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 "trace.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 sysinfo val; 48 unsigned long avail_ram; 49 unsigned long mem_size = 0; 50 bool res = false; 51 52 si_meminfo(&val); 53 54 /* only uses low memory */ 55 avail_ram = val.totalram - val.totalhigh; 56 57 /* 58 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively 59 */ 60 if (type == FREE_NIDS) { 61 mem_size = (nm_i->nid_cnt[FREE_NID] * 62 sizeof(struct free_nid)) >> PAGE_SHIFT; 63 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 64 } else if (type == NAT_ENTRIES) { 65 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >> 66 PAGE_SHIFT; 67 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 68 if (excess_cached_nats(sbi)) 69 res = false; 70 } else if (type == DIRTY_DENTS) { 71 if (sbi->sb->s_bdi->wb.dirty_exceeded) 72 return false; 73 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS); 74 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 75 } else if (type == INO_ENTRIES) { 76 int i; 77 78 for (i = 0; i < MAX_INO_ENTRY; i++) 79 mem_size += sbi->im[i].ino_num * 80 sizeof(struct ino_entry); 81 mem_size >>= PAGE_SHIFT; 82 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 83 } else if (type == EXTENT_CACHE) { 84 mem_size = (atomic_read(&sbi->total_ext_tree) * 85 sizeof(struct extent_tree) + 86 atomic_read(&sbi->total_ext_node) * 87 sizeof(struct extent_node)) >> PAGE_SHIFT; 88 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 89 } else if (type == INMEM_PAGES) { 90 /* it allows 20% / total_ram for inmemory pages */ 91 mem_size = get_pages(sbi, F2FS_INMEM_PAGES); 92 res = mem_size < (val.totalram / 5); 93 } else { 94 if (!sbi->sb->s_bdi->wb.dirty_exceeded) 95 return true; 96 } 97 return res; 98 } 99 100 static void clear_node_page_dirty(struct page *page) 101 { 102 if (PageDirty(page)) { 103 f2fs_clear_page_cache_dirty_tag(page); 104 clear_page_dirty_for_io(page); 105 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES); 106 } 107 ClearPageUptodate(page); 108 } 109 110 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 111 { 112 return f2fs_get_meta_page_nofail(sbi, current_nat_addr(sbi, nid)); 113 } 114 115 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 116 { 117 struct page *src_page; 118 struct page *dst_page; 119 pgoff_t dst_off; 120 void *src_addr; 121 void *dst_addr; 122 struct f2fs_nm_info *nm_i = NM_I(sbi); 123 124 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid)); 125 126 /* get current nat block page with lock */ 127 src_page = get_current_nat_page(sbi, nid); 128 if (IS_ERR(src_page)) 129 return src_page; 130 dst_page = f2fs_grab_meta_page(sbi, dst_off); 131 f2fs_bug_on(sbi, PageDirty(src_page)); 132 133 src_addr = page_address(src_page); 134 dst_addr = page_address(dst_page); 135 memcpy(dst_addr, src_addr, PAGE_SIZE); 136 set_page_dirty(dst_page); 137 f2fs_put_page(src_page, 1); 138 139 set_to_next_nat(nm_i, nid); 140 141 return dst_page; 142 } 143 144 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail) 145 { 146 struct nat_entry *new; 147 148 if (no_fail) 149 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO); 150 else 151 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO); 152 if (new) { 153 nat_set_nid(new, nid); 154 nat_reset_flag(new); 155 } 156 return new; 157 } 158 159 static void __free_nat_entry(struct nat_entry *e) 160 { 161 kmem_cache_free(nat_entry_slab, e); 162 } 163 164 /* must be locked by nat_tree_lock */ 165 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i, 166 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail) 167 { 168 if (no_fail) 169 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne); 170 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne)) 171 return NULL; 172 173 if (raw_ne) 174 node_info_from_raw_nat(&ne->ni, raw_ne); 175 176 spin_lock(&nm_i->nat_list_lock); 177 list_add_tail(&ne->list, &nm_i->nat_entries); 178 spin_unlock(&nm_i->nat_list_lock); 179 180 nm_i->nat_cnt++; 181 return ne; 182 } 183 184 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n) 185 { 186 struct nat_entry *ne; 187 188 ne = radix_tree_lookup(&nm_i->nat_root, n); 189 190 /* for recent accessed nat entry, move it to tail of lru list */ 191 if (ne && !get_nat_flag(ne, IS_DIRTY)) { 192 spin_lock(&nm_i->nat_list_lock); 193 if (!list_empty(&ne->list)) 194 list_move_tail(&ne->list, &nm_i->nat_entries); 195 spin_unlock(&nm_i->nat_list_lock); 196 } 197 198 return ne; 199 } 200 201 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i, 202 nid_t start, unsigned int nr, struct nat_entry **ep) 203 { 204 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr); 205 } 206 207 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e) 208 { 209 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e)); 210 nm_i->nat_cnt--; 211 __free_nat_entry(e); 212 } 213 214 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i, 215 struct nat_entry *ne) 216 { 217 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid); 218 struct nat_entry_set *head; 219 220 head = radix_tree_lookup(&nm_i->nat_set_root, set); 221 if (!head) { 222 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS); 223 224 INIT_LIST_HEAD(&head->entry_list); 225 INIT_LIST_HEAD(&head->set_list); 226 head->set = set; 227 head->entry_cnt = 0; 228 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head); 229 } 230 return head; 231 } 232 233 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i, 234 struct nat_entry *ne) 235 { 236 struct nat_entry_set *head; 237 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR; 238 239 if (!new_ne) 240 head = __grab_nat_entry_set(nm_i, ne); 241 242 /* 243 * update entry_cnt in below condition: 244 * 1. update NEW_ADDR to valid block address; 245 * 2. update old block address to new one; 246 */ 247 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) || 248 !get_nat_flag(ne, IS_DIRTY))) 249 head->entry_cnt++; 250 251 set_nat_flag(ne, IS_PREALLOC, new_ne); 252 253 if (get_nat_flag(ne, IS_DIRTY)) 254 goto refresh_list; 255 256 nm_i->dirty_nat_cnt++; 257 set_nat_flag(ne, IS_DIRTY, true); 258 refresh_list: 259 spin_lock(&nm_i->nat_list_lock); 260 if (new_ne) 261 list_del_init(&ne->list); 262 else 263 list_move_tail(&ne->list, &head->entry_list); 264 spin_unlock(&nm_i->nat_list_lock); 265 } 266 267 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i, 268 struct nat_entry_set *set, struct nat_entry *ne) 269 { 270 spin_lock(&nm_i->nat_list_lock); 271 list_move_tail(&ne->list, &nm_i->nat_entries); 272 spin_unlock(&nm_i->nat_list_lock); 273 274 set_nat_flag(ne, IS_DIRTY, false); 275 set->entry_cnt--; 276 nm_i->dirty_nat_cnt--; 277 } 278 279 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i, 280 nid_t start, unsigned int nr, struct nat_entry_set **ep) 281 { 282 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep, 283 start, nr); 284 } 285 286 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page) 287 { 288 return NODE_MAPPING(sbi) == page->mapping && 289 IS_DNODE(page) && is_cold_node(page); 290 } 291 292 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi) 293 { 294 spin_lock_init(&sbi->fsync_node_lock); 295 INIT_LIST_HEAD(&sbi->fsync_node_list); 296 sbi->fsync_seg_id = 0; 297 sbi->fsync_node_num = 0; 298 } 299 300 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi, 301 struct page *page) 302 { 303 struct fsync_node_entry *fn; 304 unsigned long flags; 305 unsigned int seq_id; 306 307 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS); 308 309 get_page(page); 310 fn->page = page; 311 INIT_LIST_HEAD(&fn->list); 312 313 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 314 list_add_tail(&fn->list, &sbi->fsync_node_list); 315 fn->seq_id = sbi->fsync_seg_id++; 316 seq_id = fn->seq_id; 317 sbi->fsync_node_num++; 318 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 319 320 return seq_id; 321 } 322 323 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page) 324 { 325 struct fsync_node_entry *fn; 326 unsigned long flags; 327 328 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 329 list_for_each_entry(fn, &sbi->fsync_node_list, list) { 330 if (fn->page == page) { 331 list_del(&fn->list); 332 sbi->fsync_node_num--; 333 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 334 kmem_cache_free(fsync_node_entry_slab, fn); 335 put_page(page); 336 return; 337 } 338 } 339 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 340 f2fs_bug_on(sbi, 1); 341 } 342 343 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi) 344 { 345 unsigned long flags; 346 347 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 348 sbi->fsync_seg_id = 0; 349 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 350 } 351 352 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid) 353 { 354 struct f2fs_nm_info *nm_i = NM_I(sbi); 355 struct nat_entry *e; 356 bool need = false; 357 358 down_read(&nm_i->nat_tree_lock); 359 e = __lookup_nat_cache(nm_i, nid); 360 if (e) { 361 if (!get_nat_flag(e, IS_CHECKPOINTED) && 362 !get_nat_flag(e, HAS_FSYNCED_INODE)) 363 need = true; 364 } 365 up_read(&nm_i->nat_tree_lock); 366 return need; 367 } 368 369 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid) 370 { 371 struct f2fs_nm_info *nm_i = NM_I(sbi); 372 struct nat_entry *e; 373 bool is_cp = true; 374 375 down_read(&nm_i->nat_tree_lock); 376 e = __lookup_nat_cache(nm_i, nid); 377 if (e && !get_nat_flag(e, IS_CHECKPOINTED)) 378 is_cp = false; 379 up_read(&nm_i->nat_tree_lock); 380 return is_cp; 381 } 382 383 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino) 384 { 385 struct f2fs_nm_info *nm_i = NM_I(sbi); 386 struct nat_entry *e; 387 bool need_update = true; 388 389 down_read(&nm_i->nat_tree_lock); 390 e = __lookup_nat_cache(nm_i, ino); 391 if (e && get_nat_flag(e, HAS_LAST_FSYNC) && 392 (get_nat_flag(e, IS_CHECKPOINTED) || 393 get_nat_flag(e, HAS_FSYNCED_INODE))) 394 need_update = false; 395 up_read(&nm_i->nat_tree_lock); 396 return need_update; 397 } 398 399 /* must be locked by nat_tree_lock */ 400 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid, 401 struct f2fs_nat_entry *ne) 402 { 403 struct f2fs_nm_info *nm_i = NM_I(sbi); 404 struct nat_entry *new, *e; 405 406 new = __alloc_nat_entry(nid, false); 407 if (!new) 408 return; 409 410 down_write(&nm_i->nat_tree_lock); 411 e = __lookup_nat_cache(nm_i, nid); 412 if (!e) 413 e = __init_nat_entry(nm_i, new, ne, false); 414 else 415 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) || 416 nat_get_blkaddr(e) != 417 le32_to_cpu(ne->block_addr) || 418 nat_get_version(e) != ne->version); 419 up_write(&nm_i->nat_tree_lock); 420 if (e != new) 421 __free_nat_entry(new); 422 } 423 424 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni, 425 block_t new_blkaddr, bool fsync_done) 426 { 427 struct f2fs_nm_info *nm_i = NM_I(sbi); 428 struct nat_entry *e; 429 struct nat_entry *new = __alloc_nat_entry(ni->nid, true); 430 431 down_write(&nm_i->nat_tree_lock); 432 e = __lookup_nat_cache(nm_i, ni->nid); 433 if (!e) { 434 e = __init_nat_entry(nm_i, new, NULL, true); 435 copy_node_info(&e->ni, ni); 436 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR); 437 } else if (new_blkaddr == NEW_ADDR) { 438 /* 439 * when nid is reallocated, 440 * previous nat entry can be remained in nat cache. 441 * So, reinitialize it with new information. 442 */ 443 copy_node_info(&e->ni, ni); 444 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR); 445 } 446 /* let's free early to reduce memory consumption */ 447 if (e != new) 448 __free_nat_entry(new); 449 450 /* sanity check */ 451 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr); 452 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR && 453 new_blkaddr == NULL_ADDR); 454 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR && 455 new_blkaddr == NEW_ADDR); 456 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) && 457 new_blkaddr == NEW_ADDR); 458 459 /* increment version no as node is removed */ 460 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) { 461 unsigned char version = nat_get_version(e); 462 nat_set_version(e, inc_node_version(version)); 463 } 464 465 /* change address */ 466 nat_set_blkaddr(e, new_blkaddr); 467 if (!__is_valid_data_blkaddr(new_blkaddr)) 468 set_nat_flag(e, IS_CHECKPOINTED, false); 469 __set_nat_cache_dirty(nm_i, e); 470 471 /* update fsync_mark if its inode nat entry is still alive */ 472 if (ni->nid != ni->ino) 473 e = __lookup_nat_cache(nm_i, ni->ino); 474 if (e) { 475 if (fsync_done && ni->nid == ni->ino) 476 set_nat_flag(e, HAS_FSYNCED_INODE, true); 477 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done); 478 } 479 up_write(&nm_i->nat_tree_lock); 480 } 481 482 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink) 483 { 484 struct f2fs_nm_info *nm_i = NM_I(sbi); 485 int nr = nr_shrink; 486 487 if (!down_write_trylock(&nm_i->nat_tree_lock)) 488 return 0; 489 490 spin_lock(&nm_i->nat_list_lock); 491 while (nr_shrink) { 492 struct nat_entry *ne; 493 494 if (list_empty(&nm_i->nat_entries)) 495 break; 496 497 ne = list_first_entry(&nm_i->nat_entries, 498 struct nat_entry, list); 499 list_del(&ne->list); 500 spin_unlock(&nm_i->nat_list_lock); 501 502 __del_from_nat_cache(nm_i, ne); 503 nr_shrink--; 504 505 spin_lock(&nm_i->nat_list_lock); 506 } 507 spin_unlock(&nm_i->nat_list_lock); 508 509 up_write(&nm_i->nat_tree_lock); 510 return nr - nr_shrink; 511 } 512 513 /* 514 * This function always returns success 515 */ 516 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid, 517 struct node_info *ni) 518 { 519 struct f2fs_nm_info *nm_i = NM_I(sbi); 520 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 521 struct f2fs_journal *journal = curseg->journal; 522 nid_t start_nid = START_NID(nid); 523 struct f2fs_nat_block *nat_blk; 524 struct page *page = NULL; 525 struct f2fs_nat_entry ne; 526 struct nat_entry *e; 527 pgoff_t index; 528 block_t blkaddr; 529 int i; 530 531 ni->nid = nid; 532 533 /* Check nat cache */ 534 down_read(&nm_i->nat_tree_lock); 535 e = __lookup_nat_cache(nm_i, nid); 536 if (e) { 537 ni->ino = nat_get_ino(e); 538 ni->blk_addr = nat_get_blkaddr(e); 539 ni->version = nat_get_version(e); 540 up_read(&nm_i->nat_tree_lock); 541 return 0; 542 } 543 544 memset(&ne, 0, sizeof(struct f2fs_nat_entry)); 545 546 /* Check current segment summary */ 547 down_read(&curseg->journal_rwsem); 548 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0); 549 if (i >= 0) { 550 ne = nat_in_journal(journal, i); 551 node_info_from_raw_nat(ni, &ne); 552 } 553 up_read(&curseg->journal_rwsem); 554 if (i >= 0) { 555 up_read(&nm_i->nat_tree_lock); 556 goto cache; 557 } 558 559 /* Fill node_info from nat page */ 560 index = current_nat_addr(sbi, nid); 561 up_read(&nm_i->nat_tree_lock); 562 563 page = f2fs_get_meta_page(sbi, index); 564 if (IS_ERR(page)) 565 return PTR_ERR(page); 566 567 nat_blk = (struct f2fs_nat_block *)page_address(page); 568 ne = nat_blk->entries[nid - start_nid]; 569 node_info_from_raw_nat(ni, &ne); 570 f2fs_put_page(page, 1); 571 cache: 572 blkaddr = le32_to_cpu(ne.block_addr); 573 if (__is_valid_data_blkaddr(blkaddr) && 574 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) 575 return -EFAULT; 576 577 /* cache nat entry */ 578 cache_nat_entry(sbi, nid, &ne); 579 return 0; 580 } 581 582 /* 583 * readahead MAX_RA_NODE number of node pages. 584 */ 585 static void f2fs_ra_node_pages(struct page *parent, int start, int n) 586 { 587 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 588 struct blk_plug plug; 589 int i, end; 590 nid_t nid; 591 592 blk_start_plug(&plug); 593 594 /* Then, try readahead for siblings of the desired node */ 595 end = start + n; 596 end = min(end, NIDS_PER_BLOCK); 597 for (i = start; i < end; i++) { 598 nid = get_nid(parent, i, false); 599 f2fs_ra_node_page(sbi, nid); 600 } 601 602 blk_finish_plug(&plug); 603 } 604 605 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs) 606 { 607 const long direct_index = ADDRS_PER_INODE(dn->inode); 608 const long direct_blks = ADDRS_PER_BLOCK(dn->inode); 609 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK; 610 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode); 611 int cur_level = dn->cur_level; 612 int max_level = dn->max_level; 613 pgoff_t base = 0; 614 615 if (!dn->max_level) 616 return pgofs + 1; 617 618 while (max_level-- > cur_level) 619 skipped_unit *= NIDS_PER_BLOCK; 620 621 switch (dn->max_level) { 622 case 3: 623 base += 2 * indirect_blks; 624 /* fall through */ 625 case 2: 626 base += 2 * direct_blks; 627 /* fall through */ 628 case 1: 629 base += direct_index; 630 break; 631 default: 632 f2fs_bug_on(F2FS_I_SB(dn->inode), 1); 633 } 634 635 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base; 636 } 637 638 /* 639 * The maximum depth is four. 640 * Offset[0] will have raw inode offset. 641 */ 642 static int get_node_path(struct inode *inode, long block, 643 int offset[4], unsigned int noffset[4]) 644 { 645 const long direct_index = ADDRS_PER_INODE(inode); 646 const long direct_blks = ADDRS_PER_BLOCK(inode); 647 const long dptrs_per_blk = NIDS_PER_BLOCK; 648 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK; 649 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK; 650 int n = 0; 651 int level = 0; 652 653 noffset[0] = 0; 654 655 if (block < direct_index) { 656 offset[n] = block; 657 goto got; 658 } 659 block -= direct_index; 660 if (block < direct_blks) { 661 offset[n++] = NODE_DIR1_BLOCK; 662 noffset[n] = 1; 663 offset[n] = block; 664 level = 1; 665 goto got; 666 } 667 block -= direct_blks; 668 if (block < direct_blks) { 669 offset[n++] = NODE_DIR2_BLOCK; 670 noffset[n] = 2; 671 offset[n] = block; 672 level = 1; 673 goto got; 674 } 675 block -= direct_blks; 676 if (block < indirect_blks) { 677 offset[n++] = NODE_IND1_BLOCK; 678 noffset[n] = 3; 679 offset[n++] = block / direct_blks; 680 noffset[n] = 4 + offset[n - 1]; 681 offset[n] = block % direct_blks; 682 level = 2; 683 goto got; 684 } 685 block -= indirect_blks; 686 if (block < indirect_blks) { 687 offset[n++] = NODE_IND2_BLOCK; 688 noffset[n] = 4 + dptrs_per_blk; 689 offset[n++] = block / direct_blks; 690 noffset[n] = 5 + dptrs_per_blk + offset[n - 1]; 691 offset[n] = block % direct_blks; 692 level = 2; 693 goto got; 694 } 695 block -= indirect_blks; 696 if (block < dindirect_blks) { 697 offset[n++] = NODE_DIND_BLOCK; 698 noffset[n] = 5 + (dptrs_per_blk * 2); 699 offset[n++] = block / indirect_blks; 700 noffset[n] = 6 + (dptrs_per_blk * 2) + 701 offset[n - 1] * (dptrs_per_blk + 1); 702 offset[n++] = (block / direct_blks) % dptrs_per_blk; 703 noffset[n] = 7 + (dptrs_per_blk * 2) + 704 offset[n - 2] * (dptrs_per_blk + 1) + 705 offset[n - 1]; 706 offset[n] = block % direct_blks; 707 level = 3; 708 goto got; 709 } else { 710 return -E2BIG; 711 } 712 got: 713 return level; 714 } 715 716 /* 717 * Caller should call f2fs_put_dnode(dn). 718 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and 719 * f2fs_unlock_op() only if ro is not set RDONLY_NODE. 720 * In the case of RDONLY_NODE, we don't need to care about mutex. 721 */ 722 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode) 723 { 724 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 725 struct page *npage[4]; 726 struct page *parent = NULL; 727 int offset[4]; 728 unsigned int noffset[4]; 729 nid_t nids[4]; 730 int level, i = 0; 731 int err = 0; 732 733 level = get_node_path(dn->inode, index, offset, noffset); 734 if (level < 0) 735 return level; 736 737 nids[0] = dn->inode->i_ino; 738 npage[0] = dn->inode_page; 739 740 if (!npage[0]) { 741 npage[0] = f2fs_get_node_page(sbi, nids[0]); 742 if (IS_ERR(npage[0])) 743 return PTR_ERR(npage[0]); 744 } 745 746 /* if inline_data is set, should not report any block indices */ 747 if (f2fs_has_inline_data(dn->inode) && index) { 748 err = -ENOENT; 749 f2fs_put_page(npage[0], 1); 750 goto release_out; 751 } 752 753 parent = npage[0]; 754 if (level != 0) 755 nids[1] = get_nid(parent, offset[0], true); 756 dn->inode_page = npage[0]; 757 dn->inode_page_locked = true; 758 759 /* get indirect or direct nodes */ 760 for (i = 1; i <= level; i++) { 761 bool done = false; 762 763 if (!nids[i] && mode == ALLOC_NODE) { 764 /* alloc new node */ 765 if (!f2fs_alloc_nid(sbi, &(nids[i]))) { 766 err = -ENOSPC; 767 goto release_pages; 768 } 769 770 dn->nid = nids[i]; 771 npage[i] = f2fs_new_node_page(dn, noffset[i]); 772 if (IS_ERR(npage[i])) { 773 f2fs_alloc_nid_failed(sbi, nids[i]); 774 err = PTR_ERR(npage[i]); 775 goto release_pages; 776 } 777 778 set_nid(parent, offset[i - 1], nids[i], i == 1); 779 f2fs_alloc_nid_done(sbi, nids[i]); 780 done = true; 781 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) { 782 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]); 783 if (IS_ERR(npage[i])) { 784 err = PTR_ERR(npage[i]); 785 goto release_pages; 786 } 787 done = true; 788 } 789 if (i == 1) { 790 dn->inode_page_locked = false; 791 unlock_page(parent); 792 } else { 793 f2fs_put_page(parent, 1); 794 } 795 796 if (!done) { 797 npage[i] = f2fs_get_node_page(sbi, nids[i]); 798 if (IS_ERR(npage[i])) { 799 err = PTR_ERR(npage[i]); 800 f2fs_put_page(npage[0], 0); 801 goto release_out; 802 } 803 } 804 if (i < level) { 805 parent = npage[i]; 806 nids[i + 1] = get_nid(parent, offset[i], false); 807 } 808 } 809 dn->nid = nids[level]; 810 dn->ofs_in_node = offset[level]; 811 dn->node_page = npage[level]; 812 dn->data_blkaddr = datablock_addr(dn->inode, 813 dn->node_page, dn->ofs_in_node); 814 return 0; 815 816 release_pages: 817 f2fs_put_page(parent, 1); 818 if (i > 1) 819 f2fs_put_page(npage[0], 0); 820 release_out: 821 dn->inode_page = NULL; 822 dn->node_page = NULL; 823 if (err == -ENOENT) { 824 dn->cur_level = i; 825 dn->max_level = level; 826 dn->ofs_in_node = offset[level]; 827 } 828 return err; 829 } 830 831 static int truncate_node(struct dnode_of_data *dn) 832 { 833 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 834 struct node_info ni; 835 int err; 836 pgoff_t index; 837 838 err = f2fs_get_node_info(sbi, dn->nid, &ni); 839 if (err) 840 return err; 841 842 /* Deallocate node address */ 843 f2fs_invalidate_blocks(sbi, ni.blk_addr); 844 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino); 845 set_node_addr(sbi, &ni, NULL_ADDR, false); 846 847 if (dn->nid == dn->inode->i_ino) { 848 f2fs_remove_orphan_inode(sbi, dn->nid); 849 dec_valid_inode_count(sbi); 850 f2fs_inode_synced(dn->inode); 851 } 852 853 clear_node_page_dirty(dn->node_page); 854 set_sbi_flag(sbi, SBI_IS_DIRTY); 855 856 index = dn->node_page->index; 857 f2fs_put_page(dn->node_page, 1); 858 859 invalidate_mapping_pages(NODE_MAPPING(sbi), 860 index, index); 861 862 dn->node_page = NULL; 863 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr); 864 865 return 0; 866 } 867 868 static int truncate_dnode(struct dnode_of_data *dn) 869 { 870 struct page *page; 871 int err; 872 873 if (dn->nid == 0) 874 return 1; 875 876 /* get direct node */ 877 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 878 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT) 879 return 1; 880 else if (IS_ERR(page)) 881 return PTR_ERR(page); 882 883 /* Make dnode_of_data for parameter */ 884 dn->node_page = page; 885 dn->ofs_in_node = 0; 886 f2fs_truncate_data_blocks(dn); 887 err = truncate_node(dn); 888 if (err) 889 return err; 890 891 return 1; 892 } 893 894 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs, 895 int ofs, int depth) 896 { 897 struct dnode_of_data rdn = *dn; 898 struct page *page; 899 struct f2fs_node *rn; 900 nid_t child_nid; 901 unsigned int child_nofs; 902 int freed = 0; 903 int i, ret; 904 905 if (dn->nid == 0) 906 return NIDS_PER_BLOCK + 1; 907 908 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr); 909 910 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 911 if (IS_ERR(page)) { 912 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page)); 913 return PTR_ERR(page); 914 } 915 916 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK); 917 918 rn = F2FS_NODE(page); 919 if (depth < 3) { 920 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) { 921 child_nid = le32_to_cpu(rn->in.nid[i]); 922 if (child_nid == 0) 923 continue; 924 rdn.nid = child_nid; 925 ret = truncate_dnode(&rdn); 926 if (ret < 0) 927 goto out_err; 928 if (set_nid(page, i, 0, false)) 929 dn->node_changed = true; 930 } 931 } else { 932 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1; 933 for (i = ofs; i < NIDS_PER_BLOCK; i++) { 934 child_nid = le32_to_cpu(rn->in.nid[i]); 935 if (child_nid == 0) { 936 child_nofs += NIDS_PER_BLOCK + 1; 937 continue; 938 } 939 rdn.nid = child_nid; 940 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1); 941 if (ret == (NIDS_PER_BLOCK + 1)) { 942 if (set_nid(page, i, 0, false)) 943 dn->node_changed = true; 944 child_nofs += ret; 945 } else if (ret < 0 && ret != -ENOENT) { 946 goto out_err; 947 } 948 } 949 freed = child_nofs; 950 } 951 952 if (!ofs) { 953 /* remove current indirect node */ 954 dn->node_page = page; 955 ret = truncate_node(dn); 956 if (ret) 957 goto out_err; 958 freed++; 959 } else { 960 f2fs_put_page(page, 1); 961 } 962 trace_f2fs_truncate_nodes_exit(dn->inode, freed); 963 return freed; 964 965 out_err: 966 f2fs_put_page(page, 1); 967 trace_f2fs_truncate_nodes_exit(dn->inode, ret); 968 return ret; 969 } 970 971 static int truncate_partial_nodes(struct dnode_of_data *dn, 972 struct f2fs_inode *ri, int *offset, int depth) 973 { 974 struct page *pages[2]; 975 nid_t nid[3]; 976 nid_t child_nid; 977 int err = 0; 978 int i; 979 int idx = depth - 2; 980 981 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 982 if (!nid[0]) 983 return 0; 984 985 /* get indirect nodes in the path */ 986 for (i = 0; i < idx + 1; i++) { 987 /* reference count'll be increased */ 988 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]); 989 if (IS_ERR(pages[i])) { 990 err = PTR_ERR(pages[i]); 991 idx = i - 1; 992 goto fail; 993 } 994 nid[i + 1] = get_nid(pages[i], offset[i + 1], false); 995 } 996 997 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK); 998 999 /* free direct nodes linked to a partial indirect node */ 1000 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) { 1001 child_nid = get_nid(pages[idx], i, false); 1002 if (!child_nid) 1003 continue; 1004 dn->nid = child_nid; 1005 err = truncate_dnode(dn); 1006 if (err < 0) 1007 goto fail; 1008 if (set_nid(pages[idx], i, 0, false)) 1009 dn->node_changed = true; 1010 } 1011 1012 if (offset[idx + 1] == 0) { 1013 dn->node_page = pages[idx]; 1014 dn->nid = nid[idx]; 1015 err = truncate_node(dn); 1016 if (err) 1017 goto fail; 1018 } else { 1019 f2fs_put_page(pages[idx], 1); 1020 } 1021 offset[idx]++; 1022 offset[idx + 1] = 0; 1023 idx--; 1024 fail: 1025 for (i = idx; i >= 0; i--) 1026 f2fs_put_page(pages[i], 1); 1027 1028 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err); 1029 1030 return err; 1031 } 1032 1033 /* 1034 * All the block addresses of data and nodes should be nullified. 1035 */ 1036 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from) 1037 { 1038 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1039 int err = 0, cont = 1; 1040 int level, offset[4], noffset[4]; 1041 unsigned int nofs = 0; 1042 struct f2fs_inode *ri; 1043 struct dnode_of_data dn; 1044 struct page *page; 1045 1046 trace_f2fs_truncate_inode_blocks_enter(inode, from); 1047 1048 level = get_node_path(inode, from, offset, noffset); 1049 if (level < 0) 1050 return level; 1051 1052 page = f2fs_get_node_page(sbi, inode->i_ino); 1053 if (IS_ERR(page)) { 1054 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page)); 1055 return PTR_ERR(page); 1056 } 1057 1058 set_new_dnode(&dn, inode, page, NULL, 0); 1059 unlock_page(page); 1060 1061 ri = F2FS_INODE(page); 1062 switch (level) { 1063 case 0: 1064 case 1: 1065 nofs = noffset[1]; 1066 break; 1067 case 2: 1068 nofs = noffset[1]; 1069 if (!offset[level - 1]) 1070 goto skip_partial; 1071 err = truncate_partial_nodes(&dn, ri, offset, level); 1072 if (err < 0 && err != -ENOENT) 1073 goto fail; 1074 nofs += 1 + NIDS_PER_BLOCK; 1075 break; 1076 case 3: 1077 nofs = 5 + 2 * NIDS_PER_BLOCK; 1078 if (!offset[level - 1]) 1079 goto skip_partial; 1080 err = truncate_partial_nodes(&dn, ri, offset, level); 1081 if (err < 0 && err != -ENOENT) 1082 goto fail; 1083 break; 1084 default: 1085 BUG(); 1086 } 1087 1088 skip_partial: 1089 while (cont) { 1090 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1091 switch (offset[0]) { 1092 case NODE_DIR1_BLOCK: 1093 case NODE_DIR2_BLOCK: 1094 err = truncate_dnode(&dn); 1095 break; 1096 1097 case NODE_IND1_BLOCK: 1098 case NODE_IND2_BLOCK: 1099 err = truncate_nodes(&dn, nofs, offset[1], 2); 1100 break; 1101 1102 case NODE_DIND_BLOCK: 1103 err = truncate_nodes(&dn, nofs, offset[1], 3); 1104 cont = 0; 1105 break; 1106 1107 default: 1108 BUG(); 1109 } 1110 if (err < 0 && err != -ENOENT) 1111 goto fail; 1112 if (offset[1] == 0 && 1113 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) { 1114 lock_page(page); 1115 BUG_ON(page->mapping != NODE_MAPPING(sbi)); 1116 f2fs_wait_on_page_writeback(page, NODE, true, true); 1117 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0; 1118 set_page_dirty(page); 1119 unlock_page(page); 1120 } 1121 offset[1] = 0; 1122 offset[0]++; 1123 nofs += err; 1124 } 1125 fail: 1126 f2fs_put_page(page, 0); 1127 trace_f2fs_truncate_inode_blocks_exit(inode, err); 1128 return err > 0 ? 0 : err; 1129 } 1130 1131 /* caller must lock inode page */ 1132 int f2fs_truncate_xattr_node(struct inode *inode) 1133 { 1134 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1135 nid_t nid = F2FS_I(inode)->i_xattr_nid; 1136 struct dnode_of_data dn; 1137 struct page *npage; 1138 int err; 1139 1140 if (!nid) 1141 return 0; 1142 1143 npage = f2fs_get_node_page(sbi, nid); 1144 if (IS_ERR(npage)) 1145 return PTR_ERR(npage); 1146 1147 set_new_dnode(&dn, inode, NULL, npage, nid); 1148 err = truncate_node(&dn); 1149 if (err) { 1150 f2fs_put_page(npage, 1); 1151 return err; 1152 } 1153 1154 f2fs_i_xnid_write(inode, 0); 1155 1156 return 0; 1157 } 1158 1159 /* 1160 * Caller should grab and release a rwsem by calling f2fs_lock_op() and 1161 * f2fs_unlock_op(). 1162 */ 1163 int f2fs_remove_inode_page(struct inode *inode) 1164 { 1165 struct dnode_of_data dn; 1166 int err; 1167 1168 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1169 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE); 1170 if (err) 1171 return err; 1172 1173 err = f2fs_truncate_xattr_node(inode); 1174 if (err) { 1175 f2fs_put_dnode(&dn); 1176 return err; 1177 } 1178 1179 /* remove potential inline_data blocks */ 1180 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1181 S_ISLNK(inode->i_mode)) 1182 f2fs_truncate_data_blocks_range(&dn, 1); 1183 1184 /* 0 is possible, after f2fs_new_inode() has failed */ 1185 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) { 1186 f2fs_put_dnode(&dn); 1187 return -EIO; 1188 } 1189 1190 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) { 1191 f2fs_warn(F2FS_I_SB(inode), "Inconsistent i_blocks, ino:%lu, iblocks:%llu", 1192 inode->i_ino, (unsigned long long)inode->i_blocks); 1193 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK); 1194 } 1195 1196 /* will put inode & node pages */ 1197 err = truncate_node(&dn); 1198 if (err) { 1199 f2fs_put_dnode(&dn); 1200 return err; 1201 } 1202 return 0; 1203 } 1204 1205 struct page *f2fs_new_inode_page(struct inode *inode) 1206 { 1207 struct dnode_of_data dn; 1208 1209 /* allocate inode page for new inode */ 1210 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1211 1212 /* caller should f2fs_put_page(page, 1); */ 1213 return f2fs_new_node_page(&dn, 0); 1214 } 1215 1216 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs) 1217 { 1218 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1219 struct node_info new_ni; 1220 struct page *page; 1221 int err; 1222 1223 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1224 return ERR_PTR(-EPERM); 1225 1226 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false); 1227 if (!page) 1228 return ERR_PTR(-ENOMEM); 1229 1230 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs)))) 1231 goto fail; 1232 1233 #ifdef CONFIG_F2FS_CHECK_FS 1234 err = f2fs_get_node_info(sbi, dn->nid, &new_ni); 1235 if (err) { 1236 dec_valid_node_count(sbi, dn->inode, !ofs); 1237 goto fail; 1238 } 1239 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR); 1240 #endif 1241 new_ni.nid = dn->nid; 1242 new_ni.ino = dn->inode->i_ino; 1243 new_ni.blk_addr = NULL_ADDR; 1244 new_ni.flag = 0; 1245 new_ni.version = 0; 1246 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 1247 1248 f2fs_wait_on_page_writeback(page, NODE, true, true); 1249 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 1250 set_cold_node(page, S_ISDIR(dn->inode->i_mode)); 1251 if (!PageUptodate(page)) 1252 SetPageUptodate(page); 1253 if (set_page_dirty(page)) 1254 dn->node_changed = true; 1255 1256 if (f2fs_has_xattr_block(ofs)) 1257 f2fs_i_xnid_write(dn->inode, dn->nid); 1258 1259 if (ofs == 0) 1260 inc_valid_inode_count(sbi); 1261 return page; 1262 1263 fail: 1264 clear_node_page_dirty(page); 1265 f2fs_put_page(page, 1); 1266 return ERR_PTR(err); 1267 } 1268 1269 /* 1270 * Caller should do after getting the following values. 1271 * 0: f2fs_put_page(page, 0) 1272 * LOCKED_PAGE or error: f2fs_put_page(page, 1) 1273 */ 1274 static int read_node_page(struct page *page, int op_flags) 1275 { 1276 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1277 struct node_info ni; 1278 struct f2fs_io_info fio = { 1279 .sbi = sbi, 1280 .type = NODE, 1281 .op = REQ_OP_READ, 1282 .op_flags = op_flags, 1283 .page = page, 1284 .encrypted_page = NULL, 1285 }; 1286 int err; 1287 1288 if (PageUptodate(page)) { 1289 if (!f2fs_inode_chksum_verify(sbi, page)) { 1290 ClearPageUptodate(page); 1291 return -EFSBADCRC; 1292 } 1293 return LOCKED_PAGE; 1294 } 1295 1296 err = f2fs_get_node_info(sbi, page->index, &ni); 1297 if (err) 1298 return err; 1299 1300 if (unlikely(ni.blk_addr == NULL_ADDR) || 1301 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) { 1302 ClearPageUptodate(page); 1303 return -ENOENT; 1304 } 1305 1306 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr; 1307 return f2fs_submit_page_bio(&fio); 1308 } 1309 1310 /* 1311 * Readahead a node page 1312 */ 1313 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 1314 { 1315 struct page *apage; 1316 int err; 1317 1318 if (!nid) 1319 return; 1320 if (f2fs_check_nid_range(sbi, nid)) 1321 return; 1322 1323 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid); 1324 if (apage) 1325 return; 1326 1327 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1328 if (!apage) 1329 return; 1330 1331 err = read_node_page(apage, REQ_RAHEAD); 1332 f2fs_put_page(apage, err ? 1 : 0); 1333 } 1334 1335 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid, 1336 struct page *parent, int start) 1337 { 1338 struct page *page; 1339 int err; 1340 1341 if (!nid) 1342 return ERR_PTR(-ENOENT); 1343 if (f2fs_check_nid_range(sbi, nid)) 1344 return ERR_PTR(-EINVAL); 1345 repeat: 1346 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1347 if (!page) 1348 return ERR_PTR(-ENOMEM); 1349 1350 err = read_node_page(page, 0); 1351 if (err < 0) { 1352 f2fs_put_page(page, 1); 1353 return ERR_PTR(err); 1354 } else if (err == LOCKED_PAGE) { 1355 err = 0; 1356 goto page_hit; 1357 } 1358 1359 if (parent) 1360 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE); 1361 1362 lock_page(page); 1363 1364 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1365 f2fs_put_page(page, 1); 1366 goto repeat; 1367 } 1368 1369 if (unlikely(!PageUptodate(page))) { 1370 err = -EIO; 1371 goto out_err; 1372 } 1373 1374 if (!f2fs_inode_chksum_verify(sbi, page)) { 1375 err = -EFSBADCRC; 1376 goto out_err; 1377 } 1378 page_hit: 1379 if(unlikely(nid != nid_of_node(page))) { 1380 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]", 1381 nid, nid_of_node(page), ino_of_node(page), 1382 ofs_of_node(page), cpver_of_node(page), 1383 next_blkaddr_of_node(page)); 1384 err = -EINVAL; 1385 out_err: 1386 ClearPageUptodate(page); 1387 f2fs_put_page(page, 1); 1388 return ERR_PTR(err); 1389 } 1390 return page; 1391 } 1392 1393 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 1394 { 1395 return __get_node_page(sbi, nid, NULL, 0); 1396 } 1397 1398 struct page *f2fs_get_node_page_ra(struct page *parent, int start) 1399 { 1400 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 1401 nid_t nid = get_nid(parent, start, false); 1402 1403 return __get_node_page(sbi, nid, parent, start); 1404 } 1405 1406 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino) 1407 { 1408 struct inode *inode; 1409 struct page *page; 1410 int ret; 1411 1412 /* should flush inline_data before evict_inode */ 1413 inode = ilookup(sbi->sb, ino); 1414 if (!inode) 1415 return; 1416 1417 page = f2fs_pagecache_get_page(inode->i_mapping, 0, 1418 FGP_LOCK|FGP_NOWAIT, 0); 1419 if (!page) 1420 goto iput_out; 1421 1422 if (!PageUptodate(page)) 1423 goto page_out; 1424 1425 if (!PageDirty(page)) 1426 goto page_out; 1427 1428 if (!clear_page_dirty_for_io(page)) 1429 goto page_out; 1430 1431 ret = f2fs_write_inline_data(inode, page); 1432 inode_dec_dirty_pages(inode); 1433 f2fs_remove_dirty_inode(inode); 1434 if (ret) 1435 set_page_dirty(page); 1436 page_out: 1437 f2fs_put_page(page, 1); 1438 iput_out: 1439 iput(inode); 1440 } 1441 1442 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino) 1443 { 1444 pgoff_t index; 1445 struct pagevec pvec; 1446 struct page *last_page = NULL; 1447 int nr_pages; 1448 1449 pagevec_init(&pvec); 1450 index = 0; 1451 1452 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1453 PAGECACHE_TAG_DIRTY))) { 1454 int i; 1455 1456 for (i = 0; i < nr_pages; i++) { 1457 struct page *page = pvec.pages[i]; 1458 1459 if (unlikely(f2fs_cp_error(sbi))) { 1460 f2fs_put_page(last_page, 0); 1461 pagevec_release(&pvec); 1462 return ERR_PTR(-EIO); 1463 } 1464 1465 if (!IS_DNODE(page) || !is_cold_node(page)) 1466 continue; 1467 if (ino_of_node(page) != ino) 1468 continue; 1469 1470 lock_page(page); 1471 1472 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1473 continue_unlock: 1474 unlock_page(page); 1475 continue; 1476 } 1477 if (ino_of_node(page) != ino) 1478 goto continue_unlock; 1479 1480 if (!PageDirty(page)) { 1481 /* someone wrote it for us */ 1482 goto continue_unlock; 1483 } 1484 1485 if (last_page) 1486 f2fs_put_page(last_page, 0); 1487 1488 get_page(page); 1489 last_page = page; 1490 unlock_page(page); 1491 } 1492 pagevec_release(&pvec); 1493 cond_resched(); 1494 } 1495 return last_page; 1496 } 1497 1498 static int __write_node_page(struct page *page, bool atomic, bool *submitted, 1499 struct writeback_control *wbc, bool do_balance, 1500 enum iostat_type io_type, unsigned int *seq_id) 1501 { 1502 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1503 nid_t nid; 1504 struct node_info ni; 1505 struct f2fs_io_info fio = { 1506 .sbi = sbi, 1507 .ino = ino_of_node(page), 1508 .type = NODE, 1509 .op = REQ_OP_WRITE, 1510 .op_flags = wbc_to_write_flags(wbc), 1511 .page = page, 1512 .encrypted_page = NULL, 1513 .submitted = false, 1514 .io_type = io_type, 1515 .io_wbc = wbc, 1516 }; 1517 unsigned int seq; 1518 1519 trace_f2fs_writepage(page, NODE); 1520 1521 if (unlikely(f2fs_cp_error(sbi))) 1522 goto redirty_out; 1523 1524 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1525 goto redirty_out; 1526 1527 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 1528 wbc->sync_mode == WB_SYNC_NONE && 1529 IS_DNODE(page) && is_cold_node(page)) 1530 goto redirty_out; 1531 1532 /* get old block addr of this node page */ 1533 nid = nid_of_node(page); 1534 f2fs_bug_on(sbi, page->index != nid); 1535 1536 if (f2fs_get_node_info(sbi, nid, &ni)) 1537 goto redirty_out; 1538 1539 if (wbc->for_reclaim) { 1540 if (!down_read_trylock(&sbi->node_write)) 1541 goto redirty_out; 1542 } else { 1543 down_read(&sbi->node_write); 1544 } 1545 1546 /* This page is already truncated */ 1547 if (unlikely(ni.blk_addr == NULL_ADDR)) { 1548 ClearPageUptodate(page); 1549 dec_page_count(sbi, F2FS_DIRTY_NODES); 1550 up_read(&sbi->node_write); 1551 unlock_page(page); 1552 return 0; 1553 } 1554 1555 if (__is_valid_data_blkaddr(ni.blk_addr) && 1556 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, 1557 DATA_GENERIC_ENHANCE)) { 1558 up_read(&sbi->node_write); 1559 goto redirty_out; 1560 } 1561 1562 if (atomic && !test_opt(sbi, NOBARRIER)) 1563 fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 1564 1565 set_page_writeback(page); 1566 ClearPageError(page); 1567 1568 if (f2fs_in_warm_node_list(sbi, page)) { 1569 seq = f2fs_add_fsync_node_entry(sbi, page); 1570 if (seq_id) 1571 *seq_id = seq; 1572 } 1573 1574 fio.old_blkaddr = ni.blk_addr; 1575 f2fs_do_write_node_page(nid, &fio); 1576 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page)); 1577 dec_page_count(sbi, F2FS_DIRTY_NODES); 1578 up_read(&sbi->node_write); 1579 1580 if (wbc->for_reclaim) { 1581 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE); 1582 submitted = NULL; 1583 } 1584 1585 unlock_page(page); 1586 1587 if (unlikely(f2fs_cp_error(sbi))) { 1588 f2fs_submit_merged_write(sbi, NODE); 1589 submitted = NULL; 1590 } 1591 if (submitted) 1592 *submitted = fio.submitted; 1593 1594 if (do_balance) 1595 f2fs_balance_fs(sbi, false); 1596 return 0; 1597 1598 redirty_out: 1599 redirty_page_for_writepage(wbc, page); 1600 return AOP_WRITEPAGE_ACTIVATE; 1601 } 1602 1603 int f2fs_move_node_page(struct page *node_page, int gc_type) 1604 { 1605 int err = 0; 1606 1607 if (gc_type == FG_GC) { 1608 struct writeback_control wbc = { 1609 .sync_mode = WB_SYNC_ALL, 1610 .nr_to_write = 1, 1611 .for_reclaim = 0, 1612 }; 1613 1614 f2fs_wait_on_page_writeback(node_page, NODE, true, true); 1615 1616 set_page_dirty(node_page); 1617 1618 if (!clear_page_dirty_for_io(node_page)) { 1619 err = -EAGAIN; 1620 goto out_page; 1621 } 1622 1623 if (__write_node_page(node_page, false, NULL, 1624 &wbc, false, FS_GC_NODE_IO, NULL)) { 1625 err = -EAGAIN; 1626 unlock_page(node_page); 1627 } 1628 goto release_page; 1629 } else { 1630 /* set page dirty and write it */ 1631 if (!PageWriteback(node_page)) 1632 set_page_dirty(node_page); 1633 } 1634 out_page: 1635 unlock_page(node_page); 1636 release_page: 1637 f2fs_put_page(node_page, 0); 1638 return err; 1639 } 1640 1641 static int f2fs_write_node_page(struct page *page, 1642 struct writeback_control *wbc) 1643 { 1644 return __write_node_page(page, false, NULL, wbc, false, 1645 FS_NODE_IO, NULL); 1646 } 1647 1648 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode, 1649 struct writeback_control *wbc, bool atomic, 1650 unsigned int *seq_id) 1651 { 1652 pgoff_t index; 1653 struct pagevec pvec; 1654 int ret = 0; 1655 struct page *last_page = NULL; 1656 bool marked = false; 1657 nid_t ino = inode->i_ino; 1658 int nr_pages; 1659 int nwritten = 0; 1660 1661 if (atomic) { 1662 last_page = last_fsync_dnode(sbi, ino); 1663 if (IS_ERR_OR_NULL(last_page)) 1664 return PTR_ERR_OR_ZERO(last_page); 1665 } 1666 retry: 1667 pagevec_init(&pvec); 1668 index = 0; 1669 1670 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1671 PAGECACHE_TAG_DIRTY))) { 1672 int i; 1673 1674 for (i = 0; i < nr_pages; i++) { 1675 struct page *page = pvec.pages[i]; 1676 bool submitted = false; 1677 1678 if (unlikely(f2fs_cp_error(sbi))) { 1679 f2fs_put_page(last_page, 0); 1680 pagevec_release(&pvec); 1681 ret = -EIO; 1682 goto out; 1683 } 1684 1685 if (!IS_DNODE(page) || !is_cold_node(page)) 1686 continue; 1687 if (ino_of_node(page) != ino) 1688 continue; 1689 1690 lock_page(page); 1691 1692 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1693 continue_unlock: 1694 unlock_page(page); 1695 continue; 1696 } 1697 if (ino_of_node(page) != ino) 1698 goto continue_unlock; 1699 1700 if (!PageDirty(page) && page != last_page) { 1701 /* someone wrote it for us */ 1702 goto continue_unlock; 1703 } 1704 1705 f2fs_wait_on_page_writeback(page, NODE, true, true); 1706 1707 set_fsync_mark(page, 0); 1708 set_dentry_mark(page, 0); 1709 1710 if (!atomic || page == last_page) { 1711 set_fsync_mark(page, 1); 1712 if (IS_INODE(page)) { 1713 if (is_inode_flag_set(inode, 1714 FI_DIRTY_INODE)) 1715 f2fs_update_inode(inode, page); 1716 set_dentry_mark(page, 1717 f2fs_need_dentry_mark(sbi, ino)); 1718 } 1719 /* may be written by other thread */ 1720 if (!PageDirty(page)) 1721 set_page_dirty(page); 1722 } 1723 1724 if (!clear_page_dirty_for_io(page)) 1725 goto continue_unlock; 1726 1727 ret = __write_node_page(page, atomic && 1728 page == last_page, 1729 &submitted, wbc, true, 1730 FS_NODE_IO, seq_id); 1731 if (ret) { 1732 unlock_page(page); 1733 f2fs_put_page(last_page, 0); 1734 break; 1735 } else if (submitted) { 1736 nwritten++; 1737 } 1738 1739 if (page == last_page) { 1740 f2fs_put_page(page, 0); 1741 marked = true; 1742 break; 1743 } 1744 } 1745 pagevec_release(&pvec); 1746 cond_resched(); 1747 1748 if (ret || marked) 1749 break; 1750 } 1751 if (!ret && atomic && !marked) { 1752 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx", 1753 ino, last_page->index); 1754 lock_page(last_page); 1755 f2fs_wait_on_page_writeback(last_page, NODE, true, true); 1756 set_page_dirty(last_page); 1757 unlock_page(last_page); 1758 goto retry; 1759 } 1760 out: 1761 if (nwritten) 1762 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE); 1763 return ret ? -EIO: 0; 1764 } 1765 1766 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data) 1767 { 1768 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1769 bool clean; 1770 1771 if (inode->i_ino != ino) 1772 return 0; 1773 1774 if (!is_inode_flag_set(inode, FI_DIRTY_INODE)) 1775 return 0; 1776 1777 spin_lock(&sbi->inode_lock[DIRTY_META]); 1778 clean = list_empty(&F2FS_I(inode)->gdirty_list); 1779 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1780 1781 if (clean) 1782 return 0; 1783 1784 inode = igrab(inode); 1785 if (!inode) 1786 return 0; 1787 return 1; 1788 } 1789 1790 static bool flush_dirty_inode(struct page *page) 1791 { 1792 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1793 struct inode *inode; 1794 nid_t ino = ino_of_node(page); 1795 1796 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL); 1797 if (!inode) 1798 return false; 1799 1800 f2fs_update_inode(inode, page); 1801 unlock_page(page); 1802 1803 iput(inode); 1804 return true; 1805 } 1806 1807 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi, 1808 struct writeback_control *wbc, 1809 bool do_balance, enum iostat_type io_type) 1810 { 1811 pgoff_t index; 1812 struct pagevec pvec; 1813 int step = 0; 1814 int nwritten = 0; 1815 int ret = 0; 1816 int nr_pages, done = 0; 1817 1818 pagevec_init(&pvec); 1819 1820 next_step: 1821 index = 0; 1822 1823 while (!done && (nr_pages = pagevec_lookup_tag(&pvec, 1824 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) { 1825 int i; 1826 1827 for (i = 0; i < nr_pages; i++) { 1828 struct page *page = pvec.pages[i]; 1829 bool submitted = false; 1830 bool may_dirty = true; 1831 1832 /* give a priority to WB_SYNC threads */ 1833 if (atomic_read(&sbi->wb_sync_req[NODE]) && 1834 wbc->sync_mode == WB_SYNC_NONE) { 1835 done = 1; 1836 break; 1837 } 1838 1839 /* 1840 * flushing sequence with step: 1841 * 0. indirect nodes 1842 * 1. dentry dnodes 1843 * 2. file dnodes 1844 */ 1845 if (step == 0 && IS_DNODE(page)) 1846 continue; 1847 if (step == 1 && (!IS_DNODE(page) || 1848 is_cold_node(page))) 1849 continue; 1850 if (step == 2 && (!IS_DNODE(page) || 1851 !is_cold_node(page))) 1852 continue; 1853 lock_node: 1854 if (wbc->sync_mode == WB_SYNC_ALL) 1855 lock_page(page); 1856 else if (!trylock_page(page)) 1857 continue; 1858 1859 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1860 continue_unlock: 1861 unlock_page(page); 1862 continue; 1863 } 1864 1865 if (!PageDirty(page)) { 1866 /* someone wrote it for us */ 1867 goto continue_unlock; 1868 } 1869 1870 /* flush inline_data */ 1871 if (is_inline_node(page)) { 1872 clear_inline_node(page); 1873 unlock_page(page); 1874 flush_inline_data(sbi, ino_of_node(page)); 1875 goto lock_node; 1876 } 1877 1878 /* flush dirty inode */ 1879 if (IS_INODE(page) && may_dirty) { 1880 may_dirty = false; 1881 if (flush_dirty_inode(page)) 1882 goto lock_node; 1883 } 1884 1885 f2fs_wait_on_page_writeback(page, NODE, true, true); 1886 1887 if (!clear_page_dirty_for_io(page)) 1888 goto continue_unlock; 1889 1890 set_fsync_mark(page, 0); 1891 set_dentry_mark(page, 0); 1892 1893 ret = __write_node_page(page, false, &submitted, 1894 wbc, do_balance, io_type, NULL); 1895 if (ret) 1896 unlock_page(page); 1897 else if (submitted) 1898 nwritten++; 1899 1900 if (--wbc->nr_to_write == 0) 1901 break; 1902 } 1903 pagevec_release(&pvec); 1904 cond_resched(); 1905 1906 if (wbc->nr_to_write == 0) { 1907 step = 2; 1908 break; 1909 } 1910 } 1911 1912 if (step < 2) { 1913 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 1914 wbc->sync_mode == WB_SYNC_NONE && step == 1) 1915 goto out; 1916 step++; 1917 goto next_step; 1918 } 1919 out: 1920 if (nwritten) 1921 f2fs_submit_merged_write(sbi, NODE); 1922 1923 if (unlikely(f2fs_cp_error(sbi))) 1924 return -EIO; 1925 return ret; 1926 } 1927 1928 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, 1929 unsigned int seq_id) 1930 { 1931 struct fsync_node_entry *fn; 1932 struct page *page; 1933 struct list_head *head = &sbi->fsync_node_list; 1934 unsigned long flags; 1935 unsigned int cur_seq_id = 0; 1936 int ret2, ret = 0; 1937 1938 while (seq_id && cur_seq_id < seq_id) { 1939 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 1940 if (list_empty(head)) { 1941 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 1942 break; 1943 } 1944 fn = list_first_entry(head, struct fsync_node_entry, list); 1945 if (fn->seq_id > seq_id) { 1946 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 1947 break; 1948 } 1949 cur_seq_id = fn->seq_id; 1950 page = fn->page; 1951 get_page(page); 1952 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 1953 1954 f2fs_wait_on_page_writeback(page, NODE, true, false); 1955 if (TestClearPageError(page)) 1956 ret = -EIO; 1957 1958 put_page(page); 1959 1960 if (ret) 1961 break; 1962 } 1963 1964 ret2 = filemap_check_errors(NODE_MAPPING(sbi)); 1965 if (!ret) 1966 ret = ret2; 1967 1968 return ret; 1969 } 1970 1971 static int f2fs_write_node_pages(struct address_space *mapping, 1972 struct writeback_control *wbc) 1973 { 1974 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 1975 struct blk_plug plug; 1976 long diff; 1977 1978 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1979 goto skip_write; 1980 1981 /* balancing f2fs's metadata in background */ 1982 f2fs_balance_fs_bg(sbi); 1983 1984 /* collect a number of dirty node pages and write together */ 1985 if (wbc->sync_mode != WB_SYNC_ALL && 1986 get_pages(sbi, F2FS_DIRTY_NODES) < 1987 nr_pages_to_skip(sbi, NODE)) 1988 goto skip_write; 1989 1990 if (wbc->sync_mode == WB_SYNC_ALL) 1991 atomic_inc(&sbi->wb_sync_req[NODE]); 1992 else if (atomic_read(&sbi->wb_sync_req[NODE])) 1993 goto skip_write; 1994 1995 trace_f2fs_writepages(mapping->host, wbc, NODE); 1996 1997 diff = nr_pages_to_write(sbi, NODE, wbc); 1998 blk_start_plug(&plug); 1999 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO); 2000 blk_finish_plug(&plug); 2001 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); 2002 2003 if (wbc->sync_mode == WB_SYNC_ALL) 2004 atomic_dec(&sbi->wb_sync_req[NODE]); 2005 return 0; 2006 2007 skip_write: 2008 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES); 2009 trace_f2fs_writepages(mapping->host, wbc, NODE); 2010 return 0; 2011 } 2012 2013 static int f2fs_set_node_page_dirty(struct page *page) 2014 { 2015 trace_f2fs_set_page_dirty(page, NODE); 2016 2017 if (!PageUptodate(page)) 2018 SetPageUptodate(page); 2019 #ifdef CONFIG_F2FS_CHECK_FS 2020 if (IS_INODE(page)) 2021 f2fs_inode_chksum_set(F2FS_P_SB(page), page); 2022 #endif 2023 if (!PageDirty(page)) { 2024 __set_page_dirty_nobuffers(page); 2025 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES); 2026 f2fs_set_page_private(page, 0); 2027 f2fs_trace_pid(page); 2028 return 1; 2029 } 2030 return 0; 2031 } 2032 2033 /* 2034 * Structure of the f2fs node operations 2035 */ 2036 const struct address_space_operations f2fs_node_aops = { 2037 .writepage = f2fs_write_node_page, 2038 .writepages = f2fs_write_node_pages, 2039 .set_page_dirty = f2fs_set_node_page_dirty, 2040 .invalidatepage = f2fs_invalidate_page, 2041 .releasepage = f2fs_release_page, 2042 #ifdef CONFIG_MIGRATION 2043 .migratepage = f2fs_migrate_page, 2044 #endif 2045 }; 2046 2047 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i, 2048 nid_t n) 2049 { 2050 return radix_tree_lookup(&nm_i->free_nid_root, n); 2051 } 2052 2053 static int __insert_free_nid(struct f2fs_sb_info *sbi, 2054 struct free_nid *i, enum nid_state state) 2055 { 2056 struct f2fs_nm_info *nm_i = NM_I(sbi); 2057 2058 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i); 2059 if (err) 2060 return err; 2061 2062 f2fs_bug_on(sbi, state != i->state); 2063 nm_i->nid_cnt[state]++; 2064 if (state == FREE_NID) 2065 list_add_tail(&i->list, &nm_i->free_nid_list); 2066 return 0; 2067 } 2068 2069 static void __remove_free_nid(struct f2fs_sb_info *sbi, 2070 struct free_nid *i, enum nid_state state) 2071 { 2072 struct f2fs_nm_info *nm_i = NM_I(sbi); 2073 2074 f2fs_bug_on(sbi, state != i->state); 2075 nm_i->nid_cnt[state]--; 2076 if (state == FREE_NID) 2077 list_del(&i->list); 2078 radix_tree_delete(&nm_i->free_nid_root, i->nid); 2079 } 2080 2081 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i, 2082 enum nid_state org_state, enum nid_state dst_state) 2083 { 2084 struct f2fs_nm_info *nm_i = NM_I(sbi); 2085 2086 f2fs_bug_on(sbi, org_state != i->state); 2087 i->state = dst_state; 2088 nm_i->nid_cnt[org_state]--; 2089 nm_i->nid_cnt[dst_state]++; 2090 2091 switch (dst_state) { 2092 case PREALLOC_NID: 2093 list_del(&i->list); 2094 break; 2095 case FREE_NID: 2096 list_add_tail(&i->list, &nm_i->free_nid_list); 2097 break; 2098 default: 2099 BUG_ON(1); 2100 } 2101 } 2102 2103 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid, 2104 bool set, bool build) 2105 { 2106 struct f2fs_nm_info *nm_i = NM_I(sbi); 2107 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid); 2108 unsigned int nid_ofs = nid - START_NID(nid); 2109 2110 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap)) 2111 return; 2112 2113 if (set) { 2114 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2115 return; 2116 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2117 nm_i->free_nid_count[nat_ofs]++; 2118 } else { 2119 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2120 return; 2121 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2122 if (!build) 2123 nm_i->free_nid_count[nat_ofs]--; 2124 } 2125 } 2126 2127 /* return if the nid is recognized as free */ 2128 static bool add_free_nid(struct f2fs_sb_info *sbi, 2129 nid_t nid, bool build, bool update) 2130 { 2131 struct f2fs_nm_info *nm_i = NM_I(sbi); 2132 struct free_nid *i, *e; 2133 struct nat_entry *ne; 2134 int err = -EINVAL; 2135 bool ret = false; 2136 2137 /* 0 nid should not be used */ 2138 if (unlikely(nid == 0)) 2139 return false; 2140 2141 if (unlikely(f2fs_check_nid_range(sbi, nid))) 2142 return false; 2143 2144 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS); 2145 i->nid = nid; 2146 i->state = FREE_NID; 2147 2148 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 2149 2150 spin_lock(&nm_i->nid_list_lock); 2151 2152 if (build) { 2153 /* 2154 * Thread A Thread B 2155 * - f2fs_create 2156 * - f2fs_new_inode 2157 * - f2fs_alloc_nid 2158 * - __insert_nid_to_list(PREALLOC_NID) 2159 * - f2fs_balance_fs_bg 2160 * - f2fs_build_free_nids 2161 * - __f2fs_build_free_nids 2162 * - scan_nat_page 2163 * - add_free_nid 2164 * - __lookup_nat_cache 2165 * - f2fs_add_link 2166 * - f2fs_init_inode_metadata 2167 * - f2fs_new_inode_page 2168 * - f2fs_new_node_page 2169 * - set_node_addr 2170 * - f2fs_alloc_nid_done 2171 * - __remove_nid_from_list(PREALLOC_NID) 2172 * - __insert_nid_to_list(FREE_NID) 2173 */ 2174 ne = __lookup_nat_cache(nm_i, nid); 2175 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) || 2176 nat_get_blkaddr(ne) != NULL_ADDR)) 2177 goto err_out; 2178 2179 e = __lookup_free_nid_list(nm_i, nid); 2180 if (e) { 2181 if (e->state == FREE_NID) 2182 ret = true; 2183 goto err_out; 2184 } 2185 } 2186 ret = true; 2187 err = __insert_free_nid(sbi, i, FREE_NID); 2188 err_out: 2189 if (update) { 2190 update_free_nid_bitmap(sbi, nid, ret, build); 2191 if (!build) 2192 nm_i->available_nids++; 2193 } 2194 spin_unlock(&nm_i->nid_list_lock); 2195 radix_tree_preload_end(); 2196 2197 if (err) 2198 kmem_cache_free(free_nid_slab, i); 2199 return ret; 2200 } 2201 2202 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid) 2203 { 2204 struct f2fs_nm_info *nm_i = NM_I(sbi); 2205 struct free_nid *i; 2206 bool need_free = false; 2207 2208 spin_lock(&nm_i->nid_list_lock); 2209 i = __lookup_free_nid_list(nm_i, nid); 2210 if (i && i->state == FREE_NID) { 2211 __remove_free_nid(sbi, i, FREE_NID); 2212 need_free = true; 2213 } 2214 spin_unlock(&nm_i->nid_list_lock); 2215 2216 if (need_free) 2217 kmem_cache_free(free_nid_slab, i); 2218 } 2219 2220 static int scan_nat_page(struct f2fs_sb_info *sbi, 2221 struct page *nat_page, nid_t start_nid) 2222 { 2223 struct f2fs_nm_info *nm_i = NM_I(sbi); 2224 struct f2fs_nat_block *nat_blk = page_address(nat_page); 2225 block_t blk_addr; 2226 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid); 2227 int i; 2228 2229 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap); 2230 2231 i = start_nid % NAT_ENTRY_PER_BLOCK; 2232 2233 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) { 2234 if (unlikely(start_nid >= nm_i->max_nid)) 2235 break; 2236 2237 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr); 2238 2239 if (blk_addr == NEW_ADDR) 2240 return -EINVAL; 2241 2242 if (blk_addr == NULL_ADDR) { 2243 add_free_nid(sbi, start_nid, true, true); 2244 } else { 2245 spin_lock(&NM_I(sbi)->nid_list_lock); 2246 update_free_nid_bitmap(sbi, start_nid, false, true); 2247 spin_unlock(&NM_I(sbi)->nid_list_lock); 2248 } 2249 } 2250 2251 return 0; 2252 } 2253 2254 static void scan_curseg_cache(struct f2fs_sb_info *sbi) 2255 { 2256 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2257 struct f2fs_journal *journal = curseg->journal; 2258 int i; 2259 2260 down_read(&curseg->journal_rwsem); 2261 for (i = 0; i < nats_in_cursum(journal); i++) { 2262 block_t addr; 2263 nid_t nid; 2264 2265 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr); 2266 nid = le32_to_cpu(nid_in_journal(journal, i)); 2267 if (addr == NULL_ADDR) 2268 add_free_nid(sbi, nid, true, false); 2269 else 2270 remove_free_nid(sbi, nid); 2271 } 2272 up_read(&curseg->journal_rwsem); 2273 } 2274 2275 static void scan_free_nid_bits(struct f2fs_sb_info *sbi) 2276 { 2277 struct f2fs_nm_info *nm_i = NM_I(sbi); 2278 unsigned int i, idx; 2279 nid_t nid; 2280 2281 down_read(&nm_i->nat_tree_lock); 2282 2283 for (i = 0; i < nm_i->nat_blocks; i++) { 2284 if (!test_bit_le(i, nm_i->nat_block_bitmap)) 2285 continue; 2286 if (!nm_i->free_nid_count[i]) 2287 continue; 2288 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) { 2289 idx = find_next_bit_le(nm_i->free_nid_bitmap[i], 2290 NAT_ENTRY_PER_BLOCK, idx); 2291 if (idx >= NAT_ENTRY_PER_BLOCK) 2292 break; 2293 2294 nid = i * NAT_ENTRY_PER_BLOCK + idx; 2295 add_free_nid(sbi, nid, true, false); 2296 2297 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS) 2298 goto out; 2299 } 2300 } 2301 out: 2302 scan_curseg_cache(sbi); 2303 2304 up_read(&nm_i->nat_tree_lock); 2305 } 2306 2307 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi, 2308 bool sync, bool mount) 2309 { 2310 struct f2fs_nm_info *nm_i = NM_I(sbi); 2311 int i = 0, ret; 2312 nid_t nid = nm_i->next_scan_nid; 2313 2314 if (unlikely(nid >= nm_i->max_nid)) 2315 nid = 0; 2316 2317 /* Enough entries */ 2318 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2319 return 0; 2320 2321 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS)) 2322 return 0; 2323 2324 if (!mount) { 2325 /* try to find free nids in free_nid_bitmap */ 2326 scan_free_nid_bits(sbi); 2327 2328 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2329 return 0; 2330 } 2331 2332 /* readahead nat pages to be scanned */ 2333 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, 2334 META_NAT, true); 2335 2336 down_read(&nm_i->nat_tree_lock); 2337 2338 while (1) { 2339 if (!test_bit_le(NAT_BLOCK_OFFSET(nid), 2340 nm_i->nat_block_bitmap)) { 2341 struct page *page = get_current_nat_page(sbi, nid); 2342 2343 if (IS_ERR(page)) { 2344 ret = PTR_ERR(page); 2345 } else { 2346 ret = scan_nat_page(sbi, page, nid); 2347 f2fs_put_page(page, 1); 2348 } 2349 2350 if (ret) { 2351 up_read(&nm_i->nat_tree_lock); 2352 f2fs_bug_on(sbi, !mount); 2353 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it"); 2354 return ret; 2355 } 2356 } 2357 2358 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK)); 2359 if (unlikely(nid >= nm_i->max_nid)) 2360 nid = 0; 2361 2362 if (++i >= FREE_NID_PAGES) 2363 break; 2364 } 2365 2366 /* go to the next free nat pages to find free nids abundantly */ 2367 nm_i->next_scan_nid = nid; 2368 2369 /* find free nids from current sum_pages */ 2370 scan_curseg_cache(sbi); 2371 2372 up_read(&nm_i->nat_tree_lock); 2373 2374 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid), 2375 nm_i->ra_nid_pages, META_NAT, false); 2376 2377 return 0; 2378 } 2379 2380 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount) 2381 { 2382 int ret; 2383 2384 mutex_lock(&NM_I(sbi)->build_lock); 2385 ret = __f2fs_build_free_nids(sbi, sync, mount); 2386 mutex_unlock(&NM_I(sbi)->build_lock); 2387 2388 return ret; 2389 } 2390 2391 /* 2392 * If this function returns success, caller can obtain a new nid 2393 * from second parameter of this function. 2394 * The returned nid could be used ino as well as nid when inode is created. 2395 */ 2396 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid) 2397 { 2398 struct f2fs_nm_info *nm_i = NM_I(sbi); 2399 struct free_nid *i = NULL; 2400 retry: 2401 if (time_to_inject(sbi, FAULT_ALLOC_NID)) { 2402 f2fs_show_injection_info(FAULT_ALLOC_NID); 2403 return false; 2404 } 2405 2406 spin_lock(&nm_i->nid_list_lock); 2407 2408 if (unlikely(nm_i->available_nids == 0)) { 2409 spin_unlock(&nm_i->nid_list_lock); 2410 return false; 2411 } 2412 2413 /* We should not use stale free nids created by f2fs_build_free_nids */ 2414 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) { 2415 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list)); 2416 i = list_first_entry(&nm_i->free_nid_list, 2417 struct free_nid, list); 2418 *nid = i->nid; 2419 2420 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID); 2421 nm_i->available_nids--; 2422 2423 update_free_nid_bitmap(sbi, *nid, false, false); 2424 2425 spin_unlock(&nm_i->nid_list_lock); 2426 return true; 2427 } 2428 spin_unlock(&nm_i->nid_list_lock); 2429 2430 /* Let's scan nat pages and its caches to get free nids */ 2431 if (!f2fs_build_free_nids(sbi, true, false)) 2432 goto retry; 2433 return false; 2434 } 2435 2436 /* 2437 * f2fs_alloc_nid() should be called prior to this function. 2438 */ 2439 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 2440 { 2441 struct f2fs_nm_info *nm_i = NM_I(sbi); 2442 struct free_nid *i; 2443 2444 spin_lock(&nm_i->nid_list_lock); 2445 i = __lookup_free_nid_list(nm_i, nid); 2446 f2fs_bug_on(sbi, !i); 2447 __remove_free_nid(sbi, i, PREALLOC_NID); 2448 spin_unlock(&nm_i->nid_list_lock); 2449 2450 kmem_cache_free(free_nid_slab, i); 2451 } 2452 2453 /* 2454 * f2fs_alloc_nid() should be called prior to this function. 2455 */ 2456 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 2457 { 2458 struct f2fs_nm_info *nm_i = NM_I(sbi); 2459 struct free_nid *i; 2460 bool need_free = false; 2461 2462 if (!nid) 2463 return; 2464 2465 spin_lock(&nm_i->nid_list_lock); 2466 i = __lookup_free_nid_list(nm_i, nid); 2467 f2fs_bug_on(sbi, !i); 2468 2469 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) { 2470 __remove_free_nid(sbi, i, PREALLOC_NID); 2471 need_free = true; 2472 } else { 2473 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID); 2474 } 2475 2476 nm_i->available_nids++; 2477 2478 update_free_nid_bitmap(sbi, nid, true, false); 2479 2480 spin_unlock(&nm_i->nid_list_lock); 2481 2482 if (need_free) 2483 kmem_cache_free(free_nid_slab, i); 2484 } 2485 2486 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink) 2487 { 2488 struct f2fs_nm_info *nm_i = NM_I(sbi); 2489 struct free_nid *i, *next; 2490 int nr = nr_shrink; 2491 2492 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2493 return 0; 2494 2495 if (!mutex_trylock(&nm_i->build_lock)) 2496 return 0; 2497 2498 spin_lock(&nm_i->nid_list_lock); 2499 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) { 2500 if (nr_shrink <= 0 || 2501 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2502 break; 2503 2504 __remove_free_nid(sbi, i, FREE_NID); 2505 kmem_cache_free(free_nid_slab, i); 2506 nr_shrink--; 2507 } 2508 spin_unlock(&nm_i->nid_list_lock); 2509 mutex_unlock(&nm_i->build_lock); 2510 2511 return nr - nr_shrink; 2512 } 2513 2514 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page) 2515 { 2516 void *src_addr, *dst_addr; 2517 size_t inline_size; 2518 struct page *ipage; 2519 struct f2fs_inode *ri; 2520 2521 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 2522 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage)); 2523 2524 ri = F2FS_INODE(page); 2525 if (ri->i_inline & F2FS_INLINE_XATTR) { 2526 set_inode_flag(inode, FI_INLINE_XATTR); 2527 } else { 2528 clear_inode_flag(inode, FI_INLINE_XATTR); 2529 goto update_inode; 2530 } 2531 2532 dst_addr = inline_xattr_addr(inode, ipage); 2533 src_addr = inline_xattr_addr(inode, page); 2534 inline_size = inline_xattr_size(inode); 2535 2536 f2fs_wait_on_page_writeback(ipage, NODE, true, true); 2537 memcpy(dst_addr, src_addr, inline_size); 2538 update_inode: 2539 f2fs_update_inode(inode, ipage); 2540 f2fs_put_page(ipage, 1); 2541 } 2542 2543 int f2fs_recover_xattr_data(struct inode *inode, struct page *page) 2544 { 2545 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2546 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid; 2547 nid_t new_xnid; 2548 struct dnode_of_data dn; 2549 struct node_info ni; 2550 struct page *xpage; 2551 int err; 2552 2553 if (!prev_xnid) 2554 goto recover_xnid; 2555 2556 /* 1: invalidate the previous xattr nid */ 2557 err = f2fs_get_node_info(sbi, prev_xnid, &ni); 2558 if (err) 2559 return err; 2560 2561 f2fs_invalidate_blocks(sbi, ni.blk_addr); 2562 dec_valid_node_count(sbi, inode, false); 2563 set_node_addr(sbi, &ni, NULL_ADDR, false); 2564 2565 recover_xnid: 2566 /* 2: update xattr nid in inode */ 2567 if (!f2fs_alloc_nid(sbi, &new_xnid)) 2568 return -ENOSPC; 2569 2570 set_new_dnode(&dn, inode, NULL, NULL, new_xnid); 2571 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET); 2572 if (IS_ERR(xpage)) { 2573 f2fs_alloc_nid_failed(sbi, new_xnid); 2574 return PTR_ERR(xpage); 2575 } 2576 2577 f2fs_alloc_nid_done(sbi, new_xnid); 2578 f2fs_update_inode_page(inode); 2579 2580 /* 3: update and set xattr node page dirty */ 2581 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE); 2582 2583 set_page_dirty(xpage); 2584 f2fs_put_page(xpage, 1); 2585 2586 return 0; 2587 } 2588 2589 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 2590 { 2591 struct f2fs_inode *src, *dst; 2592 nid_t ino = ino_of_node(page); 2593 struct node_info old_ni, new_ni; 2594 struct page *ipage; 2595 int err; 2596 2597 err = f2fs_get_node_info(sbi, ino, &old_ni); 2598 if (err) 2599 return err; 2600 2601 if (unlikely(old_ni.blk_addr != NULL_ADDR)) 2602 return -EINVAL; 2603 retry: 2604 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false); 2605 if (!ipage) { 2606 congestion_wait(BLK_RW_ASYNC, HZ/50); 2607 goto retry; 2608 } 2609 2610 /* Should not use this inode from free nid list */ 2611 remove_free_nid(sbi, ino); 2612 2613 if (!PageUptodate(ipage)) 2614 SetPageUptodate(ipage); 2615 fill_node_footer(ipage, ino, ino, 0, true); 2616 set_cold_node(ipage, false); 2617 2618 src = F2FS_INODE(page); 2619 dst = F2FS_INODE(ipage); 2620 2621 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src); 2622 dst->i_size = 0; 2623 dst->i_blocks = cpu_to_le64(1); 2624 dst->i_links = cpu_to_le32(1); 2625 dst->i_xattr_nid = 0; 2626 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR); 2627 if (dst->i_inline & F2FS_EXTRA_ATTR) { 2628 dst->i_extra_isize = src->i_extra_isize; 2629 2630 if (f2fs_sb_has_flexible_inline_xattr(sbi) && 2631 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2632 i_inline_xattr_size)) 2633 dst->i_inline_xattr_size = src->i_inline_xattr_size; 2634 2635 if (f2fs_sb_has_project_quota(sbi) && 2636 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2637 i_projid)) 2638 dst->i_projid = src->i_projid; 2639 2640 if (f2fs_sb_has_inode_crtime(sbi) && 2641 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2642 i_crtime_nsec)) { 2643 dst->i_crtime = src->i_crtime; 2644 dst->i_crtime_nsec = src->i_crtime_nsec; 2645 } 2646 } 2647 2648 new_ni = old_ni; 2649 new_ni.ino = ino; 2650 2651 if (unlikely(inc_valid_node_count(sbi, NULL, true))) 2652 WARN_ON(1); 2653 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 2654 inc_valid_inode_count(sbi); 2655 set_page_dirty(ipage); 2656 f2fs_put_page(ipage, 1); 2657 return 0; 2658 } 2659 2660 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi, 2661 unsigned int segno, struct f2fs_summary_block *sum) 2662 { 2663 struct f2fs_node *rn; 2664 struct f2fs_summary *sum_entry; 2665 block_t addr; 2666 int i, idx, last_offset, nrpages; 2667 2668 /* scan the node segment */ 2669 last_offset = sbi->blocks_per_seg; 2670 addr = START_BLOCK(sbi, segno); 2671 sum_entry = &sum->entries[0]; 2672 2673 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) { 2674 nrpages = min(last_offset - i, BIO_MAX_PAGES); 2675 2676 /* readahead node pages */ 2677 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true); 2678 2679 for (idx = addr; idx < addr + nrpages; idx++) { 2680 struct page *page = f2fs_get_tmp_page(sbi, idx); 2681 2682 if (IS_ERR(page)) 2683 return PTR_ERR(page); 2684 2685 rn = F2FS_NODE(page); 2686 sum_entry->nid = rn->footer.nid; 2687 sum_entry->version = 0; 2688 sum_entry->ofs_in_node = 0; 2689 sum_entry++; 2690 f2fs_put_page(page, 1); 2691 } 2692 2693 invalidate_mapping_pages(META_MAPPING(sbi), addr, 2694 addr + nrpages); 2695 } 2696 return 0; 2697 } 2698 2699 static void remove_nats_in_journal(struct f2fs_sb_info *sbi) 2700 { 2701 struct f2fs_nm_info *nm_i = NM_I(sbi); 2702 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2703 struct f2fs_journal *journal = curseg->journal; 2704 int i; 2705 2706 down_write(&curseg->journal_rwsem); 2707 for (i = 0; i < nats_in_cursum(journal); i++) { 2708 struct nat_entry *ne; 2709 struct f2fs_nat_entry raw_ne; 2710 nid_t nid = le32_to_cpu(nid_in_journal(journal, i)); 2711 2712 raw_ne = nat_in_journal(journal, i); 2713 2714 ne = __lookup_nat_cache(nm_i, nid); 2715 if (!ne) { 2716 ne = __alloc_nat_entry(nid, true); 2717 __init_nat_entry(nm_i, ne, &raw_ne, true); 2718 } 2719 2720 /* 2721 * if a free nat in journal has not been used after last 2722 * checkpoint, we should remove it from available nids, 2723 * since later we will add it again. 2724 */ 2725 if (!get_nat_flag(ne, IS_DIRTY) && 2726 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) { 2727 spin_lock(&nm_i->nid_list_lock); 2728 nm_i->available_nids--; 2729 spin_unlock(&nm_i->nid_list_lock); 2730 } 2731 2732 __set_nat_cache_dirty(nm_i, ne); 2733 } 2734 update_nats_in_cursum(journal, -i); 2735 up_write(&curseg->journal_rwsem); 2736 } 2737 2738 static void __adjust_nat_entry_set(struct nat_entry_set *nes, 2739 struct list_head *head, int max) 2740 { 2741 struct nat_entry_set *cur; 2742 2743 if (nes->entry_cnt >= max) 2744 goto add_out; 2745 2746 list_for_each_entry(cur, head, set_list) { 2747 if (cur->entry_cnt >= nes->entry_cnt) { 2748 list_add(&nes->set_list, cur->set_list.prev); 2749 return; 2750 } 2751 } 2752 add_out: 2753 list_add_tail(&nes->set_list, head); 2754 } 2755 2756 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid, 2757 struct page *page) 2758 { 2759 struct f2fs_nm_info *nm_i = NM_I(sbi); 2760 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK; 2761 struct f2fs_nat_block *nat_blk = page_address(page); 2762 int valid = 0; 2763 int i = 0; 2764 2765 if (!enabled_nat_bits(sbi, NULL)) 2766 return; 2767 2768 if (nat_index == 0) { 2769 valid = 1; 2770 i = 1; 2771 } 2772 for (; i < NAT_ENTRY_PER_BLOCK; i++) { 2773 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR) 2774 valid++; 2775 } 2776 if (valid == 0) { 2777 __set_bit_le(nat_index, nm_i->empty_nat_bits); 2778 __clear_bit_le(nat_index, nm_i->full_nat_bits); 2779 return; 2780 } 2781 2782 __clear_bit_le(nat_index, nm_i->empty_nat_bits); 2783 if (valid == NAT_ENTRY_PER_BLOCK) 2784 __set_bit_le(nat_index, nm_i->full_nat_bits); 2785 else 2786 __clear_bit_le(nat_index, nm_i->full_nat_bits); 2787 } 2788 2789 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi, 2790 struct nat_entry_set *set, struct cp_control *cpc) 2791 { 2792 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2793 struct f2fs_journal *journal = curseg->journal; 2794 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK; 2795 bool to_journal = true; 2796 struct f2fs_nat_block *nat_blk; 2797 struct nat_entry *ne, *cur; 2798 struct page *page = NULL; 2799 2800 /* 2801 * there are two steps to flush nat entries: 2802 * #1, flush nat entries to journal in current hot data summary block. 2803 * #2, flush nat entries to nat page. 2804 */ 2805 if (enabled_nat_bits(sbi, cpc) || 2806 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL)) 2807 to_journal = false; 2808 2809 if (to_journal) { 2810 down_write(&curseg->journal_rwsem); 2811 } else { 2812 page = get_next_nat_page(sbi, start_nid); 2813 if (IS_ERR(page)) 2814 return PTR_ERR(page); 2815 2816 nat_blk = page_address(page); 2817 f2fs_bug_on(sbi, !nat_blk); 2818 } 2819 2820 /* flush dirty nats in nat entry set */ 2821 list_for_each_entry_safe(ne, cur, &set->entry_list, list) { 2822 struct f2fs_nat_entry *raw_ne; 2823 nid_t nid = nat_get_nid(ne); 2824 int offset; 2825 2826 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR); 2827 2828 if (to_journal) { 2829 offset = f2fs_lookup_journal_in_cursum(journal, 2830 NAT_JOURNAL, nid, 1); 2831 f2fs_bug_on(sbi, offset < 0); 2832 raw_ne = &nat_in_journal(journal, offset); 2833 nid_in_journal(journal, offset) = cpu_to_le32(nid); 2834 } else { 2835 raw_ne = &nat_blk->entries[nid - start_nid]; 2836 } 2837 raw_nat_from_node_info(raw_ne, &ne->ni); 2838 nat_reset_flag(ne); 2839 __clear_nat_cache_dirty(NM_I(sbi), set, ne); 2840 if (nat_get_blkaddr(ne) == NULL_ADDR) { 2841 add_free_nid(sbi, nid, false, true); 2842 } else { 2843 spin_lock(&NM_I(sbi)->nid_list_lock); 2844 update_free_nid_bitmap(sbi, nid, false, false); 2845 spin_unlock(&NM_I(sbi)->nid_list_lock); 2846 } 2847 } 2848 2849 if (to_journal) { 2850 up_write(&curseg->journal_rwsem); 2851 } else { 2852 __update_nat_bits(sbi, start_nid, page); 2853 f2fs_put_page(page, 1); 2854 } 2855 2856 /* Allow dirty nats by node block allocation in write_begin */ 2857 if (!set->entry_cnt) { 2858 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set); 2859 kmem_cache_free(nat_entry_set_slab, set); 2860 } 2861 return 0; 2862 } 2863 2864 /* 2865 * This function is called during the checkpointing process. 2866 */ 2867 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 2868 { 2869 struct f2fs_nm_info *nm_i = NM_I(sbi); 2870 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2871 struct f2fs_journal *journal = curseg->journal; 2872 struct nat_entry_set *setvec[SETVEC_SIZE]; 2873 struct nat_entry_set *set, *tmp; 2874 unsigned int found; 2875 nid_t set_idx = 0; 2876 LIST_HEAD(sets); 2877 int err = 0; 2878 2879 /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */ 2880 if (enabled_nat_bits(sbi, cpc)) { 2881 down_write(&nm_i->nat_tree_lock); 2882 remove_nats_in_journal(sbi); 2883 up_write(&nm_i->nat_tree_lock); 2884 } 2885 2886 if (!nm_i->dirty_nat_cnt) 2887 return 0; 2888 2889 down_write(&nm_i->nat_tree_lock); 2890 2891 /* 2892 * if there are no enough space in journal to store dirty nat 2893 * entries, remove all entries from journal and merge them 2894 * into nat entry set. 2895 */ 2896 if (enabled_nat_bits(sbi, cpc) || 2897 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL)) 2898 remove_nats_in_journal(sbi); 2899 2900 while ((found = __gang_lookup_nat_set(nm_i, 2901 set_idx, SETVEC_SIZE, setvec))) { 2902 unsigned idx; 2903 set_idx = setvec[found - 1]->set + 1; 2904 for (idx = 0; idx < found; idx++) 2905 __adjust_nat_entry_set(setvec[idx], &sets, 2906 MAX_NAT_JENTRIES(journal)); 2907 } 2908 2909 /* flush dirty nats in nat entry set */ 2910 list_for_each_entry_safe(set, tmp, &sets, set_list) { 2911 err = __flush_nat_entry_set(sbi, set, cpc); 2912 if (err) 2913 break; 2914 } 2915 2916 up_write(&nm_i->nat_tree_lock); 2917 /* Allow dirty nats by node block allocation in write_begin */ 2918 2919 return err; 2920 } 2921 2922 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi) 2923 { 2924 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2925 struct f2fs_nm_info *nm_i = NM_I(sbi); 2926 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE; 2927 unsigned int i; 2928 __u64 cp_ver = cur_cp_version(ckpt); 2929 block_t nat_bits_addr; 2930 2931 if (!enabled_nat_bits(sbi, NULL)) 2932 return 0; 2933 2934 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8); 2935 nm_i->nat_bits = f2fs_kzalloc(sbi, 2936 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL); 2937 if (!nm_i->nat_bits) 2938 return -ENOMEM; 2939 2940 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg - 2941 nm_i->nat_bits_blocks; 2942 for (i = 0; i < nm_i->nat_bits_blocks; i++) { 2943 struct page *page; 2944 2945 page = f2fs_get_meta_page(sbi, nat_bits_addr++); 2946 if (IS_ERR(page)) 2947 return PTR_ERR(page); 2948 2949 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS), 2950 page_address(page), F2FS_BLKSIZE); 2951 f2fs_put_page(page, 1); 2952 } 2953 2954 cp_ver |= (cur_cp_crc(ckpt) << 32); 2955 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) { 2956 disable_nat_bits(sbi, true); 2957 return 0; 2958 } 2959 2960 nm_i->full_nat_bits = nm_i->nat_bits + 8; 2961 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes; 2962 2963 f2fs_notice(sbi, "Found nat_bits in checkpoint"); 2964 return 0; 2965 } 2966 2967 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi) 2968 { 2969 struct f2fs_nm_info *nm_i = NM_I(sbi); 2970 unsigned int i = 0; 2971 nid_t nid, last_nid; 2972 2973 if (!enabled_nat_bits(sbi, NULL)) 2974 return; 2975 2976 for (i = 0; i < nm_i->nat_blocks; i++) { 2977 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i); 2978 if (i >= nm_i->nat_blocks) 2979 break; 2980 2981 __set_bit_le(i, nm_i->nat_block_bitmap); 2982 2983 nid = i * NAT_ENTRY_PER_BLOCK; 2984 last_nid = nid + NAT_ENTRY_PER_BLOCK; 2985 2986 spin_lock(&NM_I(sbi)->nid_list_lock); 2987 for (; nid < last_nid; nid++) 2988 update_free_nid_bitmap(sbi, nid, true, true); 2989 spin_unlock(&NM_I(sbi)->nid_list_lock); 2990 } 2991 2992 for (i = 0; i < nm_i->nat_blocks; i++) { 2993 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i); 2994 if (i >= nm_i->nat_blocks) 2995 break; 2996 2997 __set_bit_le(i, nm_i->nat_block_bitmap); 2998 } 2999 } 3000 3001 static int init_node_manager(struct f2fs_sb_info *sbi) 3002 { 3003 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 3004 struct f2fs_nm_info *nm_i = NM_I(sbi); 3005 unsigned char *version_bitmap; 3006 unsigned int nat_segs; 3007 int err; 3008 3009 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 3010 3011 /* segment_count_nat includes pair segment so divide to 2. */ 3012 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 3013 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 3014 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks; 3015 3016 /* not used nids: 0, node, meta, (and root counted as valid node) */ 3017 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count - 3018 F2FS_RESERVED_NODE_NUM; 3019 nm_i->nid_cnt[FREE_NID] = 0; 3020 nm_i->nid_cnt[PREALLOC_NID] = 0; 3021 nm_i->nat_cnt = 0; 3022 nm_i->ram_thresh = DEF_RAM_THRESHOLD; 3023 nm_i->ra_nid_pages = DEF_RA_NID_PAGES; 3024 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD; 3025 3026 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC); 3027 INIT_LIST_HEAD(&nm_i->free_nid_list); 3028 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO); 3029 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO); 3030 INIT_LIST_HEAD(&nm_i->nat_entries); 3031 spin_lock_init(&nm_i->nat_list_lock); 3032 3033 mutex_init(&nm_i->build_lock); 3034 spin_lock_init(&nm_i->nid_list_lock); 3035 init_rwsem(&nm_i->nat_tree_lock); 3036 3037 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 3038 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 3039 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 3040 if (!version_bitmap) 3041 return -EFAULT; 3042 3043 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size, 3044 GFP_KERNEL); 3045 if (!nm_i->nat_bitmap) 3046 return -ENOMEM; 3047 3048 err = __get_nat_bitmaps(sbi); 3049 if (err) 3050 return err; 3051 3052 #ifdef CONFIG_F2FS_CHECK_FS 3053 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size, 3054 GFP_KERNEL); 3055 if (!nm_i->nat_bitmap_mir) 3056 return -ENOMEM; 3057 #endif 3058 3059 return 0; 3060 } 3061 3062 static int init_free_nid_cache(struct f2fs_sb_info *sbi) 3063 { 3064 struct f2fs_nm_info *nm_i = NM_I(sbi); 3065 int i; 3066 3067 nm_i->free_nid_bitmap = 3068 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *), 3069 nm_i->nat_blocks), 3070 GFP_KERNEL); 3071 if (!nm_i->free_nid_bitmap) 3072 return -ENOMEM; 3073 3074 for (i = 0; i < nm_i->nat_blocks; i++) { 3075 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi, 3076 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL); 3077 if (!nm_i->free_nid_bitmap[i]) 3078 return -ENOMEM; 3079 } 3080 3081 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8, 3082 GFP_KERNEL); 3083 if (!nm_i->nat_block_bitmap) 3084 return -ENOMEM; 3085 3086 nm_i->free_nid_count = 3087 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short), 3088 nm_i->nat_blocks), 3089 GFP_KERNEL); 3090 if (!nm_i->free_nid_count) 3091 return -ENOMEM; 3092 return 0; 3093 } 3094 3095 int f2fs_build_node_manager(struct f2fs_sb_info *sbi) 3096 { 3097 int err; 3098 3099 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info), 3100 GFP_KERNEL); 3101 if (!sbi->nm_info) 3102 return -ENOMEM; 3103 3104 err = init_node_manager(sbi); 3105 if (err) 3106 return err; 3107 3108 err = init_free_nid_cache(sbi); 3109 if (err) 3110 return err; 3111 3112 /* load free nid status from nat_bits table */ 3113 load_free_nid_bitmap(sbi); 3114 3115 return f2fs_build_free_nids(sbi, true, true); 3116 } 3117 3118 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi) 3119 { 3120 struct f2fs_nm_info *nm_i = NM_I(sbi); 3121 struct free_nid *i, *next_i; 3122 struct nat_entry *natvec[NATVEC_SIZE]; 3123 struct nat_entry_set *setvec[SETVEC_SIZE]; 3124 nid_t nid = 0; 3125 unsigned int found; 3126 3127 if (!nm_i) 3128 return; 3129 3130 /* destroy free nid list */ 3131 spin_lock(&nm_i->nid_list_lock); 3132 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 3133 __remove_free_nid(sbi, i, FREE_NID); 3134 spin_unlock(&nm_i->nid_list_lock); 3135 kmem_cache_free(free_nid_slab, i); 3136 spin_lock(&nm_i->nid_list_lock); 3137 } 3138 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]); 3139 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]); 3140 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list)); 3141 spin_unlock(&nm_i->nid_list_lock); 3142 3143 /* destroy nat cache */ 3144 down_write(&nm_i->nat_tree_lock); 3145 while ((found = __gang_lookup_nat_cache(nm_i, 3146 nid, NATVEC_SIZE, natvec))) { 3147 unsigned idx; 3148 3149 nid = nat_get_nid(natvec[found - 1]) + 1; 3150 for (idx = 0; idx < found; idx++) { 3151 spin_lock(&nm_i->nat_list_lock); 3152 list_del(&natvec[idx]->list); 3153 spin_unlock(&nm_i->nat_list_lock); 3154 3155 __del_from_nat_cache(nm_i, natvec[idx]); 3156 } 3157 } 3158 f2fs_bug_on(sbi, nm_i->nat_cnt); 3159 3160 /* destroy nat set cache */ 3161 nid = 0; 3162 while ((found = __gang_lookup_nat_set(nm_i, 3163 nid, SETVEC_SIZE, setvec))) { 3164 unsigned idx; 3165 3166 nid = setvec[found - 1]->set + 1; 3167 for (idx = 0; idx < found; idx++) { 3168 /* entry_cnt is not zero, when cp_error was occurred */ 3169 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list)); 3170 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set); 3171 kmem_cache_free(nat_entry_set_slab, setvec[idx]); 3172 } 3173 } 3174 up_write(&nm_i->nat_tree_lock); 3175 3176 kvfree(nm_i->nat_block_bitmap); 3177 if (nm_i->free_nid_bitmap) { 3178 int i; 3179 3180 for (i = 0; i < nm_i->nat_blocks; i++) 3181 kvfree(nm_i->free_nid_bitmap[i]); 3182 kvfree(nm_i->free_nid_bitmap); 3183 } 3184 kvfree(nm_i->free_nid_count); 3185 3186 kvfree(nm_i->nat_bitmap); 3187 kvfree(nm_i->nat_bits); 3188 #ifdef CONFIG_F2FS_CHECK_FS 3189 kvfree(nm_i->nat_bitmap_mir); 3190 #endif 3191 sbi->nm_info = NULL; 3192 kvfree(nm_i); 3193 } 3194 3195 int __init f2fs_create_node_manager_caches(void) 3196 { 3197 nat_entry_slab = f2fs_kmem_cache_create("nat_entry", 3198 sizeof(struct nat_entry)); 3199 if (!nat_entry_slab) 3200 goto fail; 3201 3202 free_nid_slab = f2fs_kmem_cache_create("free_nid", 3203 sizeof(struct free_nid)); 3204 if (!free_nid_slab) 3205 goto destroy_nat_entry; 3206 3207 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set", 3208 sizeof(struct nat_entry_set)); 3209 if (!nat_entry_set_slab) 3210 goto destroy_free_nid; 3211 3212 fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry", 3213 sizeof(struct fsync_node_entry)); 3214 if (!fsync_node_entry_slab) 3215 goto destroy_nat_entry_set; 3216 return 0; 3217 3218 destroy_nat_entry_set: 3219 kmem_cache_destroy(nat_entry_set_slab); 3220 destroy_free_nid: 3221 kmem_cache_destroy(free_nid_slab); 3222 destroy_nat_entry: 3223 kmem_cache_destroy(nat_entry_slab); 3224 fail: 3225 return -ENOMEM; 3226 } 3227 3228 void f2fs_destroy_node_manager_caches(void) 3229 { 3230 kmem_cache_destroy(fsync_node_entry_slab); 3231 kmem_cache_destroy(nat_entry_set_slab); 3232 kmem_cache_destroy(free_nid_slab); 3233 kmem_cache_destroy(nat_entry_slab); 3234 } 3235