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/vmalloc.h> 18 #include <linux/swap.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 /* 33 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since 34 * MSB and LSB are reversed in a byte by f2fs_set_bit. 35 */ 36 static inline unsigned long __reverse_ffs(unsigned long word) 37 { 38 int num = 0; 39 40 #if BITS_PER_LONG == 64 41 if ((word & 0xffffffff) == 0) { 42 num += 32; 43 word >>= 32; 44 } 45 #endif 46 if ((word & 0xffff) == 0) { 47 num += 16; 48 word >>= 16; 49 } 50 if ((word & 0xff) == 0) { 51 num += 8; 52 word >>= 8; 53 } 54 if ((word & 0xf0) == 0) 55 num += 4; 56 else 57 word >>= 4; 58 if ((word & 0xc) == 0) 59 num += 2; 60 else 61 word >>= 2; 62 if ((word & 0x2) == 0) 63 num += 1; 64 return num; 65 } 66 67 /* 68 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because 69 * f2fs_set_bit makes MSB and LSB reversed in a byte. 70 * Example: 71 * LSB <--> MSB 72 * f2fs_set_bit(0, bitmap) => 0000 0001 73 * f2fs_set_bit(7, bitmap) => 1000 0000 74 */ 75 static unsigned long __find_rev_next_bit(const unsigned long *addr, 76 unsigned long size, unsigned long offset) 77 { 78 const unsigned long *p = addr + BIT_WORD(offset); 79 unsigned long result = offset & ~(BITS_PER_LONG - 1); 80 unsigned long tmp; 81 unsigned long mask, submask; 82 unsigned long quot, rest; 83 84 if (offset >= size) 85 return size; 86 87 size -= result; 88 offset %= BITS_PER_LONG; 89 if (!offset) 90 goto aligned; 91 92 tmp = *(p++); 93 quot = (offset >> 3) << 3; 94 rest = offset & 0x7; 95 mask = ~0UL << quot; 96 submask = (unsigned char)(0xff << rest) >> rest; 97 submask <<= quot; 98 mask &= submask; 99 tmp &= mask; 100 if (size < BITS_PER_LONG) 101 goto found_first; 102 if (tmp) 103 goto found_middle; 104 105 size -= BITS_PER_LONG; 106 result += BITS_PER_LONG; 107 aligned: 108 while (size & ~(BITS_PER_LONG-1)) { 109 tmp = *(p++); 110 if (tmp) 111 goto found_middle; 112 result += BITS_PER_LONG; 113 size -= BITS_PER_LONG; 114 } 115 if (!size) 116 return result; 117 tmp = *p; 118 found_first: 119 tmp &= (~0UL >> (BITS_PER_LONG - size)); 120 if (tmp == 0UL) /* Are any bits set? */ 121 return result + size; /* Nope. */ 122 found_middle: 123 return result + __reverse_ffs(tmp); 124 } 125 126 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, 127 unsigned long size, unsigned long offset) 128 { 129 const unsigned long *p = addr + BIT_WORD(offset); 130 unsigned long result = offset & ~(BITS_PER_LONG - 1); 131 unsigned long tmp; 132 unsigned long mask, submask; 133 unsigned long quot, rest; 134 135 if (offset >= size) 136 return size; 137 138 size -= result; 139 offset %= BITS_PER_LONG; 140 if (!offset) 141 goto aligned; 142 143 tmp = *(p++); 144 quot = (offset >> 3) << 3; 145 rest = offset & 0x7; 146 mask = ~(~0UL << quot); 147 submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest); 148 submask <<= quot; 149 mask += submask; 150 tmp |= mask; 151 if (size < BITS_PER_LONG) 152 goto found_first; 153 if (~tmp) 154 goto found_middle; 155 156 size -= BITS_PER_LONG; 157 result += BITS_PER_LONG; 158 aligned: 159 while (size & ~(BITS_PER_LONG - 1)) { 160 tmp = *(p++); 161 if (~tmp) 162 goto found_middle; 163 result += BITS_PER_LONG; 164 size -= BITS_PER_LONG; 165 } 166 if (!size) 167 return result; 168 tmp = *p; 169 170 found_first: 171 tmp |= ~0UL << size; 172 if (tmp == ~0UL) /* Are any bits zero? */ 173 return result + size; /* Nope. */ 174 found_middle: 175 return result + __reverse_ffz(tmp); 176 } 177 178 void register_inmem_page(struct inode *inode, struct page *page) 179 { 180 struct f2fs_inode_info *fi = F2FS_I(inode); 181 struct inmem_pages *new; 182 int err; 183 184 SetPagePrivate(page); 185 f2fs_trace_pid(page); 186 187 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); 188 189 /* add atomic page indices to the list */ 190 new->page = page; 191 INIT_LIST_HEAD(&new->list); 192 retry: 193 /* increase reference count with clean state */ 194 mutex_lock(&fi->inmem_lock); 195 err = radix_tree_insert(&fi->inmem_root, page->index, new); 196 if (err == -EEXIST) { 197 mutex_unlock(&fi->inmem_lock); 198 kmem_cache_free(inmem_entry_slab, new); 199 return; 200 } else if (err) { 201 mutex_unlock(&fi->inmem_lock); 202 goto retry; 203 } 204 get_page(page); 205 list_add_tail(&new->list, &fi->inmem_pages); 206 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 207 mutex_unlock(&fi->inmem_lock); 208 } 209 210 void commit_inmem_pages(struct inode *inode, bool abort) 211 { 212 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 213 struct f2fs_inode_info *fi = F2FS_I(inode); 214 struct inmem_pages *cur, *tmp; 215 bool submit_bio = false; 216 struct f2fs_io_info fio = { 217 .type = DATA, 218 .rw = WRITE_SYNC | REQ_PRIO, 219 }; 220 221 /* 222 * The abort is true only when f2fs_evict_inode is called. 223 * Basically, the f2fs_evict_inode doesn't produce any data writes, so 224 * that we don't need to call f2fs_balance_fs. 225 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this 226 * inode becomes free by iget_locked in f2fs_iget. 227 */ 228 if (!abort) { 229 f2fs_balance_fs(sbi); 230 f2fs_lock_op(sbi); 231 } 232 233 mutex_lock(&fi->inmem_lock); 234 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { 235 if (!abort) { 236 lock_page(cur->page); 237 if (cur->page->mapping == inode->i_mapping) { 238 f2fs_wait_on_page_writeback(cur->page, DATA); 239 if (clear_page_dirty_for_io(cur->page)) 240 inode_dec_dirty_pages(inode); 241 do_write_data_page(cur->page, &fio); 242 submit_bio = true; 243 } 244 f2fs_put_page(cur->page, 1); 245 } else { 246 put_page(cur->page); 247 } 248 radix_tree_delete(&fi->inmem_root, cur->page->index); 249 list_del(&cur->list); 250 kmem_cache_free(inmem_entry_slab, cur); 251 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 252 } 253 mutex_unlock(&fi->inmem_lock); 254 255 if (!abort) { 256 f2fs_unlock_op(sbi); 257 if (submit_bio) 258 f2fs_submit_merged_bio(sbi, DATA, WRITE); 259 } 260 } 261 262 /* 263 * This function balances dirty node and dentry pages. 264 * In addition, it controls garbage collection. 265 */ 266 void f2fs_balance_fs(struct f2fs_sb_info *sbi) 267 { 268 /* 269 * We should do GC or end up with checkpoint, if there are so many dirty 270 * dir/node pages without enough free segments. 271 */ 272 if (has_not_enough_free_secs(sbi, 0)) { 273 mutex_lock(&sbi->gc_mutex); 274 f2fs_gc(sbi); 275 } 276 } 277 278 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) 279 { 280 /* check the # of cached NAT entries and prefree segments */ 281 if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) || 282 excess_prefree_segs(sbi) || 283 !available_free_memory(sbi, INO_ENTRIES)) 284 f2fs_sync_fs(sbi->sb, true); 285 } 286 287 static int issue_flush_thread(void *data) 288 { 289 struct f2fs_sb_info *sbi = data; 290 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; 291 wait_queue_head_t *q = &fcc->flush_wait_queue; 292 repeat: 293 if (kthread_should_stop()) 294 return 0; 295 296 if (!llist_empty(&fcc->issue_list)) { 297 struct bio *bio = bio_alloc(GFP_NOIO, 0); 298 struct flush_cmd *cmd, *next; 299 int ret; 300 301 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 302 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 303 304 bio->bi_bdev = sbi->sb->s_bdev; 305 ret = submit_bio_wait(WRITE_FLUSH, bio); 306 307 llist_for_each_entry_safe(cmd, next, 308 fcc->dispatch_list, llnode) { 309 cmd->ret = ret; 310 complete(&cmd->wait); 311 } 312 bio_put(bio); 313 fcc->dispatch_list = NULL; 314 } 315 316 wait_event_interruptible(*q, 317 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 318 goto repeat; 319 } 320 321 int f2fs_issue_flush(struct f2fs_sb_info *sbi) 322 { 323 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; 324 struct flush_cmd cmd; 325 326 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER), 327 test_opt(sbi, FLUSH_MERGE)); 328 329 if (test_opt(sbi, NOBARRIER)) 330 return 0; 331 332 if (!test_opt(sbi, FLUSH_MERGE)) 333 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL); 334 335 init_completion(&cmd.wait); 336 337 llist_add(&cmd.llnode, &fcc->issue_list); 338 339 if (!fcc->dispatch_list) 340 wake_up(&fcc->flush_wait_queue); 341 342 wait_for_completion(&cmd.wait); 343 344 return cmd.ret; 345 } 346 347 int create_flush_cmd_control(struct f2fs_sb_info *sbi) 348 { 349 dev_t dev = sbi->sb->s_bdev->bd_dev; 350 struct flush_cmd_control *fcc; 351 int err = 0; 352 353 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL); 354 if (!fcc) 355 return -ENOMEM; 356 init_waitqueue_head(&fcc->flush_wait_queue); 357 init_llist_head(&fcc->issue_list); 358 SM_I(sbi)->cmd_control_info = fcc; 359 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 360 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 361 if (IS_ERR(fcc->f2fs_issue_flush)) { 362 err = PTR_ERR(fcc->f2fs_issue_flush); 363 kfree(fcc); 364 SM_I(sbi)->cmd_control_info = NULL; 365 return err; 366 } 367 368 return err; 369 } 370 371 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi) 372 { 373 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; 374 375 if (fcc && fcc->f2fs_issue_flush) 376 kthread_stop(fcc->f2fs_issue_flush); 377 kfree(fcc); 378 SM_I(sbi)->cmd_control_info = NULL; 379 } 380 381 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 382 enum dirty_type dirty_type) 383 { 384 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 385 386 /* need not be added */ 387 if (IS_CURSEG(sbi, segno)) 388 return; 389 390 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 391 dirty_i->nr_dirty[dirty_type]++; 392 393 if (dirty_type == DIRTY) { 394 struct seg_entry *sentry = get_seg_entry(sbi, segno); 395 enum dirty_type t = sentry->type; 396 397 if (unlikely(t >= DIRTY)) { 398 f2fs_bug_on(sbi, 1); 399 return; 400 } 401 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 402 dirty_i->nr_dirty[t]++; 403 } 404 } 405 406 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 407 enum dirty_type dirty_type) 408 { 409 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 410 411 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 412 dirty_i->nr_dirty[dirty_type]--; 413 414 if (dirty_type == DIRTY) { 415 struct seg_entry *sentry = get_seg_entry(sbi, segno); 416 enum dirty_type t = sentry->type; 417 418 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 419 dirty_i->nr_dirty[t]--; 420 421 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0) 422 clear_bit(GET_SECNO(sbi, segno), 423 dirty_i->victim_secmap); 424 } 425 } 426 427 /* 428 * Should not occur error such as -ENOMEM. 429 * Adding dirty entry into seglist is not critical operation. 430 * If a given segment is one of current working segments, it won't be added. 431 */ 432 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 433 { 434 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 435 unsigned short valid_blocks; 436 437 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 438 return; 439 440 mutex_lock(&dirty_i->seglist_lock); 441 442 valid_blocks = get_valid_blocks(sbi, segno, 0); 443 444 if (valid_blocks == 0) { 445 __locate_dirty_segment(sbi, segno, PRE); 446 __remove_dirty_segment(sbi, segno, DIRTY); 447 } else if (valid_blocks < sbi->blocks_per_seg) { 448 __locate_dirty_segment(sbi, segno, DIRTY); 449 } else { 450 /* Recovery routine with SSR needs this */ 451 __remove_dirty_segment(sbi, segno, DIRTY); 452 } 453 454 mutex_unlock(&dirty_i->seglist_lock); 455 } 456 457 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 458 block_t blkstart, block_t blklen) 459 { 460 sector_t start = SECTOR_FROM_BLOCK(blkstart); 461 sector_t len = SECTOR_FROM_BLOCK(blklen); 462 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen); 463 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0); 464 } 465 466 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr) 467 { 468 if (f2fs_issue_discard(sbi, blkaddr, 1)) { 469 struct page *page = grab_meta_page(sbi, blkaddr); 470 /* zero-filled page */ 471 set_page_dirty(page); 472 f2fs_put_page(page, 1); 473 } 474 } 475 476 static void __add_discard_entry(struct f2fs_sb_info *sbi, 477 struct cp_control *cpc, unsigned int start, unsigned int end) 478 { 479 struct list_head *head = &SM_I(sbi)->discard_list; 480 struct discard_entry *new, *last; 481 482 if (!list_empty(head)) { 483 last = list_last_entry(head, struct discard_entry, list); 484 if (START_BLOCK(sbi, cpc->trim_start) + start == 485 last->blkaddr + last->len) { 486 last->len += end - start; 487 goto done; 488 } 489 } 490 491 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS); 492 INIT_LIST_HEAD(&new->list); 493 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start; 494 new->len = end - start; 495 list_add_tail(&new->list, head); 496 done: 497 SM_I(sbi)->nr_discards += end - start; 498 cpc->trimmed += end - start; 499 } 500 501 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc) 502 { 503 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 504 int max_blocks = sbi->blocks_per_seg; 505 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 506 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 507 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 508 unsigned long *dmap = SIT_I(sbi)->tmp_map; 509 unsigned int start = 0, end = -1; 510 bool force = (cpc->reason == CP_DISCARD); 511 int i; 512 513 if (!force && (!test_opt(sbi, DISCARD) || 514 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)) 515 return; 516 517 if (force && !se->valid_blocks) { 518 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 519 /* 520 * if this segment is registered in the prefree list, then 521 * we should skip adding a discard candidate, and let the 522 * checkpoint do that later. 523 */ 524 mutex_lock(&dirty_i->seglist_lock); 525 if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) { 526 mutex_unlock(&dirty_i->seglist_lock); 527 cpc->trimmed += sbi->blocks_per_seg; 528 return; 529 } 530 mutex_unlock(&dirty_i->seglist_lock); 531 532 __add_discard_entry(sbi, cpc, 0, sbi->blocks_per_seg); 533 return; 534 } 535 536 /* zero block will be discarded through the prefree list */ 537 if (!se->valid_blocks || se->valid_blocks == max_blocks) 538 return; 539 540 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 541 for (i = 0; i < entries; i++) 542 dmap[i] = force ? ~ckpt_map[i] : 543 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 544 545 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) { 546 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 547 if (start >= max_blocks) 548 break; 549 550 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 551 552 if (end - start < cpc->trim_minlen) 553 continue; 554 555 __add_discard_entry(sbi, cpc, start, end); 556 } 557 } 558 559 void release_discard_addrs(struct f2fs_sb_info *sbi) 560 { 561 struct list_head *head = &(SM_I(sbi)->discard_list); 562 struct discard_entry *entry, *this; 563 564 /* drop caches */ 565 list_for_each_entry_safe(entry, this, head, list) { 566 list_del(&entry->list); 567 kmem_cache_free(discard_entry_slab, entry); 568 } 569 } 570 571 /* 572 * Should call clear_prefree_segments after checkpoint is done. 573 */ 574 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 575 { 576 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 577 unsigned int segno; 578 579 mutex_lock(&dirty_i->seglist_lock); 580 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 581 __set_test_and_free(sbi, segno); 582 mutex_unlock(&dirty_i->seglist_lock); 583 } 584 585 void clear_prefree_segments(struct f2fs_sb_info *sbi) 586 { 587 struct list_head *head = &(SM_I(sbi)->discard_list); 588 struct discard_entry *entry, *this; 589 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 590 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 591 unsigned int start = 0, end = -1; 592 593 mutex_lock(&dirty_i->seglist_lock); 594 595 while (1) { 596 int i; 597 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 598 if (start >= MAIN_SEGS(sbi)) 599 break; 600 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 601 start + 1); 602 603 for (i = start; i < end; i++) 604 clear_bit(i, prefree_map); 605 606 dirty_i->nr_dirty[PRE] -= end - start; 607 608 if (!test_opt(sbi, DISCARD)) 609 continue; 610 611 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 612 (end - start) << sbi->log_blocks_per_seg); 613 } 614 mutex_unlock(&dirty_i->seglist_lock); 615 616 /* send small discards */ 617 list_for_each_entry_safe(entry, this, head, list) { 618 f2fs_issue_discard(sbi, entry->blkaddr, entry->len); 619 list_del(&entry->list); 620 SM_I(sbi)->nr_discards -= entry->len; 621 kmem_cache_free(discard_entry_slab, entry); 622 } 623 } 624 625 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 626 { 627 struct sit_info *sit_i = SIT_I(sbi); 628 629 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 630 sit_i->dirty_sentries++; 631 return false; 632 } 633 634 return true; 635 } 636 637 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 638 unsigned int segno, int modified) 639 { 640 struct seg_entry *se = get_seg_entry(sbi, segno); 641 se->type = type; 642 if (modified) 643 __mark_sit_entry_dirty(sbi, segno); 644 } 645 646 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 647 { 648 struct seg_entry *se; 649 unsigned int segno, offset; 650 long int new_vblocks; 651 652 segno = GET_SEGNO(sbi, blkaddr); 653 654 se = get_seg_entry(sbi, segno); 655 new_vblocks = se->valid_blocks + del; 656 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 657 658 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || 659 (new_vblocks > sbi->blocks_per_seg))); 660 661 se->valid_blocks = new_vblocks; 662 se->mtime = get_mtime(sbi); 663 SIT_I(sbi)->max_mtime = se->mtime; 664 665 /* Update valid block bitmap */ 666 if (del > 0) { 667 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) 668 f2fs_bug_on(sbi, 1); 669 } else { 670 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) 671 f2fs_bug_on(sbi, 1); 672 } 673 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 674 se->ckpt_valid_blocks += del; 675 676 __mark_sit_entry_dirty(sbi, segno); 677 678 /* update total number of valid blocks to be written in ckpt area */ 679 SIT_I(sbi)->written_valid_blocks += del; 680 681 if (sbi->segs_per_sec > 1) 682 get_sec_entry(sbi, segno)->valid_blocks += del; 683 } 684 685 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) 686 { 687 update_sit_entry(sbi, new, 1); 688 if (GET_SEGNO(sbi, old) != NULL_SEGNO) 689 update_sit_entry(sbi, old, -1); 690 691 locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); 692 locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); 693 } 694 695 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 696 { 697 unsigned int segno = GET_SEGNO(sbi, addr); 698 struct sit_info *sit_i = SIT_I(sbi); 699 700 f2fs_bug_on(sbi, addr == NULL_ADDR); 701 if (addr == NEW_ADDR) 702 return; 703 704 /* add it into sit main buffer */ 705 mutex_lock(&sit_i->sentry_lock); 706 707 update_sit_entry(sbi, addr, -1); 708 709 /* add it into dirty seglist */ 710 locate_dirty_segment(sbi, segno); 711 712 mutex_unlock(&sit_i->sentry_lock); 713 } 714 715 /* 716 * This function should be resided under the curseg_mutex lock 717 */ 718 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 719 struct f2fs_summary *sum) 720 { 721 struct curseg_info *curseg = CURSEG_I(sbi, type); 722 void *addr = curseg->sum_blk; 723 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 724 memcpy(addr, sum, sizeof(struct f2fs_summary)); 725 } 726 727 /* 728 * Calculate the number of current summary pages for writing 729 */ 730 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 731 { 732 int valid_sum_count = 0; 733 int i, sum_in_page; 734 735 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 736 if (sbi->ckpt->alloc_type[i] == SSR) 737 valid_sum_count += sbi->blocks_per_seg; 738 else { 739 if (for_ra) 740 valid_sum_count += le16_to_cpu( 741 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 742 else 743 valid_sum_count += curseg_blkoff(sbi, i); 744 } 745 } 746 747 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE - 748 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 749 if (valid_sum_count <= sum_in_page) 750 return 1; 751 else if ((valid_sum_count - sum_in_page) <= 752 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 753 return 2; 754 return 3; 755 } 756 757 /* 758 * Caller should put this summary page 759 */ 760 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 761 { 762 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); 763 } 764 765 static void write_sum_page(struct f2fs_sb_info *sbi, 766 struct f2fs_summary_block *sum_blk, block_t blk_addr) 767 { 768 struct page *page = grab_meta_page(sbi, blk_addr); 769 void *kaddr = page_address(page); 770 memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE); 771 set_page_dirty(page); 772 f2fs_put_page(page, 1); 773 } 774 775 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) 776 { 777 struct curseg_info *curseg = CURSEG_I(sbi, type); 778 unsigned int segno = curseg->segno + 1; 779 struct free_segmap_info *free_i = FREE_I(sbi); 780 781 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 782 return !test_bit(segno, free_i->free_segmap); 783 return 0; 784 } 785 786 /* 787 * Find a new segment from the free segments bitmap to right order 788 * This function should be returned with success, otherwise BUG 789 */ 790 static void get_new_segment(struct f2fs_sb_info *sbi, 791 unsigned int *newseg, bool new_sec, int dir) 792 { 793 struct free_segmap_info *free_i = FREE_I(sbi); 794 unsigned int segno, secno, zoneno; 795 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 796 unsigned int hint = *newseg / sbi->segs_per_sec; 797 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); 798 unsigned int left_start = hint; 799 bool init = true; 800 int go_left = 0; 801 int i; 802 803 spin_lock(&free_i->segmap_lock); 804 805 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 806 segno = find_next_zero_bit(free_i->free_segmap, 807 MAIN_SEGS(sbi), *newseg + 1); 808 if (segno - *newseg < sbi->segs_per_sec - 809 (*newseg % sbi->segs_per_sec)) 810 goto got_it; 811 } 812 find_other_zone: 813 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 814 if (secno >= MAIN_SECS(sbi)) { 815 if (dir == ALLOC_RIGHT) { 816 secno = find_next_zero_bit(free_i->free_secmap, 817 MAIN_SECS(sbi), 0); 818 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 819 } else { 820 go_left = 1; 821 left_start = hint - 1; 822 } 823 } 824 if (go_left == 0) 825 goto skip_left; 826 827 while (test_bit(left_start, free_i->free_secmap)) { 828 if (left_start > 0) { 829 left_start--; 830 continue; 831 } 832 left_start = find_next_zero_bit(free_i->free_secmap, 833 MAIN_SECS(sbi), 0); 834 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 835 break; 836 } 837 secno = left_start; 838 skip_left: 839 hint = secno; 840 segno = secno * sbi->segs_per_sec; 841 zoneno = secno / sbi->secs_per_zone; 842 843 /* give up on finding another zone */ 844 if (!init) 845 goto got_it; 846 if (sbi->secs_per_zone == 1) 847 goto got_it; 848 if (zoneno == old_zoneno) 849 goto got_it; 850 if (dir == ALLOC_LEFT) { 851 if (!go_left && zoneno + 1 >= total_zones) 852 goto got_it; 853 if (go_left && zoneno == 0) 854 goto got_it; 855 } 856 for (i = 0; i < NR_CURSEG_TYPE; i++) 857 if (CURSEG_I(sbi, i)->zone == zoneno) 858 break; 859 860 if (i < NR_CURSEG_TYPE) { 861 /* zone is in user, try another */ 862 if (go_left) 863 hint = zoneno * sbi->secs_per_zone - 1; 864 else if (zoneno + 1 >= total_zones) 865 hint = 0; 866 else 867 hint = (zoneno + 1) * sbi->secs_per_zone; 868 init = false; 869 goto find_other_zone; 870 } 871 got_it: 872 /* set it as dirty segment in free segmap */ 873 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 874 __set_inuse(sbi, segno); 875 *newseg = segno; 876 spin_unlock(&free_i->segmap_lock); 877 } 878 879 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 880 { 881 struct curseg_info *curseg = CURSEG_I(sbi, type); 882 struct summary_footer *sum_footer; 883 884 curseg->segno = curseg->next_segno; 885 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); 886 curseg->next_blkoff = 0; 887 curseg->next_segno = NULL_SEGNO; 888 889 sum_footer = &(curseg->sum_blk->footer); 890 memset(sum_footer, 0, sizeof(struct summary_footer)); 891 if (IS_DATASEG(type)) 892 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 893 if (IS_NODESEG(type)) 894 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 895 __set_sit_entry_type(sbi, type, curseg->segno, modified); 896 } 897 898 /* 899 * Allocate a current working segment. 900 * This function always allocates a free segment in LFS manner. 901 */ 902 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 903 { 904 struct curseg_info *curseg = CURSEG_I(sbi, type); 905 unsigned int segno = curseg->segno; 906 int dir = ALLOC_LEFT; 907 908 write_sum_page(sbi, curseg->sum_blk, 909 GET_SUM_BLOCK(sbi, segno)); 910 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 911 dir = ALLOC_RIGHT; 912 913 if (test_opt(sbi, NOHEAP)) 914 dir = ALLOC_RIGHT; 915 916 get_new_segment(sbi, &segno, new_sec, dir); 917 curseg->next_segno = segno; 918 reset_curseg(sbi, type, 1); 919 curseg->alloc_type = LFS; 920 } 921 922 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 923 struct curseg_info *seg, block_t start) 924 { 925 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 926 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 927 unsigned long *target_map = SIT_I(sbi)->tmp_map; 928 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 929 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 930 int i, pos; 931 932 for (i = 0; i < entries; i++) 933 target_map[i] = ckpt_map[i] | cur_map[i]; 934 935 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 936 937 seg->next_blkoff = pos; 938 } 939 940 /* 941 * If a segment is written by LFS manner, next block offset is just obtained 942 * by increasing the current block offset. However, if a segment is written by 943 * SSR manner, next block offset obtained by calling __next_free_blkoff 944 */ 945 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 946 struct curseg_info *seg) 947 { 948 if (seg->alloc_type == SSR) 949 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 950 else 951 seg->next_blkoff++; 952 } 953 954 /* 955 * This function always allocates a used segment(from dirty seglist) by SSR 956 * manner, so it should recover the existing segment information of valid blocks 957 */ 958 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) 959 { 960 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 961 struct curseg_info *curseg = CURSEG_I(sbi, type); 962 unsigned int new_segno = curseg->next_segno; 963 struct f2fs_summary_block *sum_node; 964 struct page *sum_page; 965 966 write_sum_page(sbi, curseg->sum_blk, 967 GET_SUM_BLOCK(sbi, curseg->segno)); 968 __set_test_and_inuse(sbi, new_segno); 969 970 mutex_lock(&dirty_i->seglist_lock); 971 __remove_dirty_segment(sbi, new_segno, PRE); 972 __remove_dirty_segment(sbi, new_segno, DIRTY); 973 mutex_unlock(&dirty_i->seglist_lock); 974 975 reset_curseg(sbi, type, 1); 976 curseg->alloc_type = SSR; 977 __next_free_blkoff(sbi, curseg, 0); 978 979 if (reuse) { 980 sum_page = get_sum_page(sbi, new_segno); 981 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 982 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 983 f2fs_put_page(sum_page, 1); 984 } 985 } 986 987 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 988 { 989 struct curseg_info *curseg = CURSEG_I(sbi, type); 990 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 991 992 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0)) 993 return v_ops->get_victim(sbi, 994 &(curseg)->next_segno, BG_GC, type, SSR); 995 996 /* For data segments, let's do SSR more intensively */ 997 for (; type >= CURSEG_HOT_DATA; type--) 998 if (v_ops->get_victim(sbi, &(curseg)->next_segno, 999 BG_GC, type, SSR)) 1000 return 1; 1001 return 0; 1002 } 1003 1004 /* 1005 * flush out current segment and replace it with new segment 1006 * This function should be returned with success, otherwise BUG 1007 */ 1008 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 1009 int type, bool force) 1010 { 1011 struct curseg_info *curseg = CURSEG_I(sbi, type); 1012 1013 if (force) 1014 new_curseg(sbi, type, true); 1015 else if (type == CURSEG_WARM_NODE) 1016 new_curseg(sbi, type, false); 1017 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) 1018 new_curseg(sbi, type, false); 1019 else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) 1020 change_curseg(sbi, type, true); 1021 else 1022 new_curseg(sbi, type, false); 1023 1024 stat_inc_seg_type(sbi, curseg); 1025 } 1026 1027 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type) 1028 { 1029 struct curseg_info *curseg = CURSEG_I(sbi, type); 1030 unsigned int old_segno; 1031 1032 old_segno = curseg->segno; 1033 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true); 1034 locate_dirty_segment(sbi, old_segno); 1035 } 1036 1037 void allocate_new_segments(struct f2fs_sb_info *sbi) 1038 { 1039 int i; 1040 1041 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 1042 __allocate_new_segments(sbi, i); 1043 } 1044 1045 static const struct segment_allocation default_salloc_ops = { 1046 .allocate_segment = allocate_segment_by_default, 1047 }; 1048 1049 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 1050 { 1051 __u64 start = F2FS_BYTES_TO_BLK(range->start); 1052 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 1053 unsigned int start_segno, end_segno; 1054 struct cp_control cpc; 1055 1056 if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) || 1057 range->len < sbi->blocksize) 1058 return -EINVAL; 1059 1060 cpc.trimmed = 0; 1061 if (end <= MAIN_BLKADDR(sbi)) 1062 goto out; 1063 1064 /* start/end segment number in main_area */ 1065 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 1066 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 1067 GET_SEGNO(sbi, end); 1068 cpc.reason = CP_DISCARD; 1069 cpc.trim_minlen = F2FS_BYTES_TO_BLK(range->minlen); 1070 1071 /* do checkpoint to issue discard commands safely */ 1072 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) { 1073 cpc.trim_start = start_segno; 1074 cpc.trim_end = min_t(unsigned int, rounddown(start_segno + 1075 BATCHED_TRIM_SEGMENTS(sbi), 1076 sbi->segs_per_sec) - 1, end_segno); 1077 1078 mutex_lock(&sbi->gc_mutex); 1079 write_checkpoint(sbi, &cpc); 1080 mutex_unlock(&sbi->gc_mutex); 1081 } 1082 out: 1083 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed); 1084 return 0; 1085 } 1086 1087 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 1088 { 1089 struct curseg_info *curseg = CURSEG_I(sbi, type); 1090 if (curseg->next_blkoff < sbi->blocks_per_seg) 1091 return true; 1092 return false; 1093 } 1094 1095 static int __get_segment_type_2(struct page *page, enum page_type p_type) 1096 { 1097 if (p_type == DATA) 1098 return CURSEG_HOT_DATA; 1099 else 1100 return CURSEG_HOT_NODE; 1101 } 1102 1103 static int __get_segment_type_4(struct page *page, enum page_type p_type) 1104 { 1105 if (p_type == DATA) { 1106 struct inode *inode = page->mapping->host; 1107 1108 if (S_ISDIR(inode->i_mode)) 1109 return CURSEG_HOT_DATA; 1110 else 1111 return CURSEG_COLD_DATA; 1112 } else { 1113 if (IS_DNODE(page) && is_cold_node(page)) 1114 return CURSEG_WARM_NODE; 1115 else 1116 return CURSEG_COLD_NODE; 1117 } 1118 } 1119 1120 static int __get_segment_type_6(struct page *page, enum page_type p_type) 1121 { 1122 if (p_type == DATA) { 1123 struct inode *inode = page->mapping->host; 1124 1125 if (S_ISDIR(inode->i_mode)) 1126 return CURSEG_HOT_DATA; 1127 else if (is_cold_data(page) || file_is_cold(inode)) 1128 return CURSEG_COLD_DATA; 1129 else 1130 return CURSEG_WARM_DATA; 1131 } else { 1132 if (IS_DNODE(page)) 1133 return is_cold_node(page) ? CURSEG_WARM_NODE : 1134 CURSEG_HOT_NODE; 1135 else 1136 return CURSEG_COLD_NODE; 1137 } 1138 } 1139 1140 static int __get_segment_type(struct page *page, enum page_type p_type) 1141 { 1142 switch (F2FS_P_SB(page)->active_logs) { 1143 case 2: 1144 return __get_segment_type_2(page, p_type); 1145 case 4: 1146 return __get_segment_type_4(page, p_type); 1147 } 1148 /* NR_CURSEG_TYPE(6) logs by default */ 1149 f2fs_bug_on(F2FS_P_SB(page), 1150 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE); 1151 return __get_segment_type_6(page, p_type); 1152 } 1153 1154 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 1155 block_t old_blkaddr, block_t *new_blkaddr, 1156 struct f2fs_summary *sum, int type) 1157 { 1158 struct sit_info *sit_i = SIT_I(sbi); 1159 struct curseg_info *curseg; 1160 bool direct_io = (type == CURSEG_DIRECT_IO); 1161 1162 type = direct_io ? CURSEG_WARM_DATA : type; 1163 1164 curseg = CURSEG_I(sbi, type); 1165 1166 mutex_lock(&curseg->curseg_mutex); 1167 1168 /* direct_io'ed data is aligned to the segment for better performance */ 1169 if (direct_io && curseg->next_blkoff) 1170 __allocate_new_segments(sbi, type); 1171 1172 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 1173 1174 /* 1175 * __add_sum_entry should be resided under the curseg_mutex 1176 * because, this function updates a summary entry in the 1177 * current summary block. 1178 */ 1179 __add_sum_entry(sbi, type, sum); 1180 1181 mutex_lock(&sit_i->sentry_lock); 1182 __refresh_next_blkoff(sbi, curseg); 1183 1184 stat_inc_block_count(sbi, curseg); 1185 1186 if (!__has_curseg_space(sbi, type)) 1187 sit_i->s_ops->allocate_segment(sbi, type, false); 1188 /* 1189 * SIT information should be updated before segment allocation, 1190 * since SSR needs latest valid block information. 1191 */ 1192 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); 1193 1194 mutex_unlock(&sit_i->sentry_lock); 1195 1196 if (page && IS_NODESEG(type)) 1197 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 1198 1199 mutex_unlock(&curseg->curseg_mutex); 1200 } 1201 1202 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page, 1203 struct f2fs_summary *sum, 1204 struct f2fs_io_info *fio) 1205 { 1206 int type = __get_segment_type(page, fio->type); 1207 1208 allocate_data_block(sbi, page, fio->blk_addr, &fio->blk_addr, sum, type); 1209 1210 /* writeout dirty page into bdev */ 1211 f2fs_submit_page_mbio(sbi, page, fio); 1212 } 1213 1214 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) 1215 { 1216 struct f2fs_io_info fio = { 1217 .type = META, 1218 .rw = WRITE_SYNC | REQ_META | REQ_PRIO, 1219 .blk_addr = page->index, 1220 }; 1221 1222 set_page_writeback(page); 1223 f2fs_submit_page_mbio(sbi, page, &fio); 1224 } 1225 1226 void write_node_page(struct f2fs_sb_info *sbi, struct page *page, 1227 unsigned int nid, struct f2fs_io_info *fio) 1228 { 1229 struct f2fs_summary sum; 1230 set_summary(&sum, nid, 0, 0); 1231 do_write_page(sbi, page, &sum, fio); 1232 } 1233 1234 void write_data_page(struct page *page, struct dnode_of_data *dn, 1235 struct f2fs_io_info *fio) 1236 { 1237 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1238 struct f2fs_summary sum; 1239 struct node_info ni; 1240 1241 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 1242 get_node_info(sbi, dn->nid, &ni); 1243 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 1244 do_write_page(sbi, page, &sum, fio); 1245 dn->data_blkaddr = fio->blk_addr; 1246 } 1247 1248 void rewrite_data_page(struct page *page, struct f2fs_io_info *fio) 1249 { 1250 stat_inc_inplace_blocks(F2FS_P_SB(page)); 1251 f2fs_submit_page_mbio(F2FS_P_SB(page), page, fio); 1252 } 1253 1254 void recover_data_page(struct f2fs_sb_info *sbi, 1255 struct page *page, struct f2fs_summary *sum, 1256 block_t old_blkaddr, block_t new_blkaddr) 1257 { 1258 struct sit_info *sit_i = SIT_I(sbi); 1259 struct curseg_info *curseg; 1260 unsigned int segno, old_cursegno; 1261 struct seg_entry *se; 1262 int type; 1263 1264 segno = GET_SEGNO(sbi, new_blkaddr); 1265 se = get_seg_entry(sbi, segno); 1266 type = se->type; 1267 1268 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 1269 if (old_blkaddr == NULL_ADDR) 1270 type = CURSEG_COLD_DATA; 1271 else 1272 type = CURSEG_WARM_DATA; 1273 } 1274 curseg = CURSEG_I(sbi, type); 1275 1276 mutex_lock(&curseg->curseg_mutex); 1277 mutex_lock(&sit_i->sentry_lock); 1278 1279 old_cursegno = curseg->segno; 1280 1281 /* change the current segment */ 1282 if (segno != curseg->segno) { 1283 curseg->next_segno = segno; 1284 change_curseg(sbi, type, true); 1285 } 1286 1287 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 1288 __add_sum_entry(sbi, type, sum); 1289 1290 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); 1291 locate_dirty_segment(sbi, old_cursegno); 1292 1293 mutex_unlock(&sit_i->sentry_lock); 1294 mutex_unlock(&curseg->curseg_mutex); 1295 } 1296 1297 static inline bool is_merged_page(struct f2fs_sb_info *sbi, 1298 struct page *page, enum page_type type) 1299 { 1300 enum page_type btype = PAGE_TYPE_OF_BIO(type); 1301 struct f2fs_bio_info *io = &sbi->write_io[btype]; 1302 struct bio_vec *bvec; 1303 int i; 1304 1305 down_read(&io->io_rwsem); 1306 if (!io->bio) 1307 goto out; 1308 1309 bio_for_each_segment_all(bvec, io->bio, i) { 1310 if (page == bvec->bv_page) { 1311 up_read(&io->io_rwsem); 1312 return true; 1313 } 1314 } 1315 1316 out: 1317 up_read(&io->io_rwsem); 1318 return false; 1319 } 1320 1321 void f2fs_wait_on_page_writeback(struct page *page, 1322 enum page_type type) 1323 { 1324 if (PageWriteback(page)) { 1325 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1326 1327 if (is_merged_page(sbi, page, type)) 1328 f2fs_submit_merged_bio(sbi, type, WRITE); 1329 wait_on_page_writeback(page); 1330 } 1331 } 1332 1333 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 1334 { 1335 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1336 struct curseg_info *seg_i; 1337 unsigned char *kaddr; 1338 struct page *page; 1339 block_t start; 1340 int i, j, offset; 1341 1342 start = start_sum_block(sbi); 1343 1344 page = get_meta_page(sbi, start++); 1345 kaddr = (unsigned char *)page_address(page); 1346 1347 /* Step 1: restore nat cache */ 1348 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 1349 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE); 1350 1351 /* Step 2: restore sit cache */ 1352 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 1353 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE, 1354 SUM_JOURNAL_SIZE); 1355 offset = 2 * SUM_JOURNAL_SIZE; 1356 1357 /* Step 3: restore summary entries */ 1358 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1359 unsigned short blk_off; 1360 unsigned int segno; 1361 1362 seg_i = CURSEG_I(sbi, i); 1363 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 1364 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 1365 seg_i->next_segno = segno; 1366 reset_curseg(sbi, i, 0); 1367 seg_i->alloc_type = ckpt->alloc_type[i]; 1368 seg_i->next_blkoff = blk_off; 1369 1370 if (seg_i->alloc_type == SSR) 1371 blk_off = sbi->blocks_per_seg; 1372 1373 for (j = 0; j < blk_off; j++) { 1374 struct f2fs_summary *s; 1375 s = (struct f2fs_summary *)(kaddr + offset); 1376 seg_i->sum_blk->entries[j] = *s; 1377 offset += SUMMARY_SIZE; 1378 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE - 1379 SUM_FOOTER_SIZE) 1380 continue; 1381 1382 f2fs_put_page(page, 1); 1383 page = NULL; 1384 1385 page = get_meta_page(sbi, start++); 1386 kaddr = (unsigned char *)page_address(page); 1387 offset = 0; 1388 } 1389 } 1390 f2fs_put_page(page, 1); 1391 return 0; 1392 } 1393 1394 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 1395 { 1396 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1397 struct f2fs_summary_block *sum; 1398 struct curseg_info *curseg; 1399 struct page *new; 1400 unsigned short blk_off; 1401 unsigned int segno = 0; 1402 block_t blk_addr = 0; 1403 1404 /* get segment number and block addr */ 1405 if (IS_DATASEG(type)) { 1406 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 1407 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 1408 CURSEG_HOT_DATA]); 1409 if (__exist_node_summaries(sbi)) 1410 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 1411 else 1412 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 1413 } else { 1414 segno = le32_to_cpu(ckpt->cur_node_segno[type - 1415 CURSEG_HOT_NODE]); 1416 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 1417 CURSEG_HOT_NODE]); 1418 if (__exist_node_summaries(sbi)) 1419 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 1420 type - CURSEG_HOT_NODE); 1421 else 1422 blk_addr = GET_SUM_BLOCK(sbi, segno); 1423 } 1424 1425 new = get_meta_page(sbi, blk_addr); 1426 sum = (struct f2fs_summary_block *)page_address(new); 1427 1428 if (IS_NODESEG(type)) { 1429 if (__exist_node_summaries(sbi)) { 1430 struct f2fs_summary *ns = &sum->entries[0]; 1431 int i; 1432 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 1433 ns->version = 0; 1434 ns->ofs_in_node = 0; 1435 } 1436 } else { 1437 int err; 1438 1439 err = restore_node_summary(sbi, segno, sum); 1440 if (err) { 1441 f2fs_put_page(new, 1); 1442 return err; 1443 } 1444 } 1445 } 1446 1447 /* set uncompleted segment to curseg */ 1448 curseg = CURSEG_I(sbi, type); 1449 mutex_lock(&curseg->curseg_mutex); 1450 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE); 1451 curseg->next_segno = segno; 1452 reset_curseg(sbi, type, 0); 1453 curseg->alloc_type = ckpt->alloc_type[type]; 1454 curseg->next_blkoff = blk_off; 1455 mutex_unlock(&curseg->curseg_mutex); 1456 f2fs_put_page(new, 1); 1457 return 0; 1458 } 1459 1460 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 1461 { 1462 int type = CURSEG_HOT_DATA; 1463 int err; 1464 1465 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) { 1466 int npages = npages_for_summary_flush(sbi, true); 1467 1468 if (npages >= 2) 1469 ra_meta_pages(sbi, start_sum_block(sbi), npages, 1470 META_CP); 1471 1472 /* restore for compacted data summary */ 1473 if (read_compacted_summaries(sbi)) 1474 return -EINVAL; 1475 type = CURSEG_HOT_NODE; 1476 } 1477 1478 if (__exist_node_summaries(sbi)) 1479 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), 1480 NR_CURSEG_TYPE - type, META_CP); 1481 1482 for (; type <= CURSEG_COLD_NODE; type++) { 1483 err = read_normal_summaries(sbi, type); 1484 if (err) 1485 return err; 1486 } 1487 1488 return 0; 1489 } 1490 1491 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 1492 { 1493 struct page *page; 1494 unsigned char *kaddr; 1495 struct f2fs_summary *summary; 1496 struct curseg_info *seg_i; 1497 int written_size = 0; 1498 int i, j; 1499 1500 page = grab_meta_page(sbi, blkaddr++); 1501 kaddr = (unsigned char *)page_address(page); 1502 1503 /* Step 1: write nat cache */ 1504 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 1505 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE); 1506 written_size += SUM_JOURNAL_SIZE; 1507 1508 /* Step 2: write sit cache */ 1509 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 1510 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits, 1511 SUM_JOURNAL_SIZE); 1512 written_size += SUM_JOURNAL_SIZE; 1513 1514 /* Step 3: write summary entries */ 1515 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1516 unsigned short blkoff; 1517 seg_i = CURSEG_I(sbi, i); 1518 if (sbi->ckpt->alloc_type[i] == SSR) 1519 blkoff = sbi->blocks_per_seg; 1520 else 1521 blkoff = curseg_blkoff(sbi, i); 1522 1523 for (j = 0; j < blkoff; j++) { 1524 if (!page) { 1525 page = grab_meta_page(sbi, blkaddr++); 1526 kaddr = (unsigned char *)page_address(page); 1527 written_size = 0; 1528 } 1529 summary = (struct f2fs_summary *)(kaddr + written_size); 1530 *summary = seg_i->sum_blk->entries[j]; 1531 written_size += SUMMARY_SIZE; 1532 1533 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE - 1534 SUM_FOOTER_SIZE) 1535 continue; 1536 1537 set_page_dirty(page); 1538 f2fs_put_page(page, 1); 1539 page = NULL; 1540 } 1541 } 1542 if (page) { 1543 set_page_dirty(page); 1544 f2fs_put_page(page, 1); 1545 } 1546 } 1547 1548 static void write_normal_summaries(struct f2fs_sb_info *sbi, 1549 block_t blkaddr, int type) 1550 { 1551 int i, end; 1552 if (IS_DATASEG(type)) 1553 end = type + NR_CURSEG_DATA_TYPE; 1554 else 1555 end = type + NR_CURSEG_NODE_TYPE; 1556 1557 for (i = type; i < end; i++) { 1558 struct curseg_info *sum = CURSEG_I(sbi, i); 1559 mutex_lock(&sum->curseg_mutex); 1560 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type)); 1561 mutex_unlock(&sum->curseg_mutex); 1562 } 1563 } 1564 1565 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 1566 { 1567 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) 1568 write_compacted_summaries(sbi, start_blk); 1569 else 1570 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 1571 } 1572 1573 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 1574 { 1575 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 1576 } 1577 1578 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type, 1579 unsigned int val, int alloc) 1580 { 1581 int i; 1582 1583 if (type == NAT_JOURNAL) { 1584 for (i = 0; i < nats_in_cursum(sum); i++) { 1585 if (le32_to_cpu(nid_in_journal(sum, i)) == val) 1586 return i; 1587 } 1588 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) 1589 return update_nats_in_cursum(sum, 1); 1590 } else if (type == SIT_JOURNAL) { 1591 for (i = 0; i < sits_in_cursum(sum); i++) 1592 if (le32_to_cpu(segno_in_journal(sum, i)) == val) 1593 return i; 1594 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES) 1595 return update_sits_in_cursum(sum, 1); 1596 } 1597 return -1; 1598 } 1599 1600 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 1601 unsigned int segno) 1602 { 1603 return get_meta_page(sbi, current_sit_addr(sbi, segno)); 1604 } 1605 1606 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 1607 unsigned int start) 1608 { 1609 struct sit_info *sit_i = SIT_I(sbi); 1610 struct page *src_page, *dst_page; 1611 pgoff_t src_off, dst_off; 1612 void *src_addr, *dst_addr; 1613 1614 src_off = current_sit_addr(sbi, start); 1615 dst_off = next_sit_addr(sbi, src_off); 1616 1617 /* get current sit block page without lock */ 1618 src_page = get_meta_page(sbi, src_off); 1619 dst_page = grab_meta_page(sbi, dst_off); 1620 f2fs_bug_on(sbi, PageDirty(src_page)); 1621 1622 src_addr = page_address(src_page); 1623 dst_addr = page_address(dst_page); 1624 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); 1625 1626 set_page_dirty(dst_page); 1627 f2fs_put_page(src_page, 1); 1628 1629 set_to_next_sit(sit_i, start); 1630 1631 return dst_page; 1632 } 1633 1634 static struct sit_entry_set *grab_sit_entry_set(void) 1635 { 1636 struct sit_entry_set *ses = 1637 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC); 1638 1639 ses->entry_cnt = 0; 1640 INIT_LIST_HEAD(&ses->set_list); 1641 return ses; 1642 } 1643 1644 static void release_sit_entry_set(struct sit_entry_set *ses) 1645 { 1646 list_del(&ses->set_list); 1647 kmem_cache_free(sit_entry_set_slab, ses); 1648 } 1649 1650 static void adjust_sit_entry_set(struct sit_entry_set *ses, 1651 struct list_head *head) 1652 { 1653 struct sit_entry_set *next = ses; 1654 1655 if (list_is_last(&ses->set_list, head)) 1656 return; 1657 1658 list_for_each_entry_continue(next, head, set_list) 1659 if (ses->entry_cnt <= next->entry_cnt) 1660 break; 1661 1662 list_move_tail(&ses->set_list, &next->set_list); 1663 } 1664 1665 static void add_sit_entry(unsigned int segno, struct list_head *head) 1666 { 1667 struct sit_entry_set *ses; 1668 unsigned int start_segno = START_SEGNO(segno); 1669 1670 list_for_each_entry(ses, head, set_list) { 1671 if (ses->start_segno == start_segno) { 1672 ses->entry_cnt++; 1673 adjust_sit_entry_set(ses, head); 1674 return; 1675 } 1676 } 1677 1678 ses = grab_sit_entry_set(); 1679 1680 ses->start_segno = start_segno; 1681 ses->entry_cnt++; 1682 list_add(&ses->set_list, head); 1683 } 1684 1685 static void add_sits_in_set(struct f2fs_sb_info *sbi) 1686 { 1687 struct f2fs_sm_info *sm_info = SM_I(sbi); 1688 struct list_head *set_list = &sm_info->sit_entry_set; 1689 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 1690 unsigned int segno; 1691 1692 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 1693 add_sit_entry(segno, set_list); 1694 } 1695 1696 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 1697 { 1698 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1699 struct f2fs_summary_block *sum = curseg->sum_blk; 1700 int i; 1701 1702 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) { 1703 unsigned int segno; 1704 bool dirtied; 1705 1706 segno = le32_to_cpu(segno_in_journal(sum, i)); 1707 dirtied = __mark_sit_entry_dirty(sbi, segno); 1708 1709 if (!dirtied) 1710 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 1711 } 1712 update_sits_in_cursum(sum, -sits_in_cursum(sum)); 1713 } 1714 1715 /* 1716 * CP calls this function, which flushes SIT entries including sit_journal, 1717 * and moves prefree segs to free segs. 1718 */ 1719 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1720 { 1721 struct sit_info *sit_i = SIT_I(sbi); 1722 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 1723 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1724 struct f2fs_summary_block *sum = curseg->sum_blk; 1725 struct sit_entry_set *ses, *tmp; 1726 struct list_head *head = &SM_I(sbi)->sit_entry_set; 1727 bool to_journal = true; 1728 struct seg_entry *se; 1729 1730 mutex_lock(&curseg->curseg_mutex); 1731 mutex_lock(&sit_i->sentry_lock); 1732 1733 /* 1734 * add and account sit entries of dirty bitmap in sit entry 1735 * set temporarily 1736 */ 1737 add_sits_in_set(sbi); 1738 1739 /* 1740 * if there are no enough space in journal to store dirty sit 1741 * entries, remove all entries from journal and add and account 1742 * them in sit entry set. 1743 */ 1744 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL)) 1745 remove_sits_in_journal(sbi); 1746 1747 if (!sit_i->dirty_sentries) 1748 goto out; 1749 1750 /* 1751 * there are two steps to flush sit entries: 1752 * #1, flush sit entries to journal in current cold data summary block. 1753 * #2, flush sit entries to sit page. 1754 */ 1755 list_for_each_entry_safe(ses, tmp, head, set_list) { 1756 struct page *page = NULL; 1757 struct f2fs_sit_block *raw_sit = NULL; 1758 unsigned int start_segno = ses->start_segno; 1759 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 1760 (unsigned long)MAIN_SEGS(sbi)); 1761 unsigned int segno = start_segno; 1762 1763 if (to_journal && 1764 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL)) 1765 to_journal = false; 1766 1767 if (!to_journal) { 1768 page = get_next_sit_page(sbi, start_segno); 1769 raw_sit = page_address(page); 1770 } 1771 1772 /* flush dirty sit entries in region of current sit set */ 1773 for_each_set_bit_from(segno, bitmap, end) { 1774 int offset, sit_offset; 1775 1776 se = get_seg_entry(sbi, segno); 1777 1778 /* add discard candidates */ 1779 if (cpc->reason != CP_DISCARD) { 1780 cpc->trim_start = segno; 1781 add_discard_addrs(sbi, cpc); 1782 } 1783 1784 if (to_journal) { 1785 offset = lookup_journal_in_cursum(sum, 1786 SIT_JOURNAL, segno, 1); 1787 f2fs_bug_on(sbi, offset < 0); 1788 segno_in_journal(sum, offset) = 1789 cpu_to_le32(segno); 1790 seg_info_to_raw_sit(se, 1791 &sit_in_journal(sum, offset)); 1792 } else { 1793 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 1794 seg_info_to_raw_sit(se, 1795 &raw_sit->entries[sit_offset]); 1796 } 1797 1798 __clear_bit(segno, bitmap); 1799 sit_i->dirty_sentries--; 1800 ses->entry_cnt--; 1801 } 1802 1803 if (!to_journal) 1804 f2fs_put_page(page, 1); 1805 1806 f2fs_bug_on(sbi, ses->entry_cnt); 1807 release_sit_entry_set(ses); 1808 } 1809 1810 f2fs_bug_on(sbi, !list_empty(head)); 1811 f2fs_bug_on(sbi, sit_i->dirty_sentries); 1812 out: 1813 if (cpc->reason == CP_DISCARD) { 1814 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 1815 add_discard_addrs(sbi, cpc); 1816 } 1817 mutex_unlock(&sit_i->sentry_lock); 1818 mutex_unlock(&curseg->curseg_mutex); 1819 1820 set_prefree_as_free_segments(sbi); 1821 } 1822 1823 static int build_sit_info(struct f2fs_sb_info *sbi) 1824 { 1825 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 1826 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1827 struct sit_info *sit_i; 1828 unsigned int sit_segs, start; 1829 char *src_bitmap, *dst_bitmap; 1830 unsigned int bitmap_size; 1831 1832 /* allocate memory for SIT information */ 1833 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); 1834 if (!sit_i) 1835 return -ENOMEM; 1836 1837 SM_I(sbi)->sit_info = sit_i; 1838 1839 sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry)); 1840 if (!sit_i->sentries) 1841 return -ENOMEM; 1842 1843 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 1844 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL); 1845 if (!sit_i->dirty_sentries_bitmap) 1846 return -ENOMEM; 1847 1848 for (start = 0; start < MAIN_SEGS(sbi); start++) { 1849 sit_i->sentries[start].cur_valid_map 1850 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 1851 sit_i->sentries[start].ckpt_valid_map 1852 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 1853 if (!sit_i->sentries[start].cur_valid_map 1854 || !sit_i->sentries[start].ckpt_valid_map) 1855 return -ENOMEM; 1856 } 1857 1858 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 1859 if (!sit_i->tmp_map) 1860 return -ENOMEM; 1861 1862 if (sbi->segs_per_sec > 1) { 1863 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) * 1864 sizeof(struct sec_entry)); 1865 if (!sit_i->sec_entries) 1866 return -ENOMEM; 1867 } 1868 1869 /* get information related with SIT */ 1870 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 1871 1872 /* setup SIT bitmap from ckeckpoint pack */ 1873 bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 1874 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 1875 1876 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 1877 if (!dst_bitmap) 1878 return -ENOMEM; 1879 1880 /* init SIT information */ 1881 sit_i->s_ops = &default_salloc_ops; 1882 1883 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 1884 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 1885 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count); 1886 sit_i->sit_bitmap = dst_bitmap; 1887 sit_i->bitmap_size = bitmap_size; 1888 sit_i->dirty_sentries = 0; 1889 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 1890 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 1891 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; 1892 mutex_init(&sit_i->sentry_lock); 1893 return 0; 1894 } 1895 1896 static int build_free_segmap(struct f2fs_sb_info *sbi) 1897 { 1898 struct free_segmap_info *free_i; 1899 unsigned int bitmap_size, sec_bitmap_size; 1900 1901 /* allocate memory for free segmap information */ 1902 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); 1903 if (!free_i) 1904 return -ENOMEM; 1905 1906 SM_I(sbi)->free_info = free_i; 1907 1908 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 1909 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL); 1910 if (!free_i->free_segmap) 1911 return -ENOMEM; 1912 1913 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 1914 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL); 1915 if (!free_i->free_secmap) 1916 return -ENOMEM; 1917 1918 /* set all segments as dirty temporarily */ 1919 memset(free_i->free_segmap, 0xff, bitmap_size); 1920 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 1921 1922 /* init free segmap information */ 1923 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 1924 free_i->free_segments = 0; 1925 free_i->free_sections = 0; 1926 spin_lock_init(&free_i->segmap_lock); 1927 return 0; 1928 } 1929 1930 static int build_curseg(struct f2fs_sb_info *sbi) 1931 { 1932 struct curseg_info *array; 1933 int i; 1934 1935 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL); 1936 if (!array) 1937 return -ENOMEM; 1938 1939 SM_I(sbi)->curseg_array = array; 1940 1941 for (i = 0; i < NR_CURSEG_TYPE; i++) { 1942 mutex_init(&array[i].curseg_mutex); 1943 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL); 1944 if (!array[i].sum_blk) 1945 return -ENOMEM; 1946 array[i].segno = NULL_SEGNO; 1947 array[i].next_blkoff = 0; 1948 } 1949 return restore_curseg_summaries(sbi); 1950 } 1951 1952 static void build_sit_entries(struct f2fs_sb_info *sbi) 1953 { 1954 struct sit_info *sit_i = SIT_I(sbi); 1955 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1956 struct f2fs_summary_block *sum = curseg->sum_blk; 1957 int sit_blk_cnt = SIT_BLK_CNT(sbi); 1958 unsigned int i, start, end; 1959 unsigned int readed, start_blk = 0; 1960 int nrpages = MAX_BIO_BLOCKS(sbi); 1961 1962 do { 1963 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT); 1964 1965 start = start_blk * sit_i->sents_per_block; 1966 end = (start_blk + readed) * sit_i->sents_per_block; 1967 1968 for (; start < end && start < MAIN_SEGS(sbi); start++) { 1969 struct seg_entry *se = &sit_i->sentries[start]; 1970 struct f2fs_sit_block *sit_blk; 1971 struct f2fs_sit_entry sit; 1972 struct page *page; 1973 1974 mutex_lock(&curseg->curseg_mutex); 1975 for (i = 0; i < sits_in_cursum(sum); i++) { 1976 if (le32_to_cpu(segno_in_journal(sum, i)) 1977 == start) { 1978 sit = sit_in_journal(sum, i); 1979 mutex_unlock(&curseg->curseg_mutex); 1980 goto got_it; 1981 } 1982 } 1983 mutex_unlock(&curseg->curseg_mutex); 1984 1985 page = get_current_sit_page(sbi, start); 1986 sit_blk = (struct f2fs_sit_block *)page_address(page); 1987 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 1988 f2fs_put_page(page, 1); 1989 got_it: 1990 check_block_count(sbi, start, &sit); 1991 seg_info_from_raw_sit(se, &sit); 1992 if (sbi->segs_per_sec > 1) { 1993 struct sec_entry *e = get_sec_entry(sbi, start); 1994 e->valid_blocks += se->valid_blocks; 1995 } 1996 } 1997 start_blk += readed; 1998 } while (start_blk < sit_blk_cnt); 1999 } 2000 2001 static void init_free_segmap(struct f2fs_sb_info *sbi) 2002 { 2003 unsigned int start; 2004 int type; 2005 2006 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2007 struct seg_entry *sentry = get_seg_entry(sbi, start); 2008 if (!sentry->valid_blocks) 2009 __set_free(sbi, start); 2010 } 2011 2012 /* set use the current segments */ 2013 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 2014 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 2015 __set_test_and_inuse(sbi, curseg_t->segno); 2016 } 2017 } 2018 2019 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 2020 { 2021 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2022 struct free_segmap_info *free_i = FREE_I(sbi); 2023 unsigned int segno = 0, offset = 0; 2024 unsigned short valid_blocks; 2025 2026 while (1) { 2027 /* find dirty segment based on free segmap */ 2028 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 2029 if (segno >= MAIN_SEGS(sbi)) 2030 break; 2031 offset = segno + 1; 2032 valid_blocks = get_valid_blocks(sbi, segno, 0); 2033 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) 2034 continue; 2035 if (valid_blocks > sbi->blocks_per_seg) { 2036 f2fs_bug_on(sbi, 1); 2037 continue; 2038 } 2039 mutex_lock(&dirty_i->seglist_lock); 2040 __locate_dirty_segment(sbi, segno, DIRTY); 2041 mutex_unlock(&dirty_i->seglist_lock); 2042 } 2043 } 2044 2045 static int init_victim_secmap(struct f2fs_sb_info *sbi) 2046 { 2047 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2048 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 2049 2050 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL); 2051 if (!dirty_i->victim_secmap) 2052 return -ENOMEM; 2053 return 0; 2054 } 2055 2056 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 2057 { 2058 struct dirty_seglist_info *dirty_i; 2059 unsigned int bitmap_size, i; 2060 2061 /* allocate memory for dirty segments list information */ 2062 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); 2063 if (!dirty_i) 2064 return -ENOMEM; 2065 2066 SM_I(sbi)->dirty_info = dirty_i; 2067 mutex_init(&dirty_i->seglist_lock); 2068 2069 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 2070 2071 for (i = 0; i < NR_DIRTY_TYPE; i++) { 2072 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL); 2073 if (!dirty_i->dirty_segmap[i]) 2074 return -ENOMEM; 2075 } 2076 2077 init_dirty_segmap(sbi); 2078 return init_victim_secmap(sbi); 2079 } 2080 2081 /* 2082 * Update min, max modified time for cost-benefit GC algorithm 2083 */ 2084 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 2085 { 2086 struct sit_info *sit_i = SIT_I(sbi); 2087 unsigned int segno; 2088 2089 mutex_lock(&sit_i->sentry_lock); 2090 2091 sit_i->min_mtime = LLONG_MAX; 2092 2093 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 2094 unsigned int i; 2095 unsigned long long mtime = 0; 2096 2097 for (i = 0; i < sbi->segs_per_sec; i++) 2098 mtime += get_seg_entry(sbi, segno + i)->mtime; 2099 2100 mtime = div_u64(mtime, sbi->segs_per_sec); 2101 2102 if (sit_i->min_mtime > mtime) 2103 sit_i->min_mtime = mtime; 2104 } 2105 sit_i->max_mtime = get_mtime(sbi); 2106 mutex_unlock(&sit_i->sentry_lock); 2107 } 2108 2109 int build_segment_manager(struct f2fs_sb_info *sbi) 2110 { 2111 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 2112 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2113 struct f2fs_sm_info *sm_info; 2114 int err; 2115 2116 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); 2117 if (!sm_info) 2118 return -ENOMEM; 2119 2120 /* init sm info */ 2121 sbi->sm_info = sm_info; 2122 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 2123 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 2124 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 2125 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 2126 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 2127 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 2128 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 2129 sm_info->rec_prefree_segments = sm_info->main_segments * 2130 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 2131 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 2132 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 2133 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 2134 2135 INIT_LIST_HEAD(&sm_info->discard_list); 2136 sm_info->nr_discards = 0; 2137 sm_info->max_discards = 0; 2138 2139 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS; 2140 2141 INIT_LIST_HEAD(&sm_info->sit_entry_set); 2142 2143 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) { 2144 err = create_flush_cmd_control(sbi); 2145 if (err) 2146 return err; 2147 } 2148 2149 err = build_sit_info(sbi); 2150 if (err) 2151 return err; 2152 err = build_free_segmap(sbi); 2153 if (err) 2154 return err; 2155 err = build_curseg(sbi); 2156 if (err) 2157 return err; 2158 2159 /* reinit free segmap based on SIT */ 2160 build_sit_entries(sbi); 2161 2162 init_free_segmap(sbi); 2163 err = build_dirty_segmap(sbi); 2164 if (err) 2165 return err; 2166 2167 init_min_max_mtime(sbi); 2168 return 0; 2169 } 2170 2171 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 2172 enum dirty_type dirty_type) 2173 { 2174 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2175 2176 mutex_lock(&dirty_i->seglist_lock); 2177 kfree(dirty_i->dirty_segmap[dirty_type]); 2178 dirty_i->nr_dirty[dirty_type] = 0; 2179 mutex_unlock(&dirty_i->seglist_lock); 2180 } 2181 2182 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 2183 { 2184 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2185 kfree(dirty_i->victim_secmap); 2186 } 2187 2188 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 2189 { 2190 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2191 int i; 2192 2193 if (!dirty_i) 2194 return; 2195 2196 /* discard pre-free/dirty segments list */ 2197 for (i = 0; i < NR_DIRTY_TYPE; i++) 2198 discard_dirty_segmap(sbi, i); 2199 2200 destroy_victim_secmap(sbi); 2201 SM_I(sbi)->dirty_info = NULL; 2202 kfree(dirty_i); 2203 } 2204 2205 static void destroy_curseg(struct f2fs_sb_info *sbi) 2206 { 2207 struct curseg_info *array = SM_I(sbi)->curseg_array; 2208 int i; 2209 2210 if (!array) 2211 return; 2212 SM_I(sbi)->curseg_array = NULL; 2213 for (i = 0; i < NR_CURSEG_TYPE; i++) 2214 kfree(array[i].sum_blk); 2215 kfree(array); 2216 } 2217 2218 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 2219 { 2220 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 2221 if (!free_i) 2222 return; 2223 SM_I(sbi)->free_info = NULL; 2224 kfree(free_i->free_segmap); 2225 kfree(free_i->free_secmap); 2226 kfree(free_i); 2227 } 2228 2229 static void destroy_sit_info(struct f2fs_sb_info *sbi) 2230 { 2231 struct sit_info *sit_i = SIT_I(sbi); 2232 unsigned int start; 2233 2234 if (!sit_i) 2235 return; 2236 2237 if (sit_i->sentries) { 2238 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2239 kfree(sit_i->sentries[start].cur_valid_map); 2240 kfree(sit_i->sentries[start].ckpt_valid_map); 2241 } 2242 } 2243 kfree(sit_i->tmp_map); 2244 2245 vfree(sit_i->sentries); 2246 vfree(sit_i->sec_entries); 2247 kfree(sit_i->dirty_sentries_bitmap); 2248 2249 SM_I(sbi)->sit_info = NULL; 2250 kfree(sit_i->sit_bitmap); 2251 kfree(sit_i); 2252 } 2253 2254 void destroy_segment_manager(struct f2fs_sb_info *sbi) 2255 { 2256 struct f2fs_sm_info *sm_info = SM_I(sbi); 2257 2258 if (!sm_info) 2259 return; 2260 destroy_flush_cmd_control(sbi); 2261 destroy_dirty_segmap(sbi); 2262 destroy_curseg(sbi); 2263 destroy_free_segmap(sbi); 2264 destroy_sit_info(sbi); 2265 sbi->sm_info = NULL; 2266 kfree(sm_info); 2267 } 2268 2269 int __init create_segment_manager_caches(void) 2270 { 2271 discard_entry_slab = f2fs_kmem_cache_create("discard_entry", 2272 sizeof(struct discard_entry)); 2273 if (!discard_entry_slab) 2274 goto fail; 2275 2276 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", 2277 sizeof(struct sit_entry_set)); 2278 if (!sit_entry_set_slab) 2279 goto destory_discard_entry; 2280 2281 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", 2282 sizeof(struct inmem_pages)); 2283 if (!inmem_entry_slab) 2284 goto destroy_sit_entry_set; 2285 return 0; 2286 2287 destroy_sit_entry_set: 2288 kmem_cache_destroy(sit_entry_set_slab); 2289 destory_discard_entry: 2290 kmem_cache_destroy(discard_entry_slab); 2291 fail: 2292 return -ENOMEM; 2293 } 2294 2295 void destroy_segment_manager_caches(void) 2296 { 2297 kmem_cache_destroy(sit_entry_set_slab); 2298 kmem_cache_destroy(discard_entry_slab); 2299 kmem_cache_destroy(inmem_entry_slab); 2300 } 2301