1 /* 2 * fs/f2fs/node.c 3 * 4 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 5 * http://www.samsung.com/ 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 */ 11 #include <linux/fs.h> 12 #include <linux/f2fs_fs.h> 13 #include <linux/mpage.h> 14 #include <linux/backing-dev.h> 15 #include <linux/blkdev.h> 16 #include <linux/pagevec.h> 17 #include <linux/swap.h> 18 19 #include "f2fs.h" 20 #include "node.h" 21 #include "segment.h" 22 #include <trace/events/f2fs.h> 23 24 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock) 25 26 static struct kmem_cache *nat_entry_slab; 27 static struct kmem_cache *free_nid_slab; 28 29 static inline bool available_free_memory(struct f2fs_nm_info *nm_i, int type) 30 { 31 struct sysinfo val; 32 unsigned long mem_size = 0; 33 34 si_meminfo(&val); 35 if (type == FREE_NIDS) 36 mem_size = nm_i->fcnt * sizeof(struct free_nid); 37 else if (type == NAT_ENTRIES) 38 mem_size += nm_i->nat_cnt * sizeof(struct nat_entry); 39 mem_size >>= 12; 40 41 /* give 50:50 memory for free nids and nat caches respectively */ 42 return (mem_size < ((val.totalram * nm_i->ram_thresh) >> 11)); 43 } 44 45 static void clear_node_page_dirty(struct page *page) 46 { 47 struct address_space *mapping = page->mapping; 48 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); 49 unsigned int long flags; 50 51 if (PageDirty(page)) { 52 spin_lock_irqsave(&mapping->tree_lock, flags); 53 radix_tree_tag_clear(&mapping->page_tree, 54 page_index(page), 55 PAGECACHE_TAG_DIRTY); 56 spin_unlock_irqrestore(&mapping->tree_lock, flags); 57 58 clear_page_dirty_for_io(page); 59 dec_page_count(sbi, F2FS_DIRTY_NODES); 60 } 61 ClearPageUptodate(page); 62 } 63 64 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 65 { 66 pgoff_t index = current_nat_addr(sbi, nid); 67 return get_meta_page(sbi, index); 68 } 69 70 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 71 { 72 struct page *src_page; 73 struct page *dst_page; 74 pgoff_t src_off; 75 pgoff_t dst_off; 76 void *src_addr; 77 void *dst_addr; 78 struct f2fs_nm_info *nm_i = NM_I(sbi); 79 80 src_off = current_nat_addr(sbi, nid); 81 dst_off = next_nat_addr(sbi, src_off); 82 83 /* get current nat block page with lock */ 84 src_page = get_meta_page(sbi, src_off); 85 86 /* Dirty src_page means that it is already the new target NAT page. */ 87 if (PageDirty(src_page)) 88 return src_page; 89 90 dst_page = grab_meta_page(sbi, dst_off); 91 92 src_addr = page_address(src_page); 93 dst_addr = page_address(dst_page); 94 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); 95 set_page_dirty(dst_page); 96 f2fs_put_page(src_page, 1); 97 98 set_to_next_nat(nm_i, nid); 99 100 return dst_page; 101 } 102 103 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n) 104 { 105 return radix_tree_lookup(&nm_i->nat_root, n); 106 } 107 108 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i, 109 nid_t start, unsigned int nr, struct nat_entry **ep) 110 { 111 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr); 112 } 113 114 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e) 115 { 116 list_del(&e->list); 117 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e)); 118 nm_i->nat_cnt--; 119 kmem_cache_free(nat_entry_slab, e); 120 } 121 122 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid) 123 { 124 struct f2fs_nm_info *nm_i = NM_I(sbi); 125 struct nat_entry *e; 126 int is_cp = 1; 127 128 read_lock(&nm_i->nat_tree_lock); 129 e = __lookup_nat_cache(nm_i, nid); 130 if (e && !e->checkpointed) 131 is_cp = 0; 132 read_unlock(&nm_i->nat_tree_lock); 133 return is_cp; 134 } 135 136 bool fsync_mark_done(struct f2fs_sb_info *sbi, nid_t nid) 137 { 138 struct f2fs_nm_info *nm_i = NM_I(sbi); 139 struct nat_entry *e; 140 bool fsync_done = false; 141 142 read_lock(&nm_i->nat_tree_lock); 143 e = __lookup_nat_cache(nm_i, nid); 144 if (e) 145 fsync_done = e->fsync_done; 146 read_unlock(&nm_i->nat_tree_lock); 147 return fsync_done; 148 } 149 150 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid) 151 { 152 struct nat_entry *new; 153 154 new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC); 155 if (!new) 156 return NULL; 157 if (radix_tree_insert(&nm_i->nat_root, nid, new)) { 158 kmem_cache_free(nat_entry_slab, new); 159 return NULL; 160 } 161 memset(new, 0, sizeof(struct nat_entry)); 162 nat_set_nid(new, nid); 163 new->checkpointed = true; 164 list_add_tail(&new->list, &nm_i->nat_entries); 165 nm_i->nat_cnt++; 166 return new; 167 } 168 169 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid, 170 struct f2fs_nat_entry *ne) 171 { 172 struct nat_entry *e; 173 retry: 174 write_lock(&nm_i->nat_tree_lock); 175 e = __lookup_nat_cache(nm_i, nid); 176 if (!e) { 177 e = grab_nat_entry(nm_i, nid); 178 if (!e) { 179 write_unlock(&nm_i->nat_tree_lock); 180 goto retry; 181 } 182 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr)); 183 nat_set_ino(e, le32_to_cpu(ne->ino)); 184 nat_set_version(e, ne->version); 185 } 186 write_unlock(&nm_i->nat_tree_lock); 187 } 188 189 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni, 190 block_t new_blkaddr, bool fsync_done) 191 { 192 struct f2fs_nm_info *nm_i = NM_I(sbi); 193 struct nat_entry *e; 194 retry: 195 write_lock(&nm_i->nat_tree_lock); 196 e = __lookup_nat_cache(nm_i, ni->nid); 197 if (!e) { 198 e = grab_nat_entry(nm_i, ni->nid); 199 if (!e) { 200 write_unlock(&nm_i->nat_tree_lock); 201 goto retry; 202 } 203 e->ni = *ni; 204 f2fs_bug_on(ni->blk_addr == NEW_ADDR); 205 } else if (new_blkaddr == NEW_ADDR) { 206 /* 207 * when nid is reallocated, 208 * previous nat entry can be remained in nat cache. 209 * So, reinitialize it with new information. 210 */ 211 e->ni = *ni; 212 f2fs_bug_on(ni->blk_addr != NULL_ADDR); 213 } 214 215 /* sanity check */ 216 f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr); 217 f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR && 218 new_blkaddr == NULL_ADDR); 219 f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR && 220 new_blkaddr == NEW_ADDR); 221 f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR && 222 nat_get_blkaddr(e) != NULL_ADDR && 223 new_blkaddr == NEW_ADDR); 224 225 /* increament version no as node is removed */ 226 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) { 227 unsigned char version = nat_get_version(e); 228 nat_set_version(e, inc_node_version(version)); 229 } 230 231 /* change address */ 232 nat_set_blkaddr(e, new_blkaddr); 233 __set_nat_cache_dirty(nm_i, e); 234 235 /* update fsync_mark if its inode nat entry is still alive */ 236 e = __lookup_nat_cache(nm_i, ni->ino); 237 if (e) 238 e->fsync_done = fsync_done; 239 write_unlock(&nm_i->nat_tree_lock); 240 } 241 242 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink) 243 { 244 struct f2fs_nm_info *nm_i = NM_I(sbi); 245 246 if (available_free_memory(nm_i, NAT_ENTRIES)) 247 return 0; 248 249 write_lock(&nm_i->nat_tree_lock); 250 while (nr_shrink && !list_empty(&nm_i->nat_entries)) { 251 struct nat_entry *ne; 252 ne = list_first_entry(&nm_i->nat_entries, 253 struct nat_entry, list); 254 __del_from_nat_cache(nm_i, ne); 255 nr_shrink--; 256 } 257 write_unlock(&nm_i->nat_tree_lock); 258 return nr_shrink; 259 } 260 261 /* 262 * This function returns always success 263 */ 264 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni) 265 { 266 struct f2fs_nm_info *nm_i = NM_I(sbi); 267 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 268 struct f2fs_summary_block *sum = curseg->sum_blk; 269 nid_t start_nid = START_NID(nid); 270 struct f2fs_nat_block *nat_blk; 271 struct page *page = NULL; 272 struct f2fs_nat_entry ne; 273 struct nat_entry *e; 274 int i; 275 276 memset(&ne, 0, sizeof(struct f2fs_nat_entry)); 277 ni->nid = nid; 278 279 /* Check nat cache */ 280 read_lock(&nm_i->nat_tree_lock); 281 e = __lookup_nat_cache(nm_i, nid); 282 if (e) { 283 ni->ino = nat_get_ino(e); 284 ni->blk_addr = nat_get_blkaddr(e); 285 ni->version = nat_get_version(e); 286 } 287 read_unlock(&nm_i->nat_tree_lock); 288 if (e) 289 return; 290 291 /* Check current segment summary */ 292 mutex_lock(&curseg->curseg_mutex); 293 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0); 294 if (i >= 0) { 295 ne = nat_in_journal(sum, i); 296 node_info_from_raw_nat(ni, &ne); 297 } 298 mutex_unlock(&curseg->curseg_mutex); 299 if (i >= 0) 300 goto cache; 301 302 /* Fill node_info from nat page */ 303 page = get_current_nat_page(sbi, start_nid); 304 nat_blk = (struct f2fs_nat_block *)page_address(page); 305 ne = nat_blk->entries[nid - start_nid]; 306 node_info_from_raw_nat(ni, &ne); 307 f2fs_put_page(page, 1); 308 cache: 309 /* cache nat entry */ 310 cache_nat_entry(NM_I(sbi), nid, &ne); 311 } 312 313 /* 314 * The maximum depth is four. 315 * Offset[0] will have raw inode offset. 316 */ 317 static int get_node_path(struct f2fs_inode_info *fi, long block, 318 int offset[4], unsigned int noffset[4]) 319 { 320 const long direct_index = ADDRS_PER_INODE(fi); 321 const long direct_blks = ADDRS_PER_BLOCK; 322 const long dptrs_per_blk = NIDS_PER_BLOCK; 323 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK; 324 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK; 325 int n = 0; 326 int level = 0; 327 328 noffset[0] = 0; 329 330 if (block < direct_index) { 331 offset[n] = block; 332 goto got; 333 } 334 block -= direct_index; 335 if (block < direct_blks) { 336 offset[n++] = NODE_DIR1_BLOCK; 337 noffset[n] = 1; 338 offset[n] = block; 339 level = 1; 340 goto got; 341 } 342 block -= direct_blks; 343 if (block < direct_blks) { 344 offset[n++] = NODE_DIR2_BLOCK; 345 noffset[n] = 2; 346 offset[n] = block; 347 level = 1; 348 goto got; 349 } 350 block -= direct_blks; 351 if (block < indirect_blks) { 352 offset[n++] = NODE_IND1_BLOCK; 353 noffset[n] = 3; 354 offset[n++] = block / direct_blks; 355 noffset[n] = 4 + offset[n - 1]; 356 offset[n] = block % direct_blks; 357 level = 2; 358 goto got; 359 } 360 block -= indirect_blks; 361 if (block < indirect_blks) { 362 offset[n++] = NODE_IND2_BLOCK; 363 noffset[n] = 4 + dptrs_per_blk; 364 offset[n++] = block / direct_blks; 365 noffset[n] = 5 + dptrs_per_blk + offset[n - 1]; 366 offset[n] = block % direct_blks; 367 level = 2; 368 goto got; 369 } 370 block -= indirect_blks; 371 if (block < dindirect_blks) { 372 offset[n++] = NODE_DIND_BLOCK; 373 noffset[n] = 5 + (dptrs_per_blk * 2); 374 offset[n++] = block / indirect_blks; 375 noffset[n] = 6 + (dptrs_per_blk * 2) + 376 offset[n - 1] * (dptrs_per_blk + 1); 377 offset[n++] = (block / direct_blks) % dptrs_per_blk; 378 noffset[n] = 7 + (dptrs_per_blk * 2) + 379 offset[n - 2] * (dptrs_per_blk + 1) + 380 offset[n - 1]; 381 offset[n] = block % direct_blks; 382 level = 3; 383 goto got; 384 } else { 385 BUG(); 386 } 387 got: 388 return level; 389 } 390 391 /* 392 * Caller should call f2fs_put_dnode(dn). 393 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and 394 * f2fs_unlock_op() only if ro is not set RDONLY_NODE. 395 * In the case of RDONLY_NODE, we don't need to care about mutex. 396 */ 397 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode) 398 { 399 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 400 struct page *npage[4]; 401 struct page *parent; 402 int offset[4]; 403 unsigned int noffset[4]; 404 nid_t nids[4]; 405 int level, i; 406 int err = 0; 407 408 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset); 409 410 nids[0] = dn->inode->i_ino; 411 npage[0] = dn->inode_page; 412 413 if (!npage[0]) { 414 npage[0] = get_node_page(sbi, nids[0]); 415 if (IS_ERR(npage[0])) 416 return PTR_ERR(npage[0]); 417 } 418 parent = npage[0]; 419 if (level != 0) 420 nids[1] = get_nid(parent, offset[0], true); 421 dn->inode_page = npage[0]; 422 dn->inode_page_locked = true; 423 424 /* get indirect or direct nodes */ 425 for (i = 1; i <= level; i++) { 426 bool done = false; 427 428 if (!nids[i] && mode == ALLOC_NODE) { 429 /* alloc new node */ 430 if (!alloc_nid(sbi, &(nids[i]))) { 431 err = -ENOSPC; 432 goto release_pages; 433 } 434 435 dn->nid = nids[i]; 436 npage[i] = new_node_page(dn, noffset[i], NULL); 437 if (IS_ERR(npage[i])) { 438 alloc_nid_failed(sbi, nids[i]); 439 err = PTR_ERR(npage[i]); 440 goto release_pages; 441 } 442 443 set_nid(parent, offset[i - 1], nids[i], i == 1); 444 alloc_nid_done(sbi, nids[i]); 445 done = true; 446 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) { 447 npage[i] = get_node_page_ra(parent, offset[i - 1]); 448 if (IS_ERR(npage[i])) { 449 err = PTR_ERR(npage[i]); 450 goto release_pages; 451 } 452 done = true; 453 } 454 if (i == 1) { 455 dn->inode_page_locked = false; 456 unlock_page(parent); 457 } else { 458 f2fs_put_page(parent, 1); 459 } 460 461 if (!done) { 462 npage[i] = get_node_page(sbi, nids[i]); 463 if (IS_ERR(npage[i])) { 464 err = PTR_ERR(npage[i]); 465 f2fs_put_page(npage[0], 0); 466 goto release_out; 467 } 468 } 469 if (i < level) { 470 parent = npage[i]; 471 nids[i + 1] = get_nid(parent, offset[i], false); 472 } 473 } 474 dn->nid = nids[level]; 475 dn->ofs_in_node = offset[level]; 476 dn->node_page = npage[level]; 477 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node); 478 return 0; 479 480 release_pages: 481 f2fs_put_page(parent, 1); 482 if (i > 1) 483 f2fs_put_page(npage[0], 0); 484 release_out: 485 dn->inode_page = NULL; 486 dn->node_page = NULL; 487 return err; 488 } 489 490 static void truncate_node(struct dnode_of_data *dn) 491 { 492 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 493 struct node_info ni; 494 495 get_node_info(sbi, dn->nid, &ni); 496 if (dn->inode->i_blocks == 0) { 497 f2fs_bug_on(ni.blk_addr != NULL_ADDR); 498 goto invalidate; 499 } 500 f2fs_bug_on(ni.blk_addr == NULL_ADDR); 501 502 /* Deallocate node address */ 503 invalidate_blocks(sbi, ni.blk_addr); 504 dec_valid_node_count(sbi, dn->inode); 505 set_node_addr(sbi, &ni, NULL_ADDR, false); 506 507 if (dn->nid == dn->inode->i_ino) { 508 remove_orphan_inode(sbi, dn->nid); 509 dec_valid_inode_count(sbi); 510 } else { 511 sync_inode_page(dn); 512 } 513 invalidate: 514 clear_node_page_dirty(dn->node_page); 515 F2FS_SET_SB_DIRT(sbi); 516 517 f2fs_put_page(dn->node_page, 1); 518 519 invalidate_mapping_pages(NODE_MAPPING(sbi), 520 dn->node_page->index, dn->node_page->index); 521 522 dn->node_page = NULL; 523 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr); 524 } 525 526 static int truncate_dnode(struct dnode_of_data *dn) 527 { 528 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 529 struct page *page; 530 531 if (dn->nid == 0) 532 return 1; 533 534 /* get direct node */ 535 page = get_node_page(sbi, dn->nid); 536 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT) 537 return 1; 538 else if (IS_ERR(page)) 539 return PTR_ERR(page); 540 541 /* Make dnode_of_data for parameter */ 542 dn->node_page = page; 543 dn->ofs_in_node = 0; 544 truncate_data_blocks(dn); 545 truncate_node(dn); 546 return 1; 547 } 548 549 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs, 550 int ofs, int depth) 551 { 552 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 553 struct dnode_of_data rdn = *dn; 554 struct page *page; 555 struct f2fs_node *rn; 556 nid_t child_nid; 557 unsigned int child_nofs; 558 int freed = 0; 559 int i, ret; 560 561 if (dn->nid == 0) 562 return NIDS_PER_BLOCK + 1; 563 564 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr); 565 566 page = get_node_page(sbi, dn->nid); 567 if (IS_ERR(page)) { 568 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page)); 569 return PTR_ERR(page); 570 } 571 572 rn = F2FS_NODE(page); 573 if (depth < 3) { 574 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) { 575 child_nid = le32_to_cpu(rn->in.nid[i]); 576 if (child_nid == 0) 577 continue; 578 rdn.nid = child_nid; 579 ret = truncate_dnode(&rdn); 580 if (ret < 0) 581 goto out_err; 582 set_nid(page, i, 0, false); 583 } 584 } else { 585 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1; 586 for (i = ofs; i < NIDS_PER_BLOCK; i++) { 587 child_nid = le32_to_cpu(rn->in.nid[i]); 588 if (child_nid == 0) { 589 child_nofs += NIDS_PER_BLOCK + 1; 590 continue; 591 } 592 rdn.nid = child_nid; 593 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1); 594 if (ret == (NIDS_PER_BLOCK + 1)) { 595 set_nid(page, i, 0, false); 596 child_nofs += ret; 597 } else if (ret < 0 && ret != -ENOENT) { 598 goto out_err; 599 } 600 } 601 freed = child_nofs; 602 } 603 604 if (!ofs) { 605 /* remove current indirect node */ 606 dn->node_page = page; 607 truncate_node(dn); 608 freed++; 609 } else { 610 f2fs_put_page(page, 1); 611 } 612 trace_f2fs_truncate_nodes_exit(dn->inode, freed); 613 return freed; 614 615 out_err: 616 f2fs_put_page(page, 1); 617 trace_f2fs_truncate_nodes_exit(dn->inode, ret); 618 return ret; 619 } 620 621 static int truncate_partial_nodes(struct dnode_of_data *dn, 622 struct f2fs_inode *ri, int *offset, int depth) 623 { 624 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 625 struct page *pages[2]; 626 nid_t nid[3]; 627 nid_t child_nid; 628 int err = 0; 629 int i; 630 int idx = depth - 2; 631 632 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 633 if (!nid[0]) 634 return 0; 635 636 /* get indirect nodes in the path */ 637 for (i = 0; i < idx + 1; i++) { 638 /* refernece count'll be increased */ 639 pages[i] = get_node_page(sbi, nid[i]); 640 if (IS_ERR(pages[i])) { 641 err = PTR_ERR(pages[i]); 642 idx = i - 1; 643 goto fail; 644 } 645 nid[i + 1] = get_nid(pages[i], offset[i + 1], false); 646 } 647 648 /* free direct nodes linked to a partial indirect node */ 649 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) { 650 child_nid = get_nid(pages[idx], i, false); 651 if (!child_nid) 652 continue; 653 dn->nid = child_nid; 654 err = truncate_dnode(dn); 655 if (err < 0) 656 goto fail; 657 set_nid(pages[idx], i, 0, false); 658 } 659 660 if (offset[idx + 1] == 0) { 661 dn->node_page = pages[idx]; 662 dn->nid = nid[idx]; 663 truncate_node(dn); 664 } else { 665 f2fs_put_page(pages[idx], 1); 666 } 667 offset[idx]++; 668 offset[idx + 1] = 0; 669 idx--; 670 fail: 671 for (i = idx; i >= 0; i--) 672 f2fs_put_page(pages[i], 1); 673 674 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err); 675 676 return err; 677 } 678 679 /* 680 * All the block addresses of data and nodes should be nullified. 681 */ 682 int truncate_inode_blocks(struct inode *inode, pgoff_t from) 683 { 684 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 685 int err = 0, cont = 1; 686 int level, offset[4], noffset[4]; 687 unsigned int nofs = 0; 688 struct f2fs_inode *ri; 689 struct dnode_of_data dn; 690 struct page *page; 691 692 trace_f2fs_truncate_inode_blocks_enter(inode, from); 693 694 level = get_node_path(F2FS_I(inode), from, offset, noffset); 695 restart: 696 page = get_node_page(sbi, inode->i_ino); 697 if (IS_ERR(page)) { 698 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page)); 699 return PTR_ERR(page); 700 } 701 702 set_new_dnode(&dn, inode, page, NULL, 0); 703 unlock_page(page); 704 705 ri = F2FS_INODE(page); 706 switch (level) { 707 case 0: 708 case 1: 709 nofs = noffset[1]; 710 break; 711 case 2: 712 nofs = noffset[1]; 713 if (!offset[level - 1]) 714 goto skip_partial; 715 err = truncate_partial_nodes(&dn, ri, offset, level); 716 if (err < 0 && err != -ENOENT) 717 goto fail; 718 nofs += 1 + NIDS_PER_BLOCK; 719 break; 720 case 3: 721 nofs = 5 + 2 * NIDS_PER_BLOCK; 722 if (!offset[level - 1]) 723 goto skip_partial; 724 err = truncate_partial_nodes(&dn, ri, offset, level); 725 if (err < 0 && err != -ENOENT) 726 goto fail; 727 break; 728 default: 729 BUG(); 730 } 731 732 skip_partial: 733 while (cont) { 734 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 735 switch (offset[0]) { 736 case NODE_DIR1_BLOCK: 737 case NODE_DIR2_BLOCK: 738 err = truncate_dnode(&dn); 739 break; 740 741 case NODE_IND1_BLOCK: 742 case NODE_IND2_BLOCK: 743 err = truncate_nodes(&dn, nofs, offset[1], 2); 744 break; 745 746 case NODE_DIND_BLOCK: 747 err = truncate_nodes(&dn, nofs, offset[1], 3); 748 cont = 0; 749 break; 750 751 default: 752 BUG(); 753 } 754 if (err < 0 && err != -ENOENT) 755 goto fail; 756 if (offset[1] == 0 && 757 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) { 758 lock_page(page); 759 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 760 f2fs_put_page(page, 1); 761 goto restart; 762 } 763 f2fs_wait_on_page_writeback(page, NODE); 764 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0; 765 set_page_dirty(page); 766 unlock_page(page); 767 } 768 offset[1] = 0; 769 offset[0]++; 770 nofs += err; 771 } 772 fail: 773 f2fs_put_page(page, 0); 774 trace_f2fs_truncate_inode_blocks_exit(inode, err); 775 return err > 0 ? 0 : err; 776 } 777 778 int truncate_xattr_node(struct inode *inode, struct page *page) 779 { 780 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 781 nid_t nid = F2FS_I(inode)->i_xattr_nid; 782 struct dnode_of_data dn; 783 struct page *npage; 784 785 if (!nid) 786 return 0; 787 788 npage = get_node_page(sbi, nid); 789 if (IS_ERR(npage)) 790 return PTR_ERR(npage); 791 792 F2FS_I(inode)->i_xattr_nid = 0; 793 794 /* need to do checkpoint during fsync */ 795 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi)); 796 797 set_new_dnode(&dn, inode, page, npage, nid); 798 799 if (page) 800 dn.inode_page_locked = true; 801 truncate_node(&dn); 802 return 0; 803 } 804 805 /* 806 * Caller should grab and release a rwsem by calling f2fs_lock_op() and 807 * f2fs_unlock_op(). 808 */ 809 void remove_inode_page(struct inode *inode) 810 { 811 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 812 struct page *page; 813 nid_t ino = inode->i_ino; 814 struct dnode_of_data dn; 815 816 page = get_node_page(sbi, ino); 817 if (IS_ERR(page)) 818 return; 819 820 if (truncate_xattr_node(inode, page)) { 821 f2fs_put_page(page, 1); 822 return; 823 } 824 /* 0 is possible, after f2fs_new_inode() is failed */ 825 f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1); 826 set_new_dnode(&dn, inode, page, page, ino); 827 truncate_node(&dn); 828 } 829 830 struct page *new_inode_page(struct inode *inode, const struct qstr *name) 831 { 832 struct dnode_of_data dn; 833 834 /* allocate inode page for new inode */ 835 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 836 837 /* caller should f2fs_put_page(page, 1); */ 838 return new_node_page(&dn, 0, NULL); 839 } 840 841 struct page *new_node_page(struct dnode_of_data *dn, 842 unsigned int ofs, struct page *ipage) 843 { 844 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 845 struct node_info old_ni, new_ni; 846 struct page *page; 847 int err; 848 849 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))) 850 return ERR_PTR(-EPERM); 851 852 page = grab_cache_page_write_begin(NODE_MAPPING(sbi), 853 dn->nid, AOP_FLAG_NOFS); 854 if (!page) 855 return ERR_PTR(-ENOMEM); 856 857 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) { 858 err = -ENOSPC; 859 goto fail; 860 } 861 862 get_node_info(sbi, dn->nid, &old_ni); 863 864 /* Reinitialize old_ni with new node page */ 865 f2fs_bug_on(old_ni.blk_addr != NULL_ADDR); 866 new_ni = old_ni; 867 new_ni.ino = dn->inode->i_ino; 868 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 869 870 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 871 set_cold_node(dn->inode, page); 872 SetPageUptodate(page); 873 set_page_dirty(page); 874 875 if (f2fs_has_xattr_block(ofs)) 876 F2FS_I(dn->inode)->i_xattr_nid = dn->nid; 877 878 dn->node_page = page; 879 if (ipage) 880 update_inode(dn->inode, ipage); 881 else 882 sync_inode_page(dn); 883 if (ofs == 0) 884 inc_valid_inode_count(sbi); 885 886 return page; 887 888 fail: 889 clear_node_page_dirty(page); 890 f2fs_put_page(page, 1); 891 return ERR_PTR(err); 892 } 893 894 /* 895 * Caller should do after getting the following values. 896 * 0: f2fs_put_page(page, 0) 897 * LOCKED_PAGE: f2fs_put_page(page, 1) 898 * error: nothing 899 */ 900 static int read_node_page(struct page *page, int rw) 901 { 902 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb); 903 struct node_info ni; 904 905 get_node_info(sbi, page->index, &ni); 906 907 if (unlikely(ni.blk_addr == NULL_ADDR)) { 908 f2fs_put_page(page, 1); 909 return -ENOENT; 910 } 911 912 if (PageUptodate(page)) 913 return LOCKED_PAGE; 914 915 return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw); 916 } 917 918 /* 919 * Readahead a node page 920 */ 921 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 922 { 923 struct page *apage; 924 int err; 925 926 apage = find_get_page(NODE_MAPPING(sbi), nid); 927 if (apage && PageUptodate(apage)) { 928 f2fs_put_page(apage, 0); 929 return; 930 } 931 f2fs_put_page(apage, 0); 932 933 apage = grab_cache_page(NODE_MAPPING(sbi), nid); 934 if (!apage) 935 return; 936 937 err = read_node_page(apage, READA); 938 if (err == 0) 939 f2fs_put_page(apage, 0); 940 else if (err == LOCKED_PAGE) 941 f2fs_put_page(apage, 1); 942 } 943 944 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 945 { 946 struct page *page; 947 int err; 948 repeat: 949 page = grab_cache_page_write_begin(NODE_MAPPING(sbi), 950 nid, AOP_FLAG_NOFS); 951 if (!page) 952 return ERR_PTR(-ENOMEM); 953 954 err = read_node_page(page, READ_SYNC); 955 if (err < 0) 956 return ERR_PTR(err); 957 else if (err == LOCKED_PAGE) 958 goto got_it; 959 960 lock_page(page); 961 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) { 962 f2fs_put_page(page, 1); 963 return ERR_PTR(-EIO); 964 } 965 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 966 f2fs_put_page(page, 1); 967 goto repeat; 968 } 969 got_it: 970 mark_page_accessed(page); 971 return page; 972 } 973 974 /* 975 * Return a locked page for the desired node page. 976 * And, readahead MAX_RA_NODE number of node pages. 977 */ 978 struct page *get_node_page_ra(struct page *parent, int start) 979 { 980 struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb); 981 struct blk_plug plug; 982 struct page *page; 983 int err, i, end; 984 nid_t nid; 985 986 /* First, try getting the desired direct node. */ 987 nid = get_nid(parent, start, false); 988 if (!nid) 989 return ERR_PTR(-ENOENT); 990 repeat: 991 page = grab_cache_page(NODE_MAPPING(sbi), nid); 992 if (!page) 993 return ERR_PTR(-ENOMEM); 994 995 err = read_node_page(page, READ_SYNC); 996 if (err < 0) 997 return ERR_PTR(err); 998 else if (err == LOCKED_PAGE) 999 goto page_hit; 1000 1001 blk_start_plug(&plug); 1002 1003 /* Then, try readahead for siblings of the desired node */ 1004 end = start + MAX_RA_NODE; 1005 end = min(end, NIDS_PER_BLOCK); 1006 for (i = start + 1; i < end; i++) { 1007 nid = get_nid(parent, i, false); 1008 if (!nid) 1009 continue; 1010 ra_node_page(sbi, nid); 1011 } 1012 1013 blk_finish_plug(&plug); 1014 1015 lock_page(page); 1016 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1017 f2fs_put_page(page, 1); 1018 goto repeat; 1019 } 1020 page_hit: 1021 if (unlikely(!PageUptodate(page))) { 1022 f2fs_put_page(page, 1); 1023 return ERR_PTR(-EIO); 1024 } 1025 mark_page_accessed(page); 1026 return page; 1027 } 1028 1029 void sync_inode_page(struct dnode_of_data *dn) 1030 { 1031 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) { 1032 update_inode(dn->inode, dn->node_page); 1033 } else if (dn->inode_page) { 1034 if (!dn->inode_page_locked) 1035 lock_page(dn->inode_page); 1036 update_inode(dn->inode, dn->inode_page); 1037 if (!dn->inode_page_locked) 1038 unlock_page(dn->inode_page); 1039 } else { 1040 update_inode_page(dn->inode); 1041 } 1042 } 1043 1044 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino, 1045 struct writeback_control *wbc) 1046 { 1047 pgoff_t index, end; 1048 struct pagevec pvec; 1049 int step = ino ? 2 : 0; 1050 int nwritten = 0, wrote = 0; 1051 1052 pagevec_init(&pvec, 0); 1053 1054 next_step: 1055 index = 0; 1056 end = LONG_MAX; 1057 1058 while (index <= end) { 1059 int i, nr_pages; 1060 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1061 PAGECACHE_TAG_DIRTY, 1062 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); 1063 if (nr_pages == 0) 1064 break; 1065 1066 for (i = 0; i < nr_pages; i++) { 1067 struct page *page = pvec.pages[i]; 1068 1069 /* 1070 * flushing sequence with step: 1071 * 0. indirect nodes 1072 * 1. dentry dnodes 1073 * 2. file dnodes 1074 */ 1075 if (step == 0 && IS_DNODE(page)) 1076 continue; 1077 if (step == 1 && (!IS_DNODE(page) || 1078 is_cold_node(page))) 1079 continue; 1080 if (step == 2 && (!IS_DNODE(page) || 1081 !is_cold_node(page))) 1082 continue; 1083 1084 /* 1085 * If an fsync mode, 1086 * we should not skip writing node pages. 1087 */ 1088 if (ino && ino_of_node(page) == ino) 1089 lock_page(page); 1090 else if (!trylock_page(page)) 1091 continue; 1092 1093 if (unlikely(page->mapping != NODE_MAPPING(sbi))) { 1094 continue_unlock: 1095 unlock_page(page); 1096 continue; 1097 } 1098 if (ino && ino_of_node(page) != ino) 1099 goto continue_unlock; 1100 1101 if (!PageDirty(page)) { 1102 /* someone wrote it for us */ 1103 goto continue_unlock; 1104 } 1105 1106 if (!clear_page_dirty_for_io(page)) 1107 goto continue_unlock; 1108 1109 /* called by fsync() */ 1110 if (ino && IS_DNODE(page)) { 1111 int mark = !is_checkpointed_node(sbi, ino); 1112 set_fsync_mark(page, 1); 1113 if (IS_INODE(page)) 1114 set_dentry_mark(page, mark); 1115 nwritten++; 1116 } else { 1117 set_fsync_mark(page, 0); 1118 set_dentry_mark(page, 0); 1119 } 1120 NODE_MAPPING(sbi)->a_ops->writepage(page, wbc); 1121 wrote++; 1122 1123 if (--wbc->nr_to_write == 0) 1124 break; 1125 } 1126 pagevec_release(&pvec); 1127 cond_resched(); 1128 1129 if (wbc->nr_to_write == 0) { 1130 step = 2; 1131 break; 1132 } 1133 } 1134 1135 if (step < 2) { 1136 step++; 1137 goto next_step; 1138 } 1139 1140 if (wrote) 1141 f2fs_submit_merged_bio(sbi, NODE, WRITE); 1142 return nwritten; 1143 } 1144 1145 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino) 1146 { 1147 pgoff_t index = 0, end = LONG_MAX; 1148 struct pagevec pvec; 1149 int ret2 = 0, ret = 0; 1150 1151 pagevec_init(&pvec, 0); 1152 1153 while (index <= end) { 1154 int i, nr_pages; 1155 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index, 1156 PAGECACHE_TAG_WRITEBACK, 1157 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); 1158 if (nr_pages == 0) 1159 break; 1160 1161 for (i = 0; i < nr_pages; i++) { 1162 struct page *page = pvec.pages[i]; 1163 1164 /* until radix tree lookup accepts end_index */ 1165 if (unlikely(page->index > end)) 1166 continue; 1167 1168 if (ino && ino_of_node(page) == ino) { 1169 f2fs_wait_on_page_writeback(page, NODE); 1170 if (TestClearPageError(page)) 1171 ret = -EIO; 1172 } 1173 } 1174 pagevec_release(&pvec); 1175 cond_resched(); 1176 } 1177 1178 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags))) 1179 ret2 = -ENOSPC; 1180 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags))) 1181 ret2 = -EIO; 1182 if (!ret) 1183 ret = ret2; 1184 return ret; 1185 } 1186 1187 static int f2fs_write_node_page(struct page *page, 1188 struct writeback_control *wbc) 1189 { 1190 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb); 1191 nid_t nid; 1192 block_t new_addr; 1193 struct node_info ni; 1194 struct f2fs_io_info fio = { 1195 .type = NODE, 1196 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE, 1197 }; 1198 1199 if (unlikely(sbi->por_doing)) 1200 goto redirty_out; 1201 1202 f2fs_wait_on_page_writeback(page, NODE); 1203 1204 /* get old block addr of this node page */ 1205 nid = nid_of_node(page); 1206 f2fs_bug_on(page->index != nid); 1207 1208 get_node_info(sbi, nid, &ni); 1209 1210 /* This page is already truncated */ 1211 if (unlikely(ni.blk_addr == NULL_ADDR)) { 1212 dec_page_count(sbi, F2FS_DIRTY_NODES); 1213 unlock_page(page); 1214 return 0; 1215 } 1216 1217 if (wbc->for_reclaim) 1218 goto redirty_out; 1219 1220 mutex_lock(&sbi->node_write); 1221 set_page_writeback(page); 1222 write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr); 1223 set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page)); 1224 dec_page_count(sbi, F2FS_DIRTY_NODES); 1225 mutex_unlock(&sbi->node_write); 1226 unlock_page(page); 1227 return 0; 1228 1229 redirty_out: 1230 dec_page_count(sbi, F2FS_DIRTY_NODES); 1231 wbc->pages_skipped++; 1232 account_page_redirty(page); 1233 set_page_dirty(page); 1234 return AOP_WRITEPAGE_ACTIVATE; 1235 } 1236 1237 static int f2fs_write_node_pages(struct address_space *mapping, 1238 struct writeback_control *wbc) 1239 { 1240 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); 1241 long diff; 1242 1243 /* balancing f2fs's metadata in background */ 1244 f2fs_balance_fs_bg(sbi); 1245 1246 /* collect a number of dirty node pages and write together */ 1247 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE)) 1248 goto skip_write; 1249 1250 diff = nr_pages_to_write(sbi, NODE, wbc); 1251 wbc->sync_mode = WB_SYNC_NONE; 1252 sync_node_pages(sbi, 0, wbc); 1253 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); 1254 return 0; 1255 1256 skip_write: 1257 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES); 1258 return 0; 1259 } 1260 1261 static int f2fs_set_node_page_dirty(struct page *page) 1262 { 1263 struct address_space *mapping = page->mapping; 1264 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); 1265 1266 trace_f2fs_set_page_dirty(page, NODE); 1267 1268 SetPageUptodate(page); 1269 if (!PageDirty(page)) { 1270 __set_page_dirty_nobuffers(page); 1271 inc_page_count(sbi, F2FS_DIRTY_NODES); 1272 SetPagePrivate(page); 1273 return 1; 1274 } 1275 return 0; 1276 } 1277 1278 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset, 1279 unsigned int length) 1280 { 1281 struct inode *inode = page->mapping->host; 1282 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 1283 if (PageDirty(page)) 1284 dec_page_count(sbi, F2FS_DIRTY_NODES); 1285 ClearPagePrivate(page); 1286 } 1287 1288 static int f2fs_release_node_page(struct page *page, gfp_t wait) 1289 { 1290 ClearPagePrivate(page); 1291 return 1; 1292 } 1293 1294 /* 1295 * Structure of the f2fs node operations 1296 */ 1297 const struct address_space_operations f2fs_node_aops = { 1298 .writepage = f2fs_write_node_page, 1299 .writepages = f2fs_write_node_pages, 1300 .set_page_dirty = f2fs_set_node_page_dirty, 1301 .invalidatepage = f2fs_invalidate_node_page, 1302 .releasepage = f2fs_release_node_page, 1303 }; 1304 1305 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i, 1306 nid_t n) 1307 { 1308 return radix_tree_lookup(&nm_i->free_nid_root, n); 1309 } 1310 1311 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i, 1312 struct free_nid *i) 1313 { 1314 list_del(&i->list); 1315 radix_tree_delete(&nm_i->free_nid_root, i->nid); 1316 } 1317 1318 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build) 1319 { 1320 struct free_nid *i; 1321 struct nat_entry *ne; 1322 bool allocated = false; 1323 1324 if (!available_free_memory(nm_i, FREE_NIDS)) 1325 return -1; 1326 1327 /* 0 nid should not be used */ 1328 if (unlikely(nid == 0)) 1329 return 0; 1330 1331 if (build) { 1332 /* do not add allocated nids */ 1333 read_lock(&nm_i->nat_tree_lock); 1334 ne = __lookup_nat_cache(nm_i, nid); 1335 if (ne && 1336 (!ne->checkpointed || nat_get_blkaddr(ne) != NULL_ADDR)) 1337 allocated = true; 1338 read_unlock(&nm_i->nat_tree_lock); 1339 if (allocated) 1340 return 0; 1341 } 1342 1343 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS); 1344 i->nid = nid; 1345 i->state = NID_NEW; 1346 1347 spin_lock(&nm_i->free_nid_list_lock); 1348 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) { 1349 spin_unlock(&nm_i->free_nid_list_lock); 1350 kmem_cache_free(free_nid_slab, i); 1351 return 0; 1352 } 1353 list_add_tail(&i->list, &nm_i->free_nid_list); 1354 nm_i->fcnt++; 1355 spin_unlock(&nm_i->free_nid_list_lock); 1356 return 1; 1357 } 1358 1359 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid) 1360 { 1361 struct free_nid *i; 1362 bool need_free = false; 1363 1364 spin_lock(&nm_i->free_nid_list_lock); 1365 i = __lookup_free_nid_list(nm_i, nid); 1366 if (i && i->state == NID_NEW) { 1367 __del_from_free_nid_list(nm_i, i); 1368 nm_i->fcnt--; 1369 need_free = true; 1370 } 1371 spin_unlock(&nm_i->free_nid_list_lock); 1372 1373 if (need_free) 1374 kmem_cache_free(free_nid_slab, i); 1375 } 1376 1377 static void scan_nat_page(struct f2fs_nm_info *nm_i, 1378 struct page *nat_page, nid_t start_nid) 1379 { 1380 struct f2fs_nat_block *nat_blk = page_address(nat_page); 1381 block_t blk_addr; 1382 int i; 1383 1384 i = start_nid % NAT_ENTRY_PER_BLOCK; 1385 1386 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) { 1387 1388 if (unlikely(start_nid >= nm_i->max_nid)) 1389 break; 1390 1391 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr); 1392 f2fs_bug_on(blk_addr == NEW_ADDR); 1393 if (blk_addr == NULL_ADDR) { 1394 if (add_free_nid(nm_i, start_nid, true) < 0) 1395 break; 1396 } 1397 } 1398 } 1399 1400 static void build_free_nids(struct f2fs_sb_info *sbi) 1401 { 1402 struct f2fs_nm_info *nm_i = NM_I(sbi); 1403 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 1404 struct f2fs_summary_block *sum = curseg->sum_blk; 1405 int i = 0; 1406 nid_t nid = nm_i->next_scan_nid; 1407 1408 /* Enough entries */ 1409 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK) 1410 return; 1411 1412 /* readahead nat pages to be scanned */ 1413 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT); 1414 1415 while (1) { 1416 struct page *page = get_current_nat_page(sbi, nid); 1417 1418 scan_nat_page(nm_i, page, nid); 1419 f2fs_put_page(page, 1); 1420 1421 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK)); 1422 if (unlikely(nid >= nm_i->max_nid)) 1423 nid = 0; 1424 1425 if (i++ == FREE_NID_PAGES) 1426 break; 1427 } 1428 1429 /* go to the next free nat pages to find free nids abundantly */ 1430 nm_i->next_scan_nid = nid; 1431 1432 /* find free nids from current sum_pages */ 1433 mutex_lock(&curseg->curseg_mutex); 1434 for (i = 0; i < nats_in_cursum(sum); i++) { 1435 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr); 1436 nid = le32_to_cpu(nid_in_journal(sum, i)); 1437 if (addr == NULL_ADDR) 1438 add_free_nid(nm_i, nid, true); 1439 else 1440 remove_free_nid(nm_i, nid); 1441 } 1442 mutex_unlock(&curseg->curseg_mutex); 1443 } 1444 1445 /* 1446 * If this function returns success, caller can obtain a new nid 1447 * from second parameter of this function. 1448 * The returned nid could be used ino as well as nid when inode is created. 1449 */ 1450 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid) 1451 { 1452 struct f2fs_nm_info *nm_i = NM_I(sbi); 1453 struct free_nid *i = NULL; 1454 retry: 1455 if (unlikely(sbi->total_valid_node_count + 1 >= nm_i->max_nid)) 1456 return false; 1457 1458 spin_lock(&nm_i->free_nid_list_lock); 1459 1460 /* We should not use stale free nids created by build_free_nids */ 1461 if (nm_i->fcnt && !on_build_free_nids(nm_i)) { 1462 f2fs_bug_on(list_empty(&nm_i->free_nid_list)); 1463 list_for_each_entry(i, &nm_i->free_nid_list, list) 1464 if (i->state == NID_NEW) 1465 break; 1466 1467 f2fs_bug_on(i->state != NID_NEW); 1468 *nid = i->nid; 1469 i->state = NID_ALLOC; 1470 nm_i->fcnt--; 1471 spin_unlock(&nm_i->free_nid_list_lock); 1472 return true; 1473 } 1474 spin_unlock(&nm_i->free_nid_list_lock); 1475 1476 /* Let's scan nat pages and its caches to get free nids */ 1477 mutex_lock(&nm_i->build_lock); 1478 build_free_nids(sbi); 1479 mutex_unlock(&nm_i->build_lock); 1480 goto retry; 1481 } 1482 1483 /* 1484 * alloc_nid() should be called prior to this function. 1485 */ 1486 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 1487 { 1488 struct f2fs_nm_info *nm_i = NM_I(sbi); 1489 struct free_nid *i; 1490 1491 spin_lock(&nm_i->free_nid_list_lock); 1492 i = __lookup_free_nid_list(nm_i, nid); 1493 f2fs_bug_on(!i || i->state != NID_ALLOC); 1494 __del_from_free_nid_list(nm_i, i); 1495 spin_unlock(&nm_i->free_nid_list_lock); 1496 1497 kmem_cache_free(free_nid_slab, i); 1498 } 1499 1500 /* 1501 * alloc_nid() should be called prior to this function. 1502 */ 1503 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 1504 { 1505 struct f2fs_nm_info *nm_i = NM_I(sbi); 1506 struct free_nid *i; 1507 bool need_free = false; 1508 1509 if (!nid) 1510 return; 1511 1512 spin_lock(&nm_i->free_nid_list_lock); 1513 i = __lookup_free_nid_list(nm_i, nid); 1514 f2fs_bug_on(!i || i->state != NID_ALLOC); 1515 if (!available_free_memory(nm_i, FREE_NIDS)) { 1516 __del_from_free_nid_list(nm_i, i); 1517 need_free = true; 1518 } else { 1519 i->state = NID_NEW; 1520 nm_i->fcnt++; 1521 } 1522 spin_unlock(&nm_i->free_nid_list_lock); 1523 1524 if (need_free) 1525 kmem_cache_free(free_nid_slab, i); 1526 } 1527 1528 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page, 1529 struct f2fs_summary *sum, struct node_info *ni, 1530 block_t new_blkaddr) 1531 { 1532 rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr); 1533 set_node_addr(sbi, ni, new_blkaddr, false); 1534 clear_node_page_dirty(page); 1535 } 1536 1537 void recover_inline_xattr(struct inode *inode, struct page *page) 1538 { 1539 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 1540 void *src_addr, *dst_addr; 1541 size_t inline_size; 1542 struct page *ipage; 1543 struct f2fs_inode *ri; 1544 1545 if (!f2fs_has_inline_xattr(inode)) 1546 return; 1547 1548 if (!IS_INODE(page)) 1549 return; 1550 1551 ri = F2FS_INODE(page); 1552 if (!(ri->i_inline & F2FS_INLINE_XATTR)) 1553 return; 1554 1555 ipage = get_node_page(sbi, inode->i_ino); 1556 f2fs_bug_on(IS_ERR(ipage)); 1557 1558 dst_addr = inline_xattr_addr(ipage); 1559 src_addr = inline_xattr_addr(page); 1560 inline_size = inline_xattr_size(inode); 1561 1562 memcpy(dst_addr, src_addr, inline_size); 1563 1564 update_inode(inode, ipage); 1565 f2fs_put_page(ipage, 1); 1566 } 1567 1568 bool recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr) 1569 { 1570 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 1571 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid; 1572 nid_t new_xnid = nid_of_node(page); 1573 struct node_info ni; 1574 1575 recover_inline_xattr(inode, page); 1576 1577 if (!f2fs_has_xattr_block(ofs_of_node(page))) 1578 return false; 1579 1580 /* 1: invalidate the previous xattr nid */ 1581 if (!prev_xnid) 1582 goto recover_xnid; 1583 1584 /* Deallocate node address */ 1585 get_node_info(sbi, prev_xnid, &ni); 1586 f2fs_bug_on(ni.blk_addr == NULL_ADDR); 1587 invalidate_blocks(sbi, ni.blk_addr); 1588 dec_valid_node_count(sbi, inode); 1589 set_node_addr(sbi, &ni, NULL_ADDR, false); 1590 1591 recover_xnid: 1592 /* 2: allocate new xattr nid */ 1593 if (unlikely(!inc_valid_node_count(sbi, inode))) 1594 f2fs_bug_on(1); 1595 1596 remove_free_nid(NM_I(sbi), new_xnid); 1597 get_node_info(sbi, new_xnid, &ni); 1598 ni.ino = inode->i_ino; 1599 set_node_addr(sbi, &ni, NEW_ADDR, false); 1600 F2FS_I(inode)->i_xattr_nid = new_xnid; 1601 1602 /* 3: update xattr blkaddr */ 1603 refresh_sit_entry(sbi, NEW_ADDR, blkaddr); 1604 set_node_addr(sbi, &ni, blkaddr, false); 1605 1606 update_inode_page(inode); 1607 return true; 1608 } 1609 1610 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 1611 { 1612 struct f2fs_inode *src, *dst; 1613 nid_t ino = ino_of_node(page); 1614 struct node_info old_ni, new_ni; 1615 struct page *ipage; 1616 1617 ipage = grab_cache_page(NODE_MAPPING(sbi), ino); 1618 if (!ipage) 1619 return -ENOMEM; 1620 1621 /* Should not use this inode from free nid list */ 1622 remove_free_nid(NM_I(sbi), ino); 1623 1624 get_node_info(sbi, ino, &old_ni); 1625 SetPageUptodate(ipage); 1626 fill_node_footer(ipage, ino, ino, 0, true); 1627 1628 src = F2FS_INODE(page); 1629 dst = F2FS_INODE(ipage); 1630 1631 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src); 1632 dst->i_size = 0; 1633 dst->i_blocks = cpu_to_le64(1); 1634 dst->i_links = cpu_to_le32(1); 1635 dst->i_xattr_nid = 0; 1636 1637 new_ni = old_ni; 1638 new_ni.ino = ino; 1639 1640 if (unlikely(!inc_valid_node_count(sbi, NULL))) 1641 WARN_ON(1); 1642 set_node_addr(sbi, &new_ni, NEW_ADDR, false); 1643 inc_valid_inode_count(sbi); 1644 f2fs_put_page(ipage, 1); 1645 return 0; 1646 } 1647 1648 /* 1649 * ra_sum_pages() merge contiguous pages into one bio and submit. 1650 * these pre-readed pages are linked in pages list. 1651 */ 1652 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages, 1653 int start, int nrpages) 1654 { 1655 struct page *page; 1656 int page_idx = start; 1657 struct f2fs_io_info fio = { 1658 .type = META, 1659 .rw = READ_SYNC | REQ_META | REQ_PRIO 1660 }; 1661 1662 for (; page_idx < start + nrpages; page_idx++) { 1663 /* alloc temporal page for read node summary info*/ 1664 page = alloc_page(GFP_F2FS_ZERO); 1665 if (!page) 1666 break; 1667 1668 lock_page(page); 1669 page->index = page_idx; 1670 list_add_tail(&page->lru, pages); 1671 } 1672 1673 list_for_each_entry(page, pages, lru) 1674 f2fs_submit_page_mbio(sbi, page, page->index, &fio); 1675 1676 f2fs_submit_merged_bio(sbi, META, READ); 1677 1678 return page_idx - start; 1679 } 1680 1681 int restore_node_summary(struct f2fs_sb_info *sbi, 1682 unsigned int segno, struct f2fs_summary_block *sum) 1683 { 1684 struct f2fs_node *rn; 1685 struct f2fs_summary *sum_entry; 1686 struct page *page, *tmp; 1687 block_t addr; 1688 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi)); 1689 int i, last_offset, nrpages, err = 0; 1690 LIST_HEAD(page_list); 1691 1692 /* scan the node segment */ 1693 last_offset = sbi->blocks_per_seg; 1694 addr = START_BLOCK(sbi, segno); 1695 sum_entry = &sum->entries[0]; 1696 1697 for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) { 1698 nrpages = min(last_offset - i, bio_blocks); 1699 1700 /* read ahead node pages */ 1701 nrpages = ra_sum_pages(sbi, &page_list, addr, nrpages); 1702 if (!nrpages) 1703 return -ENOMEM; 1704 1705 list_for_each_entry_safe(page, tmp, &page_list, lru) { 1706 if (err) 1707 goto skip; 1708 1709 lock_page(page); 1710 if (unlikely(!PageUptodate(page))) { 1711 err = -EIO; 1712 } else { 1713 rn = F2FS_NODE(page); 1714 sum_entry->nid = rn->footer.nid; 1715 sum_entry->version = 0; 1716 sum_entry->ofs_in_node = 0; 1717 sum_entry++; 1718 } 1719 unlock_page(page); 1720 skip: 1721 list_del(&page->lru); 1722 __free_pages(page, 0); 1723 } 1724 } 1725 return err; 1726 } 1727 1728 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi) 1729 { 1730 struct f2fs_nm_info *nm_i = NM_I(sbi); 1731 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 1732 struct f2fs_summary_block *sum = curseg->sum_blk; 1733 int i; 1734 1735 mutex_lock(&curseg->curseg_mutex); 1736 1737 if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) { 1738 mutex_unlock(&curseg->curseg_mutex); 1739 return false; 1740 } 1741 1742 for (i = 0; i < nats_in_cursum(sum); i++) { 1743 struct nat_entry *ne; 1744 struct f2fs_nat_entry raw_ne; 1745 nid_t nid = le32_to_cpu(nid_in_journal(sum, i)); 1746 1747 raw_ne = nat_in_journal(sum, i); 1748 retry: 1749 write_lock(&nm_i->nat_tree_lock); 1750 ne = __lookup_nat_cache(nm_i, nid); 1751 if (ne) { 1752 __set_nat_cache_dirty(nm_i, ne); 1753 write_unlock(&nm_i->nat_tree_lock); 1754 continue; 1755 } 1756 ne = grab_nat_entry(nm_i, nid); 1757 if (!ne) { 1758 write_unlock(&nm_i->nat_tree_lock); 1759 goto retry; 1760 } 1761 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr)); 1762 nat_set_ino(ne, le32_to_cpu(raw_ne.ino)); 1763 nat_set_version(ne, raw_ne.version); 1764 __set_nat_cache_dirty(nm_i, ne); 1765 write_unlock(&nm_i->nat_tree_lock); 1766 } 1767 update_nats_in_cursum(sum, -i); 1768 mutex_unlock(&curseg->curseg_mutex); 1769 return true; 1770 } 1771 1772 /* 1773 * This function is called during the checkpointing process. 1774 */ 1775 void flush_nat_entries(struct f2fs_sb_info *sbi) 1776 { 1777 struct f2fs_nm_info *nm_i = NM_I(sbi); 1778 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 1779 struct f2fs_summary_block *sum = curseg->sum_blk; 1780 struct nat_entry *ne, *cur; 1781 struct page *page = NULL; 1782 struct f2fs_nat_block *nat_blk = NULL; 1783 nid_t start_nid = 0, end_nid = 0; 1784 bool flushed; 1785 1786 flushed = flush_nats_in_journal(sbi); 1787 1788 if (!flushed) 1789 mutex_lock(&curseg->curseg_mutex); 1790 1791 /* 1) flush dirty nat caches */ 1792 list_for_each_entry_safe(ne, cur, &nm_i->dirty_nat_entries, list) { 1793 nid_t nid; 1794 struct f2fs_nat_entry raw_ne; 1795 int offset = -1; 1796 block_t new_blkaddr; 1797 1798 if (nat_get_blkaddr(ne) == NEW_ADDR) 1799 continue; 1800 1801 nid = nat_get_nid(ne); 1802 1803 if (flushed) 1804 goto to_nat_page; 1805 1806 /* if there is room for nat enries in curseg->sumpage */ 1807 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1); 1808 if (offset >= 0) { 1809 raw_ne = nat_in_journal(sum, offset); 1810 goto flush_now; 1811 } 1812 to_nat_page: 1813 if (!page || (start_nid > nid || nid > end_nid)) { 1814 if (page) { 1815 f2fs_put_page(page, 1); 1816 page = NULL; 1817 } 1818 start_nid = START_NID(nid); 1819 end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1; 1820 1821 /* 1822 * get nat block with dirty flag, increased reference 1823 * count, mapped and lock 1824 */ 1825 page = get_next_nat_page(sbi, start_nid); 1826 nat_blk = page_address(page); 1827 } 1828 1829 f2fs_bug_on(!nat_blk); 1830 raw_ne = nat_blk->entries[nid - start_nid]; 1831 flush_now: 1832 new_blkaddr = nat_get_blkaddr(ne); 1833 1834 raw_ne.ino = cpu_to_le32(nat_get_ino(ne)); 1835 raw_ne.block_addr = cpu_to_le32(new_blkaddr); 1836 raw_ne.version = nat_get_version(ne); 1837 1838 if (offset < 0) { 1839 nat_blk->entries[nid - start_nid] = raw_ne; 1840 } else { 1841 nat_in_journal(sum, offset) = raw_ne; 1842 nid_in_journal(sum, offset) = cpu_to_le32(nid); 1843 } 1844 1845 if (nat_get_blkaddr(ne) == NULL_ADDR && 1846 add_free_nid(NM_I(sbi), nid, false) <= 0) { 1847 write_lock(&nm_i->nat_tree_lock); 1848 __del_from_nat_cache(nm_i, ne); 1849 write_unlock(&nm_i->nat_tree_lock); 1850 } else { 1851 write_lock(&nm_i->nat_tree_lock); 1852 __clear_nat_cache_dirty(nm_i, ne); 1853 write_unlock(&nm_i->nat_tree_lock); 1854 } 1855 } 1856 if (!flushed) 1857 mutex_unlock(&curseg->curseg_mutex); 1858 f2fs_put_page(page, 1); 1859 } 1860 1861 static int init_node_manager(struct f2fs_sb_info *sbi) 1862 { 1863 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 1864 struct f2fs_nm_info *nm_i = NM_I(sbi); 1865 unsigned char *version_bitmap; 1866 unsigned int nat_segs, nat_blocks; 1867 1868 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 1869 1870 /* segment_count_nat includes pair segment so divide to 2. */ 1871 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 1872 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 1873 1874 /* not used nids: 0, node, meta, (and root counted as valid node) */ 1875 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks - 3; 1876 nm_i->fcnt = 0; 1877 nm_i->nat_cnt = 0; 1878 nm_i->ram_thresh = DEF_RAM_THRESHOLD; 1879 1880 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC); 1881 INIT_LIST_HEAD(&nm_i->free_nid_list); 1882 INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC); 1883 INIT_LIST_HEAD(&nm_i->nat_entries); 1884 INIT_LIST_HEAD(&nm_i->dirty_nat_entries); 1885 1886 mutex_init(&nm_i->build_lock); 1887 spin_lock_init(&nm_i->free_nid_list_lock); 1888 rwlock_init(&nm_i->nat_tree_lock); 1889 1890 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 1891 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 1892 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 1893 if (!version_bitmap) 1894 return -EFAULT; 1895 1896 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size, 1897 GFP_KERNEL); 1898 if (!nm_i->nat_bitmap) 1899 return -ENOMEM; 1900 return 0; 1901 } 1902 1903 int build_node_manager(struct f2fs_sb_info *sbi) 1904 { 1905 int err; 1906 1907 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL); 1908 if (!sbi->nm_info) 1909 return -ENOMEM; 1910 1911 err = init_node_manager(sbi); 1912 if (err) 1913 return err; 1914 1915 build_free_nids(sbi); 1916 return 0; 1917 } 1918 1919 void destroy_node_manager(struct f2fs_sb_info *sbi) 1920 { 1921 struct f2fs_nm_info *nm_i = NM_I(sbi); 1922 struct free_nid *i, *next_i; 1923 struct nat_entry *natvec[NATVEC_SIZE]; 1924 nid_t nid = 0; 1925 unsigned int found; 1926 1927 if (!nm_i) 1928 return; 1929 1930 /* destroy free nid list */ 1931 spin_lock(&nm_i->free_nid_list_lock); 1932 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 1933 f2fs_bug_on(i->state == NID_ALLOC); 1934 __del_from_free_nid_list(nm_i, i); 1935 nm_i->fcnt--; 1936 spin_unlock(&nm_i->free_nid_list_lock); 1937 kmem_cache_free(free_nid_slab, i); 1938 spin_lock(&nm_i->free_nid_list_lock); 1939 } 1940 f2fs_bug_on(nm_i->fcnt); 1941 spin_unlock(&nm_i->free_nid_list_lock); 1942 1943 /* destroy nat cache */ 1944 write_lock(&nm_i->nat_tree_lock); 1945 while ((found = __gang_lookup_nat_cache(nm_i, 1946 nid, NATVEC_SIZE, natvec))) { 1947 unsigned idx; 1948 nid = nat_get_nid(natvec[found - 1]) + 1; 1949 for (idx = 0; idx < found; idx++) 1950 __del_from_nat_cache(nm_i, natvec[idx]); 1951 } 1952 f2fs_bug_on(nm_i->nat_cnt); 1953 write_unlock(&nm_i->nat_tree_lock); 1954 1955 kfree(nm_i->nat_bitmap); 1956 sbi->nm_info = NULL; 1957 kfree(nm_i); 1958 } 1959 1960 int __init create_node_manager_caches(void) 1961 { 1962 nat_entry_slab = f2fs_kmem_cache_create("nat_entry", 1963 sizeof(struct nat_entry)); 1964 if (!nat_entry_slab) 1965 return -ENOMEM; 1966 1967 free_nid_slab = f2fs_kmem_cache_create("free_nid", 1968 sizeof(struct free_nid)); 1969 if (!free_nid_slab) { 1970 kmem_cache_destroy(nat_entry_slab); 1971 return -ENOMEM; 1972 } 1973 return 0; 1974 } 1975 1976 void destroy_node_manager_caches(void) 1977 { 1978 kmem_cache_destroy(free_nid_slab); 1979 kmem_cache_destroy(nat_entry_slab); 1980 } 1981