1 /* 2 * fs/f2fs/segment.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/bio.h> 14 #include <linux/blkdev.h> 15 #include <linux/prefetch.h> 16 #include <linux/kthread.h> 17 #include <linux/swap.h> 18 #include <linux/timer.h> 19 20 #include "f2fs.h" 21 #include "segment.h" 22 #include "node.h" 23 #include "trace.h" 24 #include <trace/events/f2fs.h> 25 26 #define __reverse_ffz(x) __reverse_ffs(~(x)) 27 28 static struct kmem_cache *discard_entry_slab; 29 static struct kmem_cache *sit_entry_set_slab; 30 static struct kmem_cache *inmem_entry_slab; 31 32 static unsigned long __reverse_ulong(unsigned char *str) 33 { 34 unsigned long tmp = 0; 35 int shift = 24, idx = 0; 36 37 #if BITS_PER_LONG == 64 38 shift = 56; 39 #endif 40 while (shift >= 0) { 41 tmp |= (unsigned long)str[idx++] << shift; 42 shift -= BITS_PER_BYTE; 43 } 44 return tmp; 45 } 46 47 /* 48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since 49 * MSB and LSB are reversed in a byte by f2fs_set_bit. 50 */ 51 static inline unsigned long __reverse_ffs(unsigned long word) 52 { 53 int num = 0; 54 55 #if BITS_PER_LONG == 64 56 if ((word & 0xffffffff00000000UL) == 0) 57 num += 32; 58 else 59 word >>= 32; 60 #endif 61 if ((word & 0xffff0000) == 0) 62 num += 16; 63 else 64 word >>= 16; 65 66 if ((word & 0xff00) == 0) 67 num += 8; 68 else 69 word >>= 8; 70 71 if ((word & 0xf0) == 0) 72 num += 4; 73 else 74 word >>= 4; 75 76 if ((word & 0xc) == 0) 77 num += 2; 78 else 79 word >>= 2; 80 81 if ((word & 0x2) == 0) 82 num += 1; 83 return num; 84 } 85 86 /* 87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because 88 * f2fs_set_bit makes MSB and LSB reversed in a byte. 89 * @size must be integral times of unsigned long. 90 * Example: 91 * MSB <--> LSB 92 * f2fs_set_bit(0, bitmap) => 1000 0000 93 * f2fs_set_bit(7, bitmap) => 0000 0001 94 */ 95 static unsigned long __find_rev_next_bit(const unsigned long *addr, 96 unsigned long size, unsigned long offset) 97 { 98 const unsigned long *p = addr + BIT_WORD(offset); 99 unsigned long result = size; 100 unsigned long tmp; 101 102 if (offset >= size) 103 return size; 104 105 size -= (offset & ~(BITS_PER_LONG - 1)); 106 offset %= BITS_PER_LONG; 107 108 while (1) { 109 if (*p == 0) 110 goto pass; 111 112 tmp = __reverse_ulong((unsigned char *)p); 113 114 tmp &= ~0UL >> offset; 115 if (size < BITS_PER_LONG) 116 tmp &= (~0UL << (BITS_PER_LONG - size)); 117 if (tmp) 118 goto found; 119 pass: 120 if (size <= BITS_PER_LONG) 121 break; 122 size -= BITS_PER_LONG; 123 offset = 0; 124 p++; 125 } 126 return result; 127 found: 128 return result - size + __reverse_ffs(tmp); 129 } 130 131 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, 132 unsigned long size, unsigned long offset) 133 { 134 const unsigned long *p = addr + BIT_WORD(offset); 135 unsigned long result = size; 136 unsigned long tmp; 137 138 if (offset >= size) 139 return size; 140 141 size -= (offset & ~(BITS_PER_LONG - 1)); 142 offset %= BITS_PER_LONG; 143 144 while (1) { 145 if (*p == ~0UL) 146 goto pass; 147 148 tmp = __reverse_ulong((unsigned char *)p); 149 150 if (offset) 151 tmp |= ~0UL << (BITS_PER_LONG - offset); 152 if (size < BITS_PER_LONG) 153 tmp |= ~0UL >> size; 154 if (tmp != ~0UL) 155 goto found; 156 pass: 157 if (size <= BITS_PER_LONG) 158 break; 159 size -= BITS_PER_LONG; 160 offset = 0; 161 p++; 162 } 163 return result; 164 found: 165 return result - size + __reverse_ffz(tmp); 166 } 167 168 void register_inmem_page(struct inode *inode, struct page *page) 169 { 170 struct f2fs_inode_info *fi = F2FS_I(inode); 171 struct inmem_pages *new; 172 173 f2fs_trace_pid(page); 174 175 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE); 176 SetPagePrivate(page); 177 178 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); 179 180 /* add atomic page indices to the list */ 181 new->page = page; 182 INIT_LIST_HEAD(&new->list); 183 184 /* increase reference count with clean state */ 185 mutex_lock(&fi->inmem_lock); 186 get_page(page); 187 list_add_tail(&new->list, &fi->inmem_pages); 188 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 189 mutex_unlock(&fi->inmem_lock); 190 191 trace_f2fs_register_inmem_page(page, INMEM); 192 } 193 194 int commit_inmem_pages(struct inode *inode, bool abort) 195 { 196 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 197 struct f2fs_inode_info *fi = F2FS_I(inode); 198 struct inmem_pages *cur, *tmp; 199 bool submit_bio = false; 200 struct f2fs_io_info fio = { 201 .sbi = sbi, 202 .type = DATA, 203 .rw = WRITE_SYNC | REQ_PRIO, 204 .encrypted_page = NULL, 205 }; 206 int err = 0; 207 208 /* 209 * The abort is true only when f2fs_evict_inode is called. 210 * Basically, the f2fs_evict_inode doesn't produce any data writes, so 211 * that we don't need to call f2fs_balance_fs. 212 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this 213 * inode becomes free by iget_locked in f2fs_iget. 214 */ 215 if (!abort) { 216 f2fs_balance_fs(sbi, true); 217 f2fs_lock_op(sbi); 218 } 219 220 mutex_lock(&fi->inmem_lock); 221 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { 222 lock_page(cur->page); 223 if (!abort) { 224 if (cur->page->mapping == inode->i_mapping) { 225 set_page_dirty(cur->page); 226 f2fs_wait_on_page_writeback(cur->page, DATA); 227 if (clear_page_dirty_for_io(cur->page)) 228 inode_dec_dirty_pages(inode); 229 trace_f2fs_commit_inmem_page(cur->page, INMEM); 230 fio.page = cur->page; 231 err = do_write_data_page(&fio); 232 if (err) { 233 unlock_page(cur->page); 234 break; 235 } 236 clear_cold_data(cur->page); 237 submit_bio = true; 238 } 239 } else { 240 ClearPageUptodate(cur->page); 241 trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP); 242 } 243 set_page_private(cur->page, 0); 244 ClearPagePrivate(cur->page); 245 f2fs_put_page(cur->page, 1); 246 247 list_del(&cur->list); 248 kmem_cache_free(inmem_entry_slab, cur); 249 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 250 } 251 mutex_unlock(&fi->inmem_lock); 252 253 if (!abort) { 254 f2fs_unlock_op(sbi); 255 if (submit_bio) 256 f2fs_submit_merged_bio(sbi, DATA, WRITE); 257 } 258 return err; 259 } 260 261 /* 262 * This function balances dirty node and dentry pages. 263 * In addition, it controls garbage collection. 264 */ 265 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 266 { 267 if (!need) 268 return; 269 /* 270 * We should do GC or end up with checkpoint, if there are so many dirty 271 * dir/node pages without enough free segments. 272 */ 273 if (has_not_enough_free_secs(sbi, 0)) { 274 mutex_lock(&sbi->gc_mutex); 275 f2fs_gc(sbi, false); 276 } 277 } 278 279 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) 280 { 281 /* try to shrink extent cache when there is no enough memory */ 282 if (!available_free_memory(sbi, EXTENT_CACHE)) 283 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); 284 285 /* check the # of cached NAT entries */ 286 if (!available_free_memory(sbi, NAT_ENTRIES)) 287 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 288 289 if (!available_free_memory(sbi, FREE_NIDS)) 290 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES); 291 292 /* checkpoint is the only way to shrink partial cached entries */ 293 if (!available_free_memory(sbi, NAT_ENTRIES) || 294 excess_prefree_segs(sbi) || 295 !available_free_memory(sbi, INO_ENTRIES) || 296 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) { 297 if (test_opt(sbi, DATA_FLUSH)) 298 sync_dirty_inodes(sbi, FILE_INODE); 299 f2fs_sync_fs(sbi->sb, true); 300 stat_inc_bg_cp_count(sbi->stat_info); 301 } 302 } 303 304 static int issue_flush_thread(void *data) 305 { 306 struct f2fs_sb_info *sbi = data; 307 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; 308 wait_queue_head_t *q = &fcc->flush_wait_queue; 309 repeat: 310 if (kthread_should_stop()) 311 return 0; 312 313 if (!llist_empty(&fcc->issue_list)) { 314 struct bio *bio; 315 struct flush_cmd *cmd, *next; 316 int ret; 317 318 bio = f2fs_bio_alloc(0); 319 320 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 321 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 322 323 bio->bi_bdev = sbi->sb->s_bdev; 324 ret = submit_bio_wait(WRITE_FLUSH, bio); 325 326 llist_for_each_entry_safe(cmd, next, 327 fcc->dispatch_list, llnode) { 328 cmd->ret = ret; 329 complete(&cmd->wait); 330 } 331 bio_put(bio); 332 fcc->dispatch_list = NULL; 333 } 334 335 wait_event_interruptible(*q, 336 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 337 goto repeat; 338 } 339 340 int f2fs_issue_flush(struct f2fs_sb_info *sbi) 341 { 342 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; 343 struct flush_cmd cmd; 344 345 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER), 346 test_opt(sbi, FLUSH_MERGE)); 347 348 if (test_opt(sbi, NOBARRIER)) 349 return 0; 350 351 if (!test_opt(sbi, FLUSH_MERGE)) { 352 struct bio *bio = f2fs_bio_alloc(0); 353 int ret; 354 355 bio->bi_bdev = sbi->sb->s_bdev; 356 ret = submit_bio_wait(WRITE_FLUSH, bio); 357 bio_put(bio); 358 return ret; 359 } 360 361 init_completion(&cmd.wait); 362 363 llist_add(&cmd.llnode, &fcc->issue_list); 364 365 if (!fcc->dispatch_list) 366 wake_up(&fcc->flush_wait_queue); 367 368 wait_for_completion(&cmd.wait); 369 370 return cmd.ret; 371 } 372 373 int create_flush_cmd_control(struct f2fs_sb_info *sbi) 374 { 375 dev_t dev = sbi->sb->s_bdev->bd_dev; 376 struct flush_cmd_control *fcc; 377 int err = 0; 378 379 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL); 380 if (!fcc) 381 return -ENOMEM; 382 init_waitqueue_head(&fcc->flush_wait_queue); 383 init_llist_head(&fcc->issue_list); 384 SM_I(sbi)->cmd_control_info = fcc; 385 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 386 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 387 if (IS_ERR(fcc->f2fs_issue_flush)) { 388 err = PTR_ERR(fcc->f2fs_issue_flush); 389 kfree(fcc); 390 SM_I(sbi)->cmd_control_info = NULL; 391 return err; 392 } 393 394 return err; 395 } 396 397 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi) 398 { 399 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; 400 401 if (fcc && fcc->f2fs_issue_flush) 402 kthread_stop(fcc->f2fs_issue_flush); 403 kfree(fcc); 404 SM_I(sbi)->cmd_control_info = NULL; 405 } 406 407 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 408 enum dirty_type dirty_type) 409 { 410 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 411 412 /* need not be added */ 413 if (IS_CURSEG(sbi, segno)) 414 return; 415 416 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 417 dirty_i->nr_dirty[dirty_type]++; 418 419 if (dirty_type == DIRTY) { 420 struct seg_entry *sentry = get_seg_entry(sbi, segno); 421 enum dirty_type t = sentry->type; 422 423 if (unlikely(t >= DIRTY)) { 424 f2fs_bug_on(sbi, 1); 425 return; 426 } 427 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 428 dirty_i->nr_dirty[t]++; 429 } 430 } 431 432 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 433 enum dirty_type dirty_type) 434 { 435 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 436 437 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 438 dirty_i->nr_dirty[dirty_type]--; 439 440 if (dirty_type == DIRTY) { 441 struct seg_entry *sentry = get_seg_entry(sbi, segno); 442 enum dirty_type t = sentry->type; 443 444 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 445 dirty_i->nr_dirty[t]--; 446 447 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0) 448 clear_bit(GET_SECNO(sbi, segno), 449 dirty_i->victim_secmap); 450 } 451 } 452 453 /* 454 * Should not occur error such as -ENOMEM. 455 * Adding dirty entry into seglist is not critical operation. 456 * If a given segment is one of current working segments, it won't be added. 457 */ 458 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 459 { 460 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 461 unsigned short valid_blocks; 462 463 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 464 return; 465 466 mutex_lock(&dirty_i->seglist_lock); 467 468 valid_blocks = get_valid_blocks(sbi, segno, 0); 469 470 if (valid_blocks == 0) { 471 __locate_dirty_segment(sbi, segno, PRE); 472 __remove_dirty_segment(sbi, segno, DIRTY); 473 } else if (valid_blocks < sbi->blocks_per_seg) { 474 __locate_dirty_segment(sbi, segno, DIRTY); 475 } else { 476 /* Recovery routine with SSR needs this */ 477 __remove_dirty_segment(sbi, segno, DIRTY); 478 } 479 480 mutex_unlock(&dirty_i->seglist_lock); 481 } 482 483 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 484 block_t blkstart, block_t blklen) 485 { 486 sector_t start = SECTOR_FROM_BLOCK(blkstart); 487 sector_t len = SECTOR_FROM_BLOCK(blklen); 488 struct seg_entry *se; 489 unsigned int offset; 490 block_t i; 491 492 for (i = blkstart; i < blkstart + blklen; i++) { 493 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 494 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 495 496 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 497 sbi->discard_blks--; 498 } 499 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen); 500 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0); 501 } 502 503 bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr) 504 { 505 int err = -ENOTSUPP; 506 507 if (test_opt(sbi, DISCARD)) { 508 struct seg_entry *se = get_seg_entry(sbi, 509 GET_SEGNO(sbi, blkaddr)); 510 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 511 512 if (f2fs_test_bit(offset, se->discard_map)) 513 return false; 514 515 err = f2fs_issue_discard(sbi, blkaddr, 1); 516 } 517 518 if (err) { 519 update_meta_page(sbi, NULL, blkaddr); 520 return true; 521 } 522 return false; 523 } 524 525 static void __add_discard_entry(struct f2fs_sb_info *sbi, 526 struct cp_control *cpc, struct seg_entry *se, 527 unsigned int start, unsigned int end) 528 { 529 struct list_head *head = &SM_I(sbi)->discard_list; 530 struct discard_entry *new, *last; 531 532 if (!list_empty(head)) { 533 last = list_last_entry(head, struct discard_entry, list); 534 if (START_BLOCK(sbi, cpc->trim_start) + start == 535 last->blkaddr + last->len) { 536 last->len += end - start; 537 goto done; 538 } 539 } 540 541 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS); 542 INIT_LIST_HEAD(&new->list); 543 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start; 544 new->len = end - start; 545 list_add_tail(&new->list, head); 546 done: 547 SM_I(sbi)->nr_discards += end - start; 548 } 549 550 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc) 551 { 552 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 553 int max_blocks = sbi->blocks_per_seg; 554 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 555 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 556 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 557 unsigned long *discard_map = (unsigned long *)se->discard_map; 558 unsigned long *dmap = SIT_I(sbi)->tmp_map; 559 unsigned int start = 0, end = -1; 560 bool force = (cpc->reason == CP_DISCARD); 561 int i; 562 563 if (se->valid_blocks == max_blocks) 564 return; 565 566 if (!force) { 567 if (!test_opt(sbi, DISCARD) || !se->valid_blocks || 568 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards) 569 return; 570 } 571 572 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 573 for (i = 0; i < entries; i++) 574 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 575 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 576 577 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) { 578 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 579 if (start >= max_blocks) 580 break; 581 582 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 583 __add_discard_entry(sbi, cpc, se, start, end); 584 } 585 } 586 587 void release_discard_addrs(struct f2fs_sb_info *sbi) 588 { 589 struct list_head *head = &(SM_I(sbi)->discard_list); 590 struct discard_entry *entry, *this; 591 592 /* drop caches */ 593 list_for_each_entry_safe(entry, this, head, list) { 594 list_del(&entry->list); 595 kmem_cache_free(discard_entry_slab, entry); 596 } 597 } 598 599 /* 600 * Should call clear_prefree_segments after checkpoint is done. 601 */ 602 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 603 { 604 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 605 unsigned int segno; 606 607 mutex_lock(&dirty_i->seglist_lock); 608 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 609 __set_test_and_free(sbi, segno); 610 mutex_unlock(&dirty_i->seglist_lock); 611 } 612 613 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc) 614 { 615 struct list_head *head = &(SM_I(sbi)->discard_list); 616 struct discard_entry *entry, *this; 617 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 618 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 619 unsigned int start = 0, end = -1; 620 621 mutex_lock(&dirty_i->seglist_lock); 622 623 while (1) { 624 int i; 625 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 626 if (start >= MAIN_SEGS(sbi)) 627 break; 628 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 629 start + 1); 630 631 for (i = start; i < end; i++) 632 clear_bit(i, prefree_map); 633 634 dirty_i->nr_dirty[PRE] -= end - start; 635 636 if (!test_opt(sbi, DISCARD)) 637 continue; 638 639 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 640 (end - start) << sbi->log_blocks_per_seg); 641 } 642 mutex_unlock(&dirty_i->seglist_lock); 643 644 /* send small discards */ 645 list_for_each_entry_safe(entry, this, head, list) { 646 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen) 647 goto skip; 648 f2fs_issue_discard(sbi, entry->blkaddr, entry->len); 649 cpc->trimmed += entry->len; 650 skip: 651 list_del(&entry->list); 652 SM_I(sbi)->nr_discards -= entry->len; 653 kmem_cache_free(discard_entry_slab, entry); 654 } 655 } 656 657 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 658 { 659 struct sit_info *sit_i = SIT_I(sbi); 660 661 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 662 sit_i->dirty_sentries++; 663 return false; 664 } 665 666 return true; 667 } 668 669 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 670 unsigned int segno, int modified) 671 { 672 struct seg_entry *se = get_seg_entry(sbi, segno); 673 se->type = type; 674 if (modified) 675 __mark_sit_entry_dirty(sbi, segno); 676 } 677 678 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 679 { 680 struct seg_entry *se; 681 unsigned int segno, offset; 682 long int new_vblocks; 683 684 segno = GET_SEGNO(sbi, blkaddr); 685 686 se = get_seg_entry(sbi, segno); 687 new_vblocks = se->valid_blocks + del; 688 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 689 690 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || 691 (new_vblocks > sbi->blocks_per_seg))); 692 693 se->valid_blocks = new_vblocks; 694 se->mtime = get_mtime(sbi); 695 SIT_I(sbi)->max_mtime = se->mtime; 696 697 /* Update valid block bitmap */ 698 if (del > 0) { 699 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) 700 f2fs_bug_on(sbi, 1); 701 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 702 sbi->discard_blks--; 703 } else { 704 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) 705 f2fs_bug_on(sbi, 1); 706 if (f2fs_test_and_clear_bit(offset, se->discard_map)) 707 sbi->discard_blks++; 708 } 709 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 710 se->ckpt_valid_blocks += del; 711 712 __mark_sit_entry_dirty(sbi, segno); 713 714 /* update total number of valid blocks to be written in ckpt area */ 715 SIT_I(sbi)->written_valid_blocks += del; 716 717 if (sbi->segs_per_sec > 1) 718 get_sec_entry(sbi, segno)->valid_blocks += del; 719 } 720 721 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) 722 { 723 update_sit_entry(sbi, new, 1); 724 if (GET_SEGNO(sbi, old) != NULL_SEGNO) 725 update_sit_entry(sbi, old, -1); 726 727 locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); 728 locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); 729 } 730 731 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 732 { 733 unsigned int segno = GET_SEGNO(sbi, addr); 734 struct sit_info *sit_i = SIT_I(sbi); 735 736 f2fs_bug_on(sbi, addr == NULL_ADDR); 737 if (addr == NEW_ADDR) 738 return; 739 740 /* add it into sit main buffer */ 741 mutex_lock(&sit_i->sentry_lock); 742 743 update_sit_entry(sbi, addr, -1); 744 745 /* add it into dirty seglist */ 746 locate_dirty_segment(sbi, segno); 747 748 mutex_unlock(&sit_i->sentry_lock); 749 } 750 751 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 752 { 753 struct sit_info *sit_i = SIT_I(sbi); 754 unsigned int segno, offset; 755 struct seg_entry *se; 756 bool is_cp = false; 757 758 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) 759 return true; 760 761 mutex_lock(&sit_i->sentry_lock); 762 763 segno = GET_SEGNO(sbi, blkaddr); 764 se = get_seg_entry(sbi, segno); 765 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 766 767 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 768 is_cp = true; 769 770 mutex_unlock(&sit_i->sentry_lock); 771 772 return is_cp; 773 } 774 775 /* 776 * This function should be resided under the curseg_mutex lock 777 */ 778 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 779 struct f2fs_summary *sum) 780 { 781 struct curseg_info *curseg = CURSEG_I(sbi, type); 782 void *addr = curseg->sum_blk; 783 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 784 memcpy(addr, sum, sizeof(struct f2fs_summary)); 785 } 786 787 /* 788 * Calculate the number of current summary pages for writing 789 */ 790 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 791 { 792 int valid_sum_count = 0; 793 int i, sum_in_page; 794 795 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 796 if (sbi->ckpt->alloc_type[i] == SSR) 797 valid_sum_count += sbi->blocks_per_seg; 798 else { 799 if (for_ra) 800 valid_sum_count += le16_to_cpu( 801 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 802 else 803 valid_sum_count += curseg_blkoff(sbi, i); 804 } 805 } 806 807 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE - 808 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 809 if (valid_sum_count <= sum_in_page) 810 return 1; 811 else if ((valid_sum_count - sum_in_page) <= 812 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 813 return 2; 814 return 3; 815 } 816 817 /* 818 * Caller should put this summary page 819 */ 820 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 821 { 822 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); 823 } 824 825 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr) 826 { 827 struct page *page = grab_meta_page(sbi, blk_addr); 828 void *dst = page_address(page); 829 830 if (src) 831 memcpy(dst, src, PAGE_CACHE_SIZE); 832 else 833 memset(dst, 0, PAGE_CACHE_SIZE); 834 set_page_dirty(page); 835 f2fs_put_page(page, 1); 836 } 837 838 static void write_sum_page(struct f2fs_sb_info *sbi, 839 struct f2fs_summary_block *sum_blk, block_t blk_addr) 840 { 841 update_meta_page(sbi, (void *)sum_blk, blk_addr); 842 } 843 844 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) 845 { 846 struct curseg_info *curseg = CURSEG_I(sbi, type); 847 unsigned int segno = curseg->segno + 1; 848 struct free_segmap_info *free_i = FREE_I(sbi); 849 850 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 851 return !test_bit(segno, free_i->free_segmap); 852 return 0; 853 } 854 855 /* 856 * Find a new segment from the free segments bitmap to right order 857 * This function should be returned with success, otherwise BUG 858 */ 859 static void get_new_segment(struct f2fs_sb_info *sbi, 860 unsigned int *newseg, bool new_sec, int dir) 861 { 862 struct free_segmap_info *free_i = FREE_I(sbi); 863 unsigned int segno, secno, zoneno; 864 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 865 unsigned int hint = *newseg / sbi->segs_per_sec; 866 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); 867 unsigned int left_start = hint; 868 bool init = true; 869 int go_left = 0; 870 int i; 871 872 spin_lock(&free_i->segmap_lock); 873 874 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 875 segno = find_next_zero_bit(free_i->free_segmap, 876 MAIN_SEGS(sbi), *newseg + 1); 877 if (segno - *newseg < sbi->segs_per_sec - 878 (*newseg % sbi->segs_per_sec)) 879 goto got_it; 880 } 881 find_other_zone: 882 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 883 if (secno >= MAIN_SECS(sbi)) { 884 if (dir == ALLOC_RIGHT) { 885 secno = find_next_zero_bit(free_i->free_secmap, 886 MAIN_SECS(sbi), 0); 887 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 888 } else { 889 go_left = 1; 890 left_start = hint - 1; 891 } 892 } 893 if (go_left == 0) 894 goto skip_left; 895 896 while (test_bit(left_start, free_i->free_secmap)) { 897 if (left_start > 0) { 898 left_start--; 899 continue; 900 } 901 left_start = find_next_zero_bit(free_i->free_secmap, 902 MAIN_SECS(sbi), 0); 903 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 904 break; 905 } 906 secno = left_start; 907 skip_left: 908 hint = secno; 909 segno = secno * sbi->segs_per_sec; 910 zoneno = secno / sbi->secs_per_zone; 911 912 /* give up on finding another zone */ 913 if (!init) 914 goto got_it; 915 if (sbi->secs_per_zone == 1) 916 goto got_it; 917 if (zoneno == old_zoneno) 918 goto got_it; 919 if (dir == ALLOC_LEFT) { 920 if (!go_left && zoneno + 1 >= total_zones) 921 goto got_it; 922 if (go_left && zoneno == 0) 923 goto got_it; 924 } 925 for (i = 0; i < NR_CURSEG_TYPE; i++) 926 if (CURSEG_I(sbi, i)->zone == zoneno) 927 break; 928 929 if (i < NR_CURSEG_TYPE) { 930 /* zone is in user, try another */ 931 if (go_left) 932 hint = zoneno * sbi->secs_per_zone - 1; 933 else if (zoneno + 1 >= total_zones) 934 hint = 0; 935 else 936 hint = (zoneno + 1) * sbi->secs_per_zone; 937 init = false; 938 goto find_other_zone; 939 } 940 got_it: 941 /* set it as dirty segment in free segmap */ 942 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 943 __set_inuse(sbi, segno); 944 *newseg = segno; 945 spin_unlock(&free_i->segmap_lock); 946 } 947 948 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 949 { 950 struct curseg_info *curseg = CURSEG_I(sbi, type); 951 struct summary_footer *sum_footer; 952 953 curseg->segno = curseg->next_segno; 954 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); 955 curseg->next_blkoff = 0; 956 curseg->next_segno = NULL_SEGNO; 957 958 sum_footer = &(curseg->sum_blk->footer); 959 memset(sum_footer, 0, sizeof(struct summary_footer)); 960 if (IS_DATASEG(type)) 961 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 962 if (IS_NODESEG(type)) 963 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 964 __set_sit_entry_type(sbi, type, curseg->segno, modified); 965 } 966 967 /* 968 * Allocate a current working segment. 969 * This function always allocates a free segment in LFS manner. 970 */ 971 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 972 { 973 struct curseg_info *curseg = CURSEG_I(sbi, type); 974 unsigned int segno = curseg->segno; 975 int dir = ALLOC_LEFT; 976 977 write_sum_page(sbi, curseg->sum_blk, 978 GET_SUM_BLOCK(sbi, segno)); 979 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 980 dir = ALLOC_RIGHT; 981 982 if (test_opt(sbi, NOHEAP)) 983 dir = ALLOC_RIGHT; 984 985 get_new_segment(sbi, &segno, new_sec, dir); 986 curseg->next_segno = segno; 987 reset_curseg(sbi, type, 1); 988 curseg->alloc_type = LFS; 989 } 990 991 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 992 struct curseg_info *seg, block_t start) 993 { 994 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 995 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 996 unsigned long *target_map = SIT_I(sbi)->tmp_map; 997 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 998 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 999 int i, pos; 1000 1001 for (i = 0; i < entries; i++) 1002 target_map[i] = ckpt_map[i] | cur_map[i]; 1003 1004 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 1005 1006 seg->next_blkoff = pos; 1007 } 1008 1009 /* 1010 * If a segment is written by LFS manner, next block offset is just obtained 1011 * by increasing the current block offset. However, if a segment is written by 1012 * SSR manner, next block offset obtained by calling __next_free_blkoff 1013 */ 1014 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 1015 struct curseg_info *seg) 1016 { 1017 if (seg->alloc_type == SSR) 1018 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 1019 else 1020 seg->next_blkoff++; 1021 } 1022 1023 /* 1024 * This function always allocates a used segment(from dirty seglist) by SSR 1025 * manner, so it should recover the existing segment information of valid blocks 1026 */ 1027 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) 1028 { 1029 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1030 struct curseg_info *curseg = CURSEG_I(sbi, type); 1031 unsigned int new_segno = curseg->next_segno; 1032 struct f2fs_summary_block *sum_node; 1033 struct page *sum_page; 1034 1035 write_sum_page(sbi, curseg->sum_blk, 1036 GET_SUM_BLOCK(sbi, curseg->segno)); 1037 __set_test_and_inuse(sbi, new_segno); 1038 1039 mutex_lock(&dirty_i->seglist_lock); 1040 __remove_dirty_segment(sbi, new_segno, PRE); 1041 __remove_dirty_segment(sbi, new_segno, DIRTY); 1042 mutex_unlock(&dirty_i->seglist_lock); 1043 1044 reset_curseg(sbi, type, 1); 1045 curseg->alloc_type = SSR; 1046 __next_free_blkoff(sbi, curseg, 0); 1047 1048 if (reuse) { 1049 sum_page = get_sum_page(sbi, new_segno); 1050 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 1051 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 1052 f2fs_put_page(sum_page, 1); 1053 } 1054 } 1055 1056 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 1057 { 1058 struct curseg_info *curseg = CURSEG_I(sbi, type); 1059 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 1060 1061 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0)) 1062 return v_ops->get_victim(sbi, 1063 &(curseg)->next_segno, BG_GC, type, SSR); 1064 1065 /* For data segments, let's do SSR more intensively */ 1066 for (; type >= CURSEG_HOT_DATA; type--) 1067 if (v_ops->get_victim(sbi, &(curseg)->next_segno, 1068 BG_GC, type, SSR)) 1069 return 1; 1070 return 0; 1071 } 1072 1073 /* 1074 * flush out current segment and replace it with new segment 1075 * This function should be returned with success, otherwise BUG 1076 */ 1077 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 1078 int type, bool force) 1079 { 1080 struct curseg_info *curseg = CURSEG_I(sbi, type); 1081 1082 if (force) 1083 new_curseg(sbi, type, true); 1084 else if (type == CURSEG_WARM_NODE) 1085 new_curseg(sbi, type, false); 1086 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) 1087 new_curseg(sbi, type, false); 1088 else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) 1089 change_curseg(sbi, type, true); 1090 else 1091 new_curseg(sbi, type, false); 1092 1093 stat_inc_seg_type(sbi, curseg); 1094 } 1095 1096 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type) 1097 { 1098 struct curseg_info *curseg = CURSEG_I(sbi, type); 1099 unsigned int old_segno; 1100 1101 old_segno = curseg->segno; 1102 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true); 1103 locate_dirty_segment(sbi, old_segno); 1104 } 1105 1106 void allocate_new_segments(struct f2fs_sb_info *sbi) 1107 { 1108 int i; 1109 1110 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 1111 __allocate_new_segments(sbi, i); 1112 } 1113 1114 static const struct segment_allocation default_salloc_ops = { 1115 .allocate_segment = allocate_segment_by_default, 1116 }; 1117 1118 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 1119 { 1120 __u64 start = F2FS_BYTES_TO_BLK(range->start); 1121 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 1122 unsigned int start_segno, end_segno; 1123 struct cp_control cpc; 1124 int err = 0; 1125 1126 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 1127 return -EINVAL; 1128 1129 cpc.trimmed = 0; 1130 if (end <= MAIN_BLKADDR(sbi)) 1131 goto out; 1132 1133 /* start/end segment number in main_area */ 1134 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 1135 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 1136 GET_SEGNO(sbi, end); 1137 cpc.reason = CP_DISCARD; 1138 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 1139 1140 /* do checkpoint to issue discard commands safely */ 1141 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) { 1142 cpc.trim_start = start_segno; 1143 1144 if (sbi->discard_blks == 0) 1145 break; 1146 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi)) 1147 cpc.trim_end = end_segno; 1148 else 1149 cpc.trim_end = min_t(unsigned int, 1150 rounddown(start_segno + 1151 BATCHED_TRIM_SEGMENTS(sbi), 1152 sbi->segs_per_sec) - 1, end_segno); 1153 1154 mutex_lock(&sbi->gc_mutex); 1155 err = write_checkpoint(sbi, &cpc); 1156 mutex_unlock(&sbi->gc_mutex); 1157 } 1158 out: 1159 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed); 1160 return err; 1161 } 1162 1163 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 1164 { 1165 struct curseg_info *curseg = CURSEG_I(sbi, type); 1166 if (curseg->next_blkoff < sbi->blocks_per_seg) 1167 return true; 1168 return false; 1169 } 1170 1171 static int __get_segment_type_2(struct page *page, enum page_type p_type) 1172 { 1173 if (p_type == DATA) 1174 return CURSEG_HOT_DATA; 1175 else 1176 return CURSEG_HOT_NODE; 1177 } 1178 1179 static int __get_segment_type_4(struct page *page, enum page_type p_type) 1180 { 1181 if (p_type == DATA) { 1182 struct inode *inode = page->mapping->host; 1183 1184 if (S_ISDIR(inode->i_mode)) 1185 return CURSEG_HOT_DATA; 1186 else 1187 return CURSEG_COLD_DATA; 1188 } else { 1189 if (IS_DNODE(page) && is_cold_node(page)) 1190 return CURSEG_WARM_NODE; 1191 else 1192 return CURSEG_COLD_NODE; 1193 } 1194 } 1195 1196 static int __get_segment_type_6(struct page *page, enum page_type p_type) 1197 { 1198 if (p_type == DATA) { 1199 struct inode *inode = page->mapping->host; 1200 1201 if (S_ISDIR(inode->i_mode)) 1202 return CURSEG_HOT_DATA; 1203 else if (is_cold_data(page) || file_is_cold(inode)) 1204 return CURSEG_COLD_DATA; 1205 else 1206 return CURSEG_WARM_DATA; 1207 } else { 1208 if (IS_DNODE(page)) 1209 return is_cold_node(page) ? CURSEG_WARM_NODE : 1210 CURSEG_HOT_NODE; 1211 else 1212 return CURSEG_COLD_NODE; 1213 } 1214 } 1215 1216 static int __get_segment_type(struct page *page, enum page_type p_type) 1217 { 1218 switch (F2FS_P_SB(page)->active_logs) { 1219 case 2: 1220 return __get_segment_type_2(page, p_type); 1221 case 4: 1222 return __get_segment_type_4(page, p_type); 1223 } 1224 /* NR_CURSEG_TYPE(6) logs by default */ 1225 f2fs_bug_on(F2FS_P_SB(page), 1226 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE); 1227 return __get_segment_type_6(page, p_type); 1228 } 1229 1230 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 1231 block_t old_blkaddr, block_t *new_blkaddr, 1232 struct f2fs_summary *sum, int type) 1233 { 1234 struct sit_info *sit_i = SIT_I(sbi); 1235 struct curseg_info *curseg; 1236 bool direct_io = (type == CURSEG_DIRECT_IO); 1237 1238 type = direct_io ? CURSEG_WARM_DATA : type; 1239 1240 curseg = CURSEG_I(sbi, type); 1241 1242 mutex_lock(&curseg->curseg_mutex); 1243 mutex_lock(&sit_i->sentry_lock); 1244 1245 /* direct_io'ed data is aligned to the segment for better performance */ 1246 if (direct_io && curseg->next_blkoff && 1247 !has_not_enough_free_secs(sbi, 0)) 1248 __allocate_new_segments(sbi, type); 1249 1250 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 1251 1252 /* 1253 * __add_sum_entry should be resided under the curseg_mutex 1254 * because, this function updates a summary entry in the 1255 * current summary block. 1256 */ 1257 __add_sum_entry(sbi, type, sum); 1258 1259 __refresh_next_blkoff(sbi, curseg); 1260 1261 stat_inc_block_count(sbi, curseg); 1262 1263 if (!__has_curseg_space(sbi, type)) 1264 sit_i->s_ops->allocate_segment(sbi, type, false); 1265 /* 1266 * SIT information should be updated before segment allocation, 1267 * since SSR needs latest valid block information. 1268 */ 1269 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); 1270 1271 mutex_unlock(&sit_i->sentry_lock); 1272 1273 if (page && IS_NODESEG(type)) 1274 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 1275 1276 mutex_unlock(&curseg->curseg_mutex); 1277 } 1278 1279 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 1280 { 1281 int type = __get_segment_type(fio->page, fio->type); 1282 1283 allocate_data_block(fio->sbi, fio->page, fio->blk_addr, 1284 &fio->blk_addr, sum, type); 1285 1286 /* writeout dirty page into bdev */ 1287 f2fs_submit_page_mbio(fio); 1288 } 1289 1290 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) 1291 { 1292 struct f2fs_io_info fio = { 1293 .sbi = sbi, 1294 .type = META, 1295 .rw = WRITE_SYNC | REQ_META | REQ_PRIO, 1296 .blk_addr = page->index, 1297 .page = page, 1298 .encrypted_page = NULL, 1299 }; 1300 1301 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 1302 fio.rw &= ~REQ_META; 1303 1304 set_page_writeback(page); 1305 f2fs_submit_page_mbio(&fio); 1306 } 1307 1308 void write_node_page(unsigned int nid, struct f2fs_io_info *fio) 1309 { 1310 struct f2fs_summary sum; 1311 1312 set_summary(&sum, nid, 0, 0); 1313 do_write_page(&sum, fio); 1314 } 1315 1316 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio) 1317 { 1318 struct f2fs_sb_info *sbi = fio->sbi; 1319 struct f2fs_summary sum; 1320 struct node_info ni; 1321 1322 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 1323 get_node_info(sbi, dn->nid, &ni); 1324 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 1325 do_write_page(&sum, fio); 1326 dn->data_blkaddr = fio->blk_addr; 1327 } 1328 1329 void rewrite_data_page(struct f2fs_io_info *fio) 1330 { 1331 stat_inc_inplace_blocks(fio->sbi); 1332 f2fs_submit_page_mbio(fio); 1333 } 1334 1335 static void __f2fs_replace_block(struct f2fs_sb_info *sbi, 1336 struct f2fs_summary *sum, 1337 block_t old_blkaddr, block_t new_blkaddr, 1338 bool recover_curseg) 1339 { 1340 struct sit_info *sit_i = SIT_I(sbi); 1341 struct curseg_info *curseg; 1342 unsigned int segno, old_cursegno; 1343 struct seg_entry *se; 1344 int type; 1345 unsigned short old_blkoff; 1346 1347 segno = GET_SEGNO(sbi, new_blkaddr); 1348 se = get_seg_entry(sbi, segno); 1349 type = se->type; 1350 1351 if (!recover_curseg) { 1352 /* for recovery flow */ 1353 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 1354 if (old_blkaddr == NULL_ADDR) 1355 type = CURSEG_COLD_DATA; 1356 else 1357 type = CURSEG_WARM_DATA; 1358 } 1359 } else { 1360 if (!IS_CURSEG(sbi, segno)) 1361 type = CURSEG_WARM_DATA; 1362 } 1363 1364 curseg = CURSEG_I(sbi, type); 1365 1366 mutex_lock(&curseg->curseg_mutex); 1367 mutex_lock(&sit_i->sentry_lock); 1368 1369 old_cursegno = curseg->segno; 1370 old_blkoff = curseg->next_blkoff; 1371 1372 /* change the current segment */ 1373 if (segno != curseg->segno) { 1374 curseg->next_segno = segno; 1375 change_curseg(sbi, type, true); 1376 } 1377 1378 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 1379 __add_sum_entry(sbi, type, sum); 1380 1381 if (!recover_curseg) 1382 update_sit_entry(sbi, new_blkaddr, 1); 1383 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 1384 update_sit_entry(sbi, old_blkaddr, -1); 1385 1386 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 1387 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 1388 1389 locate_dirty_segment(sbi, old_cursegno); 1390 1391 if (recover_curseg) { 1392 if (old_cursegno != curseg->segno) { 1393 curseg->next_segno = old_cursegno; 1394 change_curseg(sbi, type, true); 1395 } 1396 curseg->next_blkoff = old_blkoff; 1397 } 1398 1399 mutex_unlock(&sit_i->sentry_lock); 1400 mutex_unlock(&curseg->curseg_mutex); 1401 } 1402 1403 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 1404 block_t old_addr, block_t new_addr, 1405 unsigned char version, bool recover_curseg) 1406 { 1407 struct f2fs_summary sum; 1408 1409 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 1410 1411 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg); 1412 1413 dn->data_blkaddr = new_addr; 1414 set_data_blkaddr(dn); 1415 f2fs_update_extent_cache(dn); 1416 } 1417 1418 static inline bool is_merged_page(struct f2fs_sb_info *sbi, 1419 struct page *page, enum page_type type) 1420 { 1421 enum page_type btype = PAGE_TYPE_OF_BIO(type); 1422 struct f2fs_bio_info *io = &sbi->write_io[btype]; 1423 struct bio_vec *bvec; 1424 struct page *target; 1425 int i; 1426 1427 down_read(&io->io_rwsem); 1428 if (!io->bio) { 1429 up_read(&io->io_rwsem); 1430 return false; 1431 } 1432 1433 bio_for_each_segment_all(bvec, io->bio, i) { 1434 1435 if (bvec->bv_page->mapping) { 1436 target = bvec->bv_page; 1437 } else { 1438 struct f2fs_crypto_ctx *ctx; 1439 1440 /* encrypted page */ 1441 ctx = (struct f2fs_crypto_ctx *)page_private( 1442 bvec->bv_page); 1443 target = ctx->w.control_page; 1444 } 1445 1446 if (page == target) { 1447 up_read(&io->io_rwsem); 1448 return true; 1449 } 1450 } 1451 1452 up_read(&io->io_rwsem); 1453 return false; 1454 } 1455 1456 void f2fs_wait_on_page_writeback(struct page *page, 1457 enum page_type type) 1458 { 1459 if (PageWriteback(page)) { 1460 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1461 1462 if (is_merged_page(sbi, page, type)) 1463 f2fs_submit_merged_bio(sbi, type, WRITE); 1464 wait_on_page_writeback(page); 1465 } 1466 } 1467 1468 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi, 1469 block_t blkaddr) 1470 { 1471 struct page *cpage; 1472 1473 if (blkaddr == NEW_ADDR) 1474 return; 1475 1476 f2fs_bug_on(sbi, blkaddr == NULL_ADDR); 1477 1478 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 1479 if (cpage) { 1480 f2fs_wait_on_page_writeback(cpage, DATA); 1481 f2fs_put_page(cpage, 1); 1482 } 1483 } 1484 1485 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 1486 { 1487 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1488 struct curseg_info *seg_i; 1489 unsigned char *kaddr; 1490 struct page *page; 1491 block_t start; 1492 int i, j, offset; 1493 1494 start = start_sum_block(sbi); 1495 1496 page = get_meta_page(sbi, start++); 1497 kaddr = (unsigned char *)page_address(page); 1498 1499 /* Step 1: restore nat cache */ 1500 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 1501 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE); 1502 1503 /* Step 2: restore sit cache */ 1504 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 1505 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE, 1506 SUM_JOURNAL_SIZE); 1507 offset = 2 * SUM_JOURNAL_SIZE; 1508 1509 /* Step 3: restore summary entries */ 1510 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1511 unsigned short blk_off; 1512 unsigned int segno; 1513 1514 seg_i = CURSEG_I(sbi, i); 1515 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 1516 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 1517 seg_i->next_segno = segno; 1518 reset_curseg(sbi, i, 0); 1519 seg_i->alloc_type = ckpt->alloc_type[i]; 1520 seg_i->next_blkoff = blk_off; 1521 1522 if (seg_i->alloc_type == SSR) 1523 blk_off = sbi->blocks_per_seg; 1524 1525 for (j = 0; j < blk_off; j++) { 1526 struct f2fs_summary *s; 1527 s = (struct f2fs_summary *)(kaddr + offset); 1528 seg_i->sum_blk->entries[j] = *s; 1529 offset += SUMMARY_SIZE; 1530 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE - 1531 SUM_FOOTER_SIZE) 1532 continue; 1533 1534 f2fs_put_page(page, 1); 1535 page = NULL; 1536 1537 page = get_meta_page(sbi, start++); 1538 kaddr = (unsigned char *)page_address(page); 1539 offset = 0; 1540 } 1541 } 1542 f2fs_put_page(page, 1); 1543 return 0; 1544 } 1545 1546 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 1547 { 1548 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1549 struct f2fs_summary_block *sum; 1550 struct curseg_info *curseg; 1551 struct page *new; 1552 unsigned short blk_off; 1553 unsigned int segno = 0; 1554 block_t blk_addr = 0; 1555 1556 /* get segment number and block addr */ 1557 if (IS_DATASEG(type)) { 1558 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 1559 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 1560 CURSEG_HOT_DATA]); 1561 if (__exist_node_summaries(sbi)) 1562 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 1563 else 1564 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 1565 } else { 1566 segno = le32_to_cpu(ckpt->cur_node_segno[type - 1567 CURSEG_HOT_NODE]); 1568 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 1569 CURSEG_HOT_NODE]); 1570 if (__exist_node_summaries(sbi)) 1571 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 1572 type - CURSEG_HOT_NODE); 1573 else 1574 blk_addr = GET_SUM_BLOCK(sbi, segno); 1575 } 1576 1577 new = get_meta_page(sbi, blk_addr); 1578 sum = (struct f2fs_summary_block *)page_address(new); 1579 1580 if (IS_NODESEG(type)) { 1581 if (__exist_node_summaries(sbi)) { 1582 struct f2fs_summary *ns = &sum->entries[0]; 1583 int i; 1584 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 1585 ns->version = 0; 1586 ns->ofs_in_node = 0; 1587 } 1588 } else { 1589 int err; 1590 1591 err = restore_node_summary(sbi, segno, sum); 1592 if (err) { 1593 f2fs_put_page(new, 1); 1594 return err; 1595 } 1596 } 1597 } 1598 1599 /* set uncompleted segment to curseg */ 1600 curseg = CURSEG_I(sbi, type); 1601 mutex_lock(&curseg->curseg_mutex); 1602 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE); 1603 curseg->next_segno = segno; 1604 reset_curseg(sbi, type, 0); 1605 curseg->alloc_type = ckpt->alloc_type[type]; 1606 curseg->next_blkoff = blk_off; 1607 mutex_unlock(&curseg->curseg_mutex); 1608 f2fs_put_page(new, 1); 1609 return 0; 1610 } 1611 1612 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 1613 { 1614 int type = CURSEG_HOT_DATA; 1615 int err; 1616 1617 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) { 1618 int npages = npages_for_summary_flush(sbi, true); 1619 1620 if (npages >= 2) 1621 ra_meta_pages(sbi, start_sum_block(sbi), npages, 1622 META_CP, true); 1623 1624 /* restore for compacted data summary */ 1625 if (read_compacted_summaries(sbi)) 1626 return -EINVAL; 1627 type = CURSEG_HOT_NODE; 1628 } 1629 1630 if (__exist_node_summaries(sbi)) 1631 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), 1632 NR_CURSEG_TYPE - type, META_CP, true); 1633 1634 for (; type <= CURSEG_COLD_NODE; type++) { 1635 err = read_normal_summaries(sbi, type); 1636 if (err) 1637 return err; 1638 } 1639 1640 return 0; 1641 } 1642 1643 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 1644 { 1645 struct page *page; 1646 unsigned char *kaddr; 1647 struct f2fs_summary *summary; 1648 struct curseg_info *seg_i; 1649 int written_size = 0; 1650 int i, j; 1651 1652 page = grab_meta_page(sbi, blkaddr++); 1653 kaddr = (unsigned char *)page_address(page); 1654 1655 /* Step 1: write nat cache */ 1656 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 1657 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE); 1658 written_size += SUM_JOURNAL_SIZE; 1659 1660 /* Step 2: write sit cache */ 1661 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 1662 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits, 1663 SUM_JOURNAL_SIZE); 1664 written_size += SUM_JOURNAL_SIZE; 1665 1666 /* Step 3: write summary entries */ 1667 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1668 unsigned short blkoff; 1669 seg_i = CURSEG_I(sbi, i); 1670 if (sbi->ckpt->alloc_type[i] == SSR) 1671 blkoff = sbi->blocks_per_seg; 1672 else 1673 blkoff = curseg_blkoff(sbi, i); 1674 1675 for (j = 0; j < blkoff; j++) { 1676 if (!page) { 1677 page = grab_meta_page(sbi, blkaddr++); 1678 kaddr = (unsigned char *)page_address(page); 1679 written_size = 0; 1680 } 1681 summary = (struct f2fs_summary *)(kaddr + written_size); 1682 *summary = seg_i->sum_blk->entries[j]; 1683 written_size += SUMMARY_SIZE; 1684 1685 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE - 1686 SUM_FOOTER_SIZE) 1687 continue; 1688 1689 set_page_dirty(page); 1690 f2fs_put_page(page, 1); 1691 page = NULL; 1692 } 1693 } 1694 if (page) { 1695 set_page_dirty(page); 1696 f2fs_put_page(page, 1); 1697 } 1698 } 1699 1700 static void write_normal_summaries(struct f2fs_sb_info *sbi, 1701 block_t blkaddr, int type) 1702 { 1703 int i, end; 1704 if (IS_DATASEG(type)) 1705 end = type + NR_CURSEG_DATA_TYPE; 1706 else 1707 end = type + NR_CURSEG_NODE_TYPE; 1708 1709 for (i = type; i < end; i++) { 1710 struct curseg_info *sum = CURSEG_I(sbi, i); 1711 mutex_lock(&sum->curseg_mutex); 1712 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type)); 1713 mutex_unlock(&sum->curseg_mutex); 1714 } 1715 } 1716 1717 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 1718 { 1719 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) 1720 write_compacted_summaries(sbi, start_blk); 1721 else 1722 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 1723 } 1724 1725 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 1726 { 1727 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 1728 } 1729 1730 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type, 1731 unsigned int val, int alloc) 1732 { 1733 int i; 1734 1735 if (type == NAT_JOURNAL) { 1736 for (i = 0; i < nats_in_cursum(sum); i++) { 1737 if (le32_to_cpu(nid_in_journal(sum, i)) == val) 1738 return i; 1739 } 1740 if (alloc && __has_cursum_space(sum, 1, NAT_JOURNAL)) 1741 return update_nats_in_cursum(sum, 1); 1742 } else if (type == SIT_JOURNAL) { 1743 for (i = 0; i < sits_in_cursum(sum); i++) 1744 if (le32_to_cpu(segno_in_journal(sum, i)) == val) 1745 return i; 1746 if (alloc && __has_cursum_space(sum, 1, SIT_JOURNAL)) 1747 return update_sits_in_cursum(sum, 1); 1748 } 1749 return -1; 1750 } 1751 1752 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 1753 unsigned int segno) 1754 { 1755 return get_meta_page(sbi, current_sit_addr(sbi, segno)); 1756 } 1757 1758 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 1759 unsigned int start) 1760 { 1761 struct sit_info *sit_i = SIT_I(sbi); 1762 struct page *src_page, *dst_page; 1763 pgoff_t src_off, dst_off; 1764 void *src_addr, *dst_addr; 1765 1766 src_off = current_sit_addr(sbi, start); 1767 dst_off = next_sit_addr(sbi, src_off); 1768 1769 /* get current sit block page without lock */ 1770 src_page = get_meta_page(sbi, src_off); 1771 dst_page = grab_meta_page(sbi, dst_off); 1772 f2fs_bug_on(sbi, PageDirty(src_page)); 1773 1774 src_addr = page_address(src_page); 1775 dst_addr = page_address(dst_page); 1776 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); 1777 1778 set_page_dirty(dst_page); 1779 f2fs_put_page(src_page, 1); 1780 1781 set_to_next_sit(sit_i, start); 1782 1783 return dst_page; 1784 } 1785 1786 static struct sit_entry_set *grab_sit_entry_set(void) 1787 { 1788 struct sit_entry_set *ses = 1789 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); 1790 1791 ses->entry_cnt = 0; 1792 INIT_LIST_HEAD(&ses->set_list); 1793 return ses; 1794 } 1795 1796 static void release_sit_entry_set(struct sit_entry_set *ses) 1797 { 1798 list_del(&ses->set_list); 1799 kmem_cache_free(sit_entry_set_slab, ses); 1800 } 1801 1802 static void adjust_sit_entry_set(struct sit_entry_set *ses, 1803 struct list_head *head) 1804 { 1805 struct sit_entry_set *next = ses; 1806 1807 if (list_is_last(&ses->set_list, head)) 1808 return; 1809 1810 list_for_each_entry_continue(next, head, set_list) 1811 if (ses->entry_cnt <= next->entry_cnt) 1812 break; 1813 1814 list_move_tail(&ses->set_list, &next->set_list); 1815 } 1816 1817 static void add_sit_entry(unsigned int segno, struct list_head *head) 1818 { 1819 struct sit_entry_set *ses; 1820 unsigned int start_segno = START_SEGNO(segno); 1821 1822 list_for_each_entry(ses, head, set_list) { 1823 if (ses->start_segno == start_segno) { 1824 ses->entry_cnt++; 1825 adjust_sit_entry_set(ses, head); 1826 return; 1827 } 1828 } 1829 1830 ses = grab_sit_entry_set(); 1831 1832 ses->start_segno = start_segno; 1833 ses->entry_cnt++; 1834 list_add(&ses->set_list, head); 1835 } 1836 1837 static void add_sits_in_set(struct f2fs_sb_info *sbi) 1838 { 1839 struct f2fs_sm_info *sm_info = SM_I(sbi); 1840 struct list_head *set_list = &sm_info->sit_entry_set; 1841 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 1842 unsigned int segno; 1843 1844 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 1845 add_sit_entry(segno, set_list); 1846 } 1847 1848 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 1849 { 1850 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1851 struct f2fs_summary_block *sum = curseg->sum_blk; 1852 int i; 1853 1854 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) { 1855 unsigned int segno; 1856 bool dirtied; 1857 1858 segno = le32_to_cpu(segno_in_journal(sum, i)); 1859 dirtied = __mark_sit_entry_dirty(sbi, segno); 1860 1861 if (!dirtied) 1862 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 1863 } 1864 update_sits_in_cursum(sum, -sits_in_cursum(sum)); 1865 } 1866 1867 /* 1868 * CP calls this function, which flushes SIT entries including sit_journal, 1869 * and moves prefree segs to free segs. 1870 */ 1871 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1872 { 1873 struct sit_info *sit_i = SIT_I(sbi); 1874 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 1875 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1876 struct f2fs_summary_block *sum = curseg->sum_blk; 1877 struct sit_entry_set *ses, *tmp; 1878 struct list_head *head = &SM_I(sbi)->sit_entry_set; 1879 bool to_journal = true; 1880 struct seg_entry *se; 1881 1882 mutex_lock(&curseg->curseg_mutex); 1883 mutex_lock(&sit_i->sentry_lock); 1884 1885 if (!sit_i->dirty_sentries) 1886 goto out; 1887 1888 /* 1889 * add and account sit entries of dirty bitmap in sit entry 1890 * set temporarily 1891 */ 1892 add_sits_in_set(sbi); 1893 1894 /* 1895 * if there are no enough space in journal to store dirty sit 1896 * entries, remove all entries from journal and add and account 1897 * them in sit entry set. 1898 */ 1899 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL)) 1900 remove_sits_in_journal(sbi); 1901 1902 /* 1903 * there are two steps to flush sit entries: 1904 * #1, flush sit entries to journal in current cold data summary block. 1905 * #2, flush sit entries to sit page. 1906 */ 1907 list_for_each_entry_safe(ses, tmp, head, set_list) { 1908 struct page *page = NULL; 1909 struct f2fs_sit_block *raw_sit = NULL; 1910 unsigned int start_segno = ses->start_segno; 1911 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 1912 (unsigned long)MAIN_SEGS(sbi)); 1913 unsigned int segno = start_segno; 1914 1915 if (to_journal && 1916 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL)) 1917 to_journal = false; 1918 1919 if (!to_journal) { 1920 page = get_next_sit_page(sbi, start_segno); 1921 raw_sit = page_address(page); 1922 } 1923 1924 /* flush dirty sit entries in region of current sit set */ 1925 for_each_set_bit_from(segno, bitmap, end) { 1926 int offset, sit_offset; 1927 1928 se = get_seg_entry(sbi, segno); 1929 1930 /* add discard candidates */ 1931 if (cpc->reason != CP_DISCARD) { 1932 cpc->trim_start = segno; 1933 add_discard_addrs(sbi, cpc); 1934 } 1935 1936 if (to_journal) { 1937 offset = lookup_journal_in_cursum(sum, 1938 SIT_JOURNAL, segno, 1); 1939 f2fs_bug_on(sbi, offset < 0); 1940 segno_in_journal(sum, offset) = 1941 cpu_to_le32(segno); 1942 seg_info_to_raw_sit(se, 1943 &sit_in_journal(sum, offset)); 1944 } else { 1945 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 1946 seg_info_to_raw_sit(se, 1947 &raw_sit->entries[sit_offset]); 1948 } 1949 1950 __clear_bit(segno, bitmap); 1951 sit_i->dirty_sentries--; 1952 ses->entry_cnt--; 1953 } 1954 1955 if (!to_journal) 1956 f2fs_put_page(page, 1); 1957 1958 f2fs_bug_on(sbi, ses->entry_cnt); 1959 release_sit_entry_set(ses); 1960 } 1961 1962 f2fs_bug_on(sbi, !list_empty(head)); 1963 f2fs_bug_on(sbi, sit_i->dirty_sentries); 1964 out: 1965 if (cpc->reason == CP_DISCARD) { 1966 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 1967 add_discard_addrs(sbi, cpc); 1968 } 1969 mutex_unlock(&sit_i->sentry_lock); 1970 mutex_unlock(&curseg->curseg_mutex); 1971 1972 set_prefree_as_free_segments(sbi); 1973 } 1974 1975 static int build_sit_info(struct f2fs_sb_info *sbi) 1976 { 1977 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 1978 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1979 struct sit_info *sit_i; 1980 unsigned int sit_segs, start; 1981 char *src_bitmap, *dst_bitmap; 1982 unsigned int bitmap_size; 1983 1984 /* allocate memory for SIT information */ 1985 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); 1986 if (!sit_i) 1987 return -ENOMEM; 1988 1989 SM_I(sbi)->sit_info = sit_i; 1990 1991 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) * 1992 sizeof(struct seg_entry), GFP_KERNEL); 1993 if (!sit_i->sentries) 1994 return -ENOMEM; 1995 1996 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 1997 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 1998 if (!sit_i->dirty_sentries_bitmap) 1999 return -ENOMEM; 2000 2001 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2002 sit_i->sentries[start].cur_valid_map 2003 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2004 sit_i->sentries[start].ckpt_valid_map 2005 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2006 sit_i->sentries[start].discard_map 2007 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2008 if (!sit_i->sentries[start].cur_valid_map || 2009 !sit_i->sentries[start].ckpt_valid_map || 2010 !sit_i->sentries[start].discard_map) 2011 return -ENOMEM; 2012 } 2013 2014 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2015 if (!sit_i->tmp_map) 2016 return -ENOMEM; 2017 2018 if (sbi->segs_per_sec > 1) { 2019 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) * 2020 sizeof(struct sec_entry), GFP_KERNEL); 2021 if (!sit_i->sec_entries) 2022 return -ENOMEM; 2023 } 2024 2025 /* get information related with SIT */ 2026 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 2027 2028 /* setup SIT bitmap from ckeckpoint pack */ 2029 bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 2030 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 2031 2032 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 2033 if (!dst_bitmap) 2034 return -ENOMEM; 2035 2036 /* init SIT information */ 2037 sit_i->s_ops = &default_salloc_ops; 2038 2039 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 2040 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 2041 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count); 2042 sit_i->sit_bitmap = dst_bitmap; 2043 sit_i->bitmap_size = bitmap_size; 2044 sit_i->dirty_sentries = 0; 2045 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 2046 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 2047 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; 2048 mutex_init(&sit_i->sentry_lock); 2049 return 0; 2050 } 2051 2052 static int build_free_segmap(struct f2fs_sb_info *sbi) 2053 { 2054 struct free_segmap_info *free_i; 2055 unsigned int bitmap_size, sec_bitmap_size; 2056 2057 /* allocate memory for free segmap information */ 2058 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); 2059 if (!free_i) 2060 return -ENOMEM; 2061 2062 SM_I(sbi)->free_info = free_i; 2063 2064 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 2065 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL); 2066 if (!free_i->free_segmap) 2067 return -ENOMEM; 2068 2069 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 2070 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL); 2071 if (!free_i->free_secmap) 2072 return -ENOMEM; 2073 2074 /* set all segments as dirty temporarily */ 2075 memset(free_i->free_segmap, 0xff, bitmap_size); 2076 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 2077 2078 /* init free segmap information */ 2079 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 2080 free_i->free_segments = 0; 2081 free_i->free_sections = 0; 2082 spin_lock_init(&free_i->segmap_lock); 2083 return 0; 2084 } 2085 2086 static int build_curseg(struct f2fs_sb_info *sbi) 2087 { 2088 struct curseg_info *array; 2089 int i; 2090 2091 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL); 2092 if (!array) 2093 return -ENOMEM; 2094 2095 SM_I(sbi)->curseg_array = array; 2096 2097 for (i = 0; i < NR_CURSEG_TYPE; i++) { 2098 mutex_init(&array[i].curseg_mutex); 2099 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL); 2100 if (!array[i].sum_blk) 2101 return -ENOMEM; 2102 array[i].segno = NULL_SEGNO; 2103 array[i].next_blkoff = 0; 2104 } 2105 return restore_curseg_summaries(sbi); 2106 } 2107 2108 static void build_sit_entries(struct f2fs_sb_info *sbi) 2109 { 2110 struct sit_info *sit_i = SIT_I(sbi); 2111 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 2112 struct f2fs_summary_block *sum = curseg->sum_blk; 2113 int sit_blk_cnt = SIT_BLK_CNT(sbi); 2114 unsigned int i, start, end; 2115 unsigned int readed, start_blk = 0; 2116 int nrpages = MAX_BIO_BLOCKS(sbi); 2117 2118 do { 2119 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true); 2120 2121 start = start_blk * sit_i->sents_per_block; 2122 end = (start_blk + readed) * sit_i->sents_per_block; 2123 2124 for (; start < end && start < MAIN_SEGS(sbi); start++) { 2125 struct seg_entry *se = &sit_i->sentries[start]; 2126 struct f2fs_sit_block *sit_blk; 2127 struct f2fs_sit_entry sit; 2128 struct page *page; 2129 2130 mutex_lock(&curseg->curseg_mutex); 2131 for (i = 0; i < sits_in_cursum(sum); i++) { 2132 if (le32_to_cpu(segno_in_journal(sum, i)) 2133 == start) { 2134 sit = sit_in_journal(sum, i); 2135 mutex_unlock(&curseg->curseg_mutex); 2136 goto got_it; 2137 } 2138 } 2139 mutex_unlock(&curseg->curseg_mutex); 2140 2141 page = get_current_sit_page(sbi, start); 2142 sit_blk = (struct f2fs_sit_block *)page_address(page); 2143 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 2144 f2fs_put_page(page, 1); 2145 got_it: 2146 check_block_count(sbi, start, &sit); 2147 seg_info_from_raw_sit(se, &sit); 2148 2149 /* build discard map only one time */ 2150 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE); 2151 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks; 2152 2153 if (sbi->segs_per_sec > 1) { 2154 struct sec_entry *e = get_sec_entry(sbi, start); 2155 e->valid_blocks += se->valid_blocks; 2156 } 2157 } 2158 start_blk += readed; 2159 } while (start_blk < sit_blk_cnt); 2160 } 2161 2162 static void init_free_segmap(struct f2fs_sb_info *sbi) 2163 { 2164 unsigned int start; 2165 int type; 2166 2167 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2168 struct seg_entry *sentry = get_seg_entry(sbi, start); 2169 if (!sentry->valid_blocks) 2170 __set_free(sbi, start); 2171 } 2172 2173 /* set use the current segments */ 2174 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 2175 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 2176 __set_test_and_inuse(sbi, curseg_t->segno); 2177 } 2178 } 2179 2180 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 2181 { 2182 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2183 struct free_segmap_info *free_i = FREE_I(sbi); 2184 unsigned int segno = 0, offset = 0; 2185 unsigned short valid_blocks; 2186 2187 while (1) { 2188 /* find dirty segment based on free segmap */ 2189 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 2190 if (segno >= MAIN_SEGS(sbi)) 2191 break; 2192 offset = segno + 1; 2193 valid_blocks = get_valid_blocks(sbi, segno, 0); 2194 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) 2195 continue; 2196 if (valid_blocks > sbi->blocks_per_seg) { 2197 f2fs_bug_on(sbi, 1); 2198 continue; 2199 } 2200 mutex_lock(&dirty_i->seglist_lock); 2201 __locate_dirty_segment(sbi, segno, DIRTY); 2202 mutex_unlock(&dirty_i->seglist_lock); 2203 } 2204 } 2205 2206 static int init_victim_secmap(struct f2fs_sb_info *sbi) 2207 { 2208 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2209 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 2210 2211 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 2212 if (!dirty_i->victim_secmap) 2213 return -ENOMEM; 2214 return 0; 2215 } 2216 2217 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 2218 { 2219 struct dirty_seglist_info *dirty_i; 2220 unsigned int bitmap_size, i; 2221 2222 /* allocate memory for dirty segments list information */ 2223 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); 2224 if (!dirty_i) 2225 return -ENOMEM; 2226 2227 SM_I(sbi)->dirty_info = dirty_i; 2228 mutex_init(&dirty_i->seglist_lock); 2229 2230 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 2231 2232 for (i = 0; i < NR_DIRTY_TYPE; i++) { 2233 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 2234 if (!dirty_i->dirty_segmap[i]) 2235 return -ENOMEM; 2236 } 2237 2238 init_dirty_segmap(sbi); 2239 return init_victim_secmap(sbi); 2240 } 2241 2242 /* 2243 * Update min, max modified time for cost-benefit GC algorithm 2244 */ 2245 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 2246 { 2247 struct sit_info *sit_i = SIT_I(sbi); 2248 unsigned int segno; 2249 2250 mutex_lock(&sit_i->sentry_lock); 2251 2252 sit_i->min_mtime = LLONG_MAX; 2253 2254 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 2255 unsigned int i; 2256 unsigned long long mtime = 0; 2257 2258 for (i = 0; i < sbi->segs_per_sec; i++) 2259 mtime += get_seg_entry(sbi, segno + i)->mtime; 2260 2261 mtime = div_u64(mtime, sbi->segs_per_sec); 2262 2263 if (sit_i->min_mtime > mtime) 2264 sit_i->min_mtime = mtime; 2265 } 2266 sit_i->max_mtime = get_mtime(sbi); 2267 mutex_unlock(&sit_i->sentry_lock); 2268 } 2269 2270 int build_segment_manager(struct f2fs_sb_info *sbi) 2271 { 2272 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 2273 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2274 struct f2fs_sm_info *sm_info; 2275 int err; 2276 2277 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); 2278 if (!sm_info) 2279 return -ENOMEM; 2280 2281 /* init sm info */ 2282 sbi->sm_info = sm_info; 2283 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 2284 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 2285 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 2286 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 2287 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 2288 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 2289 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 2290 sm_info->rec_prefree_segments = sm_info->main_segments * 2291 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 2292 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 2293 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 2294 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 2295 2296 INIT_LIST_HEAD(&sm_info->discard_list); 2297 sm_info->nr_discards = 0; 2298 sm_info->max_discards = 0; 2299 2300 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS; 2301 2302 INIT_LIST_HEAD(&sm_info->sit_entry_set); 2303 2304 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) { 2305 err = create_flush_cmd_control(sbi); 2306 if (err) 2307 return err; 2308 } 2309 2310 err = build_sit_info(sbi); 2311 if (err) 2312 return err; 2313 err = build_free_segmap(sbi); 2314 if (err) 2315 return err; 2316 err = build_curseg(sbi); 2317 if (err) 2318 return err; 2319 2320 /* reinit free segmap based on SIT */ 2321 build_sit_entries(sbi); 2322 2323 init_free_segmap(sbi); 2324 err = build_dirty_segmap(sbi); 2325 if (err) 2326 return err; 2327 2328 init_min_max_mtime(sbi); 2329 return 0; 2330 } 2331 2332 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 2333 enum dirty_type dirty_type) 2334 { 2335 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2336 2337 mutex_lock(&dirty_i->seglist_lock); 2338 kvfree(dirty_i->dirty_segmap[dirty_type]); 2339 dirty_i->nr_dirty[dirty_type] = 0; 2340 mutex_unlock(&dirty_i->seglist_lock); 2341 } 2342 2343 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 2344 { 2345 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2346 kvfree(dirty_i->victim_secmap); 2347 } 2348 2349 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 2350 { 2351 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2352 int i; 2353 2354 if (!dirty_i) 2355 return; 2356 2357 /* discard pre-free/dirty segments list */ 2358 for (i = 0; i < NR_DIRTY_TYPE; i++) 2359 discard_dirty_segmap(sbi, i); 2360 2361 destroy_victim_secmap(sbi); 2362 SM_I(sbi)->dirty_info = NULL; 2363 kfree(dirty_i); 2364 } 2365 2366 static void destroy_curseg(struct f2fs_sb_info *sbi) 2367 { 2368 struct curseg_info *array = SM_I(sbi)->curseg_array; 2369 int i; 2370 2371 if (!array) 2372 return; 2373 SM_I(sbi)->curseg_array = NULL; 2374 for (i = 0; i < NR_CURSEG_TYPE; i++) 2375 kfree(array[i].sum_blk); 2376 kfree(array); 2377 } 2378 2379 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 2380 { 2381 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 2382 if (!free_i) 2383 return; 2384 SM_I(sbi)->free_info = NULL; 2385 kvfree(free_i->free_segmap); 2386 kvfree(free_i->free_secmap); 2387 kfree(free_i); 2388 } 2389 2390 static void destroy_sit_info(struct f2fs_sb_info *sbi) 2391 { 2392 struct sit_info *sit_i = SIT_I(sbi); 2393 unsigned int start; 2394 2395 if (!sit_i) 2396 return; 2397 2398 if (sit_i->sentries) { 2399 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2400 kfree(sit_i->sentries[start].cur_valid_map); 2401 kfree(sit_i->sentries[start].ckpt_valid_map); 2402 kfree(sit_i->sentries[start].discard_map); 2403 } 2404 } 2405 kfree(sit_i->tmp_map); 2406 2407 kvfree(sit_i->sentries); 2408 kvfree(sit_i->sec_entries); 2409 kvfree(sit_i->dirty_sentries_bitmap); 2410 2411 SM_I(sbi)->sit_info = NULL; 2412 kfree(sit_i->sit_bitmap); 2413 kfree(sit_i); 2414 } 2415 2416 void destroy_segment_manager(struct f2fs_sb_info *sbi) 2417 { 2418 struct f2fs_sm_info *sm_info = SM_I(sbi); 2419 2420 if (!sm_info) 2421 return; 2422 destroy_flush_cmd_control(sbi); 2423 destroy_dirty_segmap(sbi); 2424 destroy_curseg(sbi); 2425 destroy_free_segmap(sbi); 2426 destroy_sit_info(sbi); 2427 sbi->sm_info = NULL; 2428 kfree(sm_info); 2429 } 2430 2431 int __init create_segment_manager_caches(void) 2432 { 2433 discard_entry_slab = f2fs_kmem_cache_create("discard_entry", 2434 sizeof(struct discard_entry)); 2435 if (!discard_entry_slab) 2436 goto fail; 2437 2438 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", 2439 sizeof(struct sit_entry_set)); 2440 if (!sit_entry_set_slab) 2441 goto destory_discard_entry; 2442 2443 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", 2444 sizeof(struct inmem_pages)); 2445 if (!inmem_entry_slab) 2446 goto destroy_sit_entry_set; 2447 return 0; 2448 2449 destroy_sit_entry_set: 2450 kmem_cache_destroy(sit_entry_set_slab); 2451 destory_discard_entry: 2452 kmem_cache_destroy(discard_entry_slab); 2453 fail: 2454 return -ENOMEM; 2455 } 2456 2457 void destroy_segment_manager_caches(void) 2458 { 2459 kmem_cache_destroy(sit_entry_set_slab); 2460 kmem_cache_destroy(discard_entry_slab); 2461 kmem_cache_destroy(inmem_entry_slab); 2462 } 2463