1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/segment.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/f2fs_fs.h> 10 #include <linux/bio.h> 11 #include <linux/blkdev.h> 12 #include <linux/sched/mm.h> 13 #include <linux/prefetch.h> 14 #include <linux/kthread.h> 15 #include <linux/swap.h> 16 #include <linux/timer.h> 17 #include <linux/freezer.h> 18 #include <linux/sched/signal.h> 19 #include <linux/random.h> 20 21 #include "f2fs.h" 22 #include "segment.h" 23 #include "node.h" 24 #include "gc.h" 25 #include "iostat.h" 26 #include <trace/events/f2fs.h> 27 28 #define __reverse_ffz(x) __reverse_ffs(~(x)) 29 30 static struct kmem_cache *discard_entry_slab; 31 static struct kmem_cache *discard_cmd_slab; 32 static struct kmem_cache *sit_entry_set_slab; 33 static struct kmem_cache *inmem_entry_slab; 34 35 static unsigned long __reverse_ulong(unsigned char *str) 36 { 37 unsigned long tmp = 0; 38 int shift = 24, idx = 0; 39 40 #if BITS_PER_LONG == 64 41 shift = 56; 42 #endif 43 while (shift >= 0) { 44 tmp |= (unsigned long)str[idx++] << shift; 45 shift -= BITS_PER_BYTE; 46 } 47 return tmp; 48 } 49 50 /* 51 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since 52 * MSB and LSB are reversed in a byte by f2fs_set_bit. 53 */ 54 static inline unsigned long __reverse_ffs(unsigned long word) 55 { 56 int num = 0; 57 58 #if BITS_PER_LONG == 64 59 if ((word & 0xffffffff00000000UL) == 0) 60 num += 32; 61 else 62 word >>= 32; 63 #endif 64 if ((word & 0xffff0000) == 0) 65 num += 16; 66 else 67 word >>= 16; 68 69 if ((word & 0xff00) == 0) 70 num += 8; 71 else 72 word >>= 8; 73 74 if ((word & 0xf0) == 0) 75 num += 4; 76 else 77 word >>= 4; 78 79 if ((word & 0xc) == 0) 80 num += 2; 81 else 82 word >>= 2; 83 84 if ((word & 0x2) == 0) 85 num += 1; 86 return num; 87 } 88 89 /* 90 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because 91 * f2fs_set_bit makes MSB and LSB reversed in a byte. 92 * @size must be integral times of unsigned long. 93 * Example: 94 * MSB <--> LSB 95 * f2fs_set_bit(0, bitmap) => 1000 0000 96 * f2fs_set_bit(7, bitmap) => 0000 0001 97 */ 98 static unsigned long __find_rev_next_bit(const unsigned long *addr, 99 unsigned long size, unsigned long offset) 100 { 101 const unsigned long *p = addr + BIT_WORD(offset); 102 unsigned long result = size; 103 unsigned long tmp; 104 105 if (offset >= size) 106 return size; 107 108 size -= (offset & ~(BITS_PER_LONG - 1)); 109 offset %= BITS_PER_LONG; 110 111 while (1) { 112 if (*p == 0) 113 goto pass; 114 115 tmp = __reverse_ulong((unsigned char *)p); 116 117 tmp &= ~0UL >> offset; 118 if (size < BITS_PER_LONG) 119 tmp &= (~0UL << (BITS_PER_LONG - size)); 120 if (tmp) 121 goto found; 122 pass: 123 if (size <= BITS_PER_LONG) 124 break; 125 size -= BITS_PER_LONG; 126 offset = 0; 127 p++; 128 } 129 return result; 130 found: 131 return result - size + __reverse_ffs(tmp); 132 } 133 134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, 135 unsigned long size, unsigned long offset) 136 { 137 const unsigned long *p = addr + BIT_WORD(offset); 138 unsigned long result = size; 139 unsigned long tmp; 140 141 if (offset >= size) 142 return size; 143 144 size -= (offset & ~(BITS_PER_LONG - 1)); 145 offset %= BITS_PER_LONG; 146 147 while (1) { 148 if (*p == ~0UL) 149 goto pass; 150 151 tmp = __reverse_ulong((unsigned char *)p); 152 153 if (offset) 154 tmp |= ~0UL << (BITS_PER_LONG - offset); 155 if (size < BITS_PER_LONG) 156 tmp |= ~0UL >> size; 157 if (tmp != ~0UL) 158 goto found; 159 pass: 160 if (size <= BITS_PER_LONG) 161 break; 162 size -= BITS_PER_LONG; 163 offset = 0; 164 p++; 165 } 166 return result; 167 found: 168 return result - size + __reverse_ffz(tmp); 169 } 170 171 bool f2fs_need_SSR(struct f2fs_sb_info *sbi) 172 { 173 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES); 174 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS); 175 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA); 176 177 if (f2fs_lfs_mode(sbi)) 178 return false; 179 if (sbi->gc_mode == GC_URGENT_HIGH) 180 return true; 181 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 182 return true; 183 184 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs + 185 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi)); 186 } 187 188 void f2fs_register_inmem_page(struct inode *inode, struct page *page) 189 { 190 struct inmem_pages *new; 191 192 set_page_private_atomic(page); 193 194 new = f2fs_kmem_cache_alloc(inmem_entry_slab, 195 GFP_NOFS, true, NULL); 196 197 /* add atomic page indices to the list */ 198 new->page = page; 199 INIT_LIST_HEAD(&new->list); 200 201 /* increase reference count with clean state */ 202 get_page(page); 203 mutex_lock(&F2FS_I(inode)->inmem_lock); 204 list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages); 205 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 206 mutex_unlock(&F2FS_I(inode)->inmem_lock); 207 208 trace_f2fs_register_inmem_page(page, INMEM); 209 } 210 211 static int __revoke_inmem_pages(struct inode *inode, 212 struct list_head *head, bool drop, bool recover, 213 bool trylock) 214 { 215 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 216 struct inmem_pages *cur, *tmp; 217 int err = 0; 218 219 list_for_each_entry_safe(cur, tmp, head, list) { 220 struct page *page = cur->page; 221 222 if (drop) 223 trace_f2fs_commit_inmem_page(page, INMEM_DROP); 224 225 if (trylock) { 226 /* 227 * to avoid deadlock in between page lock and 228 * inmem_lock. 229 */ 230 if (!trylock_page(page)) 231 continue; 232 } else { 233 lock_page(page); 234 } 235 236 f2fs_wait_on_page_writeback(page, DATA, true, true); 237 238 if (recover) { 239 struct dnode_of_data dn; 240 struct node_info ni; 241 242 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE); 243 retry: 244 set_new_dnode(&dn, inode, NULL, NULL, 0); 245 err = f2fs_get_dnode_of_data(&dn, page->index, 246 LOOKUP_NODE); 247 if (err) { 248 if (err == -ENOMEM) { 249 memalloc_retry_wait(GFP_NOFS); 250 goto retry; 251 } 252 err = -EAGAIN; 253 goto next; 254 } 255 256 err = f2fs_get_node_info(sbi, dn.nid, &ni, false); 257 if (err) { 258 f2fs_put_dnode(&dn); 259 return err; 260 } 261 262 if (cur->old_addr == NEW_ADDR) { 263 f2fs_invalidate_blocks(sbi, dn.data_blkaddr); 264 f2fs_update_data_blkaddr(&dn, NEW_ADDR); 265 } else 266 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 267 cur->old_addr, ni.version, true, true); 268 f2fs_put_dnode(&dn); 269 } 270 next: 271 /* we don't need to invalidate this in the sccessful status */ 272 if (drop || recover) { 273 ClearPageUptodate(page); 274 clear_page_private_gcing(page); 275 } 276 detach_page_private(page); 277 set_page_private(page, 0); 278 f2fs_put_page(page, 1); 279 280 list_del(&cur->list); 281 kmem_cache_free(inmem_entry_slab, cur); 282 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 283 } 284 return err; 285 } 286 287 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure) 288 { 289 struct list_head *head = &sbi->inode_list[ATOMIC_FILE]; 290 struct inode *inode; 291 struct f2fs_inode_info *fi; 292 unsigned int count = sbi->atomic_files; 293 unsigned int looped = 0; 294 next: 295 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 296 if (list_empty(head)) { 297 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 298 return; 299 } 300 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist); 301 inode = igrab(&fi->vfs_inode); 302 if (inode) 303 list_move_tail(&fi->inmem_ilist, head); 304 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 305 306 if (inode) { 307 if (gc_failure) { 308 if (!fi->i_gc_failures[GC_FAILURE_ATOMIC]) 309 goto skip; 310 } 311 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST); 312 f2fs_drop_inmem_pages(inode); 313 skip: 314 iput(inode); 315 } 316 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT); 317 cond_resched(); 318 if (gc_failure) { 319 if (++looped >= count) 320 return; 321 } 322 goto next; 323 } 324 325 void f2fs_drop_inmem_pages(struct inode *inode) 326 { 327 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 328 struct f2fs_inode_info *fi = F2FS_I(inode); 329 330 do { 331 mutex_lock(&fi->inmem_lock); 332 if (list_empty(&fi->inmem_pages)) { 333 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0; 334 335 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 336 if (!list_empty(&fi->inmem_ilist)) 337 list_del_init(&fi->inmem_ilist); 338 if (f2fs_is_atomic_file(inode)) { 339 clear_inode_flag(inode, FI_ATOMIC_FILE); 340 sbi->atomic_files--; 341 } 342 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 343 344 mutex_unlock(&fi->inmem_lock); 345 break; 346 } 347 __revoke_inmem_pages(inode, &fi->inmem_pages, 348 true, false, true); 349 mutex_unlock(&fi->inmem_lock); 350 } while (1); 351 } 352 353 void f2fs_drop_inmem_page(struct inode *inode, struct page *page) 354 { 355 struct f2fs_inode_info *fi = F2FS_I(inode); 356 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 357 struct list_head *head = &fi->inmem_pages; 358 struct inmem_pages *cur = NULL; 359 360 f2fs_bug_on(sbi, !page_private_atomic(page)); 361 362 mutex_lock(&fi->inmem_lock); 363 list_for_each_entry(cur, head, list) { 364 if (cur->page == page) 365 break; 366 } 367 368 f2fs_bug_on(sbi, list_empty(head) || cur->page != page); 369 list_del(&cur->list); 370 mutex_unlock(&fi->inmem_lock); 371 372 dec_page_count(sbi, F2FS_INMEM_PAGES); 373 kmem_cache_free(inmem_entry_slab, cur); 374 375 ClearPageUptodate(page); 376 clear_page_private_atomic(page); 377 f2fs_put_page(page, 0); 378 379 detach_page_private(page); 380 set_page_private(page, 0); 381 382 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE); 383 } 384 385 static int __f2fs_commit_inmem_pages(struct inode *inode) 386 { 387 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 388 struct f2fs_inode_info *fi = F2FS_I(inode); 389 struct inmem_pages *cur, *tmp; 390 struct f2fs_io_info fio = { 391 .sbi = sbi, 392 .ino = inode->i_ino, 393 .type = DATA, 394 .op = REQ_OP_WRITE, 395 .op_flags = REQ_SYNC | REQ_PRIO, 396 .io_type = FS_DATA_IO, 397 }; 398 struct list_head revoke_list; 399 bool submit_bio = false; 400 int err = 0; 401 402 INIT_LIST_HEAD(&revoke_list); 403 404 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { 405 struct page *page = cur->page; 406 407 lock_page(page); 408 if (page->mapping == inode->i_mapping) { 409 trace_f2fs_commit_inmem_page(page, INMEM); 410 411 f2fs_wait_on_page_writeback(page, DATA, true, true); 412 413 set_page_dirty(page); 414 if (clear_page_dirty_for_io(page)) { 415 inode_dec_dirty_pages(inode); 416 f2fs_remove_dirty_inode(inode); 417 } 418 retry: 419 fio.page = page; 420 fio.old_blkaddr = NULL_ADDR; 421 fio.encrypted_page = NULL; 422 fio.need_lock = LOCK_DONE; 423 err = f2fs_do_write_data_page(&fio); 424 if (err) { 425 if (err == -ENOMEM) { 426 memalloc_retry_wait(GFP_NOFS); 427 goto retry; 428 } 429 unlock_page(page); 430 break; 431 } 432 /* record old blkaddr for revoking */ 433 cur->old_addr = fio.old_blkaddr; 434 submit_bio = true; 435 } 436 unlock_page(page); 437 list_move_tail(&cur->list, &revoke_list); 438 } 439 440 if (submit_bio) 441 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA); 442 443 if (err) { 444 /* 445 * try to revoke all committed pages, but still we could fail 446 * due to no memory or other reason, if that happened, EAGAIN 447 * will be returned, which means in such case, transaction is 448 * already not integrity, caller should use journal to do the 449 * recovery or rewrite & commit last transaction. For other 450 * error number, revoking was done by filesystem itself. 451 */ 452 err = __revoke_inmem_pages(inode, &revoke_list, 453 false, true, false); 454 455 /* drop all uncommitted pages */ 456 __revoke_inmem_pages(inode, &fi->inmem_pages, 457 true, false, false); 458 } else { 459 __revoke_inmem_pages(inode, &revoke_list, 460 false, false, false); 461 } 462 463 return err; 464 } 465 466 int f2fs_commit_inmem_pages(struct inode *inode) 467 { 468 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 469 struct f2fs_inode_info *fi = F2FS_I(inode); 470 int err; 471 472 f2fs_balance_fs(sbi, true); 473 474 down_write(&fi->i_gc_rwsem[WRITE]); 475 476 f2fs_lock_op(sbi); 477 set_inode_flag(inode, FI_ATOMIC_COMMIT); 478 479 mutex_lock(&fi->inmem_lock); 480 err = __f2fs_commit_inmem_pages(inode); 481 mutex_unlock(&fi->inmem_lock); 482 483 clear_inode_flag(inode, FI_ATOMIC_COMMIT); 484 485 f2fs_unlock_op(sbi); 486 up_write(&fi->i_gc_rwsem[WRITE]); 487 488 return err; 489 } 490 491 /* 492 * This function balances dirty node and dentry pages. 493 * In addition, it controls garbage collection. 494 */ 495 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 496 { 497 if (time_to_inject(sbi, FAULT_CHECKPOINT)) { 498 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT); 499 f2fs_stop_checkpoint(sbi, false); 500 } 501 502 /* balance_fs_bg is able to be pending */ 503 if (need && excess_cached_nats(sbi)) 504 f2fs_balance_fs_bg(sbi, false); 505 506 if (!f2fs_is_checkpoint_ready(sbi)) 507 return; 508 509 /* 510 * We should do GC or end up with checkpoint, if there are so many dirty 511 * dir/node pages without enough free segments. 512 */ 513 if (has_not_enough_free_secs(sbi, 0, 0)) { 514 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread && 515 sbi->gc_thread->f2fs_gc_task) { 516 DEFINE_WAIT(wait); 517 518 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait, 519 TASK_UNINTERRUPTIBLE); 520 wake_up(&sbi->gc_thread->gc_wait_queue_head); 521 io_schedule(); 522 finish_wait(&sbi->gc_thread->fggc_wq, &wait); 523 } else { 524 down_write(&sbi->gc_lock); 525 f2fs_gc(sbi, false, false, false, NULL_SEGNO); 526 } 527 } 528 } 529 530 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi) 531 { 532 int factor = rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2; 533 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS); 534 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA); 535 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES); 536 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META); 537 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA); 538 unsigned int threshold = sbi->blocks_per_seg * factor * 539 DEFAULT_DIRTY_THRESHOLD; 540 unsigned int global_threshold = threshold * 3 / 2; 541 542 if (dents >= threshold || qdata >= threshold || 543 nodes >= threshold || meta >= threshold || 544 imeta >= threshold) 545 return true; 546 return dents + qdata + nodes + meta + imeta > global_threshold; 547 } 548 549 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 550 { 551 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 552 return; 553 554 /* try to shrink extent cache when there is no enough memory */ 555 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE)) 556 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); 557 558 /* check the # of cached NAT entries */ 559 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 560 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 561 562 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 563 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 564 else 565 f2fs_build_free_nids(sbi, false, false); 566 567 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) || 568 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi)) 569 goto do_sync; 570 571 /* there is background inflight IO or foreground operation recently */ 572 if (is_inflight_io(sbi, REQ_TIME) || 573 (!f2fs_time_over(sbi, REQ_TIME) && rwsem_is_locked(&sbi->cp_rwsem))) 574 return; 575 576 /* exceed periodical checkpoint timeout threshold */ 577 if (f2fs_time_over(sbi, CP_TIME)) 578 goto do_sync; 579 580 /* checkpoint is the only way to shrink partial cached entries */ 581 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) && 582 f2fs_available_free_memory(sbi, INO_ENTRIES)) 583 return; 584 585 do_sync: 586 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 587 struct blk_plug plug; 588 589 mutex_lock(&sbi->flush_lock); 590 591 blk_start_plug(&plug); 592 f2fs_sync_dirty_inodes(sbi, FILE_INODE); 593 blk_finish_plug(&plug); 594 595 mutex_unlock(&sbi->flush_lock); 596 } 597 f2fs_sync_fs(sbi->sb, true); 598 stat_inc_bg_cp_count(sbi->stat_info); 599 } 600 601 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 602 struct block_device *bdev) 603 { 604 int ret = blkdev_issue_flush(bdev); 605 606 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 607 test_opt(sbi, FLUSH_MERGE), ret); 608 return ret; 609 } 610 611 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 612 { 613 int ret = 0; 614 int i; 615 616 if (!f2fs_is_multi_device(sbi)) 617 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 618 619 for (i = 0; i < sbi->s_ndevs; i++) { 620 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 621 continue; 622 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 623 if (ret) 624 break; 625 } 626 return ret; 627 } 628 629 static int issue_flush_thread(void *data) 630 { 631 struct f2fs_sb_info *sbi = data; 632 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 633 wait_queue_head_t *q = &fcc->flush_wait_queue; 634 repeat: 635 if (kthread_should_stop()) 636 return 0; 637 638 if (!llist_empty(&fcc->issue_list)) { 639 struct flush_cmd *cmd, *next; 640 int ret; 641 642 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 643 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 644 645 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 646 647 ret = submit_flush_wait(sbi, cmd->ino); 648 atomic_inc(&fcc->issued_flush); 649 650 llist_for_each_entry_safe(cmd, next, 651 fcc->dispatch_list, llnode) { 652 cmd->ret = ret; 653 complete(&cmd->wait); 654 } 655 fcc->dispatch_list = NULL; 656 } 657 658 wait_event_interruptible(*q, 659 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 660 goto repeat; 661 } 662 663 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 664 { 665 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 666 struct flush_cmd cmd; 667 int ret; 668 669 if (test_opt(sbi, NOBARRIER)) 670 return 0; 671 672 if (!test_opt(sbi, FLUSH_MERGE)) { 673 atomic_inc(&fcc->queued_flush); 674 ret = submit_flush_wait(sbi, ino); 675 atomic_dec(&fcc->queued_flush); 676 atomic_inc(&fcc->issued_flush); 677 return ret; 678 } 679 680 if (atomic_inc_return(&fcc->queued_flush) == 1 || 681 f2fs_is_multi_device(sbi)) { 682 ret = submit_flush_wait(sbi, ino); 683 atomic_dec(&fcc->queued_flush); 684 685 atomic_inc(&fcc->issued_flush); 686 return ret; 687 } 688 689 cmd.ino = ino; 690 init_completion(&cmd.wait); 691 692 llist_add(&cmd.llnode, &fcc->issue_list); 693 694 /* 695 * update issue_list before we wake up issue_flush thread, this 696 * smp_mb() pairs with another barrier in ___wait_event(), see 697 * more details in comments of waitqueue_active(). 698 */ 699 smp_mb(); 700 701 if (waitqueue_active(&fcc->flush_wait_queue)) 702 wake_up(&fcc->flush_wait_queue); 703 704 if (fcc->f2fs_issue_flush) { 705 wait_for_completion(&cmd.wait); 706 atomic_dec(&fcc->queued_flush); 707 } else { 708 struct llist_node *list; 709 710 list = llist_del_all(&fcc->issue_list); 711 if (!list) { 712 wait_for_completion(&cmd.wait); 713 atomic_dec(&fcc->queued_flush); 714 } else { 715 struct flush_cmd *tmp, *next; 716 717 ret = submit_flush_wait(sbi, ino); 718 719 llist_for_each_entry_safe(tmp, next, list, llnode) { 720 if (tmp == &cmd) { 721 cmd.ret = ret; 722 atomic_dec(&fcc->queued_flush); 723 continue; 724 } 725 tmp->ret = ret; 726 complete(&tmp->wait); 727 } 728 } 729 } 730 731 return cmd.ret; 732 } 733 734 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 735 { 736 dev_t dev = sbi->sb->s_bdev->bd_dev; 737 struct flush_cmd_control *fcc; 738 int err = 0; 739 740 if (SM_I(sbi)->fcc_info) { 741 fcc = SM_I(sbi)->fcc_info; 742 if (fcc->f2fs_issue_flush) 743 return err; 744 goto init_thread; 745 } 746 747 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 748 if (!fcc) 749 return -ENOMEM; 750 atomic_set(&fcc->issued_flush, 0); 751 atomic_set(&fcc->queued_flush, 0); 752 init_waitqueue_head(&fcc->flush_wait_queue); 753 init_llist_head(&fcc->issue_list); 754 SM_I(sbi)->fcc_info = fcc; 755 if (!test_opt(sbi, FLUSH_MERGE)) 756 return err; 757 758 init_thread: 759 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 760 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 761 if (IS_ERR(fcc->f2fs_issue_flush)) { 762 err = PTR_ERR(fcc->f2fs_issue_flush); 763 kfree(fcc); 764 SM_I(sbi)->fcc_info = NULL; 765 return err; 766 } 767 768 return err; 769 } 770 771 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 772 { 773 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 774 775 if (fcc && fcc->f2fs_issue_flush) { 776 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 777 778 fcc->f2fs_issue_flush = NULL; 779 kthread_stop(flush_thread); 780 } 781 if (free) { 782 kfree(fcc); 783 SM_I(sbi)->fcc_info = NULL; 784 } 785 } 786 787 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 788 { 789 int ret = 0, i; 790 791 if (!f2fs_is_multi_device(sbi)) 792 return 0; 793 794 if (test_opt(sbi, NOBARRIER)) 795 return 0; 796 797 for (i = 1; i < sbi->s_ndevs; i++) { 798 int count = DEFAULT_RETRY_IO_COUNT; 799 800 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 801 continue; 802 803 do { 804 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 805 if (ret) 806 congestion_wait(BLK_RW_ASYNC, 807 DEFAULT_IO_TIMEOUT); 808 } while (ret && --count); 809 810 if (ret) { 811 f2fs_stop_checkpoint(sbi, false); 812 break; 813 } 814 815 spin_lock(&sbi->dev_lock); 816 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 817 spin_unlock(&sbi->dev_lock); 818 } 819 820 return ret; 821 } 822 823 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 824 enum dirty_type dirty_type) 825 { 826 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 827 828 /* need not be added */ 829 if (IS_CURSEG(sbi, segno)) 830 return; 831 832 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 833 dirty_i->nr_dirty[dirty_type]++; 834 835 if (dirty_type == DIRTY) { 836 struct seg_entry *sentry = get_seg_entry(sbi, segno); 837 enum dirty_type t = sentry->type; 838 839 if (unlikely(t >= DIRTY)) { 840 f2fs_bug_on(sbi, 1); 841 return; 842 } 843 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 844 dirty_i->nr_dirty[t]++; 845 846 if (__is_large_section(sbi)) { 847 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 848 block_t valid_blocks = 849 get_valid_blocks(sbi, segno, true); 850 851 f2fs_bug_on(sbi, unlikely(!valid_blocks || 852 valid_blocks == BLKS_PER_SEC(sbi))); 853 854 if (!IS_CURSEC(sbi, secno)) 855 set_bit(secno, dirty_i->dirty_secmap); 856 } 857 } 858 } 859 860 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 861 enum dirty_type dirty_type) 862 { 863 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 864 block_t valid_blocks; 865 866 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 867 dirty_i->nr_dirty[dirty_type]--; 868 869 if (dirty_type == DIRTY) { 870 struct seg_entry *sentry = get_seg_entry(sbi, segno); 871 enum dirty_type t = sentry->type; 872 873 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 874 dirty_i->nr_dirty[t]--; 875 876 valid_blocks = get_valid_blocks(sbi, segno, true); 877 if (valid_blocks == 0) { 878 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 879 dirty_i->victim_secmap); 880 #ifdef CONFIG_F2FS_CHECK_FS 881 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 882 #endif 883 } 884 if (__is_large_section(sbi)) { 885 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 886 887 if (!valid_blocks || 888 valid_blocks == BLKS_PER_SEC(sbi)) { 889 clear_bit(secno, dirty_i->dirty_secmap); 890 return; 891 } 892 893 if (!IS_CURSEC(sbi, secno)) 894 set_bit(secno, dirty_i->dirty_secmap); 895 } 896 } 897 } 898 899 /* 900 * Should not occur error such as -ENOMEM. 901 * Adding dirty entry into seglist is not critical operation. 902 * If a given segment is one of current working segments, it won't be added. 903 */ 904 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 905 { 906 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 907 unsigned short valid_blocks, ckpt_valid_blocks; 908 unsigned int usable_blocks; 909 910 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 911 return; 912 913 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno); 914 mutex_lock(&dirty_i->seglist_lock); 915 916 valid_blocks = get_valid_blocks(sbi, segno, false); 917 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false); 918 919 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 920 ckpt_valid_blocks == usable_blocks)) { 921 __locate_dirty_segment(sbi, segno, PRE); 922 __remove_dirty_segment(sbi, segno, DIRTY); 923 } else if (valid_blocks < usable_blocks) { 924 __locate_dirty_segment(sbi, segno, DIRTY); 925 } else { 926 /* Recovery routine with SSR needs this */ 927 __remove_dirty_segment(sbi, segno, DIRTY); 928 } 929 930 mutex_unlock(&dirty_i->seglist_lock); 931 } 932 933 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 934 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 935 { 936 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 937 unsigned int segno; 938 939 mutex_lock(&dirty_i->seglist_lock); 940 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 941 if (get_valid_blocks(sbi, segno, false)) 942 continue; 943 if (IS_CURSEG(sbi, segno)) 944 continue; 945 __locate_dirty_segment(sbi, segno, PRE); 946 __remove_dirty_segment(sbi, segno, DIRTY); 947 } 948 mutex_unlock(&dirty_i->seglist_lock); 949 } 950 951 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 952 { 953 int ovp_hole_segs = 954 (overprovision_segments(sbi) - reserved_segments(sbi)); 955 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg; 956 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 957 block_t holes[2] = {0, 0}; /* DATA and NODE */ 958 block_t unusable; 959 struct seg_entry *se; 960 unsigned int segno; 961 962 mutex_lock(&dirty_i->seglist_lock); 963 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 964 se = get_seg_entry(sbi, segno); 965 if (IS_NODESEG(se->type)) 966 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) - 967 se->valid_blocks; 968 else 969 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) - 970 se->valid_blocks; 971 } 972 mutex_unlock(&dirty_i->seglist_lock); 973 974 unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE]; 975 if (unusable > ovp_holes) 976 return unusable - ovp_holes; 977 return 0; 978 } 979 980 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 981 { 982 int ovp_hole_segs = 983 (overprovision_segments(sbi) - reserved_segments(sbi)); 984 if (unusable > F2FS_OPTION(sbi).unusable_cap) 985 return -EAGAIN; 986 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 987 dirty_segments(sbi) > ovp_hole_segs) 988 return -EAGAIN; 989 return 0; 990 } 991 992 /* This is only used by SBI_CP_DISABLED */ 993 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 994 { 995 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 996 unsigned int segno = 0; 997 998 mutex_lock(&dirty_i->seglist_lock); 999 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 1000 if (get_valid_blocks(sbi, segno, false)) 1001 continue; 1002 if (get_ckpt_valid_blocks(sbi, segno, false)) 1003 continue; 1004 mutex_unlock(&dirty_i->seglist_lock); 1005 return segno; 1006 } 1007 mutex_unlock(&dirty_i->seglist_lock); 1008 return NULL_SEGNO; 1009 } 1010 1011 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 1012 struct block_device *bdev, block_t lstart, 1013 block_t start, block_t len) 1014 { 1015 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1016 struct list_head *pend_list; 1017 struct discard_cmd *dc; 1018 1019 f2fs_bug_on(sbi, !len); 1020 1021 pend_list = &dcc->pend_list[plist_idx(len)]; 1022 1023 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL); 1024 INIT_LIST_HEAD(&dc->list); 1025 dc->bdev = bdev; 1026 dc->lstart = lstart; 1027 dc->start = start; 1028 dc->len = len; 1029 dc->ref = 0; 1030 dc->state = D_PREP; 1031 dc->queued = 0; 1032 dc->error = 0; 1033 init_completion(&dc->wait); 1034 list_add_tail(&dc->list, pend_list); 1035 spin_lock_init(&dc->lock); 1036 dc->bio_ref = 0; 1037 atomic_inc(&dcc->discard_cmd_cnt); 1038 dcc->undiscard_blks += len; 1039 1040 return dc; 1041 } 1042 1043 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi, 1044 struct block_device *bdev, block_t lstart, 1045 block_t start, block_t len, 1046 struct rb_node *parent, struct rb_node **p, 1047 bool leftmost) 1048 { 1049 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1050 struct discard_cmd *dc; 1051 1052 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 1053 1054 rb_link_node(&dc->rb_node, parent, p); 1055 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 1056 1057 return dc; 1058 } 1059 1060 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 1061 struct discard_cmd *dc) 1062 { 1063 if (dc->state == D_DONE) 1064 atomic_sub(dc->queued, &dcc->queued_discard); 1065 1066 list_del(&dc->list); 1067 rb_erase_cached(&dc->rb_node, &dcc->root); 1068 dcc->undiscard_blks -= dc->len; 1069 1070 kmem_cache_free(discard_cmd_slab, dc); 1071 1072 atomic_dec(&dcc->discard_cmd_cnt); 1073 } 1074 1075 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 1076 struct discard_cmd *dc) 1077 { 1078 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1079 unsigned long flags; 1080 1081 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len); 1082 1083 spin_lock_irqsave(&dc->lock, flags); 1084 if (dc->bio_ref) { 1085 spin_unlock_irqrestore(&dc->lock, flags); 1086 return; 1087 } 1088 spin_unlock_irqrestore(&dc->lock, flags); 1089 1090 f2fs_bug_on(sbi, dc->ref); 1091 1092 if (dc->error == -EOPNOTSUPP) 1093 dc->error = 0; 1094 1095 if (dc->error) 1096 printk_ratelimited( 1097 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d", 1098 KERN_INFO, sbi->sb->s_id, 1099 dc->lstart, dc->start, dc->len, dc->error); 1100 __detach_discard_cmd(dcc, dc); 1101 } 1102 1103 static void f2fs_submit_discard_endio(struct bio *bio) 1104 { 1105 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1106 unsigned long flags; 1107 1108 spin_lock_irqsave(&dc->lock, flags); 1109 if (!dc->error) 1110 dc->error = blk_status_to_errno(bio->bi_status); 1111 dc->bio_ref--; 1112 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1113 dc->state = D_DONE; 1114 complete_all(&dc->wait); 1115 } 1116 spin_unlock_irqrestore(&dc->lock, flags); 1117 bio_put(bio); 1118 } 1119 1120 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1121 block_t start, block_t end) 1122 { 1123 #ifdef CONFIG_F2FS_CHECK_FS 1124 struct seg_entry *sentry; 1125 unsigned int segno; 1126 block_t blk = start; 1127 unsigned long offset, size, max_blocks = sbi->blocks_per_seg; 1128 unsigned long *map; 1129 1130 while (blk < end) { 1131 segno = GET_SEGNO(sbi, blk); 1132 sentry = get_seg_entry(sbi, segno); 1133 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1134 1135 if (end < START_BLOCK(sbi, segno + 1)) 1136 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1137 else 1138 size = max_blocks; 1139 map = (unsigned long *)(sentry->cur_valid_map); 1140 offset = __find_rev_next_bit(map, size, offset); 1141 f2fs_bug_on(sbi, offset != size); 1142 blk = START_BLOCK(sbi, segno + 1); 1143 } 1144 #endif 1145 } 1146 1147 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1148 struct discard_policy *dpolicy, 1149 int discard_type, unsigned int granularity) 1150 { 1151 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1152 1153 /* common policy */ 1154 dpolicy->type = discard_type; 1155 dpolicy->sync = true; 1156 dpolicy->ordered = false; 1157 dpolicy->granularity = granularity; 1158 1159 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST; 1160 dpolicy->io_aware_gran = MAX_PLIST_NUM; 1161 dpolicy->timeout = false; 1162 1163 if (discard_type == DPOLICY_BG) { 1164 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME; 1165 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME; 1166 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME; 1167 dpolicy->io_aware = true; 1168 dpolicy->sync = false; 1169 dpolicy->ordered = true; 1170 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) { 1171 dpolicy->granularity = 1; 1172 if (atomic_read(&dcc->discard_cmd_cnt)) 1173 dpolicy->max_interval = 1174 DEF_MIN_DISCARD_ISSUE_TIME; 1175 } 1176 } else if (discard_type == DPOLICY_FORCE) { 1177 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME; 1178 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME; 1179 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME; 1180 dpolicy->io_aware = false; 1181 } else if (discard_type == DPOLICY_FSTRIM) { 1182 dpolicy->io_aware = false; 1183 } else if (discard_type == DPOLICY_UMOUNT) { 1184 dpolicy->io_aware = false; 1185 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1186 dpolicy->granularity = 1; 1187 dpolicy->timeout = true; 1188 } 1189 } 1190 1191 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1192 struct block_device *bdev, block_t lstart, 1193 block_t start, block_t len); 1194 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1195 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1196 struct discard_policy *dpolicy, 1197 struct discard_cmd *dc, 1198 unsigned int *issued) 1199 { 1200 struct block_device *bdev = dc->bdev; 1201 struct request_queue *q = bdev_get_queue(bdev); 1202 unsigned int max_discard_blocks = 1203 SECTOR_TO_BLOCK(q->limits.max_discard_sectors); 1204 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1205 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1206 &(dcc->fstrim_list) : &(dcc->wait_list); 1207 int flag = dpolicy->sync ? REQ_SYNC : 0; 1208 block_t lstart, start, len, total_len; 1209 int err = 0; 1210 1211 if (dc->state != D_PREP) 1212 return 0; 1213 1214 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1215 return 0; 1216 1217 trace_f2fs_issue_discard(bdev, dc->start, dc->len); 1218 1219 lstart = dc->lstart; 1220 start = dc->start; 1221 len = dc->len; 1222 total_len = len; 1223 1224 dc->len = 0; 1225 1226 while (total_len && *issued < dpolicy->max_requests && !err) { 1227 struct bio *bio = NULL; 1228 unsigned long flags; 1229 bool last = true; 1230 1231 if (len > max_discard_blocks) { 1232 len = max_discard_blocks; 1233 last = false; 1234 } 1235 1236 (*issued)++; 1237 if (*issued == dpolicy->max_requests) 1238 last = true; 1239 1240 dc->len += len; 1241 1242 if (time_to_inject(sbi, FAULT_DISCARD)) { 1243 f2fs_show_injection_info(sbi, FAULT_DISCARD); 1244 err = -EIO; 1245 goto submit; 1246 } 1247 err = __blkdev_issue_discard(bdev, 1248 SECTOR_FROM_BLOCK(start), 1249 SECTOR_FROM_BLOCK(len), 1250 GFP_NOFS, 0, &bio); 1251 submit: 1252 if (err) { 1253 spin_lock_irqsave(&dc->lock, flags); 1254 if (dc->state == D_PARTIAL) 1255 dc->state = D_SUBMIT; 1256 spin_unlock_irqrestore(&dc->lock, flags); 1257 1258 break; 1259 } 1260 1261 f2fs_bug_on(sbi, !bio); 1262 1263 /* 1264 * should keep before submission to avoid D_DONE 1265 * right away 1266 */ 1267 spin_lock_irqsave(&dc->lock, flags); 1268 if (last) 1269 dc->state = D_SUBMIT; 1270 else 1271 dc->state = D_PARTIAL; 1272 dc->bio_ref++; 1273 spin_unlock_irqrestore(&dc->lock, flags); 1274 1275 atomic_inc(&dcc->queued_discard); 1276 dc->queued++; 1277 list_move_tail(&dc->list, wait_list); 1278 1279 /* sanity check on discard range */ 1280 __check_sit_bitmap(sbi, lstart, lstart + len); 1281 1282 bio->bi_private = dc; 1283 bio->bi_end_io = f2fs_submit_discard_endio; 1284 bio->bi_opf |= flag; 1285 submit_bio(bio); 1286 1287 atomic_inc(&dcc->issued_discard); 1288 1289 f2fs_update_iostat(sbi, FS_DISCARD, 1); 1290 1291 lstart += len; 1292 start += len; 1293 total_len -= len; 1294 len = total_len; 1295 } 1296 1297 if (!err && len) { 1298 dcc->undiscard_blks -= len; 1299 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1300 } 1301 return err; 1302 } 1303 1304 static void __insert_discard_tree(struct f2fs_sb_info *sbi, 1305 struct block_device *bdev, block_t lstart, 1306 block_t start, block_t len, 1307 struct rb_node **insert_p, 1308 struct rb_node *insert_parent) 1309 { 1310 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1311 struct rb_node **p; 1312 struct rb_node *parent = NULL; 1313 bool leftmost = true; 1314 1315 if (insert_p && insert_parent) { 1316 parent = insert_parent; 1317 p = insert_p; 1318 goto do_insert; 1319 } 1320 1321 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, 1322 lstart, &leftmost); 1323 do_insert: 1324 __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, 1325 p, leftmost); 1326 } 1327 1328 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1329 struct discard_cmd *dc) 1330 { 1331 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]); 1332 } 1333 1334 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1335 struct discard_cmd *dc, block_t blkaddr) 1336 { 1337 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1338 struct discard_info di = dc->di; 1339 bool modified = false; 1340 1341 if (dc->state == D_DONE || dc->len == 1) { 1342 __remove_discard_cmd(sbi, dc); 1343 return; 1344 } 1345 1346 dcc->undiscard_blks -= di.len; 1347 1348 if (blkaddr > di.lstart) { 1349 dc->len = blkaddr - dc->lstart; 1350 dcc->undiscard_blks += dc->len; 1351 __relocate_discard_cmd(dcc, dc); 1352 modified = true; 1353 } 1354 1355 if (blkaddr < di.lstart + di.len - 1) { 1356 if (modified) { 1357 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, 1358 di.start + blkaddr + 1 - di.lstart, 1359 di.lstart + di.len - 1 - blkaddr, 1360 NULL, NULL); 1361 } else { 1362 dc->lstart++; 1363 dc->len--; 1364 dc->start++; 1365 dcc->undiscard_blks += dc->len; 1366 __relocate_discard_cmd(dcc, dc); 1367 } 1368 } 1369 } 1370 1371 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1372 struct block_device *bdev, block_t lstart, 1373 block_t start, block_t len) 1374 { 1375 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1376 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1377 struct discard_cmd *dc; 1378 struct discard_info di = {0}; 1379 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1380 struct request_queue *q = bdev_get_queue(bdev); 1381 unsigned int max_discard_blocks = 1382 SECTOR_TO_BLOCK(q->limits.max_discard_sectors); 1383 block_t end = lstart + len; 1384 1385 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1386 NULL, lstart, 1387 (struct rb_entry **)&prev_dc, 1388 (struct rb_entry **)&next_dc, 1389 &insert_p, &insert_parent, true, NULL); 1390 if (dc) 1391 prev_dc = dc; 1392 1393 if (!prev_dc) { 1394 di.lstart = lstart; 1395 di.len = next_dc ? next_dc->lstart - lstart : len; 1396 di.len = min(di.len, len); 1397 di.start = start; 1398 } 1399 1400 while (1) { 1401 struct rb_node *node; 1402 bool merged = false; 1403 struct discard_cmd *tdc = NULL; 1404 1405 if (prev_dc) { 1406 di.lstart = prev_dc->lstart + prev_dc->len; 1407 if (di.lstart < lstart) 1408 di.lstart = lstart; 1409 if (di.lstart >= end) 1410 break; 1411 1412 if (!next_dc || next_dc->lstart > end) 1413 di.len = end - di.lstart; 1414 else 1415 di.len = next_dc->lstart - di.lstart; 1416 di.start = start + di.lstart - lstart; 1417 } 1418 1419 if (!di.len) 1420 goto next; 1421 1422 if (prev_dc && prev_dc->state == D_PREP && 1423 prev_dc->bdev == bdev && 1424 __is_discard_back_mergeable(&di, &prev_dc->di, 1425 max_discard_blocks)) { 1426 prev_dc->di.len += di.len; 1427 dcc->undiscard_blks += di.len; 1428 __relocate_discard_cmd(dcc, prev_dc); 1429 di = prev_dc->di; 1430 tdc = prev_dc; 1431 merged = true; 1432 } 1433 1434 if (next_dc && next_dc->state == D_PREP && 1435 next_dc->bdev == bdev && 1436 __is_discard_front_mergeable(&di, &next_dc->di, 1437 max_discard_blocks)) { 1438 next_dc->di.lstart = di.lstart; 1439 next_dc->di.len += di.len; 1440 next_dc->di.start = di.start; 1441 dcc->undiscard_blks += di.len; 1442 __relocate_discard_cmd(dcc, next_dc); 1443 if (tdc) 1444 __remove_discard_cmd(sbi, tdc); 1445 merged = true; 1446 } 1447 1448 if (!merged) { 1449 __insert_discard_tree(sbi, bdev, di.lstart, di.start, 1450 di.len, NULL, NULL); 1451 } 1452 next: 1453 prev_dc = next_dc; 1454 if (!prev_dc) 1455 break; 1456 1457 node = rb_next(&prev_dc->rb_node); 1458 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1459 } 1460 } 1461 1462 static int __queue_discard_cmd(struct f2fs_sb_info *sbi, 1463 struct block_device *bdev, block_t blkstart, block_t blklen) 1464 { 1465 block_t lblkstart = blkstart; 1466 1467 if (!f2fs_bdev_support_discard(bdev)) 1468 return 0; 1469 1470 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1471 1472 if (f2fs_is_multi_device(sbi)) { 1473 int devi = f2fs_target_device_index(sbi, blkstart); 1474 1475 blkstart -= FDEV(devi).start_blk; 1476 } 1477 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1478 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1479 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1480 return 0; 1481 } 1482 1483 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1484 struct discard_policy *dpolicy) 1485 { 1486 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1487 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1488 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1489 struct discard_cmd *dc; 1490 struct blk_plug plug; 1491 unsigned int pos = dcc->next_pos; 1492 unsigned int issued = 0; 1493 bool io_interrupted = false; 1494 1495 mutex_lock(&dcc->cmd_lock); 1496 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1497 NULL, pos, 1498 (struct rb_entry **)&prev_dc, 1499 (struct rb_entry **)&next_dc, 1500 &insert_p, &insert_parent, true, NULL); 1501 if (!dc) 1502 dc = next_dc; 1503 1504 blk_start_plug(&plug); 1505 1506 while (dc) { 1507 struct rb_node *node; 1508 int err = 0; 1509 1510 if (dc->state != D_PREP) 1511 goto next; 1512 1513 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1514 io_interrupted = true; 1515 break; 1516 } 1517 1518 dcc->next_pos = dc->lstart + dc->len; 1519 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1520 1521 if (issued >= dpolicy->max_requests) 1522 break; 1523 next: 1524 node = rb_next(&dc->rb_node); 1525 if (err) 1526 __remove_discard_cmd(sbi, dc); 1527 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1528 } 1529 1530 blk_finish_plug(&plug); 1531 1532 if (!dc) 1533 dcc->next_pos = 0; 1534 1535 mutex_unlock(&dcc->cmd_lock); 1536 1537 if (!issued && io_interrupted) 1538 issued = -1; 1539 1540 return issued; 1541 } 1542 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1543 struct discard_policy *dpolicy); 1544 1545 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1546 struct discard_policy *dpolicy) 1547 { 1548 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1549 struct list_head *pend_list; 1550 struct discard_cmd *dc, *tmp; 1551 struct blk_plug plug; 1552 int i, issued; 1553 bool io_interrupted = false; 1554 1555 if (dpolicy->timeout) 1556 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1557 1558 retry: 1559 issued = 0; 1560 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1561 if (dpolicy->timeout && 1562 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1563 break; 1564 1565 if (i + 1 < dpolicy->granularity) 1566 break; 1567 1568 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered) 1569 return __issue_discard_cmd_orderly(sbi, dpolicy); 1570 1571 pend_list = &dcc->pend_list[i]; 1572 1573 mutex_lock(&dcc->cmd_lock); 1574 if (list_empty(pend_list)) 1575 goto next; 1576 if (unlikely(dcc->rbtree_check)) 1577 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 1578 &dcc->root, false)); 1579 blk_start_plug(&plug); 1580 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1581 f2fs_bug_on(sbi, dc->state != D_PREP); 1582 1583 if (dpolicy->timeout && 1584 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1585 break; 1586 1587 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1588 !is_idle(sbi, DISCARD_TIME)) { 1589 io_interrupted = true; 1590 break; 1591 } 1592 1593 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1594 1595 if (issued >= dpolicy->max_requests) 1596 break; 1597 } 1598 blk_finish_plug(&plug); 1599 next: 1600 mutex_unlock(&dcc->cmd_lock); 1601 1602 if (issued >= dpolicy->max_requests || io_interrupted) 1603 break; 1604 } 1605 1606 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1607 __wait_all_discard_cmd(sbi, dpolicy); 1608 goto retry; 1609 } 1610 1611 if (!issued && io_interrupted) 1612 issued = -1; 1613 1614 return issued; 1615 } 1616 1617 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1618 { 1619 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1620 struct list_head *pend_list; 1621 struct discard_cmd *dc, *tmp; 1622 int i; 1623 bool dropped = false; 1624 1625 mutex_lock(&dcc->cmd_lock); 1626 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1627 pend_list = &dcc->pend_list[i]; 1628 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1629 f2fs_bug_on(sbi, dc->state != D_PREP); 1630 __remove_discard_cmd(sbi, dc); 1631 dropped = true; 1632 } 1633 } 1634 mutex_unlock(&dcc->cmd_lock); 1635 1636 return dropped; 1637 } 1638 1639 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1640 { 1641 __drop_discard_cmd(sbi); 1642 } 1643 1644 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1645 struct discard_cmd *dc) 1646 { 1647 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1648 unsigned int len = 0; 1649 1650 wait_for_completion_io(&dc->wait); 1651 mutex_lock(&dcc->cmd_lock); 1652 f2fs_bug_on(sbi, dc->state != D_DONE); 1653 dc->ref--; 1654 if (!dc->ref) { 1655 if (!dc->error) 1656 len = dc->len; 1657 __remove_discard_cmd(sbi, dc); 1658 } 1659 mutex_unlock(&dcc->cmd_lock); 1660 1661 return len; 1662 } 1663 1664 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1665 struct discard_policy *dpolicy, 1666 block_t start, block_t end) 1667 { 1668 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1669 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1670 &(dcc->fstrim_list) : &(dcc->wait_list); 1671 struct discard_cmd *dc, *tmp; 1672 bool need_wait; 1673 unsigned int trimmed = 0; 1674 1675 next: 1676 need_wait = false; 1677 1678 mutex_lock(&dcc->cmd_lock); 1679 list_for_each_entry_safe(dc, tmp, wait_list, list) { 1680 if (dc->lstart + dc->len <= start || end <= dc->lstart) 1681 continue; 1682 if (dc->len < dpolicy->granularity) 1683 continue; 1684 if (dc->state == D_DONE && !dc->ref) { 1685 wait_for_completion_io(&dc->wait); 1686 if (!dc->error) 1687 trimmed += dc->len; 1688 __remove_discard_cmd(sbi, dc); 1689 } else { 1690 dc->ref++; 1691 need_wait = true; 1692 break; 1693 } 1694 } 1695 mutex_unlock(&dcc->cmd_lock); 1696 1697 if (need_wait) { 1698 trimmed += __wait_one_discard_bio(sbi, dc); 1699 goto next; 1700 } 1701 1702 return trimmed; 1703 } 1704 1705 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1706 struct discard_policy *dpolicy) 1707 { 1708 struct discard_policy dp; 1709 unsigned int discard_blks; 1710 1711 if (dpolicy) 1712 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1713 1714 /* wait all */ 1715 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1); 1716 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1717 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1); 1718 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1719 1720 return discard_blks; 1721 } 1722 1723 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1724 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1725 { 1726 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1727 struct discard_cmd *dc; 1728 bool need_wait = false; 1729 1730 mutex_lock(&dcc->cmd_lock); 1731 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root, 1732 NULL, blkaddr); 1733 if (dc) { 1734 if (dc->state == D_PREP) { 1735 __punch_discard_cmd(sbi, dc, blkaddr); 1736 } else { 1737 dc->ref++; 1738 need_wait = true; 1739 } 1740 } 1741 mutex_unlock(&dcc->cmd_lock); 1742 1743 if (need_wait) 1744 __wait_one_discard_bio(sbi, dc); 1745 } 1746 1747 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1748 { 1749 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1750 1751 if (dcc && dcc->f2fs_issue_discard) { 1752 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1753 1754 dcc->f2fs_issue_discard = NULL; 1755 kthread_stop(discard_thread); 1756 } 1757 } 1758 1759 /* This comes from f2fs_put_super */ 1760 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1761 { 1762 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1763 struct discard_policy dpolicy; 1764 bool dropped; 1765 1766 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1767 dcc->discard_granularity); 1768 __issue_discard_cmd(sbi, &dpolicy); 1769 dropped = __drop_discard_cmd(sbi); 1770 1771 /* just to make sure there is no pending discard commands */ 1772 __wait_all_discard_cmd(sbi, NULL); 1773 1774 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1775 return dropped; 1776 } 1777 1778 static int issue_discard_thread(void *data) 1779 { 1780 struct f2fs_sb_info *sbi = data; 1781 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1782 wait_queue_head_t *q = &dcc->discard_wait_queue; 1783 struct discard_policy dpolicy; 1784 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME; 1785 int issued; 1786 1787 set_freezable(); 1788 1789 do { 1790 if (sbi->gc_mode == GC_URGENT_HIGH || 1791 !f2fs_available_free_memory(sbi, DISCARD_CACHE)) 1792 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1); 1793 else 1794 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1795 dcc->discard_granularity); 1796 1797 if (!atomic_read(&dcc->discard_cmd_cnt)) 1798 wait_ms = dpolicy.max_interval; 1799 1800 wait_event_interruptible_timeout(*q, 1801 kthread_should_stop() || freezing(current) || 1802 dcc->discard_wake, 1803 msecs_to_jiffies(wait_ms)); 1804 1805 if (dcc->discard_wake) 1806 dcc->discard_wake = 0; 1807 1808 /* clean up pending candidates before going to sleep */ 1809 if (atomic_read(&dcc->queued_discard)) 1810 __wait_all_discard_cmd(sbi, NULL); 1811 1812 if (try_to_freeze()) 1813 continue; 1814 if (f2fs_readonly(sbi->sb)) 1815 continue; 1816 if (kthread_should_stop()) 1817 return 0; 1818 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1819 wait_ms = dpolicy.max_interval; 1820 continue; 1821 } 1822 if (!atomic_read(&dcc->discard_cmd_cnt)) 1823 continue; 1824 1825 sb_start_intwrite(sbi->sb); 1826 1827 issued = __issue_discard_cmd(sbi, &dpolicy); 1828 if (issued > 0) { 1829 __wait_all_discard_cmd(sbi, &dpolicy); 1830 wait_ms = dpolicy.min_interval; 1831 } else if (issued == -1) { 1832 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1833 if (!wait_ms) 1834 wait_ms = dpolicy.mid_interval; 1835 } else { 1836 wait_ms = dpolicy.max_interval; 1837 } 1838 1839 sb_end_intwrite(sbi->sb); 1840 1841 } while (!kthread_should_stop()); 1842 return 0; 1843 } 1844 1845 #ifdef CONFIG_BLK_DEV_ZONED 1846 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1847 struct block_device *bdev, block_t blkstart, block_t blklen) 1848 { 1849 sector_t sector, nr_sects; 1850 block_t lblkstart = blkstart; 1851 int devi = 0; 1852 1853 if (f2fs_is_multi_device(sbi)) { 1854 devi = f2fs_target_device_index(sbi, blkstart); 1855 if (blkstart < FDEV(devi).start_blk || 1856 blkstart > FDEV(devi).end_blk) { 1857 f2fs_err(sbi, "Invalid block %x", blkstart); 1858 return -EIO; 1859 } 1860 blkstart -= FDEV(devi).start_blk; 1861 } 1862 1863 /* For sequential zones, reset the zone write pointer */ 1864 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1865 sector = SECTOR_FROM_BLOCK(blkstart); 1866 nr_sects = SECTOR_FROM_BLOCK(blklen); 1867 1868 if (sector & (bdev_zone_sectors(bdev) - 1) || 1869 nr_sects != bdev_zone_sectors(bdev)) { 1870 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1871 devi, sbi->s_ndevs ? FDEV(devi).path : "", 1872 blkstart, blklen); 1873 return -EIO; 1874 } 1875 trace_f2fs_issue_reset_zone(bdev, blkstart); 1876 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 1877 sector, nr_sects, GFP_NOFS); 1878 } 1879 1880 /* For conventional zones, use regular discard if supported */ 1881 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1882 } 1883 #endif 1884 1885 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1886 struct block_device *bdev, block_t blkstart, block_t blklen) 1887 { 1888 #ifdef CONFIG_BLK_DEV_ZONED 1889 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 1890 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1891 #endif 1892 return __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1893 } 1894 1895 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1896 block_t blkstart, block_t blklen) 1897 { 1898 sector_t start = blkstart, len = 0; 1899 struct block_device *bdev; 1900 struct seg_entry *se; 1901 unsigned int offset; 1902 block_t i; 1903 int err = 0; 1904 1905 bdev = f2fs_target_device(sbi, blkstart, NULL); 1906 1907 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1908 if (i != start) { 1909 struct block_device *bdev2 = 1910 f2fs_target_device(sbi, i, NULL); 1911 1912 if (bdev2 != bdev) { 1913 err = __issue_discard_async(sbi, bdev, 1914 start, len); 1915 if (err) 1916 return err; 1917 bdev = bdev2; 1918 start = i; 1919 len = 0; 1920 } 1921 } 1922 1923 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 1924 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 1925 1926 if (f2fs_block_unit_discard(sbi) && 1927 !f2fs_test_and_set_bit(offset, se->discard_map)) 1928 sbi->discard_blks--; 1929 } 1930 1931 if (len) 1932 err = __issue_discard_async(sbi, bdev, start, len); 1933 return err; 1934 } 1935 1936 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 1937 bool check_only) 1938 { 1939 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1940 int max_blocks = sbi->blocks_per_seg; 1941 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 1942 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1943 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1944 unsigned long *discard_map = (unsigned long *)se->discard_map; 1945 unsigned long *dmap = SIT_I(sbi)->tmp_map; 1946 unsigned int start = 0, end = -1; 1947 bool force = (cpc->reason & CP_DISCARD); 1948 struct discard_entry *de = NULL; 1949 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 1950 int i; 1951 1952 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) || 1953 !f2fs_block_unit_discard(sbi)) 1954 return false; 1955 1956 if (!force) { 1957 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks || 1958 SM_I(sbi)->dcc_info->nr_discards >= 1959 SM_I(sbi)->dcc_info->max_discards) 1960 return false; 1961 } 1962 1963 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 1964 for (i = 0; i < entries; i++) 1965 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 1966 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 1967 1968 while (force || SM_I(sbi)->dcc_info->nr_discards <= 1969 SM_I(sbi)->dcc_info->max_discards) { 1970 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 1971 if (start >= max_blocks) 1972 break; 1973 1974 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 1975 if (force && start && end != max_blocks 1976 && (end - start) < cpc->trim_minlen) 1977 continue; 1978 1979 if (check_only) 1980 return true; 1981 1982 if (!de) { 1983 de = f2fs_kmem_cache_alloc(discard_entry_slab, 1984 GFP_F2FS_ZERO, true, NULL); 1985 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 1986 list_add_tail(&de->list, head); 1987 } 1988 1989 for (i = start; i < end; i++) 1990 __set_bit_le(i, (void *)de->discard_map); 1991 1992 SM_I(sbi)->dcc_info->nr_discards += end - start; 1993 } 1994 return false; 1995 } 1996 1997 static void release_discard_addr(struct discard_entry *entry) 1998 { 1999 list_del(&entry->list); 2000 kmem_cache_free(discard_entry_slab, entry); 2001 } 2002 2003 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 2004 { 2005 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 2006 struct discard_entry *entry, *this; 2007 2008 /* drop caches */ 2009 list_for_each_entry_safe(entry, this, head, list) 2010 release_discard_addr(entry); 2011 } 2012 2013 /* 2014 * Should call f2fs_clear_prefree_segments after checkpoint is done. 2015 */ 2016 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 2017 { 2018 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2019 unsigned int segno; 2020 2021 mutex_lock(&dirty_i->seglist_lock); 2022 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 2023 __set_test_and_free(sbi, segno, false); 2024 mutex_unlock(&dirty_i->seglist_lock); 2025 } 2026 2027 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 2028 struct cp_control *cpc) 2029 { 2030 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2031 struct list_head *head = &dcc->entry_list; 2032 struct discard_entry *entry, *this; 2033 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2034 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 2035 unsigned int start = 0, end = -1; 2036 unsigned int secno, start_segno; 2037 bool force = (cpc->reason & CP_DISCARD); 2038 bool section_alignment = F2FS_OPTION(sbi).discard_unit == 2039 DISCARD_UNIT_SECTION; 2040 2041 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi)) 2042 section_alignment = true; 2043 2044 mutex_lock(&dirty_i->seglist_lock); 2045 2046 while (1) { 2047 int i; 2048 2049 if (section_alignment && end != -1) 2050 end--; 2051 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 2052 if (start >= MAIN_SEGS(sbi)) 2053 break; 2054 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 2055 start + 1); 2056 2057 if (section_alignment) { 2058 start = rounddown(start, sbi->segs_per_sec); 2059 end = roundup(end, sbi->segs_per_sec); 2060 } 2061 2062 for (i = start; i < end; i++) { 2063 if (test_and_clear_bit(i, prefree_map)) 2064 dirty_i->nr_dirty[PRE]--; 2065 } 2066 2067 if (!f2fs_realtime_discard_enable(sbi)) 2068 continue; 2069 2070 if (force && start >= cpc->trim_start && 2071 (end - 1) <= cpc->trim_end) 2072 continue; 2073 2074 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) { 2075 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 2076 (end - start) << sbi->log_blocks_per_seg); 2077 continue; 2078 } 2079 next: 2080 secno = GET_SEC_FROM_SEG(sbi, start); 2081 start_segno = GET_SEG_FROM_SEC(sbi, secno); 2082 if (!IS_CURSEC(sbi, secno) && 2083 !get_valid_blocks(sbi, start, true)) 2084 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 2085 sbi->segs_per_sec << sbi->log_blocks_per_seg); 2086 2087 start = start_segno + sbi->segs_per_sec; 2088 if (start < end) 2089 goto next; 2090 else 2091 end = start - 1; 2092 } 2093 mutex_unlock(&dirty_i->seglist_lock); 2094 2095 if (!f2fs_block_unit_discard(sbi)) 2096 goto wakeup; 2097 2098 /* send small discards */ 2099 list_for_each_entry_safe(entry, this, head, list) { 2100 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2101 bool is_valid = test_bit_le(0, entry->discard_map); 2102 2103 find_next: 2104 if (is_valid) { 2105 next_pos = find_next_zero_bit_le(entry->discard_map, 2106 sbi->blocks_per_seg, cur_pos); 2107 len = next_pos - cur_pos; 2108 2109 if (f2fs_sb_has_blkzoned(sbi) || 2110 (force && len < cpc->trim_minlen)) 2111 goto skip; 2112 2113 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2114 len); 2115 total_len += len; 2116 } else { 2117 next_pos = find_next_bit_le(entry->discard_map, 2118 sbi->blocks_per_seg, cur_pos); 2119 } 2120 skip: 2121 cur_pos = next_pos; 2122 is_valid = !is_valid; 2123 2124 if (cur_pos < sbi->blocks_per_seg) 2125 goto find_next; 2126 2127 release_discard_addr(entry); 2128 dcc->nr_discards -= total_len; 2129 } 2130 2131 wakeup: 2132 wake_up_discard_thread(sbi, false); 2133 } 2134 2135 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi) 2136 { 2137 dev_t dev = sbi->sb->s_bdev->bd_dev; 2138 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2139 int err = 0; 2140 2141 if (!f2fs_realtime_discard_enable(sbi)) 2142 return 0; 2143 2144 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2145 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2146 if (IS_ERR(dcc->f2fs_issue_discard)) 2147 err = PTR_ERR(dcc->f2fs_issue_discard); 2148 2149 return err; 2150 } 2151 2152 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2153 { 2154 struct discard_cmd_control *dcc; 2155 int err = 0, i; 2156 2157 if (SM_I(sbi)->dcc_info) { 2158 dcc = SM_I(sbi)->dcc_info; 2159 goto init_thread; 2160 } 2161 2162 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2163 if (!dcc) 2164 return -ENOMEM; 2165 2166 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2167 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT) 2168 dcc->discard_granularity = sbi->blocks_per_seg; 2169 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION) 2170 dcc->discard_granularity = BLKS_PER_SEC(sbi); 2171 2172 INIT_LIST_HEAD(&dcc->entry_list); 2173 for (i = 0; i < MAX_PLIST_NUM; i++) 2174 INIT_LIST_HEAD(&dcc->pend_list[i]); 2175 INIT_LIST_HEAD(&dcc->wait_list); 2176 INIT_LIST_HEAD(&dcc->fstrim_list); 2177 mutex_init(&dcc->cmd_lock); 2178 atomic_set(&dcc->issued_discard, 0); 2179 atomic_set(&dcc->queued_discard, 0); 2180 atomic_set(&dcc->discard_cmd_cnt, 0); 2181 dcc->nr_discards = 0; 2182 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 2183 dcc->undiscard_blks = 0; 2184 dcc->next_pos = 0; 2185 dcc->root = RB_ROOT_CACHED; 2186 dcc->rbtree_check = false; 2187 2188 init_waitqueue_head(&dcc->discard_wait_queue); 2189 SM_I(sbi)->dcc_info = dcc; 2190 init_thread: 2191 err = f2fs_start_discard_thread(sbi); 2192 if (err) { 2193 kfree(dcc); 2194 SM_I(sbi)->dcc_info = NULL; 2195 } 2196 2197 return err; 2198 } 2199 2200 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2201 { 2202 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2203 2204 if (!dcc) 2205 return; 2206 2207 f2fs_stop_discard_thread(sbi); 2208 2209 /* 2210 * Recovery can cache discard commands, so in error path of 2211 * fill_super(), it needs to give a chance to handle them. 2212 */ 2213 if (unlikely(atomic_read(&dcc->discard_cmd_cnt))) 2214 f2fs_issue_discard_timeout(sbi); 2215 2216 kfree(dcc); 2217 SM_I(sbi)->dcc_info = NULL; 2218 } 2219 2220 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2221 { 2222 struct sit_info *sit_i = SIT_I(sbi); 2223 2224 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2225 sit_i->dirty_sentries++; 2226 return false; 2227 } 2228 2229 return true; 2230 } 2231 2232 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2233 unsigned int segno, int modified) 2234 { 2235 struct seg_entry *se = get_seg_entry(sbi, segno); 2236 2237 se->type = type; 2238 if (modified) 2239 __mark_sit_entry_dirty(sbi, segno); 2240 } 2241 2242 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi, 2243 block_t blkaddr) 2244 { 2245 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2246 2247 if (segno == NULL_SEGNO) 2248 return 0; 2249 return get_seg_entry(sbi, segno)->mtime; 2250 } 2251 2252 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr, 2253 unsigned long long old_mtime) 2254 { 2255 struct seg_entry *se; 2256 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2257 unsigned long long ctime = get_mtime(sbi, false); 2258 unsigned long long mtime = old_mtime ? old_mtime : ctime; 2259 2260 if (segno == NULL_SEGNO) 2261 return; 2262 2263 se = get_seg_entry(sbi, segno); 2264 2265 if (!se->mtime) 2266 se->mtime = mtime; 2267 else 2268 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime, 2269 se->valid_blocks + 1); 2270 2271 if (ctime > SIT_I(sbi)->max_mtime) 2272 SIT_I(sbi)->max_mtime = ctime; 2273 } 2274 2275 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2276 { 2277 struct seg_entry *se; 2278 unsigned int segno, offset; 2279 long int new_vblocks; 2280 bool exist; 2281 #ifdef CONFIG_F2FS_CHECK_FS 2282 bool mir_exist; 2283 #endif 2284 2285 segno = GET_SEGNO(sbi, blkaddr); 2286 2287 se = get_seg_entry(sbi, segno); 2288 new_vblocks = se->valid_blocks + del; 2289 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2290 2291 f2fs_bug_on(sbi, (new_vblocks < 0 || 2292 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno)))); 2293 2294 se->valid_blocks = new_vblocks; 2295 2296 /* Update valid block bitmap */ 2297 if (del > 0) { 2298 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2299 #ifdef CONFIG_F2FS_CHECK_FS 2300 mir_exist = f2fs_test_and_set_bit(offset, 2301 se->cur_valid_map_mir); 2302 if (unlikely(exist != mir_exist)) { 2303 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2304 blkaddr, exist); 2305 f2fs_bug_on(sbi, 1); 2306 } 2307 #endif 2308 if (unlikely(exist)) { 2309 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", 2310 blkaddr); 2311 f2fs_bug_on(sbi, 1); 2312 se->valid_blocks--; 2313 del = 0; 2314 } 2315 2316 if (f2fs_block_unit_discard(sbi) && 2317 !f2fs_test_and_set_bit(offset, se->discard_map)) 2318 sbi->discard_blks--; 2319 2320 /* 2321 * SSR should never reuse block which is checkpointed 2322 * or newly invalidated. 2323 */ 2324 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2325 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 2326 se->ckpt_valid_blocks++; 2327 } 2328 } else { 2329 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 2330 #ifdef CONFIG_F2FS_CHECK_FS 2331 mir_exist = f2fs_test_and_clear_bit(offset, 2332 se->cur_valid_map_mir); 2333 if (unlikely(exist != mir_exist)) { 2334 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2335 blkaddr, exist); 2336 f2fs_bug_on(sbi, 1); 2337 } 2338 #endif 2339 if (unlikely(!exist)) { 2340 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", 2341 blkaddr); 2342 f2fs_bug_on(sbi, 1); 2343 se->valid_blocks++; 2344 del = 0; 2345 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2346 /* 2347 * If checkpoints are off, we must not reuse data that 2348 * was used in the previous checkpoint. If it was used 2349 * before, we must track that to know how much space we 2350 * really have. 2351 */ 2352 if (f2fs_test_bit(offset, se->ckpt_valid_map)) { 2353 spin_lock(&sbi->stat_lock); 2354 sbi->unusable_block_count++; 2355 spin_unlock(&sbi->stat_lock); 2356 } 2357 } 2358 2359 if (f2fs_block_unit_discard(sbi) && 2360 f2fs_test_and_clear_bit(offset, se->discard_map)) 2361 sbi->discard_blks++; 2362 } 2363 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 2364 se->ckpt_valid_blocks += del; 2365 2366 __mark_sit_entry_dirty(sbi, segno); 2367 2368 /* update total number of valid blocks to be written in ckpt area */ 2369 SIT_I(sbi)->written_valid_blocks += del; 2370 2371 if (__is_large_section(sbi)) 2372 get_sec_entry(sbi, segno)->valid_blocks += del; 2373 } 2374 2375 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 2376 { 2377 unsigned int segno = GET_SEGNO(sbi, addr); 2378 struct sit_info *sit_i = SIT_I(sbi); 2379 2380 f2fs_bug_on(sbi, addr == NULL_ADDR); 2381 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2382 return; 2383 2384 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr); 2385 f2fs_invalidate_compress_page(sbi, addr); 2386 2387 /* add it into sit main buffer */ 2388 down_write(&sit_i->sentry_lock); 2389 2390 update_segment_mtime(sbi, addr, 0); 2391 update_sit_entry(sbi, addr, -1); 2392 2393 /* add it into dirty seglist */ 2394 locate_dirty_segment(sbi, segno); 2395 2396 up_write(&sit_i->sentry_lock); 2397 } 2398 2399 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2400 { 2401 struct sit_info *sit_i = SIT_I(sbi); 2402 unsigned int segno, offset; 2403 struct seg_entry *se; 2404 bool is_cp = false; 2405 2406 if (!__is_valid_data_blkaddr(blkaddr)) 2407 return true; 2408 2409 down_read(&sit_i->sentry_lock); 2410 2411 segno = GET_SEGNO(sbi, blkaddr); 2412 se = get_seg_entry(sbi, segno); 2413 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2414 2415 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2416 is_cp = true; 2417 2418 up_read(&sit_i->sentry_lock); 2419 2420 return is_cp; 2421 } 2422 2423 /* 2424 * This function should be resided under the curseg_mutex lock 2425 */ 2426 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 2427 struct f2fs_summary *sum) 2428 { 2429 struct curseg_info *curseg = CURSEG_I(sbi, type); 2430 void *addr = curseg->sum_blk; 2431 2432 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 2433 memcpy(addr, sum, sizeof(struct f2fs_summary)); 2434 } 2435 2436 /* 2437 * Calculate the number of current summary pages for writing 2438 */ 2439 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2440 { 2441 int valid_sum_count = 0; 2442 int i, sum_in_page; 2443 2444 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2445 if (sbi->ckpt->alloc_type[i] == SSR) 2446 valid_sum_count += sbi->blocks_per_seg; 2447 else { 2448 if (for_ra) 2449 valid_sum_count += le16_to_cpu( 2450 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2451 else 2452 valid_sum_count += curseg_blkoff(sbi, i); 2453 } 2454 } 2455 2456 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 2457 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2458 if (valid_sum_count <= sum_in_page) 2459 return 1; 2460 else if ((valid_sum_count - sum_in_page) <= 2461 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2462 return 2; 2463 return 3; 2464 } 2465 2466 /* 2467 * Caller should put this summary page 2468 */ 2469 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 2470 { 2471 if (unlikely(f2fs_cp_error(sbi))) 2472 return ERR_PTR(-EIO); 2473 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno)); 2474 } 2475 2476 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2477 void *src, block_t blk_addr) 2478 { 2479 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2480 2481 memcpy(page_address(page), src, PAGE_SIZE); 2482 set_page_dirty(page); 2483 f2fs_put_page(page, 1); 2484 } 2485 2486 static void write_sum_page(struct f2fs_sb_info *sbi, 2487 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2488 { 2489 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2490 } 2491 2492 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2493 int type, block_t blk_addr) 2494 { 2495 struct curseg_info *curseg = CURSEG_I(sbi, type); 2496 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2497 struct f2fs_summary_block *src = curseg->sum_blk; 2498 struct f2fs_summary_block *dst; 2499 2500 dst = (struct f2fs_summary_block *)page_address(page); 2501 memset(dst, 0, PAGE_SIZE); 2502 2503 mutex_lock(&curseg->curseg_mutex); 2504 2505 down_read(&curseg->journal_rwsem); 2506 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2507 up_read(&curseg->journal_rwsem); 2508 2509 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2510 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2511 2512 mutex_unlock(&curseg->curseg_mutex); 2513 2514 set_page_dirty(page); 2515 f2fs_put_page(page, 1); 2516 } 2517 2518 static int is_next_segment_free(struct f2fs_sb_info *sbi, 2519 struct curseg_info *curseg, int type) 2520 { 2521 unsigned int segno = curseg->segno + 1; 2522 struct free_segmap_info *free_i = FREE_I(sbi); 2523 2524 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 2525 return !test_bit(segno, free_i->free_segmap); 2526 return 0; 2527 } 2528 2529 /* 2530 * Find a new segment from the free segments bitmap to right order 2531 * This function should be returned with success, otherwise BUG 2532 */ 2533 static void get_new_segment(struct f2fs_sb_info *sbi, 2534 unsigned int *newseg, bool new_sec, int dir) 2535 { 2536 struct free_segmap_info *free_i = FREE_I(sbi); 2537 unsigned int segno, secno, zoneno; 2538 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2539 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2540 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2541 unsigned int left_start = hint; 2542 bool init = true; 2543 int go_left = 0; 2544 int i; 2545 2546 spin_lock(&free_i->segmap_lock); 2547 2548 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 2549 segno = find_next_zero_bit(free_i->free_segmap, 2550 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2551 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2552 goto got_it; 2553 } 2554 find_other_zone: 2555 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2556 if (secno >= MAIN_SECS(sbi)) { 2557 if (dir == ALLOC_RIGHT) { 2558 secno = find_first_zero_bit(free_i->free_secmap, 2559 MAIN_SECS(sbi)); 2560 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 2561 } else { 2562 go_left = 1; 2563 left_start = hint - 1; 2564 } 2565 } 2566 if (go_left == 0) 2567 goto skip_left; 2568 2569 while (test_bit(left_start, free_i->free_secmap)) { 2570 if (left_start > 0) { 2571 left_start--; 2572 continue; 2573 } 2574 left_start = find_first_zero_bit(free_i->free_secmap, 2575 MAIN_SECS(sbi)); 2576 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 2577 break; 2578 } 2579 secno = left_start; 2580 skip_left: 2581 segno = GET_SEG_FROM_SEC(sbi, secno); 2582 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2583 2584 /* give up on finding another zone */ 2585 if (!init) 2586 goto got_it; 2587 if (sbi->secs_per_zone == 1) 2588 goto got_it; 2589 if (zoneno == old_zoneno) 2590 goto got_it; 2591 if (dir == ALLOC_LEFT) { 2592 if (!go_left && zoneno + 1 >= total_zones) 2593 goto got_it; 2594 if (go_left && zoneno == 0) 2595 goto got_it; 2596 } 2597 for (i = 0; i < NR_CURSEG_TYPE; i++) 2598 if (CURSEG_I(sbi, i)->zone == zoneno) 2599 break; 2600 2601 if (i < NR_CURSEG_TYPE) { 2602 /* zone is in user, try another */ 2603 if (go_left) 2604 hint = zoneno * sbi->secs_per_zone - 1; 2605 else if (zoneno + 1 >= total_zones) 2606 hint = 0; 2607 else 2608 hint = (zoneno + 1) * sbi->secs_per_zone; 2609 init = false; 2610 goto find_other_zone; 2611 } 2612 got_it: 2613 /* set it as dirty segment in free segmap */ 2614 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2615 __set_inuse(sbi, segno); 2616 *newseg = segno; 2617 spin_unlock(&free_i->segmap_lock); 2618 } 2619 2620 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2621 { 2622 struct curseg_info *curseg = CURSEG_I(sbi, type); 2623 struct summary_footer *sum_footer; 2624 unsigned short seg_type = curseg->seg_type; 2625 2626 curseg->inited = true; 2627 curseg->segno = curseg->next_segno; 2628 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2629 curseg->next_blkoff = 0; 2630 curseg->next_segno = NULL_SEGNO; 2631 2632 sum_footer = &(curseg->sum_blk->footer); 2633 memset(sum_footer, 0, sizeof(struct summary_footer)); 2634 2635 sanity_check_seg_type(sbi, seg_type); 2636 2637 if (IS_DATASEG(seg_type)) 2638 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2639 if (IS_NODESEG(seg_type)) 2640 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2641 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified); 2642 } 2643 2644 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2645 { 2646 struct curseg_info *curseg = CURSEG_I(sbi, type); 2647 unsigned short seg_type = curseg->seg_type; 2648 2649 sanity_check_seg_type(sbi, seg_type); 2650 if (f2fs_need_rand_seg(sbi)) 2651 return prandom_u32() % (MAIN_SECS(sbi) * sbi->segs_per_sec); 2652 2653 /* if segs_per_sec is large than 1, we need to keep original policy. */ 2654 if (__is_large_section(sbi)) 2655 return curseg->segno; 2656 2657 /* inmem log may not locate on any segment after mount */ 2658 if (!curseg->inited) 2659 return 0; 2660 2661 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2662 return 0; 2663 2664 if (test_opt(sbi, NOHEAP) && 2665 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))) 2666 return 0; 2667 2668 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2669 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2670 2671 /* find segments from 0 to reuse freed segments */ 2672 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2673 return 0; 2674 2675 return curseg->segno; 2676 } 2677 2678 /* 2679 * Allocate a current working segment. 2680 * This function always allocates a free segment in LFS manner. 2681 */ 2682 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2683 { 2684 struct curseg_info *curseg = CURSEG_I(sbi, type); 2685 unsigned short seg_type = curseg->seg_type; 2686 unsigned int segno = curseg->segno; 2687 int dir = ALLOC_LEFT; 2688 2689 if (curseg->inited) 2690 write_sum_page(sbi, curseg->sum_blk, 2691 GET_SUM_BLOCK(sbi, segno)); 2692 if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA) 2693 dir = ALLOC_RIGHT; 2694 2695 if (test_opt(sbi, NOHEAP)) 2696 dir = ALLOC_RIGHT; 2697 2698 segno = __get_next_segno(sbi, type); 2699 get_new_segment(sbi, &segno, new_sec, dir); 2700 curseg->next_segno = segno; 2701 reset_curseg(sbi, type, 1); 2702 curseg->alloc_type = LFS; 2703 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 2704 curseg->fragment_remained_chunk = 2705 prandom_u32() % sbi->max_fragment_chunk + 1; 2706 } 2707 2708 static int __next_free_blkoff(struct f2fs_sb_info *sbi, 2709 int segno, block_t start) 2710 { 2711 struct seg_entry *se = get_seg_entry(sbi, segno); 2712 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2713 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2714 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2715 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2716 int i; 2717 2718 for (i = 0; i < entries; i++) 2719 target_map[i] = ckpt_map[i] | cur_map[i]; 2720 2721 return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 2722 } 2723 2724 /* 2725 * If a segment is written by LFS manner, next block offset is just obtained 2726 * by increasing the current block offset. However, if a segment is written by 2727 * SSR manner, next block offset obtained by calling __next_free_blkoff 2728 */ 2729 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 2730 struct curseg_info *seg) 2731 { 2732 if (seg->alloc_type == SSR) { 2733 seg->next_blkoff = 2734 __next_free_blkoff(sbi, seg->segno, 2735 seg->next_blkoff + 1); 2736 } else { 2737 seg->next_blkoff++; 2738 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) { 2739 /* To allocate block chunks in different sizes, use random number */ 2740 if (--seg->fragment_remained_chunk <= 0) { 2741 seg->fragment_remained_chunk = 2742 prandom_u32() % sbi->max_fragment_chunk + 1; 2743 seg->next_blkoff += 2744 prandom_u32() % sbi->max_fragment_hole + 1; 2745 } 2746 } 2747 } 2748 } 2749 2750 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno) 2751 { 2752 return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg; 2753 } 2754 2755 /* 2756 * This function always allocates a used segment(from dirty seglist) by SSR 2757 * manner, so it should recover the existing segment information of valid blocks 2758 */ 2759 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush) 2760 { 2761 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2762 struct curseg_info *curseg = CURSEG_I(sbi, type); 2763 unsigned int new_segno = curseg->next_segno; 2764 struct f2fs_summary_block *sum_node; 2765 struct page *sum_page; 2766 2767 if (flush) 2768 write_sum_page(sbi, curseg->sum_blk, 2769 GET_SUM_BLOCK(sbi, curseg->segno)); 2770 2771 __set_test_and_inuse(sbi, new_segno); 2772 2773 mutex_lock(&dirty_i->seglist_lock); 2774 __remove_dirty_segment(sbi, new_segno, PRE); 2775 __remove_dirty_segment(sbi, new_segno, DIRTY); 2776 mutex_unlock(&dirty_i->seglist_lock); 2777 2778 reset_curseg(sbi, type, 1); 2779 curseg->alloc_type = SSR; 2780 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0); 2781 2782 sum_page = f2fs_get_sum_page(sbi, new_segno); 2783 if (IS_ERR(sum_page)) { 2784 /* GC won't be able to use stale summary pages by cp_error */ 2785 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE); 2786 return; 2787 } 2788 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2789 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2790 f2fs_put_page(sum_page, 1); 2791 } 2792 2793 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2794 int alloc_mode, unsigned long long age); 2795 2796 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type, 2797 int target_type, int alloc_mode, 2798 unsigned long long age) 2799 { 2800 struct curseg_info *curseg = CURSEG_I(sbi, type); 2801 2802 curseg->seg_type = target_type; 2803 2804 if (get_ssr_segment(sbi, type, alloc_mode, age)) { 2805 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno); 2806 2807 curseg->seg_type = se->type; 2808 change_curseg(sbi, type, true); 2809 } else { 2810 /* allocate cold segment by default */ 2811 curseg->seg_type = CURSEG_COLD_DATA; 2812 new_curseg(sbi, type, true); 2813 } 2814 stat_inc_seg_type(sbi, curseg); 2815 } 2816 2817 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi) 2818 { 2819 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC); 2820 2821 if (!sbi->am.atgc_enabled) 2822 return; 2823 2824 down_read(&SM_I(sbi)->curseg_lock); 2825 2826 mutex_lock(&curseg->curseg_mutex); 2827 down_write(&SIT_I(sbi)->sentry_lock); 2828 2829 get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0); 2830 2831 up_write(&SIT_I(sbi)->sentry_lock); 2832 mutex_unlock(&curseg->curseg_mutex); 2833 2834 up_read(&SM_I(sbi)->curseg_lock); 2835 2836 } 2837 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi) 2838 { 2839 __f2fs_init_atgc_curseg(sbi); 2840 } 2841 2842 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2843 { 2844 struct curseg_info *curseg = CURSEG_I(sbi, type); 2845 2846 mutex_lock(&curseg->curseg_mutex); 2847 if (!curseg->inited) 2848 goto out; 2849 2850 if (get_valid_blocks(sbi, curseg->segno, false)) { 2851 write_sum_page(sbi, curseg->sum_blk, 2852 GET_SUM_BLOCK(sbi, curseg->segno)); 2853 } else { 2854 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2855 __set_test_and_free(sbi, curseg->segno, true); 2856 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2857 } 2858 out: 2859 mutex_unlock(&curseg->curseg_mutex); 2860 } 2861 2862 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi) 2863 { 2864 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2865 2866 if (sbi->am.atgc_enabled) 2867 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2868 } 2869 2870 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2871 { 2872 struct curseg_info *curseg = CURSEG_I(sbi, type); 2873 2874 mutex_lock(&curseg->curseg_mutex); 2875 if (!curseg->inited) 2876 goto out; 2877 if (get_valid_blocks(sbi, curseg->segno, false)) 2878 goto out; 2879 2880 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2881 __set_test_and_inuse(sbi, curseg->segno); 2882 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2883 out: 2884 mutex_unlock(&curseg->curseg_mutex); 2885 } 2886 2887 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi) 2888 { 2889 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2890 2891 if (sbi->am.atgc_enabled) 2892 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2893 } 2894 2895 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2896 int alloc_mode, unsigned long long age) 2897 { 2898 struct curseg_info *curseg = CURSEG_I(sbi, type); 2899 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 2900 unsigned segno = NULL_SEGNO; 2901 unsigned short seg_type = curseg->seg_type; 2902 int i, cnt; 2903 bool reversed = false; 2904 2905 sanity_check_seg_type(sbi, seg_type); 2906 2907 /* f2fs_need_SSR() already forces to do this */ 2908 if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) { 2909 curseg->next_segno = segno; 2910 return 1; 2911 } 2912 2913 /* For node segments, let's do SSR more intensively */ 2914 if (IS_NODESEG(seg_type)) { 2915 if (seg_type >= CURSEG_WARM_NODE) { 2916 reversed = true; 2917 i = CURSEG_COLD_NODE; 2918 } else { 2919 i = CURSEG_HOT_NODE; 2920 } 2921 cnt = NR_CURSEG_NODE_TYPE; 2922 } else { 2923 if (seg_type >= CURSEG_WARM_DATA) { 2924 reversed = true; 2925 i = CURSEG_COLD_DATA; 2926 } else { 2927 i = CURSEG_HOT_DATA; 2928 } 2929 cnt = NR_CURSEG_DATA_TYPE; 2930 } 2931 2932 for (; cnt-- > 0; reversed ? i-- : i++) { 2933 if (i == seg_type) 2934 continue; 2935 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) { 2936 curseg->next_segno = segno; 2937 return 1; 2938 } 2939 } 2940 2941 /* find valid_blocks=0 in dirty list */ 2942 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2943 segno = get_free_segment(sbi); 2944 if (segno != NULL_SEGNO) { 2945 curseg->next_segno = segno; 2946 return 1; 2947 } 2948 } 2949 return 0; 2950 } 2951 2952 /* 2953 * flush out current segment and replace it with new segment 2954 * This function should be returned with success, otherwise BUG 2955 */ 2956 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 2957 int type, bool force) 2958 { 2959 struct curseg_info *curseg = CURSEG_I(sbi, type); 2960 2961 if (force) 2962 new_curseg(sbi, type, true); 2963 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 2964 curseg->seg_type == CURSEG_WARM_NODE) 2965 new_curseg(sbi, type, false); 2966 else if (curseg->alloc_type == LFS && 2967 is_next_segment_free(sbi, curseg, type) && 2968 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2969 new_curseg(sbi, type, false); 2970 else if (f2fs_need_SSR(sbi) && 2971 get_ssr_segment(sbi, type, SSR, 0)) 2972 change_curseg(sbi, type, true); 2973 else 2974 new_curseg(sbi, type, false); 2975 2976 stat_inc_seg_type(sbi, curseg); 2977 } 2978 2979 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 2980 unsigned int start, unsigned int end) 2981 { 2982 struct curseg_info *curseg = CURSEG_I(sbi, type); 2983 unsigned int segno; 2984 2985 down_read(&SM_I(sbi)->curseg_lock); 2986 mutex_lock(&curseg->curseg_mutex); 2987 down_write(&SIT_I(sbi)->sentry_lock); 2988 2989 segno = CURSEG_I(sbi, type)->segno; 2990 if (segno < start || segno > end) 2991 goto unlock; 2992 2993 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0)) 2994 change_curseg(sbi, type, true); 2995 else 2996 new_curseg(sbi, type, true); 2997 2998 stat_inc_seg_type(sbi, curseg); 2999 3000 locate_dirty_segment(sbi, segno); 3001 unlock: 3002 up_write(&SIT_I(sbi)->sentry_lock); 3003 3004 if (segno != curseg->segno) 3005 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 3006 type, segno, curseg->segno); 3007 3008 mutex_unlock(&curseg->curseg_mutex); 3009 up_read(&SM_I(sbi)->curseg_lock); 3010 } 3011 3012 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type, 3013 bool new_sec, bool force) 3014 { 3015 struct curseg_info *curseg = CURSEG_I(sbi, type); 3016 unsigned int old_segno; 3017 3018 if (!curseg->inited) 3019 goto alloc; 3020 3021 if (force || curseg->next_blkoff || 3022 get_valid_blocks(sbi, curseg->segno, new_sec)) 3023 goto alloc; 3024 3025 if (!get_ckpt_valid_blocks(sbi, curseg->segno, new_sec)) 3026 return; 3027 alloc: 3028 old_segno = curseg->segno; 3029 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true); 3030 locate_dirty_segment(sbi, old_segno); 3031 } 3032 3033 static void __allocate_new_section(struct f2fs_sb_info *sbi, 3034 int type, bool force) 3035 { 3036 __allocate_new_segment(sbi, type, true, force); 3037 } 3038 3039 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force) 3040 { 3041 down_read(&SM_I(sbi)->curseg_lock); 3042 down_write(&SIT_I(sbi)->sentry_lock); 3043 __allocate_new_section(sbi, type, force); 3044 up_write(&SIT_I(sbi)->sentry_lock); 3045 up_read(&SM_I(sbi)->curseg_lock); 3046 } 3047 3048 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 3049 { 3050 int i; 3051 3052 down_read(&SM_I(sbi)->curseg_lock); 3053 down_write(&SIT_I(sbi)->sentry_lock); 3054 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 3055 __allocate_new_segment(sbi, i, false, false); 3056 up_write(&SIT_I(sbi)->sentry_lock); 3057 up_read(&SM_I(sbi)->curseg_lock); 3058 } 3059 3060 static const struct segment_allocation default_salloc_ops = { 3061 .allocate_segment = allocate_segment_by_default, 3062 }; 3063 3064 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 3065 struct cp_control *cpc) 3066 { 3067 __u64 trim_start = cpc->trim_start; 3068 bool has_candidate = false; 3069 3070 down_write(&SIT_I(sbi)->sentry_lock); 3071 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 3072 if (add_discard_addrs(sbi, cpc, true)) { 3073 has_candidate = true; 3074 break; 3075 } 3076 } 3077 up_write(&SIT_I(sbi)->sentry_lock); 3078 3079 cpc->trim_start = trim_start; 3080 return has_candidate; 3081 } 3082 3083 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 3084 struct discard_policy *dpolicy, 3085 unsigned int start, unsigned int end) 3086 { 3087 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 3088 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 3089 struct rb_node **insert_p = NULL, *insert_parent = NULL; 3090 struct discard_cmd *dc; 3091 struct blk_plug plug; 3092 int issued; 3093 unsigned int trimmed = 0; 3094 3095 next: 3096 issued = 0; 3097 3098 mutex_lock(&dcc->cmd_lock); 3099 if (unlikely(dcc->rbtree_check)) 3100 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 3101 &dcc->root, false)); 3102 3103 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 3104 NULL, start, 3105 (struct rb_entry **)&prev_dc, 3106 (struct rb_entry **)&next_dc, 3107 &insert_p, &insert_parent, true, NULL); 3108 if (!dc) 3109 dc = next_dc; 3110 3111 blk_start_plug(&plug); 3112 3113 while (dc && dc->lstart <= end) { 3114 struct rb_node *node; 3115 int err = 0; 3116 3117 if (dc->len < dpolicy->granularity) 3118 goto skip; 3119 3120 if (dc->state != D_PREP) { 3121 list_move_tail(&dc->list, &dcc->fstrim_list); 3122 goto skip; 3123 } 3124 3125 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 3126 3127 if (issued >= dpolicy->max_requests) { 3128 start = dc->lstart + dc->len; 3129 3130 if (err) 3131 __remove_discard_cmd(sbi, dc); 3132 3133 blk_finish_plug(&plug); 3134 mutex_unlock(&dcc->cmd_lock); 3135 trimmed += __wait_all_discard_cmd(sbi, NULL); 3136 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT); 3137 goto next; 3138 } 3139 skip: 3140 node = rb_next(&dc->rb_node); 3141 if (err) 3142 __remove_discard_cmd(sbi, dc); 3143 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 3144 3145 if (fatal_signal_pending(current)) 3146 break; 3147 } 3148 3149 blk_finish_plug(&plug); 3150 mutex_unlock(&dcc->cmd_lock); 3151 3152 return trimmed; 3153 } 3154 3155 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 3156 { 3157 __u64 start = F2FS_BYTES_TO_BLK(range->start); 3158 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 3159 unsigned int start_segno, end_segno; 3160 block_t start_block, end_block; 3161 struct cp_control cpc; 3162 struct discard_policy dpolicy; 3163 unsigned long long trimmed = 0; 3164 int err = 0; 3165 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 3166 3167 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 3168 return -EINVAL; 3169 3170 if (end < MAIN_BLKADDR(sbi)) 3171 goto out; 3172 3173 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 3174 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 3175 return -EFSCORRUPTED; 3176 } 3177 3178 /* start/end segment number in main_area */ 3179 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 3180 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 3181 GET_SEGNO(sbi, end); 3182 if (need_align) { 3183 start_segno = rounddown(start_segno, sbi->segs_per_sec); 3184 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1; 3185 } 3186 3187 cpc.reason = CP_DISCARD; 3188 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 3189 cpc.trim_start = start_segno; 3190 cpc.trim_end = end_segno; 3191 3192 if (sbi->discard_blks == 0) 3193 goto out; 3194 3195 down_write(&sbi->gc_lock); 3196 err = f2fs_write_checkpoint(sbi, &cpc); 3197 up_write(&sbi->gc_lock); 3198 if (err) 3199 goto out; 3200 3201 /* 3202 * We filed discard candidates, but actually we don't need to wait for 3203 * all of them, since they'll be issued in idle time along with runtime 3204 * discard option. User configuration looks like using runtime discard 3205 * or periodic fstrim instead of it. 3206 */ 3207 if (f2fs_realtime_discard_enable(sbi)) 3208 goto out; 3209 3210 start_block = START_BLOCK(sbi, start_segno); 3211 end_block = START_BLOCK(sbi, end_segno + 1); 3212 3213 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 3214 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 3215 start_block, end_block); 3216 3217 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 3218 start_block, end_block); 3219 out: 3220 if (!err) 3221 range->len = F2FS_BLK_TO_BYTES(trimmed); 3222 return err; 3223 } 3224 3225 static bool __has_curseg_space(struct f2fs_sb_info *sbi, 3226 struct curseg_info *curseg) 3227 { 3228 return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi, 3229 curseg->segno); 3230 } 3231 3232 int f2fs_rw_hint_to_seg_type(enum rw_hint hint) 3233 { 3234 switch (hint) { 3235 case WRITE_LIFE_SHORT: 3236 return CURSEG_HOT_DATA; 3237 case WRITE_LIFE_EXTREME: 3238 return CURSEG_COLD_DATA; 3239 default: 3240 return CURSEG_WARM_DATA; 3241 } 3242 } 3243 3244 /* This returns write hints for each segment type. This hints will be 3245 * passed down to block layer. There are mapping tables which depend on 3246 * the mount option 'whint_mode'. 3247 * 3248 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET. 3249 * 3250 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users. 3251 * 3252 * User F2FS Block 3253 * ---- ---- ----- 3254 * META WRITE_LIFE_NOT_SET 3255 * HOT_NODE " 3256 * WARM_NODE " 3257 * COLD_NODE " 3258 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 3259 * extension list " " 3260 * 3261 * -- buffered io 3262 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 3263 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 3264 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 3265 * WRITE_LIFE_NONE " " 3266 * WRITE_LIFE_MEDIUM " " 3267 * WRITE_LIFE_LONG " " 3268 * 3269 * -- direct io 3270 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 3271 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 3272 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 3273 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 3274 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 3275 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 3276 * 3277 * 3) whint_mode=fs-based. F2FS passes down hints with its policy. 3278 * 3279 * User F2FS Block 3280 * ---- ---- ----- 3281 * META WRITE_LIFE_MEDIUM; 3282 * HOT_NODE WRITE_LIFE_NOT_SET 3283 * WARM_NODE " 3284 * COLD_NODE WRITE_LIFE_NONE 3285 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 3286 * extension list " " 3287 * 3288 * -- buffered io 3289 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 3290 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 3291 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG 3292 * WRITE_LIFE_NONE " " 3293 * WRITE_LIFE_MEDIUM " " 3294 * WRITE_LIFE_LONG " " 3295 * 3296 * -- direct io 3297 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 3298 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 3299 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 3300 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 3301 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 3302 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 3303 */ 3304 3305 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi, 3306 enum page_type type, enum temp_type temp) 3307 { 3308 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) { 3309 if (type == DATA) { 3310 if (temp == WARM) 3311 return WRITE_LIFE_NOT_SET; 3312 else if (temp == HOT) 3313 return WRITE_LIFE_SHORT; 3314 else if (temp == COLD) 3315 return WRITE_LIFE_EXTREME; 3316 } else { 3317 return WRITE_LIFE_NOT_SET; 3318 } 3319 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) { 3320 if (type == DATA) { 3321 if (temp == WARM) 3322 return WRITE_LIFE_LONG; 3323 else if (temp == HOT) 3324 return WRITE_LIFE_SHORT; 3325 else if (temp == COLD) 3326 return WRITE_LIFE_EXTREME; 3327 } else if (type == NODE) { 3328 if (temp == WARM || temp == HOT) 3329 return WRITE_LIFE_NOT_SET; 3330 else if (temp == COLD) 3331 return WRITE_LIFE_NONE; 3332 } else if (type == META) { 3333 return WRITE_LIFE_MEDIUM; 3334 } 3335 } 3336 return WRITE_LIFE_NOT_SET; 3337 } 3338 3339 static int __get_segment_type_2(struct f2fs_io_info *fio) 3340 { 3341 if (fio->type == DATA) 3342 return CURSEG_HOT_DATA; 3343 else 3344 return CURSEG_HOT_NODE; 3345 } 3346 3347 static int __get_segment_type_4(struct f2fs_io_info *fio) 3348 { 3349 if (fio->type == DATA) { 3350 struct inode *inode = fio->page->mapping->host; 3351 3352 if (S_ISDIR(inode->i_mode)) 3353 return CURSEG_HOT_DATA; 3354 else 3355 return CURSEG_COLD_DATA; 3356 } else { 3357 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3358 return CURSEG_WARM_NODE; 3359 else 3360 return CURSEG_COLD_NODE; 3361 } 3362 } 3363 3364 static int __get_segment_type_6(struct f2fs_io_info *fio) 3365 { 3366 if (fio->type == DATA) { 3367 struct inode *inode = fio->page->mapping->host; 3368 3369 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 3370 return CURSEG_COLD_DATA_PINNED; 3371 3372 if (page_private_gcing(fio->page)) { 3373 if (fio->sbi->am.atgc_enabled && 3374 (fio->io_type == FS_DATA_IO) && 3375 (fio->sbi->gc_mode != GC_URGENT_HIGH)) 3376 return CURSEG_ALL_DATA_ATGC; 3377 else 3378 return CURSEG_COLD_DATA; 3379 } 3380 if (file_is_cold(inode) || f2fs_need_compress_data(inode)) 3381 return CURSEG_COLD_DATA; 3382 if (file_is_hot(inode) || 3383 is_inode_flag_set(inode, FI_HOT_DATA) || 3384 f2fs_is_atomic_file(inode) || 3385 f2fs_is_volatile_file(inode)) 3386 return CURSEG_HOT_DATA; 3387 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 3388 } else { 3389 if (IS_DNODE(fio->page)) 3390 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3391 CURSEG_HOT_NODE; 3392 return CURSEG_COLD_NODE; 3393 } 3394 } 3395 3396 static int __get_segment_type(struct f2fs_io_info *fio) 3397 { 3398 int type = 0; 3399 3400 switch (F2FS_OPTION(fio->sbi).active_logs) { 3401 case 2: 3402 type = __get_segment_type_2(fio); 3403 break; 3404 case 4: 3405 type = __get_segment_type_4(fio); 3406 break; 3407 case 6: 3408 type = __get_segment_type_6(fio); 3409 break; 3410 default: 3411 f2fs_bug_on(fio->sbi, true); 3412 } 3413 3414 if (IS_HOT(type)) 3415 fio->temp = HOT; 3416 else if (IS_WARM(type)) 3417 fio->temp = WARM; 3418 else 3419 fio->temp = COLD; 3420 return type; 3421 } 3422 3423 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3424 block_t old_blkaddr, block_t *new_blkaddr, 3425 struct f2fs_summary *sum, int type, 3426 struct f2fs_io_info *fio) 3427 { 3428 struct sit_info *sit_i = SIT_I(sbi); 3429 struct curseg_info *curseg = CURSEG_I(sbi, type); 3430 unsigned long long old_mtime; 3431 bool from_gc = (type == CURSEG_ALL_DATA_ATGC); 3432 struct seg_entry *se = NULL; 3433 3434 down_read(&SM_I(sbi)->curseg_lock); 3435 3436 mutex_lock(&curseg->curseg_mutex); 3437 down_write(&sit_i->sentry_lock); 3438 3439 if (from_gc) { 3440 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO); 3441 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr)); 3442 sanity_check_seg_type(sbi, se->type); 3443 f2fs_bug_on(sbi, IS_NODESEG(se->type)); 3444 } 3445 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3446 3447 f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg); 3448 3449 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3450 3451 /* 3452 * __add_sum_entry should be resided under the curseg_mutex 3453 * because, this function updates a summary entry in the 3454 * current summary block. 3455 */ 3456 __add_sum_entry(sbi, type, sum); 3457 3458 __refresh_next_blkoff(sbi, curseg); 3459 3460 stat_inc_block_count(sbi, curseg); 3461 3462 if (from_gc) { 3463 old_mtime = get_segment_mtime(sbi, old_blkaddr); 3464 } else { 3465 update_segment_mtime(sbi, old_blkaddr, 0); 3466 old_mtime = 0; 3467 } 3468 update_segment_mtime(sbi, *new_blkaddr, old_mtime); 3469 3470 /* 3471 * SIT information should be updated before segment allocation, 3472 * since SSR needs latest valid block information. 3473 */ 3474 update_sit_entry(sbi, *new_blkaddr, 1); 3475 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 3476 update_sit_entry(sbi, old_blkaddr, -1); 3477 3478 if (!__has_curseg_space(sbi, curseg)) { 3479 if (from_gc) 3480 get_atssr_segment(sbi, type, se->type, 3481 AT_SSR, se->mtime); 3482 else 3483 sit_i->s_ops->allocate_segment(sbi, type, false); 3484 } 3485 /* 3486 * segment dirty status should be updated after segment allocation, 3487 * so we just need to update status only one time after previous 3488 * segment being closed. 3489 */ 3490 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3491 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3492 3493 up_write(&sit_i->sentry_lock); 3494 3495 if (page && IS_NODESEG(type)) { 3496 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3497 3498 f2fs_inode_chksum_set(sbi, page); 3499 } 3500 3501 if (fio) { 3502 struct f2fs_bio_info *io; 3503 3504 if (F2FS_IO_ALIGNED(sbi)) 3505 fio->retry = false; 3506 3507 INIT_LIST_HEAD(&fio->list); 3508 fio->in_list = true; 3509 io = sbi->write_io[fio->type] + fio->temp; 3510 spin_lock(&io->io_lock); 3511 list_add_tail(&fio->list, &io->io_list); 3512 spin_unlock(&io->io_lock); 3513 } 3514 3515 mutex_unlock(&curseg->curseg_mutex); 3516 3517 up_read(&SM_I(sbi)->curseg_lock); 3518 } 3519 3520 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino, 3521 block_t blkaddr, unsigned int blkcnt) 3522 { 3523 if (!f2fs_is_multi_device(sbi)) 3524 return; 3525 3526 while (1) { 3527 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr); 3528 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1; 3529 3530 /* update device state for fsync */ 3531 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO); 3532 3533 /* update device state for checkpoint */ 3534 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3535 spin_lock(&sbi->dev_lock); 3536 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3537 spin_unlock(&sbi->dev_lock); 3538 } 3539 3540 if (blkcnt <= blks) 3541 break; 3542 blkcnt -= blks; 3543 blkaddr += blks; 3544 } 3545 } 3546 3547 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3548 { 3549 int type = __get_segment_type(fio); 3550 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA); 3551 3552 if (keep_order) 3553 down_read(&fio->sbi->io_order_lock); 3554 reallocate: 3555 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3556 &fio->new_blkaddr, sum, type, fio); 3557 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) { 3558 invalidate_mapping_pages(META_MAPPING(fio->sbi), 3559 fio->old_blkaddr, fio->old_blkaddr); 3560 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr); 3561 } 3562 3563 /* writeout dirty page into bdev */ 3564 f2fs_submit_page_write(fio); 3565 if (fio->retry) { 3566 fio->old_blkaddr = fio->new_blkaddr; 3567 goto reallocate; 3568 } 3569 3570 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1); 3571 3572 if (keep_order) 3573 up_read(&fio->sbi->io_order_lock); 3574 } 3575 3576 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 3577 enum iostat_type io_type) 3578 { 3579 struct f2fs_io_info fio = { 3580 .sbi = sbi, 3581 .type = META, 3582 .temp = HOT, 3583 .op = REQ_OP_WRITE, 3584 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3585 .old_blkaddr = page->index, 3586 .new_blkaddr = page->index, 3587 .page = page, 3588 .encrypted_page = NULL, 3589 .in_list = false, 3590 }; 3591 3592 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 3593 fio.op_flags &= ~REQ_META; 3594 3595 set_page_writeback(page); 3596 ClearPageError(page); 3597 f2fs_submit_page_write(&fio); 3598 3599 stat_inc_meta_count(sbi, page->index); 3600 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE); 3601 } 3602 3603 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3604 { 3605 struct f2fs_summary sum; 3606 3607 set_summary(&sum, nid, 0, 0); 3608 do_write_page(&sum, fio); 3609 3610 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 3611 } 3612 3613 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3614 struct f2fs_io_info *fio) 3615 { 3616 struct f2fs_sb_info *sbi = fio->sbi; 3617 struct f2fs_summary sum; 3618 3619 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3620 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3621 do_write_page(&sum, fio); 3622 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3623 3624 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE); 3625 } 3626 3627 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3628 { 3629 int err; 3630 struct f2fs_sb_info *sbi = fio->sbi; 3631 unsigned int segno; 3632 3633 fio->new_blkaddr = fio->old_blkaddr; 3634 /* i/o temperature is needed for passing down write hints */ 3635 __get_segment_type(fio); 3636 3637 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3638 3639 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 3640 set_sbi_flag(sbi, SBI_NEED_FSCK); 3641 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 3642 __func__, segno); 3643 err = -EFSCORRUPTED; 3644 goto drop_bio; 3645 } 3646 3647 if (f2fs_cp_error(sbi)) { 3648 err = -EIO; 3649 goto drop_bio; 3650 } 3651 3652 invalidate_mapping_pages(META_MAPPING(sbi), 3653 fio->new_blkaddr, fio->new_blkaddr); 3654 3655 stat_inc_inplace_blocks(fio->sbi); 3656 3657 if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE))) 3658 err = f2fs_merge_page_bio(fio); 3659 else 3660 err = f2fs_submit_page_bio(fio); 3661 if (!err) { 3662 f2fs_update_device_state(fio->sbi, fio->ino, 3663 fio->new_blkaddr, 1); 3664 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 3665 } 3666 3667 return err; 3668 drop_bio: 3669 if (fio->bio && *(fio->bio)) { 3670 struct bio *bio = *(fio->bio); 3671 3672 bio->bi_status = BLK_STS_IOERR; 3673 bio_endio(bio); 3674 *(fio->bio) = NULL; 3675 } 3676 return err; 3677 } 3678 3679 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 3680 unsigned int segno) 3681 { 3682 int i; 3683 3684 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 3685 if (CURSEG_I(sbi, i)->segno == segno) 3686 break; 3687 } 3688 return i; 3689 } 3690 3691 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 3692 block_t old_blkaddr, block_t new_blkaddr, 3693 bool recover_curseg, bool recover_newaddr, 3694 bool from_gc) 3695 { 3696 struct sit_info *sit_i = SIT_I(sbi); 3697 struct curseg_info *curseg; 3698 unsigned int segno, old_cursegno; 3699 struct seg_entry *se; 3700 int type; 3701 unsigned short old_blkoff; 3702 unsigned char old_alloc_type; 3703 3704 segno = GET_SEGNO(sbi, new_blkaddr); 3705 se = get_seg_entry(sbi, segno); 3706 type = se->type; 3707 3708 down_write(&SM_I(sbi)->curseg_lock); 3709 3710 if (!recover_curseg) { 3711 /* for recovery flow */ 3712 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 3713 if (old_blkaddr == NULL_ADDR) 3714 type = CURSEG_COLD_DATA; 3715 else 3716 type = CURSEG_WARM_DATA; 3717 } 3718 } else { 3719 if (IS_CURSEG(sbi, segno)) { 3720 /* se->type is volatile as SSR allocation */ 3721 type = __f2fs_get_curseg(sbi, segno); 3722 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 3723 } else { 3724 type = CURSEG_WARM_DATA; 3725 } 3726 } 3727 3728 f2fs_bug_on(sbi, !IS_DATASEG(type)); 3729 curseg = CURSEG_I(sbi, type); 3730 3731 mutex_lock(&curseg->curseg_mutex); 3732 down_write(&sit_i->sentry_lock); 3733 3734 old_cursegno = curseg->segno; 3735 old_blkoff = curseg->next_blkoff; 3736 old_alloc_type = curseg->alloc_type; 3737 3738 /* change the current segment */ 3739 if (segno != curseg->segno) { 3740 curseg->next_segno = segno; 3741 change_curseg(sbi, type, true); 3742 } 3743 3744 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 3745 __add_sum_entry(sbi, type, sum); 3746 3747 if (!recover_curseg || recover_newaddr) { 3748 if (!from_gc) 3749 update_segment_mtime(sbi, new_blkaddr, 0); 3750 update_sit_entry(sbi, new_blkaddr, 1); 3751 } 3752 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 3753 invalidate_mapping_pages(META_MAPPING(sbi), 3754 old_blkaddr, old_blkaddr); 3755 f2fs_invalidate_compress_page(sbi, old_blkaddr); 3756 if (!from_gc) 3757 update_segment_mtime(sbi, old_blkaddr, 0); 3758 update_sit_entry(sbi, old_blkaddr, -1); 3759 } 3760 3761 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3762 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 3763 3764 locate_dirty_segment(sbi, old_cursegno); 3765 3766 if (recover_curseg) { 3767 if (old_cursegno != curseg->segno) { 3768 curseg->next_segno = old_cursegno; 3769 change_curseg(sbi, type, true); 3770 } 3771 curseg->next_blkoff = old_blkoff; 3772 curseg->alloc_type = old_alloc_type; 3773 } 3774 3775 up_write(&sit_i->sentry_lock); 3776 mutex_unlock(&curseg->curseg_mutex); 3777 up_write(&SM_I(sbi)->curseg_lock); 3778 } 3779 3780 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 3781 block_t old_addr, block_t new_addr, 3782 unsigned char version, bool recover_curseg, 3783 bool recover_newaddr) 3784 { 3785 struct f2fs_summary sum; 3786 3787 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 3788 3789 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 3790 recover_curseg, recover_newaddr, false); 3791 3792 f2fs_update_data_blkaddr(dn, new_addr); 3793 } 3794 3795 void f2fs_wait_on_page_writeback(struct page *page, 3796 enum page_type type, bool ordered, bool locked) 3797 { 3798 if (PageWriteback(page)) { 3799 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 3800 3801 /* submit cached LFS IO */ 3802 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type); 3803 /* sbumit cached IPU IO */ 3804 f2fs_submit_merged_ipu_write(sbi, NULL, page); 3805 if (ordered) { 3806 wait_on_page_writeback(page); 3807 f2fs_bug_on(sbi, locked && PageWriteback(page)); 3808 } else { 3809 wait_for_stable_page(page); 3810 } 3811 } 3812 } 3813 3814 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 3815 { 3816 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3817 struct page *cpage; 3818 3819 if (!f2fs_post_read_required(inode)) 3820 return; 3821 3822 if (!__is_valid_data_blkaddr(blkaddr)) 3823 return; 3824 3825 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3826 if (cpage) { 3827 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 3828 f2fs_put_page(cpage, 1); 3829 } 3830 } 3831 3832 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 3833 block_t len) 3834 { 3835 block_t i; 3836 3837 for (i = 0; i < len; i++) 3838 f2fs_wait_on_block_writeback(inode, blkaddr + i); 3839 } 3840 3841 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 3842 { 3843 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3844 struct curseg_info *seg_i; 3845 unsigned char *kaddr; 3846 struct page *page; 3847 block_t start; 3848 int i, j, offset; 3849 3850 start = start_sum_block(sbi); 3851 3852 page = f2fs_get_meta_page(sbi, start++); 3853 if (IS_ERR(page)) 3854 return PTR_ERR(page); 3855 kaddr = (unsigned char *)page_address(page); 3856 3857 /* Step 1: restore nat cache */ 3858 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3859 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3860 3861 /* Step 2: restore sit cache */ 3862 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3863 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3864 offset = 2 * SUM_JOURNAL_SIZE; 3865 3866 /* Step 3: restore summary entries */ 3867 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3868 unsigned short blk_off; 3869 unsigned int segno; 3870 3871 seg_i = CURSEG_I(sbi, i); 3872 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3873 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3874 seg_i->next_segno = segno; 3875 reset_curseg(sbi, i, 0); 3876 seg_i->alloc_type = ckpt->alloc_type[i]; 3877 seg_i->next_blkoff = blk_off; 3878 3879 if (seg_i->alloc_type == SSR) 3880 blk_off = sbi->blocks_per_seg; 3881 3882 for (j = 0; j < blk_off; j++) { 3883 struct f2fs_summary *s; 3884 3885 s = (struct f2fs_summary *)(kaddr + offset); 3886 seg_i->sum_blk->entries[j] = *s; 3887 offset += SUMMARY_SIZE; 3888 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3889 SUM_FOOTER_SIZE) 3890 continue; 3891 3892 f2fs_put_page(page, 1); 3893 page = NULL; 3894 3895 page = f2fs_get_meta_page(sbi, start++); 3896 if (IS_ERR(page)) 3897 return PTR_ERR(page); 3898 kaddr = (unsigned char *)page_address(page); 3899 offset = 0; 3900 } 3901 } 3902 f2fs_put_page(page, 1); 3903 return 0; 3904 } 3905 3906 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3907 { 3908 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3909 struct f2fs_summary_block *sum; 3910 struct curseg_info *curseg; 3911 struct page *new; 3912 unsigned short blk_off; 3913 unsigned int segno = 0; 3914 block_t blk_addr = 0; 3915 int err = 0; 3916 3917 /* get segment number and block addr */ 3918 if (IS_DATASEG(type)) { 3919 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3920 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3921 CURSEG_HOT_DATA]); 3922 if (__exist_node_summaries(sbi)) 3923 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type); 3924 else 3925 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3926 } else { 3927 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3928 CURSEG_HOT_NODE]); 3929 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3930 CURSEG_HOT_NODE]); 3931 if (__exist_node_summaries(sbi)) 3932 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3933 type - CURSEG_HOT_NODE); 3934 else 3935 blk_addr = GET_SUM_BLOCK(sbi, segno); 3936 } 3937 3938 new = f2fs_get_meta_page(sbi, blk_addr); 3939 if (IS_ERR(new)) 3940 return PTR_ERR(new); 3941 sum = (struct f2fs_summary_block *)page_address(new); 3942 3943 if (IS_NODESEG(type)) { 3944 if (__exist_node_summaries(sbi)) { 3945 struct f2fs_summary *ns = &sum->entries[0]; 3946 int i; 3947 3948 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3949 ns->version = 0; 3950 ns->ofs_in_node = 0; 3951 } 3952 } else { 3953 err = f2fs_restore_node_summary(sbi, segno, sum); 3954 if (err) 3955 goto out; 3956 } 3957 } 3958 3959 /* set uncompleted segment to curseg */ 3960 curseg = CURSEG_I(sbi, type); 3961 mutex_lock(&curseg->curseg_mutex); 3962 3963 /* update journal info */ 3964 down_write(&curseg->journal_rwsem); 3965 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3966 up_write(&curseg->journal_rwsem); 3967 3968 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3969 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3970 curseg->next_segno = segno; 3971 reset_curseg(sbi, type, 0); 3972 curseg->alloc_type = ckpt->alloc_type[type]; 3973 curseg->next_blkoff = blk_off; 3974 mutex_unlock(&curseg->curseg_mutex); 3975 out: 3976 f2fs_put_page(new, 1); 3977 return err; 3978 } 3979 3980 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3981 { 3982 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3983 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3984 int type = CURSEG_HOT_DATA; 3985 int err; 3986 3987 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 3988 int npages = f2fs_npages_for_summary_flush(sbi, true); 3989 3990 if (npages >= 2) 3991 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 3992 META_CP, true); 3993 3994 /* restore for compacted data summary */ 3995 err = read_compacted_summaries(sbi); 3996 if (err) 3997 return err; 3998 type = CURSEG_HOT_NODE; 3999 } 4000 4001 if (__exist_node_summaries(sbi)) 4002 f2fs_ra_meta_pages(sbi, 4003 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type), 4004 NR_CURSEG_PERSIST_TYPE - type, META_CP, true); 4005 4006 for (; type <= CURSEG_COLD_NODE; type++) { 4007 err = read_normal_summaries(sbi, type); 4008 if (err) 4009 return err; 4010 } 4011 4012 /* sanity check for summary blocks */ 4013 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 4014 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 4015 f2fs_err(sbi, "invalid journal entries nats %u sits %u", 4016 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 4017 return -EINVAL; 4018 } 4019 4020 return 0; 4021 } 4022 4023 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 4024 { 4025 struct page *page; 4026 unsigned char *kaddr; 4027 struct f2fs_summary *summary; 4028 struct curseg_info *seg_i; 4029 int written_size = 0; 4030 int i, j; 4031 4032 page = f2fs_grab_meta_page(sbi, blkaddr++); 4033 kaddr = (unsigned char *)page_address(page); 4034 memset(kaddr, 0, PAGE_SIZE); 4035 4036 /* Step 1: write nat cache */ 4037 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4038 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 4039 written_size += SUM_JOURNAL_SIZE; 4040 4041 /* Step 2: write sit cache */ 4042 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4043 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 4044 written_size += SUM_JOURNAL_SIZE; 4045 4046 /* Step 3: write summary entries */ 4047 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4048 unsigned short blkoff; 4049 4050 seg_i = CURSEG_I(sbi, i); 4051 if (sbi->ckpt->alloc_type[i] == SSR) 4052 blkoff = sbi->blocks_per_seg; 4053 else 4054 blkoff = curseg_blkoff(sbi, i); 4055 4056 for (j = 0; j < blkoff; j++) { 4057 if (!page) { 4058 page = f2fs_grab_meta_page(sbi, blkaddr++); 4059 kaddr = (unsigned char *)page_address(page); 4060 memset(kaddr, 0, PAGE_SIZE); 4061 written_size = 0; 4062 } 4063 summary = (struct f2fs_summary *)(kaddr + written_size); 4064 *summary = seg_i->sum_blk->entries[j]; 4065 written_size += SUMMARY_SIZE; 4066 4067 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 4068 SUM_FOOTER_SIZE) 4069 continue; 4070 4071 set_page_dirty(page); 4072 f2fs_put_page(page, 1); 4073 page = NULL; 4074 } 4075 } 4076 if (page) { 4077 set_page_dirty(page); 4078 f2fs_put_page(page, 1); 4079 } 4080 } 4081 4082 static void write_normal_summaries(struct f2fs_sb_info *sbi, 4083 block_t blkaddr, int type) 4084 { 4085 int i, end; 4086 4087 if (IS_DATASEG(type)) 4088 end = type + NR_CURSEG_DATA_TYPE; 4089 else 4090 end = type + NR_CURSEG_NODE_TYPE; 4091 4092 for (i = type; i < end; i++) 4093 write_current_sum_page(sbi, i, blkaddr + (i - type)); 4094 } 4095 4096 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4097 { 4098 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 4099 write_compacted_summaries(sbi, start_blk); 4100 else 4101 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 4102 } 4103 4104 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4105 { 4106 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 4107 } 4108 4109 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 4110 unsigned int val, int alloc) 4111 { 4112 int i; 4113 4114 if (type == NAT_JOURNAL) { 4115 for (i = 0; i < nats_in_cursum(journal); i++) { 4116 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 4117 return i; 4118 } 4119 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 4120 return update_nats_in_cursum(journal, 1); 4121 } else if (type == SIT_JOURNAL) { 4122 for (i = 0; i < sits_in_cursum(journal); i++) 4123 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 4124 return i; 4125 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 4126 return update_sits_in_cursum(journal, 1); 4127 } 4128 return -1; 4129 } 4130 4131 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 4132 unsigned int segno) 4133 { 4134 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno)); 4135 } 4136 4137 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 4138 unsigned int start) 4139 { 4140 struct sit_info *sit_i = SIT_I(sbi); 4141 struct page *page; 4142 pgoff_t src_off, dst_off; 4143 4144 src_off = current_sit_addr(sbi, start); 4145 dst_off = next_sit_addr(sbi, src_off); 4146 4147 page = f2fs_grab_meta_page(sbi, dst_off); 4148 seg_info_to_sit_page(sbi, page, start); 4149 4150 set_page_dirty(page); 4151 set_to_next_sit(sit_i, start); 4152 4153 return page; 4154 } 4155 4156 static struct sit_entry_set *grab_sit_entry_set(void) 4157 { 4158 struct sit_entry_set *ses = 4159 f2fs_kmem_cache_alloc(sit_entry_set_slab, 4160 GFP_NOFS, true, NULL); 4161 4162 ses->entry_cnt = 0; 4163 INIT_LIST_HEAD(&ses->set_list); 4164 return ses; 4165 } 4166 4167 static void release_sit_entry_set(struct sit_entry_set *ses) 4168 { 4169 list_del(&ses->set_list); 4170 kmem_cache_free(sit_entry_set_slab, ses); 4171 } 4172 4173 static void adjust_sit_entry_set(struct sit_entry_set *ses, 4174 struct list_head *head) 4175 { 4176 struct sit_entry_set *next = ses; 4177 4178 if (list_is_last(&ses->set_list, head)) 4179 return; 4180 4181 list_for_each_entry_continue(next, head, set_list) 4182 if (ses->entry_cnt <= next->entry_cnt) 4183 break; 4184 4185 list_move_tail(&ses->set_list, &next->set_list); 4186 } 4187 4188 static void add_sit_entry(unsigned int segno, struct list_head *head) 4189 { 4190 struct sit_entry_set *ses; 4191 unsigned int start_segno = START_SEGNO(segno); 4192 4193 list_for_each_entry(ses, head, set_list) { 4194 if (ses->start_segno == start_segno) { 4195 ses->entry_cnt++; 4196 adjust_sit_entry_set(ses, head); 4197 return; 4198 } 4199 } 4200 4201 ses = grab_sit_entry_set(); 4202 4203 ses->start_segno = start_segno; 4204 ses->entry_cnt++; 4205 list_add(&ses->set_list, head); 4206 } 4207 4208 static void add_sits_in_set(struct f2fs_sb_info *sbi) 4209 { 4210 struct f2fs_sm_info *sm_info = SM_I(sbi); 4211 struct list_head *set_list = &sm_info->sit_entry_set; 4212 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 4213 unsigned int segno; 4214 4215 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 4216 add_sit_entry(segno, set_list); 4217 } 4218 4219 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 4220 { 4221 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4222 struct f2fs_journal *journal = curseg->journal; 4223 int i; 4224 4225 down_write(&curseg->journal_rwsem); 4226 for (i = 0; i < sits_in_cursum(journal); i++) { 4227 unsigned int segno; 4228 bool dirtied; 4229 4230 segno = le32_to_cpu(segno_in_journal(journal, i)); 4231 dirtied = __mark_sit_entry_dirty(sbi, segno); 4232 4233 if (!dirtied) 4234 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 4235 } 4236 update_sits_in_cursum(journal, -i); 4237 up_write(&curseg->journal_rwsem); 4238 } 4239 4240 /* 4241 * CP calls this function, which flushes SIT entries including sit_journal, 4242 * and moves prefree segs to free segs. 4243 */ 4244 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 4245 { 4246 struct sit_info *sit_i = SIT_I(sbi); 4247 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 4248 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4249 struct f2fs_journal *journal = curseg->journal; 4250 struct sit_entry_set *ses, *tmp; 4251 struct list_head *head = &SM_I(sbi)->sit_entry_set; 4252 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 4253 struct seg_entry *se; 4254 4255 down_write(&sit_i->sentry_lock); 4256 4257 if (!sit_i->dirty_sentries) 4258 goto out; 4259 4260 /* 4261 * add and account sit entries of dirty bitmap in sit entry 4262 * set temporarily 4263 */ 4264 add_sits_in_set(sbi); 4265 4266 /* 4267 * if there are no enough space in journal to store dirty sit 4268 * entries, remove all entries from journal and add and account 4269 * them in sit entry set. 4270 */ 4271 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 4272 !to_journal) 4273 remove_sits_in_journal(sbi); 4274 4275 /* 4276 * there are two steps to flush sit entries: 4277 * #1, flush sit entries to journal in current cold data summary block. 4278 * #2, flush sit entries to sit page. 4279 */ 4280 list_for_each_entry_safe(ses, tmp, head, set_list) { 4281 struct page *page = NULL; 4282 struct f2fs_sit_block *raw_sit = NULL; 4283 unsigned int start_segno = ses->start_segno; 4284 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 4285 (unsigned long)MAIN_SEGS(sbi)); 4286 unsigned int segno = start_segno; 4287 4288 if (to_journal && 4289 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 4290 to_journal = false; 4291 4292 if (to_journal) { 4293 down_write(&curseg->journal_rwsem); 4294 } else { 4295 page = get_next_sit_page(sbi, start_segno); 4296 raw_sit = page_address(page); 4297 } 4298 4299 /* flush dirty sit entries in region of current sit set */ 4300 for_each_set_bit_from(segno, bitmap, end) { 4301 int offset, sit_offset; 4302 4303 se = get_seg_entry(sbi, segno); 4304 #ifdef CONFIG_F2FS_CHECK_FS 4305 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 4306 SIT_VBLOCK_MAP_SIZE)) 4307 f2fs_bug_on(sbi, 1); 4308 #endif 4309 4310 /* add discard candidates */ 4311 if (!(cpc->reason & CP_DISCARD)) { 4312 cpc->trim_start = segno; 4313 add_discard_addrs(sbi, cpc, false); 4314 } 4315 4316 if (to_journal) { 4317 offset = f2fs_lookup_journal_in_cursum(journal, 4318 SIT_JOURNAL, segno, 1); 4319 f2fs_bug_on(sbi, offset < 0); 4320 segno_in_journal(journal, offset) = 4321 cpu_to_le32(segno); 4322 seg_info_to_raw_sit(se, 4323 &sit_in_journal(journal, offset)); 4324 check_block_count(sbi, segno, 4325 &sit_in_journal(journal, offset)); 4326 } else { 4327 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 4328 seg_info_to_raw_sit(se, 4329 &raw_sit->entries[sit_offset]); 4330 check_block_count(sbi, segno, 4331 &raw_sit->entries[sit_offset]); 4332 } 4333 4334 __clear_bit(segno, bitmap); 4335 sit_i->dirty_sentries--; 4336 ses->entry_cnt--; 4337 } 4338 4339 if (to_journal) 4340 up_write(&curseg->journal_rwsem); 4341 else 4342 f2fs_put_page(page, 1); 4343 4344 f2fs_bug_on(sbi, ses->entry_cnt); 4345 release_sit_entry_set(ses); 4346 } 4347 4348 f2fs_bug_on(sbi, !list_empty(head)); 4349 f2fs_bug_on(sbi, sit_i->dirty_sentries); 4350 out: 4351 if (cpc->reason & CP_DISCARD) { 4352 __u64 trim_start = cpc->trim_start; 4353 4354 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 4355 add_discard_addrs(sbi, cpc, false); 4356 4357 cpc->trim_start = trim_start; 4358 } 4359 up_write(&sit_i->sentry_lock); 4360 4361 set_prefree_as_free_segments(sbi); 4362 } 4363 4364 static int build_sit_info(struct f2fs_sb_info *sbi) 4365 { 4366 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4367 struct sit_info *sit_i; 4368 unsigned int sit_segs, start; 4369 char *src_bitmap, *bitmap; 4370 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 4371 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0; 4372 4373 /* allocate memory for SIT information */ 4374 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4375 if (!sit_i) 4376 return -ENOMEM; 4377 4378 SM_I(sbi)->sit_info = sit_i; 4379 4380 sit_i->sentries = 4381 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4382 MAIN_SEGS(sbi)), 4383 GFP_KERNEL); 4384 if (!sit_i->sentries) 4385 return -ENOMEM; 4386 4387 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4388 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4389 GFP_KERNEL); 4390 if (!sit_i->dirty_sentries_bitmap) 4391 return -ENOMEM; 4392 4393 #ifdef CONFIG_F2FS_CHECK_FS 4394 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map); 4395 #else 4396 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map); 4397 #endif 4398 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4399 if (!sit_i->bitmap) 4400 return -ENOMEM; 4401 4402 bitmap = sit_i->bitmap; 4403 4404 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4405 sit_i->sentries[start].cur_valid_map = bitmap; 4406 bitmap += SIT_VBLOCK_MAP_SIZE; 4407 4408 sit_i->sentries[start].ckpt_valid_map = bitmap; 4409 bitmap += SIT_VBLOCK_MAP_SIZE; 4410 4411 #ifdef CONFIG_F2FS_CHECK_FS 4412 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4413 bitmap += SIT_VBLOCK_MAP_SIZE; 4414 #endif 4415 4416 if (discard_map) { 4417 sit_i->sentries[start].discard_map = bitmap; 4418 bitmap += SIT_VBLOCK_MAP_SIZE; 4419 } 4420 } 4421 4422 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4423 if (!sit_i->tmp_map) 4424 return -ENOMEM; 4425 4426 if (__is_large_section(sbi)) { 4427 sit_i->sec_entries = 4428 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4429 MAIN_SECS(sbi)), 4430 GFP_KERNEL); 4431 if (!sit_i->sec_entries) 4432 return -ENOMEM; 4433 } 4434 4435 /* get information related with SIT */ 4436 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4437 4438 /* setup SIT bitmap from ckeckpoint pack */ 4439 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4440 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4441 4442 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4443 if (!sit_i->sit_bitmap) 4444 return -ENOMEM; 4445 4446 #ifdef CONFIG_F2FS_CHECK_FS 4447 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4448 sit_bitmap_size, GFP_KERNEL); 4449 if (!sit_i->sit_bitmap_mir) 4450 return -ENOMEM; 4451 4452 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4453 main_bitmap_size, GFP_KERNEL); 4454 if (!sit_i->invalid_segmap) 4455 return -ENOMEM; 4456 #endif 4457 4458 /* init SIT information */ 4459 sit_i->s_ops = &default_salloc_ops; 4460 4461 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4462 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 4463 sit_i->written_valid_blocks = 0; 4464 sit_i->bitmap_size = sit_bitmap_size; 4465 sit_i->dirty_sentries = 0; 4466 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4467 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4468 sit_i->mounted_time = ktime_get_boottime_seconds(); 4469 init_rwsem(&sit_i->sentry_lock); 4470 return 0; 4471 } 4472 4473 static int build_free_segmap(struct f2fs_sb_info *sbi) 4474 { 4475 struct free_segmap_info *free_i; 4476 unsigned int bitmap_size, sec_bitmap_size; 4477 4478 /* allocate memory for free segmap information */ 4479 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4480 if (!free_i) 4481 return -ENOMEM; 4482 4483 SM_I(sbi)->free_info = free_i; 4484 4485 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4486 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4487 if (!free_i->free_segmap) 4488 return -ENOMEM; 4489 4490 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4491 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4492 if (!free_i->free_secmap) 4493 return -ENOMEM; 4494 4495 /* set all segments as dirty temporarily */ 4496 memset(free_i->free_segmap, 0xff, bitmap_size); 4497 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4498 4499 /* init free segmap information */ 4500 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4501 free_i->free_segments = 0; 4502 free_i->free_sections = 0; 4503 spin_lock_init(&free_i->segmap_lock); 4504 return 0; 4505 } 4506 4507 static int build_curseg(struct f2fs_sb_info *sbi) 4508 { 4509 struct curseg_info *array; 4510 int i; 4511 4512 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, 4513 sizeof(*array)), GFP_KERNEL); 4514 if (!array) 4515 return -ENOMEM; 4516 4517 SM_I(sbi)->curseg_array = array; 4518 4519 for (i = 0; i < NO_CHECK_TYPE; i++) { 4520 mutex_init(&array[i].curseg_mutex); 4521 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4522 if (!array[i].sum_blk) 4523 return -ENOMEM; 4524 init_rwsem(&array[i].journal_rwsem); 4525 array[i].journal = f2fs_kzalloc(sbi, 4526 sizeof(struct f2fs_journal), GFP_KERNEL); 4527 if (!array[i].journal) 4528 return -ENOMEM; 4529 if (i < NR_PERSISTENT_LOG) 4530 array[i].seg_type = CURSEG_HOT_DATA + i; 4531 else if (i == CURSEG_COLD_DATA_PINNED) 4532 array[i].seg_type = CURSEG_COLD_DATA; 4533 else if (i == CURSEG_ALL_DATA_ATGC) 4534 array[i].seg_type = CURSEG_COLD_DATA; 4535 array[i].segno = NULL_SEGNO; 4536 array[i].next_blkoff = 0; 4537 array[i].inited = false; 4538 } 4539 return restore_curseg_summaries(sbi); 4540 } 4541 4542 static int build_sit_entries(struct f2fs_sb_info *sbi) 4543 { 4544 struct sit_info *sit_i = SIT_I(sbi); 4545 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4546 struct f2fs_journal *journal = curseg->journal; 4547 struct seg_entry *se; 4548 struct f2fs_sit_entry sit; 4549 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4550 unsigned int i, start, end; 4551 unsigned int readed, start_blk = 0; 4552 int err = 0; 4553 block_t total_node_blocks = 0; 4554 4555 do { 4556 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS, 4557 META_SIT, true); 4558 4559 start = start_blk * sit_i->sents_per_block; 4560 end = (start_blk + readed) * sit_i->sents_per_block; 4561 4562 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4563 struct f2fs_sit_block *sit_blk; 4564 struct page *page; 4565 4566 se = &sit_i->sentries[start]; 4567 page = get_current_sit_page(sbi, start); 4568 if (IS_ERR(page)) 4569 return PTR_ERR(page); 4570 sit_blk = (struct f2fs_sit_block *)page_address(page); 4571 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4572 f2fs_put_page(page, 1); 4573 4574 err = check_block_count(sbi, start, &sit); 4575 if (err) 4576 return err; 4577 seg_info_from_raw_sit(se, &sit); 4578 if (IS_NODESEG(se->type)) 4579 total_node_blocks += se->valid_blocks; 4580 4581 if (f2fs_block_unit_discard(sbi)) { 4582 /* build discard map only one time */ 4583 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4584 memset(se->discard_map, 0xff, 4585 SIT_VBLOCK_MAP_SIZE); 4586 } else { 4587 memcpy(se->discard_map, 4588 se->cur_valid_map, 4589 SIT_VBLOCK_MAP_SIZE); 4590 sbi->discard_blks += 4591 sbi->blocks_per_seg - 4592 se->valid_blocks; 4593 } 4594 } 4595 4596 if (__is_large_section(sbi)) 4597 get_sec_entry(sbi, start)->valid_blocks += 4598 se->valid_blocks; 4599 } 4600 start_blk += readed; 4601 } while (start_blk < sit_blk_cnt); 4602 4603 down_read(&curseg->journal_rwsem); 4604 for (i = 0; i < sits_in_cursum(journal); i++) { 4605 unsigned int old_valid_blocks; 4606 4607 start = le32_to_cpu(segno_in_journal(journal, i)); 4608 if (start >= MAIN_SEGS(sbi)) { 4609 f2fs_err(sbi, "Wrong journal entry on segno %u", 4610 start); 4611 err = -EFSCORRUPTED; 4612 break; 4613 } 4614 4615 se = &sit_i->sentries[start]; 4616 sit = sit_in_journal(journal, i); 4617 4618 old_valid_blocks = se->valid_blocks; 4619 if (IS_NODESEG(se->type)) 4620 total_node_blocks -= old_valid_blocks; 4621 4622 err = check_block_count(sbi, start, &sit); 4623 if (err) 4624 break; 4625 seg_info_from_raw_sit(se, &sit); 4626 if (IS_NODESEG(se->type)) 4627 total_node_blocks += se->valid_blocks; 4628 4629 if (f2fs_block_unit_discard(sbi)) { 4630 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4631 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 4632 } else { 4633 memcpy(se->discard_map, se->cur_valid_map, 4634 SIT_VBLOCK_MAP_SIZE); 4635 sbi->discard_blks += old_valid_blocks; 4636 sbi->discard_blks -= se->valid_blocks; 4637 } 4638 } 4639 4640 if (__is_large_section(sbi)) { 4641 get_sec_entry(sbi, start)->valid_blocks += 4642 se->valid_blocks; 4643 get_sec_entry(sbi, start)->valid_blocks -= 4644 old_valid_blocks; 4645 } 4646 } 4647 up_read(&curseg->journal_rwsem); 4648 4649 if (!err && total_node_blocks != valid_node_count(sbi)) { 4650 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 4651 total_node_blocks, valid_node_count(sbi)); 4652 err = -EFSCORRUPTED; 4653 } 4654 4655 return err; 4656 } 4657 4658 static void init_free_segmap(struct f2fs_sb_info *sbi) 4659 { 4660 unsigned int start; 4661 int type; 4662 struct seg_entry *sentry; 4663 4664 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4665 if (f2fs_usable_blks_in_seg(sbi, start) == 0) 4666 continue; 4667 sentry = get_seg_entry(sbi, start); 4668 if (!sentry->valid_blocks) 4669 __set_free(sbi, start); 4670 else 4671 SIT_I(sbi)->written_valid_blocks += 4672 sentry->valid_blocks; 4673 } 4674 4675 /* set use the current segments */ 4676 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 4677 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 4678 4679 __set_test_and_inuse(sbi, curseg_t->segno); 4680 } 4681 } 4682 4683 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 4684 { 4685 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4686 struct free_segmap_info *free_i = FREE_I(sbi); 4687 unsigned int segno = 0, offset = 0, secno; 4688 block_t valid_blocks, usable_blks_in_seg; 4689 block_t blks_per_sec = BLKS_PER_SEC(sbi); 4690 4691 while (1) { 4692 /* find dirty segment based on free segmap */ 4693 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 4694 if (segno >= MAIN_SEGS(sbi)) 4695 break; 4696 offset = segno + 1; 4697 valid_blocks = get_valid_blocks(sbi, segno, false); 4698 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 4699 if (valid_blocks == usable_blks_in_seg || !valid_blocks) 4700 continue; 4701 if (valid_blocks > usable_blks_in_seg) { 4702 f2fs_bug_on(sbi, 1); 4703 continue; 4704 } 4705 mutex_lock(&dirty_i->seglist_lock); 4706 __locate_dirty_segment(sbi, segno, DIRTY); 4707 mutex_unlock(&dirty_i->seglist_lock); 4708 } 4709 4710 if (!__is_large_section(sbi)) 4711 return; 4712 4713 mutex_lock(&dirty_i->seglist_lock); 4714 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 4715 valid_blocks = get_valid_blocks(sbi, segno, true); 4716 secno = GET_SEC_FROM_SEG(sbi, segno); 4717 4718 if (!valid_blocks || valid_blocks == blks_per_sec) 4719 continue; 4720 if (IS_CURSEC(sbi, secno)) 4721 continue; 4722 set_bit(secno, dirty_i->dirty_secmap); 4723 } 4724 mutex_unlock(&dirty_i->seglist_lock); 4725 } 4726 4727 static int init_victim_secmap(struct f2fs_sb_info *sbi) 4728 { 4729 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4730 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4731 4732 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4733 if (!dirty_i->victim_secmap) 4734 return -ENOMEM; 4735 return 0; 4736 } 4737 4738 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 4739 { 4740 struct dirty_seglist_info *dirty_i; 4741 unsigned int bitmap_size, i; 4742 4743 /* allocate memory for dirty segments list information */ 4744 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 4745 GFP_KERNEL); 4746 if (!dirty_i) 4747 return -ENOMEM; 4748 4749 SM_I(sbi)->dirty_info = dirty_i; 4750 mutex_init(&dirty_i->seglist_lock); 4751 4752 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4753 4754 for (i = 0; i < NR_DIRTY_TYPE; i++) { 4755 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 4756 GFP_KERNEL); 4757 if (!dirty_i->dirty_segmap[i]) 4758 return -ENOMEM; 4759 } 4760 4761 if (__is_large_section(sbi)) { 4762 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4763 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi, 4764 bitmap_size, GFP_KERNEL); 4765 if (!dirty_i->dirty_secmap) 4766 return -ENOMEM; 4767 } 4768 4769 init_dirty_segmap(sbi); 4770 return init_victim_secmap(sbi); 4771 } 4772 4773 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 4774 { 4775 int i; 4776 4777 /* 4778 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 4779 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 4780 */ 4781 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 4782 struct curseg_info *curseg = CURSEG_I(sbi, i); 4783 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 4784 unsigned int blkofs = curseg->next_blkoff; 4785 4786 if (f2fs_sb_has_readonly(sbi) && 4787 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE) 4788 continue; 4789 4790 sanity_check_seg_type(sbi, curseg->seg_type); 4791 4792 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 4793 goto out; 4794 4795 if (curseg->alloc_type == SSR) 4796 continue; 4797 4798 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) { 4799 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 4800 continue; 4801 out: 4802 f2fs_err(sbi, 4803 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 4804 i, curseg->segno, curseg->alloc_type, 4805 curseg->next_blkoff, blkofs); 4806 return -EFSCORRUPTED; 4807 } 4808 } 4809 return 0; 4810 } 4811 4812 #ifdef CONFIG_BLK_DEV_ZONED 4813 4814 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 4815 struct f2fs_dev_info *fdev, 4816 struct blk_zone *zone) 4817 { 4818 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno; 4819 block_t zone_block, wp_block, last_valid_block; 4820 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4821 int i, s, b, ret; 4822 struct seg_entry *se; 4823 4824 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4825 return 0; 4826 4827 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block); 4828 wp_segno = GET_SEGNO(sbi, wp_block); 4829 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4830 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 4831 zone_segno = GET_SEGNO(sbi, zone_block); 4832 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno); 4833 4834 if (zone_segno >= MAIN_SEGS(sbi)) 4835 return 0; 4836 4837 /* 4838 * Skip check of zones cursegs point to, since 4839 * fix_curseg_write_pointer() checks them. 4840 */ 4841 for (i = 0; i < NO_CHECK_TYPE; i++) 4842 if (zone_secno == GET_SEC_FROM_SEG(sbi, 4843 CURSEG_I(sbi, i)->segno)) 4844 return 0; 4845 4846 /* 4847 * Get last valid block of the zone. 4848 */ 4849 last_valid_block = zone_block - 1; 4850 for (s = sbi->segs_per_sec - 1; s >= 0; s--) { 4851 segno = zone_segno + s; 4852 se = get_seg_entry(sbi, segno); 4853 for (b = sbi->blocks_per_seg - 1; b >= 0; b--) 4854 if (f2fs_test_bit(b, se->cur_valid_map)) { 4855 last_valid_block = START_BLOCK(sbi, segno) + b; 4856 break; 4857 } 4858 if (last_valid_block >= zone_block) 4859 break; 4860 } 4861 4862 /* 4863 * If last valid block is beyond the write pointer, report the 4864 * inconsistency. This inconsistency does not cause write error 4865 * because the zone will not be selected for write operation until 4866 * it get discarded. Just report it. 4867 */ 4868 if (last_valid_block >= wp_block) { 4869 f2fs_notice(sbi, "Valid block beyond write pointer: " 4870 "valid block[0x%x,0x%x] wp[0x%x,0x%x]", 4871 GET_SEGNO(sbi, last_valid_block), 4872 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block), 4873 wp_segno, wp_blkoff); 4874 return 0; 4875 } 4876 4877 /* 4878 * If there is no valid block in the zone and if write pointer is 4879 * not at zone start, reset the write pointer. 4880 */ 4881 if (last_valid_block + 1 == zone_block && zone->wp != zone->start) { 4882 f2fs_notice(sbi, 4883 "Zone without valid block has non-zero write " 4884 "pointer. Reset the write pointer: wp[0x%x,0x%x]", 4885 wp_segno, wp_blkoff); 4886 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 4887 zone->len >> log_sectors_per_block); 4888 if (ret) { 4889 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4890 fdev->path, ret); 4891 return ret; 4892 } 4893 } 4894 4895 return 0; 4896 } 4897 4898 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 4899 block_t zone_blkaddr) 4900 { 4901 int i; 4902 4903 for (i = 0; i < sbi->s_ndevs; i++) { 4904 if (!bdev_is_zoned(FDEV(i).bdev)) 4905 continue; 4906 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 4907 zone_blkaddr <= FDEV(i).end_blk)) 4908 return &FDEV(i); 4909 } 4910 4911 return NULL; 4912 } 4913 4914 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 4915 void *data) 4916 { 4917 memcpy(data, zone, sizeof(struct blk_zone)); 4918 return 0; 4919 } 4920 4921 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 4922 { 4923 struct curseg_info *cs = CURSEG_I(sbi, type); 4924 struct f2fs_dev_info *zbd; 4925 struct blk_zone zone; 4926 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 4927 block_t cs_zone_block, wp_block; 4928 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4929 sector_t zone_sector; 4930 int err; 4931 4932 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4933 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4934 4935 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4936 if (!zbd) 4937 return 0; 4938 4939 /* report zone for the sector the curseg points to */ 4940 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4941 << log_sectors_per_block; 4942 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4943 report_one_zone_cb, &zone); 4944 if (err != 1) { 4945 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4946 zbd->path, err); 4947 return err; 4948 } 4949 4950 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4951 return 0; 4952 4953 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 4954 wp_segno = GET_SEGNO(sbi, wp_block); 4955 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4956 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 4957 4958 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 4959 wp_sector_off == 0) 4960 return 0; 4961 4962 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 4963 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", 4964 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff); 4965 4966 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 4967 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff); 4968 4969 f2fs_allocate_new_section(sbi, type, true); 4970 4971 /* check consistency of the zone curseg pointed to */ 4972 if (check_zone_write_pointer(sbi, zbd, &zone)) 4973 return -EIO; 4974 4975 /* check newly assigned zone */ 4976 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4977 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4978 4979 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4980 if (!zbd) 4981 return 0; 4982 4983 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4984 << log_sectors_per_block; 4985 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4986 report_one_zone_cb, &zone); 4987 if (err != 1) { 4988 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4989 zbd->path, err); 4990 return err; 4991 } 4992 4993 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4994 return 0; 4995 4996 if (zone.wp != zone.start) { 4997 f2fs_notice(sbi, 4998 "New zone for curseg[%d] is not yet discarded. " 4999 "Reset the zone: curseg[0x%x,0x%x]", 5000 type, cs->segno, cs->next_blkoff); 5001 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, 5002 zone_sector >> log_sectors_per_block, 5003 zone.len >> log_sectors_per_block); 5004 if (err) { 5005 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5006 zbd->path, err); 5007 return err; 5008 } 5009 } 5010 5011 return 0; 5012 } 5013 5014 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5015 { 5016 int i, ret; 5017 5018 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5019 ret = fix_curseg_write_pointer(sbi, i); 5020 if (ret) 5021 return ret; 5022 } 5023 5024 return 0; 5025 } 5026 5027 struct check_zone_write_pointer_args { 5028 struct f2fs_sb_info *sbi; 5029 struct f2fs_dev_info *fdev; 5030 }; 5031 5032 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 5033 void *data) 5034 { 5035 struct check_zone_write_pointer_args *args; 5036 5037 args = (struct check_zone_write_pointer_args *)data; 5038 5039 return check_zone_write_pointer(args->sbi, args->fdev, zone); 5040 } 5041 5042 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5043 { 5044 int i, ret; 5045 struct check_zone_write_pointer_args args; 5046 5047 for (i = 0; i < sbi->s_ndevs; i++) { 5048 if (!bdev_is_zoned(FDEV(i).bdev)) 5049 continue; 5050 5051 args.sbi = sbi; 5052 args.fdev = &FDEV(i); 5053 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 5054 check_zone_write_pointer_cb, &args); 5055 if (ret < 0) 5056 return ret; 5057 } 5058 5059 return 0; 5060 } 5061 5062 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx, 5063 unsigned int dev_idx) 5064 { 5065 if (!bdev_is_zoned(FDEV(dev_idx).bdev)) 5066 return true; 5067 return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq); 5068 } 5069 5070 /* Return the zone index in the given device */ 5071 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno, 5072 int dev_idx) 5073 { 5074 block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5075 5076 return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >> 5077 sbi->log_blocks_per_blkz; 5078 } 5079 5080 /* 5081 * Return the usable segments in a section based on the zone's 5082 * corresponding zone capacity. Zone is equal to a section. 5083 */ 5084 static inline unsigned int f2fs_usable_zone_segs_in_sec( 5085 struct f2fs_sb_info *sbi, unsigned int segno) 5086 { 5087 unsigned int dev_idx, zone_idx, unusable_segs_in_sec; 5088 5089 dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno)); 5090 zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx); 5091 5092 /* Conventional zone's capacity is always equal to zone size */ 5093 if (is_conv_zone(sbi, zone_idx, dev_idx)) 5094 return sbi->segs_per_sec; 5095 5096 /* 5097 * If the zone_capacity_blocks array is NULL, then zone capacity 5098 * is equal to the zone size for all zones 5099 */ 5100 if (!FDEV(dev_idx).zone_capacity_blocks) 5101 return sbi->segs_per_sec; 5102 5103 /* Get the segment count beyond zone capacity block */ 5104 unusable_segs_in_sec = (sbi->blocks_per_blkz - 5105 FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >> 5106 sbi->log_blocks_per_seg; 5107 return sbi->segs_per_sec - unusable_segs_in_sec; 5108 } 5109 5110 /* 5111 * Return the number of usable blocks in a segment. The number of blocks 5112 * returned is always equal to the number of blocks in a segment for 5113 * segments fully contained within a sequential zone capacity or a 5114 * conventional zone. For segments partially contained in a sequential 5115 * zone capacity, the number of usable blocks up to the zone capacity 5116 * is returned. 0 is returned in all other cases. 5117 */ 5118 static inline unsigned int f2fs_usable_zone_blks_in_seg( 5119 struct f2fs_sb_info *sbi, unsigned int segno) 5120 { 5121 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr; 5122 unsigned int zone_idx, dev_idx, secno; 5123 5124 secno = GET_SEC_FROM_SEG(sbi, segno); 5125 seg_start = START_BLOCK(sbi, segno); 5126 dev_idx = f2fs_target_device_index(sbi, seg_start); 5127 zone_idx = get_zone_idx(sbi, secno, dev_idx); 5128 5129 /* 5130 * Conventional zone's capacity is always equal to zone size, 5131 * so, blocks per segment is unchanged. 5132 */ 5133 if (is_conv_zone(sbi, zone_idx, dev_idx)) 5134 return sbi->blocks_per_seg; 5135 5136 if (!FDEV(dev_idx).zone_capacity_blocks) 5137 return sbi->blocks_per_seg; 5138 5139 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5140 sec_cap_blkaddr = sec_start_blkaddr + 5141 FDEV(dev_idx).zone_capacity_blocks[zone_idx]; 5142 5143 /* 5144 * If segment starts before zone capacity and spans beyond 5145 * zone capacity, then usable blocks are from seg start to 5146 * zone capacity. If the segment starts after the zone capacity, 5147 * then there are no usable blocks. 5148 */ 5149 if (seg_start >= sec_cap_blkaddr) 5150 return 0; 5151 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr) 5152 return sec_cap_blkaddr - seg_start; 5153 5154 return sbi->blocks_per_seg; 5155 } 5156 #else 5157 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5158 { 5159 return 0; 5160 } 5161 5162 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5163 { 5164 return 0; 5165 } 5166 5167 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi, 5168 unsigned int segno) 5169 { 5170 return 0; 5171 } 5172 5173 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi, 5174 unsigned int segno) 5175 { 5176 return 0; 5177 } 5178 #endif 5179 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi, 5180 unsigned int segno) 5181 { 5182 if (f2fs_sb_has_blkzoned(sbi)) 5183 return f2fs_usable_zone_blks_in_seg(sbi, segno); 5184 5185 return sbi->blocks_per_seg; 5186 } 5187 5188 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi, 5189 unsigned int segno) 5190 { 5191 if (f2fs_sb_has_blkzoned(sbi)) 5192 return f2fs_usable_zone_segs_in_sec(sbi, segno); 5193 5194 return sbi->segs_per_sec; 5195 } 5196 5197 /* 5198 * Update min, max modified time for cost-benefit GC algorithm 5199 */ 5200 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 5201 { 5202 struct sit_info *sit_i = SIT_I(sbi); 5203 unsigned int segno; 5204 5205 down_write(&sit_i->sentry_lock); 5206 5207 sit_i->min_mtime = ULLONG_MAX; 5208 5209 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 5210 unsigned int i; 5211 unsigned long long mtime = 0; 5212 5213 for (i = 0; i < sbi->segs_per_sec; i++) 5214 mtime += get_seg_entry(sbi, segno + i)->mtime; 5215 5216 mtime = div_u64(mtime, sbi->segs_per_sec); 5217 5218 if (sit_i->min_mtime > mtime) 5219 sit_i->min_mtime = mtime; 5220 } 5221 sit_i->max_mtime = get_mtime(sbi, false); 5222 sit_i->dirty_max_mtime = 0; 5223 up_write(&sit_i->sentry_lock); 5224 } 5225 5226 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 5227 { 5228 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 5229 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 5230 struct f2fs_sm_info *sm_info; 5231 int err; 5232 5233 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 5234 if (!sm_info) 5235 return -ENOMEM; 5236 5237 /* init sm info */ 5238 sbi->sm_info = sm_info; 5239 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 5240 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 5241 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 5242 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 5243 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 5244 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 5245 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 5246 sm_info->rec_prefree_segments = sm_info->main_segments * 5247 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 5248 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 5249 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 5250 5251 if (!f2fs_lfs_mode(sbi)) 5252 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 5253 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 5254 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 5255 sm_info->min_seq_blocks = sbi->blocks_per_seg; 5256 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 5257 sm_info->min_ssr_sections = reserved_sections(sbi); 5258 5259 INIT_LIST_HEAD(&sm_info->sit_entry_set); 5260 5261 init_rwsem(&sm_info->curseg_lock); 5262 5263 if (!f2fs_readonly(sbi->sb)) { 5264 err = f2fs_create_flush_cmd_control(sbi); 5265 if (err) 5266 return err; 5267 } 5268 5269 err = create_discard_cmd_control(sbi); 5270 if (err) 5271 return err; 5272 5273 err = build_sit_info(sbi); 5274 if (err) 5275 return err; 5276 err = build_free_segmap(sbi); 5277 if (err) 5278 return err; 5279 err = build_curseg(sbi); 5280 if (err) 5281 return err; 5282 5283 /* reinit free segmap based on SIT */ 5284 err = build_sit_entries(sbi); 5285 if (err) 5286 return err; 5287 5288 init_free_segmap(sbi); 5289 err = build_dirty_segmap(sbi); 5290 if (err) 5291 return err; 5292 5293 err = sanity_check_curseg(sbi); 5294 if (err) 5295 return err; 5296 5297 init_min_max_mtime(sbi); 5298 return 0; 5299 } 5300 5301 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 5302 enum dirty_type dirty_type) 5303 { 5304 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5305 5306 mutex_lock(&dirty_i->seglist_lock); 5307 kvfree(dirty_i->dirty_segmap[dirty_type]); 5308 dirty_i->nr_dirty[dirty_type] = 0; 5309 mutex_unlock(&dirty_i->seglist_lock); 5310 } 5311 5312 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 5313 { 5314 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5315 5316 kvfree(dirty_i->victim_secmap); 5317 } 5318 5319 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 5320 { 5321 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5322 int i; 5323 5324 if (!dirty_i) 5325 return; 5326 5327 /* discard pre-free/dirty segments list */ 5328 for (i = 0; i < NR_DIRTY_TYPE; i++) 5329 discard_dirty_segmap(sbi, i); 5330 5331 if (__is_large_section(sbi)) { 5332 mutex_lock(&dirty_i->seglist_lock); 5333 kvfree(dirty_i->dirty_secmap); 5334 mutex_unlock(&dirty_i->seglist_lock); 5335 } 5336 5337 destroy_victim_secmap(sbi); 5338 SM_I(sbi)->dirty_info = NULL; 5339 kfree(dirty_i); 5340 } 5341 5342 static void destroy_curseg(struct f2fs_sb_info *sbi) 5343 { 5344 struct curseg_info *array = SM_I(sbi)->curseg_array; 5345 int i; 5346 5347 if (!array) 5348 return; 5349 SM_I(sbi)->curseg_array = NULL; 5350 for (i = 0; i < NR_CURSEG_TYPE; i++) { 5351 kfree(array[i].sum_blk); 5352 kfree(array[i].journal); 5353 } 5354 kfree(array); 5355 } 5356 5357 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 5358 { 5359 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 5360 5361 if (!free_i) 5362 return; 5363 SM_I(sbi)->free_info = NULL; 5364 kvfree(free_i->free_segmap); 5365 kvfree(free_i->free_secmap); 5366 kfree(free_i); 5367 } 5368 5369 static void destroy_sit_info(struct f2fs_sb_info *sbi) 5370 { 5371 struct sit_info *sit_i = SIT_I(sbi); 5372 5373 if (!sit_i) 5374 return; 5375 5376 if (sit_i->sentries) 5377 kvfree(sit_i->bitmap); 5378 kfree(sit_i->tmp_map); 5379 5380 kvfree(sit_i->sentries); 5381 kvfree(sit_i->sec_entries); 5382 kvfree(sit_i->dirty_sentries_bitmap); 5383 5384 SM_I(sbi)->sit_info = NULL; 5385 kvfree(sit_i->sit_bitmap); 5386 #ifdef CONFIG_F2FS_CHECK_FS 5387 kvfree(sit_i->sit_bitmap_mir); 5388 kvfree(sit_i->invalid_segmap); 5389 #endif 5390 kfree(sit_i); 5391 } 5392 5393 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 5394 { 5395 struct f2fs_sm_info *sm_info = SM_I(sbi); 5396 5397 if (!sm_info) 5398 return; 5399 f2fs_destroy_flush_cmd_control(sbi, true); 5400 destroy_discard_cmd_control(sbi); 5401 destroy_dirty_segmap(sbi); 5402 destroy_curseg(sbi); 5403 destroy_free_segmap(sbi); 5404 destroy_sit_info(sbi); 5405 sbi->sm_info = NULL; 5406 kfree(sm_info); 5407 } 5408 5409 int __init f2fs_create_segment_manager_caches(void) 5410 { 5411 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 5412 sizeof(struct discard_entry)); 5413 if (!discard_entry_slab) 5414 goto fail; 5415 5416 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 5417 sizeof(struct discard_cmd)); 5418 if (!discard_cmd_slab) 5419 goto destroy_discard_entry; 5420 5421 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 5422 sizeof(struct sit_entry_set)); 5423 if (!sit_entry_set_slab) 5424 goto destroy_discard_cmd; 5425 5426 inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry", 5427 sizeof(struct inmem_pages)); 5428 if (!inmem_entry_slab) 5429 goto destroy_sit_entry_set; 5430 return 0; 5431 5432 destroy_sit_entry_set: 5433 kmem_cache_destroy(sit_entry_set_slab); 5434 destroy_discard_cmd: 5435 kmem_cache_destroy(discard_cmd_slab); 5436 destroy_discard_entry: 5437 kmem_cache_destroy(discard_entry_slab); 5438 fail: 5439 return -ENOMEM; 5440 } 5441 5442 void f2fs_destroy_segment_manager_caches(void) 5443 { 5444 kmem_cache_destroy(sit_entry_set_slab); 5445 kmem_cache_destroy(discard_cmd_slab); 5446 kmem_cache_destroy(discard_entry_slab); 5447 kmem_cache_destroy(inmem_entry_slab); 5448 } 5449