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 23 static struct kmem_cache *nat_entry_slab; 24 static struct kmem_cache *free_nid_slab; 25 26 static void clear_node_page_dirty(struct page *page) 27 { 28 struct address_space *mapping = page->mapping; 29 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); 30 unsigned int long flags; 31 32 if (PageDirty(page)) { 33 spin_lock_irqsave(&mapping->tree_lock, flags); 34 radix_tree_tag_clear(&mapping->page_tree, 35 page_index(page), 36 PAGECACHE_TAG_DIRTY); 37 spin_unlock_irqrestore(&mapping->tree_lock, flags); 38 39 clear_page_dirty_for_io(page); 40 dec_page_count(sbi, F2FS_DIRTY_NODES); 41 } 42 ClearPageUptodate(page); 43 } 44 45 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 46 { 47 pgoff_t index = current_nat_addr(sbi, nid); 48 return get_meta_page(sbi, index); 49 } 50 51 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid) 52 { 53 struct page *src_page; 54 struct page *dst_page; 55 pgoff_t src_off; 56 pgoff_t dst_off; 57 void *src_addr; 58 void *dst_addr; 59 struct f2fs_nm_info *nm_i = NM_I(sbi); 60 61 src_off = current_nat_addr(sbi, nid); 62 dst_off = next_nat_addr(sbi, src_off); 63 64 /* get current nat block page with lock */ 65 src_page = get_meta_page(sbi, src_off); 66 67 /* Dirty src_page means that it is already the new target NAT page. */ 68 if (PageDirty(src_page)) 69 return src_page; 70 71 dst_page = grab_meta_page(sbi, dst_off); 72 73 src_addr = page_address(src_page); 74 dst_addr = page_address(dst_page); 75 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); 76 set_page_dirty(dst_page); 77 f2fs_put_page(src_page, 1); 78 79 set_to_next_nat(nm_i, nid); 80 81 return dst_page; 82 } 83 84 /* 85 * Readahead NAT pages 86 */ 87 static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid) 88 { 89 struct address_space *mapping = sbi->meta_inode->i_mapping; 90 struct f2fs_nm_info *nm_i = NM_I(sbi); 91 struct page *page; 92 pgoff_t index; 93 int i; 94 95 for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) { 96 if (nid >= nm_i->max_nid) 97 nid = 0; 98 index = current_nat_addr(sbi, nid); 99 100 page = grab_cache_page(mapping, index); 101 if (!page) 102 continue; 103 if (f2fs_readpage(sbi, page, index, READ)) { 104 f2fs_put_page(page, 1); 105 continue; 106 } 107 f2fs_put_page(page, 0); 108 } 109 } 110 111 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n) 112 { 113 return radix_tree_lookup(&nm_i->nat_root, n); 114 } 115 116 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i, 117 nid_t start, unsigned int nr, struct nat_entry **ep) 118 { 119 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr); 120 } 121 122 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e) 123 { 124 list_del(&e->list); 125 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e)); 126 nm_i->nat_cnt--; 127 kmem_cache_free(nat_entry_slab, e); 128 } 129 130 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid) 131 { 132 struct f2fs_nm_info *nm_i = NM_I(sbi); 133 struct nat_entry *e; 134 int is_cp = 1; 135 136 read_lock(&nm_i->nat_tree_lock); 137 e = __lookup_nat_cache(nm_i, nid); 138 if (e && !e->checkpointed) 139 is_cp = 0; 140 read_unlock(&nm_i->nat_tree_lock); 141 return is_cp; 142 } 143 144 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid) 145 { 146 struct nat_entry *new; 147 148 new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC); 149 if (!new) 150 return NULL; 151 if (radix_tree_insert(&nm_i->nat_root, nid, new)) { 152 kmem_cache_free(nat_entry_slab, new); 153 return NULL; 154 } 155 memset(new, 0, sizeof(struct nat_entry)); 156 nat_set_nid(new, nid); 157 list_add_tail(&new->list, &nm_i->nat_entries); 158 nm_i->nat_cnt++; 159 return new; 160 } 161 162 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid, 163 struct f2fs_nat_entry *ne) 164 { 165 struct nat_entry *e; 166 retry: 167 write_lock(&nm_i->nat_tree_lock); 168 e = __lookup_nat_cache(nm_i, nid); 169 if (!e) { 170 e = grab_nat_entry(nm_i, nid); 171 if (!e) { 172 write_unlock(&nm_i->nat_tree_lock); 173 goto retry; 174 } 175 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr)); 176 nat_set_ino(e, le32_to_cpu(ne->ino)); 177 nat_set_version(e, ne->version); 178 e->checkpointed = true; 179 } 180 write_unlock(&nm_i->nat_tree_lock); 181 } 182 183 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni, 184 block_t new_blkaddr) 185 { 186 struct f2fs_nm_info *nm_i = NM_I(sbi); 187 struct nat_entry *e; 188 retry: 189 write_lock(&nm_i->nat_tree_lock); 190 e = __lookup_nat_cache(nm_i, ni->nid); 191 if (!e) { 192 e = grab_nat_entry(nm_i, ni->nid); 193 if (!e) { 194 write_unlock(&nm_i->nat_tree_lock); 195 goto retry; 196 } 197 e->ni = *ni; 198 e->checkpointed = true; 199 BUG_ON(ni->blk_addr == NEW_ADDR); 200 } else if (new_blkaddr == NEW_ADDR) { 201 /* 202 * when nid is reallocated, 203 * previous nat entry can be remained in nat cache. 204 * So, reinitialize it with new information. 205 */ 206 e->ni = *ni; 207 BUG_ON(ni->blk_addr != NULL_ADDR); 208 } 209 210 if (new_blkaddr == NEW_ADDR) 211 e->checkpointed = false; 212 213 /* sanity check */ 214 BUG_ON(nat_get_blkaddr(e) != ni->blk_addr); 215 BUG_ON(nat_get_blkaddr(e) == NULL_ADDR && 216 new_blkaddr == NULL_ADDR); 217 BUG_ON(nat_get_blkaddr(e) == NEW_ADDR && 218 new_blkaddr == NEW_ADDR); 219 BUG_ON(nat_get_blkaddr(e) != NEW_ADDR && 220 nat_get_blkaddr(e) != NULL_ADDR && 221 new_blkaddr == NEW_ADDR); 222 223 /* increament version no as node is removed */ 224 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) { 225 unsigned char version = nat_get_version(e); 226 nat_set_version(e, inc_node_version(version)); 227 } 228 229 /* change address */ 230 nat_set_blkaddr(e, new_blkaddr); 231 __set_nat_cache_dirty(nm_i, e); 232 write_unlock(&nm_i->nat_tree_lock); 233 } 234 235 static int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink) 236 { 237 struct f2fs_nm_info *nm_i = NM_I(sbi); 238 239 if (nm_i->nat_cnt < 2 * NM_WOUT_THRESHOLD) 240 return 0; 241 242 write_lock(&nm_i->nat_tree_lock); 243 while (nr_shrink && !list_empty(&nm_i->nat_entries)) { 244 struct nat_entry *ne; 245 ne = list_first_entry(&nm_i->nat_entries, 246 struct nat_entry, list); 247 __del_from_nat_cache(nm_i, ne); 248 nr_shrink--; 249 } 250 write_unlock(&nm_i->nat_tree_lock); 251 return nr_shrink; 252 } 253 254 /* 255 * This function returns always success 256 */ 257 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni) 258 { 259 struct f2fs_nm_info *nm_i = NM_I(sbi); 260 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 261 struct f2fs_summary_block *sum = curseg->sum_blk; 262 nid_t start_nid = START_NID(nid); 263 struct f2fs_nat_block *nat_blk; 264 struct page *page = NULL; 265 struct f2fs_nat_entry ne; 266 struct nat_entry *e; 267 int i; 268 269 memset(&ne, 0, sizeof(struct f2fs_nat_entry)); 270 ni->nid = nid; 271 272 /* Check nat cache */ 273 read_lock(&nm_i->nat_tree_lock); 274 e = __lookup_nat_cache(nm_i, nid); 275 if (e) { 276 ni->ino = nat_get_ino(e); 277 ni->blk_addr = nat_get_blkaddr(e); 278 ni->version = nat_get_version(e); 279 } 280 read_unlock(&nm_i->nat_tree_lock); 281 if (e) 282 return; 283 284 /* Check current segment summary */ 285 mutex_lock(&curseg->curseg_mutex); 286 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0); 287 if (i >= 0) { 288 ne = nat_in_journal(sum, i); 289 node_info_from_raw_nat(ni, &ne); 290 } 291 mutex_unlock(&curseg->curseg_mutex); 292 if (i >= 0) 293 goto cache; 294 295 /* Fill node_info from nat page */ 296 page = get_current_nat_page(sbi, start_nid); 297 nat_blk = (struct f2fs_nat_block *)page_address(page); 298 ne = nat_blk->entries[nid - start_nid]; 299 node_info_from_raw_nat(ni, &ne); 300 f2fs_put_page(page, 1); 301 cache: 302 /* cache nat entry */ 303 cache_nat_entry(NM_I(sbi), nid, &ne); 304 } 305 306 /* 307 * The maximum depth is four. 308 * Offset[0] will have raw inode offset. 309 */ 310 static int get_node_path(long block, int offset[4], unsigned int noffset[4]) 311 { 312 const long direct_index = ADDRS_PER_INODE; 313 const long direct_blks = ADDRS_PER_BLOCK; 314 const long dptrs_per_blk = NIDS_PER_BLOCK; 315 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK; 316 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK; 317 int n = 0; 318 int level = 0; 319 320 noffset[0] = 0; 321 322 if (block < direct_index) { 323 offset[n++] = block; 324 level = 0; 325 goto got; 326 } 327 block -= direct_index; 328 if (block < direct_blks) { 329 offset[n++] = NODE_DIR1_BLOCK; 330 noffset[n] = 1; 331 offset[n++] = block; 332 level = 1; 333 goto got; 334 } 335 block -= direct_blks; 336 if (block < direct_blks) { 337 offset[n++] = NODE_DIR2_BLOCK; 338 noffset[n] = 2; 339 offset[n++] = block; 340 level = 1; 341 goto got; 342 } 343 block -= direct_blks; 344 if (block < indirect_blks) { 345 offset[n++] = NODE_IND1_BLOCK; 346 noffset[n] = 3; 347 offset[n++] = block / direct_blks; 348 noffset[n] = 4 + offset[n - 1]; 349 offset[n++] = block % direct_blks; 350 level = 2; 351 goto got; 352 } 353 block -= indirect_blks; 354 if (block < indirect_blks) { 355 offset[n++] = NODE_IND2_BLOCK; 356 noffset[n] = 4 + dptrs_per_blk; 357 offset[n++] = block / direct_blks; 358 noffset[n] = 5 + dptrs_per_blk + offset[n - 1]; 359 offset[n++] = block % direct_blks; 360 level = 2; 361 goto got; 362 } 363 block -= indirect_blks; 364 if (block < dindirect_blks) { 365 offset[n++] = NODE_DIND_BLOCK; 366 noffset[n] = 5 + (dptrs_per_blk * 2); 367 offset[n++] = block / indirect_blks; 368 noffset[n] = 6 + (dptrs_per_blk * 2) + 369 offset[n - 1] * (dptrs_per_blk + 1); 370 offset[n++] = (block / direct_blks) % dptrs_per_blk; 371 noffset[n] = 7 + (dptrs_per_blk * 2) + 372 offset[n - 2] * (dptrs_per_blk + 1) + 373 offset[n - 1]; 374 offset[n++] = block % direct_blks; 375 level = 3; 376 goto got; 377 } else { 378 BUG(); 379 } 380 got: 381 return level; 382 } 383 384 /* 385 * Caller should call f2fs_put_dnode(dn). 386 */ 387 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode) 388 { 389 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 390 struct page *npage[4]; 391 struct page *parent; 392 int offset[4]; 393 unsigned int noffset[4]; 394 nid_t nids[4]; 395 int level, i; 396 int err = 0; 397 398 level = get_node_path(index, offset, noffset); 399 400 nids[0] = dn->inode->i_ino; 401 npage[0] = get_node_page(sbi, nids[0]); 402 if (IS_ERR(npage[0])) 403 return PTR_ERR(npage[0]); 404 405 parent = npage[0]; 406 if (level != 0) 407 nids[1] = get_nid(parent, offset[0], true); 408 dn->inode_page = npage[0]; 409 dn->inode_page_locked = true; 410 411 /* get indirect or direct nodes */ 412 for (i = 1; i <= level; i++) { 413 bool done = false; 414 415 if (!nids[i] && mode == ALLOC_NODE) { 416 mutex_lock_op(sbi, NODE_NEW); 417 418 /* alloc new node */ 419 if (!alloc_nid(sbi, &(nids[i]))) { 420 mutex_unlock_op(sbi, NODE_NEW); 421 err = -ENOSPC; 422 goto release_pages; 423 } 424 425 dn->nid = nids[i]; 426 npage[i] = new_node_page(dn, noffset[i]); 427 if (IS_ERR(npage[i])) { 428 alloc_nid_failed(sbi, nids[i]); 429 mutex_unlock_op(sbi, NODE_NEW); 430 err = PTR_ERR(npage[i]); 431 goto release_pages; 432 } 433 434 set_nid(parent, offset[i - 1], nids[i], i == 1); 435 alloc_nid_done(sbi, nids[i]); 436 mutex_unlock_op(sbi, NODE_NEW); 437 done = true; 438 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) { 439 npage[i] = get_node_page_ra(parent, offset[i - 1]); 440 if (IS_ERR(npage[i])) { 441 err = PTR_ERR(npage[i]); 442 goto release_pages; 443 } 444 done = true; 445 } 446 if (i == 1) { 447 dn->inode_page_locked = false; 448 unlock_page(parent); 449 } else { 450 f2fs_put_page(parent, 1); 451 } 452 453 if (!done) { 454 npage[i] = get_node_page(sbi, nids[i]); 455 if (IS_ERR(npage[i])) { 456 err = PTR_ERR(npage[i]); 457 f2fs_put_page(npage[0], 0); 458 goto release_out; 459 } 460 } 461 if (i < level) { 462 parent = npage[i]; 463 nids[i + 1] = get_nid(parent, offset[i], false); 464 } 465 } 466 dn->nid = nids[level]; 467 dn->ofs_in_node = offset[level]; 468 dn->node_page = npage[level]; 469 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node); 470 return 0; 471 472 release_pages: 473 f2fs_put_page(parent, 1); 474 if (i > 1) 475 f2fs_put_page(npage[0], 0); 476 release_out: 477 dn->inode_page = NULL; 478 dn->node_page = NULL; 479 return err; 480 } 481 482 static void truncate_node(struct dnode_of_data *dn) 483 { 484 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 485 struct node_info ni; 486 487 get_node_info(sbi, dn->nid, &ni); 488 if (dn->inode->i_blocks == 0) { 489 BUG_ON(ni.blk_addr != NULL_ADDR); 490 goto invalidate; 491 } 492 BUG_ON(ni.blk_addr == NULL_ADDR); 493 494 /* Deallocate node address */ 495 invalidate_blocks(sbi, ni.blk_addr); 496 dec_valid_node_count(sbi, dn->inode, 1); 497 set_node_addr(sbi, &ni, NULL_ADDR); 498 499 if (dn->nid == dn->inode->i_ino) { 500 remove_orphan_inode(sbi, dn->nid); 501 dec_valid_inode_count(sbi); 502 } else { 503 sync_inode_page(dn); 504 } 505 invalidate: 506 clear_node_page_dirty(dn->node_page); 507 F2FS_SET_SB_DIRT(sbi); 508 509 f2fs_put_page(dn->node_page, 1); 510 dn->node_page = NULL; 511 } 512 513 static int truncate_dnode(struct dnode_of_data *dn) 514 { 515 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 516 struct page *page; 517 518 if (dn->nid == 0) 519 return 1; 520 521 /* get direct node */ 522 page = get_node_page(sbi, dn->nid); 523 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT) 524 return 1; 525 else if (IS_ERR(page)) 526 return PTR_ERR(page); 527 528 /* Make dnode_of_data for parameter */ 529 dn->node_page = page; 530 dn->ofs_in_node = 0; 531 truncate_data_blocks(dn); 532 truncate_node(dn); 533 return 1; 534 } 535 536 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs, 537 int ofs, int depth) 538 { 539 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 540 struct dnode_of_data rdn = *dn; 541 struct page *page; 542 struct f2fs_node *rn; 543 nid_t child_nid; 544 unsigned int child_nofs; 545 int freed = 0; 546 int i, ret; 547 548 if (dn->nid == 0) 549 return NIDS_PER_BLOCK + 1; 550 551 page = get_node_page(sbi, dn->nid); 552 if (IS_ERR(page)) 553 return PTR_ERR(page); 554 555 rn = (struct f2fs_node *)page_address(page); 556 if (depth < 3) { 557 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) { 558 child_nid = le32_to_cpu(rn->in.nid[i]); 559 if (child_nid == 0) 560 continue; 561 rdn.nid = child_nid; 562 ret = truncate_dnode(&rdn); 563 if (ret < 0) 564 goto out_err; 565 set_nid(page, i, 0, false); 566 } 567 } else { 568 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1; 569 for (i = ofs; i < NIDS_PER_BLOCK; i++) { 570 child_nid = le32_to_cpu(rn->in.nid[i]); 571 if (child_nid == 0) { 572 child_nofs += NIDS_PER_BLOCK + 1; 573 continue; 574 } 575 rdn.nid = child_nid; 576 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1); 577 if (ret == (NIDS_PER_BLOCK + 1)) { 578 set_nid(page, i, 0, false); 579 child_nofs += ret; 580 } else if (ret < 0 && ret != -ENOENT) { 581 goto out_err; 582 } 583 } 584 freed = child_nofs; 585 } 586 587 if (!ofs) { 588 /* remove current indirect node */ 589 dn->node_page = page; 590 truncate_node(dn); 591 freed++; 592 } else { 593 f2fs_put_page(page, 1); 594 } 595 return freed; 596 597 out_err: 598 f2fs_put_page(page, 1); 599 return ret; 600 } 601 602 static int truncate_partial_nodes(struct dnode_of_data *dn, 603 struct f2fs_inode *ri, int *offset, int depth) 604 { 605 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 606 struct page *pages[2]; 607 nid_t nid[3]; 608 nid_t child_nid; 609 int err = 0; 610 int i; 611 int idx = depth - 2; 612 613 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]); 614 if (!nid[0]) 615 return 0; 616 617 /* get indirect nodes in the path */ 618 for (i = 0; i < depth - 1; i++) { 619 /* refernece count'll be increased */ 620 pages[i] = get_node_page(sbi, nid[i]); 621 if (IS_ERR(pages[i])) { 622 depth = i + 1; 623 err = PTR_ERR(pages[i]); 624 goto fail; 625 } 626 nid[i + 1] = get_nid(pages[i], offset[i + 1], false); 627 } 628 629 /* free direct nodes linked to a partial indirect node */ 630 for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) { 631 child_nid = get_nid(pages[idx], i, false); 632 if (!child_nid) 633 continue; 634 dn->nid = child_nid; 635 err = truncate_dnode(dn); 636 if (err < 0) 637 goto fail; 638 set_nid(pages[idx], i, 0, false); 639 } 640 641 if (offset[depth - 1] == 0) { 642 dn->node_page = pages[idx]; 643 dn->nid = nid[idx]; 644 truncate_node(dn); 645 } else { 646 f2fs_put_page(pages[idx], 1); 647 } 648 offset[idx]++; 649 offset[depth - 1] = 0; 650 fail: 651 for (i = depth - 3; i >= 0; i--) 652 f2fs_put_page(pages[i], 1); 653 return err; 654 } 655 656 /* 657 * All the block addresses of data and nodes should be nullified. 658 */ 659 int truncate_inode_blocks(struct inode *inode, pgoff_t from) 660 { 661 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 662 int err = 0, cont = 1; 663 int level, offset[4], noffset[4]; 664 unsigned int nofs = 0; 665 struct f2fs_node *rn; 666 struct dnode_of_data dn; 667 struct page *page; 668 669 level = get_node_path(from, offset, noffset); 670 671 page = get_node_page(sbi, inode->i_ino); 672 if (IS_ERR(page)) 673 return PTR_ERR(page); 674 675 set_new_dnode(&dn, inode, page, NULL, 0); 676 unlock_page(page); 677 678 rn = page_address(page); 679 switch (level) { 680 case 0: 681 case 1: 682 nofs = noffset[1]; 683 break; 684 case 2: 685 nofs = noffset[1]; 686 if (!offset[level - 1]) 687 goto skip_partial; 688 err = truncate_partial_nodes(&dn, &rn->i, offset, level); 689 if (err < 0 && err != -ENOENT) 690 goto fail; 691 nofs += 1 + NIDS_PER_BLOCK; 692 break; 693 case 3: 694 nofs = 5 + 2 * NIDS_PER_BLOCK; 695 if (!offset[level - 1]) 696 goto skip_partial; 697 err = truncate_partial_nodes(&dn, &rn->i, offset, level); 698 if (err < 0 && err != -ENOENT) 699 goto fail; 700 break; 701 default: 702 BUG(); 703 } 704 705 skip_partial: 706 while (cont) { 707 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]); 708 switch (offset[0]) { 709 case NODE_DIR1_BLOCK: 710 case NODE_DIR2_BLOCK: 711 err = truncate_dnode(&dn); 712 break; 713 714 case NODE_IND1_BLOCK: 715 case NODE_IND2_BLOCK: 716 err = truncate_nodes(&dn, nofs, offset[1], 2); 717 break; 718 719 case NODE_DIND_BLOCK: 720 err = truncate_nodes(&dn, nofs, offset[1], 3); 721 cont = 0; 722 break; 723 724 default: 725 BUG(); 726 } 727 if (err < 0 && err != -ENOENT) 728 goto fail; 729 if (offset[1] == 0 && 730 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) { 731 lock_page(page); 732 wait_on_page_writeback(page); 733 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0; 734 set_page_dirty(page); 735 unlock_page(page); 736 } 737 offset[1] = 0; 738 offset[0]++; 739 nofs += err; 740 } 741 fail: 742 f2fs_put_page(page, 0); 743 return err > 0 ? 0 : err; 744 } 745 746 int remove_inode_page(struct inode *inode) 747 { 748 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 749 struct page *page; 750 nid_t ino = inode->i_ino; 751 struct dnode_of_data dn; 752 753 mutex_lock_op(sbi, NODE_TRUNC); 754 page = get_node_page(sbi, ino); 755 if (IS_ERR(page)) { 756 mutex_unlock_op(sbi, NODE_TRUNC); 757 return PTR_ERR(page); 758 } 759 760 if (F2FS_I(inode)->i_xattr_nid) { 761 nid_t nid = F2FS_I(inode)->i_xattr_nid; 762 struct page *npage = get_node_page(sbi, nid); 763 764 if (IS_ERR(npage)) { 765 mutex_unlock_op(sbi, NODE_TRUNC); 766 return PTR_ERR(npage); 767 } 768 769 F2FS_I(inode)->i_xattr_nid = 0; 770 set_new_dnode(&dn, inode, page, npage, nid); 771 dn.inode_page_locked = 1; 772 truncate_node(&dn); 773 } 774 775 /* 0 is possible, after f2fs_new_inode() is failed */ 776 BUG_ON(inode->i_blocks != 0 && inode->i_blocks != 1); 777 set_new_dnode(&dn, inode, page, page, ino); 778 truncate_node(&dn); 779 780 mutex_unlock_op(sbi, NODE_TRUNC); 781 return 0; 782 } 783 784 int new_inode_page(struct inode *inode, const struct qstr *name) 785 { 786 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 787 struct page *page; 788 struct dnode_of_data dn; 789 790 /* allocate inode page for new inode */ 791 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino); 792 mutex_lock_op(sbi, NODE_NEW); 793 page = new_node_page(&dn, 0); 794 init_dent_inode(name, page); 795 mutex_unlock_op(sbi, NODE_NEW); 796 if (IS_ERR(page)) 797 return PTR_ERR(page); 798 f2fs_put_page(page, 1); 799 return 0; 800 } 801 802 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs) 803 { 804 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 805 struct address_space *mapping = sbi->node_inode->i_mapping; 806 struct node_info old_ni, new_ni; 807 struct page *page; 808 int err; 809 810 if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)) 811 return ERR_PTR(-EPERM); 812 813 page = grab_cache_page(mapping, dn->nid); 814 if (!page) 815 return ERR_PTR(-ENOMEM); 816 817 get_node_info(sbi, dn->nid, &old_ni); 818 819 SetPageUptodate(page); 820 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true); 821 822 /* Reinitialize old_ni with new node page */ 823 BUG_ON(old_ni.blk_addr != NULL_ADDR); 824 new_ni = old_ni; 825 new_ni.ino = dn->inode->i_ino; 826 827 if (!inc_valid_node_count(sbi, dn->inode, 1)) { 828 err = -ENOSPC; 829 goto fail; 830 } 831 set_node_addr(sbi, &new_ni, NEW_ADDR); 832 set_cold_node(dn->inode, page); 833 834 dn->node_page = page; 835 sync_inode_page(dn); 836 set_page_dirty(page); 837 if (ofs == 0) 838 inc_valid_inode_count(sbi); 839 840 return page; 841 842 fail: 843 clear_node_page_dirty(page); 844 f2fs_put_page(page, 1); 845 return ERR_PTR(err); 846 } 847 848 static int read_node_page(struct page *page, int type) 849 { 850 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb); 851 struct node_info ni; 852 853 get_node_info(sbi, page->index, &ni); 854 855 if (ni.blk_addr == NULL_ADDR) 856 return -ENOENT; 857 return f2fs_readpage(sbi, page, ni.blk_addr, type); 858 } 859 860 /* 861 * Readahead a node page 862 */ 863 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid) 864 { 865 struct address_space *mapping = sbi->node_inode->i_mapping; 866 struct page *apage; 867 868 apage = find_get_page(mapping, nid); 869 if (apage && PageUptodate(apage)) 870 goto release_out; 871 f2fs_put_page(apage, 0); 872 873 apage = grab_cache_page(mapping, nid); 874 if (!apage) 875 return; 876 877 if (read_node_page(apage, READA)) 878 unlock_page(apage); 879 880 release_out: 881 f2fs_put_page(apage, 0); 882 return; 883 } 884 885 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid) 886 { 887 int err; 888 struct page *page; 889 struct address_space *mapping = sbi->node_inode->i_mapping; 890 891 page = grab_cache_page(mapping, nid); 892 if (!page) 893 return ERR_PTR(-ENOMEM); 894 895 err = read_node_page(page, READ_SYNC); 896 if (err) { 897 f2fs_put_page(page, 1); 898 return ERR_PTR(err); 899 } 900 901 BUG_ON(nid != nid_of_node(page)); 902 mark_page_accessed(page); 903 return page; 904 } 905 906 /* 907 * Return a locked page for the desired node page. 908 * And, readahead MAX_RA_NODE number of node pages. 909 */ 910 struct page *get_node_page_ra(struct page *parent, int start) 911 { 912 struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb); 913 struct address_space *mapping = sbi->node_inode->i_mapping; 914 int i, end; 915 int err = 0; 916 nid_t nid; 917 struct page *page; 918 919 /* First, try getting the desired direct node. */ 920 nid = get_nid(parent, start, false); 921 if (!nid) 922 return ERR_PTR(-ENOENT); 923 924 repeat: 925 page = grab_cache_page(mapping, nid); 926 if (!page) 927 return ERR_PTR(-ENOMEM); 928 else if (PageUptodate(page)) 929 goto page_hit; 930 931 err = read_node_page(page, READ_SYNC); 932 unlock_page(page); 933 if (err) { 934 f2fs_put_page(page, 0); 935 return ERR_PTR(err); 936 } 937 938 /* Then, try readahead for siblings of the desired node */ 939 end = start + MAX_RA_NODE; 940 end = min(end, NIDS_PER_BLOCK); 941 for (i = start + 1; i < end; i++) { 942 nid = get_nid(parent, i, false); 943 if (!nid) 944 continue; 945 ra_node_page(sbi, nid); 946 } 947 948 lock_page(page); 949 950 page_hit: 951 if (PageError(page)) { 952 f2fs_put_page(page, 1); 953 return ERR_PTR(-EIO); 954 } 955 956 /* Has the page been truncated? */ 957 if (page->mapping != mapping) { 958 f2fs_put_page(page, 1); 959 goto repeat; 960 } 961 return page; 962 } 963 964 void sync_inode_page(struct dnode_of_data *dn) 965 { 966 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) { 967 update_inode(dn->inode, dn->node_page); 968 } else if (dn->inode_page) { 969 if (!dn->inode_page_locked) 970 lock_page(dn->inode_page); 971 update_inode(dn->inode, dn->inode_page); 972 if (!dn->inode_page_locked) 973 unlock_page(dn->inode_page); 974 } else { 975 f2fs_write_inode(dn->inode, NULL); 976 } 977 } 978 979 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino, 980 struct writeback_control *wbc) 981 { 982 struct address_space *mapping = sbi->node_inode->i_mapping; 983 pgoff_t index, end; 984 struct pagevec pvec; 985 int step = ino ? 2 : 0; 986 int nwritten = 0, wrote = 0; 987 988 pagevec_init(&pvec, 0); 989 990 next_step: 991 index = 0; 992 end = LONG_MAX; 993 994 while (index <= end) { 995 int i, nr_pages; 996 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 997 PAGECACHE_TAG_DIRTY, 998 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); 999 if (nr_pages == 0) 1000 break; 1001 1002 for (i = 0; i < nr_pages; i++) { 1003 struct page *page = pvec.pages[i]; 1004 1005 /* 1006 * flushing sequence with step: 1007 * 0. indirect nodes 1008 * 1. dentry dnodes 1009 * 2. file dnodes 1010 */ 1011 if (step == 0 && IS_DNODE(page)) 1012 continue; 1013 if (step == 1 && (!IS_DNODE(page) || 1014 is_cold_node(page))) 1015 continue; 1016 if (step == 2 && (!IS_DNODE(page) || 1017 !is_cold_node(page))) 1018 continue; 1019 1020 /* 1021 * If an fsync mode, 1022 * we should not skip writing node pages. 1023 */ 1024 if (ino && ino_of_node(page) == ino) 1025 lock_page(page); 1026 else if (!trylock_page(page)) 1027 continue; 1028 1029 if (unlikely(page->mapping != mapping)) { 1030 continue_unlock: 1031 unlock_page(page); 1032 continue; 1033 } 1034 if (ino && ino_of_node(page) != ino) 1035 goto continue_unlock; 1036 1037 if (!PageDirty(page)) { 1038 /* someone wrote it for us */ 1039 goto continue_unlock; 1040 } 1041 1042 if (!clear_page_dirty_for_io(page)) 1043 goto continue_unlock; 1044 1045 /* called by fsync() */ 1046 if (ino && IS_DNODE(page)) { 1047 int mark = !is_checkpointed_node(sbi, ino); 1048 set_fsync_mark(page, 1); 1049 if (IS_INODE(page)) 1050 set_dentry_mark(page, mark); 1051 nwritten++; 1052 } else { 1053 set_fsync_mark(page, 0); 1054 set_dentry_mark(page, 0); 1055 } 1056 mapping->a_ops->writepage(page, wbc); 1057 wrote++; 1058 1059 if (--wbc->nr_to_write == 0) 1060 break; 1061 } 1062 pagevec_release(&pvec); 1063 cond_resched(); 1064 1065 if (wbc->nr_to_write == 0) { 1066 step = 2; 1067 break; 1068 } 1069 } 1070 1071 if (step < 2) { 1072 step++; 1073 goto next_step; 1074 } 1075 1076 if (wrote) 1077 f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL); 1078 1079 return nwritten; 1080 } 1081 1082 static int f2fs_write_node_page(struct page *page, 1083 struct writeback_control *wbc) 1084 { 1085 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb); 1086 nid_t nid; 1087 block_t new_addr; 1088 struct node_info ni; 1089 1090 if (wbc->for_reclaim) { 1091 dec_page_count(sbi, F2FS_DIRTY_NODES); 1092 wbc->pages_skipped++; 1093 set_page_dirty(page); 1094 return AOP_WRITEPAGE_ACTIVATE; 1095 } 1096 1097 wait_on_page_writeback(page); 1098 1099 mutex_lock_op(sbi, NODE_WRITE); 1100 1101 /* get old block addr of this node page */ 1102 nid = nid_of_node(page); 1103 BUG_ON(page->index != nid); 1104 1105 get_node_info(sbi, nid, &ni); 1106 1107 /* This page is already truncated */ 1108 if (ni.blk_addr == NULL_ADDR) 1109 goto out; 1110 1111 set_page_writeback(page); 1112 1113 /* insert node offset */ 1114 write_node_page(sbi, page, nid, ni.blk_addr, &new_addr); 1115 set_node_addr(sbi, &ni, new_addr); 1116 out: 1117 dec_page_count(sbi, F2FS_DIRTY_NODES); 1118 mutex_unlock_op(sbi, NODE_WRITE); 1119 unlock_page(page); 1120 return 0; 1121 } 1122 1123 /* 1124 * It is very important to gather dirty pages and write at once, so that we can 1125 * submit a big bio without interfering other data writes. 1126 * Be default, 512 pages (2MB), a segment size, is quite reasonable. 1127 */ 1128 #define COLLECT_DIRTY_NODES 512 1129 static int f2fs_write_node_pages(struct address_space *mapping, 1130 struct writeback_control *wbc) 1131 { 1132 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); 1133 struct block_device *bdev = sbi->sb->s_bdev; 1134 long nr_to_write = wbc->nr_to_write; 1135 1136 /* First check balancing cached NAT entries */ 1137 if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) { 1138 write_checkpoint(sbi, false); 1139 return 0; 1140 } 1141 1142 /* collect a number of dirty node pages and write together */ 1143 if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES) 1144 return 0; 1145 1146 /* if mounting is failed, skip writing node pages */ 1147 wbc->nr_to_write = bio_get_nr_vecs(bdev); 1148 sync_node_pages(sbi, 0, wbc); 1149 wbc->nr_to_write = nr_to_write - 1150 (bio_get_nr_vecs(bdev) - wbc->nr_to_write); 1151 return 0; 1152 } 1153 1154 static int f2fs_set_node_page_dirty(struct page *page) 1155 { 1156 struct address_space *mapping = page->mapping; 1157 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); 1158 1159 SetPageUptodate(page); 1160 if (!PageDirty(page)) { 1161 __set_page_dirty_nobuffers(page); 1162 inc_page_count(sbi, F2FS_DIRTY_NODES); 1163 SetPagePrivate(page); 1164 return 1; 1165 } 1166 return 0; 1167 } 1168 1169 static void f2fs_invalidate_node_page(struct page *page, unsigned long offset) 1170 { 1171 struct inode *inode = page->mapping->host; 1172 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 1173 if (PageDirty(page)) 1174 dec_page_count(sbi, F2FS_DIRTY_NODES); 1175 ClearPagePrivate(page); 1176 } 1177 1178 static int f2fs_release_node_page(struct page *page, gfp_t wait) 1179 { 1180 ClearPagePrivate(page); 1181 return 0; 1182 } 1183 1184 /* 1185 * Structure of the f2fs node operations 1186 */ 1187 const struct address_space_operations f2fs_node_aops = { 1188 .writepage = f2fs_write_node_page, 1189 .writepages = f2fs_write_node_pages, 1190 .set_page_dirty = f2fs_set_node_page_dirty, 1191 .invalidatepage = f2fs_invalidate_node_page, 1192 .releasepage = f2fs_release_node_page, 1193 }; 1194 1195 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head) 1196 { 1197 struct list_head *this; 1198 struct free_nid *i; 1199 list_for_each(this, head) { 1200 i = list_entry(this, struct free_nid, list); 1201 if (i->nid == n) 1202 return i; 1203 } 1204 return NULL; 1205 } 1206 1207 static void __del_from_free_nid_list(struct free_nid *i) 1208 { 1209 list_del(&i->list); 1210 kmem_cache_free(free_nid_slab, i); 1211 } 1212 1213 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid) 1214 { 1215 struct free_nid *i; 1216 1217 if (nm_i->fcnt > 2 * MAX_FREE_NIDS) 1218 return 0; 1219 retry: 1220 i = kmem_cache_alloc(free_nid_slab, GFP_NOFS); 1221 if (!i) { 1222 cond_resched(); 1223 goto retry; 1224 } 1225 i->nid = nid; 1226 i->state = NID_NEW; 1227 1228 spin_lock(&nm_i->free_nid_list_lock); 1229 if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) { 1230 spin_unlock(&nm_i->free_nid_list_lock); 1231 kmem_cache_free(free_nid_slab, i); 1232 return 0; 1233 } 1234 list_add_tail(&i->list, &nm_i->free_nid_list); 1235 nm_i->fcnt++; 1236 spin_unlock(&nm_i->free_nid_list_lock); 1237 return 1; 1238 } 1239 1240 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid) 1241 { 1242 struct free_nid *i; 1243 spin_lock(&nm_i->free_nid_list_lock); 1244 i = __lookup_free_nid_list(nid, &nm_i->free_nid_list); 1245 if (i && i->state == NID_NEW) { 1246 __del_from_free_nid_list(i); 1247 nm_i->fcnt--; 1248 } 1249 spin_unlock(&nm_i->free_nid_list_lock); 1250 } 1251 1252 static int scan_nat_page(struct f2fs_nm_info *nm_i, 1253 struct page *nat_page, nid_t start_nid) 1254 { 1255 struct f2fs_nat_block *nat_blk = page_address(nat_page); 1256 block_t blk_addr; 1257 int fcnt = 0; 1258 int i; 1259 1260 /* 0 nid should not be used */ 1261 if (start_nid == 0) 1262 ++start_nid; 1263 1264 i = start_nid % NAT_ENTRY_PER_BLOCK; 1265 1266 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) { 1267 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr); 1268 BUG_ON(blk_addr == NEW_ADDR); 1269 if (blk_addr == NULL_ADDR) 1270 fcnt += add_free_nid(nm_i, start_nid); 1271 } 1272 return fcnt; 1273 } 1274 1275 static void build_free_nids(struct f2fs_sb_info *sbi) 1276 { 1277 struct free_nid *fnid, *next_fnid; 1278 struct f2fs_nm_info *nm_i = NM_I(sbi); 1279 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 1280 struct f2fs_summary_block *sum = curseg->sum_blk; 1281 nid_t nid = 0; 1282 bool is_cycled = false; 1283 int fcnt = 0; 1284 int i; 1285 1286 nid = nm_i->next_scan_nid; 1287 nm_i->init_scan_nid = nid; 1288 1289 ra_nat_pages(sbi, nid); 1290 1291 while (1) { 1292 struct page *page = get_current_nat_page(sbi, nid); 1293 1294 fcnt += scan_nat_page(nm_i, page, nid); 1295 f2fs_put_page(page, 1); 1296 1297 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK)); 1298 1299 if (nid >= nm_i->max_nid) { 1300 nid = 0; 1301 is_cycled = true; 1302 } 1303 if (fcnt > MAX_FREE_NIDS) 1304 break; 1305 if (is_cycled && nm_i->init_scan_nid <= nid) 1306 break; 1307 } 1308 1309 nm_i->next_scan_nid = nid; 1310 1311 /* find free nids from current sum_pages */ 1312 mutex_lock(&curseg->curseg_mutex); 1313 for (i = 0; i < nats_in_cursum(sum); i++) { 1314 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr); 1315 nid = le32_to_cpu(nid_in_journal(sum, i)); 1316 if (addr == NULL_ADDR) 1317 add_free_nid(nm_i, nid); 1318 else 1319 remove_free_nid(nm_i, nid); 1320 } 1321 mutex_unlock(&curseg->curseg_mutex); 1322 1323 /* remove the free nids from current allocated nids */ 1324 list_for_each_entry_safe(fnid, next_fnid, &nm_i->free_nid_list, list) { 1325 struct nat_entry *ne; 1326 1327 read_lock(&nm_i->nat_tree_lock); 1328 ne = __lookup_nat_cache(nm_i, fnid->nid); 1329 if (ne && nat_get_blkaddr(ne) != NULL_ADDR) 1330 remove_free_nid(nm_i, fnid->nid); 1331 read_unlock(&nm_i->nat_tree_lock); 1332 } 1333 } 1334 1335 /* 1336 * If this function returns success, caller can obtain a new nid 1337 * from second parameter of this function. 1338 * The returned nid could be used ino as well as nid when inode is created. 1339 */ 1340 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid) 1341 { 1342 struct f2fs_nm_info *nm_i = NM_I(sbi); 1343 struct free_nid *i = NULL; 1344 struct list_head *this; 1345 retry: 1346 mutex_lock(&nm_i->build_lock); 1347 if (!nm_i->fcnt) { 1348 /* scan NAT in order to build free nid list */ 1349 build_free_nids(sbi); 1350 if (!nm_i->fcnt) { 1351 mutex_unlock(&nm_i->build_lock); 1352 return false; 1353 } 1354 } 1355 mutex_unlock(&nm_i->build_lock); 1356 1357 /* 1358 * We check fcnt again since previous check is racy as 1359 * we didn't hold free_nid_list_lock. So other thread 1360 * could consume all of free nids. 1361 */ 1362 spin_lock(&nm_i->free_nid_list_lock); 1363 if (!nm_i->fcnt) { 1364 spin_unlock(&nm_i->free_nid_list_lock); 1365 goto retry; 1366 } 1367 1368 BUG_ON(list_empty(&nm_i->free_nid_list)); 1369 list_for_each(this, &nm_i->free_nid_list) { 1370 i = list_entry(this, struct free_nid, list); 1371 if (i->state == NID_NEW) 1372 break; 1373 } 1374 1375 BUG_ON(i->state != NID_NEW); 1376 *nid = i->nid; 1377 i->state = NID_ALLOC; 1378 nm_i->fcnt--; 1379 spin_unlock(&nm_i->free_nid_list_lock); 1380 return true; 1381 } 1382 1383 /* 1384 * alloc_nid() should be called prior to this function. 1385 */ 1386 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid) 1387 { 1388 struct f2fs_nm_info *nm_i = NM_I(sbi); 1389 struct free_nid *i; 1390 1391 spin_lock(&nm_i->free_nid_list_lock); 1392 i = __lookup_free_nid_list(nid, &nm_i->free_nid_list); 1393 if (i) { 1394 BUG_ON(i->state != NID_ALLOC); 1395 __del_from_free_nid_list(i); 1396 } 1397 spin_unlock(&nm_i->free_nid_list_lock); 1398 } 1399 1400 /* 1401 * alloc_nid() should be called prior to this function. 1402 */ 1403 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid) 1404 { 1405 alloc_nid_done(sbi, nid); 1406 add_free_nid(NM_I(sbi), nid); 1407 } 1408 1409 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page, 1410 struct f2fs_summary *sum, struct node_info *ni, 1411 block_t new_blkaddr) 1412 { 1413 rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr); 1414 set_node_addr(sbi, ni, new_blkaddr); 1415 clear_node_page_dirty(page); 1416 } 1417 1418 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page) 1419 { 1420 struct address_space *mapping = sbi->node_inode->i_mapping; 1421 struct f2fs_node *src, *dst; 1422 nid_t ino = ino_of_node(page); 1423 struct node_info old_ni, new_ni; 1424 struct page *ipage; 1425 1426 ipage = grab_cache_page(mapping, ino); 1427 if (!ipage) 1428 return -ENOMEM; 1429 1430 /* Should not use this inode from free nid list */ 1431 remove_free_nid(NM_I(sbi), ino); 1432 1433 get_node_info(sbi, ino, &old_ni); 1434 SetPageUptodate(ipage); 1435 fill_node_footer(ipage, ino, ino, 0, true); 1436 1437 src = (struct f2fs_node *)page_address(page); 1438 dst = (struct f2fs_node *)page_address(ipage); 1439 1440 memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i); 1441 dst->i.i_size = 0; 1442 dst->i.i_blocks = cpu_to_le64(1); 1443 dst->i.i_links = cpu_to_le32(1); 1444 dst->i.i_xattr_nid = 0; 1445 1446 new_ni = old_ni; 1447 new_ni.ino = ino; 1448 1449 set_node_addr(sbi, &new_ni, NEW_ADDR); 1450 inc_valid_inode_count(sbi); 1451 1452 f2fs_put_page(ipage, 1); 1453 return 0; 1454 } 1455 1456 int restore_node_summary(struct f2fs_sb_info *sbi, 1457 unsigned int segno, struct f2fs_summary_block *sum) 1458 { 1459 struct f2fs_node *rn; 1460 struct f2fs_summary *sum_entry; 1461 struct page *page; 1462 block_t addr; 1463 int i, last_offset; 1464 1465 /* alloc temporal page for read node */ 1466 page = alloc_page(GFP_NOFS | __GFP_ZERO); 1467 if (IS_ERR(page)) 1468 return PTR_ERR(page); 1469 lock_page(page); 1470 1471 /* scan the node segment */ 1472 last_offset = sbi->blocks_per_seg; 1473 addr = START_BLOCK(sbi, segno); 1474 sum_entry = &sum->entries[0]; 1475 1476 for (i = 0; i < last_offset; i++, sum_entry++) { 1477 if (f2fs_readpage(sbi, page, addr, READ_SYNC)) 1478 goto out; 1479 1480 rn = (struct f2fs_node *)page_address(page); 1481 sum_entry->nid = rn->footer.nid; 1482 sum_entry->version = 0; 1483 sum_entry->ofs_in_node = 0; 1484 addr++; 1485 1486 /* 1487 * In order to read next node page, 1488 * we must clear PageUptodate flag. 1489 */ 1490 ClearPageUptodate(page); 1491 } 1492 out: 1493 unlock_page(page); 1494 __free_pages(page, 0); 1495 return 0; 1496 } 1497 1498 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi) 1499 { 1500 struct f2fs_nm_info *nm_i = NM_I(sbi); 1501 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 1502 struct f2fs_summary_block *sum = curseg->sum_blk; 1503 int i; 1504 1505 mutex_lock(&curseg->curseg_mutex); 1506 1507 if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) { 1508 mutex_unlock(&curseg->curseg_mutex); 1509 return false; 1510 } 1511 1512 for (i = 0; i < nats_in_cursum(sum); i++) { 1513 struct nat_entry *ne; 1514 struct f2fs_nat_entry raw_ne; 1515 nid_t nid = le32_to_cpu(nid_in_journal(sum, i)); 1516 1517 raw_ne = nat_in_journal(sum, i); 1518 retry: 1519 write_lock(&nm_i->nat_tree_lock); 1520 ne = __lookup_nat_cache(nm_i, nid); 1521 if (ne) { 1522 __set_nat_cache_dirty(nm_i, ne); 1523 write_unlock(&nm_i->nat_tree_lock); 1524 continue; 1525 } 1526 ne = grab_nat_entry(nm_i, nid); 1527 if (!ne) { 1528 write_unlock(&nm_i->nat_tree_lock); 1529 goto retry; 1530 } 1531 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr)); 1532 nat_set_ino(ne, le32_to_cpu(raw_ne.ino)); 1533 nat_set_version(ne, raw_ne.version); 1534 __set_nat_cache_dirty(nm_i, ne); 1535 write_unlock(&nm_i->nat_tree_lock); 1536 } 1537 update_nats_in_cursum(sum, -i); 1538 mutex_unlock(&curseg->curseg_mutex); 1539 return true; 1540 } 1541 1542 /* 1543 * This function is called during the checkpointing process. 1544 */ 1545 void flush_nat_entries(struct f2fs_sb_info *sbi) 1546 { 1547 struct f2fs_nm_info *nm_i = NM_I(sbi); 1548 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA); 1549 struct f2fs_summary_block *sum = curseg->sum_blk; 1550 struct list_head *cur, *n; 1551 struct page *page = NULL; 1552 struct f2fs_nat_block *nat_blk = NULL; 1553 nid_t start_nid = 0, end_nid = 0; 1554 bool flushed; 1555 1556 flushed = flush_nats_in_journal(sbi); 1557 1558 if (!flushed) 1559 mutex_lock(&curseg->curseg_mutex); 1560 1561 /* 1) flush dirty nat caches */ 1562 list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) { 1563 struct nat_entry *ne; 1564 nid_t nid; 1565 struct f2fs_nat_entry raw_ne; 1566 int offset = -1; 1567 block_t new_blkaddr; 1568 1569 ne = list_entry(cur, struct nat_entry, list); 1570 nid = nat_get_nid(ne); 1571 1572 if (nat_get_blkaddr(ne) == NEW_ADDR) 1573 continue; 1574 if (flushed) 1575 goto to_nat_page; 1576 1577 /* if there is room for nat enries in curseg->sumpage */ 1578 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1); 1579 if (offset >= 0) { 1580 raw_ne = nat_in_journal(sum, offset); 1581 goto flush_now; 1582 } 1583 to_nat_page: 1584 if (!page || (start_nid > nid || nid > end_nid)) { 1585 if (page) { 1586 f2fs_put_page(page, 1); 1587 page = NULL; 1588 } 1589 start_nid = START_NID(nid); 1590 end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1; 1591 1592 /* 1593 * get nat block with dirty flag, increased reference 1594 * count, mapped and lock 1595 */ 1596 page = get_next_nat_page(sbi, start_nid); 1597 nat_blk = page_address(page); 1598 } 1599 1600 BUG_ON(!nat_blk); 1601 raw_ne = nat_blk->entries[nid - start_nid]; 1602 flush_now: 1603 new_blkaddr = nat_get_blkaddr(ne); 1604 1605 raw_ne.ino = cpu_to_le32(nat_get_ino(ne)); 1606 raw_ne.block_addr = cpu_to_le32(new_blkaddr); 1607 raw_ne.version = nat_get_version(ne); 1608 1609 if (offset < 0) { 1610 nat_blk->entries[nid - start_nid] = raw_ne; 1611 } else { 1612 nat_in_journal(sum, offset) = raw_ne; 1613 nid_in_journal(sum, offset) = cpu_to_le32(nid); 1614 } 1615 1616 if (nat_get_blkaddr(ne) == NULL_ADDR) { 1617 write_lock(&nm_i->nat_tree_lock); 1618 __del_from_nat_cache(nm_i, ne); 1619 write_unlock(&nm_i->nat_tree_lock); 1620 1621 /* We can reuse this freed nid at this point */ 1622 add_free_nid(NM_I(sbi), nid); 1623 } else { 1624 write_lock(&nm_i->nat_tree_lock); 1625 __clear_nat_cache_dirty(nm_i, ne); 1626 ne->checkpointed = true; 1627 write_unlock(&nm_i->nat_tree_lock); 1628 } 1629 } 1630 if (!flushed) 1631 mutex_unlock(&curseg->curseg_mutex); 1632 f2fs_put_page(page, 1); 1633 1634 /* 2) shrink nat caches if necessary */ 1635 try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD); 1636 } 1637 1638 static int init_node_manager(struct f2fs_sb_info *sbi) 1639 { 1640 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi); 1641 struct f2fs_nm_info *nm_i = NM_I(sbi); 1642 unsigned char *version_bitmap; 1643 unsigned int nat_segs, nat_blocks; 1644 1645 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr); 1646 1647 /* segment_count_nat includes pair segment so divide to 2. */ 1648 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1; 1649 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg); 1650 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks; 1651 nm_i->fcnt = 0; 1652 nm_i->nat_cnt = 0; 1653 1654 INIT_LIST_HEAD(&nm_i->free_nid_list); 1655 INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC); 1656 INIT_LIST_HEAD(&nm_i->nat_entries); 1657 INIT_LIST_HEAD(&nm_i->dirty_nat_entries); 1658 1659 mutex_init(&nm_i->build_lock); 1660 spin_lock_init(&nm_i->free_nid_list_lock); 1661 rwlock_init(&nm_i->nat_tree_lock); 1662 1663 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP); 1664 nm_i->init_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 1665 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid); 1666 1667 nm_i->nat_bitmap = kzalloc(nm_i->bitmap_size, GFP_KERNEL); 1668 if (!nm_i->nat_bitmap) 1669 return -ENOMEM; 1670 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP); 1671 if (!version_bitmap) 1672 return -EFAULT; 1673 1674 /* copy version bitmap */ 1675 memcpy(nm_i->nat_bitmap, version_bitmap, nm_i->bitmap_size); 1676 return 0; 1677 } 1678 1679 int build_node_manager(struct f2fs_sb_info *sbi) 1680 { 1681 int err; 1682 1683 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL); 1684 if (!sbi->nm_info) 1685 return -ENOMEM; 1686 1687 err = init_node_manager(sbi); 1688 if (err) 1689 return err; 1690 1691 build_free_nids(sbi); 1692 return 0; 1693 } 1694 1695 void destroy_node_manager(struct f2fs_sb_info *sbi) 1696 { 1697 struct f2fs_nm_info *nm_i = NM_I(sbi); 1698 struct free_nid *i, *next_i; 1699 struct nat_entry *natvec[NATVEC_SIZE]; 1700 nid_t nid = 0; 1701 unsigned int found; 1702 1703 if (!nm_i) 1704 return; 1705 1706 /* destroy free nid list */ 1707 spin_lock(&nm_i->free_nid_list_lock); 1708 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) { 1709 BUG_ON(i->state == NID_ALLOC); 1710 __del_from_free_nid_list(i); 1711 nm_i->fcnt--; 1712 } 1713 BUG_ON(nm_i->fcnt); 1714 spin_unlock(&nm_i->free_nid_list_lock); 1715 1716 /* destroy nat cache */ 1717 write_lock(&nm_i->nat_tree_lock); 1718 while ((found = __gang_lookup_nat_cache(nm_i, 1719 nid, NATVEC_SIZE, natvec))) { 1720 unsigned idx; 1721 for (idx = 0; idx < found; idx++) { 1722 struct nat_entry *e = natvec[idx]; 1723 nid = nat_get_nid(e) + 1; 1724 __del_from_nat_cache(nm_i, e); 1725 } 1726 } 1727 BUG_ON(nm_i->nat_cnt); 1728 write_unlock(&nm_i->nat_tree_lock); 1729 1730 kfree(nm_i->nat_bitmap); 1731 sbi->nm_info = NULL; 1732 kfree(nm_i); 1733 } 1734 1735 int __init create_node_manager_caches(void) 1736 { 1737 nat_entry_slab = f2fs_kmem_cache_create("nat_entry", 1738 sizeof(struct nat_entry), NULL); 1739 if (!nat_entry_slab) 1740 return -ENOMEM; 1741 1742 free_nid_slab = f2fs_kmem_cache_create("free_nid", 1743 sizeof(struct free_nid), NULL); 1744 if (!free_nid_slab) { 1745 kmem_cache_destroy(nat_entry_slab); 1746 return -ENOMEM; 1747 } 1748 return 0; 1749 } 1750 1751 void destroy_node_manager_caches(void) 1752 { 1753 kmem_cache_destroy(free_nid_slab); 1754 kmem_cache_destroy(nat_entry_slab); 1755 } 1756