct f2fs_nm_info *nm_i = NM_I(sbi); 431 struct nat_entry *new, *e; 432 433 new = __alloc_nat_entry(sbi, nid, false); 434 if (!new) 435 return; 436 437 down_write(&nm_i->nat_tree_lock); 438 e = __lookup_nat_cache(nm_i, nid); 439 if (!e) 440 e = __init_nat_entry(nm_i, new, ne, false); 441 else 442 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) || 443 nat_get_blkaddr(e) != 444 le32_to_cpu(ne->block_addr) || 445 nat_get_version(e) != ne->version); 446 up_write(&nm_i->nat_tree_lock); 447 if (e != new) 448 __free_nat_entry(new); 449 } 450 451 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni, 452 block_t new_blkaddr, bool fsync_done) 453 { 454 struct f2fs_nm_info *nm_i = NM_I(sbi); 455 struct nat_entry *e; 456 struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true); 457 458 down_write(&nm_i->nat_tree_lock); 459 e = __lookup_nat_cache(nm_i, ni->nid); 460 if (!e) { 461 e = __init_nat_entry(nm_i, new, NULL, true); 462 copy_node_info(&e->ni, ni); 463 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR); 464 } else if (new_blkaddr == NEW_ADDR) { 465 /* 466 * when nid is reallocated, 467 * previous nat entry can be remained in nat cache. 468 * So, reinitialize it with new information. 469 */ 470 copy_node_info(&e->ni, ni); 471 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR); 472 } 473 /* let's free early to reduce memory consumption */ 474 if (e != new) 475 __free_nat_entry(new); 476 477 /* sanity check */ 478 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr); 479 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR && 480 new_blkaddr == NULL_ADDR); 481 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR && 482 new_blkaddr == NEW_ADDR); 483 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) && 484 new_blkaddr == NEW_ADDR); 485 486 /* increment version no as node is removed */ 487 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) { 488 unsigned char version = nat_get_version(e); 489 490 nat_set_version(e, inc_node_version(version)); 491 } 492 493 /* change address */ 494 nat_set_blkaddr(e, new_blkaddr); 495 if (!__is_valid_data_blkaddr(new_blkaddr)) 496 set_nat_flag(e, IS_CHECKPOINTED, false); 497 __set_nat_cache_dirty(nm_i, e); 498 499 /* update fsync_mark if its inode nat entry is still alive */ 500 if (ni->nid != ni->ino) 501 e = __lookup_nat_cache(nm_i, ni->ino); 502 if (e) { 503 if (fsync_done && ni->nid == ni->ino) 504 set_nat_flag(e, HAS_FSYNCED_INODE, true); 505 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done); 506 } 507 up_write(&nm_i->nat_tree_lock); 508 } 509 510 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink) 511 { 512 struct f2fs_nm_info *nm_i = NM_I(sbi); 513 int nr = nr_shrink; 514 515 if (!down_write_trylock(&nm_i->nat_tree_lock)) 516 return 0; 517 518 spin_lock(&nm_i->nat_list_lock); 519 while (nr_shrink) { 520 struct nat_entry *ne; 521 522 if (list_empty(&nm_i->nat_entries)) 523 break; 524 525 ne = list_first_entry(&nm_i->nat_entries, 526 struct nat_entry, list); 527 list_del(&ne->list); 528 spin_unlock(&nm_i->nat_list_lock); 529 530 __del_from_nat_cache(nm_i, ne); 531 nr_shrink--; 532 533 spin_lock(&nm_i->nat_list_lock); 534 } 535 spin_unlock(&nm_i->nat_list_lock); 536 537 up_write(&nm_i->nat_tree_lock); 538 return nr - nr_shrink; 539 } 540 541 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid, 542 struct node_info *ni) 543 { 544 struct f2fs_nm_info *nm_i = NM_I(sbi); 545 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 546 struct f2fs_journal *journal = curseg->journal; 547 nid_t start_nid = START_NID(nid); 548 struct f2fs_nat_block *nat_blk; 549 struct page *page = NULL; 550 struct f2fs_nat_entry ne; 551 struct nat_entry *e; 552 pgoff_t index; 553 block_t blkaddr; 554 int i; 555 556 ni->nid = nid; 557 retry: 558 /* Check nat cache */ 559 down_read(&nm_i->nat_tree_lock); 560 e = __lookup_nat_cache(nm_i, nid); 561 if (e) { 562 ni->ino = nat_get_ino(e); 563 ni->blk_addr = nat_get_blkaddr(e); 564 ni->version = nat_get_version(e); 565 up_read(&nm_i->nat_tree_lock); 566 return 0; 567 } 568 569 /* 570 * Check current segment summary by trying to grab journal_rwsem first. 571 * This sem is on the critical path on the checkpoint requiring the above 572 * nat_tree_lock. Therefore, we should retry, if we failed to grab here 573 * while not bothering checkpoint. 574 */ 575 if (!rwsem_is_locked(&sbi->cp_global_sem)) { 576 down_read(&curseg->journal_rwsem); 577 } else if (!down_read_trylock(&curseg->journal_rwsem)) { 578 up_read(&nm_i->nat_tree_lock); 579 goto retry; 580 } 581 582 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0); 583 if (i >= 0) { 584 ne = nat_in_journal(journal, i); 585 node_info_from_raw_nat(ni, &ne); 586 } 587 up_read(&curseg->journal_rwsem); 588 if (i >= 0) { 589 up_read(&nm_i->nat_tree_lock); 590 goto cache; 591 } 592 593 /* Fill node_info from nat page */ 594 index = current_nat_addr(sbi, nid); 595 up_read(&nm_i->nat_tree_lock); 596 597 page = f2fs_get_meta_page(sbi, index); 598 if (IS_ERR(page)) 599 return PTR_ERR(page); 600 601 nat_blk = (struct f2fs_nat_block *)page_address(page); 602 ne = nat_blk->entries[nid - start_nid]; 603 node_info_from_raw_nat(ni, &ne); 604 f2fs_put_page(page, 1); 605 cache: 606 blkaddr = le32_to_cpu(ne.block_addr); 607 if (__is_valid_data_blkaddr(blkaddr) && 608 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) 609 return -EFAULT; 610 611 /* cache nat entry */ 612 cache_nat_entry(sbi, nid, &ne); 613 return 0; 614 } 615 616 /* 617 * readahead MAX_RA_NODE number of node pages. 618 */ 619 static void f2fs_ra_node_pages(struct page *parent, int start, int n) 620 { 621 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 622 struct blk_plug plug; 623 int i, end; 624 nid_t nid; 625 626 blk_start_plug(&plug); 627 628 /* Then, try readahead for siblings of the desired node */ 629 end = start + n; 630 end = min(end, NIDS_PER_BLOCK); 631 for (i = start; i < end; i++) { 632 nid = get_nid(parent, i, false); 633 f2fs_ra_node_page(sbi, nid); 634 } 635 636 blk_finish_plug(&plug); 637 } 638 639 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs) 640 { 641 const long direct_index = ADDRS_PER_INODE(dn->inode); 642 const long direct_blks = ADDRS_PER_BLOCK(dn->inode); 643 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK; 644 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode); 645 int cur_level = dn->cur_level; 646 int max_level = dn->max_level; 647 pgoff_t base = 0; 648 649 if (!dn->max_level) 650 return pgofs + 1; 651 652 while (max_level-- > cur_level) 653 skipped_unit *= NIDS_PER_BLOCK; 654 655 switch (dn->max_level) { 656 case 3: 657 base += 2 * indirect_blks; 658 fallthrough; 659 case 2: 660 base += 2 * direct_blks; 661 fallthrough; 662 case 1: 663 base += direct_index; 664 break; 665 default: 666 f2fs_bug_on(F2FS_I_SB(dn->inode), 1); 667 } 668 669 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base; 670 } 671 672 /* 673 * The maximum depth is four. 674 * Offset[0] will have raw inode offset. 675 */ 676 static int get_node_path(struct inode *inode, long block, 677 int offset[4], unsigned int noffset[4]) 678 { 679 const long direct_index = ADDRS_PER_INODE(inode); 680 const long direct_blks = ADDRS_PER_BLOCK(inode); 681 const long dptrs_per_blk = NIDS_PER_BLOCK; 682 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK; 683 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK; 684 int n = 0; 685 int level = 0; 686 687 noffset[0] = 0; 688 689 if (block < direct_index) { 690 offset[n] = block; 691 goto got; 692 } 693 block -= direct_index; 694 if (block < direct_blks) { 695 offset[n++] = NODE_DIR1_BLOCK; 696 noffset[n] = 1; 697 offset[n] = block; 698 level = 1; 699 goto got; 700 } 701 block -= direct_blks; 702 if (block < direct_blks) { 703 offset[n++] = NODE_DIR2_BLOCK; 704 noffset[n] = 2; 705 offset[n] = block; 706 level = 1; 707 goto got; 708 } 709 block -= direct_blks; 710 if (block < indirect_blks) { 711 offset[n++] = NODE_IND1_BLOCK; 712 noffset[n] = 3; 713 offset[n++] = block / direct_blks; 714 noffset[n] = 4 + offset[n - 1]; 715 offset[n] = block % direct_blks; 716 level = 2; 717 goto got; 718 } 719 block -= indirect_blks; 720 if (block < indirect_blks) { 721 offset[n++] = NODE_IND2_BLOCK; 722 noffset[n] = 4 + dptrs_per_blk; 723 offset[n++] = block / direct_blks; 724 noffset[n] = 5 + dptrs_per_blk + offset[n - 1]; 725 offset[n] = block % direct_blks; 726 level = 2; 727 goto got; 728 } 729 block -= indirect_blks; 730 if (block < dindirect_blks) { 731 offset[n++] = NODE_DIND_BLOCK; 732 noffset[n] = 5 + (dptrs_per_blk * 2); 733 offset[n++] = block / indirect_blks; 734 noffset[n] = 6 + (dptrs_per_blk * 2) + 735 offset[n - 1] * (dptrs_per_blk + 1); 736 offset[n++] = (block / direct_blks) % dptrs_per_blk; 737 noffset[n] = 7 + (dptrs_per_blk * 2) + 738 offset[n - 2] * (dptrs_per_blk + 1) + 739 offset[n - 1]; 740 offset[n] = block % direct_blks; 741 level = 3; 742 goto got; 743 } else { 744 return -E2BIG; 745 } 746 got: 747 return level; 748 } 749 750 /* 751 * Caller should call f2fs_put_dnode(dn). 752 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and 753 * f2fs_unlock_op() only if mode is set with ALLOC_NODE. 754 */ 755 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode) 756 { 757 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 758 struct page *npage[4]; 759 struct page *parent = NULL; 760 int offset[4]; 761 unsigned int noffset[4]; 762 nid_t nids[4]; 763 int level, i = 0; 764 int err = 0; 765 766 level = get_node_path(dn->inode, index, offset, noffset); 767 if (level < 0) 768 return level; 769 770 nids[0] = dn->inode->i_ino; 771 npage[0] = dn->inode_page; 772 773 if (!npage[0]) { 774 npage[0] = f2fs_get_node_page(sbi, nids[0]); 775 if (IS_ERR(npage[0])) 776 return PTR_ERR(npage[0]); 777 } 778 779 /* if inline_data is set, should not report any block indices */ 780 if (f2fs_has_inline_data(dn->inode) && index) { 781 err = -ENOENT; 782 f2fs_put_page(npage[0], 1); 783 goto release_out; 784 } 785 786 parent = npage[0]; 787 if (level != 0) 788 nids[1] = get_nid(parent, offset[0], true); 789 dn->inode_page = npage[0]; 790 dn->inode_page_locked = true; 791 792 /* get indirect or direct nodes */ 793 for (i = 1; i <= level; i++) { 794 bool done = false; 795 796 if (!nids[i] && mode == ALLOC_NODE) { 797 /* alloc new node */ 798 if (!f2fs_alloc_nid(sbi, &(nids[i]))) { 799 err = -ENOSPC; 800 goto release_pages; 801 } 802 803 dn->nid = nids[i]; 804 npage[i] = f2fs_new_node_page(dn, noffset[i]); 805 if (IS_ERR(npage[i])) { 806 f2fs_alloc_nid_failed(sbi, nids[i]); 807 err = PTR_ERR(npage[i]); 808 goto release_pages; 809 } 810 811 set_nid(parent, offset[i - 1], nids[i], i == 1); 812 f2fs_alloc_nid_done(sbi, nids[i]); 813 done = true; 814 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) { 815 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]); 816 if (IS_ERR(npage[i])) { 817 err = PTR_ERR(npage[i]); 818 goto release_pages; 819 } 820 done = true; 821 } 822 if (i == 1) { 823 dn->inode_page_locked = false; 824 unlock_page(parent); 825 } else { 826 f2fs_put_page(parent, 1); 827 } 828 829 if (!done) { 830 npage[i] = f2fs_get_node_page(sbi, nids[i]); 831 if (IS_ERR(npage[i])) { 832 err = PTR_ERR(npage[i]); 833 f2fs_put_page(npage[0], 0); 834 goto release_out; 835 } 836 } 837 if (i < level) { 838 parent = npage[i]; 839 nids[i + 1] = get_nid(parent, offset[i], false); 840 } 841 } 842 dn->nid = nids[level]; 843 dn->ofs_in_node = offset[level]; 844 dn->node_page = npage[level]; 845 dn->data_blkaddr = f2fs_data_blkaddr(dn); 846 847 if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) && 848 f2fs_sb_has_readonly(sbi)) { 849 unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn); 850 block_t blkaddr; 851 852 if (!c_len) 853 goto out; 854 855 blkaddr = f2fs_data_blkaddr(dn); 856 if (blkaddr == COMPRESS_ADDR) 857 blkaddr = data_blkaddr(dn->inode, dn->node_page, 858 dn->ofs_in_node + 1); 859 860 f2fs_update_extent_tree_range_compressed(dn->inode, 861 index, blkaddr, 862 F2FS_I(dn->inode)->i_cluster_size, 863 c_len); 864 } 865 out: 866 return 0; 867 868 release_pages: 869 f2fs_put_page(parent, 1); 870 if (i > 1) 871 f2fs_put_page(npage[0], 0); 872 release_out: 873 dn->inode_page = NULL; 874 dn->node_page = NULL; 875 if (err == -ENOENT) { 876 dn->cur_level = i; 877 dn->max_level = level; 878 dn->ofs_in_node = offset[level]; 879 } 880 return err; 881 } 882 883 static int truncate_node(struct dnode_of_data *dn) 884 { 885 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 886 struct node_info ni; 887 int err; 888 pgoff_t index; 889 890 err = f2fs_get_node_info(sbi, dn->nid, &ni); 891 if (err) 892 return err; 893 894 /* Deallocate node address */ 895 f2fs_invalidate_blocks(sbi, ni.blk_addr); 896 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino); 897 set_node_addr(sbi, &ni, NULL_ADDR, false); 898 899 if (dn->nid == dn->inode->i_ino) { 900 f2fs_remove_orphan_inode(sbi, dn->nid); 901 dec_valid_inode_count(sbi); 902 f2fs_inode_synced(dn->inode); 903 } 904 905 clear_node_page_dirty(dn->node_page); 906 set_sbi_flag(sbi, SBI_IS_DIRTY); 907 908 index = dn->node_page->index; 909 f2fs_put_page(dn->node_page, 1); 910 911 invalidate_mapping_pages(NODE_MAPPING(sbi), 912 index, index); 913 914 dn->node_page = NULL; 915 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr); 916 917 return 0; 918 } 919 920 static int truncate_dnode(struct dnode_of_data *dn) 921 { 922 struct page *page; 923 int err; 924 925 if (dn->nid == 0) 926 return 1; 927 928 /* get direct node */ 929 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 930 if (PTR_ERR(page) == -ENOENT) 931 return 1; 932 else if (IS_ERR(page)) 933 return PTR_ERR(page); 934 935 /* Make dnode_of_data for parameter */ 936 dn->node_page = page; 937 dn->ofs_in_node = 0; 938 f2fs_truncate_data_blocks(dn); 939 err = truncate_node(dn); 940 if (err) 941 return err; 942 943 return 1; 944 } 945 946 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs, 947 int ofs, int depth) 948 { 949 struct dnode_of_data rdn = *dn; 950 struct page *page; 951 struct f2fs_node *rn; 952 nid_t child_nid; 953 unsigned int child_nofs; 954 int freed = 0; 955 int i, ret; 956 957 if (dn->nid == 0) 958 return NIDS_PER_BLOCK + 1; 959 960 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr); 961 962 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 963 if (IS_ERR(page)) { 964 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page)); 965 return PTR_ERR(page); 966 } 967 968 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK); 969 970 rn = F2FS_NODE(page); 971 if (depth < 3) { 972 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) { 973 child_nid = le32_to_cpu(rn->in.nid[i]); 974 if (child_nid == 0) 975 continue; 976 rdn.nid = child_nid; 977 ret = truncate_dnode(&rdn); 978 if (ret < 0) 979 goto out_err; 980 if (set_nid(page, i, 0, false)) 981 dn->node_changed = true; 982 } 983 } else { 984 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1; 985 for (i = ofs; i < NIDS_PER_BLOCK; i++) { 986 child_nid = le32_to_cpu(rn->in.nid[i]); 987 if (child_nid == 0) { 988 child_nofs += NIDS_PER_BLOCK + 1; 989 continue; 990 } 991 rdn.nid = child_nid; 992 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1); 993 if (ret == (NIDS_PER_BLOCK + 1)) { 994 if (set_nid(page, i, 0, false)) 995 dn->node_changed = true; 996 child_nofs += ret; 997 } else if (ret < 0 && ret != -ENOENT) { 998 goto out_err; 999 } 1000 } 1001 freed = child_nofs; 1002 } 1003 1004 if (!ofs) { 1005 /* remove current indirect node */ 1006 dn->node_page = page; 1007 ret = truncate_node(dn); 1008 if (ret) 1009 goto out_err; 1010 freed++; 1011 } else { 1012 f2fs_put_page(page, 1); 1013 } 1014 trace_f2fs_truncate_nodes_exit(dn->inode, freed); 1015 return freed; 1016 1017 out_err: 1018 f2fs_put_page(page, 1); 1019 trace_f2fs_truncate_nodes_exit(dn->inode, ret); 1020 return ret; 1021 } 1022 1023 static int truncate_partial_nodes(struct dnode_of_data *dn, 1024 struct f2fs_inode *ri, int *offset, int depth) 1025 { 1026 struct page *pages[2]; 1027 nid_t nid[3]; 1028 nid_t child_nid; 1029 int err = 0; 1030 int i; 1031 int idx = depth - 2; 1032 1033 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1034 if (!nid[0]) 1035 return 0; 1036 1037 /* get indirect nodes in the path */ 1038 for (i = 0; i < idx + 1; i++) { 1039 /* reference count'll be increased */ 1040 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]); 1041 if (IS_ERR(pages[i])) { 1042 err = PTR_ERR(pages[i]); 1043 idx = i - 1; 1044 goto fail; 1045 } 1046 nid[i + 1] = get_nid(pages[i], offset[i + 1], false); 1047 } 1048 1049 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK); 1050 1051 /* free direct nodes linked to a partial indirect node */ 1052 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) { 1053 child_nid = get_nid(pages[idx], i, false); 1054 if (!child_nid) 1055 continue; 1056 dn->nid = child_nid; 1057 err = truncate_dnode(dn); 1058 if (err < 0) 1059 goto fail; 1060 if (set_nid(pages[idx], i, 0, false)) 1061 dn->node_changed = true; 1062 } 1063 1064 if (offset[idx + 1] == 0) { 1065 dn->node_page = pages[idx]; 1066 dn->nid = nid[idx]; 1067 err = truncate_node(dn); 1068 if (err) 1069 goto fail; 1070 } else { 1071 f2fs_put_page(pages[idx], 1); 1072 } 1073 offset[idx]++; 1074 offset[idx + 1] = 0; 1075 idx--; 1076 fail: 1077 for (i = idx; i >= 0; i--) 1078 f2fs_put_page(pages[i], 1); 1079 1080 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err); 1081 1082 return err; 1083 } 1084 1085 /* 1086 * All the block addresses of data and nodes should be nullified. 1087 */ 1088 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from) 1089 { 1090 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1091 int err = 0, cont = 1; 1092 int level, offset[4], noffset[4]; 1093 unsigned int nofs = 0; 1094 struct f2fs_inode *ri; 1095 struct dnode_of_data dn; 1096 struct page *page; 1097 1098 trace_f2fs_truncate_inode_blocks_enter(inode, from); 1099 1100 level = get_node_path(inode, from, offset, noffset); 1101 if (level < 0) { 1102 trace_f2fs_truncate_inode_blocks_exit(inode, level); 1103 return level; 1104 } 1105 1106 page = f2fs_get_node_page(sbi, inode->i_ino); 1107 if (IS_ERR(page)) { 1108 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page)); 1109 return PTR_ERR(page); 1110 } 1111 1112 set_new_dnode(&dn, inode, page, NULL, 0); 1113 unlock_page(page); 1114 1115 ri = F2FS_INODE(page); 1116 switch (level) { 1117 case 0: 1118 case 1: 1119 nofs = noffset[1]; 1120 break; 1121 case 2: 1122 nofs = noffset[1]; 1123 if (!offset[level - 1]) 1124 goto skip_partial; 1125 err = truncate_partial_nodes(&dn, ri, offset, level); 1126 if (err < 0 && err != -ENOENT) 1127 goto fail; 1128 nofs += 1 + NIDS_PER_BLOCK; 1129 break; 1130 case 3: 1131 nofs = 5 + 2 * NIDS_PER_BLOCK; 1132 if (!offset[level - 1]) 1133 goto skip_partial; 1134 err = truncate_partial_nodes(&dn, ri, offset, level); 1135 if (err < 0 && err != -ENOENT) 1136 goto fail; 1137 break; 1138 default: 1139 BUG(); 1140 } 1141 1142 skip_partial: 1143 while (cont) { 1144 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1145 switch (offset[0]) { 1146 case NODE_DIR1_BLOCK: 1147 case NODE_DIR2_BLOCK: 1148 err = truncate_dnode(&dn); 1149 break; 1150 1151 case NODE_IND1_BLOCK: 1152 case NODE_IND2_BLOCK: 1153 err = truncate_nodes(&dn, nofs, offset[1], 2); 1154 break; 1155 1156 case NODE_DIND_BLOCK: 1157 err = truncate_nodes(&dn, nofs, offset[1], 3); 1158 cont = 0; 1159 break; 1160 1161 default: 1162 BUG(); 1163 } 1164 if (err < 0 && err != -ENOENT) 1165 goto fail; 1166 if (offset[1] == 0 && 1167 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) { 1168 lock_page(page); 1169 BUG_ON(page->mapping != NODE_MAPPING(sbi)); 1170 f2fs_wait_on_page_writeback(page, NODE, true, true); 1171 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0; 1172 set_page_dirty(page); 1173 unlock_page(page); 1174 } 1175 offset[1] = 0; 1176 offset[0]++; 1177 nofs += err; 1178 } 1179 fail: 1180 f2fs_put_page(page, 0); 1181 trace_f2fs_truncate_inode_blocks_exit(inode, err); 1182 return err > 0 ? 0 : err; 1183 } 1184 1185 /* caller must lock inode page */ 1186 int f2fs_truncate_xattr_node(struct inode *inode) 1187 { 1188 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1189 nid_t nid = F2FS_I(inode)->i_xattr_nid; 1190 struct dnode_of_data dn; 1191 struct page *npage; 1192 int err; 1193 1194 if (!nid) 1195 return 0; 1196 1197 npage = f2fs_get_node_page(sbi, nid); 1198 if (IS_ERR(npage)) 1199 return PTR_ERR(npage); 1200 1201 set_new_dnode(&dn, inode, NULL, npage, nid); 1202 err = truncate_node(&dn); 1203 if (err) { 1204 f2fs_put_page(npage, 1); 1205 return err; 1206 } 1207 1208 f2fs_i_xnid_write(inode, 0); 1209 1210 return 0; 1211 } 1212 1213 /* 1214 * Caller should grab and release a rwsem by calling f2fs_lock_op() and 1215 * f2fs_unlock_op(). 1216 */ 1217 int f2fs_remove_inode_page(struct inode *inode) 1218 { 1219 struct dnode_of_data dn; 1220 int err; 1221 1222 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1223 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE); 1224 if (err) 1225 return err; 1226 1227 err = f2fs_truncate_xattr_node(inode); 1228 if (err) { 1229 f2fs_put_dnode(&dn); 1230 return err; 1231 } 1232 1233 /* remove potential inline_data blocks */ 1234 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1235 S_ISLNK(inode->i_mode)) 1236 f2fs_truncate_data_blocks_range(&dn, 1); 1237 1238 /* 0 is possible, after f2fs_new_inode() has failed */ 1239 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) { 1240 f2fs_put_dnode(&dn); 1241 return -EIO; 1242 } 1243 1244 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) { 1245 f2fs_warn(F2FS_I_SB(inode), 1246 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu", 1247 inode->i_ino, (unsigned long long)inode->i_blocks); 1248 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK); 1249 } 1250 1251 /* will put inode & node pages */ 1252 err = truncate_node(&dn); 1253 if (err) { 1254 f2fs_put_dnode(&dn); 1255 return err; 1256 } 1257 return 0; 1258 } 1259 1260 struct page *f2fs_new_inode_page(struct inode *inode) 1261 { 1262 struct dnode_of_data dn; 1263 1264 /* allocate inode page for new inode */ 1265 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1266 1267 /* caller should f2fs_put_page(page, 1); */ 1268 return f2fs_new_node_page(&dn, 0); 1269 } 1270 1271 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs) 1272 { 1273 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1274 struct node_info new_ni; 1275 struct page *page; 1276 int err; 1277 1278 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1279 return ERR_PTR(-EPERM); 1280 1281 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false); 1282 if (!page) 1283 return ERR_PTR(-ENOMEM); 1284 1285 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs)))) 1286 goto fail; 1287 1288 #ifdef CONFIG_F2FS_CHECK_FS 1289 err = f2fs_get_node_info(sbi, dn->nid, &new_ni); 1290 if (err) { 1291 dec_valid_node_count(sbi, dn->inode, !ofs); 1292 goto fail; 1293 } 1294 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR); 1295 #endif 1296 new_ni.nid = dn->nid; 1297 new_ni.ino = dn->inode->i_ino; 1298 new_ni.blk_addr = NULL_ADDR; 1299 new_ni.flag = 0; 1300 new_ni.version = 0; 1301 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 1302 1303 f2fs_wait_on_page_writeback(page, NODE, true, true); 1304 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 1305 set_cold_node(page, S_ISDIR(dn->inode->i_mode)); 1306 if (!PageUptodate(page)) 1307 SetPageUptodate(page); 1308 if (set_page_dirty(page)) 1309 dn->node_changed = true; 1310 1311 if (f2fs_has_xattr_block(ofs)) 1312 f2fs_i_xnid_write(dn->inode, dn->nid); 1313 1314 if (ofs == 0) 1315 inc_valid_inode_count(sbi); 1316 return page; 1317 1318 fail: 1319 clear_node_page_dirty(page); 1320 f2fs_put_page(page, 1); 1321 return ERR_PTR(err); 1322 } 1323 1324 /* 1325 * Caller should do after getting the following values. 1326 * 0: f2fs_put_page(page, 0) 1327 * LOCKED_PAGE or error: f2fs_put_page(page, 1) 1328 */ 1329 static int read_node_page(struct page *page, int op_flags) 1330 { 1331 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1332 struct node_info ni; 1333 struct f2fs_io_info fio = { 1334 .sbi = sbi, 1335 .type = NODE, 1336 .op = REQ_OP_READ, 1337 .op_flags = op_flags, 1338 .page = page, 1339 .encrypted_page = NULL, 1340 }; 1341 int err; 1342 1343 if (PageUptodate(page)) { 1344 if (!f2fs_inode_chksum_verify(sbi, page)) { 1345 ClearPageUptodate(page); 1346 return -EFSBADCRC; 1347 } 1348 return LOCKED_PAGE; 1349 } 1350 1351 err = f2fs_get_node_info(sbi, page->index, &ni); 1352 if (err) 1353 return err; 1354 1355 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */ 1356 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR) || 1357 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) { 1358 ClearPageUptodate(page); 1359 return -ENOENT; 1360 } 1361 1362 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr; 1363 1364 err = f2fs_submit_page_bio(&fio); 1365 1366 if (!err) 1367 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE); 1368 1369 return err; 1370 } 1371 1372 /* 1373 * Readahead a node page 1374 */ 1375 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 1376 { 1377 struct page *apage; 1378 int err; 1379 1380 if (!nid) 1381 return; 1382 if (f2fs_check_nid_range(sbi, nid)) 1383 return; 1384 1385 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid); 1386 if (apage) 1387 return; 1388 1389 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1390 if (!apage) 1391 return; 1392 1393 err = read_node_page(apage, REQ_RAHEAD); 1394 f2fs_put_page(apage, err ? 1 : 0); 1395 } 1396 1397 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid, 1398 struct page *parent, int start) 1399 { 1400 struct page *page; 1401 int err; 1402 1403 if (!nid) 1404 return ERR_PTR(-ENOENT); 1405 if (f2fs_check_nid_range(sbi, nid)) 1406 return ERR_PTR(-EINVAL); 1407 repeat: 1408 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1409 if (!page) 1410 return ERR_PTR(-ENOMEM); 1411 1412 err = read_node_page(page, 0); 1413 if (err < 0) { 1414 f2fs_put_page(page, 1); 1415 return ERR_PTR(err); 1416 } else if (err == LOCKED_PAGE) { 1417 err = 0; 1418 goto page_hit; 1419 } 1420 1421 if (parent) 1422 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE); 1423 1424 lock_page(page); 1425 1426 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1427 f2fs_put_page(page, 1); 1428 goto repeat; 1429 } 1430 1431 if (unlikely(!PageUptodate(page))) { 1432 err = -EIO; 1433 goto out_err; 1434 } 1435 1436 if (!f2fs_inode_chksum_verify(sbi, page)) { 1437 err = -EFSBADCRC; 1438 goto out_err; 1439 } 1440 page_hit: 1441 if (unlikely(nid != nid_of_node(page))) { 1442 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]", 1443 nid, nid_of_node(page), ino_of_node(page), 1444 ofs_of_node(page), cpver_of_node(page), 1445 next_blkaddr_of_node(page)); 1446 err = -EINVAL; 1447 out_err: 1448 ClearPageUptodate(page); 1449 f2fs_put_page(page, 1); 1450 return ERR_PTR(err); 1451 } 1452 return page; 1453 } 1454 1455 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 1456 { 1457 return __get_node_page(sbi, nid, NULL, 0); 1458 } 1459 1460 struct page *f2fs_get_node_page_ra(struct page *parent, int start) 1461 { 1462 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 1463 nid_t nid = get_nid(parent, start, false); 1464 1465 return __get_node_page(sbi, nid, parent, start); 1466 } 1467 1468 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino) 1469 { 1470 struct inode *inode; 1471 struct page *page; 1472 int ret; 1473 1474 /* should flush inline_data before evict_inode */ 1475 inode = ilookup(sbi->sb, ino); 1476 if (!inode) 1477 return; 1478 1479 page = f2fs_pagecache_get_page(inode->i_mapping, 0, 1480 FGP_LOCK|FGP_NOWAIT, 0); 1481 if (!page) 1482 goto iput_out; 1483 1484 if (!PageUptodate(page)) 1485 goto page_out; 1486 1487 if (!PageDirty(page)) 1488 goto page_out; 1489 1490 if (!clear_page_dirty_for_io(page)) 1491 goto page_out; 1492 1493 ret = f2fs_write_inline_data(inode, page); 1494 inode_dec_dirty_pages(inode); 1495 f2fs_remove_dirty_inode(inode); 1496 if (ret) 1497 set_page_dirty(page); 1498 page_out: 1499 f2fs_put_page(page, 1); 1500 iput_out: 1501 iput(inode); 1502 } 1503 1504 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino) 1505 { 1506 pgoff_t index; 1507 struct pagevec pvec; 1508 struct page *last_page = NULL; 1509 int nr_pages; 1510 1511 pagevec_init(&pvec); 1512 index = 0; 1513 1514 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1515 PAGECACHE_TAG_DIRTY))) { 1516 int i; 1517 1518 for (i = 0; i < nr_pages; i++) { 1519 struct page *page = pvec.pages[i]; 1520 1521 if (unlikely(f2fs_cp_error(sbi))) { 1522 f2fs_put_page(last_page, 0); 1523 pagevec_release(&pvec); 1524 return ERR_PTR(-EIO); 1525 } 1526 1527 if (!IS_DNODE(page) || !is_cold_node(page)) 1528 continue; 1529 if (ino_of_node(page) != ino) 1530 continue; 1531 1532 lock_page(page); 1533 1534 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1535 continue_unlock: 1536 unlock_page(page); 1537 continue; 1538 } 1539 if (ino_of_node(page) != ino) 1540 goto continue_unlock; 1541 1542 if (!PageDirty(page)) { 1543 /* someone wrote it for us */ 1544 goto continue_unlock; 1545 } 1546 1547 if (last_page) 1548 f2fs_put_page(last_page, 0); 1549 1550 get_page(page); 1551 last_page = page; 1552 unlock_page(page); 1553 } 1554 pagevec_release(&pvec); 1555 cond_resched(); 1556 } 1557 return last_page; 1558 } 1559 1560 static int __write_node_page(struct page *page, bool atomic, bool *submitted, 1561 struct writeback_control *wbc, bool do_balance, 1562 enum iostat_type io_type, unsigned int *seq_id) 1563 { 1564 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1565 nid_t nid; 1566 struct node_info ni; 1567 struct f2fs_io_info fio = { 1568 .sbi = sbi, 1569 .ino = ino_of_node(page), 1570 .type = NODE, 1571 .op = REQ_OP_WRITE, 1572 .op_flags = wbc_to_write_flags(wbc), 1573 .page = page, 1574 .encrypted_page = NULL, 1575 .submitted = false, 1576 .io_type = io_type, 1577 .io_wbc = wbc, 1578 }; 1579 unsigned int seq; 1580 1581 trace_f2fs_writepage(page, NODE); 1582 1583 if (unlikely(f2fs_cp_error(sbi))) { 1584 ClearPageUptodate(page); 1585 dec_page_count(sbi, F2FS_DIRTY_NODES); 1586 unlock_page(page); 1587 return 0; 1588 } 1589 1590 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1591 goto redirty_out; 1592 1593 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 1594 wbc->sync_mode == WB_SYNC_NONE && 1595 IS_DNODE(page) && is_cold_node(page)) 1596 goto redirty_out; 1597 1598 /* get old block addr of this node page */ 1599 nid = nid_of_node(page); 1600 f2fs_bug_on(sbi, page->index != nid); 1601 1602 if (f2fs_get_node_info(sbi, nid, &ni)) 1603 goto redirty_out; 1604 1605 if (wbc->for_reclaim) { 1606 if (!down_read_trylock(&sbi->node_write)) 1607 goto redirty_out; 1608 } else { 1609 down_read(&sbi->node_write); 1610 } 1611 1612 /* This page is already truncated */ 1613 if (unlikely(ni.blk_addr == NULL_ADDR)) { 1614 ClearPageUptodate(page); 1615 dec_page_count(sbi, F2FS_DIRTY_NODES); 1616 up_read(&sbi->node_write); 1617 unlock_page(page); 1618 return 0; 1619 } 1620 1621 if (__is_valid_data_blkaddr(ni.blk_addr) && 1622 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, 1623 DATA_GENERIC_ENHANCE)) { 1624 up_read(&sbi->node_write); 1625 goto redirty_out; 1626 } 1627 1628 if (atomic && !test_opt(sbi, NOBARRIER)) 1629 fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 1630 1631 /* should add to global list before clearing PAGECACHE status */ 1632 if (f2fs_in_warm_node_list(sbi, page)) { 1633 seq = f2fs_add_fsync_node_entry(sbi, page); 1634 if (seq_id) 1635 *seq_id = seq; 1636 } 1637 1638 set_page_writeback(page); 1639 ClearPageError(page); 1640 1641 fio.old_blkaddr = ni.blk_addr; 1642 f2fs_do_write_node_page(nid, &fio); 1643 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page)); 1644 dec_page_count(sbi, F2FS_DIRTY_NODES); 1645 up_read(&sbi->node_write); 1646 1647 if (wbc->for_reclaim) { 1648 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE); 1649 submitted = NULL; 1650 } 1651 1652 unlock_page(page); 1653 1654 if (unlikely(f2fs_cp_error(sbi))) { 1655 f2fs_submit_merged_write(sbi, NODE); 1656 submitted = NULL; 1657 } 1658 if (submitted) 1659 *submitted = fio.submitted; 1660 1661 if (do_balance) 1662 f2fs_balance_fs(sbi, false); 1663 return 0; 1664 1665 redirty_out: 1666 redirty_page_for_writepage(wbc, page); 1667 return AOP_WRITEPAGE_ACTIVATE; 1668 } 1669 1670 int f2fs_move_node_page(struct page *node_page, int gc_type) 1671 { 1672 int err = 0; 1673 1674 if (gc_type == FG_GC) { 1675 struct writeback_control wbc = { 1676 .sync_mode = WB_SYNC_ALL, 1677 .nr_to_write = 1, 1678 .for_reclaim = 0, 1679 }; 1680 1681 f2fs_wait_on_page_writeback(node_page, NODE, true, true); 1682 1683 set_page_dirty(node_page); 1684 1685 if (!clear_page_dirty_for_io(node_page)) { 1686 err = -EAGAIN; 1687 goto out_page; 1688 } 1689 1690 if (__write_node_page(node_page, false, NULL, 1691 &wbc, false, FS_GC_NODE_IO, NULL)) { 1692 err = -EAGAIN; 1693 unlock_page(node_page); 1694 } 1695 goto release_page; 1696 } else { 1697 /* set page dirty and write it */ 1698 if (!PageWriteback(node_page)) 1699 set_page_dirty(node_page); 1700 } 1701 out_page: 1702 unlock_page(node_page); 1703 release_page: 1704 f2fs_put_page(node_page, 0); 1705 return err; 1706 } 1707 1708 static int f2fs_write_node_page(struct page *page, 1709 struct writeback_control *wbc) 1710 { 1711 return __write_node_page(page, false, NULL, wbc, false, 1712 FS_NODE_IO, NULL); 1713 } 1714 1715 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode, 1716 struct writeback_control *wbc, bool atomic, 1717 unsigned int *seq_id) 1718 { 1719 pgoff_t index; 1720 struct pagevec pvec; 1721 int ret = 0; 1722 struct page *last_page = NULL; 1723 bool marked = false; 1724 nid_t ino = inode->i_ino; 1725 int nr_pages; 1726 int nwritten = 0; 1727 1728 if (atomic) { 1729 last_page = last_fsync_dnode(sbi, ino); 1730 if (IS_ERR_OR_NULL(last_page)) 1731 return PTR_ERR_OR_ZERO(last_page); 1732 } 1733 retry: 1734 pagevec_init(&pvec); 1735 index = 0; 1736 1737 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1738 PAGECACHE_TAG_DIRTY))) { 1739 int i; 1740 1741 for (i = 0; i < nr_pages; i++) { 1742 struct page *page = pvec.pages[i]; 1743 bool submitted = false; 1744 1745 if (unlikely(f2fs_cp_error(sbi))) { 1746 f2fs_put_page(last_page, 0); 1747 pagevec_release(&pvec); 1748 ret = -EIO; 1749 goto out; 1750 } 1751 1752 if (!IS_DNODE(page) || !is_cold_node(page)) 1753 continue; 1754 if (ino_of_node(page) != ino) 1755 continue; 1756 1757 lock_page(page); 1758 1759 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1760 continue_unlock: 1761 unlock_page(page); 1762 continue; 1763 } 1764 if (ino_of_node(page) != ino) 1765 goto continue_unlock; 1766 1767 if (!PageDirty(page) && page != last_page) { 1768 /* someone wrote it for us */ 1769 goto continue_unlock; 1770 } 1771 1772 f2fs_wait_on_page_writeback(page, NODE, true, true); 1773 1774 set_fsync_mark(page, 0); 1775 set_dentry_mark(page, 0); 1776 1777 if (!atomic || page == last_page) { 1778 set_fsync_mark(page, 1); 1779 if (IS_INODE(page)) { 1780 if (is_inode_flag_set(inode, 1781 FI_DIRTY_INODE)) 1782 f2fs_update_inode(inode, page); 1783 set_dentry_mark(page, 1784 f2fs_need_dentry_mark(sbi, ino)); 1785 } 1786 /* may be written by other thread */ 1787 if (!PageDirty(page)) 1788 set_page_dirty(page); 1789 } 1790 1791 if (!clear_page_dirty_for_io(page)) 1792 goto continue_unlock; 1793 1794 ret = __write_node_page(page, atomic && 1795 page == last_page, 1796 &submitted, wbc, true, 1797 FS_NODE_IO, seq_id); 1798 if (ret) { 1799 unlock_page(page); 1800 f2fs_put_page(last_page, 0); 1801 break; 1802 } else if (submitted) { 1803 nwritten++; 1804 } 1805 1806 if (page == last_page) { 1807 f2fs_put_page(page, 0); 1808 marked = true; 1809 break; 1810 } 1811 } 1812 pagevec_release(&pvec); 1813 cond_resched(); 1814 1815 if (ret || marked) 1816 break; 1817 } 1818 if (!ret && atomic && !marked) { 1819 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx", 1820 ino, last_page->index); 1821 lock_page(last_page); 1822 f2fs_wait_on_page_writeback(last_page, NODE, true, true); 1823 set_page_dirty(last_page); 1824 unlock_page(last_page); 1825 goto retry; 1826 } 1827 out: 1828 if (nwritten) 1829 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE); 1830 return ret ? -EIO : 0; 1831 } 1832 1833 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data) 1834 { 1835 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1836 bool clean; 1837 1838 if (inode->i_ino != ino) 1839 return 0; 1840 1841 if (!is_inode_flag_set(inode, FI_DIRTY_INODE)) 1842 return 0; 1843 1844 spin_lock(&sbi->inode_lock[DIRTY_META]); 1845 clean = list_empty(&F2FS_I(inode)->gdirty_list); 1846 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1847 1848 if (clean) 1849 return 0; 1850 1851 inode = igrab(inode); 1852 if (!inode) 1853 return 0; 1854 return 1; 1855 } 1856 1857 static bool flush_dirty_inode(struct page *page) 1858 { 1859 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1860 struct inode *inode; 1861 nid_t ino = ino_of_node(page); 1862 1863 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL); 1864 if (!inode) 1865 return false; 1866 1867 f2fs_update_inode(inode, page); 1868 unlock_page(page); 1869 1870 iput(inode); 1871 return true; 1872 } 1873 1874 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi) 1875 { 1876 pgoff_t index = 0; 1877 struct pagevec pvec; 1878 int nr_pages; 1879 1880 pagevec_init(&pvec); 1881 1882 while ((nr_pages = pagevec_lookup_tag(&pvec, 1883 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) { 1884 int i; 1885 1886 for (i = 0; i < nr_pages; i++) { 1887 struct page *page = pvec.pages[i]; 1888 1889 if (!IS_DNODE(page)) 1890 continue; 1891 1892 lock_page(page); 1893 1894 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1895 continue_unlock: 1896 unlock_page(page); 1897 continue; 1898 } 1899 1900 if (!PageDirty(page)) { 1901 /* someone wrote it for us */ 1902 goto continue_unlock; 1903 } 1904 1905 /* flush inline_data, if it's async context. */ 1906 if (page_private_inline(page)) { 1907 clear_page_private_inline(page); 1908 unlock_page(page); 1909 flush_inline_data(sbi, ino_of_node(page)); 1910 continue; 1911 } 1912 unlock_page(page); 1913 } 1914 pagevec_release(&pvec); 1915 cond_resched(); 1916 } 1917 } 1918 1919 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi, 1920 struct writeback_control *wbc, 1921 bool do_balance, enum iostat_type io_type) 1922 { 1923 pgoff_t index; 1924 struct pagevec pvec; 1925 int step = 0; 1926 int nwritten = 0; 1927 int ret = 0; 1928 int nr_pages, done = 0; 1929 1930 pagevec_init(&pvec); 1931 1932 next_step: 1933 index = 0; 1934 1935 while (!done && (nr_pages = pagevec_lookup_tag(&pvec, 1936 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) { 1937 int i; 1938 1939 for (i = 0; i < nr_pages; i++) { 1940 struct page *page = pvec.pages[i]; 1941 bool submitted = false; 1942 bool may_dirty = true; 1943 1944 /* give a priority to WB_SYNC threads */ 1945 if (atomic_read(&sbi->wb_sync_req[NODE]) && 1946 wbc->sync_mode == WB_SYNC_NONE) { 1947 done = 1; 1948 break; 1949 } 1950 1951 /* 1952 * flushing sequence with step: 1953 * 0. indirect nodes 1954 * 1. dentry dnodes 1955 * 2. file dnodes 1956 */ 1957 if (step == 0 && IS_DNODE(page)) 1958 continue; 1959 if (step == 1 && (!IS_DNODE(page) || 1960 is_cold_node(page))) 1961 continue; 1962 if (step == 2 && (!IS_DNODE(page) || 1963 !is_cold_node(page))) 1964 continue; 1965 lock_node: 1966 if (wbc->sync_mode == WB_SYNC_ALL) 1967 lock_page(page); 1968 else if (!trylock_page(page)) 1969 continue; 1970 1971 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1972 continue_unlock: 1973 unlock_page(page); 1974 continue; 1975 } 1976 1977 if (!PageDirty(page)) { 1978 /* someone wrote it for us */ 1979 goto continue_unlock; 1980 } 1981 1982 /* flush inline_data/inode, if it's async context. */ 1983 if (!do_balance) 1984 goto write_node; 1985 1986 /* flush inline_data */ 1987 if (page_private_inline(page)) { 1988 clear_page_private_inline(page); 1989 unlock_page(page); 1990 flush_inline_data(sbi, ino_of_node(page)); 1991 goto lock_node; 1992 } 1993 1994 /* flush dirty inode */ 1995 if (IS_INODE(page) && may_dirty) { 1996 may_dirty = false; 1997 if (flush_dirty_inode(page)) 1998 goto lock_node; 1999 } 2000 write_node: 2001 f2fs_wait_on_page_writeback(page, NODE, true, true); 2002 2003 if (!clear_page_dirty_for_io(page)) 2004 goto continue_unlock; 2005 2006 set_fsync_mark(page, 0); 2007 set_dentry_mark(page, 0); 2008 2009 ret = __write_node_page(page, false, &submitted, 2010 wbc, do_balance, io_type, NULL); 2011 if (ret) 2012 unlock_page(page); 2013 else if (submitted) 2014 nwritten++; 2015 2016 if (--wbc->nr_to_write == 0) 2017 break; 2018 } 2019 pagevec_release(&pvec); 2020 cond_resched(); 2021 2022 if (wbc->nr_to_write == 0) { 2023 step = 2; 2024 break; 2025 } 2026 } 2027 2028 if (step < 2) { 2029 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2030 wbc->sync_mode == WB_SYNC_NONE && step == 1) 2031 goto out; 2032 step++; 2033 goto next_step; 2034 } 2035 out: 2036 if (nwritten) 2037 f2fs_submit_merged_write(sbi, NODE); 2038 2039 if (unlikely(f2fs_cp_error(sbi))) 2040 return -EIO; 2041 return ret; 2042 } 2043 2044 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, 2045 unsigned int seq_id) 2046 { 2047 struct fsync_node_entry *fn; 2048 struct page *page; 2049 struct list_head *head = &sbi->fsync_node_list; 2050 unsigned long flags; 2051 unsigned int cur_seq_id = 0; 2052 int ret2, ret = 0; 2053 2054 while (seq_id && cur_seq_id < seq_id) { 2055 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 2056 if (list_empty(head)) { 2057 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2058 break; 2059 } 2060 fn = list_first_entry(head, struct fsync_node_entry, list); 2061 if (fn->seq_id > seq_id) { 2062 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2063 break; 2064 } 2065 cur_seq_id = fn->seq_id; 2066 page = fn->page; 2067 get_page(page); 2068 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 2069 2070 f2fs_wait_on_page_writeback(page, NODE, true, false); 2071 if (TestClearPageError(page)) 2072 ret = -EIO; 2073 2074 put_page(page); 2075 2076 if (ret) 2077 break; 2078 } 2079 2080 ret2 = filemap_check_errors(NODE_MAPPING(sbi)); 2081 if (!ret) 2082 ret = ret2; 2083 2084 return ret; 2085 } 2086 2087 static int f2fs_write_node_pages(struct address_space *mapping, 2088 struct writeback_control *wbc) 2089 { 2090 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2091 struct blk_plug plug; 2092 long diff; 2093 2094 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2095 goto skip_write; 2096 2097 /* balancing f2fs's metadata in background */ 2098 f2fs_balance_fs_bg(sbi, true); 2099 2100 /* collect a number of dirty node pages and write together */ 2101 if (wbc->sync_mode != WB_SYNC_ALL && 2102 get_pages(sbi, F2FS_DIRTY_NODES) < 2103 nr_pages_to_skip(sbi, NODE)) 2104 goto skip_write; 2105 2106 if (wbc->sync_mode == WB_SYNC_ALL) 2107 atomic_inc(&sbi->wb_sync_req[NODE]); 2108 else if (atomic_read(&sbi->wb_sync_req[NODE])) 2109 goto skip_write; 2110 2111 trace_f2fs_writepages(mapping->host, wbc, NODE); 2112 2113 diff = nr_pages_to_write(sbi, NODE, wbc); 2114 blk_start_plug(&plug); 2115 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO); 2116 blk_finish_plug(&plug); 2117 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); 2118 2119 if (wbc->sync_mode == WB_SYNC_ALL) 2120 atomic_dec(&sbi->wb_sync_req[NODE]); 2121 return 0; 2122 2123 skip_write: 2124 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES); 2125 trace_f2fs_writepages(mapping->host, wbc, NODE); 2126 return 0; 2127 } 2128 2129 static int f2fs_set_node_page_dirty(struct page *page) 2130 { 2131 trace_f2fs_set_page_dirty(page, NODE); 2132 2133 if (!PageUptodate(page)) 2134 SetPageUptodate(page); 2135 #ifdef CONFIG_F2FS_CHECK_FS 2136 if (IS_INODE(page)) 2137 f2fs_inode_chksum_set(F2FS_P_SB(page), page); 2138 #endif 2139 if (!PageDirty(page)) { 2140 __set_page_dirty_nobuffers(page); 2141 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES); 2142 set_page_private_reference(page); 2143 return 1; 2144 } 2145 return 0; 2146 } 2147 2148 /* 2149 * Structure of the f2fs node operations 2150 */ 2151 const struct address_space_operations f2fs_node_aops = { 2152 .writepage = f2fs_write_node_page, 2153 .writepages = f2fs_write_node_pages, 2154 .set_page_dirty = f2fs_set_node_page_dirty, 2155 .invalidatepage = f2fs_invalidate_page, 2156 .releasepage = f2fs_release_page, 2157 #ifdef CONFIG_MIGRATION 2158 .migratepage = f2fs_migrate_page, 2159 #endif 2160 }; 2161 2162 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i, 2163 nid_t n) 2164 { 2165 return radix_tree_lookup(&nm_i->free_nid_root, n); 2166 } 2167 2168 static int __insert_free_nid(struct f2fs_sb_info *sbi, 2169 struct free_nid *i) 2170 { 2171 struct f2fs_nm_info *nm_i = NM_I(sbi); 2172 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i); 2173 2174 if (err) 2175 return err; 2176 2177 nm_i->nid_cnt[FREE_NID]++; 2178 list_add_tail(&i->list, &nm_i->free_nid_list); 2179 return 0; 2180 } 2181 2182 static void __remove_free_nid(struct f2fs_sb_info *sbi, 2183 struct free_nid *i, enum nid_state state) 2184 { 2185 struct f2fs_nm_info *nm_i = NM_I(sbi); 2186 2187 f2fs_bug_on(sbi, state != i->state); 2188 nm_i->nid_cnt[state]--; 2189 if (state == FREE_NID) 2190 list_del(&i->list); 2191 radix_tree_delete(&nm_i->free_nid_root, i->nid); 2192 } 2193 2194 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i, 2195 enum nid_state org_state, enum nid_state dst_state) 2196 { 2197 struct f2fs_nm_info *nm_i = NM_I(sbi); 2198 2199 f2fs_bug_on(sbi, org_state != i->state); 2200 i->state = dst_state; 2201 nm_i->nid_cnt[org_state]--; 2202 nm_i->nid_cnt[dst_state]++; 2203 2204 switch (dst_state) { 2205 case PREALLOC_NID: 2206 list_del(&i->list); 2207 break; 2208 case FREE_NID: 2209 list_add_tail(&i->list, &nm_i->free_nid_list); 2210 break; 2211 default: 2212 BUG_ON(1); 2213 } 2214 } 2215 2216 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi) 2217 { 2218 struct f2fs_nm_info *nm_i = NM_I(sbi); 2219 unsigned int i; 2220 bool ret = true; 2221 2222 down_read(&nm_i->nat_tree_lock); 2223 for (i = 0; i < nm_i->nat_blocks; i++) { 2224 if (!test_bit_le(i, nm_i->nat_block_bitmap)) { 2225 ret = false; 2226 break; 2227 } 2228 } 2229 up_read(&nm_i->nat_tree_lock); 2230 2231 return ret; 2232 } 2233 2234 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid, 2235 bool set, bool build) 2236 { 2237 struct f2fs_nm_info *nm_i = NM_I(sbi); 2238 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid); 2239 unsigned int nid_ofs = nid - START_NID(nid); 2240 2241 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap)) 2242 return; 2243 2244 if (set) { 2245 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2246 return; 2247 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2248 nm_i->free_nid_count[nat_ofs]++; 2249 } else { 2250 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2251 return; 2252 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2253 if (!build) 2254 nm_i->free_nid_count[nat_ofs]--; 2255 } 2256 } 2257 2258 /* return if the nid is recognized as free */ 2259 static bool add_free_nid(struct f2fs_sb_info *sbi, 2260 nid_t nid, bool build, bool update) 2261 { 2262 struct f2fs_nm_info *nm_i = NM_I(sbi); 2263 struct free_nid *i, *e; 2264 struct nat_entry *ne; 2265 int err = -EINVAL; 2266 bool ret = false; 2267 2268 /* 0 nid should not be used */ 2269 if (unlikely(nid == 0)) 2270 return false; 2271 2272 if (unlikely(f2fs_check_nid_range(sbi, nid))) 2273 return false; 2274 2275 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL); 2276 i->nid = nid; 2277 i->state = FREE_NID; 2278 2279 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 2280 2281 spin_lock(&nm_i->nid_list_lock); 2282 2283 if (build) { 2284 /* 2285 * Thread A Thread B 2286 * - f2fs_create 2287 * - f2fs_new_inode 2288 * - f2fs_alloc_nid 2289 * - __insert_nid_to_list(PREALLOC_NID) 2290 * - f2fs_balance_fs_bg 2291 * - f2fs_build_free_nids 2292 * - __f2fs_build_free_nids 2293 * - scan_nat_page 2294 * - add_free_nid 2295 * - __lookup_nat_cache 2296 * - f2fs_add_link 2297 * - f2fs_init_inode_metadata 2298 * - f2fs_new_inode_page 2299 * - f2fs_new_node_page 2300 * - set_node_addr 2301 * - f2fs_alloc_nid_done 2302 * - __remove_nid_from_list(PREALLOC_NID) 2303 * - __insert_nid_to_list(FREE_NID) 2304 */ 2305 ne = __lookup_nat_cache(nm_i, nid); 2306 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) || 2307 nat_get_blkaddr(ne) != NULL_ADDR)) 2308 goto err_out; 2309 2310 e = __lookup_free_nid_list(nm_i, nid); 2311 if (e) { 2312 if (e->state == FREE_NID) 2313 ret = true; 2314 goto err_out; 2315 } 2316 } 2317 ret = true; 2318 err = __insert_free_nid(sbi, i); 2319 err_out: 2320 if (update) { 2321 update_free_nid_bitmap(sbi, nid, ret, build); 2322 if (!build) 2323 nm_i->available_nids++; 2324 } 2325 spin_unlock(&nm_i->nid_list_lock); 2326 radix_tree_preload_end(); 2327 2328 if (err) 2329 kmem_cache_free(free_nid_slab, i); 2330 return ret; 2331 } 2332 2333 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid) 2334 { 2335 struct f2fs_nm_info *nm_i = NM_I(sbi); 2336 struct free_nid *i; 2337 bool need_free = false; 2338 2339 spin_lock(&nm_i->nid_list_lock); 2340 i = __lookup_free_nid_list(nm_i, nid); 2341 if (i && i->state == FREE_NID) { 2342 __remove_free_nid(sbi, i, FREE_NID); 2343 need_free = true; 2344 } 2345 spin_unlock(&nm_i->nid_list_lock); 2346 2347 if (need_free) 2348 kmem_cache_free(free_nid_slab, i); 2349 } 2350 2351 static int scan_nat_page(struct f2fs_sb_info *sbi, 2352 struct page *nat_page, nid_t start_nid) 2353 { 2354 struct f2fs_nm_info *nm_i = NM_I(sbi); 2355 struct f2fs_nat_block *nat_blk = page_address(nat_page); 2356 block_t blk_addr; 2357 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid); 2358 int i; 2359 2360 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap); 2361 2362 i = start_nid % NAT_ENTRY_PER_BLOCK; 2363 2364 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) { 2365 if (unlikely(start_nid >= nm_i->max_nid)) 2366 break; 2367 2368 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr); 2369 2370 if (blk_addr == NEW_ADDR) 2371 return -EINVAL; 2372 2373 if (blk_addr == NULL_ADDR) { 2374 add_free_nid(sbi, start_nid, true, true); 2375 } else { 2376 spin_lock(&NM_I(sbi)->nid_list_lock); 2377 update_free_nid_bitmap(sbi, start_nid, false, true); 2378 spin_unlock(&NM_I(sbi)->nid_list_lock); 2379 } 2380 } 2381 2382 return 0; 2383 } 2384 2385 static void scan_curseg_cache(struct f2fs_sb_info *sbi) 2386 { 2387 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2388 struct f2fs_journal *journal = curseg->journal; 2389 int i; 2390 2391 down_read(&curseg->journal_rwsem); 2392 for (i = 0; i < nats_in_cursum(journal); i++) { 2393 block_t addr; 2394 nid_t nid; 2395 2396 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr); 2397 nid = le32_to_cpu(nid_in_journal(journal, i)); 2398 if (addr == NULL_ADDR) 2399 add_free_nid(sbi, nid, true, false); 2400 else 2401 remove_free_nid(sbi, nid); 2402 } 2403 up_read(&curseg->journal_rwsem); 2404 } 2405 2406 static void scan_free_nid_bits(struct f2fs_sb_info *sbi) 2407 { 2408 struct f2fs_nm_info *nm_i = NM_I(sbi); 2409 unsigned int i, idx; 2410 nid_t nid; 2411 2412 down_read(&nm_i->nat_tree_lock); 2413 2414 for (i = 0; i < nm_i->nat_blocks; i++) { 2415 if (!test_bit_le(i, nm_i->nat_block_bitmap)) 2416 continue; 2417 if (!nm_i->free_nid_count[i]) 2418 continue; 2419 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) { 2420 idx = find_next_bit_le(nm_i->free_nid_bitmap[i], 2421 NAT_ENTRY_PER_BLOCK, idx); 2422 if (idx >= NAT_ENTRY_PER_BLOCK) 2423 break; 2424 2425 nid = i * NAT_ENTRY_PER_BLOCK + idx; 2426 add_free_nid(sbi, nid, true, false); 2427 2428 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS) 2429 goto out; 2430 } 2431 } 2432 out: 2433 scan_curseg_cache(sbi); 2434 2435 up_read(&nm_i->nat_tree_lock); 2436 } 2437 2438 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi, 2439 bool sync, bool mount) 2440 { 2441 struct f2fs_nm_info *nm_i = NM_I(sbi); 2442 int i = 0, ret; 2443 nid_t nid = nm_i->next_scan_nid; 2444 2445 if (unlikely(nid >= nm_i->max_nid)) 2446 nid = 0; 2447 2448 if (unlikely(nid % NAT_ENTRY_PER_BLOCK)) 2449 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK; 2450 2451 /* Enough entries */ 2452 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2453 return 0; 2454 2455 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS)) 2456 return 0; 2457 2458 if (!mount) { 2459 /* try to find free nids in free_nid_bitmap */ 2460 scan_free_nid_bits(sbi); 2461 2462 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2463 return 0; 2464 } 2465 2466 /* readahead nat pages to be scanned */ 2467 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, 2468 META_NAT, true); 2469 2470 down_read(&nm_i->nat_tree_lock); 2471 2472 while (1) { 2473 if (!test_bit_le(NAT_BLOCK_OFFSET(nid), 2474 nm_i->nat_block_bitmap)) { 2475 struct page *page = get_current_nat_page(sbi, nid); 2476 2477 if (IS_ERR(page)) { 2478 ret = PTR_ERR(page); 2479 } else { 2480 ret = scan_nat_page(sbi, page, nid); 2481 f2fs_put_page(page, 1); 2482 } 2483 2484 if (ret) { 2485 up_read(&nm_i->nat_tree_lock); 2486 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it"); 2487 return ret; 2488 } 2489 } 2490 2491 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK)); 2492 if (unlikely(nid >= nm_i->max_nid)) 2493 nid = 0; 2494 2495 if (++i >= FREE_NID_PAGES) 2496 break; 2497 } 2498 2499 /* go to the next free nat pages to find free nids abundantly */ 2500 nm_i->next_scan_nid = nid; 2501 2502 /* find free nids from current sum_pages */ 2503 scan_curseg_cache(sbi); 2504 2505 up_read(&nm_i->nat_tree_lock); 2506 2507 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid), 2508 nm_i->ra_nid_pages, META_NAT, false); 2509 2510 return 0; 2511 } 2512 2513 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount) 2514 { 2515 int ret; 2516 2517 mutex_lock(&NM_I(sbi)->build_lock); 2518 ret = __f2fs_build_free_nids(sbi, sync, mount); 2519 mutex_unlock(&NM_I(sbi)->build_lock); 2520 2521 return ret; 2522 } 2523 2524 /* 2525 * If this function returns success, caller can obtain a new nid 2526 * from second parameter of this function. 2527 * The returned nid could be used ino as well as nid when inode is created. 2528 */ 2529 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid) 2530 { 2531 struct f2fs_nm_info *nm_i = NM_I(sbi); 2532 struct free_nid *i = NULL; 2533 retry: 2534 if (time_to_inject(sbi, FAULT_ALLOC_NID)) { 2535 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID); 2536 return false; 2537 } 2538 2539 spin_lock(&nm_i->nid_list_lock); 2540 2541 if (unlikely(nm_i->available_nids == 0)) { 2542 spin_unlock(&nm_i->nid_list_lock); 2543 return false; 2544 } 2545 2546 /* We should not use stale free nids created by f2fs_build_free_nids */ 2547 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) { 2548 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list)); 2549 i = list_first_entry(&nm_i->free_nid_list, 2550 struct free_nid, list); 2551 *nid = i->nid; 2552 2553 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID); 2554 nm_i->available_nids--; 2555 2556 update_free_nid_bitmap(sbi, *nid, false, false); 2557 2558 spin_unlock(&nm_i->nid_list_lock); 2559 return true; 2560 } 2561 spin_unlock(&nm_i->nid_list_lock); 2562 2563 /* Let's scan nat pages and its caches to get free nids */ 2564 if (!f2fs_build_free_nids(sbi, true, false)) 2565 goto retry; 2566 return false; 2567 } 2568 2569 /* 2570 * f2fs_alloc_nid() should be called prior to this function. 2571 */ 2572 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 2573 { 2574 struct f2fs_nm_info *nm_i = NM_I(sbi); 2575 struct free_nid *i; 2576 2577 spin_lock(&nm_i->nid_list_lock); 2578 i = __lookup_free_nid_list(nm_i, nid); 2579 f2fs_bug_on(sbi, !i); 2580 __remove_free_nid(sbi, i, PREALLOC_NID); 2581 spin_unlock(&nm_i->nid_list_lock); 2582 2583 kmem_cache_free(free_nid_slab, i); 2584 } 2585 2586 /* 2587 * f2fs_alloc_nid() should be called prior to this function. 2588 */ 2589 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 2590 { 2591 struct f2fs_nm_info *nm_i = NM_I(sbi); 2592 struct free_nid *i; 2593 bool need_free = false; 2594 2595 if (!nid) 2596 return; 2597 2598 spin_lock(&nm_i->nid_list_lock); 2599 i = __lookup_free_nid_list(nm_i, nid); 2600 f2fs_bug_on(sbi, !i); 2601 2602 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) { 2603 __remove_free_nid(sbi, i, PREALLOC_NID); 2604 need_free = true; 2605 } else { 2606 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID); 2607 } 2608 2609 nm_i->available_nids++; 2610 2611 update_free_nid_bitmap(sbi, nid, true, false); 2612 2613 spin_unlock(&nm_i->nid_list_lock); 2614 2615 if (need_free) 2616 kmem_cache_free(free_nid_slab, i); 2617 } 2618 2619 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink) 2620 { 2621 struct f2fs_nm_info *nm_i = NM_I(sbi); 2622 int nr = nr_shrink; 2623 2624 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2625 return 0; 2626 2627 if (!mutex_trylock(&nm_i->build_lock)) 2628 return 0; 2629 2630 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) { 2631 struct free_nid *i, *next; 2632 unsigned int batch = SHRINK_NID_BATCH_SIZE; 2633 2634 spin_lock(&nm_i->nid_list_lock); 2635 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) { 2636 if (!nr_shrink || !batch || 2637 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2638 break; 2639 __remove_free_nid(sbi, i, FREE_NID); 2640 kmem_cache_free(free_nid_slab, i); 2641 nr_shrink--; 2642 batch--; 2643 } 2644 spin_unlock(&nm_i->nid_list_lock); 2645 } 2646 2647 mutex_unlock(&nm_i->build_lock); 2648 2649 return nr - nr_shrink; 2650 } 2651 2652 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page) 2653 { 2654 void *src_addr, *dst_addr; 2655 size_t inline_size; 2656 struct page *ipage; 2657 struct f2fs_inode *ri; 2658 2659 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 2660 if (IS_ERR(ipage)) 2661 return PTR_ERR(ipage); 2662 2663 ri = F2FS_INODE(page); 2664 if (ri->i_inline & F2FS_INLINE_XATTR) { 2665 if (!f2fs_has_inline_xattr(inode)) { 2666 set_inode_flag(inode, FI_INLINE_XATTR); 2667 stat_inc_inline_xattr(inode); 2668 } 2669 } else { 2670 if (f2fs_has_inline_xattr(inode)) { 2671 stat_dec_inline_xattr(inode); 2672 clear_inode_flag(inode, FI_INLINE_XATTR); 2673 } 2674 goto update_inode; 2675 } 2676 2677 dst_addr = inline_xattr_addr(inode, ipage); 2678 src_addr = inline_xattr_addr(inode, page); 2679 inline_size = inline_xattr_size(inode); 2680 2681 f2fs_wait_on_page_writeback(ipage, NODE, true, true); 2682 memcpy(dst_addr, src_addr, inline_size); 2683 update_inode: 2684 f2fs_update_inode(inode, ipage); 2685 f2fs_put_page(ipage, 1); 2686 return 0; 2687 } 2688 2689 int f2fs_recover_xattr_data(struct inode *inode, struct page *page) 2690 { 2691 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2692 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid; 2693 nid_t new_xnid; 2694 struct dnode_of_data dn; 2695 struct node_info ni; 2696 struct page *xpage; 2697 int err; 2698 2699 if (!prev_xnid) 2700 goto recover_xnid; 2701 2702 /* 1: invalidate the previous xattr nid */ 2703 err = f2fs_get_node_info(sbi, prev_xnid, &ni); 2704 if (err) 2705 return err; 2706 2707 f2fs_invalidate_blocks(sbi, ni.blk_addr); 2708 dec_valid_node_count(sbi, inode, false); 2709 set_node_addr(sbi, &ni, NULL_ADDR, false); 2710 2711 recover_xnid: 2712 /* 2: update xattr nid in inode */ 2713 if (!f2fs_alloc_nid(sbi, &new_xnid)) 2714 return -ENOSPC; 2715 2716 set_new_dnode(&dn, inode, NULL, NULL, new_xnid); 2717 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET); 2718 if (IS_ERR(xpage)) { 2719 f2fs_alloc_nid_failed(sbi, new_xnid); 2720 return PTR_ERR(xpage); 2721 } 2722 2723 f2fs_alloc_nid_done(sbi, new_xnid); 2724 f2fs_update_inode_page(inode); 2725 2726 /* 3: update and set xattr node page dirty */ 2727 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE); 2728 2729 set_page_dirty(xpage); 2730 f2fs_put_page(xpage, 1); 2731 2732 return 0; 2733 } 2734 2735 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 2736 { 2737 struct f2fs_inode *src, *dst; 2738 nid_t ino = ino_of_node(page); 2739 struct node_info old_ni, new_ni; 2740 struct page *ipage; 2741 int err; 2742 2743 err = f2fs_get_node_info(sbi, ino, &old_ni); 2744 if (err) 2745 return err; 2746 2747 if (unlikely(old_ni.blk_addr != NULL_ADDR)) 2748 return -EINVAL; 2749 retry: 2750 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false); 2751 if (!ipage) { 2752 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT); 2753 goto retry; 2754 } 2755 2756 /* Should not use this inode from free nid list */ 2757 remove_free_nid(sbi, ino); 2758 2759 if (!PageUptodate(ipage)) 2760 SetPageUptodate(ipage); 2761 fill_node_footer(ipage, ino, ino, 0, true); 2762 set_cold_node(ipage, false); 2763 2764 src = F2FS_INODE(page); 2765 dst = F2FS_INODE(ipage); 2766 2767 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext)); 2768 dst->i_size = 0; 2769 dst->i_blocks = cpu_to_le64(1); 2770 dst->i_links = cpu_to_le32(1); 2771 dst->i_xattr_nid = 0; 2772 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR); 2773 if (dst->i_inline & F2FS_EXTRA_ATTR) { 2774 dst->i_extra_isize = src->i_extra_isize; 2775 2776 if (f2fs_sb_has_flexible_inline_xattr(sbi) && 2777 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2778 i_inline_xattr_size)) 2779 dst->i_inline_xattr_size = src->i_inline_xattr_size; 2780 2781 if (f2fs_sb_has_project_quota(sbi) && 2782 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2783 i_projid)) 2784 dst->i_projid = src->i_projid; 2785 2786 if (f2fs_sb_has_inode_crtime(sbi) && 2787 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2788 i_crtime_nsec)) { 2789 dst->i_crtime = src->i_crtime; 2790 dst->i_crtime_nsec = src->i_crtime_nsec; 2791 } 2792 } 2793 2794 new_ni = old_ni; 2795 new_ni.ino = ino; 2796 2797 if (unlikely(inc_valid_node_count(sbi, NULL, true))) 2798 WARN_ON(1); 2799 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 2800 inc_valid_inode_count(sbi); 2801 set_page_dirty(ipage); 2802 f2fs_put_page(ipage, 1); 2803 return 0; 2804 } 2805 2806 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi, 2807 unsigned int segno, struct f2fs_summary_block *sum) 2808 { 2809 struct f2fs_node *rn; 2810 struct f2fs_summary *sum_entry; 2811 block_t addr; 2812 int i, idx, last_offset, nrpages; 2813 2814 /* scan the node segment */ 2815 last_offset = sbi->blocks_per_seg; 2816 addr = START_BLOCK(sbi, segno); 2817 sum_entry = &sum->entries[0]; 2818 2819 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) { 2820 nrpages = bio_max_segs(last_offset - i); 2821 2822 /* readahead node pages */ 2823 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true); 2824 2825 for (idx = addr; idx < addr + nrpages; idx++) { 2826 struct page *page = f2fs_get_tmp_page(sbi, idx); 2827 2828 if (IS_ERR(page)) 2829 return PTR_ERR(page); 2830 2831 rn = F2FS_NODE(page); 2832 sum_entry->nid = rn->footer.nid; 2833 sum_entry->version = 0; 2834 sum_entry->ofs_in_node = 0; 2835 sum_entry++; 2836 f2fs_put_page(page, 1); 2837 } 2838 2839 invalidate_mapping_pages(META_MAPPING(sbi), addr, 2840 addr + nrpages); 2841 } 2842 return 0; 2843 } 2844 2845 static void remove_nats_in_journal(struct f2fs_sb_info *sbi) 2846 { 2847 struct f2fs_nm_info *nm_i = NM_I(sbi); 2848 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2849 struct f2fs_journal *journal = curseg->journal; 2850 int i; 2851 2852 down_write(&curseg->journal_rwsem); 2853 for (i = 0; i < nats_in_cursum(journal); i++) { 2854 struct nat_entry *ne; 2855 struct f2fs_nat_entry raw_ne; 2856 nid_t nid = le32_to_cpu(nid_in_journal(journal, i)); 2857 2858 if (f2fs_check_nid_range(sbi, nid)) 2859 continue; 2860 2861 raw_ne = nat_in_journal(journal, i); 2862 2863 ne = __lookup_nat_cache(nm_i, nid); 2864 if (!ne) { 2865 ne = __alloc_nat_entry(sbi, nid, true); 2866 __init_nat_entry(nm_i, ne, &raw_ne, true); 2867 } 2868 2869 /* 2870 * if a free nat in journal has not been used after last 2871 * checkpoint, we should remove it from available nids, 2872 * since later we will add it again. 2873 */ 2874 if (!get_nat_flag(ne, IS_DIRTY) && 2875 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) { 2876 spin_lock(&nm_i->nid_list_lock); 2877 nm_i->available_nids--; 2878 spin_unlock(&nm_i->nid_list_lock); 2879 } 2880 2881 __set_nat_cache_dirty(nm_i, ne); 2882 } 2883 update_nats_in_cursum(journal, -i); 2884 up_write(&curseg->journal_rwsem); 2885 } 2886 2887 static void __adjust_nat_entry_set(struct nat_entry_set *nes, 2888 struct list_head *head, int max) 2889 { 2890 struct nat_entry_set *cur; 2891 2892 if (nes->entry_cnt >= max) 2893 goto add_out; 2894 2895 list_for_each_entry(cur, head, set_list) { 2896 if (cur->entry_cnt >= nes->entry_cnt) { 2897 list_add(&nes->set_list, cur->set_list.prev); 2898 return; 2899 } 2900 } 2901 add_out: 2902 list_add_tail(&nes->set_list, head); 2903 } 2904 2905 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs, 2906 unsigned int valid) 2907 { 2908 if (valid == 0) { 2909 __set_bit_le(nat_ofs, nm_i->empty_nat_bits); 2910 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2911 return; 2912 } 2913 2914 __clear_bit_le(nat_ofs, nm_i->empty_nat_bits); 2915 if (valid == NAT_ENTRY_PER_BLOCK) 2916 __set_bit_le(nat_ofs, nm_i->full_nat_bits); 2917 else 2918 __clear_bit_le(nat_ofs, nm_i->full_nat_bits); 2919 } 2920 2921 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid, 2922 struct page *page) 2923 { 2924 struct f2fs_nm_info *nm_i = NM_I(sbi); 2925 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK; 2926 struct f2fs_nat_block *nat_blk = page_address(page); 2927 int valid = 0; 2928 int i = 0; 2929 2930 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 2931 return; 2932 2933 if (nat_index == 0) { 2934 valid = 1; 2935 i = 1; 2936 } 2937 for (; i < NAT_ENTRY_PER_BLOCK; i++) { 2938 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR) 2939 valid++; 2940 } 2941 2942 __update_nat_bits(nm_i, nat_index, valid); 2943 } 2944 2945 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi) 2946 { 2947 struct f2fs_nm_info *nm_i = NM_I(sbi); 2948 unsigned int nat_ofs; 2949 2950 down_read(&nm_i->nat_tree_lock); 2951 2952 for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) { 2953 unsigned int valid = 0, nid_ofs = 0; 2954 2955 /* handle nid zero due to it should never be used */ 2956 if (unlikely(nat_ofs == 0)) { 2957 valid = 1; 2958 nid_ofs = 1; 2959 } 2960 2961 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) { 2962 if (!test_bit_le(nid_ofs, 2963 nm_i->free_nid_bitmap[nat_ofs])) 2964 valid++; 2965 } 2966 2967 __update_nat_bits(nm_i, nat_ofs, valid); 2968 } 2969 2970 up_read(&nm_i->nat_tree_lock); 2971 } 2972 2973 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi, 2974 struct nat_entry_set *set, struct cp_control *cpc) 2975 { 2976 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2977 struct f2fs_journal *journal = curseg->journal; 2978 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK; 2979 bool to_journal = true; 2980 struct f2fs_nat_block *nat_blk; 2981 struct nat_entry *ne, *cur; 2982 struct page *page = NULL; 2983 2984 /* 2985 * there are two steps to flush nat entries: 2986 * #1, flush nat entries to journal in current hot data summary block. 2987 * #2, flush nat entries to nat page. 2988 */ 2989 if ((cpc->reason & CP_UMOUNT) || 2990 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL)) 2991 to_journal = false; 2992 2993 if (to_journal) { 2994 down_write(&curseg->journal_rwsem); 2995 } else { 2996 page = get_next_nat_page(sbi, start_nid); 2997 if (IS_ERR(page)) 2998 return PTR_ERR(page); 2999 3000 nat_blk = page_address(page); 3001 f2fs_bug_on(sbi, !nat_blk); 3002 } 3003 3004 /* flush dirty nats in nat entry set */ 3005 list_for_each_entry_safe(ne, cur, &set->entry_list, list) { 3006 struct f2fs_nat_entry *raw_ne; 3007 nid_t nid = nat_get_nid(ne); 3008 int offset; 3009 3010 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR); 3011 3012 if (to_journal) { 3013 offset = f2fs_lookup_journal_in_cursum(journal, 3014 NAT_JOURNAL, nid, 1); 3015 f2fs_bug_on(sbi, offset < 0); 3016 raw_ne = &nat_in_journal(journal, offset); 3017 nid_in_journal(journal, offset) = cpu_to_le32(nid); 3018 } else { 3019 raw_ne = &nat_blk->entries[nid - start_nid]; 3020 } 3021 raw_nat_from_node_info(raw_ne, &ne->ni); 3022 nat_reset_flag(ne); 3023 __clear_nat_cache_dirty(NM_I(sbi), set, ne); 3024 if (nat_get_blkaddr(ne) == NULL_ADDR) { 3025 add_free_nid(sbi, nid, false, true); 3026 } else { 3027 spin_lock(&NM_I(sbi)->nid_list_lock); 3028 update_free_nid_bitmap(sbi, nid, false, false); 3029 spin_unlock(&NM_I(sbi)->nid_list_lock); 3030 } 3031 } 3032 3033 if (to_journal) { 3034 up_write(&curseg->journal_rwsem); 3035 } else { 3036 update_nat_bits(sbi, start_nid, page); 3037 f2fs_put_page(page, 1); 3038 } 3039 3040 /* Allow dirty nats by node block allocation in write_begin */ 3041 if (!set->entry_cnt) { 3042 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set); 3043 kmem_cache_free(nat_entry_set_slab, set); 3044 } 3045 return 0; 3046 } 3047 3048 /* 3049 * This function is called during the checkpointing process. 3050 */ 3051 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3052 { 3053 struct f2fs_nm_info *nm_i = NM_I(sbi); 3054 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 3055 struct f2fs_journal *journal = curseg->journal; 3056 struct nat_entry_set *setvec[SETVEC_SIZE]; 3057 struct nat_entry_set *set, *tmp; 3058 unsigned int found; 3059 nid_t set_idx = 0; 3060 LIST_HEAD(sets); 3061 int err = 0; 3062 3063 /* 3064 * during unmount, let's flush nat_bits before checking 3065 * nat_cnt[DIRTY_NAT]. 3066 */ 3067 if (cpc->reason & CP_UMOUNT) { 3068 down_write(&nm_i->nat_tree_lock); 3069 remove_nats_in_journal(sbi); 3070 up_write(&nm_i->nat_tree_lock); 3071 } 3072 3073 if (!nm_i->nat_cnt[DIRTY_NAT]) 3074 return 0; 3075 3076 down_write(&nm_i->nat_tree_lock); 3077 3078 /* 3079 * if there are no enough space in journal to store dirty nat 3080 * entries, remove all entries from journal and merge them 3081 * into nat entry set. 3082 */ 3083 if (cpc->reason & CP_UMOUNT || 3084 !__has_cursum_space(journal, 3085 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL)) 3086 remove_nats_in_journal(sbi); 3087 3088 while ((found = __gang_lookup_nat_set(nm_i, 3089 set_idx, SETVEC_SIZE, setvec))) { 3090 unsigned idx; 3091 3092 set_idx = setvec[found - 1]->set + 1; 3093 for (idx = 0; idx < found; idx++) 3094 __adjust_nat_entry_set(setvec[idx], &sets, 3095 MAX_NAT_JENTRIES(journal)); 3096 } 3097 3098 /* flush dirty nats in nat entry set */ 3099 list_for_each_entry_safe(set, tmp, &sets, set_list) { 3100 err = __flush_nat_entry_set(sbi, set, cpc); 3101 if (err) 3102 break; 3103 } 3104 3105 up_write(&nm_i->nat_tree_lock); 3106 /* Allow dirty nats by node block allocation in write_begin */ 3107 3108 return err; 3109 } 3110 3111 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi) 3112 { 3113 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3114 struct f2fs_nm_info *nm_i = NM_I(sbi); 3115 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE; 3116 unsigned int i; 3117 __u64 cp_ver = cur_cp_version(ckpt); 3118 block_t nat_bits_addr; 3119 3120 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8); 3121 nm_i->nat_bits = f2fs_kvzalloc(sbi, 3122 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL); 3123 if (!nm_i->nat_bits) 3124 return -ENOMEM; 3125 3126 nm_i->full_nat_bits = nm_i->nat_bits + 8; 3127 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes; 3128 3129 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3130 return 0; 3131 3132 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg - 3133 nm_i->nat_bits_blocks; 3134 for (i = 0; i < nm_i->nat_bits_blocks; i++) { 3135 struct page *page; 3136 3137 page = f2fs_get_meta_page(sbi, nat_bits_addr++); 3138 if (IS_ERR(page)) 3139 return PTR_ERR(page); 3140 3141 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS), 3142 page_address(page), F2FS_BLKSIZE); 3143 f2fs_put_page(page, 1); 3144 } 3145 3146 cp_ver |= (cur_cp_crc(ckpt) << 32); 3147 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) { 3148 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG); 3149 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)", 3150 cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits)); 3151 return 0; 3152 } 3153 3154 f2fs_notice(sbi, "Found nat_bits in checkpoint"); 3155 return 0; 3156 } 3157 3158 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi) 3159 { 3160 struct f2fs_nm_info *nm_i = NM_I(sbi); 3161 unsigned int i = 0; 3162 nid_t nid, last_nid; 3163 3164 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) 3165 return; 3166 3167 for (i = 0; i < nm_i->nat_blocks; i++) { 3168 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i); 3169 if (i >= nm_i->nat_blocks) 3170 break; 3171 3172 __set_bit_le(i, nm_i->nat_block_bitmap); 3173 3174 nid = i * NAT_ENTRY_PER_BLOCK; 3175 last_nid = nid + NAT_ENTRY_PER_BLOCK; 3176 3177 spin_lock(&NM_I(sbi)->nid_list_lock); 3178 for (; nid < last_nid; nid++) 3179 update_free_nid_bitmap(sbi, nid, true, true); 3180 spin_unlock(&NM_I(sbi)->nid_list_lock); 3181 } 3182 3183 for (i = 0; i < nm_i->nat_blocks; i++) { 3184 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i); 3185 if (i >= nm_i->nat_blocks) 3186 break; 3187 3188 __set_bit_le(i, nm_i->nat_block_bitmap); 3189 } 3190 } 3191 3192 static int init_node_manager(struct f2fs_sb_info *sbi) 3193 { 3194 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 3195 struct f2fs_nm_info *nm_i = NM_I(sbi); 3196 unsigned char *version_bitmap; 3197 unsigned int nat_segs; 3198 int err; 3199 3200 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 3201 3202 /* segment_count_nat includes pair segment so divide to 2. */ 3203 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 3204 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 3205 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks; 3206 3207 /* not used nids: 0, node, meta, (and root counted as valid node) */ 3208 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count - 3209 F2FS_RESERVED_NODE_NUM; 3210 nm_i->nid_cnt[FREE_NID] = 0; 3211 nm_i->nid_cnt[PREALLOC_NID] = 0; 3212 nm_i->ram_thresh = DEF_RAM_THRESHOLD; 3213 nm_i->ra_nid_pages = DEF_RA_NID_PAGES; 3214 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD; 3215 3216 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC); 3217 INIT_LIST_HEAD(&nm_i->free_nid_list); 3218 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO); 3219 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO); 3220 INIT_LIST_HEAD(&nm_i->nat_entries); 3221 spin_lock_init(&nm_i->nat_list_lock); 3222 3223 mutex_init(&nm_i->build_lock); 3224 spin_lock_init(&nm_i->nid_list_lock); 3225 init_rwsem(&nm_i->nat_tree_lock); 3226 3227 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 3228 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 3229 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 3230 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size, 3231 GFP_KERNEL); 3232 if (!nm_i->nat_bitmap) 3233 return -ENOMEM; 3234 3235 err = __get_nat_bitmaps(sbi); 3236 if (err) 3237 return err; 3238 3239 #ifdef CONFIG_F2FS_CHECK_FS 3240 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size, 3241 GFP_KERNEL); 3242 if (!nm_i->nat_bitmap_mir) 3243 return -ENOMEM; 3244 #endif 3245 3246 return 0; 3247 } 3248 3249 static int init_free_nid_cache(struct f2fs_sb_info *sbi) 3250 { 3251 struct f2fs_nm_info *nm_i = NM_I(sbi); 3252 int i; 3253 3254 nm_i->free_nid_bitmap = 3255 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *), 3256 nm_i->nat_blocks), 3257 GFP_KERNEL); 3258 if (!nm_i->free_nid_bitmap) 3259 return -ENOMEM; 3260 3261 for (i = 0; i < nm_i->nat_blocks; i++) { 3262 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi, 3263 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL); 3264 if (!nm_i->free_nid_bitmap[i]) 3265 return -ENOMEM; 3266 } 3267 3268 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8, 3269 GFP_KERNEL); 3270 if (!nm_i->nat_block_bitmap) 3271 return -ENOMEM; 3272 3273 nm_i->free_nid_count = 3274 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short), 3275 nm_i->nat_blocks), 3276 GFP_KERNEL); 3277 if (!nm_i->free_nid_count) 3278 return -ENOMEM; 3279 return 0; 3280 } 3281 3282 int f2fs_build_node_manager(struct f2fs_sb_info *sbi) 3283 { 3284 int err; 3285 3286 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info), 3287 GFP_KERNEL); 3288 if (!sbi->nm_info) 3289 return -ENOMEM; 3290 3291 err = init_node_manager(sbi); 3292 if (err) 3293 return err; 3294 3295 err = init_free_nid_cache(sbi); 3296 if (err) 3297 return err; 3298 3299 /* load free nid status from nat_bits table */ 3300 load_free_nid_bitmap(sbi); 3301 3302 return f2fs_build_free_nids(sbi, true, true); 3303 } 3304 3305 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi) 3306 { 3307 struct f2fs_nm_info *nm_i = NM_I(sbi); 3308 struct free_nid *i, *next_i; 3309 struct nat_entry *natvec[NATVEC_SIZE]; 3310 struct nat_entry_set *setvec[SETVEC_SIZE]; 3311 nid_t nid = 0; 3312 unsigned int found; 3313 3314 if (!nm_i) 3315 return; 3316 3317 /* destroy free nid list */ 3318 spin_lock(&nm_i->nid_list_lock); 3319 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 3320 __remove_free_nid(sbi, i, FREE_NID); 3321 spin_unlock(&nm_i->nid_list_lock); 3322 kmem_cache_free(free_nid_slab, i); 3323 spin_lock(&nm_i->nid_list_lock); 3324 } 3325 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]); 3326 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]); 3327 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list)); 3328 spin_unlock(&nm_i->nid_list_lock); 3329 3330 /* destroy nat cache */ 3331 down_write(&nm_i->nat_tree_lock); 3332 while ((found = __gang_lookup_nat_cache(nm_i, 3333 nid, NATVEC_SIZE, natvec))) { 3334 unsigned idx; 3335 3336 nid = nat_get_nid(natvec[found - 1]) + 1; 3337 for (idx = 0; idx < found; idx++) { 3338 spin_lock(&nm_i->nat_list_lock); 3339 list_del(&natvec[idx]->list); 3340 spin_unlock(&nm_i->nat_list_lock); 3341 3342 __del_from_nat_cache(nm_i, natvec[idx]); 3343 } 3344 } 3345 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]); 3346 3347 /* destroy nat set cache */ 3348 nid = 0; 3349 while ((found = __gang_lookup_nat_set(nm_i, 3350 nid, SETVEC_SIZE, setvec))) { 3351 unsigned idx; 3352 3353 nid = setvec[found - 1]->set + 1; 3354 for (idx = 0; idx < found; idx++) { 3355 /* entry_cnt is not zero, when cp_error was occurred */ 3356 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list)); 3357 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set); 3358 kmem_cache_free(nat_entry_set_slab, setvec[idx]); 3359 } 3360 } 3361 up_write(&nm_i->nat_tree_lock); 3362 3363 kvfree(nm_i->nat_block_bitmap); 3364 if (nm_i->free_nid_bitmap) { 3365 int i; 3366 3367 for (i = 0; i < nm_i->nat_blocks; i++) 3368 kvfree(nm_i->free_nid_bitmap[i]); 3369 kvfree(nm_i->free_nid_bitmap); 3370 } 3371 kvfree(nm_i->free_nid_count); 3372 3373 kvfree(nm_i->nat_bitmap); 3374 kvfree(nm_i->nat_bits); 3375 #ifdef CONFIG_F2FS_CHECK_FS 3376 kvfree(nm_i->nat_bitmap_mir); 3377 #endif 3378 sbi->nm_info = NULL; 3379 kfree(nm_i); 3380 } 3381 3382 int __init f2fs_create_node_manager_caches(void) 3383 { 3384 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry", 3385 sizeof(struct nat_entry)); 3386 if (!nat_entry_slab) 3387 goto fail; 3388 3389 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid", 3390 sizeof(struct free_nid)); 3391 if (!free_nid_slab) 3392 goto destroy_nat_entry; 3393 3394 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set", 3395 sizeof(struct nat_entry_set)); 3396 if (!nat_entry_set_slab) 3397 goto destroy_free_nid; 3398 3399 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry", 3400 sizeof(struct fsync_node_entry)); 3401 if (!fsync_node_entry_slab) 3402 goto destroy_nat_entry_set; 3403 return 0; 3404 3405 destroy_nat_entry_set: 3406 kmem_cache_destroy(nat_entry_set_slab); 3407 destroy_free_nid: 3408 kmem_cache_destroy(free_nid_slab); 3409 destroy_nat_entry: 3410 kmem_cache_destroy(nat_entry_slab); 3411 fail: 3412 return -ENOMEM; 3413 } 3414 3415 void f2fs_destroy_node_manager_caches(void) 3416 { 3417 kmem_cache_destroy(fsync_node_entry_slab); 3418 kmem_cache_destroy(nat_entry_set_slab); 3419 kmem_cache_destroy(free_nid_slab); 3420 kmem_cache_destroy(nat_entry_slab); 3421 } 3422