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; 681 struct discard_cmd *dc; 682 683 f2fs_bug_on(sbi, !len); 684 685 pend_list = &dcc->pend_list[plist_idx(len)]; 686 687 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS); 688 INIT_LIST_HEAD(&dc->list); 689 dc->bdev = bdev; 690 dc->lstart = lstart; 691 dc->start = start; 692 dc->len = len; 693 dc->state = D_PREP; 694 dc->error = 0; 695 init_completion(&dc->wait); 696 list_add_tail(&dc->list, pend_list); 697 atomic_inc(&dcc->discard_cmd_cnt); 698 699 return dc; 700 } 701 702 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi, 703 struct block_device *bdev, block_t lstart, 704 block_t start, block_t len, 705 struct rb_node *parent, struct rb_node **p) 706 { 707 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 708 struct discard_cmd *dc; 709 710 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 711 712 rb_link_node(&dc->rb_node, parent, p); 713 rb_insert_color(&dc->rb_node, &dcc->root); 714 715 return dc; 716 } 717 718 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 719 struct discard_cmd *dc) 720 { 721 if (dc->state == D_DONE) 722 atomic_dec(&dcc->issing_discard); 723 724 list_del(&dc->list); 725 rb_erase(&dc->rb_node, &dcc->root); 726 727 kmem_cache_free(discard_cmd_slab, dc); 728 729 atomic_dec(&dcc->discard_cmd_cnt); 730 } 731 732 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 733 struct discard_cmd *dc) 734 { 735 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 736 737 if (dc->error == -EOPNOTSUPP) 738 dc->error = 0; 739 740 if (dc->error) 741 f2fs_msg(sbi->sb, KERN_INFO, 742 "Issue discard failed, ret: %d", dc->error); 743 __detach_discard_cmd(dcc, dc); 744 } 745 746 static void f2fs_submit_discard_endio(struct bio *bio) 747 { 748 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 749 750 dc->error = bio->bi_error; 751 dc->state = D_DONE; 752 complete(&dc->wait); 753 bio_put(bio); 754 } 755 756 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 757 static void __submit_discard_cmd(struct f2fs_sb_info *sbi, 758 struct discard_cmd *dc) 759 { 760 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 761 struct bio *bio = NULL; 762 763 if (dc->state != D_PREP) 764 return; 765 766 dc->error = __blkdev_issue_discard(dc->bdev, 767 SECTOR_FROM_BLOCK(dc->start), 768 SECTOR_FROM_BLOCK(dc->len), 769 GFP_NOFS, 0, &bio); 770 if (!dc->error) { 771 /* should keep before submission to avoid D_DONE right away */ 772 dc->state = D_SUBMIT; 773 atomic_inc(&dcc->issued_discard); 774 atomic_inc(&dcc->issing_discard); 775 if (bio) { 776 bio->bi_private = dc; 777 bio->bi_end_io = f2fs_submit_discard_endio; 778 bio->bi_opf |= REQ_SYNC; 779 submit_bio(bio); 780 list_move_tail(&dc->list, &dcc->wait_list); 781 } 782 } else { 783 __remove_discard_cmd(sbi, dc); 784 } 785 } 786 787 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi, 788 struct block_device *bdev, block_t lstart, 789 block_t start, block_t len, 790 struct rb_node **insert_p, 791 struct rb_node *insert_parent) 792 { 793 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 794 struct rb_node **p = &dcc->root.rb_node; 795 struct rb_node *parent = NULL; 796 struct discard_cmd *dc = NULL; 797 798 if (insert_p && insert_parent) { 799 parent = insert_parent; 800 p = insert_p; 801 goto do_insert; 802 } 803 804 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart); 805 do_insert: 806 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p); 807 if (!dc) 808 return NULL; 809 810 return dc; 811 } 812 813 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 814 struct discard_cmd *dc) 815 { 816 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]); 817 } 818 819 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 820 struct discard_cmd *dc, block_t blkaddr) 821 { 822 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 823 struct discard_info di = dc->di; 824 bool modified = false; 825 826 if (dc->state == D_DONE || dc->len == 1) { 827 __remove_discard_cmd(sbi, dc); 828 return; 829 } 830 831 if (blkaddr > di.lstart) { 832 dc->len = blkaddr - dc->lstart; 833 __relocate_discard_cmd(dcc, dc); 834 modified = true; 835 } 836 837 if (blkaddr < di.lstart + di.len - 1) { 838 if (modified) { 839 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, 840 di.start + blkaddr + 1 - di.lstart, 841 di.lstart + di.len - 1 - blkaddr, 842 NULL, NULL); 843 } else { 844 dc->lstart++; 845 dc->len--; 846 dc->start++; 847 __relocate_discard_cmd(dcc, dc); 848 } 849 } 850 } 851 852 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 853 struct block_device *bdev, block_t lstart, 854 block_t start, block_t len) 855 { 856 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 857 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 858 struct discard_cmd *dc; 859 struct discard_info di = {0}; 860 struct rb_node **insert_p = NULL, *insert_parent = NULL; 861 block_t end = lstart + len; 862 863 mutex_lock(&dcc->cmd_lock); 864 865 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root, 866 NULL, lstart, 867 (struct rb_entry **)&prev_dc, 868 (struct rb_entry **)&next_dc, 869 &insert_p, &insert_parent, true); 870 if (dc) 871 prev_dc = dc; 872 873 if (!prev_dc) { 874 di.lstart = lstart; 875 di.len = next_dc ? next_dc->lstart - lstart : len; 876 di.len = min(di.len, len); 877 di.start = start; 878 } 879 880 while (1) { 881 struct rb_node *node; 882 bool merged = false; 883 struct discard_cmd *tdc = NULL; 884 885 if (prev_dc) { 886 di.lstart = prev_dc->lstart + prev_dc->len; 887 if (di.lstart < lstart) 888 di.lstart = lstart; 889 if (di.lstart >= end) 890 break; 891 892 if (!next_dc || next_dc->lstart > end) 893 di.len = end - di.lstart; 894 else 895 di.len = next_dc->lstart - di.lstart; 896 di.start = start + di.lstart - lstart; 897 } 898 899 if (!di.len) 900 goto next; 901 902 if (prev_dc && prev_dc->state == D_PREP && 903 prev_dc->bdev == bdev && 904 __is_discard_back_mergeable(&di, &prev_dc->di)) { 905 prev_dc->di.len += di.len; 906 __relocate_discard_cmd(dcc, prev_dc); 907 di = prev_dc->di; 908 tdc = prev_dc; 909 merged = true; 910 } 911 912 if (next_dc && next_dc->state == D_PREP && 913 next_dc->bdev == bdev && 914 __is_discard_front_mergeable(&di, &next_dc->di)) { 915 next_dc->di.lstart = di.lstart; 916 next_dc->di.len += di.len; 917 next_dc->di.start = di.start; 918 __relocate_discard_cmd(dcc, next_dc); 919 if (tdc) 920 __remove_discard_cmd(sbi, tdc); 921 922 merged = true; 923 } 924 925 if (!merged) 926 __insert_discard_tree(sbi, bdev, di.lstart, di.start, 927 di.len, NULL, NULL); 928 next: 929 prev_dc = next_dc; 930 if (!prev_dc) 931 break; 932 933 node = rb_next(&prev_dc->rb_node); 934 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 935 } 936 937 mutex_unlock(&dcc->cmd_lock); 938 } 939 940 static int __queue_discard_cmd(struct f2fs_sb_info *sbi, 941 struct block_device *bdev, block_t blkstart, block_t blklen) 942 { 943 block_t lblkstart = blkstart; 944 945 trace_f2fs_issue_discard(bdev, blkstart, blklen); 946 947 if (sbi->s_ndevs) { 948 int devi = f2fs_target_device_index(sbi, blkstart); 949 950 blkstart -= FDEV(devi).start_blk; 951 } 952 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 953 wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue); 954 return 0; 955 } 956 957 /* This should be covered by global mutex, &sit_i->sentry_lock */ 958 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 959 { 960 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 961 struct discard_cmd *dc; 962 963 mutex_lock(&dcc->cmd_lock); 964 965 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr); 966 if (dc) { 967 if (dc->state != D_PREP) 968 wait_for_completion_io(&dc->wait); 969 __punch_discard_cmd(sbi, dc, blkaddr); 970 } 971 972 mutex_unlock(&dcc->cmd_lock); 973 } 974 975 /* This comes from f2fs_put_super */ 976 void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi) 977 { 978 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 979 struct list_head *pend_list; 980 struct list_head *wait_list = &(dcc->wait_list); 981 struct discard_cmd *dc, *tmp; 982 struct blk_plug plug; 983 int i; 984 985 mutex_lock(&dcc->cmd_lock); 986 987 blk_start_plug(&plug); 988 for (i = 0; i < MAX_PLIST_NUM; i++) { 989 pend_list = &dcc->pend_list[i]; 990 list_for_each_entry_safe(dc, tmp, pend_list, list) 991 __submit_discard_cmd(sbi, dc); 992 } 993 blk_finish_plug(&plug); 994 995 list_for_each_entry_safe(dc, tmp, wait_list, list) { 996 wait_for_completion_io(&dc->wait); 997 __remove_discard_cmd(sbi, dc); 998 } 999 1000 mutex_unlock(&dcc->cmd_lock); 1001 } 1002 1003 static int issue_discard_thread(void *data) 1004 { 1005 struct f2fs_sb_info *sbi = data; 1006 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1007 wait_queue_head_t *q = &dcc->discard_wait_queue; 1008 struct list_head *pend_list; 1009 struct list_head *wait_list = &dcc->wait_list; 1010 struct discard_cmd *dc, *tmp; 1011 struct blk_plug plug; 1012 int iter = 0, i; 1013 repeat: 1014 if (kthread_should_stop()) 1015 return 0; 1016 1017 mutex_lock(&dcc->cmd_lock); 1018 blk_start_plug(&plug); 1019 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1020 pend_list = &dcc->pend_list[i]; 1021 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1022 f2fs_bug_on(sbi, dc->state != D_PREP); 1023 1024 if (is_idle(sbi)) 1025 __submit_discard_cmd(sbi, dc); 1026 1027 if (iter++ > DISCARD_ISSUE_RATE) 1028 goto next_step; 1029 } 1030 } 1031 next_step: 1032 blk_finish_plug(&plug); 1033 1034 list_for_each_entry_safe(dc, tmp, wait_list, list) { 1035 if (dc->state == D_DONE) { 1036 wait_for_completion_io(&dc->wait); 1037 __remove_discard_cmd(sbi, dc); 1038 } 1039 } 1040 mutex_unlock(&dcc->cmd_lock); 1041 1042 iter = 0; 1043 congestion_wait(BLK_RW_SYNC, HZ/50); 1044 1045 wait_event_interruptible(*q, kthread_should_stop() || 1046 atomic_read(&dcc->discard_cmd_cnt)); 1047 goto repeat; 1048 } 1049 1050 #ifdef CONFIG_BLK_DEV_ZONED 1051 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1052 struct block_device *bdev, block_t blkstart, block_t blklen) 1053 { 1054 sector_t sector, nr_sects; 1055 block_t lblkstart = blkstart; 1056 int devi = 0; 1057 1058 if (sbi->s_ndevs) { 1059 devi = f2fs_target_device_index(sbi, blkstart); 1060 blkstart -= FDEV(devi).start_blk; 1061 } 1062 1063 /* 1064 * We need to know the type of the zone: for conventional zones, 1065 * use regular discard if the drive supports it. For sequential 1066 * zones, reset the zone write pointer. 1067 */ 1068 switch (get_blkz_type(sbi, bdev, blkstart)) { 1069 1070 case BLK_ZONE_TYPE_CONVENTIONAL: 1071 if (!blk_queue_discard(bdev_get_queue(bdev))) 1072 return 0; 1073 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1074 case BLK_ZONE_TYPE_SEQWRITE_REQ: 1075 case BLK_ZONE_TYPE_SEQWRITE_PREF: 1076 sector = SECTOR_FROM_BLOCK(blkstart); 1077 nr_sects = SECTOR_FROM_BLOCK(blklen); 1078 1079 if (sector & (bdev_zone_sectors(bdev) - 1) || 1080 nr_sects != bdev_zone_sectors(bdev)) { 1081 f2fs_msg(sbi->sb, KERN_INFO, 1082 "(%d) %s: Unaligned discard attempted (block %x + %x)", 1083 devi, sbi->s_ndevs ? FDEV(devi).path: "", 1084 blkstart, blklen); 1085 return -EIO; 1086 } 1087 trace_f2fs_issue_reset_zone(bdev, blkstart); 1088 return blkdev_reset_zones(bdev, sector, 1089 nr_sects, GFP_NOFS); 1090 default: 1091 /* Unknown zone type: broken device ? */ 1092 return -EIO; 1093 } 1094 } 1095 #endif 1096 1097 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1098 struct block_device *bdev, block_t blkstart, block_t blklen) 1099 { 1100 #ifdef CONFIG_BLK_DEV_ZONED 1101 if (f2fs_sb_mounted_blkzoned(sbi->sb) && 1102 bdev_zoned_model(bdev) != BLK_ZONED_NONE) 1103 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1104 #endif 1105 return __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1106 } 1107 1108 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1109 block_t blkstart, block_t blklen) 1110 { 1111 sector_t start = blkstart, len = 0; 1112 struct block_device *bdev; 1113 struct seg_entry *se; 1114 unsigned int offset; 1115 block_t i; 1116 int err = 0; 1117 1118 bdev = f2fs_target_device(sbi, blkstart, NULL); 1119 1120 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1121 if (i != start) { 1122 struct block_device *bdev2 = 1123 f2fs_target_device(sbi, i, NULL); 1124 1125 if (bdev2 != bdev) { 1126 err = __issue_discard_async(sbi, bdev, 1127 start, len); 1128 if (err) 1129 return err; 1130 bdev = bdev2; 1131 start = i; 1132 len = 0; 1133 } 1134 } 1135 1136 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 1137 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 1138 1139 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 1140 sbi->discard_blks--; 1141 } 1142 1143 if (len) 1144 err = __issue_discard_async(sbi, bdev, start, len); 1145 return err; 1146 } 1147 1148 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 1149 bool check_only) 1150 { 1151 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1152 int max_blocks = sbi->blocks_per_seg; 1153 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 1154 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1155 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1156 unsigned long *discard_map = (unsigned long *)se->discard_map; 1157 unsigned long *dmap = SIT_I(sbi)->tmp_map; 1158 unsigned int start = 0, end = -1; 1159 bool force = (cpc->reason == CP_DISCARD); 1160 struct discard_entry *de = NULL; 1161 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 1162 int i; 1163 1164 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi)) 1165 return false; 1166 1167 if (!force) { 1168 if (!test_opt(sbi, DISCARD) || !se->valid_blocks || 1169 SM_I(sbi)->dcc_info->nr_discards >= 1170 SM_I(sbi)->dcc_info->max_discards) 1171 return false; 1172 } 1173 1174 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 1175 for (i = 0; i < entries; i++) 1176 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 1177 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 1178 1179 while (force || SM_I(sbi)->dcc_info->nr_discards <= 1180 SM_I(sbi)->dcc_info->max_discards) { 1181 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 1182 if (start >= max_blocks) 1183 break; 1184 1185 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 1186 if (force && start && end != max_blocks 1187 && (end - start) < cpc->trim_minlen) 1188 continue; 1189 1190 if (check_only) 1191 return true; 1192 1193 if (!de) { 1194 de = f2fs_kmem_cache_alloc(discard_entry_slab, 1195 GFP_F2FS_ZERO); 1196 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 1197 list_add_tail(&de->list, head); 1198 } 1199 1200 for (i = start; i < end; i++) 1201 __set_bit_le(i, (void *)de->discard_map); 1202 1203 SM_I(sbi)->dcc_info->nr_discards += end - start; 1204 } 1205 return false; 1206 } 1207 1208 void release_discard_addrs(struct f2fs_sb_info *sbi) 1209 { 1210 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 1211 struct discard_entry *entry, *this; 1212 1213 /* drop caches */ 1214 list_for_each_entry_safe(entry, this, head, list) { 1215 list_del(&entry->list); 1216 kmem_cache_free(discard_entry_slab, entry); 1217 } 1218 } 1219 1220 /* 1221 * Should call clear_prefree_segments after checkpoint is done. 1222 */ 1223 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 1224 { 1225 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1226 unsigned int segno; 1227 1228 mutex_lock(&dirty_i->seglist_lock); 1229 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 1230 __set_test_and_free(sbi, segno); 1231 mutex_unlock(&dirty_i->seglist_lock); 1232 } 1233 1234 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1235 { 1236 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 1237 struct discard_entry *entry, *this; 1238 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1239 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 1240 unsigned int start = 0, end = -1; 1241 unsigned int secno, start_segno; 1242 bool force = (cpc->reason == CP_DISCARD); 1243 1244 mutex_lock(&dirty_i->seglist_lock); 1245 1246 while (1) { 1247 int i; 1248 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 1249 if (start >= MAIN_SEGS(sbi)) 1250 break; 1251 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 1252 start + 1); 1253 1254 for (i = start; i < end; i++) 1255 clear_bit(i, prefree_map); 1256 1257 dirty_i->nr_dirty[PRE] -= end - start; 1258 1259 if (!test_opt(sbi, DISCARD)) 1260 continue; 1261 1262 if (force && start >= cpc->trim_start && 1263 (end - 1) <= cpc->trim_end) 1264 continue; 1265 1266 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) { 1267 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 1268 (end - start) << sbi->log_blocks_per_seg); 1269 continue; 1270 } 1271 next: 1272 secno = GET_SEC_FROM_SEG(sbi, start); 1273 start_segno = GET_SEG_FROM_SEC(sbi, secno); 1274 if (!IS_CURSEC(sbi, secno) && 1275 !get_valid_blocks(sbi, start, true)) 1276 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 1277 sbi->segs_per_sec << sbi->log_blocks_per_seg); 1278 1279 start = start_segno + sbi->segs_per_sec; 1280 if (start < end) 1281 goto next; 1282 else 1283 end = start - 1; 1284 } 1285 mutex_unlock(&dirty_i->seglist_lock); 1286 1287 /* send small discards */ 1288 list_for_each_entry_safe(entry, this, head, list) { 1289 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 1290 bool is_valid = test_bit_le(0, entry->discard_map); 1291 1292 find_next: 1293 if (is_valid) { 1294 next_pos = find_next_zero_bit_le(entry->discard_map, 1295 sbi->blocks_per_seg, cur_pos); 1296 len = next_pos - cur_pos; 1297 1298 if (force && len < cpc->trim_minlen) 1299 goto skip; 1300 1301 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 1302 len); 1303 cpc->trimmed += len; 1304 total_len += len; 1305 } else { 1306 next_pos = find_next_bit_le(entry->discard_map, 1307 sbi->blocks_per_seg, cur_pos); 1308 } 1309 skip: 1310 cur_pos = next_pos; 1311 is_valid = !is_valid; 1312 1313 if (cur_pos < sbi->blocks_per_seg) 1314 goto find_next; 1315 1316 list_del(&entry->list); 1317 SM_I(sbi)->dcc_info->nr_discards -= total_len; 1318 kmem_cache_free(discard_entry_slab, entry); 1319 } 1320 } 1321 1322 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 1323 { 1324 dev_t dev = sbi->sb->s_bdev->bd_dev; 1325 struct discard_cmd_control *dcc; 1326 int err = 0, i; 1327 1328 if (SM_I(sbi)->dcc_info) { 1329 dcc = SM_I(sbi)->dcc_info; 1330 goto init_thread; 1331 } 1332 1333 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL); 1334 if (!dcc) 1335 return -ENOMEM; 1336 1337 INIT_LIST_HEAD(&dcc->entry_list); 1338 for (i = 0; i < MAX_PLIST_NUM; i++) 1339 INIT_LIST_HEAD(&dcc->pend_list[i]); 1340 INIT_LIST_HEAD(&dcc->wait_list); 1341 mutex_init(&dcc->cmd_lock); 1342 atomic_set(&dcc->issued_discard, 0); 1343 atomic_set(&dcc->issing_discard, 0); 1344 atomic_set(&dcc->discard_cmd_cnt, 0); 1345 dcc->nr_discards = 0; 1346 dcc->max_discards = 0; 1347 dcc->root = RB_ROOT; 1348 1349 init_waitqueue_head(&dcc->discard_wait_queue); 1350 SM_I(sbi)->dcc_info = dcc; 1351 init_thread: 1352 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 1353 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 1354 if (IS_ERR(dcc->f2fs_issue_discard)) { 1355 err = PTR_ERR(dcc->f2fs_issue_discard); 1356 kfree(dcc); 1357 SM_I(sbi)->dcc_info = NULL; 1358 return err; 1359 } 1360 1361 return err; 1362 } 1363 1364 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 1365 { 1366 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1367 1368 if (!dcc) 1369 return; 1370 1371 if (dcc->f2fs_issue_discard) { 1372 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1373 1374 dcc->f2fs_issue_discard = NULL; 1375 kthread_stop(discard_thread); 1376 } 1377 1378 kfree(dcc); 1379 SM_I(sbi)->dcc_info = NULL; 1380 } 1381 1382 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 1383 { 1384 struct sit_info *sit_i = SIT_I(sbi); 1385 1386 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 1387 sit_i->dirty_sentries++; 1388 return false; 1389 } 1390 1391 return true; 1392 } 1393 1394 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 1395 unsigned int segno, int modified) 1396 { 1397 struct seg_entry *se = get_seg_entry(sbi, segno); 1398 se->type = type; 1399 if (modified) 1400 __mark_sit_entry_dirty(sbi, segno); 1401 } 1402 1403 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 1404 { 1405 struct seg_entry *se; 1406 unsigned int segno, offset; 1407 long int new_vblocks; 1408 1409 segno = GET_SEGNO(sbi, blkaddr); 1410 1411 se = get_seg_entry(sbi, segno); 1412 new_vblocks = se->valid_blocks + del; 1413 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1414 1415 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || 1416 (new_vblocks > sbi->blocks_per_seg))); 1417 1418 se->valid_blocks = new_vblocks; 1419 se->mtime = get_mtime(sbi); 1420 SIT_I(sbi)->max_mtime = se->mtime; 1421 1422 /* Update valid block bitmap */ 1423 if (del > 0) { 1424 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) { 1425 #ifdef CONFIG_F2FS_CHECK_FS 1426 if (f2fs_test_and_set_bit(offset, 1427 se->cur_valid_map_mir)) 1428 f2fs_bug_on(sbi, 1); 1429 else 1430 WARN_ON(1); 1431 #else 1432 f2fs_bug_on(sbi, 1); 1433 #endif 1434 } 1435 if (f2fs_discard_en(sbi) && 1436 !f2fs_test_and_set_bit(offset, se->discard_map)) 1437 sbi->discard_blks--; 1438 1439 /* don't overwrite by SSR to keep node chain */ 1440 if (se->type == CURSEG_WARM_NODE) { 1441 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 1442 se->ckpt_valid_blocks++; 1443 } 1444 } else { 1445 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) { 1446 #ifdef CONFIG_F2FS_CHECK_FS 1447 if (!f2fs_test_and_clear_bit(offset, 1448 se->cur_valid_map_mir)) 1449 f2fs_bug_on(sbi, 1); 1450 else 1451 WARN_ON(1); 1452 #else 1453 f2fs_bug_on(sbi, 1); 1454 #endif 1455 } 1456 if (f2fs_discard_en(sbi) && 1457 f2fs_test_and_clear_bit(offset, se->discard_map)) 1458 sbi->discard_blks++; 1459 } 1460 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 1461 se->ckpt_valid_blocks += del; 1462 1463 __mark_sit_entry_dirty(sbi, segno); 1464 1465 /* update total number of valid blocks to be written in ckpt area */ 1466 SIT_I(sbi)->written_valid_blocks += del; 1467 1468 if (sbi->segs_per_sec > 1) 1469 get_sec_entry(sbi, segno)->valid_blocks += del; 1470 } 1471 1472 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) 1473 { 1474 update_sit_entry(sbi, new, 1); 1475 if (GET_SEGNO(sbi, old) != NULL_SEGNO) 1476 update_sit_entry(sbi, old, -1); 1477 1478 locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); 1479 locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); 1480 } 1481 1482 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 1483 { 1484 unsigned int segno = GET_SEGNO(sbi, addr); 1485 struct sit_info *sit_i = SIT_I(sbi); 1486 1487 f2fs_bug_on(sbi, addr == NULL_ADDR); 1488 if (addr == NEW_ADDR) 1489 return; 1490 1491 /* add it into sit main buffer */ 1492 mutex_lock(&sit_i->sentry_lock); 1493 1494 update_sit_entry(sbi, addr, -1); 1495 1496 /* add it into dirty seglist */ 1497 locate_dirty_segment(sbi, segno); 1498 1499 mutex_unlock(&sit_i->sentry_lock); 1500 } 1501 1502 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 1503 { 1504 struct sit_info *sit_i = SIT_I(sbi); 1505 unsigned int segno, offset; 1506 struct seg_entry *se; 1507 bool is_cp = false; 1508 1509 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) 1510 return true; 1511 1512 mutex_lock(&sit_i->sentry_lock); 1513 1514 segno = GET_SEGNO(sbi, blkaddr); 1515 se = get_seg_entry(sbi, segno); 1516 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1517 1518 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 1519 is_cp = true; 1520 1521 mutex_unlock(&sit_i->sentry_lock); 1522 1523 return is_cp; 1524 } 1525 1526 /* 1527 * This function should be resided under the curseg_mutex lock 1528 */ 1529 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 1530 struct f2fs_summary *sum) 1531 { 1532 struct curseg_info *curseg = CURSEG_I(sbi, type); 1533 void *addr = curseg->sum_blk; 1534 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 1535 memcpy(addr, sum, sizeof(struct f2fs_summary)); 1536 } 1537 1538 /* 1539 * Calculate the number of current summary pages for writing 1540 */ 1541 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 1542 { 1543 int valid_sum_count = 0; 1544 int i, sum_in_page; 1545 1546 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1547 if (sbi->ckpt->alloc_type[i] == SSR) 1548 valid_sum_count += sbi->blocks_per_seg; 1549 else { 1550 if (for_ra) 1551 valid_sum_count += le16_to_cpu( 1552 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 1553 else 1554 valid_sum_count += curseg_blkoff(sbi, i); 1555 } 1556 } 1557 1558 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 1559 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 1560 if (valid_sum_count <= sum_in_page) 1561 return 1; 1562 else if ((valid_sum_count - sum_in_page) <= 1563 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 1564 return 2; 1565 return 3; 1566 } 1567 1568 /* 1569 * Caller should put this summary page 1570 */ 1571 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 1572 { 1573 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); 1574 } 1575 1576 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr) 1577 { 1578 struct page *page = grab_meta_page(sbi, blk_addr); 1579 void *dst = page_address(page); 1580 1581 if (src) 1582 memcpy(dst, src, PAGE_SIZE); 1583 else 1584 memset(dst, 0, PAGE_SIZE); 1585 set_page_dirty(page); 1586 f2fs_put_page(page, 1); 1587 } 1588 1589 static void write_sum_page(struct f2fs_sb_info *sbi, 1590 struct f2fs_summary_block *sum_blk, block_t blk_addr) 1591 { 1592 update_meta_page(sbi, (void *)sum_blk, blk_addr); 1593 } 1594 1595 static void write_current_sum_page(struct f2fs_sb_info *sbi, 1596 int type, block_t blk_addr) 1597 { 1598 struct curseg_info *curseg = CURSEG_I(sbi, type); 1599 struct page *page = grab_meta_page(sbi, blk_addr); 1600 struct f2fs_summary_block *src = curseg->sum_blk; 1601 struct f2fs_summary_block *dst; 1602 1603 dst = (struct f2fs_summary_block *)page_address(page); 1604 1605 mutex_lock(&curseg->curseg_mutex); 1606 1607 down_read(&curseg->journal_rwsem); 1608 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 1609 up_read(&curseg->journal_rwsem); 1610 1611 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 1612 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 1613 1614 mutex_unlock(&curseg->curseg_mutex); 1615 1616 set_page_dirty(page); 1617 f2fs_put_page(page, 1); 1618 } 1619 1620 /* 1621 * Find a new segment from the free segments bitmap to right order 1622 * This function should be returned with success, otherwise BUG 1623 */ 1624 static void get_new_segment(struct f2fs_sb_info *sbi, 1625 unsigned int *newseg, bool new_sec, int dir) 1626 { 1627 struct free_segmap_info *free_i = FREE_I(sbi); 1628 unsigned int segno, secno, zoneno; 1629 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 1630 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 1631 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 1632 unsigned int left_start = hint; 1633 bool init = true; 1634 int go_left = 0; 1635 int i; 1636 1637 spin_lock(&free_i->segmap_lock); 1638 1639 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 1640 segno = find_next_zero_bit(free_i->free_segmap, 1641 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 1642 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 1643 goto got_it; 1644 } 1645 find_other_zone: 1646 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 1647 if (secno >= MAIN_SECS(sbi)) { 1648 if (dir == ALLOC_RIGHT) { 1649 secno = find_next_zero_bit(free_i->free_secmap, 1650 MAIN_SECS(sbi), 0); 1651 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 1652 } else { 1653 go_left = 1; 1654 left_start = hint - 1; 1655 } 1656 } 1657 if (go_left == 0) 1658 goto skip_left; 1659 1660 while (test_bit(left_start, free_i->free_secmap)) { 1661 if (left_start > 0) { 1662 left_start--; 1663 continue; 1664 } 1665 left_start = find_next_zero_bit(free_i->free_secmap, 1666 MAIN_SECS(sbi), 0); 1667 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 1668 break; 1669 } 1670 secno = left_start; 1671 skip_left: 1672 hint = secno; 1673 segno = GET_SEG_FROM_SEC(sbi, secno); 1674 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 1675 1676 /* give up on finding another zone */ 1677 if (!init) 1678 goto got_it; 1679 if (sbi->secs_per_zone == 1) 1680 goto got_it; 1681 if (zoneno == old_zoneno) 1682 goto got_it; 1683 if (dir == ALLOC_LEFT) { 1684 if (!go_left && zoneno + 1 >= total_zones) 1685 goto got_it; 1686 if (go_left && zoneno == 0) 1687 goto got_it; 1688 } 1689 for (i = 0; i < NR_CURSEG_TYPE; i++) 1690 if (CURSEG_I(sbi, i)->zone == zoneno) 1691 break; 1692 1693 if (i < NR_CURSEG_TYPE) { 1694 /* zone is in user, try another */ 1695 if (go_left) 1696 hint = zoneno * sbi->secs_per_zone - 1; 1697 else if (zoneno + 1 >= total_zones) 1698 hint = 0; 1699 else 1700 hint = (zoneno + 1) * sbi->secs_per_zone; 1701 init = false; 1702 goto find_other_zone; 1703 } 1704 got_it: 1705 /* set it as dirty segment in free segmap */ 1706 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 1707 __set_inuse(sbi, segno); 1708 *newseg = segno; 1709 spin_unlock(&free_i->segmap_lock); 1710 } 1711 1712 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 1713 { 1714 struct curseg_info *curseg = CURSEG_I(sbi, type); 1715 struct summary_footer *sum_footer; 1716 1717 curseg->segno = curseg->next_segno; 1718 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 1719 curseg->next_blkoff = 0; 1720 curseg->next_segno = NULL_SEGNO; 1721 1722 sum_footer = &(curseg->sum_blk->footer); 1723 memset(sum_footer, 0, sizeof(struct summary_footer)); 1724 if (IS_DATASEG(type)) 1725 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 1726 if (IS_NODESEG(type)) 1727 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 1728 __set_sit_entry_type(sbi, type, curseg->segno, modified); 1729 } 1730 1731 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 1732 { 1733 if (type == CURSEG_HOT_DATA || IS_NODESEG(type)) 1734 return 0; 1735 1736 return CURSEG_I(sbi, type)->segno; 1737 } 1738 1739 /* 1740 * Allocate a current working segment. 1741 * This function always allocates a free segment in LFS manner. 1742 */ 1743 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 1744 { 1745 struct curseg_info *curseg = CURSEG_I(sbi, type); 1746 unsigned int segno = curseg->segno; 1747 int dir = ALLOC_LEFT; 1748 1749 write_sum_page(sbi, curseg->sum_blk, 1750 GET_SUM_BLOCK(sbi, segno)); 1751 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 1752 dir = ALLOC_RIGHT; 1753 1754 if (test_opt(sbi, NOHEAP)) 1755 dir = ALLOC_RIGHT; 1756 1757 segno = __get_next_segno(sbi, type); 1758 get_new_segment(sbi, &segno, new_sec, dir); 1759 curseg->next_segno = segno; 1760 reset_curseg(sbi, type, 1); 1761 curseg->alloc_type = LFS; 1762 } 1763 1764 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 1765 struct curseg_info *seg, block_t start) 1766 { 1767 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 1768 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1769 unsigned long *target_map = SIT_I(sbi)->tmp_map; 1770 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1771 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1772 int i, pos; 1773 1774 for (i = 0; i < entries; i++) 1775 target_map[i] = ckpt_map[i] | cur_map[i]; 1776 1777 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 1778 1779 seg->next_blkoff = pos; 1780 } 1781 1782 /* 1783 * If a segment is written by LFS manner, next block offset is just obtained 1784 * by increasing the current block offset. However, if a segment is written by 1785 * SSR manner, next block offset obtained by calling __next_free_blkoff 1786 */ 1787 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 1788 struct curseg_info *seg) 1789 { 1790 if (seg->alloc_type == SSR) 1791 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 1792 else 1793 seg->next_blkoff++; 1794 } 1795 1796 /* 1797 * This function always allocates a used segment(from dirty seglist) by SSR 1798 * manner, so it should recover the existing segment information of valid blocks 1799 */ 1800 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) 1801 { 1802 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1803 struct curseg_info *curseg = CURSEG_I(sbi, type); 1804 unsigned int new_segno = curseg->next_segno; 1805 struct f2fs_summary_block *sum_node; 1806 struct page *sum_page; 1807 1808 write_sum_page(sbi, curseg->sum_blk, 1809 GET_SUM_BLOCK(sbi, curseg->segno)); 1810 __set_test_and_inuse(sbi, new_segno); 1811 1812 mutex_lock(&dirty_i->seglist_lock); 1813 __remove_dirty_segment(sbi, new_segno, PRE); 1814 __remove_dirty_segment(sbi, new_segno, DIRTY); 1815 mutex_unlock(&dirty_i->seglist_lock); 1816 1817 reset_curseg(sbi, type, 1); 1818 curseg->alloc_type = SSR; 1819 __next_free_blkoff(sbi, curseg, 0); 1820 1821 if (reuse) { 1822 sum_page = get_sum_page(sbi, new_segno); 1823 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 1824 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 1825 f2fs_put_page(sum_page, 1); 1826 } 1827 } 1828 1829 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 1830 { 1831 struct curseg_info *curseg = CURSEG_I(sbi, type); 1832 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 1833 int i, cnt; 1834 bool reversed = false; 1835 1836 /* need_SSR() already forces to do this */ 1837 if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR)) 1838 return 1; 1839 1840 /* For node segments, let's do SSR more intensively */ 1841 if (IS_NODESEG(type)) { 1842 if (type >= CURSEG_WARM_NODE) { 1843 reversed = true; 1844 i = CURSEG_COLD_NODE; 1845 } else { 1846 i = CURSEG_HOT_NODE; 1847 } 1848 cnt = NR_CURSEG_NODE_TYPE; 1849 } else { 1850 if (type >= CURSEG_WARM_DATA) { 1851 reversed = true; 1852 i = CURSEG_COLD_DATA; 1853 } else { 1854 i = CURSEG_HOT_DATA; 1855 } 1856 cnt = NR_CURSEG_DATA_TYPE; 1857 } 1858 1859 for (; cnt-- > 0; reversed ? i-- : i++) { 1860 if (i == type) 1861 continue; 1862 if (v_ops->get_victim(sbi, &(curseg)->next_segno, 1863 BG_GC, i, SSR)) 1864 return 1; 1865 } 1866 return 0; 1867 } 1868 1869 /* 1870 * flush out current segment and replace it with new segment 1871 * This function should be returned with success, otherwise BUG 1872 */ 1873 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 1874 int type, bool force) 1875 { 1876 if (force) 1877 new_curseg(sbi, type, true); 1878 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 1879 type == CURSEG_WARM_NODE) 1880 new_curseg(sbi, type, false); 1881 else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) 1882 change_curseg(sbi, type, true); 1883 else 1884 new_curseg(sbi, type, false); 1885 1886 stat_inc_seg_type(sbi, CURSEG_I(sbi, type)); 1887 } 1888 1889 void allocate_new_segments(struct f2fs_sb_info *sbi) 1890 { 1891 struct curseg_info *curseg; 1892 unsigned int old_segno; 1893 int i; 1894 1895 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1896 curseg = CURSEG_I(sbi, i); 1897 old_segno = curseg->segno; 1898 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); 1899 locate_dirty_segment(sbi, old_segno); 1900 } 1901 } 1902 1903 static const struct segment_allocation default_salloc_ops = { 1904 .allocate_segment = allocate_segment_by_default, 1905 }; 1906 1907 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1908 { 1909 __u64 trim_start = cpc->trim_start; 1910 bool has_candidate = false; 1911 1912 mutex_lock(&SIT_I(sbi)->sentry_lock); 1913 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 1914 if (add_discard_addrs(sbi, cpc, true)) { 1915 has_candidate = true; 1916 break; 1917 } 1918 } 1919 mutex_unlock(&SIT_I(sbi)->sentry_lock); 1920 1921 cpc->trim_start = trim_start; 1922 return has_candidate; 1923 } 1924 1925 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 1926 { 1927 __u64 start = F2FS_BYTES_TO_BLK(range->start); 1928 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 1929 unsigned int start_segno, end_segno; 1930 struct cp_control cpc; 1931 int err = 0; 1932 1933 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 1934 return -EINVAL; 1935 1936 cpc.trimmed = 0; 1937 if (end <= MAIN_BLKADDR(sbi)) 1938 goto out; 1939 1940 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1941 f2fs_msg(sbi->sb, KERN_WARNING, 1942 "Found FS corruption, run fsck to fix."); 1943 goto out; 1944 } 1945 1946 /* start/end segment number in main_area */ 1947 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 1948 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 1949 GET_SEGNO(sbi, end); 1950 cpc.reason = CP_DISCARD; 1951 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 1952 1953 /* do checkpoint to issue discard commands safely */ 1954 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) { 1955 cpc.trim_start = start_segno; 1956 1957 if (sbi->discard_blks == 0) 1958 break; 1959 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi)) 1960 cpc.trim_end = end_segno; 1961 else 1962 cpc.trim_end = min_t(unsigned int, 1963 rounddown(start_segno + 1964 BATCHED_TRIM_SEGMENTS(sbi), 1965 sbi->segs_per_sec) - 1, end_segno); 1966 1967 mutex_lock(&sbi->gc_mutex); 1968 err = write_checkpoint(sbi, &cpc); 1969 mutex_unlock(&sbi->gc_mutex); 1970 if (err) 1971 break; 1972 1973 schedule(); 1974 } 1975 out: 1976 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed); 1977 return err; 1978 } 1979 1980 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 1981 { 1982 struct curseg_info *curseg = CURSEG_I(sbi, type); 1983 if (curseg->next_blkoff < sbi->blocks_per_seg) 1984 return true; 1985 return false; 1986 } 1987 1988 static int __get_segment_type_2(struct page *page, enum page_type p_type) 1989 { 1990 if (p_type == DATA) 1991 return CURSEG_HOT_DATA; 1992 else 1993 return CURSEG_HOT_NODE; 1994 } 1995 1996 static int __get_segment_type_4(struct page *page, enum page_type p_type) 1997 { 1998 if (p_type == DATA) { 1999 struct inode *inode = page->mapping->host; 2000 2001 if (S_ISDIR(inode->i_mode)) 2002 return CURSEG_HOT_DATA; 2003 else 2004 return CURSEG_COLD_DATA; 2005 } else { 2006 if (IS_DNODE(page) && is_cold_node(page)) 2007 return CURSEG_WARM_NODE; 2008 else 2009 return CURSEG_COLD_NODE; 2010 } 2011 } 2012 2013 static int __get_segment_type_6(struct page *page, enum page_type p_type) 2014 { 2015 if (p_type == DATA) { 2016 struct inode *inode = page->mapping->host; 2017 2018 if (is_cold_data(page) || file_is_cold(inode)) 2019 return CURSEG_COLD_DATA; 2020 if (is_inode_flag_set(inode, FI_HOT_DATA)) 2021 return CURSEG_HOT_DATA; 2022 return CURSEG_WARM_DATA; 2023 } else { 2024 if (IS_DNODE(page)) 2025 return is_cold_node(page) ? CURSEG_WARM_NODE : 2026 CURSEG_HOT_NODE; 2027 return CURSEG_COLD_NODE; 2028 } 2029 } 2030 2031 static int __get_segment_type(struct page *page, enum page_type p_type) 2032 { 2033 switch (F2FS_P_SB(page)->active_logs) { 2034 case 2: 2035 return __get_segment_type_2(page, p_type); 2036 case 4: 2037 return __get_segment_type_4(page, p_type); 2038 } 2039 /* NR_CURSEG_TYPE(6) logs by default */ 2040 f2fs_bug_on(F2FS_P_SB(page), 2041 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE); 2042 return __get_segment_type_6(page, p_type); 2043 } 2044 2045 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 2046 block_t old_blkaddr, block_t *new_blkaddr, 2047 struct f2fs_summary *sum, int type) 2048 { 2049 struct sit_info *sit_i = SIT_I(sbi); 2050 struct curseg_info *curseg = CURSEG_I(sbi, type); 2051 2052 mutex_lock(&curseg->curseg_mutex); 2053 mutex_lock(&sit_i->sentry_lock); 2054 2055 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 2056 2057 f2fs_wait_discard_bio(sbi, *new_blkaddr); 2058 2059 /* 2060 * __add_sum_entry should be resided under the curseg_mutex 2061 * because, this function updates a summary entry in the 2062 * current summary block. 2063 */ 2064 __add_sum_entry(sbi, type, sum); 2065 2066 __refresh_next_blkoff(sbi, curseg); 2067 2068 stat_inc_block_count(sbi, curseg); 2069 2070 if (!__has_curseg_space(sbi, type)) 2071 sit_i->s_ops->allocate_segment(sbi, type, false); 2072 /* 2073 * SIT information should be updated after segment allocation, 2074 * since we need to keep dirty segments precisely under SSR. 2075 */ 2076 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); 2077 2078 mutex_unlock(&sit_i->sentry_lock); 2079 2080 if (page && IS_NODESEG(type)) 2081 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 2082 2083 mutex_unlock(&curseg->curseg_mutex); 2084 } 2085 2086 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 2087 { 2088 int type = __get_segment_type(fio->page, fio->type); 2089 int err; 2090 2091 if (fio->type == NODE || fio->type == DATA) 2092 mutex_lock(&fio->sbi->wio_mutex[fio->type]); 2093 reallocate: 2094 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 2095 &fio->new_blkaddr, sum, type); 2096 2097 /* writeout dirty page into bdev */ 2098 err = f2fs_submit_page_mbio(fio); 2099 if (err == -EAGAIN) { 2100 fio->old_blkaddr = fio->new_blkaddr; 2101 goto reallocate; 2102 } 2103 2104 if (fio->type == NODE || fio->type == DATA) 2105 mutex_unlock(&fio->sbi->wio_mutex[fio->type]); 2106 } 2107 2108 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) 2109 { 2110 struct f2fs_io_info fio = { 2111 .sbi = sbi, 2112 .type = META, 2113 .op = REQ_OP_WRITE, 2114 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 2115 .old_blkaddr = page->index, 2116 .new_blkaddr = page->index, 2117 .page = page, 2118 .encrypted_page = NULL, 2119 }; 2120 2121 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 2122 fio.op_flags &= ~REQ_META; 2123 2124 set_page_writeback(page); 2125 f2fs_submit_page_mbio(&fio); 2126 } 2127 2128 void write_node_page(unsigned int nid, struct f2fs_io_info *fio) 2129 { 2130 struct f2fs_summary sum; 2131 2132 set_summary(&sum, nid, 0, 0); 2133 do_write_page(&sum, fio); 2134 } 2135 2136 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio) 2137 { 2138 struct f2fs_sb_info *sbi = fio->sbi; 2139 struct f2fs_summary sum; 2140 struct node_info ni; 2141 2142 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 2143 get_node_info(sbi, dn->nid, &ni); 2144 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 2145 do_write_page(&sum, fio); 2146 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 2147 } 2148 2149 int rewrite_data_page(struct f2fs_io_info *fio) 2150 { 2151 fio->new_blkaddr = fio->old_blkaddr; 2152 stat_inc_inplace_blocks(fio->sbi); 2153 return f2fs_submit_page_bio(fio); 2154 } 2155 2156 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 2157 block_t old_blkaddr, block_t new_blkaddr, 2158 bool recover_curseg, bool recover_newaddr) 2159 { 2160 struct sit_info *sit_i = SIT_I(sbi); 2161 struct curseg_info *curseg; 2162 unsigned int segno, old_cursegno; 2163 struct seg_entry *se; 2164 int type; 2165 unsigned short old_blkoff; 2166 2167 segno = GET_SEGNO(sbi, new_blkaddr); 2168 se = get_seg_entry(sbi, segno); 2169 type = se->type; 2170 2171 if (!recover_curseg) { 2172 /* for recovery flow */ 2173 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 2174 if (old_blkaddr == NULL_ADDR) 2175 type = CURSEG_COLD_DATA; 2176 else 2177 type = CURSEG_WARM_DATA; 2178 } 2179 } else { 2180 if (!IS_CURSEG(sbi, segno)) 2181 type = CURSEG_WARM_DATA; 2182 } 2183 2184 curseg = CURSEG_I(sbi, type); 2185 2186 mutex_lock(&curseg->curseg_mutex); 2187 mutex_lock(&sit_i->sentry_lock); 2188 2189 old_cursegno = curseg->segno; 2190 old_blkoff = curseg->next_blkoff; 2191 2192 /* change the current segment */ 2193 if (segno != curseg->segno) { 2194 curseg->next_segno = segno; 2195 change_curseg(sbi, type, true); 2196 } 2197 2198 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 2199 __add_sum_entry(sbi, type, sum); 2200 2201 if (!recover_curseg || recover_newaddr) 2202 update_sit_entry(sbi, new_blkaddr, 1); 2203 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 2204 update_sit_entry(sbi, old_blkaddr, -1); 2205 2206 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 2207 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 2208 2209 locate_dirty_segment(sbi, old_cursegno); 2210 2211 if (recover_curseg) { 2212 if (old_cursegno != curseg->segno) { 2213 curseg->next_segno = old_cursegno; 2214 change_curseg(sbi, type, true); 2215 } 2216 curseg->next_blkoff = old_blkoff; 2217 } 2218 2219 mutex_unlock(&sit_i->sentry_lock); 2220 mutex_unlock(&curseg->curseg_mutex); 2221 } 2222 2223 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 2224 block_t old_addr, block_t new_addr, 2225 unsigned char version, bool recover_curseg, 2226 bool recover_newaddr) 2227 { 2228 struct f2fs_summary sum; 2229 2230 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 2231 2232 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, 2233 recover_curseg, recover_newaddr); 2234 2235 f2fs_update_data_blkaddr(dn, new_addr); 2236 } 2237 2238 void f2fs_wait_on_page_writeback(struct page *page, 2239 enum page_type type, bool ordered) 2240 { 2241 if (PageWriteback(page)) { 2242 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 2243 2244 f2fs_submit_merged_bio_cond(sbi, page->mapping->host, 2245 0, page->index, type, WRITE); 2246 if (ordered) 2247 wait_on_page_writeback(page); 2248 else 2249 wait_for_stable_page(page); 2250 } 2251 } 2252 2253 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi, 2254 block_t blkaddr) 2255 { 2256 struct page *cpage; 2257 2258 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) 2259 return; 2260 2261 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 2262 if (cpage) { 2263 f2fs_wait_on_page_writeback(cpage, DATA, true); 2264 f2fs_put_page(cpage, 1); 2265 } 2266 } 2267 2268 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 2269 { 2270 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2271 struct curseg_info *seg_i; 2272 unsigned char *kaddr; 2273 struct page *page; 2274 block_t start; 2275 int i, j, offset; 2276 2277 start = start_sum_block(sbi); 2278 2279 page = get_meta_page(sbi, start++); 2280 kaddr = (unsigned char *)page_address(page); 2281 2282 /* Step 1: restore nat cache */ 2283 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 2284 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 2285 2286 /* Step 2: restore sit cache */ 2287 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 2288 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 2289 offset = 2 * SUM_JOURNAL_SIZE; 2290 2291 /* Step 3: restore summary entries */ 2292 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2293 unsigned short blk_off; 2294 unsigned int segno; 2295 2296 seg_i = CURSEG_I(sbi, i); 2297 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 2298 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 2299 seg_i->next_segno = segno; 2300 reset_curseg(sbi, i, 0); 2301 seg_i->alloc_type = ckpt->alloc_type[i]; 2302 seg_i->next_blkoff = blk_off; 2303 2304 if (seg_i->alloc_type == SSR) 2305 blk_off = sbi->blocks_per_seg; 2306 2307 for (j = 0; j < blk_off; j++) { 2308 struct f2fs_summary *s; 2309 s = (struct f2fs_summary *)(kaddr + offset); 2310 seg_i->sum_blk->entries[j] = *s; 2311 offset += SUMMARY_SIZE; 2312 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 2313 SUM_FOOTER_SIZE) 2314 continue; 2315 2316 f2fs_put_page(page, 1); 2317 page = NULL; 2318 2319 page = get_meta_page(sbi, start++); 2320 kaddr = (unsigned char *)page_address(page); 2321 offset = 0; 2322 } 2323 } 2324 f2fs_put_page(page, 1); 2325 return 0; 2326 } 2327 2328 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 2329 { 2330 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2331 struct f2fs_summary_block *sum; 2332 struct curseg_info *curseg; 2333 struct page *new; 2334 unsigned short blk_off; 2335 unsigned int segno = 0; 2336 block_t blk_addr = 0; 2337 2338 /* get segment number and block addr */ 2339 if (IS_DATASEG(type)) { 2340 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 2341 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 2342 CURSEG_HOT_DATA]); 2343 if (__exist_node_summaries(sbi)) 2344 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 2345 else 2346 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 2347 } else { 2348 segno = le32_to_cpu(ckpt->cur_node_segno[type - 2349 CURSEG_HOT_NODE]); 2350 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 2351 CURSEG_HOT_NODE]); 2352 if (__exist_node_summaries(sbi)) 2353 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 2354 type - CURSEG_HOT_NODE); 2355 else 2356 blk_addr = GET_SUM_BLOCK(sbi, segno); 2357 } 2358 2359 new = get_meta_page(sbi, blk_addr); 2360 sum = (struct f2fs_summary_block *)page_address(new); 2361 2362 if (IS_NODESEG(type)) { 2363 if (__exist_node_summaries(sbi)) { 2364 struct f2fs_summary *ns = &sum->entries[0]; 2365 int i; 2366 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 2367 ns->version = 0; 2368 ns->ofs_in_node = 0; 2369 } 2370 } else { 2371 int err; 2372 2373 err = restore_node_summary(sbi, segno, sum); 2374 if (err) { 2375 f2fs_put_page(new, 1); 2376 return err; 2377 } 2378 } 2379 } 2380 2381 /* set uncompleted segment to curseg */ 2382 curseg = CURSEG_I(sbi, type); 2383 mutex_lock(&curseg->curseg_mutex); 2384 2385 /* update journal info */ 2386 down_write(&curseg->journal_rwsem); 2387 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 2388 up_write(&curseg->journal_rwsem); 2389 2390 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 2391 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 2392 curseg->next_segno = segno; 2393 reset_curseg(sbi, type, 0); 2394 curseg->alloc_type = ckpt->alloc_type[type]; 2395 curseg->next_blkoff = blk_off; 2396 mutex_unlock(&curseg->curseg_mutex); 2397 f2fs_put_page(new, 1); 2398 return 0; 2399 } 2400 2401 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 2402 { 2403 int type = CURSEG_HOT_DATA; 2404 int err; 2405 2406 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 2407 int npages = npages_for_summary_flush(sbi, true); 2408 2409 if (npages >= 2) 2410 ra_meta_pages(sbi, start_sum_block(sbi), npages, 2411 META_CP, true); 2412 2413 /* restore for compacted data summary */ 2414 if (read_compacted_summaries(sbi)) 2415 return -EINVAL; 2416 type = CURSEG_HOT_NODE; 2417 } 2418 2419 if (__exist_node_summaries(sbi)) 2420 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), 2421 NR_CURSEG_TYPE - type, META_CP, true); 2422 2423 for (; type <= CURSEG_COLD_NODE; type++) { 2424 err = read_normal_summaries(sbi, type); 2425 if (err) 2426 return err; 2427 } 2428 2429 return 0; 2430 } 2431 2432 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 2433 { 2434 struct page *page; 2435 unsigned char *kaddr; 2436 struct f2fs_summary *summary; 2437 struct curseg_info *seg_i; 2438 int written_size = 0; 2439 int i, j; 2440 2441 page = grab_meta_page(sbi, blkaddr++); 2442 kaddr = (unsigned char *)page_address(page); 2443 2444 /* Step 1: write nat cache */ 2445 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 2446 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 2447 written_size += SUM_JOURNAL_SIZE; 2448 2449 /* Step 2: write sit cache */ 2450 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 2451 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 2452 written_size += SUM_JOURNAL_SIZE; 2453 2454 /* Step 3: write summary entries */ 2455 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2456 unsigned short blkoff; 2457 seg_i = CURSEG_I(sbi, i); 2458 if (sbi->ckpt->alloc_type[i] == SSR) 2459 blkoff = sbi->blocks_per_seg; 2460 else 2461 blkoff = curseg_blkoff(sbi, i); 2462 2463 for (j = 0; j < blkoff; j++) { 2464 if (!page) { 2465 page = grab_meta_page(sbi, blkaddr++); 2466 kaddr = (unsigned char *)page_address(page); 2467 written_size = 0; 2468 } 2469 summary = (struct f2fs_summary *)(kaddr + written_size); 2470 *summary = seg_i->sum_blk->entries[j]; 2471 written_size += SUMMARY_SIZE; 2472 2473 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 2474 SUM_FOOTER_SIZE) 2475 continue; 2476 2477 set_page_dirty(page); 2478 f2fs_put_page(page, 1); 2479 page = NULL; 2480 } 2481 } 2482 if (page) { 2483 set_page_dirty(page); 2484 f2fs_put_page(page, 1); 2485 } 2486 } 2487 2488 static void write_normal_summaries(struct f2fs_sb_info *sbi, 2489 block_t blkaddr, int type) 2490 { 2491 int i, end; 2492 if (IS_DATASEG(type)) 2493 end = type + NR_CURSEG_DATA_TYPE; 2494 else 2495 end = type + NR_CURSEG_NODE_TYPE; 2496 2497 for (i = type; i < end; i++) 2498 write_current_sum_page(sbi, i, blkaddr + (i - type)); 2499 } 2500 2501 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 2502 { 2503 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 2504 write_compacted_summaries(sbi, start_blk); 2505 else 2506 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 2507 } 2508 2509 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 2510 { 2511 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 2512 } 2513 2514 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 2515 unsigned int val, int alloc) 2516 { 2517 int i; 2518 2519 if (type == NAT_JOURNAL) { 2520 for (i = 0; i < nats_in_cursum(journal); i++) { 2521 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 2522 return i; 2523 } 2524 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 2525 return update_nats_in_cursum(journal, 1); 2526 } else if (type == SIT_JOURNAL) { 2527 for (i = 0; i < sits_in_cursum(journal); i++) 2528 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 2529 return i; 2530 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 2531 return update_sits_in_cursum(journal, 1); 2532 } 2533 return -1; 2534 } 2535 2536 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 2537 unsigned int segno) 2538 { 2539 return get_meta_page(sbi, current_sit_addr(sbi, segno)); 2540 } 2541 2542 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 2543 unsigned int start) 2544 { 2545 struct sit_info *sit_i = SIT_I(sbi); 2546 struct page *src_page, *dst_page; 2547 pgoff_t src_off, dst_off; 2548 void *src_addr, *dst_addr; 2549 2550 src_off = current_sit_addr(sbi, start); 2551 dst_off = next_sit_addr(sbi, src_off); 2552 2553 /* get current sit block page without lock */ 2554 src_page = get_meta_page(sbi, src_off); 2555 dst_page = grab_meta_page(sbi, dst_off); 2556 f2fs_bug_on(sbi, PageDirty(src_page)); 2557 2558 src_addr = page_address(src_page); 2559 dst_addr = page_address(dst_page); 2560 memcpy(dst_addr, src_addr, PAGE_SIZE); 2561 2562 set_page_dirty(dst_page); 2563 f2fs_put_page(src_page, 1); 2564 2565 set_to_next_sit(sit_i, start); 2566 2567 return dst_page; 2568 } 2569 2570 static struct sit_entry_set *grab_sit_entry_set(void) 2571 { 2572 struct sit_entry_set *ses = 2573 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); 2574 2575 ses->entry_cnt = 0; 2576 INIT_LIST_HEAD(&ses->set_list); 2577 return ses; 2578 } 2579 2580 static void release_sit_entry_set(struct sit_entry_set *ses) 2581 { 2582 list_del(&ses->set_list); 2583 kmem_cache_free(sit_entry_set_slab, ses); 2584 } 2585 2586 static void adjust_sit_entry_set(struct sit_entry_set *ses, 2587 struct list_head *head) 2588 { 2589 struct sit_entry_set *next = ses; 2590 2591 if (list_is_last(&ses->set_list, head)) 2592 return; 2593 2594 list_for_each_entry_continue(next, head, set_list) 2595 if (ses->entry_cnt <= next->entry_cnt) 2596 break; 2597 2598 list_move_tail(&ses->set_list, &next->set_list); 2599 } 2600 2601 static void add_sit_entry(unsigned int segno, struct list_head *head) 2602 { 2603 struct sit_entry_set *ses; 2604 unsigned int start_segno = START_SEGNO(segno); 2605 2606 list_for_each_entry(ses, head, set_list) { 2607 if (ses->start_segno == start_segno) { 2608 ses->entry_cnt++; 2609 adjust_sit_entry_set(ses, head); 2610 return; 2611 } 2612 } 2613 2614 ses = grab_sit_entry_set(); 2615 2616 ses->start_segno = start_segno; 2617 ses->entry_cnt++; 2618 list_add(&ses->set_list, head); 2619 } 2620 2621 static void add_sits_in_set(struct f2fs_sb_info *sbi) 2622 { 2623 struct f2fs_sm_info *sm_info = SM_I(sbi); 2624 struct list_head *set_list = &sm_info->sit_entry_set; 2625 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 2626 unsigned int segno; 2627 2628 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 2629 add_sit_entry(segno, set_list); 2630 } 2631 2632 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 2633 { 2634 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 2635 struct f2fs_journal *journal = curseg->journal; 2636 int i; 2637 2638 down_write(&curseg->journal_rwsem); 2639 for (i = 0; i < sits_in_cursum(journal); i++) { 2640 unsigned int segno; 2641 bool dirtied; 2642 2643 segno = le32_to_cpu(segno_in_journal(journal, i)); 2644 dirtied = __mark_sit_entry_dirty(sbi, segno); 2645 2646 if (!dirtied) 2647 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 2648 } 2649 update_sits_in_cursum(journal, -i); 2650 up_write(&curseg->journal_rwsem); 2651 } 2652 2653 /* 2654 * CP calls this function, which flushes SIT entries including sit_journal, 2655 * and moves prefree segs to free segs. 2656 */ 2657 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 2658 { 2659 struct sit_info *sit_i = SIT_I(sbi); 2660 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 2661 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 2662 struct f2fs_journal *journal = curseg->journal; 2663 struct sit_entry_set *ses, *tmp; 2664 struct list_head *head = &SM_I(sbi)->sit_entry_set; 2665 bool to_journal = true; 2666 struct seg_entry *se; 2667 2668 mutex_lock(&sit_i->sentry_lock); 2669 2670 if (!sit_i->dirty_sentries) 2671 goto out; 2672 2673 /* 2674 * add and account sit entries of dirty bitmap in sit entry 2675 * set temporarily 2676 */ 2677 add_sits_in_set(sbi); 2678 2679 /* 2680 * if there are no enough space in journal to store dirty sit 2681 * entries, remove all entries from journal and add and account 2682 * them in sit entry set. 2683 */ 2684 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL)) 2685 remove_sits_in_journal(sbi); 2686 2687 /* 2688 * there are two steps to flush sit entries: 2689 * #1, flush sit entries to journal in current cold data summary block. 2690 * #2, flush sit entries to sit page. 2691 */ 2692 list_for_each_entry_safe(ses, tmp, head, set_list) { 2693 struct page *page = NULL; 2694 struct f2fs_sit_block *raw_sit = NULL; 2695 unsigned int start_segno = ses->start_segno; 2696 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 2697 (unsigned long)MAIN_SEGS(sbi)); 2698 unsigned int segno = start_segno; 2699 2700 if (to_journal && 2701 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 2702 to_journal = false; 2703 2704 if (to_journal) { 2705 down_write(&curseg->journal_rwsem); 2706 } else { 2707 page = get_next_sit_page(sbi, start_segno); 2708 raw_sit = page_address(page); 2709 } 2710 2711 /* flush dirty sit entries in region of current sit set */ 2712 for_each_set_bit_from(segno, bitmap, end) { 2713 int offset, sit_offset; 2714 2715 se = get_seg_entry(sbi, segno); 2716 2717 /* add discard candidates */ 2718 if (cpc->reason != CP_DISCARD) { 2719 cpc->trim_start = segno; 2720 add_discard_addrs(sbi, cpc, false); 2721 } 2722 2723 if (to_journal) { 2724 offset = lookup_journal_in_cursum(journal, 2725 SIT_JOURNAL, segno, 1); 2726 f2fs_bug_on(sbi, offset < 0); 2727 segno_in_journal(journal, offset) = 2728 cpu_to_le32(segno); 2729 seg_info_to_raw_sit(se, 2730 &sit_in_journal(journal, offset)); 2731 } else { 2732 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 2733 seg_info_to_raw_sit(se, 2734 &raw_sit->entries[sit_offset]); 2735 } 2736 2737 __clear_bit(segno, bitmap); 2738 sit_i->dirty_sentries--; 2739 ses->entry_cnt--; 2740 } 2741 2742 if (to_journal) 2743 up_write(&curseg->journal_rwsem); 2744 else 2745 f2fs_put_page(page, 1); 2746 2747 f2fs_bug_on(sbi, ses->entry_cnt); 2748 release_sit_entry_set(ses); 2749 } 2750 2751 f2fs_bug_on(sbi, !list_empty(head)); 2752 f2fs_bug_on(sbi, sit_i->dirty_sentries); 2753 out: 2754 if (cpc->reason == CP_DISCARD) { 2755 __u64 trim_start = cpc->trim_start; 2756 2757 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 2758 add_discard_addrs(sbi, cpc, false); 2759 2760 cpc->trim_start = trim_start; 2761 } 2762 mutex_unlock(&sit_i->sentry_lock); 2763 2764 set_prefree_as_free_segments(sbi); 2765 } 2766 2767 static int build_sit_info(struct f2fs_sb_info *sbi) 2768 { 2769 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 2770 struct sit_info *sit_i; 2771 unsigned int sit_segs, start; 2772 char *src_bitmap; 2773 unsigned int bitmap_size; 2774 2775 /* allocate memory for SIT information */ 2776 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); 2777 if (!sit_i) 2778 return -ENOMEM; 2779 2780 SM_I(sbi)->sit_info = sit_i; 2781 2782 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) * 2783 sizeof(struct seg_entry), GFP_KERNEL); 2784 if (!sit_i->sentries) 2785 return -ENOMEM; 2786 2787 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 2788 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 2789 if (!sit_i->dirty_sentries_bitmap) 2790 return -ENOMEM; 2791 2792 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2793 sit_i->sentries[start].cur_valid_map 2794 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2795 sit_i->sentries[start].ckpt_valid_map 2796 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2797 if (!sit_i->sentries[start].cur_valid_map || 2798 !sit_i->sentries[start].ckpt_valid_map) 2799 return -ENOMEM; 2800 2801 #ifdef CONFIG_F2FS_CHECK_FS 2802 sit_i->sentries[start].cur_valid_map_mir 2803 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2804 if (!sit_i->sentries[start].cur_valid_map_mir) 2805 return -ENOMEM; 2806 #endif 2807 2808 if (f2fs_discard_en(sbi)) { 2809 sit_i->sentries[start].discard_map 2810 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2811 if (!sit_i->sentries[start].discard_map) 2812 return -ENOMEM; 2813 } 2814 } 2815 2816 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2817 if (!sit_i->tmp_map) 2818 return -ENOMEM; 2819 2820 if (sbi->segs_per_sec > 1) { 2821 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) * 2822 sizeof(struct sec_entry), GFP_KERNEL); 2823 if (!sit_i->sec_entries) 2824 return -ENOMEM; 2825 } 2826 2827 /* get information related with SIT */ 2828 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 2829 2830 /* setup SIT bitmap from ckeckpoint pack */ 2831 bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 2832 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 2833 2834 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 2835 if (!sit_i->sit_bitmap) 2836 return -ENOMEM; 2837 2838 #ifdef CONFIG_F2FS_CHECK_FS 2839 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 2840 if (!sit_i->sit_bitmap_mir) 2841 return -ENOMEM; 2842 #endif 2843 2844 /* init SIT information */ 2845 sit_i->s_ops = &default_salloc_ops; 2846 2847 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 2848 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 2849 sit_i->written_valid_blocks = 0; 2850 sit_i->bitmap_size = bitmap_size; 2851 sit_i->dirty_sentries = 0; 2852 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 2853 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 2854 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; 2855 mutex_init(&sit_i->sentry_lock); 2856 return 0; 2857 } 2858 2859 static int build_free_segmap(struct f2fs_sb_info *sbi) 2860 { 2861 struct free_segmap_info *free_i; 2862 unsigned int bitmap_size, sec_bitmap_size; 2863 2864 /* allocate memory for free segmap information */ 2865 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); 2866 if (!free_i) 2867 return -ENOMEM; 2868 2869 SM_I(sbi)->free_info = free_i; 2870 2871 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 2872 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL); 2873 if (!free_i->free_segmap) 2874 return -ENOMEM; 2875 2876 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 2877 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL); 2878 if (!free_i->free_secmap) 2879 return -ENOMEM; 2880 2881 /* set all segments as dirty temporarily */ 2882 memset(free_i->free_segmap, 0xff, bitmap_size); 2883 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 2884 2885 /* init free segmap information */ 2886 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 2887 free_i->free_segments = 0; 2888 free_i->free_sections = 0; 2889 spin_lock_init(&free_i->segmap_lock); 2890 return 0; 2891 } 2892 2893 static int build_curseg(struct f2fs_sb_info *sbi) 2894 { 2895 struct curseg_info *array; 2896 int i; 2897 2898 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL); 2899 if (!array) 2900 return -ENOMEM; 2901 2902 SM_I(sbi)->curseg_array = array; 2903 2904 for (i = 0; i < NR_CURSEG_TYPE; i++) { 2905 mutex_init(&array[i].curseg_mutex); 2906 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL); 2907 if (!array[i].sum_blk) 2908 return -ENOMEM; 2909 init_rwsem(&array[i].journal_rwsem); 2910 array[i].journal = kzalloc(sizeof(struct f2fs_journal), 2911 GFP_KERNEL); 2912 if (!array[i].journal) 2913 return -ENOMEM; 2914 array[i].segno = NULL_SEGNO; 2915 array[i].next_blkoff = 0; 2916 } 2917 return restore_curseg_summaries(sbi); 2918 } 2919 2920 static void build_sit_entries(struct f2fs_sb_info *sbi) 2921 { 2922 struct sit_info *sit_i = SIT_I(sbi); 2923 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 2924 struct f2fs_journal *journal = curseg->journal; 2925 struct seg_entry *se; 2926 struct f2fs_sit_entry sit; 2927 int sit_blk_cnt = SIT_BLK_CNT(sbi); 2928 unsigned int i, start, end; 2929 unsigned int readed, start_blk = 0; 2930 2931 do { 2932 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES, 2933 META_SIT, true); 2934 2935 start = start_blk * sit_i->sents_per_block; 2936 end = (start_blk + readed) * sit_i->sents_per_block; 2937 2938 for (; start < end && start < MAIN_SEGS(sbi); start++) { 2939 struct f2fs_sit_block *sit_blk; 2940 struct page *page; 2941 2942 se = &sit_i->sentries[start]; 2943 page = get_current_sit_page(sbi, start); 2944 sit_blk = (struct f2fs_sit_block *)page_address(page); 2945 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 2946 f2fs_put_page(page, 1); 2947 2948 check_block_count(sbi, start, &sit); 2949 seg_info_from_raw_sit(se, &sit); 2950 2951 /* build discard map only one time */ 2952 if (f2fs_discard_en(sbi)) { 2953 memcpy(se->discard_map, se->cur_valid_map, 2954 SIT_VBLOCK_MAP_SIZE); 2955 sbi->discard_blks += sbi->blocks_per_seg - 2956 se->valid_blocks; 2957 } 2958 2959 if (sbi->segs_per_sec > 1) 2960 get_sec_entry(sbi, start)->valid_blocks += 2961 se->valid_blocks; 2962 } 2963 start_blk += readed; 2964 } while (start_blk < sit_blk_cnt); 2965 2966 down_read(&curseg->journal_rwsem); 2967 for (i = 0; i < sits_in_cursum(journal); i++) { 2968 unsigned int old_valid_blocks; 2969 2970 start = le32_to_cpu(segno_in_journal(journal, i)); 2971 se = &sit_i->sentries[start]; 2972 sit = sit_in_journal(journal, i); 2973 2974 old_valid_blocks = se->valid_blocks; 2975 2976 check_block_count(sbi, start, &sit); 2977 seg_info_from_raw_sit(se, &sit); 2978 2979 if (f2fs_discard_en(sbi)) { 2980 memcpy(se->discard_map, se->cur_valid_map, 2981 SIT_VBLOCK_MAP_SIZE); 2982 sbi->discard_blks += old_valid_blocks - 2983 se->valid_blocks; 2984 } 2985 2986 if (sbi->segs_per_sec > 1) 2987 get_sec_entry(sbi, start)->valid_blocks += 2988 se->valid_blocks - old_valid_blocks; 2989 } 2990 up_read(&curseg->journal_rwsem); 2991 } 2992 2993 static void init_free_segmap(struct f2fs_sb_info *sbi) 2994 { 2995 unsigned int start; 2996 int type; 2997 2998 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2999 struct seg_entry *sentry = get_seg_entry(sbi, start); 3000 if (!sentry->valid_blocks) 3001 __set_free(sbi, start); 3002 else 3003 SIT_I(sbi)->written_valid_blocks += 3004 sentry->valid_blocks; 3005 } 3006 3007 /* set use the current segments */ 3008 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 3009 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 3010 __set_test_and_inuse(sbi, curseg_t->segno); 3011 } 3012 } 3013 3014 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 3015 { 3016 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3017 struct free_segmap_info *free_i = FREE_I(sbi); 3018 unsigned int segno = 0, offset = 0; 3019 unsigned short valid_blocks; 3020 3021 while (1) { 3022 /* find dirty segment based on free segmap */ 3023 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 3024 if (segno >= MAIN_SEGS(sbi)) 3025 break; 3026 offset = segno + 1; 3027 valid_blocks = get_valid_blocks(sbi, segno, false); 3028 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) 3029 continue; 3030 if (valid_blocks > sbi->blocks_per_seg) { 3031 f2fs_bug_on(sbi, 1); 3032 continue; 3033 } 3034 mutex_lock(&dirty_i->seglist_lock); 3035 __locate_dirty_segment(sbi, segno, DIRTY); 3036 mutex_unlock(&dirty_i->seglist_lock); 3037 } 3038 } 3039 3040 static int init_victim_secmap(struct f2fs_sb_info *sbi) 3041 { 3042 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3043 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 3044 3045 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 3046 if (!dirty_i->victim_secmap) 3047 return -ENOMEM; 3048 return 0; 3049 } 3050 3051 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 3052 { 3053 struct dirty_seglist_info *dirty_i; 3054 unsigned int bitmap_size, i; 3055 3056 /* allocate memory for dirty segments list information */ 3057 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); 3058 if (!dirty_i) 3059 return -ENOMEM; 3060 3061 SM_I(sbi)->dirty_info = dirty_i; 3062 mutex_init(&dirty_i->seglist_lock); 3063 3064 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 3065 3066 for (i = 0; i < NR_DIRTY_TYPE; i++) { 3067 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 3068 if (!dirty_i->dirty_segmap[i]) 3069 return -ENOMEM; 3070 } 3071 3072 init_dirty_segmap(sbi); 3073 return init_victim_secmap(sbi); 3074 } 3075 3076 /* 3077 * Update min, max modified time for cost-benefit GC algorithm 3078 */ 3079 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 3080 { 3081 struct sit_info *sit_i = SIT_I(sbi); 3082 unsigned int segno; 3083 3084 mutex_lock(&sit_i->sentry_lock); 3085 3086 sit_i->min_mtime = LLONG_MAX; 3087 3088 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 3089 unsigned int i; 3090 unsigned long long mtime = 0; 3091 3092 for (i = 0; i < sbi->segs_per_sec; i++) 3093 mtime += get_seg_entry(sbi, segno + i)->mtime; 3094 3095 mtime = div_u64(mtime, sbi->segs_per_sec); 3096 3097 if (sit_i->min_mtime > mtime) 3098 sit_i->min_mtime = mtime; 3099 } 3100 sit_i->max_mtime = get_mtime(sbi); 3101 mutex_unlock(&sit_i->sentry_lock); 3102 } 3103 3104 int build_segment_manager(struct f2fs_sb_info *sbi) 3105 { 3106 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 3107 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3108 struct f2fs_sm_info *sm_info; 3109 int err; 3110 3111 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); 3112 if (!sm_info) 3113 return -ENOMEM; 3114 3115 /* init sm info */ 3116 sbi->sm_info = sm_info; 3117 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 3118 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 3119 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 3120 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 3121 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 3122 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 3123 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 3124 sm_info->rec_prefree_segments = sm_info->main_segments * 3125 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 3126 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 3127 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 3128 3129 if (!test_opt(sbi, LFS)) 3130 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 3131 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 3132 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 3133 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 3134 3135 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS; 3136 3137 INIT_LIST_HEAD(&sm_info->sit_entry_set); 3138 3139 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) { 3140 err = create_flush_cmd_control(sbi); 3141 if (err) 3142 return err; 3143 } 3144 3145 err = create_discard_cmd_control(sbi); 3146 if (err) 3147 return err; 3148 3149 err = build_sit_info(sbi); 3150 if (err) 3151 return err; 3152 err = build_free_segmap(sbi); 3153 if (err) 3154 return err; 3155 err = build_curseg(sbi); 3156 if (err) 3157 return err; 3158 3159 /* reinit free segmap based on SIT */ 3160 build_sit_entries(sbi); 3161 3162 init_free_segmap(sbi); 3163 err = build_dirty_segmap(sbi); 3164 if (err) 3165 return err; 3166 3167 init_min_max_mtime(sbi); 3168 return 0; 3169 } 3170 3171 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 3172 enum dirty_type dirty_type) 3173 { 3174 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3175 3176 mutex_lock(&dirty_i->seglist_lock); 3177 kvfree(dirty_i->dirty_segmap[dirty_type]); 3178 dirty_i->nr_dirty[dirty_type] = 0; 3179 mutex_unlock(&dirty_i->seglist_lock); 3180 } 3181 3182 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 3183 { 3184 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3185 kvfree(dirty_i->victim_secmap); 3186 } 3187 3188 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 3189 { 3190 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3191 int i; 3192 3193 if (!dirty_i) 3194 return; 3195 3196 /* discard pre-free/dirty segments list */ 3197 for (i = 0; i < NR_DIRTY_TYPE; i++) 3198 discard_dirty_segmap(sbi, i); 3199 3200 destroy_victim_secmap(sbi); 3201 SM_I(sbi)->dirty_info = NULL; 3202 kfree(dirty_i); 3203 } 3204 3205 static void destroy_curseg(struct f2fs_sb_info *sbi) 3206 { 3207 struct curseg_info *array = SM_I(sbi)->curseg_array; 3208 int i; 3209 3210 if (!array) 3211 return; 3212 SM_I(sbi)->curseg_array = NULL; 3213 for (i = 0; i < NR_CURSEG_TYPE; i++) { 3214 kfree(array[i].sum_blk); 3215 kfree(array[i].journal); 3216 } 3217 kfree(array); 3218 } 3219 3220 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 3221 { 3222 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 3223 if (!free_i) 3224 return; 3225 SM_I(sbi)->free_info = NULL; 3226 kvfree(free_i->free_segmap); 3227 kvfree(free_i->free_secmap); 3228 kfree(free_i); 3229 } 3230 3231 static void destroy_sit_info(struct f2fs_sb_info *sbi) 3232 { 3233 struct sit_info *sit_i = SIT_I(sbi); 3234 unsigned int start; 3235 3236 if (!sit_i) 3237 return; 3238 3239 if (sit_i->sentries) { 3240 for (start = 0; start < MAIN_SEGS(sbi); start++) { 3241 kfree(sit_i->sentries[start].cur_valid_map); 3242 #ifdef CONFIG_F2FS_CHECK_FS 3243 kfree(sit_i->sentries[start].cur_valid_map_mir); 3244 #endif 3245 kfree(sit_i->sentries[start].ckpt_valid_map); 3246 kfree(sit_i->sentries[start].discard_map); 3247 } 3248 } 3249 kfree(sit_i->tmp_map); 3250 3251 kvfree(sit_i->sentries); 3252 kvfree(sit_i->sec_entries); 3253 kvfree(sit_i->dirty_sentries_bitmap); 3254 3255 SM_I(sbi)->sit_info = NULL; 3256 kfree(sit_i->sit_bitmap); 3257 #ifdef CONFIG_F2FS_CHECK_FS 3258 kfree(sit_i->sit_bitmap_mir); 3259 #endif 3260 kfree(sit_i); 3261 } 3262 3263 void destroy_segment_manager(struct f2fs_sb_info *sbi) 3264 { 3265 struct f2fs_sm_info *sm_info = SM_I(sbi); 3266 3267 if (!sm_info) 3268 return; 3269 destroy_flush_cmd_control(sbi, true); 3270 destroy_discard_cmd_control(sbi); 3271 destroy_dirty_segmap(sbi); 3272 destroy_curseg(sbi); 3273 destroy_free_segmap(sbi); 3274 destroy_sit_info(sbi); 3275 sbi->sm_info = NULL; 3276 kfree(sm_info); 3277 } 3278 3279 int __init create_segment_manager_caches(void) 3280 { 3281 discard_entry_slab = f2fs_kmem_cache_create("discard_entry", 3282 sizeof(struct discard_entry)); 3283 if (!discard_entry_slab) 3284 goto fail; 3285 3286 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd", 3287 sizeof(struct discard_cmd)); 3288 if (!discard_cmd_slab) 3289 goto destroy_discard_entry; 3290 3291 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", 3292 sizeof(struct sit_entry_set)); 3293 if (!sit_entry_set_slab) 3294 goto destroy_discard_cmd; 3295 3296 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", 3297 sizeof(struct inmem_pages)); 3298 if (!inmem_entry_slab) 3299 goto destroy_sit_entry_set; 3300 return 0; 3301 3302 destroy_sit_entry_set: 3303 kmem_cache_destroy(sit_entry_set_slab); 3304 destroy_discard_cmd: 3305 kmem_cache_destroy(discard_cmd_slab); 3306 destroy_discard_entry: 3307 kmem_cache_destroy(discard_entry_slab); 3308 fail: 3309 return -ENOMEM; 3310 } 3311 3312 void destroy_segment_manager_caches(void) 3313 { 3314 kmem_cache_destroy(sit_entry_set_slab); 3315 kmem_cache_destroy(discard_cmd_slab); 3316 kmem_cache_destroy(discard_entry_slab); 3317 kmem_cache_destroy(inmem_entry_slab); 3318 } 3319