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 *discard_cmd_slab; 30 static struct kmem_cache *sit_entry_set_slab; 31 static struct kmem_cache *inmem_entry_slab; 32 33 static unsigned long __reverse_ulong(unsigned char *str) 34 { 35 unsigned long tmp = 0; 36 int shift = 24, idx = 0; 37 38 #if BITS_PER_LONG == 64 39 shift = 56; 40 #endif 41 while (shift >= 0) { 42 tmp |= (unsigned long)str[idx++] << shift; 43 shift -= BITS_PER_BYTE; 44 } 45 return tmp; 46 } 47 48 /* 49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since 50 * MSB and LSB are reversed in a byte by f2fs_set_bit. 51 */ 52 static inline unsigned long __reverse_ffs(unsigned long word) 53 { 54 int num = 0; 55 56 #if BITS_PER_LONG == 64 57 if ((word & 0xffffffff00000000UL) == 0) 58 num += 32; 59 else 60 word >>= 32; 61 #endif 62 if ((word & 0xffff0000) == 0) 63 num += 16; 64 else 65 word >>= 16; 66 67 if ((word & 0xff00) == 0) 68 num += 8; 69 else 70 word >>= 8; 71 72 if ((word & 0xf0) == 0) 73 num += 4; 74 else 75 word >>= 4; 76 77 if ((word & 0xc) == 0) 78 num += 2; 79 else 80 word >>= 2; 81 82 if ((word & 0x2) == 0) 83 num += 1; 84 return num; 85 } 86 87 /* 88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because 89 * f2fs_set_bit makes MSB and LSB reversed in a byte. 90 * @size must be integral times of unsigned long. 91 * Example: 92 * MSB <--> LSB 93 * f2fs_set_bit(0, bitmap) => 1000 0000 94 * f2fs_set_bit(7, bitmap) => 0000 0001 95 */ 96 static unsigned long __find_rev_next_bit(const unsigned long *addr, 97 unsigned long size, unsigned long offset) 98 { 99 const unsigned long *p = addr + BIT_WORD(offset); 100 unsigned long result = size; 101 unsigned long tmp; 102 103 if (offset >= size) 104 return size; 105 106 size -= (offset & ~(BITS_PER_LONG - 1)); 107 offset %= BITS_PER_LONG; 108 109 while (1) { 110 if (*p == 0) 111 goto pass; 112 113 tmp = __reverse_ulong((unsigned char *)p); 114 115 tmp &= ~0UL >> offset; 116 if (size < BITS_PER_LONG) 117 tmp &= (~0UL << (BITS_PER_LONG - size)); 118 if (tmp) 119 goto found; 120 pass: 121 if (size <= BITS_PER_LONG) 122 break; 123 size -= BITS_PER_LONG; 124 offset = 0; 125 p++; 126 } 127 return result; 128 found: 129 return result - size + __reverse_ffs(tmp); 130 } 131 132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, 133 unsigned long size, unsigned long offset) 134 { 135 const unsigned long *p = addr + BIT_WORD(offset); 136 unsigned long result = size; 137 unsigned long tmp; 138 139 if (offset >= size) 140 return size; 141 142 size -= (offset & ~(BITS_PER_LONG - 1)); 143 offset %= BITS_PER_LONG; 144 145 while (1) { 146 if (*p == ~0UL) 147 goto pass; 148 149 tmp = __reverse_ulong((unsigned char *)p); 150 151 if (offset) 152 tmp |= ~0UL << (BITS_PER_LONG - offset); 153 if (size < BITS_PER_LONG) 154 tmp |= ~0UL >> size; 155 if (tmp != ~0UL) 156 goto found; 157 pass: 158 if (size <= BITS_PER_LONG) 159 break; 160 size -= BITS_PER_LONG; 161 offset = 0; 162 p++; 163 } 164 return result; 165 found: 166 return result - size + __reverse_ffz(tmp); 167 } 168 169 void register_inmem_page(struct inode *inode, struct page *page) 170 { 171 struct f2fs_inode_info *fi = F2FS_I(inode); 172 struct inmem_pages *new; 173 174 f2fs_trace_pid(page); 175 176 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE); 177 SetPagePrivate(page); 178 179 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); 180 181 /* add atomic page indices to the list */ 182 new->page = page; 183 INIT_LIST_HEAD(&new->list); 184 185 /* increase reference count with clean state */ 186 mutex_lock(&fi->inmem_lock); 187 get_page(page); 188 list_add_tail(&new->list, &fi->inmem_pages); 189 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 190 mutex_unlock(&fi->inmem_lock); 191 192 trace_f2fs_register_inmem_page(page, INMEM); 193 } 194 195 static int __revoke_inmem_pages(struct inode *inode, 196 struct list_head *head, bool drop, bool recover) 197 { 198 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 199 struct inmem_pages *cur, *tmp; 200 int err = 0; 201 202 list_for_each_entry_safe(cur, tmp, head, list) { 203 struct page *page = cur->page; 204 205 if (drop) 206 trace_f2fs_commit_inmem_page(page, INMEM_DROP); 207 208 lock_page(page); 209 210 if (recover) { 211 struct dnode_of_data dn; 212 struct node_info ni; 213 214 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE); 215 216 set_new_dnode(&dn, inode, NULL, NULL, 0); 217 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) { 218 err = -EAGAIN; 219 goto next; 220 } 221 get_node_info(sbi, dn.nid, &ni); 222 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 223 cur->old_addr, ni.version, true, true); 224 f2fs_put_dnode(&dn); 225 } 226 next: 227 /* we don't need to invalidate this in the sccessful status */ 228 if (drop || recover) 229 ClearPageUptodate(page); 230 set_page_private(page, 0); 231 ClearPagePrivate(page); 232 f2fs_put_page(page, 1); 233 234 list_del(&cur->list); 235 kmem_cache_free(inmem_entry_slab, cur); 236 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 237 } 238 return err; 239 } 240 241 void drop_inmem_pages(struct inode *inode) 242 { 243 struct f2fs_inode_info *fi = F2FS_I(inode); 244 245 mutex_lock(&fi->inmem_lock); 246 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); 247 mutex_unlock(&fi->inmem_lock); 248 249 clear_inode_flag(inode, FI_ATOMIC_FILE); 250 stat_dec_atomic_write(inode); 251 } 252 253 void drop_inmem_page(struct inode *inode, struct page *page) 254 { 255 struct f2fs_inode_info *fi = F2FS_I(inode); 256 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 257 struct list_head *head = &fi->inmem_pages; 258 struct inmem_pages *cur = NULL; 259 260 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page)); 261 262 mutex_lock(&fi->inmem_lock); 263 list_for_each_entry(cur, head, list) { 264 if (cur->page == page) 265 break; 266 } 267 268 f2fs_bug_on(sbi, !cur || cur->page != page); 269 list_del(&cur->list); 270 mutex_unlock(&fi->inmem_lock); 271 272 dec_page_count(sbi, F2FS_INMEM_PAGES); 273 kmem_cache_free(inmem_entry_slab, cur); 274 275 ClearPageUptodate(page); 276 set_page_private(page, 0); 277 ClearPagePrivate(page); 278 f2fs_put_page(page, 0); 279 280 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE); 281 } 282 283 static int __commit_inmem_pages(struct inode *inode, 284 struct list_head *revoke_list) 285 { 286 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 287 struct f2fs_inode_info *fi = F2FS_I(inode); 288 struct inmem_pages *cur, *tmp; 289 struct f2fs_io_info fio = { 290 .sbi = sbi, 291 .type = DATA, 292 .op = REQ_OP_WRITE, 293 .op_flags = REQ_SYNC | REQ_PRIO, 294 .encrypted_page = NULL, 295 }; 296 pgoff_t last_idx = ULONG_MAX; 297 int err = 0; 298 299 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { 300 struct page *page = cur->page; 301 302 lock_page(page); 303 if (page->mapping == inode->i_mapping) { 304 trace_f2fs_commit_inmem_page(page, INMEM); 305 306 set_page_dirty(page); 307 f2fs_wait_on_page_writeback(page, DATA, true); 308 if (clear_page_dirty_for_io(page)) { 309 inode_dec_dirty_pages(inode); 310 remove_dirty_inode(inode); 311 } 312 313 fio.page = page; 314 err = do_write_data_page(&fio); 315 if (err) { 316 unlock_page(page); 317 break; 318 } 319 320 /* record old blkaddr for revoking */ 321 cur->old_addr = fio.old_blkaddr; 322 last_idx = page->index; 323 } 324 unlock_page(page); 325 list_move_tail(&cur->list, revoke_list); 326 } 327 328 if (last_idx != ULONG_MAX) 329 f2fs_submit_merged_bio_cond(sbi, inode, 0, last_idx, 330 DATA, WRITE); 331 332 if (!err) 333 __revoke_inmem_pages(inode, revoke_list, false, false); 334 335 return err; 336 } 337 338 int commit_inmem_pages(struct inode *inode) 339 { 340 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 341 struct f2fs_inode_info *fi = F2FS_I(inode); 342 struct list_head revoke_list; 343 int err; 344 345 INIT_LIST_HEAD(&revoke_list); 346 f2fs_balance_fs(sbi, true); 347 f2fs_lock_op(sbi); 348 349 set_inode_flag(inode, FI_ATOMIC_COMMIT); 350 351 mutex_lock(&fi->inmem_lock); 352 err = __commit_inmem_pages(inode, &revoke_list); 353 if (err) { 354 int ret; 355 /* 356 * try to revoke all committed pages, but still we could fail 357 * due to no memory or other reason, if that happened, EAGAIN 358 * will be returned, which means in such case, transaction is 359 * already not integrity, caller should use journal to do the 360 * recovery or rewrite & commit last transaction. For other 361 * error number, revoking was done by filesystem itself. 362 */ 363 ret = __revoke_inmem_pages(inode, &revoke_list, false, true); 364 if (ret) 365 err = ret; 366 367 /* drop all uncommitted pages */ 368 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); 369 } 370 mutex_unlock(&fi->inmem_lock); 371 372 clear_inode_flag(inode, FI_ATOMIC_COMMIT); 373 374 f2fs_unlock_op(sbi); 375 return err; 376 } 377 378 /* 379 * This function balances dirty node and dentry pages. 380 * In addition, it controls garbage collection. 381 */ 382 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 383 { 384 #ifdef CONFIG_F2FS_FAULT_INJECTION 385 if (time_to_inject(sbi, FAULT_CHECKPOINT)) { 386 f2fs_show_injection_info(FAULT_CHECKPOINT); 387 f2fs_stop_checkpoint(sbi, false); 388 } 389 #endif 390 391 if (!need) 392 return; 393 394 /* balance_fs_bg is able to be pending */ 395 if (excess_cached_nats(sbi)) 396 f2fs_balance_fs_bg(sbi); 397 398 /* 399 * We should do GC or end up with checkpoint, if there are so many dirty 400 * dir/node pages without enough free segments. 401 */ 402 if (has_not_enough_free_secs(sbi, 0, 0)) { 403 mutex_lock(&sbi->gc_mutex); 404 f2fs_gc(sbi, false, false); 405 } 406 } 407 408 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) 409 { 410 /* try to shrink extent cache when there is no enough memory */ 411 if (!available_free_memory(sbi, EXTENT_CACHE)) 412 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); 413 414 /* check the # of cached NAT entries */ 415 if (!available_free_memory(sbi, NAT_ENTRIES)) 416 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 417 418 if (!available_free_memory(sbi, FREE_NIDS)) 419 try_to_free_nids(sbi, MAX_FREE_NIDS); 420 else 421 build_free_nids(sbi, false, false); 422 423 if (!is_idle(sbi)) 424 return; 425 426 /* checkpoint is the only way to shrink partial cached entries */ 427 if (!available_free_memory(sbi, NAT_ENTRIES) || 428 !available_free_memory(sbi, INO_ENTRIES) || 429 excess_prefree_segs(sbi) || 430 excess_dirty_nats(sbi) || 431 f2fs_time_over(sbi, CP_TIME)) { 432 if (test_opt(sbi, DATA_FLUSH)) { 433 struct blk_plug plug; 434 435 blk_start_plug(&plug); 436 sync_dirty_inodes(sbi, FILE_INODE); 437 blk_finish_plug(&plug); 438 } 439 f2fs_sync_fs(sbi->sb, true); 440 stat_inc_bg_cp_count(sbi->stat_info); 441 } 442 } 443 444 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 445 struct block_device *bdev) 446 { 447 struct bio *bio = f2fs_bio_alloc(0); 448 int ret; 449 450 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; 451 bio->bi_bdev = bdev; 452 ret = submit_bio_wait(bio); 453 bio_put(bio); 454 455 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 456 test_opt(sbi, FLUSH_MERGE), ret); 457 return ret; 458 } 459 460 static int submit_flush_wait(struct f2fs_sb_info *sbi) 461 { 462 int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev); 463 int i; 464 465 if (!sbi->s_ndevs || ret) 466 return ret; 467 468 for (i = 1; i < sbi->s_ndevs; i++) { 469 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 470 if (ret) 471 break; 472 } 473 return ret; 474 } 475 476 static int issue_flush_thread(void *data) 477 { 478 struct f2fs_sb_info *sbi = data; 479 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 480 wait_queue_head_t *q = &fcc->flush_wait_queue; 481 repeat: 482 if (kthread_should_stop()) 483 return 0; 484 485 if (!llist_empty(&fcc->issue_list)) { 486 struct flush_cmd *cmd, *next; 487 int ret; 488 489 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 490 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 491 492 ret = submit_flush_wait(sbi); 493 atomic_inc(&fcc->issued_flush); 494 495 llist_for_each_entry_safe(cmd, next, 496 fcc->dispatch_list, llnode) { 497 cmd->ret = ret; 498 complete(&cmd->wait); 499 } 500 fcc->dispatch_list = NULL; 501 } 502 503 wait_event_interruptible(*q, 504 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 505 goto repeat; 506 } 507 508 int f2fs_issue_flush(struct f2fs_sb_info *sbi) 509 { 510 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 511 struct flush_cmd cmd; 512 int ret; 513 514 if (test_opt(sbi, NOBARRIER)) 515 return 0; 516 517 if (!test_opt(sbi, FLUSH_MERGE)) { 518 ret = submit_flush_wait(sbi); 519 atomic_inc(&fcc->issued_flush); 520 return ret; 521 } 522 523 if (!atomic_read(&fcc->issing_flush)) { 524 atomic_inc(&fcc->issing_flush); 525 ret = submit_flush_wait(sbi); 526 atomic_dec(&fcc->issing_flush); 527 528 atomic_inc(&fcc->issued_flush); 529 return ret; 530 } 531 532 init_completion(&cmd.wait); 533 534 atomic_inc(&fcc->issing_flush); 535 llist_add(&cmd.llnode, &fcc->issue_list); 536 537 if (!fcc->dispatch_list) 538 wake_up(&fcc->flush_wait_queue); 539 540 if (fcc->f2fs_issue_flush) { 541 wait_for_completion(&cmd.wait); 542 atomic_dec(&fcc->issing_flush); 543 } else { 544 llist_del_all(&fcc->issue_list); 545 atomic_set(&fcc->issing_flush, 0); 546 } 547 548 return cmd.ret; 549 } 550 551 int create_flush_cmd_control(struct f2fs_sb_info *sbi) 552 { 553 dev_t dev = sbi->sb->s_bdev->bd_dev; 554 struct flush_cmd_control *fcc; 555 int err = 0; 556 557 if (SM_I(sbi)->fcc_info) { 558 fcc = SM_I(sbi)->fcc_info; 559 goto init_thread; 560 } 561 562 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL); 563 if (!fcc) 564 return -ENOMEM; 565 atomic_set(&fcc->issued_flush, 0); 566 atomic_set(&fcc->issing_flush, 0); 567 init_waitqueue_head(&fcc->flush_wait_queue); 568 init_llist_head(&fcc->issue_list); 569 SM_I(sbi)->fcc_info = fcc; 570 init_thread: 571 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 572 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 573 if (IS_ERR(fcc->f2fs_issue_flush)) { 574 err = PTR_ERR(fcc->f2fs_issue_flush); 575 kfree(fcc); 576 SM_I(sbi)->fcc_info = NULL; 577 return err; 578 } 579 580 return err; 581 } 582 583 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 584 { 585 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 586 587 if (fcc && fcc->f2fs_issue_flush) { 588 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 589 590 fcc->f2fs_issue_flush = NULL; 591 kthread_stop(flush_thread); 592 } 593 if (free) { 594 kfree(fcc); 595 SM_I(sbi)->fcc_info = NULL; 596 } 597 } 598 599 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 600 enum dirty_type dirty_type) 601 { 602 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 603 604 /* need not be added */ 605 if (IS_CURSEG(sbi, segno)) 606 return; 607 608 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 609 dirty_i->nr_dirty[dirty_type]++; 610 611 if (dirty_type == DIRTY) { 612 struct seg_entry *sentry = get_seg_entry(sbi, segno); 613 enum dirty_type t = sentry->type; 614 615 if (unlikely(t >= DIRTY)) { 616 f2fs_bug_on(sbi, 1); 617 return; 618 } 619 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 620 dirty_i->nr_dirty[t]++; 621 } 622 } 623 624 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 625 enum dirty_type dirty_type) 626 { 627 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 628 629 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 630 dirty_i->nr_dirty[dirty_type]--; 631 632 if (dirty_type == DIRTY) { 633 struct seg_entry *sentry = get_seg_entry(sbi, segno); 634 enum dirty_type t = sentry->type; 635 636 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 637 dirty_i->nr_dirty[t]--; 638 639 if (get_valid_blocks(sbi, segno, true) == 0) 640 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 641 dirty_i->victim_secmap); 642 } 643 } 644 645 /* 646 * Should not occur error such as -ENOMEM. 647 * Adding dirty entry into seglist is not critical operation. 648 * If a given segment is one of current working segments, it won't be added. 649 */ 650 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 651 { 652 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 653 unsigned short valid_blocks; 654 655 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 656 return; 657 658 mutex_lock(&dirty_i->seglist_lock); 659 660 valid_blocks = get_valid_blocks(sbi, segno, false); 661 662 if (valid_blocks == 0) { 663 __locate_dirty_segment(sbi, segno, PRE); 664 __remove_dirty_segment(sbi, segno, DIRTY); 665 } else if (valid_blocks < sbi->blocks_per_seg) { 666 __locate_dirty_segment(sbi, segno, DIRTY); 667 } else { 668 /* Recovery routine with SSR needs this */ 669 __remove_dirty_segment(sbi, segno, DIRTY); 670 } 671 672 mutex_unlock(&dirty_i->seglist_lock); 673 } 674 675 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 676 struct block_device *bdev, block_t lstart, 677 block_t start, block_t len) 678 { 679 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 680 struct list_head *pend_list = &(dcc->pend_list); 681 struct discard_cmd *dc; 682 683 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS); 684 INIT_LIST_HEAD(&dc->list); 685 dc->bdev = bdev; 686 dc->lstart = lstart; 687 dc->start = start; 688 dc->len = len; 689 dc->state = D_PREP; 690 dc->error = 0; 691 init_completion(&dc->wait); 692 list_add_tail(&dc->list, pend_list); 693 atomic_inc(&dcc->discard_cmd_cnt); 694 695 return dc; 696 } 697 698 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi, 699 struct block_device *bdev, block_t lstart, 700 block_t start, block_t len, 701 struct rb_node *parent, struct rb_node **p) 702 { 703 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 704 struct discard_cmd *dc; 705 706 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 707 708 rb_link_node(&dc->rb_node, parent, p); 709 rb_insert_color(&dc->rb_node, &dcc->root); 710 711 return dc; 712 } 713 714 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 715 struct discard_cmd *dc) 716 { 717 if (dc->state == D_DONE) 718 atomic_dec(&dcc->issing_discard); 719 720 list_del(&dc->list); 721 rb_erase(&dc->rb_node, &dcc->root); 722 723 kmem_cache_free(discard_cmd_slab, dc); 724 725 atomic_dec(&dcc->discard_cmd_cnt); 726 } 727 728 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 729 struct discard_cmd *dc) 730 { 731 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 732 733 if (dc->error == -EOPNOTSUPP) 734 dc->error = 0; 735 736 if (dc->error) 737 f2fs_msg(sbi->sb, KERN_INFO, 738 "Issue discard failed, ret: %d", dc->error); 739 __detach_discard_cmd(dcc, dc); 740 } 741 742 static void f2fs_submit_discard_endio(struct bio *bio) 743 { 744 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 745 746 dc->error = bio->bi_error; 747 dc->state = D_DONE; 748 complete(&dc->wait); 749 bio_put(bio); 750 } 751 752 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 753 static void __submit_discard_cmd(struct f2fs_sb_info *sbi, 754 struct discard_cmd *dc) 755 { 756 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 757 struct bio *bio = NULL; 758 759 if (dc->state != D_PREP) 760 return; 761 762 dc->error = __blkdev_issue_discard(dc->bdev, 763 SECTOR_FROM_BLOCK(dc->start), 764 SECTOR_FROM_BLOCK(dc->len), 765 GFP_NOFS, 0, &bio); 766 if (!dc->error) { 767 /* should keep before submission to avoid D_DONE right away */ 768 dc->state = D_SUBMIT; 769 atomic_inc(&dcc->issued_discard); 770 atomic_inc(&dcc->issing_discard); 771 if (bio) { 772 bio->bi_private = dc; 773 bio->bi_end_io = f2fs_submit_discard_endio; 774 bio->bi_opf |= REQ_SYNC; 775 submit_bio(bio); 776 list_move_tail(&dc->list, &dcc->wait_list); 777 } 778 } else { 779 __remove_discard_cmd(sbi, dc); 780 } 781 } 782 783 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi, 784 struct block_device *bdev, block_t lstart, 785 block_t start, block_t len, 786 struct rb_node **insert_p, 787 struct rb_node *insert_parent) 788 { 789 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 790 struct rb_node **p = &dcc->root.rb_node; 791 struct rb_node *parent = NULL; 792 struct discard_cmd *dc = NULL; 793 794 if (insert_p && insert_parent) { 795 parent = insert_parent; 796 p = insert_p; 797 goto do_insert; 798 } 799 800 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart); 801 do_insert: 802 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p); 803 if (!dc) 804 return NULL; 805 806 return dc; 807 } 808 809 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 810 struct discard_cmd *dc, block_t blkaddr) 811 { 812 struct discard_info di = dc->di; 813 bool modified = false; 814 815 if (dc->state == D_DONE || dc->len == 1) { 816 __remove_discard_cmd(sbi, dc); 817 return; 818 } 819 820 if (blkaddr > di.lstart) { 821 dc->len = blkaddr - dc->lstart; 822 modified = true; 823 } 824 825 if (blkaddr < di.lstart + di.len - 1) { 826 if (modified) { 827 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, 828 di.start + blkaddr + 1 - di.lstart, 829 di.lstart + di.len - 1 - blkaddr, 830 NULL, NULL); 831 } else { 832 dc->lstart++; 833 dc->len--; 834 dc->start++; 835 } 836 } 837 } 838 839 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 840 struct block_device *bdev, block_t lstart, 841 block_t start, block_t len) 842 { 843 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 844 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 845 struct discard_cmd *dc; 846 struct discard_info di = {0}; 847 struct rb_node **insert_p = NULL, *insert_parent = NULL; 848 block_t end = lstart + len; 849 850 mutex_lock(&dcc->cmd_lock); 851 852 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root, 853 NULL, lstart, 854 (struct rb_entry **)&prev_dc, 855 (struct rb_entry **)&next_dc, 856 &insert_p, &insert_parent, true); 857 if (dc) 858 prev_dc = dc; 859 860 if (!prev_dc) { 861 di.lstart = lstart; 862 di.len = next_dc ? next_dc->lstart - lstart : len; 863 di.len = min(di.len, len); 864 di.start = start; 865 } 866 867 while (1) { 868 struct rb_node *node; 869 bool merged = false; 870 struct discard_cmd *tdc = NULL; 871 872 if (prev_dc) { 873 di.lstart = prev_dc->lstart + prev_dc->len; 874 if (di.lstart < lstart) 875 di.lstart = lstart; 876 if (di.lstart >= end) 877 break; 878 879 if (!next_dc || next_dc->lstart > end) 880 di.len = end - di.lstart; 881 else 882 di.len = next_dc->lstart - di.lstart; 883 di.start = start + di.lstart - lstart; 884 } 885 886 if (!di.len) 887 goto next; 888 889 if (prev_dc && prev_dc->state == D_PREP && 890 prev_dc->bdev == bdev && 891 __is_discard_back_mergeable(&di, &prev_dc->di)) { 892 prev_dc->di.len += di.len; 893 di = prev_dc->di; 894 tdc = prev_dc; 895 merged = true; 896 } 897 898 if (next_dc && next_dc->state == D_PREP && 899 next_dc->bdev == bdev && 900 __is_discard_front_mergeable(&di, &next_dc->di)) { 901 next_dc->di.lstart = di.lstart; 902 next_dc->di.len += di.len; 903 next_dc->di.start = di.start; 904 if (tdc) 905 __remove_discard_cmd(sbi, tdc); 906 907 merged = true; 908 } 909 910 if (!merged) 911 __insert_discard_tree(sbi, bdev, di.lstart, di.start, 912 di.len, NULL, NULL); 913 next: 914 prev_dc = next_dc; 915 if (!prev_dc) 916 break; 917 918 node = rb_next(&prev_dc->rb_node); 919 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 920 } 921 922 mutex_unlock(&dcc->cmd_lock); 923 } 924 925 static int __queue_discard_cmd(struct f2fs_sb_info *sbi, 926 struct block_device *bdev, block_t blkstart, block_t blklen) 927 { 928 block_t lblkstart = blkstart; 929 930 trace_f2fs_issue_discard(bdev, blkstart, blklen); 931 932 if (sbi->s_ndevs) { 933 int devi = f2fs_target_device_index(sbi, blkstart); 934 935 blkstart -= FDEV(devi).start_blk; 936 } 937 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 938 wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue); 939 return 0; 940 } 941 942 /* This should be covered by global mutex, &sit_i->sentry_lock */ 943 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 944 { 945 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 946 struct discard_cmd *dc; 947 948 mutex_lock(&dcc->cmd_lock); 949 950 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr); 951 if (dc) { 952 if (dc->state != D_PREP) 953 wait_for_completion_io(&dc->wait); 954 __punch_discard_cmd(sbi, dc, blkaddr); 955 } 956 957 mutex_unlock(&dcc->cmd_lock); 958 } 959 960 /* This comes from f2fs_put_super */ 961 void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi) 962 { 963 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 964 struct list_head *pend_list = &(dcc->pend_list); 965 struct list_head *wait_list = &(dcc->wait_list); 966 struct discard_cmd *dc, *tmp; 967 struct blk_plug plug; 968 969 mutex_lock(&dcc->cmd_lock); 970 971 blk_start_plug(&plug); 972 list_for_each_entry_safe(dc, tmp, pend_list, list) 973 __submit_discard_cmd(sbi, dc); 974 blk_finish_plug(&plug); 975 976 list_for_each_entry_safe(dc, tmp, wait_list, list) { 977 wait_for_completion_io(&dc->wait); 978 __remove_discard_cmd(sbi, dc); 979 } 980 981 mutex_unlock(&dcc->cmd_lock); 982 } 983 984 static int issue_discard_thread(void *data) 985 { 986 struct f2fs_sb_info *sbi = data; 987 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 988 wait_queue_head_t *q = &dcc->discard_wait_queue; 989 struct list_head *pend_list = &dcc->pend_list; 990 struct list_head *wait_list = &dcc->wait_list; 991 struct discard_cmd *dc, *tmp; 992 struct blk_plug plug; 993 int iter = 0; 994 repeat: 995 if (kthread_should_stop()) 996 return 0; 997 998 mutex_lock(&dcc->cmd_lock); 999 blk_start_plug(&plug); 1000 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1001 f2fs_bug_on(sbi, dc->state != D_PREP); 1002 1003 if (is_idle(sbi)) 1004 __submit_discard_cmd(sbi, dc); 1005 1006 if (iter++ > DISCARD_ISSUE_RATE) 1007 break; 1008 } 1009 blk_finish_plug(&plug); 1010 1011 list_for_each_entry_safe(dc, tmp, wait_list, list) { 1012 if (dc->state == D_DONE) { 1013 wait_for_completion_io(&dc->wait); 1014 __remove_discard_cmd(sbi, dc); 1015 } 1016 } 1017 mutex_unlock(&dcc->cmd_lock); 1018 1019 iter = 0; 1020 congestion_wait(BLK_RW_SYNC, HZ/50); 1021 1022 wait_event_interruptible(*q, kthread_should_stop() || 1023 !list_empty(pend_list) || !list_empty(wait_list)); 1024 goto repeat; 1025 } 1026 1027 #ifdef CONFIG_BLK_DEV_ZONED 1028 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1029 struct block_device *bdev, block_t blkstart, block_t blklen) 1030 { 1031 sector_t sector, nr_sects; 1032 block_t lblkstart = blkstart; 1033 int devi = 0; 1034 1035 if (sbi->s_ndevs) { 1036 devi = f2fs_target_device_index(sbi, blkstart); 1037 blkstart -= FDEV(devi).start_blk; 1038 } 1039 1040 /* 1041 * We need to know the type of the zone: for conventional zones, 1042 * use regular discard if the drive supports it. For sequential 1043 * zones, reset the zone write pointer. 1044 */ 1045 switch (get_blkz_type(sbi, bdev, blkstart)) { 1046 1047 case BLK_ZONE_TYPE_CONVENTIONAL: 1048 if (!blk_queue_discard(bdev_get_queue(bdev))) 1049 return 0; 1050 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1051 case BLK_ZONE_TYPE_SEQWRITE_REQ: 1052 case BLK_ZONE_TYPE_SEQWRITE_PREF: 1053 sector = SECTOR_FROM_BLOCK(blkstart); 1054 nr_sects = SECTOR_FROM_BLOCK(blklen); 1055 1056 if (sector & (bdev_zone_sectors(bdev) - 1) || 1057 nr_sects != bdev_zone_sectors(bdev)) { 1058 f2fs_msg(sbi->sb, KERN_INFO, 1059 "(%d) %s: Unaligned discard attempted (block %x + %x)", 1060 devi, sbi->s_ndevs ? FDEV(devi).path: "", 1061 blkstart, blklen); 1062 return -EIO; 1063 } 1064 trace_f2fs_issue_reset_zone(bdev, blkstart); 1065 return blkdev_reset_zones(bdev, sector, 1066 nr_sects, GFP_NOFS); 1067 default: 1068 /* Unknown zone type: broken device ? */ 1069 return -EIO; 1070 } 1071 } 1072 #endif 1073 1074 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1075 struct block_device *bdev, block_t blkstart, block_t blklen) 1076 { 1077 #ifdef CONFIG_BLK_DEV_ZONED 1078 if (f2fs_sb_mounted_blkzoned(sbi->sb) && 1079 bdev_zoned_model(bdev) != BLK_ZONED_NONE) 1080 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1081 #endif 1082 return __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1083 } 1084 1085 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1086 block_t blkstart, block_t blklen) 1087 { 1088 sector_t start = blkstart, len = 0; 1089 struct block_device *bdev; 1090 struct seg_entry *se; 1091 unsigned int offset; 1092 block_t i; 1093 int err = 0; 1094 1095 bdev = f2fs_target_device(sbi, blkstart, NULL); 1096 1097 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1098 if (i != start) { 1099 struct block_device *bdev2 = 1100 f2fs_target_device(sbi, i, NULL); 1101 1102 if (bdev2 != bdev) { 1103 err = __issue_discard_async(sbi, bdev, 1104 start, len); 1105 if (err) 1106 return err; 1107 bdev = bdev2; 1108 start = i; 1109 len = 0; 1110 } 1111 } 1112 1113 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 1114 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 1115 1116 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 1117 sbi->discard_blks--; 1118 } 1119 1120 if (len) 1121 err = __issue_discard_async(sbi, bdev, start, len); 1122 return err; 1123 } 1124 1125 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 1126 bool check_only) 1127 { 1128 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1129 int max_blocks = sbi->blocks_per_seg; 1130 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 1131 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1132 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1133 unsigned long *discard_map = (unsigned long *)se->discard_map; 1134 unsigned long *dmap = SIT_I(sbi)->tmp_map; 1135 unsigned int start = 0, end = -1; 1136 bool force = (cpc->reason == CP_DISCARD); 1137 struct discard_entry *de = NULL; 1138 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 1139 int i; 1140 1141 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi)) 1142 return false; 1143 1144 if (!force) { 1145 if (!test_opt(sbi, DISCARD) || !se->valid_blocks || 1146 SM_I(sbi)->dcc_info->nr_discards >= 1147 SM_I(sbi)->dcc_info->max_discards) 1148 return false; 1149 } 1150 1151 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 1152 for (i = 0; i < entries; i++) 1153 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 1154 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 1155 1156 while (force || SM_I(sbi)->dcc_info->nr_discards <= 1157 SM_I(sbi)->dcc_info->max_discards) { 1158 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 1159 if (start >= max_blocks) 1160 break; 1161 1162 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 1163 if (force && start && end != max_blocks 1164 && (end - start) < cpc->trim_minlen) 1165 continue; 1166 1167 if (check_only) 1168 return true; 1169 1170 if (!de) { 1171 de = f2fs_kmem_cache_alloc(discard_entry_slab, 1172 GFP_F2FS_ZERO); 1173 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 1174 list_add_tail(&de->list, head); 1175 } 1176 1177 for (i = start; i < end; i++) 1178 __set_bit_le(i, (void *)de->discard_map); 1179 1180 SM_I(sbi)->dcc_info->nr_discards += end - start; 1181 } 1182 return false; 1183 } 1184 1185 void release_discard_addrs(struct f2fs_sb_info *sbi) 1186 { 1187 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 1188 struct discard_entry *entry, *this; 1189 1190 /* drop caches */ 1191 list_for_each_entry_safe(entry, this, head, list) { 1192 list_del(&entry->list); 1193 kmem_cache_free(discard_entry_slab, entry); 1194 } 1195 } 1196 1197 /* 1198 * Should call clear_prefree_segments after checkpoint is done. 1199 */ 1200 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 1201 { 1202 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1203 unsigned int segno; 1204 1205 mutex_lock(&dirty_i->seglist_lock); 1206 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 1207 __set_test_and_free(sbi, segno); 1208 mutex_unlock(&dirty_i->seglist_lock); 1209 } 1210 1211 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1212 { 1213 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 1214 struct discard_entry *entry, *this; 1215 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1216 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 1217 unsigned int start = 0, end = -1; 1218 unsigned int secno, start_segno; 1219 bool force = (cpc->reason == CP_DISCARD); 1220 1221 mutex_lock(&dirty_i->seglist_lock); 1222 1223 while (1) { 1224 int i; 1225 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 1226 if (start >= MAIN_SEGS(sbi)) 1227 break; 1228 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 1229 start + 1); 1230 1231 for (i = start; i < end; i++) 1232 clear_bit(i, prefree_map); 1233 1234 dirty_i->nr_dirty[PRE] -= end - start; 1235 1236 if (!test_opt(sbi, DISCARD)) 1237 continue; 1238 1239 if (force && start >= cpc->trim_start && 1240 (end - 1) <= cpc->trim_end) 1241 continue; 1242 1243 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) { 1244 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 1245 (end - start) << sbi->log_blocks_per_seg); 1246 continue; 1247 } 1248 next: 1249 secno = GET_SEC_FROM_SEG(sbi, start); 1250 start_segno = GET_SEG_FROM_SEC(sbi, secno); 1251 if (!IS_CURSEC(sbi, secno) && 1252 !get_valid_blocks(sbi, start, true)) 1253 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 1254 sbi->segs_per_sec << sbi->log_blocks_per_seg); 1255 1256 start = start_segno + sbi->segs_per_sec; 1257 if (start < end) 1258 goto next; 1259 else 1260 end = start - 1; 1261 } 1262 mutex_unlock(&dirty_i->seglist_lock); 1263 1264 /* send small discards */ 1265 list_for_each_entry_safe(entry, this, head, list) { 1266 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 1267 bool is_valid = test_bit_le(0, entry->discard_map); 1268 1269 find_next: 1270 if (is_valid) { 1271 next_pos = find_next_zero_bit_le(entry->discard_map, 1272 sbi->blocks_per_seg, cur_pos); 1273 len = next_pos - cur_pos; 1274 1275 if (force && len < cpc->trim_minlen) 1276 goto skip; 1277 1278 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 1279 len); 1280 cpc->trimmed += len; 1281 total_len += len; 1282 } else { 1283 next_pos = find_next_bit_le(entry->discard_map, 1284 sbi->blocks_per_seg, cur_pos); 1285 } 1286 skip: 1287 cur_pos = next_pos; 1288 is_valid = !is_valid; 1289 1290 if (cur_pos < sbi->blocks_per_seg) 1291 goto find_next; 1292 1293 list_del(&entry->list); 1294 SM_I(sbi)->dcc_info->nr_discards -= total_len; 1295 kmem_cache_free(discard_entry_slab, entry); 1296 } 1297 } 1298 1299 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 1300 { 1301 dev_t dev = sbi->sb->s_bdev->bd_dev; 1302 struct discard_cmd_control *dcc; 1303 int err = 0; 1304 1305 if (SM_I(sbi)->dcc_info) { 1306 dcc = SM_I(sbi)->dcc_info; 1307 goto init_thread; 1308 } 1309 1310 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL); 1311 if (!dcc) 1312 return -ENOMEM; 1313 1314 INIT_LIST_HEAD(&dcc->entry_list); 1315 INIT_LIST_HEAD(&dcc->pend_list); 1316 INIT_LIST_HEAD(&dcc->wait_list); 1317 mutex_init(&dcc->cmd_lock); 1318 atomic_set(&dcc->issued_discard, 0); 1319 atomic_set(&dcc->issing_discard, 0); 1320 atomic_set(&dcc->discard_cmd_cnt, 0); 1321 dcc->nr_discards = 0; 1322 dcc->max_discards = 0; 1323 dcc->root = RB_ROOT; 1324 1325 init_waitqueue_head(&dcc->discard_wait_queue); 1326 SM_I(sbi)->dcc_info = dcc; 1327 init_thread: 1328 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 1329 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 1330 if (IS_ERR(dcc->f2fs_issue_discard)) { 1331 err = PTR_ERR(dcc->f2fs_issue_discard); 1332 kfree(dcc); 1333 SM_I(sbi)->dcc_info = NULL; 1334 return err; 1335 } 1336 1337 return err; 1338 } 1339 1340 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 1341 { 1342 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1343 1344 if (!dcc) 1345 return; 1346 1347 if (dcc->f2fs_issue_discard) { 1348 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1349 1350 dcc->f2fs_issue_discard = NULL; 1351 kthread_stop(discard_thread); 1352 } 1353 1354 kfree(dcc); 1355 SM_I(sbi)->dcc_info = NULL; 1356 } 1357 1358 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 1359 { 1360 struct sit_info *sit_i = SIT_I(sbi); 1361 1362 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 1363 sit_i->dirty_sentries++; 1364 return false; 1365 } 1366 1367 return true; 1368 } 1369 1370 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 1371 unsigned int segno, int modified) 1372 { 1373 struct seg_entry *se = get_seg_entry(sbi, segno); 1374 se->type = type; 1375 if (modified) 1376 __mark_sit_entry_dirty(sbi, segno); 1377 } 1378 1379 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 1380 { 1381 struct seg_entry *se; 1382 unsigned int segno, offset; 1383 long int new_vblocks; 1384 1385 segno = GET_SEGNO(sbi, blkaddr); 1386 1387 se = get_seg_entry(sbi, segno); 1388 new_vblocks = se->valid_blocks + del; 1389 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1390 1391 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || 1392 (new_vblocks > sbi->blocks_per_seg))); 1393 1394 se->valid_blocks = new_vblocks; 1395 se->mtime = get_mtime(sbi); 1396 SIT_I(sbi)->max_mtime = se->mtime; 1397 1398 /* Update valid block bitmap */ 1399 if (del > 0) { 1400 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) { 1401 #ifdef CONFIG_F2FS_CHECK_FS 1402 if (f2fs_test_and_set_bit(offset, 1403 se->cur_valid_map_mir)) 1404 f2fs_bug_on(sbi, 1); 1405 else 1406 WARN_ON(1); 1407 #else 1408 f2fs_bug_on(sbi, 1); 1409 #endif 1410 } 1411 if (f2fs_discard_en(sbi) && 1412 !f2fs_test_and_set_bit(offset, se->discard_map)) 1413 sbi->discard_blks--; 1414 1415 /* don't overwrite by SSR to keep node chain */ 1416 if (se->type == CURSEG_WARM_NODE) { 1417 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 1418 se->ckpt_valid_blocks++; 1419 } 1420 } else { 1421 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) { 1422 #ifdef CONFIG_F2FS_CHECK_FS 1423 if (!f2fs_test_and_clear_bit(offset, 1424 se->cur_valid_map_mir)) 1425 f2fs_bug_on(sbi, 1); 1426 else 1427 WARN_ON(1); 1428 #else 1429 f2fs_bug_on(sbi, 1); 1430 #endif 1431 } 1432 if (f2fs_discard_en(sbi) && 1433 f2fs_test_and_clear_bit(offset, se->discard_map)) 1434 sbi->discard_blks++; 1435 } 1436 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 1437 se->ckpt_valid_blocks += del; 1438 1439 __mark_sit_entry_dirty(sbi, segno); 1440 1441 /* update total number of valid blocks to be written in ckpt area */ 1442 SIT_I(sbi)->written_valid_blocks += del; 1443 1444 if (sbi->segs_per_sec > 1) 1445 get_sec_entry(sbi, segno)->valid_blocks += del; 1446 } 1447 1448 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) 1449 { 1450 update_sit_entry(sbi, new, 1); 1451 if (GET_SEGNO(sbi, old) != NULL_SEGNO) 1452 update_sit_entry(sbi, old, -1); 1453 1454 locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); 1455 locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); 1456 } 1457 1458 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 1459 { 1460 unsigned int segno = GET_SEGNO(sbi, addr); 1461 struct sit_info *sit_i = SIT_I(sbi); 1462 1463 f2fs_bug_on(sbi, addr == NULL_ADDR); 1464 if (addr == NEW_ADDR) 1465 return; 1466 1467 /* add it into sit main buffer */ 1468 mutex_lock(&sit_i->sentry_lock); 1469 1470 update_sit_entry(sbi, addr, -1); 1471 1472 /* add it into dirty seglist */ 1473 locate_dirty_segment(sbi, segno); 1474 1475 mutex_unlock(&sit_i->sentry_lock); 1476 } 1477 1478 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 1479 { 1480 struct sit_info *sit_i = SIT_I(sbi); 1481 unsigned int segno, offset; 1482 struct seg_entry *se; 1483 bool is_cp = false; 1484 1485 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) 1486 return true; 1487 1488 mutex_lock(&sit_i->sentry_lock); 1489 1490 segno = GET_SEGNO(sbi, blkaddr); 1491 se = get_seg_entry(sbi, segno); 1492 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1493 1494 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 1495 is_cp = true; 1496 1497 mutex_unlock(&sit_i->sentry_lock); 1498 1499 return is_cp; 1500 } 1501 1502 /* 1503 * This function should be resided under the curseg_mutex lock 1504 */ 1505 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 1506 struct f2fs_summary *sum) 1507 { 1508 struct curseg_info *curseg = CURSEG_I(sbi, type); 1509 void *addr = curseg->sum_blk; 1510 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 1511 memcpy(addr, sum, sizeof(struct f2fs_summary)); 1512 } 1513 1514 /* 1515 * Calculate the number of current summary pages for writing 1516 */ 1517 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 1518 { 1519 int valid_sum_count = 0; 1520 int i, sum_in_page; 1521 1522 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1523 if (sbi->ckpt->alloc_type[i] == SSR) 1524 valid_sum_count += sbi->blocks_per_seg; 1525 else { 1526 if (for_ra) 1527 valid_sum_count += le16_to_cpu( 1528 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 1529 else 1530 valid_sum_count += curseg_blkoff(sbi, i); 1531 } 1532 } 1533 1534 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 1535 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 1536 if (valid_sum_count <= sum_in_page) 1537 return 1; 1538 else if ((valid_sum_count - sum_in_page) <= 1539 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 1540 return 2; 1541 return 3; 1542 } 1543 1544 /* 1545 * Caller should put this summary page 1546 */ 1547 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 1548 { 1549 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); 1550 } 1551 1552 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr) 1553 { 1554 struct page *page = grab_meta_page(sbi, blk_addr); 1555 void *dst = page_address(page); 1556 1557 if (src) 1558 memcpy(dst, src, PAGE_SIZE); 1559 else 1560 memset(dst, 0, PAGE_SIZE); 1561 set_page_dirty(page); 1562 f2fs_put_page(page, 1); 1563 } 1564 1565 static void write_sum_page(struct f2fs_sb_info *sbi, 1566 struct f2fs_summary_block *sum_blk, block_t blk_addr) 1567 { 1568 update_meta_page(sbi, (void *)sum_blk, blk_addr); 1569 } 1570 1571 static void write_current_sum_page(struct f2fs_sb_info *sbi, 1572 int type, block_t blk_addr) 1573 { 1574 struct curseg_info *curseg = CURSEG_I(sbi, type); 1575 struct page *page = grab_meta_page(sbi, blk_addr); 1576 struct f2fs_summary_block *src = curseg->sum_blk; 1577 struct f2fs_summary_block *dst; 1578 1579 dst = (struct f2fs_summary_block *)page_address(page); 1580 1581 mutex_lock(&curseg->curseg_mutex); 1582 1583 down_read(&curseg->journal_rwsem); 1584 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 1585 up_read(&curseg->journal_rwsem); 1586 1587 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 1588 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 1589 1590 mutex_unlock(&curseg->curseg_mutex); 1591 1592 set_page_dirty(page); 1593 f2fs_put_page(page, 1); 1594 } 1595 1596 /* 1597 * Find a new segment from the free segments bitmap to right order 1598 * This function should be returned with success, otherwise BUG 1599 */ 1600 static void get_new_segment(struct f2fs_sb_info *sbi, 1601 unsigned int *newseg, bool new_sec, int dir) 1602 { 1603 struct free_segmap_info *free_i = FREE_I(sbi); 1604 unsigned int segno, secno, zoneno; 1605 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 1606 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 1607 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 1608 unsigned int left_start = hint; 1609 bool init = true; 1610 int go_left = 0; 1611 int i; 1612 1613 spin_lock(&free_i->segmap_lock); 1614 1615 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 1616 segno = find_next_zero_bit(free_i->free_segmap, 1617 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 1618 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 1619 goto got_it; 1620 } 1621 find_other_zone: 1622 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 1623 if (secno >= MAIN_SECS(sbi)) { 1624 if (dir == ALLOC_RIGHT) { 1625 secno = find_next_zero_bit(free_i->free_secmap, 1626 MAIN_SECS(sbi), 0); 1627 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 1628 } else { 1629 go_left = 1; 1630 left_start = hint - 1; 1631 } 1632 } 1633 if (go_left == 0) 1634 goto skip_left; 1635 1636 while (test_bit(left_start, free_i->free_secmap)) { 1637 if (left_start > 0) { 1638 left_start--; 1639 continue; 1640 } 1641 left_start = find_next_zero_bit(free_i->free_secmap, 1642 MAIN_SECS(sbi), 0); 1643 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 1644 break; 1645 } 1646 secno = left_start; 1647 skip_left: 1648 hint = secno; 1649 segno = GET_SEG_FROM_SEC(sbi, secno); 1650 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 1651 1652 /* give up on finding another zone */ 1653 if (!init) 1654 goto got_it; 1655 if (sbi->secs_per_zone == 1) 1656 goto got_it; 1657 if (zoneno == old_zoneno) 1658 goto got_it; 1659 if (dir == ALLOC_LEFT) { 1660 if (!go_left && zoneno + 1 >= total_zones) 1661 goto got_it; 1662 if (go_left && zoneno == 0) 1663 goto got_it; 1664 } 1665 for (i = 0; i < NR_CURSEG_TYPE; i++) 1666 if (CURSEG_I(sbi, i)->zone == zoneno) 1667 break; 1668 1669 if (i < NR_CURSEG_TYPE) { 1670 /* zone is in user, try another */ 1671 if (go_left) 1672 hint = zoneno * sbi->secs_per_zone - 1; 1673 else if (zoneno + 1 >= total_zones) 1674 hint = 0; 1675 else 1676 hint = (zoneno + 1) * sbi->secs_per_zone; 1677 init = false; 1678 goto find_other_zone; 1679 } 1680 got_it: 1681 /* set it as dirty segment in free segmap */ 1682 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 1683 __set_inuse(sbi, segno); 1684 *newseg = segno; 1685 spin_unlock(&free_i->segmap_lock); 1686 } 1687 1688 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 1689 { 1690 struct curseg_info *curseg = CURSEG_I(sbi, type); 1691 struct summary_footer *sum_footer; 1692 1693 curseg->segno = curseg->next_segno; 1694 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 1695 curseg->next_blkoff = 0; 1696 curseg->next_segno = NULL_SEGNO; 1697 1698 sum_footer = &(curseg->sum_blk->footer); 1699 memset(sum_footer, 0, sizeof(struct summary_footer)); 1700 if (IS_DATASEG(type)) 1701 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 1702 if (IS_NODESEG(type)) 1703 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 1704 __set_sit_entry_type(sbi, type, curseg->segno, modified); 1705 } 1706 1707 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 1708 { 1709 if (type == CURSEG_HOT_DATA || IS_NODESEG(type)) 1710 return 0; 1711 1712 return CURSEG_I(sbi, type)->segno; 1713 } 1714 1715 /* 1716 * Allocate a current working segment. 1717 * This function always allocates a free segment in LFS manner. 1718 */ 1719 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 1720 { 1721 struct curseg_info *curseg = CURSEG_I(sbi, type); 1722 unsigned int segno = curseg->segno; 1723 int dir = ALLOC_LEFT; 1724 1725 write_sum_page(sbi, curseg->sum_blk, 1726 GET_SUM_BLOCK(sbi, segno)); 1727 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 1728 dir = ALLOC_RIGHT; 1729 1730 if (test_opt(sbi, NOHEAP)) 1731 dir = ALLOC_RIGHT; 1732 1733 segno = __get_next_segno(sbi, type); 1734 get_new_segment(sbi, &segno, new_sec, dir); 1735 curseg->next_segno = segno; 1736 reset_curseg(sbi, type, 1); 1737 curseg->alloc_type = LFS; 1738 } 1739 1740 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 1741 struct curseg_info *seg, block_t start) 1742 { 1743 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 1744 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1745 unsigned long *target_map = SIT_I(sbi)->tmp_map; 1746 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1747 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1748 int i, pos; 1749 1750 for (i = 0; i < entries; i++) 1751 target_map[i] = ckpt_map[i] | cur_map[i]; 1752 1753 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 1754 1755 seg->next_blkoff = pos; 1756 } 1757 1758 /* 1759 * If a segment is written by LFS manner, next block offset is just obtained 1760 * by increasing the current block offset. However, if a segment is written by 1761 * SSR manner, next block offset obtained by calling __next_free_blkoff 1762 */ 1763 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 1764 struct curseg_info *seg) 1765 { 1766 if (seg->alloc_type == SSR) 1767 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 1768 else 1769 seg->next_blkoff++; 1770 } 1771 1772 /* 1773 * This function always allocates a used segment(from dirty seglist) by SSR 1774 * manner, so it should recover the existing segment information of valid blocks 1775 */ 1776 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) 1777 { 1778 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1779 struct curseg_info *curseg = CURSEG_I(sbi, type); 1780 unsigned int new_segno = curseg->next_segno; 1781 struct f2fs_summary_block *sum_node; 1782 struct page *sum_page; 1783 1784 write_sum_page(sbi, curseg->sum_blk, 1785 GET_SUM_BLOCK(sbi, curseg->segno)); 1786 __set_test_and_inuse(sbi, new_segno); 1787 1788 mutex_lock(&dirty_i->seglist_lock); 1789 __remove_dirty_segment(sbi, new_segno, PRE); 1790 __remove_dirty_segment(sbi, new_segno, DIRTY); 1791 mutex_unlock(&dirty_i->seglist_lock); 1792 1793 reset_curseg(sbi, type, 1); 1794 curseg->alloc_type = SSR; 1795 __next_free_blkoff(sbi, curseg, 0); 1796 1797 if (reuse) { 1798 sum_page = get_sum_page(sbi, new_segno); 1799 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 1800 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 1801 f2fs_put_page(sum_page, 1); 1802 } 1803 } 1804 1805 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 1806 { 1807 struct curseg_info *curseg = CURSEG_I(sbi, type); 1808 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 1809 int i, cnt; 1810 bool reversed = false; 1811 1812 /* need_SSR() already forces to do this */ 1813 if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR)) 1814 return 1; 1815 1816 /* For node segments, let's do SSR more intensively */ 1817 if (IS_NODESEG(type)) { 1818 if (type >= CURSEG_WARM_NODE) { 1819 reversed = true; 1820 i = CURSEG_COLD_NODE; 1821 } else { 1822 i = CURSEG_HOT_NODE; 1823 } 1824 cnt = NR_CURSEG_NODE_TYPE; 1825 } else { 1826 if (type >= CURSEG_WARM_DATA) { 1827 reversed = true; 1828 i = CURSEG_COLD_DATA; 1829 } else { 1830 i = CURSEG_HOT_DATA; 1831 } 1832 cnt = NR_CURSEG_DATA_TYPE; 1833 } 1834 1835 for (; cnt-- > 0; reversed ? i-- : i++) { 1836 if (i == type) 1837 continue; 1838 if (v_ops->get_victim(sbi, &(curseg)->next_segno, 1839 BG_GC, i, SSR)) 1840 return 1; 1841 } 1842 return 0; 1843 } 1844 1845 /* 1846 * flush out current segment and replace it with new segment 1847 * This function should be returned with success, otherwise BUG 1848 */ 1849 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 1850 int type, bool force) 1851 { 1852 if (force) 1853 new_curseg(sbi, type, true); 1854 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 1855 type == CURSEG_WARM_NODE) 1856 new_curseg(sbi, type, false); 1857 else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) 1858 change_curseg(sbi, type, true); 1859 else 1860 new_curseg(sbi, type, false); 1861 1862 stat_inc_seg_type(sbi, CURSEG_I(sbi, type)); 1863 } 1864 1865 void allocate_new_segments(struct f2fs_sb_info *sbi) 1866 { 1867 struct curseg_info *curseg; 1868 unsigned int old_segno; 1869 int i; 1870 1871 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1872 curseg = CURSEG_I(sbi, i); 1873 old_segno = curseg->segno; 1874 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); 1875 locate_dirty_segment(sbi, old_segno); 1876 } 1877 } 1878 1879 static const struct segment_allocation default_salloc_ops = { 1880 .allocate_segment = allocate_segment_by_default, 1881 }; 1882 1883 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1884 { 1885 __u64 trim_start = cpc->trim_start; 1886 bool has_candidate = false; 1887 1888 mutex_lock(&SIT_I(sbi)->sentry_lock); 1889 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 1890 if (add_discard_addrs(sbi, cpc, true)) { 1891 has_candidate = true; 1892 break; 1893 } 1894 } 1895 mutex_unlock(&SIT_I(sbi)->sentry_lock); 1896 1897 cpc->trim_start = trim_start; 1898 return has_candidate; 1899 } 1900 1901 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 1902 { 1903 __u64 start = F2FS_BYTES_TO_BLK(range->start); 1904 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 1905 unsigned int start_segno, end_segno; 1906 struct cp_control cpc; 1907 int err = 0; 1908 1909 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 1910 return -EINVAL; 1911 1912 cpc.trimmed = 0; 1913 if (end <= MAIN_BLKADDR(sbi)) 1914 goto out; 1915 1916 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1917 f2fs_msg(sbi->sb, KERN_WARNING, 1918 "Found FS corruption, run fsck to fix."); 1919 goto out; 1920 } 1921 1922 /* start/end segment number in main_area */ 1923 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 1924 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 1925 GET_SEGNO(sbi, end); 1926 cpc.reason = CP_DISCARD; 1927 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 1928 1929 /* do checkpoint to issue discard commands safely */ 1930 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) { 1931 cpc.trim_start = start_segno; 1932 1933 if (sbi->discard_blks == 0) 1934 break; 1935 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi)) 1936 cpc.trim_end = end_segno; 1937 else 1938 cpc.trim_end = min_t(unsigned int, 1939 rounddown(start_segno + 1940 BATCHED_TRIM_SEGMENTS(sbi), 1941 sbi->segs_per_sec) - 1, end_segno); 1942 1943 mutex_lock(&sbi->gc_mutex); 1944 err = write_checkpoint(sbi, &cpc); 1945 mutex_unlock(&sbi->gc_mutex); 1946 if (err) 1947 break; 1948 1949 schedule(); 1950 } 1951 out: 1952 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed); 1953 return err; 1954 } 1955 1956 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 1957 { 1958 struct curseg_info *curseg = CURSEG_I(sbi, type); 1959 if (curseg->next_blkoff < sbi->blocks_per_seg) 1960 return true; 1961 return false; 1962 } 1963 1964 static int __get_segment_type_2(struct page *page, enum page_type p_type) 1965 { 1966 if (p_type == DATA) 1967 return CURSEG_HOT_DATA; 1968 else 1969 return CURSEG_HOT_NODE; 1970 } 1971 1972 static int __get_segment_type_4(struct page *page, enum page_type p_type) 1973 { 1974 if (p_type == DATA) { 1975 struct inode *inode = page->mapping->host; 1976 1977 if (S_ISDIR(inode->i_mode)) 1978 return CURSEG_HOT_DATA; 1979 else 1980 return CURSEG_COLD_DATA; 1981 } else { 1982 if (IS_DNODE(page) && is_cold_node(page)) 1983 return CURSEG_WARM_NODE; 1984 else 1985 return CURSEG_COLD_NODE; 1986 } 1987 } 1988 1989 static int __get_segment_type_6(struct page *page, enum page_type p_type) 1990 { 1991 if (p_type == DATA) { 1992 struct inode *inode = page->mapping->host; 1993 1994 if (is_cold_data(page) || file_is_cold(inode)) 1995 return CURSEG_COLD_DATA; 1996 if (is_inode_flag_set(inode, FI_HOT_DATA)) 1997 return CURSEG_HOT_DATA; 1998 return CURSEG_WARM_DATA; 1999 } else { 2000 if (IS_DNODE(page)) 2001 return is_cold_node(page) ? CURSEG_WARM_NODE : 2002 CURSEG_HOT_NODE; 2003 return CURSEG_COLD_NODE; 2004 } 2005 } 2006 2007 static int __get_segment_type(struct page *page, enum page_type p_type) 2008 { 2009 switch (F2FS_P_SB(page)->active_logs) { 2010 case 2: 2011 return __get_segment_type_2(page, p_type); 2012 case 4: 2013 return __get_segment_type_4(page, p_type); 2014 } 2015 /* NR_CURSEG_TYPE(6) logs by default */ 2016 f2fs_bug_on(F2FS_P_SB(page), 2017 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE); 2018 return __get_segment_type_6(page, p_type); 2019 } 2020 2021 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 2022 block_t old_blkaddr, block_t *new_blkaddr, 2023 struct f2fs_summary *sum, int type) 2024 { 2025 struct sit_info *sit_i = SIT_I(sbi); 2026 struct curseg_info *curseg = CURSEG_I(sbi, type); 2027 2028 mutex_lock(&curseg->curseg_mutex); 2029 mutex_lock(&sit_i->sentry_lock); 2030 2031 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 2032 2033 f2fs_wait_discard_bio(sbi, *new_blkaddr); 2034 2035 /* 2036 * __add_sum_entry should be resided under the curseg_mutex 2037 * because, this function updates a summary entry in the 2038 * current summary block. 2039 */ 2040 __add_sum_entry(sbi, type, sum); 2041 2042 __refresh_next_blkoff(sbi, curseg); 2043 2044 stat_inc_block_count(sbi, curseg); 2045 2046 if (!__has_curseg_space(sbi, type)) 2047 sit_i->s_ops->allocate_segment(sbi, type, false); 2048 /* 2049 * SIT information should be updated after segment allocation, 2050 * since we need to keep dirty segments precisely under SSR. 2051 */ 2052 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); 2053 2054 mutex_unlock(&sit_i->sentry_lock); 2055 2056 if (page && IS_NODESEG(type)) 2057 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 2058 2059 mutex_unlock(&curseg->curseg_mutex); 2060 } 2061 2062 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 2063 { 2064 int type = __get_segment_type(fio->page, fio->type); 2065 int err; 2066 2067 if (fio->type == NODE || fio->type == DATA) 2068 mutex_lock(&fio->sbi->wio_mutex[fio->type]); 2069 reallocate: 2070 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 2071 &fio->new_blkaddr, sum, type); 2072 2073 /* writeout dirty page into bdev */ 2074 err = f2fs_submit_page_mbio(fio); 2075 if (err == -EAGAIN) { 2076 fio->old_blkaddr = fio->new_blkaddr; 2077 goto reallocate; 2078 } 2079 2080 if (fio->type == NODE || fio->type == DATA) 2081 mutex_unlock(&fio->sbi->wio_mutex[fio->type]); 2082 } 2083 2084 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) 2085 { 2086 struct f2fs_io_info fio = { 2087 .sbi = sbi, 2088 .type = META, 2089 .op = REQ_OP_WRITE, 2090 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 2091 .old_blkaddr = page->index, 2092 .new_blkaddr = page->index, 2093 .page = page, 2094 .encrypted_page = NULL, 2095 }; 2096 2097 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 2098 fio.op_flags &= ~REQ_META; 2099 2100 set_page_writeback(page); 2101 f2fs_submit_page_mbio(&fio); 2102 } 2103 2104 void write_node_page(unsigned int nid, struct f2fs_io_info *fio) 2105 { 2106 struct f2fs_summary sum; 2107 2108 set_summary(&sum, nid, 0, 0); 2109 do_write_page(&sum, fio); 2110 } 2111 2112 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio) 2113 { 2114 struct f2fs_sb_info *sbi = fio->sbi; 2115 struct f2fs_summary sum; 2116 struct node_info ni; 2117 2118 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 2119 get_node_info(sbi, dn->nid, &ni); 2120 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 2121 do_write_page(&sum, fio); 2122 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 2123 } 2124 2125 int rewrite_data_page(struct f2fs_io_info *fio) 2126 { 2127 fio->new_blkaddr = fio->old_blkaddr; 2128 stat_inc_inplace_blocks(fio->sbi); 2129 return f2fs_submit_page_bio(fio); 2130 } 2131 2132 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 2133 block_t old_blkaddr, block_t new_blkaddr, 2134 bool recover_curseg, bool recover_newaddr) 2135 { 2136 struct sit_info *sit_i = SIT_I(sbi); 2137 struct curseg_info *curseg; 2138 unsigned int segno, old_cursegno; 2139 struct seg_entry *se; 2140 int type; 2141 unsigned short old_blkoff; 2142 2143 segno = GET_SEGNO(sbi, new_blkaddr); 2144 se = get_seg_entry(sbi, segno); 2145 type = se->type; 2146 2147 if (!recover_curseg) { 2148 /* for recovery flow */ 2149 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 2150 if (old_blkaddr == NULL_ADDR) 2151 type = CURSEG_COLD_DATA; 2152 else 2153 type = CURSEG_WARM_DATA; 2154 } 2155 } else { 2156 if (!IS_CURSEG(sbi, segno)) 2157 type = CURSEG_WARM_DATA; 2158 } 2159 2160 curseg = CURSEG_I(sbi, type); 2161 2162 mutex_lock(&curseg->curseg_mutex); 2163 mutex_lock(&sit_i->sentry_lock); 2164 2165 old_cursegno = curseg->segno; 2166 old_blkoff = curseg->next_blkoff; 2167 2168 /* change the current segment */ 2169 if (segno != curseg->segno) { 2170 curseg->next_segno = segno; 2171 change_curseg(sbi, type, true); 2172 } 2173 2174 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 2175 __add_sum_entry(sbi, type, sum); 2176 2177 if (!recover_curseg || recover_newaddr) 2178 update_sit_entry(sbi, new_blkaddr, 1); 2179 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 2180 update_sit_entry(sbi, old_blkaddr, -1); 2181 2182 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 2183 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 2184 2185 locate_dirty_segment(sbi, old_cursegno); 2186 2187 if (recover_curseg) { 2188 if (old_cursegno != curseg->segno) { 2189 curseg->next_segno = old_cursegno; 2190 change_curseg(sbi, type, true); 2191 } 2192 curseg->next_blkoff = old_blkoff; 2193 } 2194 2195 mutex_unlock(&sit_i->sentry_lock); 2196 mutex_unlock(&curseg->curseg_mutex); 2197 } 2198 2199 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 2200 block_t old_addr, block_t new_addr, 2201 unsigned char version, bool recover_curseg, 2202 bool recover_newaddr) 2203 { 2204 struct f2fs_summary sum; 2205 2206 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 2207 2208 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, 2209 recover_curseg, recover_newaddr); 2210 2211 f2fs_update_data_blkaddr(dn, new_addr); 2212 } 2213 2214 void f2fs_wait_on_page_writeback(struct page *page, 2215 enum page_type type, bool ordered) 2216 { 2217 if (PageWriteback(page)) { 2218 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 2219 2220 f2fs_submit_merged_bio_cond(sbi, page->mapping->host, 2221 0, page->index, type, WRITE); 2222 if (ordered) 2223 wait_on_page_writeback(page); 2224 else 2225 wait_for_stable_page(page); 2226 } 2227 } 2228 2229 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi, 2230 block_t blkaddr) 2231 { 2232 struct page *cpage; 2233 2234 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) 2235 return; 2236 2237 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 2238 if (cpage) { 2239 f2fs_wait_on_page_writeback(cpage, DATA, true); 2240 f2fs_put_page(cpage, 1); 2241 } 2242 } 2243 2244 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 2245 { 2246 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2247 struct curseg_info *seg_i; 2248 unsigned char *kaddr; 2249 struct page *page; 2250 block_t start; 2251 int i, j, offset; 2252 2253 start = start_sum_block(sbi); 2254 2255 page = get_meta_page(sbi, start++); 2256 kaddr = (unsigned char *)page_address(page); 2257 2258 /* Step 1: restore nat cache */ 2259 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 2260 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 2261 2262 /* Step 2: restore sit cache */ 2263 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 2264 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 2265 offset = 2 * SUM_JOURNAL_SIZE; 2266 2267 /* Step 3: restore summary entries */ 2268 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2269 unsigned short blk_off; 2270 unsigned int segno; 2271 2272 seg_i = CURSEG_I(sbi, i); 2273 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 2274 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 2275 seg_i->next_segno = segno; 2276 reset_curseg(sbi, i, 0); 2277 seg_i->alloc_type = ckpt->alloc_type[i]; 2278 seg_i->next_blkoff = blk_off; 2279 2280 if (seg_i->alloc_type == SSR) 2281 blk_off = sbi->blocks_per_seg; 2282 2283 for (j = 0; j < blk_off; j++) { 2284 struct f2fs_summary *s; 2285 s = (struct f2fs_summary *)(kaddr + offset); 2286 seg_i->sum_blk->entries[j] = *s; 2287 offset += SUMMARY_SIZE; 2288 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 2289 SUM_FOOTER_SIZE) 2290 continue; 2291 2292 f2fs_put_page(page, 1); 2293 page = NULL; 2294 2295 page = get_meta_page(sbi, start++); 2296 kaddr = (unsigned char *)page_address(page); 2297 offset = 0; 2298 } 2299 } 2300 f2fs_put_page(page, 1); 2301 return 0; 2302 } 2303 2304 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 2305 { 2306 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2307 struct f2fs_summary_block *sum; 2308 struct curseg_info *curseg; 2309 struct page *new; 2310 unsigned short blk_off; 2311 unsigned int segno = 0; 2312 block_t blk_addr = 0; 2313 2314 /* get segment number and block addr */ 2315 if (IS_DATASEG(type)) { 2316 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 2317 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 2318 CURSEG_HOT_DATA]); 2319 if (__exist_node_summaries(sbi)) 2320 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 2321 else 2322 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 2323 } else { 2324 segno = le32_to_cpu(ckpt->cur_node_segno[type - 2325 CURSEG_HOT_NODE]); 2326 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 2327 CURSEG_HOT_NODE]); 2328 if (__exist_node_summaries(sbi)) 2329 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 2330 type - CURSEG_HOT_NODE); 2331 else 2332 blk_addr = GET_SUM_BLOCK(sbi, segno); 2333 } 2334 2335 new = get_meta_page(sbi, blk_addr); 2336 sum = (struct f2fs_summary_block *)page_address(new); 2337 2338 if (IS_NODESEG(type)) { 2339 if (__exist_node_summaries(sbi)) { 2340 struct f2fs_summary *ns = &sum->entries[0]; 2341 int i; 2342 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 2343 ns->version = 0; 2344 ns->ofs_in_node = 0; 2345 } 2346 } else { 2347 int err; 2348 2349 err = restore_node_summary(sbi, segno, sum); 2350 if (err) { 2351 f2fs_put_page(new, 1); 2352 return err; 2353 } 2354 } 2355 } 2356 2357 /* set uncompleted segment to curseg */ 2358 curseg = CURSEG_I(sbi, type); 2359 mutex_lock(&curseg->curseg_mutex); 2360 2361 /* update journal info */ 2362 down_write(&curseg->journal_rwsem); 2363 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 2364 up_write(&curseg->journal_rwsem); 2365 2366 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 2367 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 2368 curseg->next_segno = segno; 2369 reset_curseg(sbi, type, 0); 2370 curseg->alloc_type = ckpt->alloc_type[type]; 2371 curseg->next_blkoff = blk_off; 2372 mutex_unlock(&curseg->curseg_mutex); 2373 f2fs_put_page(new, 1); 2374 return 0; 2375 } 2376 2377 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 2378 { 2379 int type = CURSEG_HOT_DATA; 2380 int err; 2381 2382 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 2383 int npages = npages_for_summary_flush(sbi, true); 2384 2385 if (npages >= 2) 2386 ra_meta_pages(sbi, start_sum_block(sbi), npages, 2387 META_CP, true); 2388 2389 /* restore for compacted data summary */ 2390 if (read_compacted_summaries(sbi)) 2391 return -EINVAL; 2392 type = CURSEG_HOT_NODE; 2393 } 2394 2395 if (__exist_node_summaries(sbi)) 2396 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), 2397 NR_CURSEG_TYPE - type, META_CP, true); 2398 2399 for (; type <= CURSEG_COLD_NODE; type++) { 2400 err = read_normal_summaries(sbi, type); 2401 if (err) 2402 return err; 2403 } 2404 2405 return 0; 2406 } 2407 2408 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 2409 { 2410 struct page *page; 2411 unsigned char *kaddr; 2412 struct f2fs_summary *summary; 2413 struct curseg_info *seg_i; 2414 int written_size = 0; 2415 int i, j; 2416 2417 page = grab_meta_page(sbi, blkaddr++); 2418 kaddr = (unsigned char *)page_address(page); 2419 2420 /* Step 1: write nat cache */ 2421 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 2422 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 2423 written_size += SUM_JOURNAL_SIZE; 2424 2425 /* Step 2: write sit cache */ 2426 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 2427 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 2428 written_size += SUM_JOURNAL_SIZE; 2429 2430 /* Step 3: write summary entries */ 2431 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2432 unsigned short blkoff; 2433 seg_i = CURSEG_I(sbi, i); 2434 if (sbi->ckpt->alloc_type[i] == SSR) 2435 blkoff = sbi->blocks_per_seg; 2436 else 2437 blkoff = curseg_blkoff(sbi, i); 2438 2439 for (j = 0; j < blkoff; j++) { 2440 if (!page) { 2441 page = grab_meta_page(sbi, blkaddr++); 2442 kaddr = (unsigned char *)page_address(page); 2443 written_size = 0; 2444 } 2445 summary = (struct f2fs_summary *)(kaddr + written_size); 2446 *summary = seg_i->sum_blk->entries[j]; 2447 written_size += SUMMARY_SIZE; 2448 2449 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 2450 SUM_FOOTER_SIZE) 2451 continue; 2452 2453 set_page_dirty(page); 2454 f2fs_put_page(page, 1); 2455 page = NULL; 2456 } 2457 } 2458 if (page) { 2459 set_page_dirty(page); 2460 f2fs_put_page(page, 1); 2461 } 2462 } 2463 2464 static void write_normal_summaries(struct f2fs_sb_info *sbi, 2465 block_t blkaddr, int type) 2466 { 2467 int i, end; 2468 if (IS_DATASEG(type)) 2469 end = type + NR_CURSEG_DATA_TYPE; 2470 else 2471 end = type + NR_CURSEG_NODE_TYPE; 2472 2473 for (i = type; i < end; i++) 2474 write_current_sum_page(sbi, i, blkaddr + (i - type)); 2475 } 2476 2477 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 2478 { 2479 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 2480 write_compacted_summaries(sbi, start_blk); 2481 else 2482 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 2483 } 2484 2485 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 2486 { 2487 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 2488 } 2489 2490 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 2491 unsigned int val, int alloc) 2492 { 2493 int i; 2494 2495 if (type == NAT_JOURNAL) { 2496 for (i = 0; i < nats_in_cursum(journal); i++) { 2497 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 2498 return i; 2499 } 2500 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 2501 return update_nats_in_cursum(journal, 1); 2502 } else if (type == SIT_JOURNAL) { 2503 for (i = 0; i < sits_in_cursum(journal); i++) 2504 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 2505 return i; 2506 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 2507 return update_sits_in_cursum(journal, 1); 2508 } 2509 return -1; 2510 } 2511 2512 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 2513 unsigned int segno) 2514 { 2515 return get_meta_page(sbi, current_sit_addr(sbi, segno)); 2516 } 2517 2518 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 2519 unsigned int start) 2520 { 2521 struct sit_info *sit_i = SIT_I(sbi); 2522 struct page *src_page, *dst_page; 2523 pgoff_t src_off, dst_off; 2524 void *src_addr, *dst_addr; 2525 2526 src_off = current_sit_addr(sbi, start); 2527 dst_off = next_sit_addr(sbi, src_off); 2528 2529 /* get current sit block page without lock */ 2530 src_page = get_meta_page(sbi, src_off); 2531 dst_page = grab_meta_page(sbi, dst_off); 2532 f2fs_bug_on(sbi, PageDirty(src_page)); 2533 2534 src_addr = page_address(src_page); 2535 dst_addr = page_address(dst_page); 2536 memcpy(dst_addr, src_addr, PAGE_SIZE); 2537 2538 set_page_dirty(dst_page); 2539 f2fs_put_page(src_page, 1); 2540 2541 set_to_next_sit(sit_i, start); 2542 2543 return dst_page; 2544 } 2545 2546 static struct sit_entry_set *grab_sit_entry_set(void) 2547 { 2548 struct sit_entry_set *ses = 2549 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); 2550 2551 ses->entry_cnt = 0; 2552 INIT_LIST_HEAD(&ses->set_list); 2553 return ses; 2554 } 2555 2556 static void release_sit_entry_set(struct sit_entry_set *ses) 2557 { 2558 list_del(&ses->set_list); 2559 kmem_cache_free(sit_entry_set_slab, ses); 2560 } 2561 2562 static void adjust_sit_entry_set(struct sit_entry_set *ses, 2563 struct list_head *head) 2564 { 2565 struct sit_entry_set *next = ses; 2566 2567 if (list_is_last(&ses->set_list, head)) 2568 return; 2569 2570 list_for_each_entry_continue(next, head, set_list) 2571 if (ses->entry_cnt <= next->entry_cnt) 2572 break; 2573 2574 list_move_tail(&ses->set_list, &next->set_list); 2575 } 2576 2577 static void add_sit_entry(unsigned int segno, struct list_head *head) 2578 { 2579 struct sit_entry_set *ses; 2580 unsigned int start_segno = START_SEGNO(segno); 2581 2582 list_for_each_entry(ses, head, set_list) { 2583 if (ses->start_segno == start_segno) { 2584 ses->entry_cnt++; 2585 adjust_sit_entry_set(ses, head); 2586 return; 2587 } 2588 } 2589 2590 ses = grab_sit_entry_set(); 2591 2592 ses->start_segno = start_segno; 2593 ses->entry_cnt++; 2594 list_add(&ses->set_list, head); 2595 } 2596 2597 static void add_sits_in_set(struct f2fs_sb_info *sbi) 2598 { 2599 struct f2fs_sm_info *sm_info = SM_I(sbi); 2600 struct list_head *set_list = &sm_info->sit_entry_set; 2601 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 2602 unsigned int segno; 2603 2604 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 2605 add_sit_entry(segno, set_list); 2606 } 2607 2608 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 2609 { 2610 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 2611 struct f2fs_journal *journal = curseg->journal; 2612 int i; 2613 2614 down_write(&curseg->journal_rwsem); 2615 for (i = 0; i < sits_in_cursum(journal); i++) { 2616 unsigned int segno; 2617 bool dirtied; 2618 2619 segno = le32_to_cpu(segno_in_journal(journal, i)); 2620 dirtied = __mark_sit_entry_dirty(sbi, segno); 2621 2622 if (!dirtied) 2623 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 2624 } 2625 update_sits_in_cursum(journal, -i); 2626 up_write(&curseg->journal_rwsem); 2627 } 2628 2629 /* 2630 * CP calls this function, which flushes SIT entries including sit_journal, 2631 * and moves prefree segs to free segs. 2632 */ 2633 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 2634 { 2635 struct sit_info *sit_i = SIT_I(sbi); 2636 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 2637 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 2638 struct f2fs_journal *journal = curseg->journal; 2639 struct sit_entry_set *ses, *tmp; 2640 struct list_head *head = &SM_I(sbi)->sit_entry_set; 2641 bool to_journal = true; 2642 struct seg_entry *se; 2643 2644 mutex_lock(&sit_i->sentry_lock); 2645 2646 if (!sit_i->dirty_sentries) 2647 goto out; 2648 2649 /* 2650 * add and account sit entries of dirty bitmap in sit entry 2651 * set temporarily 2652 */ 2653 add_sits_in_set(sbi); 2654 2655 /* 2656 * if there are no enough space in journal to store dirty sit 2657 * entries, remove all entries from journal and add and account 2658 * them in sit entry set. 2659 */ 2660 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL)) 2661 remove_sits_in_journal(sbi); 2662 2663 /* 2664 * there are two steps to flush sit entries: 2665 * #1, flush sit entries to journal in current cold data summary block. 2666 * #2, flush sit entries to sit page. 2667 */ 2668 list_for_each_entry_safe(ses, tmp, head, set_list) { 2669 struct page *page = NULL; 2670 struct f2fs_sit_block *raw_sit = NULL; 2671 unsigned int start_segno = ses->start_segno; 2672 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 2673 (unsigned long)MAIN_SEGS(sbi)); 2674 unsigned int segno = start_segno; 2675 2676 if (to_journal && 2677 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 2678 to_journal = false; 2679 2680 if (to_journal) { 2681 down_write(&curseg->journal_rwsem); 2682 } else { 2683 page = get_next_sit_page(sbi, start_segno); 2684 raw_sit = page_address(page); 2685 } 2686 2687 /* flush dirty sit entries in region of current sit set */ 2688 for_each_set_bit_from(segno, bitmap, end) { 2689 int offset, sit_offset; 2690 2691 se = get_seg_entry(sbi, segno); 2692 2693 /* add discard candidates */ 2694 if (cpc->reason != CP_DISCARD) { 2695 cpc->trim_start = segno; 2696 add_discard_addrs(sbi, cpc, false); 2697 } 2698 2699 if (to_journal) { 2700 offset = lookup_journal_in_cursum(journal, 2701 SIT_JOURNAL, segno, 1); 2702 f2fs_bug_on(sbi, offset < 0); 2703 segno_in_journal(journal, offset) = 2704 cpu_to_le32(segno); 2705 seg_info_to_raw_sit(se, 2706 &sit_in_journal(journal, offset)); 2707 } else { 2708 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 2709 seg_info_to_raw_sit(se, 2710 &raw_sit->entries[sit_offset]); 2711 } 2712 2713 __clear_bit(segno, bitmap); 2714 sit_i->dirty_sentries--; 2715 ses->entry_cnt--; 2716 } 2717 2718 if (to_journal) 2719 up_write(&curseg->journal_rwsem); 2720 else 2721 f2fs_put_page(page, 1); 2722 2723 f2fs_bug_on(sbi, ses->entry_cnt); 2724 release_sit_entry_set(ses); 2725 } 2726 2727 f2fs_bug_on(sbi, !list_empty(head)); 2728 f2fs_bug_on(sbi, sit_i->dirty_sentries); 2729 out: 2730 if (cpc->reason == CP_DISCARD) { 2731 __u64 trim_start = cpc->trim_start; 2732 2733 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 2734 add_discard_addrs(sbi, cpc, false); 2735 2736 cpc->trim_start = trim_start; 2737 } 2738 mutex_unlock(&sit_i->sentry_lock); 2739 2740 set_prefree_as_free_segments(sbi); 2741 } 2742 2743 static int build_sit_info(struct f2fs_sb_info *sbi) 2744 { 2745 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 2746 struct sit_info *sit_i; 2747 unsigned int sit_segs, start; 2748 char *src_bitmap; 2749 unsigned int bitmap_size; 2750 2751 /* allocate memory for SIT information */ 2752 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); 2753 if (!sit_i) 2754 return -ENOMEM; 2755 2756 SM_I(sbi)->sit_info = sit_i; 2757 2758 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) * 2759 sizeof(struct seg_entry), GFP_KERNEL); 2760 if (!sit_i->sentries) 2761 return -ENOMEM; 2762 2763 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 2764 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 2765 if (!sit_i->dirty_sentries_bitmap) 2766 return -ENOMEM; 2767 2768 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2769 sit_i->sentries[start].cur_valid_map 2770 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2771 sit_i->sentries[start].ckpt_valid_map 2772 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2773 if (!sit_i->sentries[start].cur_valid_map || 2774 !sit_i->sentries[start].ckpt_valid_map) 2775 return -ENOMEM; 2776 2777 #ifdef CONFIG_F2FS_CHECK_FS 2778 sit_i->sentries[start].cur_valid_map_mir 2779 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2780 if (!sit_i->sentries[start].cur_valid_map_mir) 2781 return -ENOMEM; 2782 #endif 2783 2784 if (f2fs_discard_en(sbi)) { 2785 sit_i->sentries[start].discard_map 2786 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2787 if (!sit_i->sentries[start].discard_map) 2788 return -ENOMEM; 2789 } 2790 } 2791 2792 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2793 if (!sit_i->tmp_map) 2794 return -ENOMEM; 2795 2796 if (sbi->segs_per_sec > 1) { 2797 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) * 2798 sizeof(struct sec_entry), GFP_KERNEL); 2799 if (!sit_i->sec_entries) 2800 return -ENOMEM; 2801 } 2802 2803 /* get information related with SIT */ 2804 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 2805 2806 /* setup SIT bitmap from ckeckpoint pack */ 2807 bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 2808 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 2809 2810 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 2811 if (!sit_i->sit_bitmap) 2812 return -ENOMEM; 2813 2814 #ifdef CONFIG_F2FS_CHECK_FS 2815 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 2816 if (!sit_i->sit_bitmap_mir) 2817 return -ENOMEM; 2818 #endif 2819 2820 /* init SIT information */ 2821 sit_i->s_ops = &default_salloc_ops; 2822 2823 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 2824 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 2825 sit_i->written_valid_blocks = 0; 2826 sit_i->bitmap_size = bitmap_size; 2827 sit_i->dirty_sentries = 0; 2828 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 2829 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 2830 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; 2831 mutex_init(&sit_i->sentry_lock); 2832 return 0; 2833 } 2834 2835 static int build_free_segmap(struct f2fs_sb_info *sbi) 2836 { 2837 struct free_segmap_info *free_i; 2838 unsigned int bitmap_size, sec_bitmap_size; 2839 2840 /* allocate memory for free segmap information */ 2841 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); 2842 if (!free_i) 2843 return -ENOMEM; 2844 2845 SM_I(sbi)->free_info = free_i; 2846 2847 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 2848 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL); 2849 if (!free_i->free_segmap) 2850 return -ENOMEM; 2851 2852 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 2853 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL); 2854 if (!free_i->free_secmap) 2855 return -ENOMEM; 2856 2857 /* set all segments as dirty temporarily */ 2858 memset(free_i->free_segmap, 0xff, bitmap_size); 2859 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 2860 2861 /* init free segmap information */ 2862 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 2863 free_i->free_segments = 0; 2864 free_i->free_sections = 0; 2865 spin_lock_init(&free_i->segmap_lock); 2866 return 0; 2867 } 2868 2869 static int build_curseg(struct f2fs_sb_info *sbi) 2870 { 2871 struct curseg_info *array; 2872 int i; 2873 2874 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL); 2875 if (!array) 2876 return -ENOMEM; 2877 2878 SM_I(sbi)->curseg_array = array; 2879 2880 for (i = 0; i < NR_CURSEG_TYPE; i++) { 2881 mutex_init(&array[i].curseg_mutex); 2882 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL); 2883 if (!array[i].sum_blk) 2884 return -ENOMEM; 2885 init_rwsem(&array[i].journal_rwsem); 2886 array[i].journal = kzalloc(sizeof(struct f2fs_journal), 2887 GFP_KERNEL); 2888 if (!array[i].journal) 2889 return -ENOMEM; 2890 array[i].segno = NULL_SEGNO; 2891 array[i].next_blkoff = 0; 2892 } 2893 return restore_curseg_summaries(sbi); 2894 } 2895 2896 static void build_sit_entries(struct f2fs_sb_info *sbi) 2897 { 2898 struct sit_info *sit_i = SIT_I(sbi); 2899 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 2900 struct f2fs_journal *journal = curseg->journal; 2901 struct seg_entry *se; 2902 struct f2fs_sit_entry sit; 2903 int sit_blk_cnt = SIT_BLK_CNT(sbi); 2904 unsigned int i, start, end; 2905 unsigned int readed, start_blk = 0; 2906 2907 do { 2908 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES, 2909 META_SIT, true); 2910 2911 start = start_blk * sit_i->sents_per_block; 2912 end = (start_blk + readed) * sit_i->sents_per_block; 2913 2914 for (; start < end && start < MAIN_SEGS(sbi); start++) { 2915 struct f2fs_sit_block *sit_blk; 2916 struct page *page; 2917 2918 se = &sit_i->sentries[start]; 2919 page = get_current_sit_page(sbi, start); 2920 sit_blk = (struct f2fs_sit_block *)page_address(page); 2921 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 2922 f2fs_put_page(page, 1); 2923 2924 check_block_count(sbi, start, &sit); 2925 seg_info_from_raw_sit(se, &sit); 2926 2927 /* build discard map only one time */ 2928 if (f2fs_discard_en(sbi)) { 2929 memcpy(se->discard_map, se->cur_valid_map, 2930 SIT_VBLOCK_MAP_SIZE); 2931 sbi->discard_blks += sbi->blocks_per_seg - 2932 se->valid_blocks; 2933 } 2934 2935 if (sbi->segs_per_sec > 1) 2936 get_sec_entry(sbi, start)->valid_blocks += 2937 se->valid_blocks; 2938 } 2939 start_blk += readed; 2940 } while (start_blk < sit_blk_cnt); 2941 2942 down_read(&curseg->journal_rwsem); 2943 for (i = 0; i < sits_in_cursum(journal); i++) { 2944 unsigned int old_valid_blocks; 2945 2946 start = le32_to_cpu(segno_in_journal(journal, i)); 2947 se = &sit_i->sentries[start]; 2948 sit = sit_in_journal(journal, i); 2949 2950 old_valid_blocks = se->valid_blocks; 2951 2952 check_block_count(sbi, start, &sit); 2953 seg_info_from_raw_sit(se, &sit); 2954 2955 if (f2fs_discard_en(sbi)) { 2956 memcpy(se->discard_map, se->cur_valid_map, 2957 SIT_VBLOCK_MAP_SIZE); 2958 sbi->discard_blks += old_valid_blocks - 2959 se->valid_blocks; 2960 } 2961 2962 if (sbi->segs_per_sec > 1) 2963 get_sec_entry(sbi, start)->valid_blocks += 2964 se->valid_blocks - old_valid_blocks; 2965 } 2966 up_read(&curseg->journal_rwsem); 2967 } 2968 2969 static void init_free_segmap(struct f2fs_sb_info *sbi) 2970 { 2971 unsigned int start; 2972 int type; 2973 2974 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2975 struct seg_entry *sentry = get_seg_entry(sbi, start); 2976 if (!sentry->valid_blocks) 2977 __set_free(sbi, start); 2978 else 2979 SIT_I(sbi)->written_valid_blocks += 2980 sentry->valid_blocks; 2981 } 2982 2983 /* set use the current segments */ 2984 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 2985 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 2986 __set_test_and_inuse(sbi, curseg_t->segno); 2987 } 2988 } 2989 2990 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 2991 { 2992 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2993 struct free_segmap_info *free_i = FREE_I(sbi); 2994 unsigned int segno = 0, offset = 0; 2995 unsigned short valid_blocks; 2996 2997 while (1) { 2998 /* find dirty segment based on free segmap */ 2999 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 3000 if (segno >= MAIN_SEGS(sbi)) 3001 break; 3002 offset = segno + 1; 3003 valid_blocks = get_valid_blocks(sbi, segno, false); 3004 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) 3005 continue; 3006 if (valid_blocks > sbi->blocks_per_seg) { 3007 f2fs_bug_on(sbi, 1); 3008 continue; 3009 } 3010 mutex_lock(&dirty_i->seglist_lock); 3011 __locate_dirty_segment(sbi, segno, DIRTY); 3012 mutex_unlock(&dirty_i->seglist_lock); 3013 } 3014 } 3015 3016 static int init_victim_secmap(struct f2fs_sb_info *sbi) 3017 { 3018 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3019 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 3020 3021 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 3022 if (!dirty_i->victim_secmap) 3023 return -ENOMEM; 3024 return 0; 3025 } 3026 3027 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 3028 { 3029 struct dirty_seglist_info *dirty_i; 3030 unsigned int bitmap_size, i; 3031 3032 /* allocate memory for dirty segments list information */ 3033 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); 3034 if (!dirty_i) 3035 return -ENOMEM; 3036 3037 SM_I(sbi)->dirty_info = dirty_i; 3038 mutex_init(&dirty_i->seglist_lock); 3039 3040 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 3041 3042 for (i = 0; i < NR_DIRTY_TYPE; i++) { 3043 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 3044 if (!dirty_i->dirty_segmap[i]) 3045 return -ENOMEM; 3046 } 3047 3048 init_dirty_segmap(sbi); 3049 return init_victim_secmap(sbi); 3050 } 3051 3052 /* 3053 * Update min, max modified time for cost-benefit GC algorithm 3054 */ 3055 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 3056 { 3057 struct sit_info *sit_i = SIT_I(sbi); 3058 unsigned int segno; 3059 3060 mutex_lock(&sit_i->sentry_lock); 3061 3062 sit_i->min_mtime = LLONG_MAX; 3063 3064 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 3065 unsigned int i; 3066 unsigned long long mtime = 0; 3067 3068 for (i = 0; i < sbi->segs_per_sec; i++) 3069 mtime += get_seg_entry(sbi, segno + i)->mtime; 3070 3071 mtime = div_u64(mtime, sbi->segs_per_sec); 3072 3073 if (sit_i->min_mtime > mtime) 3074 sit_i->min_mtime = mtime; 3075 } 3076 sit_i->max_mtime = get_mtime(sbi); 3077 mutex_unlock(&sit_i->sentry_lock); 3078 } 3079 3080 int build_segment_manager(struct f2fs_sb_info *sbi) 3081 { 3082 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 3083 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3084 struct f2fs_sm_info *sm_info; 3085 int err; 3086 3087 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); 3088 if (!sm_info) 3089 return -ENOMEM; 3090 3091 /* init sm info */ 3092 sbi->sm_info = sm_info; 3093 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 3094 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 3095 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 3096 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 3097 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 3098 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 3099 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 3100 sm_info->rec_prefree_segments = sm_info->main_segments * 3101 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 3102 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 3103 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 3104 3105 if (!test_opt(sbi, LFS)) 3106 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 3107 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 3108 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 3109 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 3110 3111 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS; 3112 3113 INIT_LIST_HEAD(&sm_info->sit_entry_set); 3114 3115 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) { 3116 err = create_flush_cmd_control(sbi); 3117 if (err) 3118 return err; 3119 } 3120 3121 err = create_discard_cmd_control(sbi); 3122 if (err) 3123 return err; 3124 3125 err = build_sit_info(sbi); 3126 if (err) 3127 return err; 3128 err = build_free_segmap(sbi); 3129 if (err) 3130 return err; 3131 err = build_curseg(sbi); 3132 if (err) 3133 return err; 3134 3135 /* reinit free segmap based on SIT */ 3136 build_sit_entries(sbi); 3137 3138 init_free_segmap(sbi); 3139 err = build_dirty_segmap(sbi); 3140 if (err) 3141 return err; 3142 3143 init_min_max_mtime(sbi); 3144 return 0; 3145 } 3146 3147 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 3148 enum dirty_type dirty_type) 3149 { 3150 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3151 3152 mutex_lock(&dirty_i->seglist_lock); 3153 kvfree(dirty_i->dirty_segmap[dirty_type]); 3154 dirty_i->nr_dirty[dirty_type] = 0; 3155 mutex_unlock(&dirty_i->seglist_lock); 3156 } 3157 3158 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 3159 { 3160 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3161 kvfree(dirty_i->victim_secmap); 3162 } 3163 3164 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 3165 { 3166 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3167 int i; 3168 3169 if (!dirty_i) 3170 return; 3171 3172 /* discard pre-free/dirty segments list */ 3173 for (i = 0; i < NR_DIRTY_TYPE; i++) 3174 discard_dirty_segmap(sbi, i); 3175 3176 destroy_victim_secmap(sbi); 3177 SM_I(sbi)->dirty_info = NULL; 3178 kfree(dirty_i); 3179 } 3180 3181 static void destroy_curseg(struct f2fs_sb_info *sbi) 3182 { 3183 struct curseg_info *array = SM_I(sbi)->curseg_array; 3184 int i; 3185 3186 if (!array) 3187 return; 3188 SM_I(sbi)->curseg_array = NULL; 3189 for (i = 0; i < NR_CURSEG_TYPE; i++) { 3190 kfree(array[i].sum_blk); 3191 kfree(array[i].journal); 3192 } 3193 kfree(array); 3194 } 3195 3196 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 3197 { 3198 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 3199 if (!free_i) 3200 return; 3201 SM_I(sbi)->free_info = NULL; 3202 kvfree(free_i->free_segmap); 3203 kvfree(free_i->free_secmap); 3204 kfree(free_i); 3205 } 3206 3207 static void destroy_sit_info(struct f2fs_sb_info *sbi) 3208 { 3209 struct sit_info *sit_i = SIT_I(sbi); 3210 unsigned int start; 3211 3212 if (!sit_i) 3213 return; 3214 3215 if (sit_i->sentries) { 3216 for (start = 0; start < MAIN_SEGS(sbi); start++) { 3217 kfree(sit_i->sentries[start].cur_valid_map); 3218 #ifdef CONFIG_F2FS_CHECK_FS 3219 kfree(sit_i->sentries[start].cur_valid_map_mir); 3220 #endif 3221 kfree(sit_i->sentries[start].ckpt_valid_map); 3222 kfree(sit_i->sentries[start].discard_map); 3223 } 3224 } 3225 kfree(sit_i->tmp_map); 3226 3227 kvfree(sit_i->sentries); 3228 kvfree(sit_i->sec_entries); 3229 kvfree(sit_i->dirty_sentries_bitmap); 3230 3231 SM_I(sbi)->sit_info = NULL; 3232 kfree(sit_i->sit_bitmap); 3233 #ifdef CONFIG_F2FS_CHECK_FS 3234 kfree(sit_i->sit_bitmap_mir); 3235 #endif 3236 kfree(sit_i); 3237 } 3238 3239 void destroy_segment_manager(struct f2fs_sb_info *sbi) 3240 { 3241 struct f2fs_sm_info *sm_info = SM_I(sbi); 3242 3243 if (!sm_info) 3244 return; 3245 destroy_flush_cmd_control(sbi, true); 3246 destroy_discard_cmd_control(sbi); 3247 destroy_dirty_segmap(sbi); 3248 destroy_curseg(sbi); 3249 destroy_free_segmap(sbi); 3250 destroy_sit_info(sbi); 3251 sbi->sm_info = NULL; 3252 kfree(sm_info); 3253 } 3254 3255 int __init create_segment_manager_caches(void) 3256 { 3257 discard_entry_slab = f2fs_kmem_cache_create("discard_entry", 3258 sizeof(struct discard_entry)); 3259 if (!discard_entry_slab) 3260 goto fail; 3261 3262 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd", 3263 sizeof(struct discard_cmd)); 3264 if (!discard_cmd_slab) 3265 goto destroy_discard_entry; 3266 3267 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", 3268 sizeof(struct sit_entry_set)); 3269 if (!sit_entry_set_slab) 3270 goto destroy_discard_cmd; 3271 3272 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", 3273 sizeof(struct inmem_pages)); 3274 if (!inmem_entry_slab) 3275 goto destroy_sit_entry_set; 3276 return 0; 3277 3278 destroy_sit_entry_set: 3279 kmem_cache_destroy(sit_entry_set_slab); 3280 destroy_discard_cmd: 3281 kmem_cache_destroy(discard_cmd_slab); 3282 destroy_discard_entry: 3283 kmem_cache_destroy(discard_entry_slab); 3284 fail: 3285 return -ENOMEM; 3286 } 3287 3288 void destroy_segment_manager_caches(void) 3289 { 3290 kmem_cache_destroy(sit_entry_set_slab); 3291 kmem_cache_destroy(discard_cmd_slab); 3292 kmem_cache_destroy(discard_entry_slab); 3293 kmem_cache_destroy(inmem_entry_slab); 3294 } 3295