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_msg(sbi->sb, KERN_WARNING, 38 "%s: out-of-range nid=%x, run fsck to fix.", 39 __func__, nid); 40 return -EINVAL; 41 } 42 return 0; 43 } 44 45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type) 46 { 47 struct f2fs_nm_info *nm_i = NM_I(sbi); 48 struct sysinfo val; 49 unsigned long avail_ram; 50 unsigned long mem_size = 0; 51 bool res = false; 52 53 si_meminfo(&val); 54 55 /* only uses low memory */ 56 avail_ram = val.totalram - val.totalhigh; 57 58 /* 59 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively 60 */ 61 if (type == FREE_NIDS) { 62 mem_size = (nm_i->nid_cnt[FREE_NID] * 63 sizeof(struct free_nid)) >> PAGE_SHIFT; 64 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 65 } else if (type == NAT_ENTRIES) { 66 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >> 67 PAGE_SHIFT; 68 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2); 69 if (excess_cached_nats(sbi)) 70 res = false; 71 } else if (type == DIRTY_DENTS) { 72 if (sbi->sb->s_bdi->wb.dirty_exceeded) 73 return false; 74 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS); 75 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 76 } else if (type == INO_ENTRIES) { 77 int i; 78 79 for (i = 0; i < MAX_INO_ENTRY; i++) 80 mem_size += sbi->im[i].ino_num * 81 sizeof(struct ino_entry); 82 mem_size >>= PAGE_SHIFT; 83 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 84 } else if (type == EXTENT_CACHE) { 85 mem_size = (atomic_read(&sbi->total_ext_tree) * 86 sizeof(struct extent_tree) + 87 atomic_read(&sbi->total_ext_node) * 88 sizeof(struct extent_node)) >> PAGE_SHIFT; 89 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1); 90 } else if (type == INMEM_PAGES) { 91 /* it allows 20% / total_ram for inmemory pages */ 92 mem_size = get_pages(sbi, F2FS_INMEM_PAGES); 93 res = mem_size < (val.totalram / 5); 94 } else { 95 if (!sbi->sb->s_bdi->wb.dirty_exceeded) 96 return true; 97 } 98 return res; 99 } 100 101 static void clear_node_page_dirty(struct page *page) 102 { 103 if (PageDirty(page)) { 104 f2fs_clear_page_cache_dirty_tag(page); 105 clear_page_dirty_for_io(page); 106 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES); 107 } 108 ClearPageUptodate(page); 109 } 110 111 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 112 { 113 return f2fs_get_meta_page_nofail(sbi, current_nat_addr(sbi, nid)); 114 } 115 116 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 117 { 118 struct page *src_page; 119 struct page *dst_page; 120 pgoff_t dst_off; 121 void *src_addr; 122 void *dst_addr; 123 struct f2fs_nm_info *nm_i = NM_I(sbi); 124 125 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid)); 126 127 /* get current nat block page with lock */ 128 src_page = get_current_nat_page(sbi, nid); 129 if (IS_ERR(src_page)) 130 return src_page; 131 dst_page = f2fs_grab_meta_page(sbi, dst_off); 132 f2fs_bug_on(sbi, PageDirty(src_page)); 133 134 src_addr = page_address(src_page); 135 dst_addr = page_address(dst_page); 136 memcpy(dst_addr, src_addr, PAGE_SIZE); 137 set_page_dirty(dst_page); 138 f2fs_put_page(src_page, 1); 139 140 set_to_next_nat(nm_i, nid); 141 142 return dst_page; 143 } 144 145 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail) 146 { 147 struct nat_entry *new; 148 149 if (no_fail) 150 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO); 151 else 152 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO); 153 if (new) { 154 nat_set_nid(new, nid); 155 nat_reset_flag(new); 156 } 157 return new; 158 } 159 160 static void __free_nat_entry(struct nat_entry *e) 161 { 162 kmem_cache_free(nat_entry_slab, e); 163 } 164 165 /* must be locked by nat_tree_lock */ 166 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i, 167 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail) 168 { 169 if (no_fail) 170 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne); 171 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne)) 172 return NULL; 173 174 if (raw_ne) 175 node_info_from_raw_nat(&ne->ni, raw_ne); 176 177 spin_lock(&nm_i->nat_list_lock); 178 list_add_tail(&ne->list, &nm_i->nat_entries); 179 spin_unlock(&nm_i->nat_list_lock); 180 181 nm_i->nat_cnt++; 182 return ne; 183 } 184 185 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n) 186 { 187 struct nat_entry *ne; 188 189 ne = radix_tree_lookup(&nm_i->nat_root, n); 190 191 /* for recent accessed nat entry, move it to tail of lru list */ 192 if (ne && !get_nat_flag(ne, IS_DIRTY)) { 193 spin_lock(&nm_i->nat_list_lock); 194 if (!list_empty(&ne->list)) 195 list_move_tail(&ne->list, &nm_i->nat_entries); 196 spin_unlock(&nm_i->nat_list_lock); 197 } 198 199 return ne; 200 } 201 202 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i, 203 nid_t start, unsigned int nr, struct nat_entry **ep) 204 { 205 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr); 206 } 207 208 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e) 209 { 210 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e)); 211 nm_i->nat_cnt--; 212 __free_nat_entry(e); 213 } 214 215 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i, 216 struct nat_entry *ne) 217 { 218 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid); 219 struct nat_entry_set *head; 220 221 head = radix_tree_lookup(&nm_i->nat_set_root, set); 222 if (!head) { 223 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS); 224 225 INIT_LIST_HEAD(&head->entry_list); 226 INIT_LIST_HEAD(&head->set_list); 227 head->set = set; 228 head->entry_cnt = 0; 229 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head); 230 } 231 return head; 232 } 233 234 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i, 235 struct nat_entry *ne) 236 { 237 struct nat_entry_set *head; 238 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR; 239 240 if (!new_ne) 241 head = __grab_nat_entry_set(nm_i, ne); 242 243 /* 244 * update entry_cnt in below condition: 245 * 1. update NEW_ADDR to valid block address; 246 * 2. update old block address to new one; 247 */ 248 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) || 249 !get_nat_flag(ne, IS_DIRTY))) 250 head->entry_cnt++; 251 252 set_nat_flag(ne, IS_PREALLOC, new_ne); 253 254 if (get_nat_flag(ne, IS_DIRTY)) 255 goto refresh_list; 256 257 nm_i->dirty_nat_cnt++; 258 set_nat_flag(ne, IS_DIRTY, true); 259 refresh_list: 260 spin_lock(&nm_i->nat_list_lock); 261 if (new_ne) 262 list_del_init(&ne->list); 263 else 264 list_move_tail(&ne->list, &head->entry_list); 265 spin_unlock(&nm_i->nat_list_lock); 266 } 267 268 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i, 269 struct nat_entry_set *set, struct nat_entry *ne) 270 { 271 spin_lock(&nm_i->nat_list_lock); 272 list_move_tail(&ne->list, &nm_i->nat_entries); 273 spin_unlock(&nm_i->nat_list_lock); 274 275 set_nat_flag(ne, IS_DIRTY, false); 276 set->entry_cnt--; 277 nm_i->dirty_nat_cnt--; 278 } 279 280 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i, 281 nid_t start, unsigned int nr, struct nat_entry_set **ep) 282 { 283 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep, 284 start, nr); 285 } 286 287 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page) 288 { 289 return NODE_MAPPING(sbi) == page->mapping && 290 IS_DNODE(page) && is_cold_node(page); 291 } 292 293 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi) 294 { 295 spin_lock_init(&sbi->fsync_node_lock); 296 INIT_LIST_HEAD(&sbi->fsync_node_list); 297 sbi->fsync_seg_id = 0; 298 sbi->fsync_node_num = 0; 299 } 300 301 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi, 302 struct page *page) 303 { 304 struct fsync_node_entry *fn; 305 unsigned long flags; 306 unsigned int seq_id; 307 308 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS); 309 310 get_page(page); 311 fn->page = page; 312 INIT_LIST_HEAD(&fn->list); 313 314 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 315 list_add_tail(&fn->list, &sbi->fsync_node_list); 316 fn->seq_id = sbi->fsync_seg_id++; 317 seq_id = fn->seq_id; 318 sbi->fsync_node_num++; 319 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 320 321 return seq_id; 322 } 323 324 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page) 325 { 326 struct fsync_node_entry *fn; 327 unsigned long flags; 328 329 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 330 list_for_each_entry(fn, &sbi->fsync_node_list, list) { 331 if (fn->page == page) { 332 list_del(&fn->list); 333 sbi->fsync_node_num--; 334 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 335 kmem_cache_free(fsync_node_entry_slab, fn); 336 put_page(page); 337 return; 338 } 339 } 340 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 341 f2fs_bug_on(sbi, 1); 342 } 343 344 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi) 345 { 346 unsigned long flags; 347 348 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 349 sbi->fsync_seg_id = 0; 350 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 351 } 352 353 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid) 354 { 355 struct f2fs_nm_info *nm_i = NM_I(sbi); 356 struct nat_entry *e; 357 bool need = false; 358 359 down_read(&nm_i->nat_tree_lock); 360 e = __lookup_nat_cache(nm_i, nid); 361 if (e) { 362 if (!get_nat_flag(e, IS_CHECKPOINTED) && 363 !get_nat_flag(e, HAS_FSYNCED_INODE)) 364 need = true; 365 } 366 up_read(&nm_i->nat_tree_lock); 367 return need; 368 } 369 370 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid) 371 { 372 struct f2fs_nm_info *nm_i = NM_I(sbi); 373 struct nat_entry *e; 374 bool is_cp = true; 375 376 down_read(&nm_i->nat_tree_lock); 377 e = __lookup_nat_cache(nm_i, nid); 378 if (e && !get_nat_flag(e, IS_CHECKPOINTED)) 379 is_cp = false; 380 up_read(&nm_i->nat_tree_lock); 381 return is_cp; 382 } 383 384 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino) 385 { 386 struct f2fs_nm_info *nm_i = NM_I(sbi); 387 struct nat_entry *e; 388 bool need_update = true; 389 390 down_read(&nm_i->nat_tree_lock); 391 e = __lookup_nat_cache(nm_i, ino); 392 if (e && get_nat_flag(e, HAS_LAST_FSYNC) && 393 (get_nat_flag(e, IS_CHECKPOINTED) || 394 get_nat_flag(e, HAS_FSYNCED_INODE))) 395 need_update = false; 396 up_read(&nm_i->nat_tree_lock); 397 return need_update; 398 } 399 400 /* must be locked by nat_tree_lock */ 401 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid, 402 struct f2fs_nat_entry *ne) 403 { 404 struct f2fs_nm_info *nm_i = NM_I(sbi); 405 struct nat_entry *new, *e; 406 407 new = __alloc_nat_entry(nid, false); 408 if (!new) 409 return; 410 411 down_write(&nm_i->nat_tree_lock); 412 e = __lookup_nat_cache(nm_i, nid); 413 if (!e) 414 e = __init_nat_entry(nm_i, new, ne, false); 415 else 416 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) || 417 nat_get_blkaddr(e) != 418 le32_to_cpu(ne->block_addr) || 419 nat_get_version(e) != ne->version); 420 up_write(&nm_i->nat_tree_lock); 421 if (e != new) 422 __free_nat_entry(new); 423 } 424 425 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni, 426 block_t new_blkaddr, bool fsync_done) 427 { 428 struct f2fs_nm_info *nm_i = NM_I(sbi); 429 struct nat_entry *e; 430 struct nat_entry *new = __alloc_nat_entry(ni->nid, true); 431 432 down_write(&nm_i->nat_tree_lock); 433 e = __lookup_nat_cache(nm_i, ni->nid); 434 if (!e) { 435 e = __init_nat_entry(nm_i, new, NULL, true); 436 copy_node_info(&e->ni, ni); 437 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR); 438 } else if (new_blkaddr == NEW_ADDR) { 439 /* 440 * when nid is reallocated, 441 * previous nat entry can be remained in nat cache. 442 * So, reinitialize it with new information. 443 */ 444 copy_node_info(&e->ni, ni); 445 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR); 446 } 447 /* let's free early to reduce memory consumption */ 448 if (e != new) 449 __free_nat_entry(new); 450 451 /* sanity check */ 452 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr); 453 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR && 454 new_blkaddr == NULL_ADDR); 455 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR && 456 new_blkaddr == NEW_ADDR); 457 f2fs_bug_on(sbi, is_valid_data_blkaddr(sbi, nat_get_blkaddr(e)) && 458 new_blkaddr == NEW_ADDR); 459 460 /* increment version no as node is removed */ 461 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) { 462 unsigned char version = nat_get_version(e); 463 nat_set_version(e, inc_node_version(version)); 464 } 465 466 /* change address */ 467 nat_set_blkaddr(e, new_blkaddr); 468 if (!is_valid_data_blkaddr(sbi, new_blkaddr)) 469 set_nat_flag(e, IS_CHECKPOINTED, false); 470 __set_nat_cache_dirty(nm_i, e); 471 472 /* update fsync_mark if its inode nat entry is still alive */ 473 if (ni->nid != ni->ino) 474 e = __lookup_nat_cache(nm_i, ni->ino); 475 if (e) { 476 if (fsync_done && ni->nid == ni->ino) 477 set_nat_flag(e, HAS_FSYNCED_INODE, true); 478 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done); 479 } 480 up_write(&nm_i->nat_tree_lock); 481 } 482 483 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink) 484 { 485 struct f2fs_nm_info *nm_i = NM_I(sbi); 486 int nr = nr_shrink; 487 488 if (!down_write_trylock(&nm_i->nat_tree_lock)) 489 return 0; 490 491 spin_lock(&nm_i->nat_list_lock); 492 while (nr_shrink) { 493 struct nat_entry *ne; 494 495 if (list_empty(&nm_i->nat_entries)) 496 break; 497 498 ne = list_first_entry(&nm_i->nat_entries, 499 struct nat_entry, list); 500 list_del(&ne->list); 501 spin_unlock(&nm_i->nat_list_lock); 502 503 __del_from_nat_cache(nm_i, ne); 504 nr_shrink--; 505 506 spin_lock(&nm_i->nat_list_lock); 507 } 508 spin_unlock(&nm_i->nat_list_lock); 509 510 up_write(&nm_i->nat_tree_lock); 511 return nr - nr_shrink; 512 } 513 514 /* 515 * This function always returns success 516 */ 517 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid, 518 struct node_info *ni) 519 { 520 struct f2fs_nm_info *nm_i = NM_I(sbi); 521 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 522 struct f2fs_journal *journal = curseg->journal; 523 nid_t start_nid = START_NID(nid); 524 struct f2fs_nat_block *nat_blk; 525 struct page *page = NULL; 526 struct f2fs_nat_entry ne; 527 struct nat_entry *e; 528 pgoff_t index; 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 /* cache nat entry */ 573 cache_nat_entry(sbi, nid, &ne); 574 return 0; 575 } 576 577 /* 578 * readahead MAX_RA_NODE number of node pages. 579 */ 580 static void f2fs_ra_node_pages(struct page *parent, int start, int n) 581 { 582 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 583 struct blk_plug plug; 584 int i, end; 585 nid_t nid; 586 587 blk_start_plug(&plug); 588 589 /* Then, try readahead for siblings of the desired node */ 590 end = start + n; 591 end = min(end, NIDS_PER_BLOCK); 592 for (i = start; i < end; i++) { 593 nid = get_nid(parent, i, false); 594 f2fs_ra_node_page(sbi, nid); 595 } 596 597 blk_finish_plug(&plug); 598 } 599 600 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs) 601 { 602 const long direct_index = ADDRS_PER_INODE(dn->inode); 603 const long direct_blks = ADDRS_PER_BLOCK; 604 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK; 605 unsigned int skipped_unit = ADDRS_PER_BLOCK; 606 int cur_level = dn->cur_level; 607 int max_level = dn->max_level; 608 pgoff_t base = 0; 609 610 if (!dn->max_level) 611 return pgofs + 1; 612 613 while (max_level-- > cur_level) 614 skipped_unit *= NIDS_PER_BLOCK; 615 616 switch (dn->max_level) { 617 case 3: 618 base += 2 * indirect_blks; 619 case 2: 620 base += 2 * direct_blks; 621 case 1: 622 base += direct_index; 623 break; 624 default: 625 f2fs_bug_on(F2FS_I_SB(dn->inode), 1); 626 } 627 628 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base; 629 } 630 631 /* 632 * The maximum depth is four. 633 * Offset[0] will have raw inode offset. 634 */ 635 static int get_node_path(struct inode *inode, long block, 636 int offset[4], unsigned int noffset[4]) 637 { 638 const long direct_index = ADDRS_PER_INODE(inode); 639 const long direct_blks = ADDRS_PER_BLOCK; 640 const long dptrs_per_blk = NIDS_PER_BLOCK; 641 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK; 642 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK; 643 int n = 0; 644 int level = 0; 645 646 noffset[0] = 0; 647 648 if (block < direct_index) { 649 offset[n] = block; 650 goto got; 651 } 652 block -= direct_index; 653 if (block < direct_blks) { 654 offset[n++] = NODE_DIR1_BLOCK; 655 noffset[n] = 1; 656 offset[n] = block; 657 level = 1; 658 goto got; 659 } 660 block -= direct_blks; 661 if (block < direct_blks) { 662 offset[n++] = NODE_DIR2_BLOCK; 663 noffset[n] = 2; 664 offset[n] = block; 665 level = 1; 666 goto got; 667 } 668 block -= direct_blks; 669 if (block < indirect_blks) { 670 offset[n++] = NODE_IND1_BLOCK; 671 noffset[n] = 3; 672 offset[n++] = block / direct_blks; 673 noffset[n] = 4 + offset[n - 1]; 674 offset[n] = block % direct_blks; 675 level = 2; 676 goto got; 677 } 678 block -= indirect_blks; 679 if (block < indirect_blks) { 680 offset[n++] = NODE_IND2_BLOCK; 681 noffset[n] = 4 + dptrs_per_blk; 682 offset[n++] = block / direct_blks; 683 noffset[n] = 5 + dptrs_per_blk + offset[n - 1]; 684 offset[n] = block % direct_blks; 685 level = 2; 686 goto got; 687 } 688 block -= indirect_blks; 689 if (block < dindirect_blks) { 690 offset[n++] = NODE_DIND_BLOCK; 691 noffset[n] = 5 + (dptrs_per_blk * 2); 692 offset[n++] = block / indirect_blks; 693 noffset[n] = 6 + (dptrs_per_blk * 2) + 694 offset[n - 1] * (dptrs_per_blk + 1); 695 offset[n++] = (block / direct_blks) % dptrs_per_blk; 696 noffset[n] = 7 + (dptrs_per_blk * 2) + 697 offset[n - 2] * (dptrs_per_blk + 1) + 698 offset[n - 1]; 699 offset[n] = block % direct_blks; 700 level = 3; 701 goto got; 702 } else { 703 return -E2BIG; 704 } 705 got: 706 return level; 707 } 708 709 /* 710 * Caller should call f2fs_put_dnode(dn). 711 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and 712 * f2fs_unlock_op() only if ro is not set RDONLY_NODE. 713 * In the case of RDONLY_NODE, we don't need to care about mutex. 714 */ 715 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode) 716 { 717 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 718 struct page *npage[4]; 719 struct page *parent = NULL; 720 int offset[4]; 721 unsigned int noffset[4]; 722 nid_t nids[4]; 723 int level, i = 0; 724 int err = 0; 725 726 level = get_node_path(dn->inode, index, offset, noffset); 727 if (level < 0) 728 return level; 729 730 nids[0] = dn->inode->i_ino; 731 npage[0] = dn->inode_page; 732 733 if (!npage[0]) { 734 npage[0] = f2fs_get_node_page(sbi, nids[0]); 735 if (IS_ERR(npage[0])) 736 return PTR_ERR(npage[0]); 737 } 738 739 /* if inline_data is set, should not report any block indices */ 740 if (f2fs_has_inline_data(dn->inode) && index) { 741 err = -ENOENT; 742 f2fs_put_page(npage[0], 1); 743 goto release_out; 744 } 745 746 parent = npage[0]; 747 if (level != 0) 748 nids[1] = get_nid(parent, offset[0], true); 749 dn->inode_page = npage[0]; 750 dn->inode_page_locked = true; 751 752 /* get indirect or direct nodes */ 753 for (i = 1; i <= level; i++) { 754 bool done = false; 755 756 if (!nids[i] && mode == ALLOC_NODE) { 757 /* alloc new node */ 758 if (!f2fs_alloc_nid(sbi, &(nids[i]))) { 759 err = -ENOSPC; 760 goto release_pages; 761 } 762 763 dn->nid = nids[i]; 764 npage[i] = f2fs_new_node_page(dn, noffset[i]); 765 if (IS_ERR(npage[i])) { 766 f2fs_alloc_nid_failed(sbi, nids[i]); 767 err = PTR_ERR(npage[i]); 768 goto release_pages; 769 } 770 771 set_nid(parent, offset[i - 1], nids[i], i == 1); 772 f2fs_alloc_nid_done(sbi, nids[i]); 773 done = true; 774 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) { 775 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]); 776 if (IS_ERR(npage[i])) { 777 err = PTR_ERR(npage[i]); 778 goto release_pages; 779 } 780 done = true; 781 } 782 if (i == 1) { 783 dn->inode_page_locked = false; 784 unlock_page(parent); 785 } else { 786 f2fs_put_page(parent, 1); 787 } 788 789 if (!done) { 790 npage[i] = f2fs_get_node_page(sbi, nids[i]); 791 if (IS_ERR(npage[i])) { 792 err = PTR_ERR(npage[i]); 793 f2fs_put_page(npage[0], 0); 794 goto release_out; 795 } 796 } 797 if (i < level) { 798 parent = npage[i]; 799 nids[i + 1] = get_nid(parent, offset[i], false); 800 } 801 } 802 dn->nid = nids[level]; 803 dn->ofs_in_node = offset[level]; 804 dn->node_page = npage[level]; 805 dn->data_blkaddr = datablock_addr(dn->inode, 806 dn->node_page, dn->ofs_in_node); 807 return 0; 808 809 release_pages: 810 f2fs_put_page(parent, 1); 811 if (i > 1) 812 f2fs_put_page(npage[0], 0); 813 release_out: 814 dn->inode_page = NULL; 815 dn->node_page = NULL; 816 if (err == -ENOENT) { 817 dn->cur_level = i; 818 dn->max_level = level; 819 dn->ofs_in_node = offset[level]; 820 } 821 return err; 822 } 823 824 static int truncate_node(struct dnode_of_data *dn) 825 { 826 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 827 struct node_info ni; 828 int err; 829 pgoff_t index; 830 831 err = f2fs_get_node_info(sbi, dn->nid, &ni); 832 if (err) 833 return err; 834 835 /* Deallocate node address */ 836 f2fs_invalidate_blocks(sbi, ni.blk_addr); 837 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino); 838 set_node_addr(sbi, &ni, NULL_ADDR, false); 839 840 if (dn->nid == dn->inode->i_ino) { 841 f2fs_remove_orphan_inode(sbi, dn->nid); 842 dec_valid_inode_count(sbi); 843 f2fs_inode_synced(dn->inode); 844 } 845 846 clear_node_page_dirty(dn->node_page); 847 set_sbi_flag(sbi, SBI_IS_DIRTY); 848 849 index = dn->node_page->index; 850 f2fs_put_page(dn->node_page, 1); 851 852 invalidate_mapping_pages(NODE_MAPPING(sbi), 853 index, index); 854 855 dn->node_page = NULL; 856 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr); 857 858 return 0; 859 } 860 861 static int truncate_dnode(struct dnode_of_data *dn) 862 { 863 struct page *page; 864 int err; 865 866 if (dn->nid == 0) 867 return 1; 868 869 /* get direct node */ 870 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 871 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT) 872 return 1; 873 else if (IS_ERR(page)) 874 return PTR_ERR(page); 875 876 /* Make dnode_of_data for parameter */ 877 dn->node_page = page; 878 dn->ofs_in_node = 0; 879 f2fs_truncate_data_blocks(dn); 880 err = truncate_node(dn); 881 if (err) 882 return err; 883 884 return 1; 885 } 886 887 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs, 888 int ofs, int depth) 889 { 890 struct dnode_of_data rdn = *dn; 891 struct page *page; 892 struct f2fs_node *rn; 893 nid_t child_nid; 894 unsigned int child_nofs; 895 int freed = 0; 896 int i, ret; 897 898 if (dn->nid == 0) 899 return NIDS_PER_BLOCK + 1; 900 901 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr); 902 903 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid); 904 if (IS_ERR(page)) { 905 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page)); 906 return PTR_ERR(page); 907 } 908 909 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK); 910 911 rn = F2FS_NODE(page); 912 if (depth < 3) { 913 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) { 914 child_nid = le32_to_cpu(rn->in.nid[i]); 915 if (child_nid == 0) 916 continue; 917 rdn.nid = child_nid; 918 ret = truncate_dnode(&rdn); 919 if (ret < 0) 920 goto out_err; 921 if (set_nid(page, i, 0, false)) 922 dn->node_changed = true; 923 } 924 } else { 925 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1; 926 for (i = ofs; i < NIDS_PER_BLOCK; i++) { 927 child_nid = le32_to_cpu(rn->in.nid[i]); 928 if (child_nid == 0) { 929 child_nofs += NIDS_PER_BLOCK + 1; 930 continue; 931 } 932 rdn.nid = child_nid; 933 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1); 934 if (ret == (NIDS_PER_BLOCK + 1)) { 935 if (set_nid(page, i, 0, false)) 936 dn->node_changed = true; 937 child_nofs += ret; 938 } else if (ret < 0 && ret != -ENOENT) { 939 goto out_err; 940 } 941 } 942 freed = child_nofs; 943 } 944 945 if (!ofs) { 946 /* remove current indirect node */ 947 dn->node_page = page; 948 ret = truncate_node(dn); 949 if (ret) 950 goto out_err; 951 freed++; 952 } else { 953 f2fs_put_page(page, 1); 954 } 955 trace_f2fs_truncate_nodes_exit(dn->inode, freed); 956 return freed; 957 958 out_err: 959 f2fs_put_page(page, 1); 960 trace_f2fs_truncate_nodes_exit(dn->inode, ret); 961 return ret; 962 } 963 964 static int truncate_partial_nodes(struct dnode_of_data *dn, 965 struct f2fs_inode *ri, int *offset, int depth) 966 { 967 struct page *pages[2]; 968 nid_t nid[3]; 969 nid_t child_nid; 970 int err = 0; 971 int i; 972 int idx = depth - 2; 973 974 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 975 if (!nid[0]) 976 return 0; 977 978 /* get indirect nodes in the path */ 979 for (i = 0; i < idx + 1; i++) { 980 /* reference count'll be increased */ 981 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]); 982 if (IS_ERR(pages[i])) { 983 err = PTR_ERR(pages[i]); 984 idx = i - 1; 985 goto fail; 986 } 987 nid[i + 1] = get_nid(pages[i], offset[i + 1], false); 988 } 989 990 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK); 991 992 /* free direct nodes linked to a partial indirect node */ 993 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) { 994 child_nid = get_nid(pages[idx], i, false); 995 if (!child_nid) 996 continue; 997 dn->nid = child_nid; 998 err = truncate_dnode(dn); 999 if (err < 0) 1000 goto fail; 1001 if (set_nid(pages[idx], i, 0, false)) 1002 dn->node_changed = true; 1003 } 1004 1005 if (offset[idx + 1] == 0) { 1006 dn->node_page = pages[idx]; 1007 dn->nid = nid[idx]; 1008 err = truncate_node(dn); 1009 if (err) 1010 goto fail; 1011 } else { 1012 f2fs_put_page(pages[idx], 1); 1013 } 1014 offset[idx]++; 1015 offset[idx + 1] = 0; 1016 idx--; 1017 fail: 1018 for (i = idx; i >= 0; i--) 1019 f2fs_put_page(pages[i], 1); 1020 1021 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err); 1022 1023 return err; 1024 } 1025 1026 /* 1027 * All the block addresses of data and nodes should be nullified. 1028 */ 1029 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from) 1030 { 1031 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1032 int err = 0, cont = 1; 1033 int level, offset[4], noffset[4]; 1034 unsigned int nofs = 0; 1035 struct f2fs_inode *ri; 1036 struct dnode_of_data dn; 1037 struct page *page; 1038 1039 trace_f2fs_truncate_inode_blocks_enter(inode, from); 1040 1041 level = get_node_path(inode, from, offset, noffset); 1042 if (level < 0) 1043 return level; 1044 1045 page = f2fs_get_node_page(sbi, inode->i_ino); 1046 if (IS_ERR(page)) { 1047 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page)); 1048 return PTR_ERR(page); 1049 } 1050 1051 set_new_dnode(&dn, inode, page, NULL, 0); 1052 unlock_page(page); 1053 1054 ri = F2FS_INODE(page); 1055 switch (level) { 1056 case 0: 1057 case 1: 1058 nofs = noffset[1]; 1059 break; 1060 case 2: 1061 nofs = noffset[1]; 1062 if (!offset[level - 1]) 1063 goto skip_partial; 1064 err = truncate_partial_nodes(&dn, ri, offset, level); 1065 if (err < 0 && err != -ENOENT) 1066 goto fail; 1067 nofs += 1 + NIDS_PER_BLOCK; 1068 break; 1069 case 3: 1070 nofs = 5 + 2 * NIDS_PER_BLOCK; 1071 if (!offset[level - 1]) 1072 goto skip_partial; 1073 err = truncate_partial_nodes(&dn, ri, offset, level); 1074 if (err < 0 && err != -ENOENT) 1075 goto fail; 1076 break; 1077 default: 1078 BUG(); 1079 } 1080 1081 skip_partial: 1082 while (cont) { 1083 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 1084 switch (offset[0]) { 1085 case NODE_DIR1_BLOCK: 1086 case NODE_DIR2_BLOCK: 1087 err = truncate_dnode(&dn); 1088 break; 1089 1090 case NODE_IND1_BLOCK: 1091 case NODE_IND2_BLOCK: 1092 err = truncate_nodes(&dn, nofs, offset[1], 2); 1093 break; 1094 1095 case NODE_DIND_BLOCK: 1096 err = truncate_nodes(&dn, nofs, offset[1], 3); 1097 cont = 0; 1098 break; 1099 1100 default: 1101 BUG(); 1102 } 1103 if (err < 0 && err != -ENOENT) 1104 goto fail; 1105 if (offset[1] == 0 && 1106 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) { 1107 lock_page(page); 1108 BUG_ON(page->mapping != NODE_MAPPING(sbi)); 1109 f2fs_wait_on_page_writeback(page, NODE, true, true); 1110 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0; 1111 set_page_dirty(page); 1112 unlock_page(page); 1113 } 1114 offset[1] = 0; 1115 offset[0]++; 1116 nofs += err; 1117 } 1118 fail: 1119 f2fs_put_page(page, 0); 1120 trace_f2fs_truncate_inode_blocks_exit(inode, err); 1121 return err > 0 ? 0 : err; 1122 } 1123 1124 /* caller must lock inode page */ 1125 int f2fs_truncate_xattr_node(struct inode *inode) 1126 { 1127 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1128 nid_t nid = F2FS_I(inode)->i_xattr_nid; 1129 struct dnode_of_data dn; 1130 struct page *npage; 1131 int err; 1132 1133 if (!nid) 1134 return 0; 1135 1136 npage = f2fs_get_node_page(sbi, nid); 1137 if (IS_ERR(npage)) 1138 return PTR_ERR(npage); 1139 1140 set_new_dnode(&dn, inode, NULL, npage, nid); 1141 err = truncate_node(&dn); 1142 if (err) { 1143 f2fs_put_page(npage, 1); 1144 return err; 1145 } 1146 1147 f2fs_i_xnid_write(inode, 0); 1148 1149 return 0; 1150 } 1151 1152 /* 1153 * Caller should grab and release a rwsem by calling f2fs_lock_op() and 1154 * f2fs_unlock_op(). 1155 */ 1156 int f2fs_remove_inode_page(struct inode *inode) 1157 { 1158 struct dnode_of_data dn; 1159 int err; 1160 1161 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1162 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE); 1163 if (err) 1164 return err; 1165 1166 err = f2fs_truncate_xattr_node(inode); 1167 if (err) { 1168 f2fs_put_dnode(&dn); 1169 return err; 1170 } 1171 1172 /* remove potential inline_data blocks */ 1173 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1174 S_ISLNK(inode->i_mode)) 1175 f2fs_truncate_data_blocks_range(&dn, 1); 1176 1177 /* 0 is possible, after f2fs_new_inode() has failed */ 1178 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) { 1179 f2fs_put_dnode(&dn); 1180 return -EIO; 1181 } 1182 f2fs_bug_on(F2FS_I_SB(inode), 1183 inode->i_blocks != 0 && inode->i_blocks != 8); 1184 1185 /* will put inode & node pages */ 1186 err = truncate_node(&dn); 1187 if (err) { 1188 f2fs_put_dnode(&dn); 1189 return err; 1190 } 1191 return 0; 1192 } 1193 1194 struct page *f2fs_new_inode_page(struct inode *inode) 1195 { 1196 struct dnode_of_data dn; 1197 1198 /* allocate inode page for new inode */ 1199 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 1200 1201 /* caller should f2fs_put_page(page, 1); */ 1202 return f2fs_new_node_page(&dn, 0); 1203 } 1204 1205 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs) 1206 { 1207 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1208 struct node_info new_ni; 1209 struct page *page; 1210 int err; 1211 1212 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1213 return ERR_PTR(-EPERM); 1214 1215 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false); 1216 if (!page) 1217 return ERR_PTR(-ENOMEM); 1218 1219 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs)))) 1220 goto fail; 1221 1222 #ifdef CONFIG_F2FS_CHECK_FS 1223 err = f2fs_get_node_info(sbi, dn->nid, &new_ni); 1224 if (err) { 1225 dec_valid_node_count(sbi, dn->inode, !ofs); 1226 goto fail; 1227 } 1228 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR); 1229 #endif 1230 new_ni.nid = dn->nid; 1231 new_ni.ino = dn->inode->i_ino; 1232 new_ni.blk_addr = NULL_ADDR; 1233 new_ni.flag = 0; 1234 new_ni.version = 0; 1235 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 1236 1237 f2fs_wait_on_page_writeback(page, NODE, true, true); 1238 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 1239 set_cold_node(page, S_ISDIR(dn->inode->i_mode)); 1240 if (!PageUptodate(page)) 1241 SetPageUptodate(page); 1242 if (set_page_dirty(page)) 1243 dn->node_changed = true; 1244 1245 if (f2fs_has_xattr_block(ofs)) 1246 f2fs_i_xnid_write(dn->inode, dn->nid); 1247 1248 if (ofs == 0) 1249 inc_valid_inode_count(sbi); 1250 return page; 1251 1252 fail: 1253 clear_node_page_dirty(page); 1254 f2fs_put_page(page, 1); 1255 return ERR_PTR(err); 1256 } 1257 1258 /* 1259 * Caller should do after getting the following values. 1260 * 0: f2fs_put_page(page, 0) 1261 * LOCKED_PAGE or error: f2fs_put_page(page, 1) 1262 */ 1263 static int read_node_page(struct page *page, int op_flags) 1264 { 1265 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1266 struct node_info ni; 1267 struct f2fs_io_info fio = { 1268 .sbi = sbi, 1269 .type = NODE, 1270 .op = REQ_OP_READ, 1271 .op_flags = op_flags, 1272 .page = page, 1273 .encrypted_page = NULL, 1274 }; 1275 int err; 1276 1277 if (PageUptodate(page)) { 1278 #ifdef CONFIG_F2FS_CHECK_FS 1279 f2fs_bug_on(sbi, !f2fs_inode_chksum_verify(sbi, page)); 1280 #endif 1281 return LOCKED_PAGE; 1282 } 1283 1284 err = f2fs_get_node_info(sbi, page->index, &ni); 1285 if (err) 1286 return err; 1287 1288 if (unlikely(ni.blk_addr == NULL_ADDR) || 1289 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) { 1290 ClearPageUptodate(page); 1291 return -ENOENT; 1292 } 1293 1294 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr; 1295 return f2fs_submit_page_bio(&fio); 1296 } 1297 1298 /* 1299 * Readahead a node page 1300 */ 1301 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 1302 { 1303 struct page *apage; 1304 int err; 1305 1306 if (!nid) 1307 return; 1308 if (f2fs_check_nid_range(sbi, nid)) 1309 return; 1310 1311 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid); 1312 if (apage) 1313 return; 1314 1315 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1316 if (!apage) 1317 return; 1318 1319 err = read_node_page(apage, REQ_RAHEAD); 1320 f2fs_put_page(apage, err ? 1 : 0); 1321 } 1322 1323 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid, 1324 struct page *parent, int start) 1325 { 1326 struct page *page; 1327 int err; 1328 1329 if (!nid) 1330 return ERR_PTR(-ENOENT); 1331 if (f2fs_check_nid_range(sbi, nid)) 1332 return ERR_PTR(-EINVAL); 1333 repeat: 1334 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false); 1335 if (!page) 1336 return ERR_PTR(-ENOMEM); 1337 1338 err = read_node_page(page, 0); 1339 if (err < 0) { 1340 f2fs_put_page(page, 1); 1341 return ERR_PTR(err); 1342 } else if (err == LOCKED_PAGE) { 1343 err = 0; 1344 goto page_hit; 1345 } 1346 1347 if (parent) 1348 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE); 1349 1350 lock_page(page); 1351 1352 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1353 f2fs_put_page(page, 1); 1354 goto repeat; 1355 } 1356 1357 if (unlikely(!PageUptodate(page))) { 1358 err = -EIO; 1359 goto out_err; 1360 } 1361 1362 if (!f2fs_inode_chksum_verify(sbi, page)) { 1363 err = -EBADMSG; 1364 goto out_err; 1365 } 1366 page_hit: 1367 if(unlikely(nid != nid_of_node(page))) { 1368 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, " 1369 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]", 1370 nid, nid_of_node(page), ino_of_node(page), 1371 ofs_of_node(page), cpver_of_node(page), 1372 next_blkaddr_of_node(page)); 1373 err = -EINVAL; 1374 out_err: 1375 ClearPageUptodate(page); 1376 f2fs_put_page(page, 1); 1377 return ERR_PTR(err); 1378 } 1379 return page; 1380 } 1381 1382 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 1383 { 1384 return __get_node_page(sbi, nid, NULL, 0); 1385 } 1386 1387 struct page *f2fs_get_node_page_ra(struct page *parent, int start) 1388 { 1389 struct f2fs_sb_info *sbi = F2FS_P_SB(parent); 1390 nid_t nid = get_nid(parent, start, false); 1391 1392 return __get_node_page(sbi, nid, parent, start); 1393 } 1394 1395 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino) 1396 { 1397 struct inode *inode; 1398 struct page *page; 1399 int ret; 1400 1401 /* should flush inline_data before evict_inode */ 1402 inode = ilookup(sbi->sb, ino); 1403 if (!inode) 1404 return; 1405 1406 page = f2fs_pagecache_get_page(inode->i_mapping, 0, 1407 FGP_LOCK|FGP_NOWAIT, 0); 1408 if (!page) 1409 goto iput_out; 1410 1411 if (!PageUptodate(page)) 1412 goto page_out; 1413 1414 if (!PageDirty(page)) 1415 goto page_out; 1416 1417 if (!clear_page_dirty_for_io(page)) 1418 goto page_out; 1419 1420 ret = f2fs_write_inline_data(inode, page); 1421 inode_dec_dirty_pages(inode); 1422 f2fs_remove_dirty_inode(inode); 1423 if (ret) 1424 set_page_dirty(page); 1425 page_out: 1426 f2fs_put_page(page, 1); 1427 iput_out: 1428 iput(inode); 1429 } 1430 1431 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino) 1432 { 1433 pgoff_t index; 1434 struct pagevec pvec; 1435 struct page *last_page = NULL; 1436 int nr_pages; 1437 1438 pagevec_init(&pvec); 1439 index = 0; 1440 1441 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1442 PAGECACHE_TAG_DIRTY))) { 1443 int i; 1444 1445 for (i = 0; i < nr_pages; i++) { 1446 struct page *page = pvec.pages[i]; 1447 1448 if (unlikely(f2fs_cp_error(sbi))) { 1449 f2fs_put_page(last_page, 0); 1450 pagevec_release(&pvec); 1451 return ERR_PTR(-EIO); 1452 } 1453 1454 if (!IS_DNODE(page) || !is_cold_node(page)) 1455 continue; 1456 if (ino_of_node(page) != ino) 1457 continue; 1458 1459 lock_page(page); 1460 1461 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1462 continue_unlock: 1463 unlock_page(page); 1464 continue; 1465 } 1466 if (ino_of_node(page) != ino) 1467 goto continue_unlock; 1468 1469 if (!PageDirty(page)) { 1470 /* someone wrote it for us */ 1471 goto continue_unlock; 1472 } 1473 1474 if (last_page) 1475 f2fs_put_page(last_page, 0); 1476 1477 get_page(page); 1478 last_page = page; 1479 unlock_page(page); 1480 } 1481 pagevec_release(&pvec); 1482 cond_resched(); 1483 } 1484 return last_page; 1485 } 1486 1487 static int __write_node_page(struct page *page, bool atomic, bool *submitted, 1488 struct writeback_control *wbc, bool do_balance, 1489 enum iostat_type io_type, unsigned int *seq_id) 1490 { 1491 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1492 nid_t nid; 1493 struct node_info ni; 1494 struct f2fs_io_info fio = { 1495 .sbi = sbi, 1496 .ino = ino_of_node(page), 1497 .type = NODE, 1498 .op = REQ_OP_WRITE, 1499 .op_flags = wbc_to_write_flags(wbc), 1500 .page = page, 1501 .encrypted_page = NULL, 1502 .submitted = false, 1503 .io_type = io_type, 1504 .io_wbc = wbc, 1505 }; 1506 unsigned int seq; 1507 1508 trace_f2fs_writepage(page, NODE); 1509 1510 if (unlikely(f2fs_cp_error(sbi))) 1511 goto redirty_out; 1512 1513 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1514 goto redirty_out; 1515 1516 if (wbc->sync_mode == WB_SYNC_NONE && 1517 IS_DNODE(page) && is_cold_node(page)) 1518 goto redirty_out; 1519 1520 /* get old block addr of this node page */ 1521 nid = nid_of_node(page); 1522 f2fs_bug_on(sbi, page->index != nid); 1523 1524 if (f2fs_get_node_info(sbi, nid, &ni)) 1525 goto redirty_out; 1526 1527 if (wbc->for_reclaim) { 1528 if (!down_read_trylock(&sbi->node_write)) 1529 goto redirty_out; 1530 } else { 1531 down_read(&sbi->node_write); 1532 } 1533 1534 /* This page is already truncated */ 1535 if (unlikely(ni.blk_addr == NULL_ADDR)) { 1536 ClearPageUptodate(page); 1537 dec_page_count(sbi, F2FS_DIRTY_NODES); 1538 up_read(&sbi->node_write); 1539 unlock_page(page); 1540 return 0; 1541 } 1542 1543 if (__is_valid_data_blkaddr(ni.blk_addr) && 1544 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC)) { 1545 up_read(&sbi->node_write); 1546 goto redirty_out; 1547 } 1548 1549 if (atomic && !test_opt(sbi, NOBARRIER)) 1550 fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 1551 1552 set_page_writeback(page); 1553 ClearPageError(page); 1554 1555 if (f2fs_in_warm_node_list(sbi, page)) { 1556 seq = f2fs_add_fsync_node_entry(sbi, page); 1557 if (seq_id) 1558 *seq_id = seq; 1559 } 1560 1561 fio.old_blkaddr = ni.blk_addr; 1562 f2fs_do_write_node_page(nid, &fio); 1563 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page)); 1564 dec_page_count(sbi, F2FS_DIRTY_NODES); 1565 up_read(&sbi->node_write); 1566 1567 if (wbc->for_reclaim) { 1568 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE); 1569 submitted = NULL; 1570 } 1571 1572 unlock_page(page); 1573 1574 if (unlikely(f2fs_cp_error(sbi))) { 1575 f2fs_submit_merged_write(sbi, NODE); 1576 submitted = NULL; 1577 } 1578 if (submitted) 1579 *submitted = fio.submitted; 1580 1581 if (do_balance) 1582 f2fs_balance_fs(sbi, false); 1583 return 0; 1584 1585 redirty_out: 1586 redirty_page_for_writepage(wbc, page); 1587 return AOP_WRITEPAGE_ACTIVATE; 1588 } 1589 1590 int f2fs_move_node_page(struct page *node_page, int gc_type) 1591 { 1592 int err = 0; 1593 1594 if (gc_type == FG_GC) { 1595 struct writeback_control wbc = { 1596 .sync_mode = WB_SYNC_ALL, 1597 .nr_to_write = 1, 1598 .for_reclaim = 0, 1599 }; 1600 1601 f2fs_wait_on_page_writeback(node_page, NODE, true, true); 1602 1603 set_page_dirty(node_page); 1604 1605 if (!clear_page_dirty_for_io(node_page)) { 1606 err = -EAGAIN; 1607 goto out_page; 1608 } 1609 1610 if (__write_node_page(node_page, false, NULL, 1611 &wbc, false, FS_GC_NODE_IO, NULL)) { 1612 err = -EAGAIN; 1613 unlock_page(node_page); 1614 } 1615 goto release_page; 1616 } else { 1617 /* set page dirty and write it */ 1618 if (!PageWriteback(node_page)) 1619 set_page_dirty(node_page); 1620 } 1621 out_page: 1622 unlock_page(node_page); 1623 release_page: 1624 f2fs_put_page(node_page, 0); 1625 return err; 1626 } 1627 1628 static int f2fs_write_node_page(struct page *page, 1629 struct writeback_control *wbc) 1630 { 1631 return __write_node_page(page, false, NULL, wbc, false, 1632 FS_NODE_IO, NULL); 1633 } 1634 1635 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode, 1636 struct writeback_control *wbc, bool atomic, 1637 unsigned int *seq_id) 1638 { 1639 pgoff_t index; 1640 struct pagevec pvec; 1641 int ret = 0; 1642 struct page *last_page = NULL; 1643 bool marked = false; 1644 nid_t ino = inode->i_ino; 1645 int nr_pages; 1646 int nwritten = 0; 1647 1648 if (atomic) { 1649 last_page = last_fsync_dnode(sbi, ino); 1650 if (IS_ERR_OR_NULL(last_page)) 1651 return PTR_ERR_OR_ZERO(last_page); 1652 } 1653 retry: 1654 pagevec_init(&pvec); 1655 index = 0; 1656 1657 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1658 PAGECACHE_TAG_DIRTY))) { 1659 int i; 1660 1661 for (i = 0; i < nr_pages; i++) { 1662 struct page *page = pvec.pages[i]; 1663 bool submitted = false; 1664 1665 if (unlikely(f2fs_cp_error(sbi))) { 1666 f2fs_put_page(last_page, 0); 1667 pagevec_release(&pvec); 1668 ret = -EIO; 1669 goto out; 1670 } 1671 1672 if (!IS_DNODE(page) || !is_cold_node(page)) 1673 continue; 1674 if (ino_of_node(page) != ino) 1675 continue; 1676 1677 lock_page(page); 1678 1679 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1680 continue_unlock: 1681 unlock_page(page); 1682 continue; 1683 } 1684 if (ino_of_node(page) != ino) 1685 goto continue_unlock; 1686 1687 if (!PageDirty(page) && page != last_page) { 1688 /* someone wrote it for us */ 1689 goto continue_unlock; 1690 } 1691 1692 f2fs_wait_on_page_writeback(page, NODE, true, true); 1693 1694 set_fsync_mark(page, 0); 1695 set_dentry_mark(page, 0); 1696 1697 if (!atomic || page == last_page) { 1698 set_fsync_mark(page, 1); 1699 if (IS_INODE(page)) { 1700 if (is_inode_flag_set(inode, 1701 FI_DIRTY_INODE)) 1702 f2fs_update_inode(inode, page); 1703 set_dentry_mark(page, 1704 f2fs_need_dentry_mark(sbi, ino)); 1705 } 1706 /* may be written by other thread */ 1707 if (!PageDirty(page)) 1708 set_page_dirty(page); 1709 } 1710 1711 if (!clear_page_dirty_for_io(page)) 1712 goto continue_unlock; 1713 1714 ret = __write_node_page(page, atomic && 1715 page == last_page, 1716 &submitted, wbc, true, 1717 FS_NODE_IO, seq_id); 1718 if (ret) { 1719 unlock_page(page); 1720 f2fs_put_page(last_page, 0); 1721 break; 1722 } else if (submitted) { 1723 nwritten++; 1724 } 1725 1726 if (page == last_page) { 1727 f2fs_put_page(page, 0); 1728 marked = true; 1729 break; 1730 } 1731 } 1732 pagevec_release(&pvec); 1733 cond_resched(); 1734 1735 if (ret || marked) 1736 break; 1737 } 1738 if (!ret && atomic && !marked) { 1739 f2fs_msg(sbi->sb, KERN_DEBUG, 1740 "Retry to write fsync mark: ino=%u, idx=%lx", 1741 ino, last_page->index); 1742 lock_page(last_page); 1743 f2fs_wait_on_page_writeback(last_page, NODE, true, true); 1744 set_page_dirty(last_page); 1745 unlock_page(last_page); 1746 goto retry; 1747 } 1748 out: 1749 if (nwritten) 1750 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE); 1751 return ret ? -EIO: 0; 1752 } 1753 1754 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi, 1755 struct writeback_control *wbc, 1756 bool do_balance, enum iostat_type io_type) 1757 { 1758 pgoff_t index; 1759 struct pagevec pvec; 1760 int step = 0; 1761 int nwritten = 0; 1762 int ret = 0; 1763 int nr_pages, done = 0; 1764 1765 pagevec_init(&pvec); 1766 1767 next_step: 1768 index = 0; 1769 1770 while (!done && (nr_pages = pagevec_lookup_tag(&pvec, 1771 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) { 1772 int i; 1773 1774 for (i = 0; i < nr_pages; i++) { 1775 struct page *page = pvec.pages[i]; 1776 bool submitted = false; 1777 1778 /* give a priority to WB_SYNC threads */ 1779 if (atomic_read(&sbi->wb_sync_req[NODE]) && 1780 wbc->sync_mode == WB_SYNC_NONE) { 1781 done = 1; 1782 break; 1783 } 1784 1785 /* 1786 * flushing sequence with step: 1787 * 0. indirect nodes 1788 * 1. dentry dnodes 1789 * 2. file dnodes 1790 */ 1791 if (step == 0 && IS_DNODE(page)) 1792 continue; 1793 if (step == 1 && (!IS_DNODE(page) || 1794 is_cold_node(page))) 1795 continue; 1796 if (step == 2 && (!IS_DNODE(page) || 1797 !is_cold_node(page))) 1798 continue; 1799 lock_node: 1800 if (wbc->sync_mode == WB_SYNC_ALL) 1801 lock_page(page); 1802 else if (!trylock_page(page)) 1803 continue; 1804 1805 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1806 continue_unlock: 1807 unlock_page(page); 1808 continue; 1809 } 1810 1811 if (!PageDirty(page)) { 1812 /* someone wrote it for us */ 1813 goto continue_unlock; 1814 } 1815 1816 /* flush inline_data */ 1817 if (is_inline_node(page)) { 1818 clear_inline_node(page); 1819 unlock_page(page); 1820 flush_inline_data(sbi, ino_of_node(page)); 1821 goto lock_node; 1822 } 1823 1824 f2fs_wait_on_page_writeback(page, NODE, true, true); 1825 1826 if (!clear_page_dirty_for_io(page)) 1827 goto continue_unlock; 1828 1829 set_fsync_mark(page, 0); 1830 set_dentry_mark(page, 0); 1831 1832 ret = __write_node_page(page, false, &submitted, 1833 wbc, do_balance, io_type, NULL); 1834 if (ret) 1835 unlock_page(page); 1836 else if (submitted) 1837 nwritten++; 1838 1839 if (--wbc->nr_to_write == 0) 1840 break; 1841 } 1842 pagevec_release(&pvec); 1843 cond_resched(); 1844 1845 if (wbc->nr_to_write == 0) { 1846 step = 2; 1847 break; 1848 } 1849 } 1850 1851 if (step < 2) { 1852 if (wbc->sync_mode == WB_SYNC_NONE && step == 1) 1853 goto out; 1854 step++; 1855 goto next_step; 1856 } 1857 out: 1858 if (nwritten) 1859 f2fs_submit_merged_write(sbi, NODE); 1860 1861 if (unlikely(f2fs_cp_error(sbi))) 1862 return -EIO; 1863 return ret; 1864 } 1865 1866 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, 1867 unsigned int seq_id) 1868 { 1869 struct fsync_node_entry *fn; 1870 struct page *page; 1871 struct list_head *head = &sbi->fsync_node_list; 1872 unsigned long flags; 1873 unsigned int cur_seq_id = 0; 1874 int ret2, ret = 0; 1875 1876 while (seq_id && cur_seq_id < seq_id) { 1877 spin_lock_irqsave(&sbi->fsync_node_lock, flags); 1878 if (list_empty(head)) { 1879 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 1880 break; 1881 } 1882 fn = list_first_entry(head, struct fsync_node_entry, list); 1883 if (fn->seq_id > seq_id) { 1884 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 1885 break; 1886 } 1887 cur_seq_id = fn->seq_id; 1888 page = fn->page; 1889 get_page(page); 1890 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags); 1891 1892 f2fs_wait_on_page_writeback(page, NODE, true, false); 1893 if (TestClearPageError(page)) 1894 ret = -EIO; 1895 1896 put_page(page); 1897 1898 if (ret) 1899 break; 1900 } 1901 1902 ret2 = filemap_check_errors(NODE_MAPPING(sbi)); 1903 if (!ret) 1904 ret = ret2; 1905 1906 return ret; 1907 } 1908 1909 static int f2fs_write_node_pages(struct address_space *mapping, 1910 struct writeback_control *wbc) 1911 { 1912 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 1913 struct blk_plug plug; 1914 long diff; 1915 1916 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1917 goto skip_write; 1918 1919 /* balancing f2fs's metadata in background */ 1920 f2fs_balance_fs_bg(sbi); 1921 1922 /* collect a number of dirty node pages and write together */ 1923 if (wbc->sync_mode != WB_SYNC_ALL && 1924 get_pages(sbi, F2FS_DIRTY_NODES) < 1925 nr_pages_to_skip(sbi, NODE)) 1926 goto skip_write; 1927 1928 if (wbc->sync_mode == WB_SYNC_ALL) 1929 atomic_inc(&sbi->wb_sync_req[NODE]); 1930 else if (atomic_read(&sbi->wb_sync_req[NODE])) 1931 goto skip_write; 1932 1933 trace_f2fs_writepages(mapping->host, wbc, NODE); 1934 1935 diff = nr_pages_to_write(sbi, NODE, wbc); 1936 blk_start_plug(&plug); 1937 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO); 1938 blk_finish_plug(&plug); 1939 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); 1940 1941 if (wbc->sync_mode == WB_SYNC_ALL) 1942 atomic_dec(&sbi->wb_sync_req[NODE]); 1943 return 0; 1944 1945 skip_write: 1946 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES); 1947 trace_f2fs_writepages(mapping->host, wbc, NODE); 1948 return 0; 1949 } 1950 1951 static int f2fs_set_node_page_dirty(struct page *page) 1952 { 1953 trace_f2fs_set_page_dirty(page, NODE); 1954 1955 if (!PageUptodate(page)) 1956 SetPageUptodate(page); 1957 #ifdef CONFIG_F2FS_CHECK_FS 1958 if (IS_INODE(page)) 1959 f2fs_inode_chksum_set(F2FS_P_SB(page), page); 1960 #endif 1961 if (!PageDirty(page)) { 1962 __set_page_dirty_nobuffers(page); 1963 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES); 1964 f2fs_set_page_private(page, 0); 1965 f2fs_trace_pid(page); 1966 return 1; 1967 } 1968 return 0; 1969 } 1970 1971 /* 1972 * Structure of the f2fs node operations 1973 */ 1974 const struct address_space_operations f2fs_node_aops = { 1975 .writepage = f2fs_write_node_page, 1976 .writepages = f2fs_write_node_pages, 1977 .set_page_dirty = f2fs_set_node_page_dirty, 1978 .invalidatepage = f2fs_invalidate_page, 1979 .releasepage = f2fs_release_page, 1980 #ifdef CONFIG_MIGRATION 1981 .migratepage = f2fs_migrate_page, 1982 #endif 1983 }; 1984 1985 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i, 1986 nid_t n) 1987 { 1988 return radix_tree_lookup(&nm_i->free_nid_root, n); 1989 } 1990 1991 static int __insert_free_nid(struct f2fs_sb_info *sbi, 1992 struct free_nid *i, enum nid_state state) 1993 { 1994 struct f2fs_nm_info *nm_i = NM_I(sbi); 1995 1996 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i); 1997 if (err) 1998 return err; 1999 2000 f2fs_bug_on(sbi, state != i->state); 2001 nm_i->nid_cnt[state]++; 2002 if (state == FREE_NID) 2003 list_add_tail(&i->list, &nm_i->free_nid_list); 2004 return 0; 2005 } 2006 2007 static void __remove_free_nid(struct f2fs_sb_info *sbi, 2008 struct free_nid *i, enum nid_state state) 2009 { 2010 struct f2fs_nm_info *nm_i = NM_I(sbi); 2011 2012 f2fs_bug_on(sbi, state != i->state); 2013 nm_i->nid_cnt[state]--; 2014 if (state == FREE_NID) 2015 list_del(&i->list); 2016 radix_tree_delete(&nm_i->free_nid_root, i->nid); 2017 } 2018 2019 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i, 2020 enum nid_state org_state, enum nid_state dst_state) 2021 { 2022 struct f2fs_nm_info *nm_i = NM_I(sbi); 2023 2024 f2fs_bug_on(sbi, org_state != i->state); 2025 i->state = dst_state; 2026 nm_i->nid_cnt[org_state]--; 2027 nm_i->nid_cnt[dst_state]++; 2028 2029 switch (dst_state) { 2030 case PREALLOC_NID: 2031 list_del(&i->list); 2032 break; 2033 case FREE_NID: 2034 list_add_tail(&i->list, &nm_i->free_nid_list); 2035 break; 2036 default: 2037 BUG_ON(1); 2038 } 2039 } 2040 2041 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid, 2042 bool set, bool build) 2043 { 2044 struct f2fs_nm_info *nm_i = NM_I(sbi); 2045 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid); 2046 unsigned int nid_ofs = nid - START_NID(nid); 2047 2048 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap)) 2049 return; 2050 2051 if (set) { 2052 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2053 return; 2054 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2055 nm_i->free_nid_count[nat_ofs]++; 2056 } else { 2057 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs])) 2058 return; 2059 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]); 2060 if (!build) 2061 nm_i->free_nid_count[nat_ofs]--; 2062 } 2063 } 2064 2065 /* return if the nid is recognized as free */ 2066 static bool add_free_nid(struct f2fs_sb_info *sbi, 2067 nid_t nid, bool build, bool update) 2068 { 2069 struct f2fs_nm_info *nm_i = NM_I(sbi); 2070 struct free_nid *i, *e; 2071 struct nat_entry *ne; 2072 int err = -EINVAL; 2073 bool ret = false; 2074 2075 /* 0 nid should not be used */ 2076 if (unlikely(nid == 0)) 2077 return false; 2078 2079 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS); 2080 i->nid = nid; 2081 i->state = FREE_NID; 2082 2083 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 2084 2085 spin_lock(&nm_i->nid_list_lock); 2086 2087 if (build) { 2088 /* 2089 * Thread A Thread B 2090 * - f2fs_create 2091 * - f2fs_new_inode 2092 * - f2fs_alloc_nid 2093 * - __insert_nid_to_list(PREALLOC_NID) 2094 * - f2fs_balance_fs_bg 2095 * - f2fs_build_free_nids 2096 * - __f2fs_build_free_nids 2097 * - scan_nat_page 2098 * - add_free_nid 2099 * - __lookup_nat_cache 2100 * - f2fs_add_link 2101 * - f2fs_init_inode_metadata 2102 * - f2fs_new_inode_page 2103 * - f2fs_new_node_page 2104 * - set_node_addr 2105 * - f2fs_alloc_nid_done 2106 * - __remove_nid_from_list(PREALLOC_NID) 2107 * - __insert_nid_to_list(FREE_NID) 2108 */ 2109 ne = __lookup_nat_cache(nm_i, nid); 2110 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) || 2111 nat_get_blkaddr(ne) != NULL_ADDR)) 2112 goto err_out; 2113 2114 e = __lookup_free_nid_list(nm_i, nid); 2115 if (e) { 2116 if (e->state == FREE_NID) 2117 ret = true; 2118 goto err_out; 2119 } 2120 } 2121 ret = true; 2122 err = __insert_free_nid(sbi, i, FREE_NID); 2123 err_out: 2124 if (update) { 2125 update_free_nid_bitmap(sbi, nid, ret, build); 2126 if (!build) 2127 nm_i->available_nids++; 2128 } 2129 spin_unlock(&nm_i->nid_list_lock); 2130 radix_tree_preload_end(); 2131 2132 if (err) 2133 kmem_cache_free(free_nid_slab, i); 2134 return ret; 2135 } 2136 2137 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid) 2138 { 2139 struct f2fs_nm_info *nm_i = NM_I(sbi); 2140 struct free_nid *i; 2141 bool need_free = false; 2142 2143 spin_lock(&nm_i->nid_list_lock); 2144 i = __lookup_free_nid_list(nm_i, nid); 2145 if (i && i->state == FREE_NID) { 2146 __remove_free_nid(sbi, i, FREE_NID); 2147 need_free = true; 2148 } 2149 spin_unlock(&nm_i->nid_list_lock); 2150 2151 if (need_free) 2152 kmem_cache_free(free_nid_slab, i); 2153 } 2154 2155 static int scan_nat_page(struct f2fs_sb_info *sbi, 2156 struct page *nat_page, nid_t start_nid) 2157 { 2158 struct f2fs_nm_info *nm_i = NM_I(sbi); 2159 struct f2fs_nat_block *nat_blk = page_address(nat_page); 2160 block_t blk_addr; 2161 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid); 2162 int i; 2163 2164 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap); 2165 2166 i = start_nid % NAT_ENTRY_PER_BLOCK; 2167 2168 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) { 2169 if (unlikely(start_nid >= nm_i->max_nid)) 2170 break; 2171 2172 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr); 2173 2174 if (blk_addr == NEW_ADDR) 2175 return -EINVAL; 2176 2177 if (blk_addr == NULL_ADDR) { 2178 add_free_nid(sbi, start_nid, true, true); 2179 } else { 2180 spin_lock(&NM_I(sbi)->nid_list_lock); 2181 update_free_nid_bitmap(sbi, start_nid, false, true); 2182 spin_unlock(&NM_I(sbi)->nid_list_lock); 2183 } 2184 } 2185 2186 return 0; 2187 } 2188 2189 static void scan_curseg_cache(struct f2fs_sb_info *sbi) 2190 { 2191 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2192 struct f2fs_journal *journal = curseg->journal; 2193 int i; 2194 2195 down_read(&curseg->journal_rwsem); 2196 for (i = 0; i < nats_in_cursum(journal); i++) { 2197 block_t addr; 2198 nid_t nid; 2199 2200 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr); 2201 nid = le32_to_cpu(nid_in_journal(journal, i)); 2202 if (addr == NULL_ADDR) 2203 add_free_nid(sbi, nid, true, false); 2204 else 2205 remove_free_nid(sbi, nid); 2206 } 2207 up_read(&curseg->journal_rwsem); 2208 } 2209 2210 static void scan_free_nid_bits(struct f2fs_sb_info *sbi) 2211 { 2212 struct f2fs_nm_info *nm_i = NM_I(sbi); 2213 unsigned int i, idx; 2214 nid_t nid; 2215 2216 down_read(&nm_i->nat_tree_lock); 2217 2218 for (i = 0; i < nm_i->nat_blocks; i++) { 2219 if (!test_bit_le(i, nm_i->nat_block_bitmap)) 2220 continue; 2221 if (!nm_i->free_nid_count[i]) 2222 continue; 2223 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) { 2224 idx = find_next_bit_le(nm_i->free_nid_bitmap[i], 2225 NAT_ENTRY_PER_BLOCK, idx); 2226 if (idx >= NAT_ENTRY_PER_BLOCK) 2227 break; 2228 2229 nid = i * NAT_ENTRY_PER_BLOCK + idx; 2230 add_free_nid(sbi, nid, true, false); 2231 2232 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS) 2233 goto out; 2234 } 2235 } 2236 out: 2237 scan_curseg_cache(sbi); 2238 2239 up_read(&nm_i->nat_tree_lock); 2240 } 2241 2242 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi, 2243 bool sync, bool mount) 2244 { 2245 struct f2fs_nm_info *nm_i = NM_I(sbi); 2246 int i = 0, ret; 2247 nid_t nid = nm_i->next_scan_nid; 2248 2249 if (unlikely(nid >= nm_i->max_nid)) 2250 nid = 0; 2251 2252 /* Enough entries */ 2253 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2254 return 0; 2255 2256 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS)) 2257 return 0; 2258 2259 if (!mount) { 2260 /* try to find free nids in free_nid_bitmap */ 2261 scan_free_nid_bits(sbi); 2262 2263 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK) 2264 return 0; 2265 } 2266 2267 /* readahead nat pages to be scanned */ 2268 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, 2269 META_NAT, true); 2270 2271 down_read(&nm_i->nat_tree_lock); 2272 2273 while (1) { 2274 if (!test_bit_le(NAT_BLOCK_OFFSET(nid), 2275 nm_i->nat_block_bitmap)) { 2276 struct page *page = get_current_nat_page(sbi, nid); 2277 2278 if (IS_ERR(page)) { 2279 ret = PTR_ERR(page); 2280 } else { 2281 ret = scan_nat_page(sbi, page, nid); 2282 f2fs_put_page(page, 1); 2283 } 2284 2285 if (ret) { 2286 up_read(&nm_i->nat_tree_lock); 2287 f2fs_bug_on(sbi, !mount); 2288 f2fs_msg(sbi->sb, KERN_ERR, 2289 "NAT is corrupt, run fsck to fix it"); 2290 return ret; 2291 } 2292 } 2293 2294 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK)); 2295 if (unlikely(nid >= nm_i->max_nid)) 2296 nid = 0; 2297 2298 if (++i >= FREE_NID_PAGES) 2299 break; 2300 } 2301 2302 /* go to the next free nat pages to find free nids abundantly */ 2303 nm_i->next_scan_nid = nid; 2304 2305 /* find free nids from current sum_pages */ 2306 scan_curseg_cache(sbi); 2307 2308 up_read(&nm_i->nat_tree_lock); 2309 2310 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid), 2311 nm_i->ra_nid_pages, META_NAT, false); 2312 2313 return 0; 2314 } 2315 2316 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount) 2317 { 2318 int ret; 2319 2320 mutex_lock(&NM_I(sbi)->build_lock); 2321 ret = __f2fs_build_free_nids(sbi, sync, mount); 2322 mutex_unlock(&NM_I(sbi)->build_lock); 2323 2324 return ret; 2325 } 2326 2327 /* 2328 * If this function returns success, caller can obtain a new nid 2329 * from second parameter of this function. 2330 * The returned nid could be used ino as well as nid when inode is created. 2331 */ 2332 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid) 2333 { 2334 struct f2fs_nm_info *nm_i = NM_I(sbi); 2335 struct free_nid *i = NULL; 2336 retry: 2337 if (time_to_inject(sbi, FAULT_ALLOC_NID)) { 2338 f2fs_show_injection_info(FAULT_ALLOC_NID); 2339 return false; 2340 } 2341 2342 spin_lock(&nm_i->nid_list_lock); 2343 2344 if (unlikely(nm_i->available_nids == 0)) { 2345 spin_unlock(&nm_i->nid_list_lock); 2346 return false; 2347 } 2348 2349 /* We should not use stale free nids created by f2fs_build_free_nids */ 2350 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) { 2351 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list)); 2352 i = list_first_entry(&nm_i->free_nid_list, 2353 struct free_nid, list); 2354 *nid = i->nid; 2355 2356 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID); 2357 nm_i->available_nids--; 2358 2359 update_free_nid_bitmap(sbi, *nid, false, false); 2360 2361 spin_unlock(&nm_i->nid_list_lock); 2362 return true; 2363 } 2364 spin_unlock(&nm_i->nid_list_lock); 2365 2366 /* Let's scan nat pages and its caches to get free nids */ 2367 if (!f2fs_build_free_nids(sbi, true, false)) 2368 goto retry; 2369 return false; 2370 } 2371 2372 /* 2373 * f2fs_alloc_nid() should be called prior to this function. 2374 */ 2375 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 2376 { 2377 struct f2fs_nm_info *nm_i = NM_I(sbi); 2378 struct free_nid *i; 2379 2380 spin_lock(&nm_i->nid_list_lock); 2381 i = __lookup_free_nid_list(nm_i, nid); 2382 f2fs_bug_on(sbi, !i); 2383 __remove_free_nid(sbi, i, PREALLOC_NID); 2384 spin_unlock(&nm_i->nid_list_lock); 2385 2386 kmem_cache_free(free_nid_slab, i); 2387 } 2388 2389 /* 2390 * f2fs_alloc_nid() should be called prior to this function. 2391 */ 2392 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 2393 { 2394 struct f2fs_nm_info *nm_i = NM_I(sbi); 2395 struct free_nid *i; 2396 bool need_free = false; 2397 2398 if (!nid) 2399 return; 2400 2401 spin_lock(&nm_i->nid_list_lock); 2402 i = __lookup_free_nid_list(nm_i, nid); 2403 f2fs_bug_on(sbi, !i); 2404 2405 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) { 2406 __remove_free_nid(sbi, i, PREALLOC_NID); 2407 need_free = true; 2408 } else { 2409 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID); 2410 } 2411 2412 nm_i->available_nids++; 2413 2414 update_free_nid_bitmap(sbi, nid, true, false); 2415 2416 spin_unlock(&nm_i->nid_list_lock); 2417 2418 if (need_free) 2419 kmem_cache_free(free_nid_slab, i); 2420 } 2421 2422 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink) 2423 { 2424 struct f2fs_nm_info *nm_i = NM_I(sbi); 2425 struct free_nid *i, *next; 2426 int nr = nr_shrink; 2427 2428 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2429 return 0; 2430 2431 if (!mutex_trylock(&nm_i->build_lock)) 2432 return 0; 2433 2434 spin_lock(&nm_i->nid_list_lock); 2435 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) { 2436 if (nr_shrink <= 0 || 2437 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS) 2438 break; 2439 2440 __remove_free_nid(sbi, i, FREE_NID); 2441 kmem_cache_free(free_nid_slab, i); 2442 nr_shrink--; 2443 } 2444 spin_unlock(&nm_i->nid_list_lock); 2445 mutex_unlock(&nm_i->build_lock); 2446 2447 return nr - nr_shrink; 2448 } 2449 2450 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page) 2451 { 2452 void *src_addr, *dst_addr; 2453 size_t inline_size; 2454 struct page *ipage; 2455 struct f2fs_inode *ri; 2456 2457 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 2458 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage)); 2459 2460 ri = F2FS_INODE(page); 2461 if (ri->i_inline & F2FS_INLINE_XATTR) { 2462 set_inode_flag(inode, FI_INLINE_XATTR); 2463 } else { 2464 clear_inode_flag(inode, FI_INLINE_XATTR); 2465 goto update_inode; 2466 } 2467 2468 dst_addr = inline_xattr_addr(inode, ipage); 2469 src_addr = inline_xattr_addr(inode, page); 2470 inline_size = inline_xattr_size(inode); 2471 2472 f2fs_wait_on_page_writeback(ipage, NODE, true, true); 2473 memcpy(dst_addr, src_addr, inline_size); 2474 update_inode: 2475 f2fs_update_inode(inode, ipage); 2476 f2fs_put_page(ipage, 1); 2477 } 2478 2479 int f2fs_recover_xattr_data(struct inode *inode, struct page *page) 2480 { 2481 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2482 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid; 2483 nid_t new_xnid; 2484 struct dnode_of_data dn; 2485 struct node_info ni; 2486 struct page *xpage; 2487 int err; 2488 2489 if (!prev_xnid) 2490 goto recover_xnid; 2491 2492 /* 1: invalidate the previous xattr nid */ 2493 err = f2fs_get_node_info(sbi, prev_xnid, &ni); 2494 if (err) 2495 return err; 2496 2497 f2fs_invalidate_blocks(sbi, ni.blk_addr); 2498 dec_valid_node_count(sbi, inode, false); 2499 set_node_addr(sbi, &ni, NULL_ADDR, false); 2500 2501 recover_xnid: 2502 /* 2: update xattr nid in inode */ 2503 if (!f2fs_alloc_nid(sbi, &new_xnid)) 2504 return -ENOSPC; 2505 2506 set_new_dnode(&dn, inode, NULL, NULL, new_xnid); 2507 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET); 2508 if (IS_ERR(xpage)) { 2509 f2fs_alloc_nid_failed(sbi, new_xnid); 2510 return PTR_ERR(xpage); 2511 } 2512 2513 f2fs_alloc_nid_done(sbi, new_xnid); 2514 f2fs_update_inode_page(inode); 2515 2516 /* 3: update and set xattr node page dirty */ 2517 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE); 2518 2519 set_page_dirty(xpage); 2520 f2fs_put_page(xpage, 1); 2521 2522 return 0; 2523 } 2524 2525 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 2526 { 2527 struct f2fs_inode *src, *dst; 2528 nid_t ino = ino_of_node(page); 2529 struct node_info old_ni, new_ni; 2530 struct page *ipage; 2531 int err; 2532 2533 err = f2fs_get_node_info(sbi, ino, &old_ni); 2534 if (err) 2535 return err; 2536 2537 if (unlikely(old_ni.blk_addr != NULL_ADDR)) 2538 return -EINVAL; 2539 retry: 2540 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false); 2541 if (!ipage) { 2542 congestion_wait(BLK_RW_ASYNC, HZ/50); 2543 goto retry; 2544 } 2545 2546 /* Should not use this inode from free nid list */ 2547 remove_free_nid(sbi, ino); 2548 2549 if (!PageUptodate(ipage)) 2550 SetPageUptodate(ipage); 2551 fill_node_footer(ipage, ino, ino, 0, true); 2552 set_cold_node(ipage, false); 2553 2554 src = F2FS_INODE(page); 2555 dst = F2FS_INODE(ipage); 2556 2557 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src); 2558 dst->i_size = 0; 2559 dst->i_blocks = cpu_to_le64(1); 2560 dst->i_links = cpu_to_le32(1); 2561 dst->i_xattr_nid = 0; 2562 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR); 2563 if (dst->i_inline & F2FS_EXTRA_ATTR) { 2564 dst->i_extra_isize = src->i_extra_isize; 2565 2566 if (f2fs_sb_has_flexible_inline_xattr(sbi) && 2567 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2568 i_inline_xattr_size)) 2569 dst->i_inline_xattr_size = src->i_inline_xattr_size; 2570 2571 if (f2fs_sb_has_project_quota(sbi) && 2572 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2573 i_projid)) 2574 dst->i_projid = src->i_projid; 2575 2576 if (f2fs_sb_has_inode_crtime(sbi) && 2577 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize), 2578 i_crtime_nsec)) { 2579 dst->i_crtime = src->i_crtime; 2580 dst->i_crtime_nsec = src->i_crtime_nsec; 2581 } 2582 } 2583 2584 new_ni = old_ni; 2585 new_ni.ino = ino; 2586 2587 if (unlikely(inc_valid_node_count(sbi, NULL, true))) 2588 WARN_ON(1); 2589 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 2590 inc_valid_inode_count(sbi); 2591 set_page_dirty(ipage); 2592 f2fs_put_page(ipage, 1); 2593 return 0; 2594 } 2595 2596 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi, 2597 unsigned int segno, struct f2fs_summary_block *sum) 2598 { 2599 struct f2fs_node *rn; 2600 struct f2fs_summary *sum_entry; 2601 block_t addr; 2602 int i, idx, last_offset, nrpages; 2603 2604 /* scan the node segment */ 2605 last_offset = sbi->blocks_per_seg; 2606 addr = START_BLOCK(sbi, segno); 2607 sum_entry = &sum->entries[0]; 2608 2609 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) { 2610 nrpages = min(last_offset - i, BIO_MAX_PAGES); 2611 2612 /* readahead node pages */ 2613 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true); 2614 2615 for (idx = addr; idx < addr + nrpages; idx++) { 2616 struct page *page = f2fs_get_tmp_page(sbi, idx); 2617 2618 if (IS_ERR(page)) 2619 return PTR_ERR(page); 2620 2621 rn = F2FS_NODE(page); 2622 sum_entry->nid = rn->footer.nid; 2623 sum_entry->version = 0; 2624 sum_entry->ofs_in_node = 0; 2625 sum_entry++; 2626 f2fs_put_page(page, 1); 2627 } 2628 2629 invalidate_mapping_pages(META_MAPPING(sbi), addr, 2630 addr + nrpages); 2631 } 2632 return 0; 2633 } 2634 2635 static void remove_nats_in_journal(struct f2fs_sb_info *sbi) 2636 { 2637 struct f2fs_nm_info *nm_i = NM_I(sbi); 2638 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2639 struct f2fs_journal *journal = curseg->journal; 2640 int i; 2641 2642 down_write(&curseg->journal_rwsem); 2643 for (i = 0; i < nats_in_cursum(journal); i++) { 2644 struct nat_entry *ne; 2645 struct f2fs_nat_entry raw_ne; 2646 nid_t nid = le32_to_cpu(nid_in_journal(journal, i)); 2647 2648 raw_ne = nat_in_journal(journal, i); 2649 2650 ne = __lookup_nat_cache(nm_i, nid); 2651 if (!ne) { 2652 ne = __alloc_nat_entry(nid, true); 2653 __init_nat_entry(nm_i, ne, &raw_ne, true); 2654 } 2655 2656 /* 2657 * if a free nat in journal has not been used after last 2658 * checkpoint, we should remove it from available nids, 2659 * since later we will add it again. 2660 */ 2661 if (!get_nat_flag(ne, IS_DIRTY) && 2662 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) { 2663 spin_lock(&nm_i->nid_list_lock); 2664 nm_i->available_nids--; 2665 spin_unlock(&nm_i->nid_list_lock); 2666 } 2667 2668 __set_nat_cache_dirty(nm_i, ne); 2669 } 2670 update_nats_in_cursum(journal, -i); 2671 up_write(&curseg->journal_rwsem); 2672 } 2673 2674 static void __adjust_nat_entry_set(struct nat_entry_set *nes, 2675 struct list_head *head, int max) 2676 { 2677 struct nat_entry_set *cur; 2678 2679 if (nes->entry_cnt >= max) 2680 goto add_out; 2681 2682 list_for_each_entry(cur, head, set_list) { 2683 if (cur->entry_cnt >= nes->entry_cnt) { 2684 list_add(&nes->set_list, cur->set_list.prev); 2685 return; 2686 } 2687 } 2688 add_out: 2689 list_add_tail(&nes->set_list, head); 2690 } 2691 2692 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid, 2693 struct page *page) 2694 { 2695 struct f2fs_nm_info *nm_i = NM_I(sbi); 2696 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK; 2697 struct f2fs_nat_block *nat_blk = page_address(page); 2698 int valid = 0; 2699 int i = 0; 2700 2701 if (!enabled_nat_bits(sbi, NULL)) 2702 return; 2703 2704 if (nat_index == 0) { 2705 valid = 1; 2706 i = 1; 2707 } 2708 for (; i < NAT_ENTRY_PER_BLOCK; i++) { 2709 if (nat_blk->entries[i].block_addr != NULL_ADDR) 2710 valid++; 2711 } 2712 if (valid == 0) { 2713 __set_bit_le(nat_index, nm_i->empty_nat_bits); 2714 __clear_bit_le(nat_index, nm_i->full_nat_bits); 2715 return; 2716 } 2717 2718 __clear_bit_le(nat_index, nm_i->empty_nat_bits); 2719 if (valid == NAT_ENTRY_PER_BLOCK) 2720 __set_bit_le(nat_index, nm_i->full_nat_bits); 2721 else 2722 __clear_bit_le(nat_index, nm_i->full_nat_bits); 2723 } 2724 2725 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi, 2726 struct nat_entry_set *set, struct cp_control *cpc) 2727 { 2728 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2729 struct f2fs_journal *journal = curseg->journal; 2730 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK; 2731 bool to_journal = true; 2732 struct f2fs_nat_block *nat_blk; 2733 struct nat_entry *ne, *cur; 2734 struct page *page = NULL; 2735 2736 /* 2737 * there are two steps to flush nat entries: 2738 * #1, flush nat entries to journal in current hot data summary block. 2739 * #2, flush nat entries to nat page. 2740 */ 2741 if (enabled_nat_bits(sbi, cpc) || 2742 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL)) 2743 to_journal = false; 2744 2745 if (to_journal) { 2746 down_write(&curseg->journal_rwsem); 2747 } else { 2748 page = get_next_nat_page(sbi, start_nid); 2749 if (IS_ERR(page)) 2750 return PTR_ERR(page); 2751 2752 nat_blk = page_address(page); 2753 f2fs_bug_on(sbi, !nat_blk); 2754 } 2755 2756 /* flush dirty nats in nat entry set */ 2757 list_for_each_entry_safe(ne, cur, &set->entry_list, list) { 2758 struct f2fs_nat_entry *raw_ne; 2759 nid_t nid = nat_get_nid(ne); 2760 int offset; 2761 2762 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR); 2763 2764 if (to_journal) { 2765 offset = f2fs_lookup_journal_in_cursum(journal, 2766 NAT_JOURNAL, nid, 1); 2767 f2fs_bug_on(sbi, offset < 0); 2768 raw_ne = &nat_in_journal(journal, offset); 2769 nid_in_journal(journal, offset) = cpu_to_le32(nid); 2770 } else { 2771 raw_ne = &nat_blk->entries[nid - start_nid]; 2772 } 2773 raw_nat_from_node_info(raw_ne, &ne->ni); 2774 nat_reset_flag(ne); 2775 __clear_nat_cache_dirty(NM_I(sbi), set, ne); 2776 if (nat_get_blkaddr(ne) == NULL_ADDR) { 2777 add_free_nid(sbi, nid, false, true); 2778 } else { 2779 spin_lock(&NM_I(sbi)->nid_list_lock); 2780 update_free_nid_bitmap(sbi, nid, false, false); 2781 spin_unlock(&NM_I(sbi)->nid_list_lock); 2782 } 2783 } 2784 2785 if (to_journal) { 2786 up_write(&curseg->journal_rwsem); 2787 } else { 2788 __update_nat_bits(sbi, start_nid, page); 2789 f2fs_put_page(page, 1); 2790 } 2791 2792 /* Allow dirty nats by node block allocation in write_begin */ 2793 if (!set->entry_cnt) { 2794 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set); 2795 kmem_cache_free(nat_entry_set_slab, set); 2796 } 2797 return 0; 2798 } 2799 2800 /* 2801 * This function is called during the checkpointing process. 2802 */ 2803 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 2804 { 2805 struct f2fs_nm_info *nm_i = NM_I(sbi); 2806 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 2807 struct f2fs_journal *journal = curseg->journal; 2808 struct nat_entry_set *setvec[SETVEC_SIZE]; 2809 struct nat_entry_set *set, *tmp; 2810 unsigned int found; 2811 nid_t set_idx = 0; 2812 LIST_HEAD(sets); 2813 int err = 0; 2814 2815 /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */ 2816 if (enabled_nat_bits(sbi, cpc)) { 2817 down_write(&nm_i->nat_tree_lock); 2818 remove_nats_in_journal(sbi); 2819 up_write(&nm_i->nat_tree_lock); 2820 } 2821 2822 if (!nm_i->dirty_nat_cnt) 2823 return 0; 2824 2825 down_write(&nm_i->nat_tree_lock); 2826 2827 /* 2828 * if there are no enough space in journal to store dirty nat 2829 * entries, remove all entries from journal and merge them 2830 * into nat entry set. 2831 */ 2832 if (enabled_nat_bits(sbi, cpc) || 2833 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL)) 2834 remove_nats_in_journal(sbi); 2835 2836 while ((found = __gang_lookup_nat_set(nm_i, 2837 set_idx, SETVEC_SIZE, setvec))) { 2838 unsigned idx; 2839 set_idx = setvec[found - 1]->set + 1; 2840 for (idx = 0; idx < found; idx++) 2841 __adjust_nat_entry_set(setvec[idx], &sets, 2842 MAX_NAT_JENTRIES(journal)); 2843 } 2844 2845 /* flush dirty nats in nat entry set */ 2846 list_for_each_entry_safe(set, tmp, &sets, set_list) { 2847 err = __flush_nat_entry_set(sbi, set, cpc); 2848 if (err) 2849 break; 2850 } 2851 2852 up_write(&nm_i->nat_tree_lock); 2853 /* Allow dirty nats by node block allocation in write_begin */ 2854 2855 return err; 2856 } 2857 2858 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi) 2859 { 2860 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2861 struct f2fs_nm_info *nm_i = NM_I(sbi); 2862 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE; 2863 unsigned int i; 2864 __u64 cp_ver = cur_cp_version(ckpt); 2865 block_t nat_bits_addr; 2866 2867 if (!enabled_nat_bits(sbi, NULL)) 2868 return 0; 2869 2870 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8); 2871 nm_i->nat_bits = f2fs_kzalloc(sbi, 2872 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL); 2873 if (!nm_i->nat_bits) 2874 return -ENOMEM; 2875 2876 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg - 2877 nm_i->nat_bits_blocks; 2878 for (i = 0; i < nm_i->nat_bits_blocks; i++) { 2879 struct page *page; 2880 2881 page = f2fs_get_meta_page(sbi, nat_bits_addr++); 2882 if (IS_ERR(page)) 2883 return PTR_ERR(page); 2884 2885 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS), 2886 page_address(page), F2FS_BLKSIZE); 2887 f2fs_put_page(page, 1); 2888 } 2889 2890 cp_ver |= (cur_cp_crc(ckpt) << 32); 2891 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) { 2892 disable_nat_bits(sbi, true); 2893 return 0; 2894 } 2895 2896 nm_i->full_nat_bits = nm_i->nat_bits + 8; 2897 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes; 2898 2899 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint"); 2900 return 0; 2901 } 2902 2903 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi) 2904 { 2905 struct f2fs_nm_info *nm_i = NM_I(sbi); 2906 unsigned int i = 0; 2907 nid_t nid, last_nid; 2908 2909 if (!enabled_nat_bits(sbi, NULL)) 2910 return; 2911 2912 for (i = 0; i < nm_i->nat_blocks; i++) { 2913 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i); 2914 if (i >= nm_i->nat_blocks) 2915 break; 2916 2917 __set_bit_le(i, nm_i->nat_block_bitmap); 2918 2919 nid = i * NAT_ENTRY_PER_BLOCK; 2920 last_nid = nid + NAT_ENTRY_PER_BLOCK; 2921 2922 spin_lock(&NM_I(sbi)->nid_list_lock); 2923 for (; nid < last_nid; nid++) 2924 update_free_nid_bitmap(sbi, nid, true, true); 2925 spin_unlock(&NM_I(sbi)->nid_list_lock); 2926 } 2927 2928 for (i = 0; i < nm_i->nat_blocks; i++) { 2929 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i); 2930 if (i >= nm_i->nat_blocks) 2931 break; 2932 2933 __set_bit_le(i, nm_i->nat_block_bitmap); 2934 } 2935 } 2936 2937 static int init_node_manager(struct f2fs_sb_info *sbi) 2938 { 2939 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 2940 struct f2fs_nm_info *nm_i = NM_I(sbi); 2941 unsigned char *version_bitmap; 2942 unsigned int nat_segs; 2943 int err; 2944 2945 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 2946 2947 /* segment_count_nat includes pair segment so divide to 2. */ 2948 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 2949 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 2950 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks; 2951 2952 /* not used nids: 0, node, meta, (and root counted as valid node) */ 2953 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count - 2954 sbi->nquota_files - F2FS_RESERVED_NODE_NUM; 2955 nm_i->nid_cnt[FREE_NID] = 0; 2956 nm_i->nid_cnt[PREALLOC_NID] = 0; 2957 nm_i->nat_cnt = 0; 2958 nm_i->ram_thresh = DEF_RAM_THRESHOLD; 2959 nm_i->ra_nid_pages = DEF_RA_NID_PAGES; 2960 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD; 2961 2962 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC); 2963 INIT_LIST_HEAD(&nm_i->free_nid_list); 2964 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO); 2965 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO); 2966 INIT_LIST_HEAD(&nm_i->nat_entries); 2967 spin_lock_init(&nm_i->nat_list_lock); 2968 2969 mutex_init(&nm_i->build_lock); 2970 spin_lock_init(&nm_i->nid_list_lock); 2971 init_rwsem(&nm_i->nat_tree_lock); 2972 2973 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 2974 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 2975 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 2976 if (!version_bitmap) 2977 return -EFAULT; 2978 2979 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size, 2980 GFP_KERNEL); 2981 if (!nm_i->nat_bitmap) 2982 return -ENOMEM; 2983 2984 err = __get_nat_bitmaps(sbi); 2985 if (err) 2986 return err; 2987 2988 #ifdef CONFIG_F2FS_CHECK_FS 2989 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size, 2990 GFP_KERNEL); 2991 if (!nm_i->nat_bitmap_mir) 2992 return -ENOMEM; 2993 #endif 2994 2995 return 0; 2996 } 2997 2998 static int init_free_nid_cache(struct f2fs_sb_info *sbi) 2999 { 3000 struct f2fs_nm_info *nm_i = NM_I(sbi); 3001 int i; 3002 3003 nm_i->free_nid_bitmap = 3004 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *), 3005 nm_i->nat_blocks), 3006 GFP_KERNEL); 3007 if (!nm_i->free_nid_bitmap) 3008 return -ENOMEM; 3009 3010 for (i = 0; i < nm_i->nat_blocks; i++) { 3011 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi, 3012 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL); 3013 if (!nm_i->free_nid_bitmap[i]) 3014 return -ENOMEM; 3015 } 3016 3017 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8, 3018 GFP_KERNEL); 3019 if (!nm_i->nat_block_bitmap) 3020 return -ENOMEM; 3021 3022 nm_i->free_nid_count = 3023 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short), 3024 nm_i->nat_blocks), 3025 GFP_KERNEL); 3026 if (!nm_i->free_nid_count) 3027 return -ENOMEM; 3028 return 0; 3029 } 3030 3031 int f2fs_build_node_manager(struct f2fs_sb_info *sbi) 3032 { 3033 int err; 3034 3035 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info), 3036 GFP_KERNEL); 3037 if (!sbi->nm_info) 3038 return -ENOMEM; 3039 3040 err = init_node_manager(sbi); 3041 if (err) 3042 return err; 3043 3044 err = init_free_nid_cache(sbi); 3045 if (err) 3046 return err; 3047 3048 /* load free nid status from nat_bits table */ 3049 load_free_nid_bitmap(sbi); 3050 3051 return f2fs_build_free_nids(sbi, true, true); 3052 } 3053 3054 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi) 3055 { 3056 struct f2fs_nm_info *nm_i = NM_I(sbi); 3057 struct free_nid *i, *next_i; 3058 struct nat_entry *natvec[NATVEC_SIZE]; 3059 struct nat_entry_set *setvec[SETVEC_SIZE]; 3060 nid_t nid = 0; 3061 unsigned int found; 3062 3063 if (!nm_i) 3064 return; 3065 3066 /* destroy free nid list */ 3067 spin_lock(&nm_i->nid_list_lock); 3068 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 3069 __remove_free_nid(sbi, i, FREE_NID); 3070 spin_unlock(&nm_i->nid_list_lock); 3071 kmem_cache_free(free_nid_slab, i); 3072 spin_lock(&nm_i->nid_list_lock); 3073 } 3074 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]); 3075 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]); 3076 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list)); 3077 spin_unlock(&nm_i->nid_list_lock); 3078 3079 /* destroy nat cache */ 3080 down_write(&nm_i->nat_tree_lock); 3081 while ((found = __gang_lookup_nat_cache(nm_i, 3082 nid, NATVEC_SIZE, natvec))) { 3083 unsigned idx; 3084 3085 nid = nat_get_nid(natvec[found - 1]) + 1; 3086 for (idx = 0; idx < found; idx++) { 3087 spin_lock(&nm_i->nat_list_lock); 3088 list_del(&natvec[idx]->list); 3089 spin_unlock(&nm_i->nat_list_lock); 3090 3091 __del_from_nat_cache(nm_i, natvec[idx]); 3092 } 3093 } 3094 f2fs_bug_on(sbi, nm_i->nat_cnt); 3095 3096 /* destroy nat set cache */ 3097 nid = 0; 3098 while ((found = __gang_lookup_nat_set(nm_i, 3099 nid, SETVEC_SIZE, setvec))) { 3100 unsigned idx; 3101 3102 nid = setvec[found - 1]->set + 1; 3103 for (idx = 0; idx < found; idx++) { 3104 /* entry_cnt is not zero, when cp_error was occurred */ 3105 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list)); 3106 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set); 3107 kmem_cache_free(nat_entry_set_slab, setvec[idx]); 3108 } 3109 } 3110 up_write(&nm_i->nat_tree_lock); 3111 3112 kvfree(nm_i->nat_block_bitmap); 3113 if (nm_i->free_nid_bitmap) { 3114 int i; 3115 3116 for (i = 0; i < nm_i->nat_blocks; i++) 3117 kvfree(nm_i->free_nid_bitmap[i]); 3118 kvfree(nm_i->free_nid_bitmap); 3119 } 3120 kvfree(nm_i->free_nid_count); 3121 3122 kvfree(nm_i->nat_bitmap); 3123 kvfree(nm_i->nat_bits); 3124 #ifdef CONFIG_F2FS_CHECK_FS 3125 kvfree(nm_i->nat_bitmap_mir); 3126 #endif 3127 sbi->nm_info = NULL; 3128 kvfree(nm_i); 3129 } 3130 3131 int __init f2fs_create_node_manager_caches(void) 3132 { 3133 nat_entry_slab = f2fs_kmem_cache_create("nat_entry", 3134 sizeof(struct nat_entry)); 3135 if (!nat_entry_slab) 3136 goto fail; 3137 3138 free_nid_slab = f2fs_kmem_cache_create("free_nid", 3139 sizeof(struct free_nid)); 3140 if (!free_nid_slab) 3141 goto destroy_nat_entry; 3142 3143 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set", 3144 sizeof(struct nat_entry_set)); 3145 if (!nat_entry_set_slab) 3146 goto destroy_free_nid; 3147 3148 fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry", 3149 sizeof(struct fsync_node_entry)); 3150 if (!fsync_node_entry_slab) 3151 goto destroy_nat_entry_set; 3152 return 0; 3153 3154 destroy_nat_entry_set: 3155 kmem_cache_destroy(nat_entry_set_slab); 3156 destroy_free_nid: 3157 kmem_cache_destroy(free_nid_slab); 3158 destroy_nat_entry: 3159 kmem_cache_destroy(nat_entry_slab); 3160 fail: 3161 return -ENOMEM; 3162 } 3163 3164 void f2fs_destroy_node_manager_caches(void) 3165 { 3166 kmem_cache_destroy(fsync_node_entry_slab); 3167 kmem_cache_destroy(nat_entry_set_slab); 3168 kmem_cache_destroy(free_nid_slab); 3169 kmem_cache_destroy(nat_entry_slab); 3170 } 3171