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