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 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 317 if (gc_failure) { 318 if (++looped >= count) 319 return; 320 } 321 goto next; 322 } 323 324 void f2fs_drop_inmem_pages(struct inode *inode) 325 { 326 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 327 struct f2fs_inode_info *fi = F2FS_I(inode); 328 329 do { 330 mutex_lock(&fi->inmem_lock); 331 if (list_empty(&fi->inmem_pages)) { 332 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0; 333 334 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 335 if (!list_empty(&fi->inmem_ilist)) 336 list_del_init(&fi->inmem_ilist); 337 if (f2fs_is_atomic_file(inode)) { 338 clear_inode_flag(inode, FI_ATOMIC_FILE); 339 sbi->atomic_files--; 340 } 341 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 342 343 mutex_unlock(&fi->inmem_lock); 344 break; 345 } 346 __revoke_inmem_pages(inode, &fi->inmem_pages, 347 true, false, true); 348 mutex_unlock(&fi->inmem_lock); 349 } while (1); 350 } 351 352 void f2fs_drop_inmem_page(struct inode *inode, struct page *page) 353 { 354 struct f2fs_inode_info *fi = F2FS_I(inode); 355 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 356 struct list_head *head = &fi->inmem_pages; 357 struct inmem_pages *cur = NULL; 358 359 f2fs_bug_on(sbi, !page_private_atomic(page)); 360 361 mutex_lock(&fi->inmem_lock); 362 list_for_each_entry(cur, head, list) { 363 if (cur->page == page) 364 break; 365 } 366 367 f2fs_bug_on(sbi, list_empty(head) || cur->page != page); 368 list_del(&cur->list); 369 mutex_unlock(&fi->inmem_lock); 370 371 dec_page_count(sbi, F2FS_INMEM_PAGES); 372 kmem_cache_free(inmem_entry_slab, cur); 373 374 ClearPageUptodate(page); 375 clear_page_private_atomic(page); 376 f2fs_put_page(page, 0); 377 378 detach_page_private(page); 379 set_page_private(page, 0); 380 381 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE); 382 } 383 384 static int __f2fs_commit_inmem_pages(struct inode *inode) 385 { 386 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 387 struct f2fs_inode_info *fi = F2FS_I(inode); 388 struct inmem_pages *cur, *tmp; 389 struct f2fs_io_info fio = { 390 .sbi = sbi, 391 .ino = inode->i_ino, 392 .type = DATA, 393 .op = REQ_OP_WRITE, 394 .op_flags = REQ_SYNC | REQ_PRIO, 395 .io_type = FS_DATA_IO, 396 }; 397 struct list_head revoke_list; 398 bool submit_bio = false; 399 int err = 0; 400 401 INIT_LIST_HEAD(&revoke_list); 402 403 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { 404 struct page *page = cur->page; 405 406 lock_page(page); 407 if (page->mapping == inode->i_mapping) { 408 trace_f2fs_commit_inmem_page(page, INMEM); 409 410 f2fs_wait_on_page_writeback(page, DATA, true, true); 411 412 set_page_dirty(page); 413 if (clear_page_dirty_for_io(page)) { 414 inode_dec_dirty_pages(inode); 415 f2fs_remove_dirty_inode(inode); 416 } 417 retry: 418 fio.page = page; 419 fio.old_blkaddr = NULL_ADDR; 420 fio.encrypted_page = NULL; 421 fio.need_lock = LOCK_DONE; 422 err = f2fs_do_write_data_page(&fio); 423 if (err) { 424 if (err == -ENOMEM) { 425 memalloc_retry_wait(GFP_NOFS); 426 goto retry; 427 } 428 unlock_page(page); 429 break; 430 } 431 /* record old blkaddr for revoking */ 432 cur->old_addr = fio.old_blkaddr; 433 submit_bio = true; 434 } 435 unlock_page(page); 436 list_move_tail(&cur->list, &revoke_list); 437 } 438 439 if (submit_bio) 440 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA); 441 442 if (err) { 443 /* 444 * try to revoke all committed pages, but still we could fail 445 * due to no memory or other reason, if that happened, EAGAIN 446 * will be returned, which means in such case, transaction is 447 * already not integrity, caller should use journal to do the 448 * recovery or rewrite & commit last transaction. For other 449 * error number, revoking was done by filesystem itself. 450 */ 451 err = __revoke_inmem_pages(inode, &revoke_list, 452 false, true, false); 453 454 /* drop all uncommitted pages */ 455 __revoke_inmem_pages(inode, &fi->inmem_pages, 456 true, false, false); 457 } else { 458 __revoke_inmem_pages(inode, &revoke_list, 459 false, false, false); 460 } 461 462 return err; 463 } 464 465 int f2fs_commit_inmem_pages(struct inode *inode) 466 { 467 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 468 struct f2fs_inode_info *fi = F2FS_I(inode); 469 int err; 470 471 f2fs_balance_fs(sbi, true); 472 473 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 474 475 f2fs_lock_op(sbi); 476 set_inode_flag(inode, FI_ATOMIC_COMMIT); 477 478 mutex_lock(&fi->inmem_lock); 479 err = __f2fs_commit_inmem_pages(inode); 480 mutex_unlock(&fi->inmem_lock); 481 482 clear_inode_flag(inode, FI_ATOMIC_COMMIT); 483 484 f2fs_unlock_op(sbi); 485 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 486 487 return err; 488 } 489 490 /* 491 * This function balances dirty node and dentry pages. 492 * In addition, it controls garbage collection. 493 */ 494 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 495 { 496 if (time_to_inject(sbi, FAULT_CHECKPOINT)) { 497 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT); 498 f2fs_stop_checkpoint(sbi, false); 499 } 500 501 /* balance_fs_bg is able to be pending */ 502 if (need && excess_cached_nats(sbi)) 503 f2fs_balance_fs_bg(sbi, false); 504 505 if (!f2fs_is_checkpoint_ready(sbi)) 506 return; 507 508 /* 509 * We should do GC or end up with checkpoint, if there are so many dirty 510 * dir/node pages without enough free segments. 511 */ 512 if (has_not_enough_free_secs(sbi, 0, 0)) { 513 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread && 514 sbi->gc_thread->f2fs_gc_task) { 515 DEFINE_WAIT(wait); 516 517 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait, 518 TASK_UNINTERRUPTIBLE); 519 wake_up(&sbi->gc_thread->gc_wait_queue_head); 520 io_schedule(); 521 finish_wait(&sbi->gc_thread->fggc_wq, &wait); 522 } else { 523 f2fs_down_write(&sbi->gc_lock); 524 f2fs_gc(sbi, false, false, false, NULL_SEGNO); 525 } 526 } 527 } 528 529 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi) 530 { 531 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2; 532 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS); 533 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA); 534 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES); 535 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META); 536 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA); 537 unsigned int threshold = sbi->blocks_per_seg * factor * 538 DEFAULT_DIRTY_THRESHOLD; 539 unsigned int global_threshold = threshold * 3 / 2; 540 541 if (dents >= threshold || qdata >= threshold || 542 nodes >= threshold || meta >= threshold || 543 imeta >= threshold) 544 return true; 545 return dents + qdata + nodes + meta + imeta > global_threshold; 546 } 547 548 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 549 { 550 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 551 return; 552 553 /* try to shrink extent cache when there is no enough memory */ 554 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE)) 555 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); 556 557 /* check the # of cached NAT entries */ 558 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 559 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 560 561 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 562 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 563 else 564 f2fs_build_free_nids(sbi, false, false); 565 566 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) || 567 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi)) 568 goto do_sync; 569 570 /* there is background inflight IO or foreground operation recently */ 571 if (is_inflight_io(sbi, REQ_TIME) || 572 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem))) 573 return; 574 575 /* exceed periodical checkpoint timeout threshold */ 576 if (f2fs_time_over(sbi, CP_TIME)) 577 goto do_sync; 578 579 /* checkpoint is the only way to shrink partial cached entries */ 580 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) && 581 f2fs_available_free_memory(sbi, INO_ENTRIES)) 582 return; 583 584 do_sync: 585 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 586 struct blk_plug plug; 587 588 mutex_lock(&sbi->flush_lock); 589 590 blk_start_plug(&plug); 591 f2fs_sync_dirty_inodes(sbi, FILE_INODE); 592 blk_finish_plug(&plug); 593 594 mutex_unlock(&sbi->flush_lock); 595 } 596 f2fs_sync_fs(sbi->sb, true); 597 stat_inc_bg_cp_count(sbi->stat_info); 598 } 599 600 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 601 struct block_device *bdev) 602 { 603 int ret = blkdev_issue_flush(bdev); 604 605 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 606 test_opt(sbi, FLUSH_MERGE), ret); 607 return ret; 608 } 609 610 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 611 { 612 int ret = 0; 613 int i; 614 615 if (!f2fs_is_multi_device(sbi)) 616 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 617 618 for (i = 0; i < sbi->s_ndevs; i++) { 619 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 620 continue; 621 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 622 if (ret) 623 break; 624 } 625 return ret; 626 } 627 628 static int issue_flush_thread(void *data) 629 { 630 struct f2fs_sb_info *sbi = data; 631 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 632 wait_queue_head_t *q = &fcc->flush_wait_queue; 633 repeat: 634 if (kthread_should_stop()) 635 return 0; 636 637 if (!llist_empty(&fcc->issue_list)) { 638 struct flush_cmd *cmd, *next; 639 int ret; 640 641 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 642 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 643 644 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 645 646 ret = submit_flush_wait(sbi, cmd->ino); 647 atomic_inc(&fcc->issued_flush); 648 649 llist_for_each_entry_safe(cmd, next, 650 fcc->dispatch_list, llnode) { 651 cmd->ret = ret; 652 complete(&cmd->wait); 653 } 654 fcc->dispatch_list = NULL; 655 } 656 657 wait_event_interruptible(*q, 658 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 659 goto repeat; 660 } 661 662 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 663 { 664 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 665 struct flush_cmd cmd; 666 int ret; 667 668 if (test_opt(sbi, NOBARRIER)) 669 return 0; 670 671 if (!test_opt(sbi, FLUSH_MERGE)) { 672 atomic_inc(&fcc->queued_flush); 673 ret = submit_flush_wait(sbi, ino); 674 atomic_dec(&fcc->queued_flush); 675 atomic_inc(&fcc->issued_flush); 676 return ret; 677 } 678 679 if (atomic_inc_return(&fcc->queued_flush) == 1 || 680 f2fs_is_multi_device(sbi)) { 681 ret = submit_flush_wait(sbi, ino); 682 atomic_dec(&fcc->queued_flush); 683 684 atomic_inc(&fcc->issued_flush); 685 return ret; 686 } 687 688 cmd.ino = ino; 689 init_completion(&cmd.wait); 690 691 llist_add(&cmd.llnode, &fcc->issue_list); 692 693 /* 694 * update issue_list before we wake up issue_flush thread, this 695 * smp_mb() pairs with another barrier in ___wait_event(), see 696 * more details in comments of waitqueue_active(). 697 */ 698 smp_mb(); 699 700 if (waitqueue_active(&fcc->flush_wait_queue)) 701 wake_up(&fcc->flush_wait_queue); 702 703 if (fcc->f2fs_issue_flush) { 704 wait_for_completion(&cmd.wait); 705 atomic_dec(&fcc->queued_flush); 706 } else { 707 struct llist_node *list; 708 709 list = llist_del_all(&fcc->issue_list); 710 if (!list) { 711 wait_for_completion(&cmd.wait); 712 atomic_dec(&fcc->queued_flush); 713 } else { 714 struct flush_cmd *tmp, *next; 715 716 ret = submit_flush_wait(sbi, ino); 717 718 llist_for_each_entry_safe(tmp, next, list, llnode) { 719 if (tmp == &cmd) { 720 cmd.ret = ret; 721 atomic_dec(&fcc->queued_flush); 722 continue; 723 } 724 tmp->ret = ret; 725 complete(&tmp->wait); 726 } 727 } 728 } 729 730 return cmd.ret; 731 } 732 733 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 734 { 735 dev_t dev = sbi->sb->s_bdev->bd_dev; 736 struct flush_cmd_control *fcc; 737 int err = 0; 738 739 if (SM_I(sbi)->fcc_info) { 740 fcc = SM_I(sbi)->fcc_info; 741 if (fcc->f2fs_issue_flush) 742 return err; 743 goto init_thread; 744 } 745 746 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 747 if (!fcc) 748 return -ENOMEM; 749 atomic_set(&fcc->issued_flush, 0); 750 atomic_set(&fcc->queued_flush, 0); 751 init_waitqueue_head(&fcc->flush_wait_queue); 752 init_llist_head(&fcc->issue_list); 753 SM_I(sbi)->fcc_info = fcc; 754 if (!test_opt(sbi, FLUSH_MERGE)) 755 return err; 756 757 init_thread: 758 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 759 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 760 if (IS_ERR(fcc->f2fs_issue_flush)) { 761 err = PTR_ERR(fcc->f2fs_issue_flush); 762 kfree(fcc); 763 SM_I(sbi)->fcc_info = NULL; 764 return err; 765 } 766 767 return err; 768 } 769 770 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 771 { 772 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 773 774 if (fcc && fcc->f2fs_issue_flush) { 775 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 776 777 fcc->f2fs_issue_flush = NULL; 778 kthread_stop(flush_thread); 779 } 780 if (free) { 781 kfree(fcc); 782 SM_I(sbi)->fcc_info = NULL; 783 } 784 } 785 786 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 787 { 788 int ret = 0, i; 789 790 if (!f2fs_is_multi_device(sbi)) 791 return 0; 792 793 if (test_opt(sbi, NOBARRIER)) 794 return 0; 795 796 for (i = 1; i < sbi->s_ndevs; i++) { 797 int count = DEFAULT_RETRY_IO_COUNT; 798 799 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 800 continue; 801 802 do { 803 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 804 if (ret) 805 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 806 } while (ret && --count); 807 808 if (ret) { 809 f2fs_stop_checkpoint(sbi, false); 810 break; 811 } 812 813 spin_lock(&sbi->dev_lock); 814 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 815 spin_unlock(&sbi->dev_lock); 816 } 817 818 return ret; 819 } 820 821 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 822 enum dirty_type dirty_type) 823 { 824 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 825 826 /* need not be added */ 827 if (IS_CURSEG(sbi, segno)) 828 return; 829 830 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 831 dirty_i->nr_dirty[dirty_type]++; 832 833 if (dirty_type == DIRTY) { 834 struct seg_entry *sentry = get_seg_entry(sbi, segno); 835 enum dirty_type t = sentry->type; 836 837 if (unlikely(t >= DIRTY)) { 838 f2fs_bug_on(sbi, 1); 839 return; 840 } 841 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 842 dirty_i->nr_dirty[t]++; 843 844 if (__is_large_section(sbi)) { 845 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 846 block_t valid_blocks = 847 get_valid_blocks(sbi, segno, true); 848 849 f2fs_bug_on(sbi, unlikely(!valid_blocks || 850 valid_blocks == BLKS_PER_SEC(sbi))); 851 852 if (!IS_CURSEC(sbi, secno)) 853 set_bit(secno, dirty_i->dirty_secmap); 854 } 855 } 856 } 857 858 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 859 enum dirty_type dirty_type) 860 { 861 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 862 block_t valid_blocks; 863 864 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 865 dirty_i->nr_dirty[dirty_type]--; 866 867 if (dirty_type == DIRTY) { 868 struct seg_entry *sentry = get_seg_entry(sbi, segno); 869 enum dirty_type t = sentry->type; 870 871 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 872 dirty_i->nr_dirty[t]--; 873 874 valid_blocks = get_valid_blocks(sbi, segno, true); 875 if (valid_blocks == 0) { 876 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 877 dirty_i->victim_secmap); 878 #ifdef CONFIG_F2FS_CHECK_FS 879 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 880 #endif 881 } 882 if (__is_large_section(sbi)) { 883 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 884 885 if (!valid_blocks || 886 valid_blocks == BLKS_PER_SEC(sbi)) { 887 clear_bit(secno, dirty_i->dirty_secmap); 888 return; 889 } 890 891 if (!IS_CURSEC(sbi, secno)) 892 set_bit(secno, dirty_i->dirty_secmap); 893 } 894 } 895 } 896 897 /* 898 * Should not occur error such as -ENOMEM. 899 * Adding dirty entry into seglist is not critical operation. 900 * If a given segment is one of current working segments, it won't be added. 901 */ 902 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 903 { 904 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 905 unsigned short valid_blocks, ckpt_valid_blocks; 906 unsigned int usable_blocks; 907 908 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 909 return; 910 911 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno); 912 mutex_lock(&dirty_i->seglist_lock); 913 914 valid_blocks = get_valid_blocks(sbi, segno, false); 915 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false); 916 917 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 918 ckpt_valid_blocks == usable_blocks)) { 919 __locate_dirty_segment(sbi, segno, PRE); 920 __remove_dirty_segment(sbi, segno, DIRTY); 921 } else if (valid_blocks < usable_blocks) { 922 __locate_dirty_segment(sbi, segno, DIRTY); 923 } else { 924 /* Recovery routine with SSR needs this */ 925 __remove_dirty_segment(sbi, segno, DIRTY); 926 } 927 928 mutex_unlock(&dirty_i->seglist_lock); 929 } 930 931 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 932 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 933 { 934 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 935 unsigned int segno; 936 937 mutex_lock(&dirty_i->seglist_lock); 938 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 939 if (get_valid_blocks(sbi, segno, false)) 940 continue; 941 if (IS_CURSEG(sbi, segno)) 942 continue; 943 __locate_dirty_segment(sbi, segno, PRE); 944 __remove_dirty_segment(sbi, segno, DIRTY); 945 } 946 mutex_unlock(&dirty_i->seglist_lock); 947 } 948 949 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 950 { 951 int ovp_hole_segs = 952 (overprovision_segments(sbi) - reserved_segments(sbi)); 953 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg; 954 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 955 block_t holes[2] = {0, 0}; /* DATA and NODE */ 956 block_t unusable; 957 struct seg_entry *se; 958 unsigned int segno; 959 960 mutex_lock(&dirty_i->seglist_lock); 961 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 962 se = get_seg_entry(sbi, segno); 963 if (IS_NODESEG(se->type)) 964 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) - 965 se->valid_blocks; 966 else 967 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) - 968 se->valid_blocks; 969 } 970 mutex_unlock(&dirty_i->seglist_lock); 971 972 unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE]; 973 if (unusable > ovp_holes) 974 return unusable - ovp_holes; 975 return 0; 976 } 977 978 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 979 { 980 int ovp_hole_segs = 981 (overprovision_segments(sbi) - reserved_segments(sbi)); 982 if (unusable > F2FS_OPTION(sbi).unusable_cap) 983 return -EAGAIN; 984 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 985 dirty_segments(sbi) > ovp_hole_segs) 986 return -EAGAIN; 987 return 0; 988 } 989 990 /* This is only used by SBI_CP_DISABLED */ 991 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 992 { 993 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 994 unsigned int segno = 0; 995 996 mutex_lock(&dirty_i->seglist_lock); 997 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 998 if (get_valid_blocks(sbi, segno, false)) 999 continue; 1000 if (get_ckpt_valid_blocks(sbi, segno, false)) 1001 continue; 1002 mutex_unlock(&dirty_i->seglist_lock); 1003 return segno; 1004 } 1005 mutex_unlock(&dirty_i->seglist_lock); 1006 return NULL_SEGNO; 1007 } 1008 1009 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 1010 struct block_device *bdev, block_t lstart, 1011 block_t start, block_t len) 1012 { 1013 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1014 struct list_head *pend_list; 1015 struct discard_cmd *dc; 1016 1017 f2fs_bug_on(sbi, !len); 1018 1019 pend_list = &dcc->pend_list[plist_idx(len)]; 1020 1021 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL); 1022 INIT_LIST_HEAD(&dc->list); 1023 dc->bdev = bdev; 1024 dc->lstart = lstart; 1025 dc->start = start; 1026 dc->len = len; 1027 dc->ref = 0; 1028 dc->state = D_PREP; 1029 dc->queued = 0; 1030 dc->error = 0; 1031 init_completion(&dc->wait); 1032 list_add_tail(&dc->list, pend_list); 1033 spin_lock_init(&dc->lock); 1034 dc->bio_ref = 0; 1035 atomic_inc(&dcc->discard_cmd_cnt); 1036 dcc->undiscard_blks += len; 1037 1038 return dc; 1039 } 1040 1041 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi, 1042 struct block_device *bdev, block_t lstart, 1043 block_t start, block_t len, 1044 struct rb_node *parent, struct rb_node **p, 1045 bool leftmost) 1046 { 1047 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1048 struct discard_cmd *dc; 1049 1050 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 1051 1052 rb_link_node(&dc->rb_node, parent, p); 1053 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 1054 1055 return dc; 1056 } 1057 1058 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 1059 struct discard_cmd *dc) 1060 { 1061 if (dc->state == D_DONE) 1062 atomic_sub(dc->queued, &dcc->queued_discard); 1063 1064 list_del(&dc->list); 1065 rb_erase_cached(&dc->rb_node, &dcc->root); 1066 dcc->undiscard_blks -= dc->len; 1067 1068 kmem_cache_free(discard_cmd_slab, dc); 1069 1070 atomic_dec(&dcc->discard_cmd_cnt); 1071 } 1072 1073 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 1074 struct discard_cmd *dc) 1075 { 1076 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1077 unsigned long flags; 1078 1079 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len); 1080 1081 spin_lock_irqsave(&dc->lock, flags); 1082 if (dc->bio_ref) { 1083 spin_unlock_irqrestore(&dc->lock, flags); 1084 return; 1085 } 1086 spin_unlock_irqrestore(&dc->lock, flags); 1087 1088 f2fs_bug_on(sbi, dc->ref); 1089 1090 if (dc->error == -EOPNOTSUPP) 1091 dc->error = 0; 1092 1093 if (dc->error) 1094 printk_ratelimited( 1095 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d", 1096 KERN_INFO, sbi->sb->s_id, 1097 dc->lstart, dc->start, dc->len, dc->error); 1098 __detach_discard_cmd(dcc, dc); 1099 } 1100 1101 static void f2fs_submit_discard_endio(struct bio *bio) 1102 { 1103 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1104 unsigned long flags; 1105 1106 spin_lock_irqsave(&dc->lock, flags); 1107 if (!dc->error) 1108 dc->error = blk_status_to_errno(bio->bi_status); 1109 dc->bio_ref--; 1110 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1111 dc->state = D_DONE; 1112 complete_all(&dc->wait); 1113 } 1114 spin_unlock_irqrestore(&dc->lock, flags); 1115 bio_put(bio); 1116 } 1117 1118 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1119 block_t start, block_t end) 1120 { 1121 #ifdef CONFIG_F2FS_CHECK_FS 1122 struct seg_entry *sentry; 1123 unsigned int segno; 1124 block_t blk = start; 1125 unsigned long offset, size, max_blocks = sbi->blocks_per_seg; 1126 unsigned long *map; 1127 1128 while (blk < end) { 1129 segno = GET_SEGNO(sbi, blk); 1130 sentry = get_seg_entry(sbi, segno); 1131 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1132 1133 if (end < START_BLOCK(sbi, segno + 1)) 1134 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1135 else 1136 size = max_blocks; 1137 map = (unsigned long *)(sentry->cur_valid_map); 1138 offset = __find_rev_next_bit(map, size, offset); 1139 f2fs_bug_on(sbi, offset != size); 1140 blk = START_BLOCK(sbi, segno + 1); 1141 } 1142 #endif 1143 } 1144 1145 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1146 struct discard_policy *dpolicy, 1147 int discard_type, unsigned int granularity) 1148 { 1149 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1150 1151 /* common policy */ 1152 dpolicy->type = discard_type; 1153 dpolicy->sync = true; 1154 dpolicy->ordered = false; 1155 dpolicy->granularity = granularity; 1156 1157 dpolicy->max_requests = dcc->max_discard_request; 1158 dpolicy->io_aware_gran = MAX_PLIST_NUM; 1159 dpolicy->timeout = false; 1160 1161 if (discard_type == DPOLICY_BG) { 1162 dpolicy->min_interval = dcc->min_discard_issue_time; 1163 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1164 dpolicy->max_interval = dcc->max_discard_issue_time; 1165 dpolicy->io_aware = true; 1166 dpolicy->sync = false; 1167 dpolicy->ordered = true; 1168 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) { 1169 dpolicy->granularity = 1; 1170 if (atomic_read(&dcc->discard_cmd_cnt)) 1171 dpolicy->max_interval = 1172 dcc->min_discard_issue_time; 1173 } 1174 } else if (discard_type == DPOLICY_FORCE) { 1175 dpolicy->min_interval = dcc->min_discard_issue_time; 1176 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1177 dpolicy->max_interval = dcc->max_discard_issue_time; 1178 dpolicy->io_aware = false; 1179 } else if (discard_type == DPOLICY_FSTRIM) { 1180 dpolicy->io_aware = false; 1181 } else if (discard_type == DPOLICY_UMOUNT) { 1182 dpolicy->io_aware = false; 1183 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1184 dpolicy->granularity = 1; 1185 dpolicy->timeout = true; 1186 } 1187 } 1188 1189 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1190 struct block_device *bdev, block_t lstart, 1191 block_t start, block_t len); 1192 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1193 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1194 struct discard_policy *dpolicy, 1195 struct discard_cmd *dc, 1196 unsigned int *issued) 1197 { 1198 struct block_device *bdev = dc->bdev; 1199 unsigned int max_discard_blocks = 1200 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1201 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1202 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1203 &(dcc->fstrim_list) : &(dcc->wait_list); 1204 int flag = dpolicy->sync ? REQ_SYNC : 0; 1205 block_t lstart, start, len, total_len; 1206 int err = 0; 1207 1208 if (dc->state != D_PREP) 1209 return 0; 1210 1211 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1212 return 0; 1213 1214 trace_f2fs_issue_discard(bdev, dc->start, dc->len); 1215 1216 lstart = dc->lstart; 1217 start = dc->start; 1218 len = dc->len; 1219 total_len = len; 1220 1221 dc->len = 0; 1222 1223 while (total_len && *issued < dpolicy->max_requests && !err) { 1224 struct bio *bio = NULL; 1225 unsigned long flags; 1226 bool last = true; 1227 1228 if (len > max_discard_blocks) { 1229 len = max_discard_blocks; 1230 last = false; 1231 } 1232 1233 (*issued)++; 1234 if (*issued == dpolicy->max_requests) 1235 last = true; 1236 1237 dc->len += len; 1238 1239 if (time_to_inject(sbi, FAULT_DISCARD)) { 1240 f2fs_show_injection_info(sbi, FAULT_DISCARD); 1241 err = -EIO; 1242 goto submit; 1243 } 1244 err = __blkdev_issue_discard(bdev, 1245 SECTOR_FROM_BLOCK(start), 1246 SECTOR_FROM_BLOCK(len), 1247 GFP_NOFS, &bio); 1248 submit: 1249 if (err) { 1250 spin_lock_irqsave(&dc->lock, flags); 1251 if (dc->state == D_PARTIAL) 1252 dc->state = D_SUBMIT; 1253 spin_unlock_irqrestore(&dc->lock, flags); 1254 1255 break; 1256 } 1257 1258 f2fs_bug_on(sbi, !bio); 1259 1260 /* 1261 * should keep before submission to avoid D_DONE 1262 * right away 1263 */ 1264 spin_lock_irqsave(&dc->lock, flags); 1265 if (last) 1266 dc->state = D_SUBMIT; 1267 else 1268 dc->state = D_PARTIAL; 1269 dc->bio_ref++; 1270 spin_unlock_irqrestore(&dc->lock, flags); 1271 1272 atomic_inc(&dcc->queued_discard); 1273 dc->queued++; 1274 list_move_tail(&dc->list, wait_list); 1275 1276 /* sanity check on discard range */ 1277 __check_sit_bitmap(sbi, lstart, lstart + len); 1278 1279 bio->bi_private = dc; 1280 bio->bi_end_io = f2fs_submit_discard_endio; 1281 bio->bi_opf |= flag; 1282 submit_bio(bio); 1283 1284 atomic_inc(&dcc->issued_discard); 1285 1286 f2fs_update_iostat(sbi, FS_DISCARD, 1); 1287 1288 lstart += len; 1289 start += len; 1290 total_len -= len; 1291 len = total_len; 1292 } 1293 1294 if (!err && len) { 1295 dcc->undiscard_blks -= len; 1296 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1297 } 1298 return err; 1299 } 1300 1301 static void __insert_discard_tree(struct f2fs_sb_info *sbi, 1302 struct block_device *bdev, block_t lstart, 1303 block_t start, block_t len, 1304 struct rb_node **insert_p, 1305 struct rb_node *insert_parent) 1306 { 1307 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1308 struct rb_node **p; 1309 struct rb_node *parent = NULL; 1310 bool leftmost = true; 1311 1312 if (insert_p && insert_parent) { 1313 parent = insert_parent; 1314 p = insert_p; 1315 goto do_insert; 1316 } 1317 1318 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, 1319 lstart, &leftmost); 1320 do_insert: 1321 __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, 1322 p, leftmost); 1323 } 1324 1325 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1326 struct discard_cmd *dc) 1327 { 1328 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]); 1329 } 1330 1331 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1332 struct discard_cmd *dc, block_t blkaddr) 1333 { 1334 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1335 struct discard_info di = dc->di; 1336 bool modified = false; 1337 1338 if (dc->state == D_DONE || dc->len == 1) { 1339 __remove_discard_cmd(sbi, dc); 1340 return; 1341 } 1342 1343 dcc->undiscard_blks -= di.len; 1344 1345 if (blkaddr > di.lstart) { 1346 dc->len = blkaddr - dc->lstart; 1347 dcc->undiscard_blks += dc->len; 1348 __relocate_discard_cmd(dcc, dc); 1349 modified = true; 1350 } 1351 1352 if (blkaddr < di.lstart + di.len - 1) { 1353 if (modified) { 1354 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, 1355 di.start + blkaddr + 1 - di.lstart, 1356 di.lstart + di.len - 1 - blkaddr, 1357 NULL, NULL); 1358 } else { 1359 dc->lstart++; 1360 dc->len--; 1361 dc->start++; 1362 dcc->undiscard_blks += dc->len; 1363 __relocate_discard_cmd(dcc, dc); 1364 } 1365 } 1366 } 1367 1368 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1369 struct block_device *bdev, block_t lstart, 1370 block_t start, block_t len) 1371 { 1372 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1373 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1374 struct discard_cmd *dc; 1375 struct discard_info di = {0}; 1376 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1377 unsigned int max_discard_blocks = 1378 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1379 block_t end = lstart + len; 1380 1381 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1382 NULL, lstart, 1383 (struct rb_entry **)&prev_dc, 1384 (struct rb_entry **)&next_dc, 1385 &insert_p, &insert_parent, true, NULL); 1386 if (dc) 1387 prev_dc = dc; 1388 1389 if (!prev_dc) { 1390 di.lstart = lstart; 1391 di.len = next_dc ? next_dc->lstart - lstart : len; 1392 di.len = min(di.len, len); 1393 di.start = start; 1394 } 1395 1396 while (1) { 1397 struct rb_node *node; 1398 bool merged = false; 1399 struct discard_cmd *tdc = NULL; 1400 1401 if (prev_dc) { 1402 di.lstart = prev_dc->lstart + prev_dc->len; 1403 if (di.lstart < lstart) 1404 di.lstart = lstart; 1405 if (di.lstart >= end) 1406 break; 1407 1408 if (!next_dc || next_dc->lstart > end) 1409 di.len = end - di.lstart; 1410 else 1411 di.len = next_dc->lstart - di.lstart; 1412 di.start = start + di.lstart - lstart; 1413 } 1414 1415 if (!di.len) 1416 goto next; 1417 1418 if (prev_dc && prev_dc->state == D_PREP && 1419 prev_dc->bdev == bdev && 1420 __is_discard_back_mergeable(&di, &prev_dc->di, 1421 max_discard_blocks)) { 1422 prev_dc->di.len += di.len; 1423 dcc->undiscard_blks += di.len; 1424 __relocate_discard_cmd(dcc, prev_dc); 1425 di = prev_dc->di; 1426 tdc = prev_dc; 1427 merged = true; 1428 } 1429 1430 if (next_dc && next_dc->state == D_PREP && 1431 next_dc->bdev == bdev && 1432 __is_discard_front_mergeable(&di, &next_dc->di, 1433 max_discard_blocks)) { 1434 next_dc->di.lstart = di.lstart; 1435 next_dc->di.len += di.len; 1436 next_dc->di.start = di.start; 1437 dcc->undiscard_blks += di.len; 1438 __relocate_discard_cmd(dcc, next_dc); 1439 if (tdc) 1440 __remove_discard_cmd(sbi, tdc); 1441 merged = true; 1442 } 1443 1444 if (!merged) { 1445 __insert_discard_tree(sbi, bdev, di.lstart, di.start, 1446 di.len, NULL, NULL); 1447 } 1448 next: 1449 prev_dc = next_dc; 1450 if (!prev_dc) 1451 break; 1452 1453 node = rb_next(&prev_dc->rb_node); 1454 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1455 } 1456 } 1457 1458 static int __queue_discard_cmd(struct f2fs_sb_info *sbi, 1459 struct block_device *bdev, block_t blkstart, block_t blklen) 1460 { 1461 block_t lblkstart = blkstart; 1462 1463 if (!f2fs_bdev_support_discard(bdev)) 1464 return 0; 1465 1466 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1467 1468 if (f2fs_is_multi_device(sbi)) { 1469 int devi = f2fs_target_device_index(sbi, blkstart); 1470 1471 blkstart -= FDEV(devi).start_blk; 1472 } 1473 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1474 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1475 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1476 return 0; 1477 } 1478 1479 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1480 struct discard_policy *dpolicy) 1481 { 1482 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1483 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1484 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1485 struct discard_cmd *dc; 1486 struct blk_plug plug; 1487 unsigned int pos = dcc->next_pos; 1488 unsigned int issued = 0; 1489 bool io_interrupted = false; 1490 1491 mutex_lock(&dcc->cmd_lock); 1492 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1493 NULL, pos, 1494 (struct rb_entry **)&prev_dc, 1495 (struct rb_entry **)&next_dc, 1496 &insert_p, &insert_parent, true, NULL); 1497 if (!dc) 1498 dc = next_dc; 1499 1500 blk_start_plug(&plug); 1501 1502 while (dc) { 1503 struct rb_node *node; 1504 int err = 0; 1505 1506 if (dc->state != D_PREP) 1507 goto next; 1508 1509 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1510 io_interrupted = true; 1511 break; 1512 } 1513 1514 dcc->next_pos = dc->lstart + dc->len; 1515 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1516 1517 if (issued >= dpolicy->max_requests) 1518 break; 1519 next: 1520 node = rb_next(&dc->rb_node); 1521 if (err) 1522 __remove_discard_cmd(sbi, dc); 1523 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1524 } 1525 1526 blk_finish_plug(&plug); 1527 1528 if (!dc) 1529 dcc->next_pos = 0; 1530 1531 mutex_unlock(&dcc->cmd_lock); 1532 1533 if (!issued && io_interrupted) 1534 issued = -1; 1535 1536 return issued; 1537 } 1538 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1539 struct discard_policy *dpolicy); 1540 1541 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1542 struct discard_policy *dpolicy) 1543 { 1544 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1545 struct list_head *pend_list; 1546 struct discard_cmd *dc, *tmp; 1547 struct blk_plug plug; 1548 int i, issued; 1549 bool io_interrupted = false; 1550 1551 if (dpolicy->timeout) 1552 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1553 1554 retry: 1555 issued = 0; 1556 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1557 if (dpolicy->timeout && 1558 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1559 break; 1560 1561 if (i + 1 < dpolicy->granularity) 1562 break; 1563 1564 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered) 1565 return __issue_discard_cmd_orderly(sbi, dpolicy); 1566 1567 pend_list = &dcc->pend_list[i]; 1568 1569 mutex_lock(&dcc->cmd_lock); 1570 if (list_empty(pend_list)) 1571 goto next; 1572 if (unlikely(dcc->rbtree_check)) 1573 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 1574 &dcc->root, false)); 1575 blk_start_plug(&plug); 1576 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1577 f2fs_bug_on(sbi, dc->state != D_PREP); 1578 1579 if (dpolicy->timeout && 1580 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1581 break; 1582 1583 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1584 !is_idle(sbi, DISCARD_TIME)) { 1585 io_interrupted = true; 1586 break; 1587 } 1588 1589 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1590 1591 if (issued >= dpolicy->max_requests) 1592 break; 1593 } 1594 blk_finish_plug(&plug); 1595 next: 1596 mutex_unlock(&dcc->cmd_lock); 1597 1598 if (issued >= dpolicy->max_requests || io_interrupted) 1599 break; 1600 } 1601 1602 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1603 __wait_all_discard_cmd(sbi, dpolicy); 1604 goto retry; 1605 } 1606 1607 if (!issued && io_interrupted) 1608 issued = -1; 1609 1610 return issued; 1611 } 1612 1613 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1614 { 1615 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1616 struct list_head *pend_list; 1617 struct discard_cmd *dc, *tmp; 1618 int i; 1619 bool dropped = false; 1620 1621 mutex_lock(&dcc->cmd_lock); 1622 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1623 pend_list = &dcc->pend_list[i]; 1624 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1625 f2fs_bug_on(sbi, dc->state != D_PREP); 1626 __remove_discard_cmd(sbi, dc); 1627 dropped = true; 1628 } 1629 } 1630 mutex_unlock(&dcc->cmd_lock); 1631 1632 return dropped; 1633 } 1634 1635 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1636 { 1637 __drop_discard_cmd(sbi); 1638 } 1639 1640 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1641 struct discard_cmd *dc) 1642 { 1643 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1644 unsigned int len = 0; 1645 1646 wait_for_completion_io(&dc->wait); 1647 mutex_lock(&dcc->cmd_lock); 1648 f2fs_bug_on(sbi, dc->state != D_DONE); 1649 dc->ref--; 1650 if (!dc->ref) { 1651 if (!dc->error) 1652 len = dc->len; 1653 __remove_discard_cmd(sbi, dc); 1654 } 1655 mutex_unlock(&dcc->cmd_lock); 1656 1657 return len; 1658 } 1659 1660 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1661 struct discard_policy *dpolicy, 1662 block_t start, block_t end) 1663 { 1664 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1665 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1666 &(dcc->fstrim_list) : &(dcc->wait_list); 1667 struct discard_cmd *dc, *tmp; 1668 bool need_wait; 1669 unsigned int trimmed = 0; 1670 1671 next: 1672 need_wait = false; 1673 1674 mutex_lock(&dcc->cmd_lock); 1675 list_for_each_entry_safe(dc, tmp, wait_list, list) { 1676 if (dc->lstart + dc->len <= start || end <= dc->lstart) 1677 continue; 1678 if (dc->len < dpolicy->granularity) 1679 continue; 1680 if (dc->state == D_DONE && !dc->ref) { 1681 wait_for_completion_io(&dc->wait); 1682 if (!dc->error) 1683 trimmed += dc->len; 1684 __remove_discard_cmd(sbi, dc); 1685 } else { 1686 dc->ref++; 1687 need_wait = true; 1688 break; 1689 } 1690 } 1691 mutex_unlock(&dcc->cmd_lock); 1692 1693 if (need_wait) { 1694 trimmed += __wait_one_discard_bio(sbi, dc); 1695 goto next; 1696 } 1697 1698 return trimmed; 1699 } 1700 1701 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1702 struct discard_policy *dpolicy) 1703 { 1704 struct discard_policy dp; 1705 unsigned int discard_blks; 1706 1707 if (dpolicy) 1708 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1709 1710 /* wait all */ 1711 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1); 1712 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1713 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1); 1714 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1715 1716 return discard_blks; 1717 } 1718 1719 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1720 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1721 { 1722 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1723 struct discard_cmd *dc; 1724 bool need_wait = false; 1725 1726 mutex_lock(&dcc->cmd_lock); 1727 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root, 1728 NULL, blkaddr); 1729 if (dc) { 1730 if (dc->state == D_PREP) { 1731 __punch_discard_cmd(sbi, dc, blkaddr); 1732 } else { 1733 dc->ref++; 1734 need_wait = true; 1735 } 1736 } 1737 mutex_unlock(&dcc->cmd_lock); 1738 1739 if (need_wait) 1740 __wait_one_discard_bio(sbi, dc); 1741 } 1742 1743 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1744 { 1745 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1746 1747 if (dcc && dcc->f2fs_issue_discard) { 1748 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1749 1750 dcc->f2fs_issue_discard = NULL; 1751 kthread_stop(discard_thread); 1752 } 1753 } 1754 1755 /* This comes from f2fs_put_super */ 1756 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1757 { 1758 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1759 struct discard_policy dpolicy; 1760 bool dropped; 1761 1762 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1763 dcc->discard_granularity); 1764 __issue_discard_cmd(sbi, &dpolicy); 1765 dropped = __drop_discard_cmd(sbi); 1766 1767 /* just to make sure there is no pending discard commands */ 1768 __wait_all_discard_cmd(sbi, NULL); 1769 1770 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1771 return dropped; 1772 } 1773 1774 static int issue_discard_thread(void *data) 1775 { 1776 struct f2fs_sb_info *sbi = data; 1777 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1778 wait_queue_head_t *q = &dcc->discard_wait_queue; 1779 struct discard_policy dpolicy; 1780 unsigned int wait_ms = dcc->min_discard_issue_time; 1781 int issued; 1782 1783 set_freezable(); 1784 1785 do { 1786 if (sbi->gc_mode == GC_URGENT_HIGH || 1787 !f2fs_available_free_memory(sbi, DISCARD_CACHE)) 1788 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1); 1789 else 1790 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1791 dcc->discard_granularity); 1792 1793 if (!atomic_read(&dcc->discard_cmd_cnt)) 1794 wait_ms = dpolicy.max_interval; 1795 1796 wait_event_interruptible_timeout(*q, 1797 kthread_should_stop() || freezing(current) || 1798 dcc->discard_wake, 1799 msecs_to_jiffies(wait_ms)); 1800 1801 if (dcc->discard_wake) 1802 dcc->discard_wake = 0; 1803 1804 /* clean up pending candidates before going to sleep */ 1805 if (atomic_read(&dcc->queued_discard)) 1806 __wait_all_discard_cmd(sbi, NULL); 1807 1808 if (try_to_freeze()) 1809 continue; 1810 if (f2fs_readonly(sbi->sb)) 1811 continue; 1812 if (kthread_should_stop()) 1813 return 0; 1814 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1815 wait_ms = dpolicy.max_interval; 1816 continue; 1817 } 1818 if (!atomic_read(&dcc->discard_cmd_cnt)) 1819 continue; 1820 1821 sb_start_intwrite(sbi->sb); 1822 1823 issued = __issue_discard_cmd(sbi, &dpolicy); 1824 if (issued > 0) { 1825 __wait_all_discard_cmd(sbi, &dpolicy); 1826 wait_ms = dpolicy.min_interval; 1827 } else if (issued == -1) { 1828 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1829 if (!wait_ms) 1830 wait_ms = dpolicy.mid_interval; 1831 } else { 1832 wait_ms = dpolicy.max_interval; 1833 } 1834 1835 sb_end_intwrite(sbi->sb); 1836 1837 } while (!kthread_should_stop()); 1838 return 0; 1839 } 1840 1841 #ifdef CONFIG_BLK_DEV_ZONED 1842 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1843 struct block_device *bdev, block_t blkstart, block_t blklen) 1844 { 1845 sector_t sector, nr_sects; 1846 block_t lblkstart = blkstart; 1847 int devi = 0; 1848 1849 if (f2fs_is_multi_device(sbi)) { 1850 devi = f2fs_target_device_index(sbi, blkstart); 1851 if (blkstart < FDEV(devi).start_blk || 1852 blkstart > FDEV(devi).end_blk) { 1853 f2fs_err(sbi, "Invalid block %x", blkstart); 1854 return -EIO; 1855 } 1856 blkstart -= FDEV(devi).start_blk; 1857 } 1858 1859 /* For sequential zones, reset the zone write pointer */ 1860 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1861 sector = SECTOR_FROM_BLOCK(blkstart); 1862 nr_sects = SECTOR_FROM_BLOCK(blklen); 1863 1864 if (sector & (bdev_zone_sectors(bdev) - 1) || 1865 nr_sects != bdev_zone_sectors(bdev)) { 1866 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1867 devi, sbi->s_ndevs ? FDEV(devi).path : "", 1868 blkstart, blklen); 1869 return -EIO; 1870 } 1871 trace_f2fs_issue_reset_zone(bdev, blkstart); 1872 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 1873 sector, nr_sects, GFP_NOFS); 1874 } 1875 1876 /* For conventional zones, use regular discard if supported */ 1877 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1878 } 1879 #endif 1880 1881 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1882 struct block_device *bdev, block_t blkstart, block_t blklen) 1883 { 1884 #ifdef CONFIG_BLK_DEV_ZONED 1885 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 1886 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1887 #endif 1888 return __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1889 } 1890 1891 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1892 block_t blkstart, block_t blklen) 1893 { 1894 sector_t start = blkstart, len = 0; 1895 struct block_device *bdev; 1896 struct seg_entry *se; 1897 unsigned int offset; 1898 block_t i; 1899 int err = 0; 1900 1901 bdev = f2fs_target_device(sbi, blkstart, NULL); 1902 1903 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1904 if (i != start) { 1905 struct block_device *bdev2 = 1906 f2fs_target_device(sbi, i, NULL); 1907 1908 if (bdev2 != bdev) { 1909 err = __issue_discard_async(sbi, bdev, 1910 start, len); 1911 if (err) 1912 return err; 1913 bdev = bdev2; 1914 start = i; 1915 len = 0; 1916 } 1917 } 1918 1919 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 1920 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 1921 1922 if (f2fs_block_unit_discard(sbi) && 1923 !f2fs_test_and_set_bit(offset, se->discard_map)) 1924 sbi->discard_blks--; 1925 } 1926 1927 if (len) 1928 err = __issue_discard_async(sbi, bdev, start, len); 1929 return err; 1930 } 1931 1932 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 1933 bool check_only) 1934 { 1935 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1936 int max_blocks = sbi->blocks_per_seg; 1937 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 1938 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1939 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1940 unsigned long *discard_map = (unsigned long *)se->discard_map; 1941 unsigned long *dmap = SIT_I(sbi)->tmp_map; 1942 unsigned int start = 0, end = -1; 1943 bool force = (cpc->reason & CP_DISCARD); 1944 struct discard_entry *de = NULL; 1945 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 1946 int i; 1947 1948 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) || 1949 !f2fs_block_unit_discard(sbi)) 1950 return false; 1951 1952 if (!force) { 1953 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks || 1954 SM_I(sbi)->dcc_info->nr_discards >= 1955 SM_I(sbi)->dcc_info->max_discards) 1956 return false; 1957 } 1958 1959 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 1960 for (i = 0; i < entries; i++) 1961 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 1962 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 1963 1964 while (force || SM_I(sbi)->dcc_info->nr_discards <= 1965 SM_I(sbi)->dcc_info->max_discards) { 1966 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 1967 if (start >= max_blocks) 1968 break; 1969 1970 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 1971 if (force && start && end != max_blocks 1972 && (end - start) < cpc->trim_minlen) 1973 continue; 1974 1975 if (check_only) 1976 return true; 1977 1978 if (!de) { 1979 de = f2fs_kmem_cache_alloc(discard_entry_slab, 1980 GFP_F2FS_ZERO, true, NULL); 1981 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 1982 list_add_tail(&de->list, head); 1983 } 1984 1985 for (i = start; i < end; i++) 1986 __set_bit_le(i, (void *)de->discard_map); 1987 1988 SM_I(sbi)->dcc_info->nr_discards += end - start; 1989 } 1990 return false; 1991 } 1992 1993 static void release_discard_addr(struct discard_entry *entry) 1994 { 1995 list_del(&entry->list); 1996 kmem_cache_free(discard_entry_slab, entry); 1997 } 1998 1999 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 2000 { 2001 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 2002 struct discard_entry *entry, *this; 2003 2004 /* drop caches */ 2005 list_for_each_entry_safe(entry, this, head, list) 2006 release_discard_addr(entry); 2007 } 2008 2009 /* 2010 * Should call f2fs_clear_prefree_segments after checkpoint is done. 2011 */ 2012 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 2013 { 2014 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2015 unsigned int segno; 2016 2017 mutex_lock(&dirty_i->seglist_lock); 2018 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 2019 __set_test_and_free(sbi, segno, false); 2020 mutex_unlock(&dirty_i->seglist_lock); 2021 } 2022 2023 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 2024 struct cp_control *cpc) 2025 { 2026 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2027 struct list_head *head = &dcc->entry_list; 2028 struct discard_entry *entry, *this; 2029 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2030 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 2031 unsigned int start = 0, end = -1; 2032 unsigned int secno, start_segno; 2033 bool force = (cpc->reason & CP_DISCARD); 2034 bool section_alignment = F2FS_OPTION(sbi).discard_unit == 2035 DISCARD_UNIT_SECTION; 2036 2037 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi)) 2038 section_alignment = true; 2039 2040 mutex_lock(&dirty_i->seglist_lock); 2041 2042 while (1) { 2043 int i; 2044 2045 if (section_alignment && end != -1) 2046 end--; 2047 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 2048 if (start >= MAIN_SEGS(sbi)) 2049 break; 2050 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 2051 start + 1); 2052 2053 if (section_alignment) { 2054 start = rounddown(start, sbi->segs_per_sec); 2055 end = roundup(end, sbi->segs_per_sec); 2056 } 2057 2058 for (i = start; i < end; i++) { 2059 if (test_and_clear_bit(i, prefree_map)) 2060 dirty_i->nr_dirty[PRE]--; 2061 } 2062 2063 if (!f2fs_realtime_discard_enable(sbi)) 2064 continue; 2065 2066 if (force && start >= cpc->trim_start && 2067 (end - 1) <= cpc->trim_end) 2068 continue; 2069 2070 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) { 2071 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 2072 (end - start) << sbi->log_blocks_per_seg); 2073 continue; 2074 } 2075 next: 2076 secno = GET_SEC_FROM_SEG(sbi, start); 2077 start_segno = GET_SEG_FROM_SEC(sbi, secno); 2078 if (!IS_CURSEC(sbi, secno) && 2079 !get_valid_blocks(sbi, start, true)) 2080 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 2081 sbi->segs_per_sec << sbi->log_blocks_per_seg); 2082 2083 start = start_segno + sbi->segs_per_sec; 2084 if (start < end) 2085 goto next; 2086 else 2087 end = start - 1; 2088 } 2089 mutex_unlock(&dirty_i->seglist_lock); 2090 2091 if (!f2fs_block_unit_discard(sbi)) 2092 goto wakeup; 2093 2094 /* send small discards */ 2095 list_for_each_entry_safe(entry, this, head, list) { 2096 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2097 bool is_valid = test_bit_le(0, entry->discard_map); 2098 2099 find_next: 2100 if (is_valid) { 2101 next_pos = find_next_zero_bit_le(entry->discard_map, 2102 sbi->blocks_per_seg, cur_pos); 2103 len = next_pos - cur_pos; 2104 2105 if (f2fs_sb_has_blkzoned(sbi) || 2106 (force && len < cpc->trim_minlen)) 2107 goto skip; 2108 2109 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2110 len); 2111 total_len += len; 2112 } else { 2113 next_pos = find_next_bit_le(entry->discard_map, 2114 sbi->blocks_per_seg, cur_pos); 2115 } 2116 skip: 2117 cur_pos = next_pos; 2118 is_valid = !is_valid; 2119 2120 if (cur_pos < sbi->blocks_per_seg) 2121 goto find_next; 2122 2123 release_discard_addr(entry); 2124 dcc->nr_discards -= total_len; 2125 } 2126 2127 wakeup: 2128 wake_up_discard_thread(sbi, false); 2129 } 2130 2131 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi) 2132 { 2133 dev_t dev = sbi->sb->s_bdev->bd_dev; 2134 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2135 int err = 0; 2136 2137 if (!f2fs_realtime_discard_enable(sbi)) 2138 return 0; 2139 2140 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2141 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2142 if (IS_ERR(dcc->f2fs_issue_discard)) 2143 err = PTR_ERR(dcc->f2fs_issue_discard); 2144 2145 return err; 2146 } 2147 2148 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2149 { 2150 struct discard_cmd_control *dcc; 2151 int err = 0, i; 2152 2153 if (SM_I(sbi)->dcc_info) { 2154 dcc = SM_I(sbi)->dcc_info; 2155 goto init_thread; 2156 } 2157 2158 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2159 if (!dcc) 2160 return -ENOMEM; 2161 2162 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2163 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT) 2164 dcc->discard_granularity = sbi->blocks_per_seg; 2165 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION) 2166 dcc->discard_granularity = BLKS_PER_SEC(sbi); 2167 2168 INIT_LIST_HEAD(&dcc->entry_list); 2169 for (i = 0; i < MAX_PLIST_NUM; i++) 2170 INIT_LIST_HEAD(&dcc->pend_list[i]); 2171 INIT_LIST_HEAD(&dcc->wait_list); 2172 INIT_LIST_HEAD(&dcc->fstrim_list); 2173 mutex_init(&dcc->cmd_lock); 2174 atomic_set(&dcc->issued_discard, 0); 2175 atomic_set(&dcc->queued_discard, 0); 2176 atomic_set(&dcc->discard_cmd_cnt, 0); 2177 dcc->nr_discards = 0; 2178 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 2179 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST; 2180 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME; 2181 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME; 2182 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME; 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 f2fs_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 f2fs_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 f2fs_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 f2fs_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 f2fs_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 f2fs_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 f2fs_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 f2fs_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 f2fs_io_schedule_timeout(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 f2fs_down_write(&sbi->gc_lock); 3196 err = f2fs_write_checkpoint(sbi, &cpc); 3197 f2fs_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 static int __get_segment_type_2(struct f2fs_io_info *fio) 3245 { 3246 if (fio->type == DATA) 3247 return CURSEG_HOT_DATA; 3248 else 3249 return CURSEG_HOT_NODE; 3250 } 3251 3252 static int __get_segment_type_4(struct f2fs_io_info *fio) 3253 { 3254 if (fio->type == DATA) { 3255 struct inode *inode = fio->page->mapping->host; 3256 3257 if (S_ISDIR(inode->i_mode)) 3258 return CURSEG_HOT_DATA; 3259 else 3260 return CURSEG_COLD_DATA; 3261 } else { 3262 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3263 return CURSEG_WARM_NODE; 3264 else 3265 return CURSEG_COLD_NODE; 3266 } 3267 } 3268 3269 static int __get_segment_type_6(struct f2fs_io_info *fio) 3270 { 3271 if (fio->type == DATA) { 3272 struct inode *inode = fio->page->mapping->host; 3273 3274 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 3275 return CURSEG_COLD_DATA_PINNED; 3276 3277 if (page_private_gcing(fio->page)) { 3278 if (fio->sbi->am.atgc_enabled && 3279 (fio->io_type == FS_DATA_IO) && 3280 (fio->sbi->gc_mode != GC_URGENT_HIGH)) 3281 return CURSEG_ALL_DATA_ATGC; 3282 else 3283 return CURSEG_COLD_DATA; 3284 } 3285 if (file_is_cold(inode) || f2fs_need_compress_data(inode)) 3286 return CURSEG_COLD_DATA; 3287 if (file_is_hot(inode) || 3288 is_inode_flag_set(inode, FI_HOT_DATA) || 3289 f2fs_is_atomic_file(inode) || 3290 f2fs_is_volatile_file(inode)) 3291 return CURSEG_HOT_DATA; 3292 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 3293 } else { 3294 if (IS_DNODE(fio->page)) 3295 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3296 CURSEG_HOT_NODE; 3297 return CURSEG_COLD_NODE; 3298 } 3299 } 3300 3301 static int __get_segment_type(struct f2fs_io_info *fio) 3302 { 3303 int type = 0; 3304 3305 switch (F2FS_OPTION(fio->sbi).active_logs) { 3306 case 2: 3307 type = __get_segment_type_2(fio); 3308 break; 3309 case 4: 3310 type = __get_segment_type_4(fio); 3311 break; 3312 case 6: 3313 type = __get_segment_type_6(fio); 3314 break; 3315 default: 3316 f2fs_bug_on(fio->sbi, true); 3317 } 3318 3319 if (IS_HOT(type)) 3320 fio->temp = HOT; 3321 else if (IS_WARM(type)) 3322 fio->temp = WARM; 3323 else 3324 fio->temp = COLD; 3325 return type; 3326 } 3327 3328 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3329 block_t old_blkaddr, block_t *new_blkaddr, 3330 struct f2fs_summary *sum, int type, 3331 struct f2fs_io_info *fio) 3332 { 3333 struct sit_info *sit_i = SIT_I(sbi); 3334 struct curseg_info *curseg = CURSEG_I(sbi, type); 3335 unsigned long long old_mtime; 3336 bool from_gc = (type == CURSEG_ALL_DATA_ATGC); 3337 struct seg_entry *se = NULL; 3338 3339 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3340 3341 mutex_lock(&curseg->curseg_mutex); 3342 down_write(&sit_i->sentry_lock); 3343 3344 if (from_gc) { 3345 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO); 3346 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr)); 3347 sanity_check_seg_type(sbi, se->type); 3348 f2fs_bug_on(sbi, IS_NODESEG(se->type)); 3349 } 3350 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3351 3352 f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg); 3353 3354 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3355 3356 /* 3357 * __add_sum_entry should be resided under the curseg_mutex 3358 * because, this function updates a summary entry in the 3359 * current summary block. 3360 */ 3361 __add_sum_entry(sbi, type, sum); 3362 3363 __refresh_next_blkoff(sbi, curseg); 3364 3365 stat_inc_block_count(sbi, curseg); 3366 3367 if (from_gc) { 3368 old_mtime = get_segment_mtime(sbi, old_blkaddr); 3369 } else { 3370 update_segment_mtime(sbi, old_blkaddr, 0); 3371 old_mtime = 0; 3372 } 3373 update_segment_mtime(sbi, *new_blkaddr, old_mtime); 3374 3375 /* 3376 * SIT information should be updated before segment allocation, 3377 * since SSR needs latest valid block information. 3378 */ 3379 update_sit_entry(sbi, *new_blkaddr, 1); 3380 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 3381 update_sit_entry(sbi, old_blkaddr, -1); 3382 3383 if (!__has_curseg_space(sbi, curseg)) { 3384 if (from_gc) 3385 get_atssr_segment(sbi, type, se->type, 3386 AT_SSR, se->mtime); 3387 else 3388 sit_i->s_ops->allocate_segment(sbi, type, false); 3389 } 3390 /* 3391 * segment dirty status should be updated after segment allocation, 3392 * so we just need to update status only one time after previous 3393 * segment being closed. 3394 */ 3395 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3396 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3397 3398 up_write(&sit_i->sentry_lock); 3399 3400 if (page && IS_NODESEG(type)) { 3401 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3402 3403 f2fs_inode_chksum_set(sbi, page); 3404 } 3405 3406 if (fio) { 3407 struct f2fs_bio_info *io; 3408 3409 if (F2FS_IO_ALIGNED(sbi)) 3410 fio->retry = false; 3411 3412 INIT_LIST_HEAD(&fio->list); 3413 fio->in_list = true; 3414 io = sbi->write_io[fio->type] + fio->temp; 3415 spin_lock(&io->io_lock); 3416 list_add_tail(&fio->list, &io->io_list); 3417 spin_unlock(&io->io_lock); 3418 } 3419 3420 mutex_unlock(&curseg->curseg_mutex); 3421 3422 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3423 } 3424 3425 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino, 3426 block_t blkaddr, unsigned int blkcnt) 3427 { 3428 if (!f2fs_is_multi_device(sbi)) 3429 return; 3430 3431 while (1) { 3432 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr); 3433 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1; 3434 3435 /* update device state for fsync */ 3436 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO); 3437 3438 /* update device state for checkpoint */ 3439 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3440 spin_lock(&sbi->dev_lock); 3441 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3442 spin_unlock(&sbi->dev_lock); 3443 } 3444 3445 if (blkcnt <= blks) 3446 break; 3447 blkcnt -= blks; 3448 blkaddr += blks; 3449 } 3450 } 3451 3452 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3453 { 3454 int type = __get_segment_type(fio); 3455 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA); 3456 3457 if (keep_order) 3458 f2fs_down_read(&fio->sbi->io_order_lock); 3459 reallocate: 3460 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3461 &fio->new_blkaddr, sum, type, fio); 3462 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) { 3463 invalidate_mapping_pages(META_MAPPING(fio->sbi), 3464 fio->old_blkaddr, fio->old_blkaddr); 3465 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr); 3466 } 3467 3468 /* writeout dirty page into bdev */ 3469 f2fs_submit_page_write(fio); 3470 if (fio->retry) { 3471 fio->old_blkaddr = fio->new_blkaddr; 3472 goto reallocate; 3473 } 3474 3475 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1); 3476 3477 if (keep_order) 3478 f2fs_up_read(&fio->sbi->io_order_lock); 3479 } 3480 3481 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 3482 enum iostat_type io_type) 3483 { 3484 struct f2fs_io_info fio = { 3485 .sbi = sbi, 3486 .type = META, 3487 .temp = HOT, 3488 .op = REQ_OP_WRITE, 3489 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3490 .old_blkaddr = page->index, 3491 .new_blkaddr = page->index, 3492 .page = page, 3493 .encrypted_page = NULL, 3494 .in_list = false, 3495 }; 3496 3497 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 3498 fio.op_flags &= ~REQ_META; 3499 3500 set_page_writeback(page); 3501 ClearPageError(page); 3502 f2fs_submit_page_write(&fio); 3503 3504 stat_inc_meta_count(sbi, page->index); 3505 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE); 3506 } 3507 3508 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3509 { 3510 struct f2fs_summary sum; 3511 3512 set_summary(&sum, nid, 0, 0); 3513 do_write_page(&sum, fio); 3514 3515 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 3516 } 3517 3518 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3519 struct f2fs_io_info *fio) 3520 { 3521 struct f2fs_sb_info *sbi = fio->sbi; 3522 struct f2fs_summary sum; 3523 3524 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3525 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3526 do_write_page(&sum, fio); 3527 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3528 3529 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE); 3530 } 3531 3532 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3533 { 3534 int err; 3535 struct f2fs_sb_info *sbi = fio->sbi; 3536 unsigned int segno; 3537 3538 fio->new_blkaddr = fio->old_blkaddr; 3539 /* i/o temperature is needed for passing down write hints */ 3540 __get_segment_type(fio); 3541 3542 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3543 3544 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 3545 set_sbi_flag(sbi, SBI_NEED_FSCK); 3546 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 3547 __func__, segno); 3548 err = -EFSCORRUPTED; 3549 goto drop_bio; 3550 } 3551 3552 if (f2fs_cp_error(sbi)) { 3553 err = -EIO; 3554 goto drop_bio; 3555 } 3556 3557 invalidate_mapping_pages(META_MAPPING(sbi), 3558 fio->new_blkaddr, fio->new_blkaddr); 3559 3560 stat_inc_inplace_blocks(fio->sbi); 3561 3562 if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE))) 3563 err = f2fs_merge_page_bio(fio); 3564 else 3565 err = f2fs_submit_page_bio(fio); 3566 if (!err) { 3567 f2fs_update_device_state(fio->sbi, fio->ino, 3568 fio->new_blkaddr, 1); 3569 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 3570 } 3571 3572 return err; 3573 drop_bio: 3574 if (fio->bio && *(fio->bio)) { 3575 struct bio *bio = *(fio->bio); 3576 3577 bio->bi_status = BLK_STS_IOERR; 3578 bio_endio(bio); 3579 *(fio->bio) = NULL; 3580 } 3581 return err; 3582 } 3583 3584 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 3585 unsigned int segno) 3586 { 3587 int i; 3588 3589 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 3590 if (CURSEG_I(sbi, i)->segno == segno) 3591 break; 3592 } 3593 return i; 3594 } 3595 3596 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 3597 block_t old_blkaddr, block_t new_blkaddr, 3598 bool recover_curseg, bool recover_newaddr, 3599 bool from_gc) 3600 { 3601 struct sit_info *sit_i = SIT_I(sbi); 3602 struct curseg_info *curseg; 3603 unsigned int segno, old_cursegno; 3604 struct seg_entry *se; 3605 int type; 3606 unsigned short old_blkoff; 3607 unsigned char old_alloc_type; 3608 3609 segno = GET_SEGNO(sbi, new_blkaddr); 3610 se = get_seg_entry(sbi, segno); 3611 type = se->type; 3612 3613 f2fs_down_write(&SM_I(sbi)->curseg_lock); 3614 3615 if (!recover_curseg) { 3616 /* for recovery flow */ 3617 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 3618 if (old_blkaddr == NULL_ADDR) 3619 type = CURSEG_COLD_DATA; 3620 else 3621 type = CURSEG_WARM_DATA; 3622 } 3623 } else { 3624 if (IS_CURSEG(sbi, segno)) { 3625 /* se->type is volatile as SSR allocation */ 3626 type = __f2fs_get_curseg(sbi, segno); 3627 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 3628 } else { 3629 type = CURSEG_WARM_DATA; 3630 } 3631 } 3632 3633 f2fs_bug_on(sbi, !IS_DATASEG(type)); 3634 curseg = CURSEG_I(sbi, type); 3635 3636 mutex_lock(&curseg->curseg_mutex); 3637 down_write(&sit_i->sentry_lock); 3638 3639 old_cursegno = curseg->segno; 3640 old_blkoff = curseg->next_blkoff; 3641 old_alloc_type = curseg->alloc_type; 3642 3643 /* change the current segment */ 3644 if (segno != curseg->segno) { 3645 curseg->next_segno = segno; 3646 change_curseg(sbi, type, true); 3647 } 3648 3649 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 3650 __add_sum_entry(sbi, type, sum); 3651 3652 if (!recover_curseg || recover_newaddr) { 3653 if (!from_gc) 3654 update_segment_mtime(sbi, new_blkaddr, 0); 3655 update_sit_entry(sbi, new_blkaddr, 1); 3656 } 3657 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 3658 invalidate_mapping_pages(META_MAPPING(sbi), 3659 old_blkaddr, old_blkaddr); 3660 f2fs_invalidate_compress_page(sbi, old_blkaddr); 3661 if (!from_gc) 3662 update_segment_mtime(sbi, old_blkaddr, 0); 3663 update_sit_entry(sbi, old_blkaddr, -1); 3664 } 3665 3666 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3667 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 3668 3669 locate_dirty_segment(sbi, old_cursegno); 3670 3671 if (recover_curseg) { 3672 if (old_cursegno != curseg->segno) { 3673 curseg->next_segno = old_cursegno; 3674 change_curseg(sbi, type, true); 3675 } 3676 curseg->next_blkoff = old_blkoff; 3677 curseg->alloc_type = old_alloc_type; 3678 } 3679 3680 up_write(&sit_i->sentry_lock); 3681 mutex_unlock(&curseg->curseg_mutex); 3682 f2fs_up_write(&SM_I(sbi)->curseg_lock); 3683 } 3684 3685 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 3686 block_t old_addr, block_t new_addr, 3687 unsigned char version, bool recover_curseg, 3688 bool recover_newaddr) 3689 { 3690 struct f2fs_summary sum; 3691 3692 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 3693 3694 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 3695 recover_curseg, recover_newaddr, false); 3696 3697 f2fs_update_data_blkaddr(dn, new_addr); 3698 } 3699 3700 void f2fs_wait_on_page_writeback(struct page *page, 3701 enum page_type type, bool ordered, bool locked) 3702 { 3703 if (PageWriteback(page)) { 3704 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 3705 3706 /* submit cached LFS IO */ 3707 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type); 3708 /* sbumit cached IPU IO */ 3709 f2fs_submit_merged_ipu_write(sbi, NULL, page); 3710 if (ordered) { 3711 wait_on_page_writeback(page); 3712 f2fs_bug_on(sbi, locked && PageWriteback(page)); 3713 } else { 3714 wait_for_stable_page(page); 3715 } 3716 } 3717 } 3718 3719 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 3720 { 3721 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3722 struct page *cpage; 3723 3724 if (!f2fs_post_read_required(inode)) 3725 return; 3726 3727 if (!__is_valid_data_blkaddr(blkaddr)) 3728 return; 3729 3730 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3731 if (cpage) { 3732 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 3733 f2fs_put_page(cpage, 1); 3734 } 3735 } 3736 3737 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 3738 block_t len) 3739 { 3740 block_t i; 3741 3742 for (i = 0; i < len; i++) 3743 f2fs_wait_on_block_writeback(inode, blkaddr + i); 3744 } 3745 3746 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 3747 { 3748 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3749 struct curseg_info *seg_i; 3750 unsigned char *kaddr; 3751 struct page *page; 3752 block_t start; 3753 int i, j, offset; 3754 3755 start = start_sum_block(sbi); 3756 3757 page = f2fs_get_meta_page(sbi, start++); 3758 if (IS_ERR(page)) 3759 return PTR_ERR(page); 3760 kaddr = (unsigned char *)page_address(page); 3761 3762 /* Step 1: restore nat cache */ 3763 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3764 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3765 3766 /* Step 2: restore sit cache */ 3767 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3768 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3769 offset = 2 * SUM_JOURNAL_SIZE; 3770 3771 /* Step 3: restore summary entries */ 3772 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3773 unsigned short blk_off; 3774 unsigned int segno; 3775 3776 seg_i = CURSEG_I(sbi, i); 3777 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3778 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3779 seg_i->next_segno = segno; 3780 reset_curseg(sbi, i, 0); 3781 seg_i->alloc_type = ckpt->alloc_type[i]; 3782 seg_i->next_blkoff = blk_off; 3783 3784 if (seg_i->alloc_type == SSR) 3785 blk_off = sbi->blocks_per_seg; 3786 3787 for (j = 0; j < blk_off; j++) { 3788 struct f2fs_summary *s; 3789 3790 s = (struct f2fs_summary *)(kaddr + offset); 3791 seg_i->sum_blk->entries[j] = *s; 3792 offset += SUMMARY_SIZE; 3793 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3794 SUM_FOOTER_SIZE) 3795 continue; 3796 3797 f2fs_put_page(page, 1); 3798 page = NULL; 3799 3800 page = f2fs_get_meta_page(sbi, start++); 3801 if (IS_ERR(page)) 3802 return PTR_ERR(page); 3803 kaddr = (unsigned char *)page_address(page); 3804 offset = 0; 3805 } 3806 } 3807 f2fs_put_page(page, 1); 3808 return 0; 3809 } 3810 3811 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3812 { 3813 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3814 struct f2fs_summary_block *sum; 3815 struct curseg_info *curseg; 3816 struct page *new; 3817 unsigned short blk_off; 3818 unsigned int segno = 0; 3819 block_t blk_addr = 0; 3820 int err = 0; 3821 3822 /* get segment number and block addr */ 3823 if (IS_DATASEG(type)) { 3824 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3825 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3826 CURSEG_HOT_DATA]); 3827 if (__exist_node_summaries(sbi)) 3828 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type); 3829 else 3830 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3831 } else { 3832 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3833 CURSEG_HOT_NODE]); 3834 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3835 CURSEG_HOT_NODE]); 3836 if (__exist_node_summaries(sbi)) 3837 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3838 type - CURSEG_HOT_NODE); 3839 else 3840 blk_addr = GET_SUM_BLOCK(sbi, segno); 3841 } 3842 3843 new = f2fs_get_meta_page(sbi, blk_addr); 3844 if (IS_ERR(new)) 3845 return PTR_ERR(new); 3846 sum = (struct f2fs_summary_block *)page_address(new); 3847 3848 if (IS_NODESEG(type)) { 3849 if (__exist_node_summaries(sbi)) { 3850 struct f2fs_summary *ns = &sum->entries[0]; 3851 int i; 3852 3853 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3854 ns->version = 0; 3855 ns->ofs_in_node = 0; 3856 } 3857 } else { 3858 err = f2fs_restore_node_summary(sbi, segno, sum); 3859 if (err) 3860 goto out; 3861 } 3862 } 3863 3864 /* set uncompleted segment to curseg */ 3865 curseg = CURSEG_I(sbi, type); 3866 mutex_lock(&curseg->curseg_mutex); 3867 3868 /* update journal info */ 3869 down_write(&curseg->journal_rwsem); 3870 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3871 up_write(&curseg->journal_rwsem); 3872 3873 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3874 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3875 curseg->next_segno = segno; 3876 reset_curseg(sbi, type, 0); 3877 curseg->alloc_type = ckpt->alloc_type[type]; 3878 curseg->next_blkoff = blk_off; 3879 mutex_unlock(&curseg->curseg_mutex); 3880 out: 3881 f2fs_put_page(new, 1); 3882 return err; 3883 } 3884 3885 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3886 { 3887 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3888 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3889 int type = CURSEG_HOT_DATA; 3890 int err; 3891 3892 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 3893 int npages = f2fs_npages_for_summary_flush(sbi, true); 3894 3895 if (npages >= 2) 3896 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 3897 META_CP, true); 3898 3899 /* restore for compacted data summary */ 3900 err = read_compacted_summaries(sbi); 3901 if (err) 3902 return err; 3903 type = CURSEG_HOT_NODE; 3904 } 3905 3906 if (__exist_node_summaries(sbi)) 3907 f2fs_ra_meta_pages(sbi, 3908 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type), 3909 NR_CURSEG_PERSIST_TYPE - type, META_CP, true); 3910 3911 for (; type <= CURSEG_COLD_NODE; type++) { 3912 err = read_normal_summaries(sbi, type); 3913 if (err) 3914 return err; 3915 } 3916 3917 /* sanity check for summary blocks */ 3918 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 3919 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 3920 f2fs_err(sbi, "invalid journal entries nats %u sits %u", 3921 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 3922 return -EINVAL; 3923 } 3924 3925 return 0; 3926 } 3927 3928 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 3929 { 3930 struct page *page; 3931 unsigned char *kaddr; 3932 struct f2fs_summary *summary; 3933 struct curseg_info *seg_i; 3934 int written_size = 0; 3935 int i, j; 3936 3937 page = f2fs_grab_meta_page(sbi, blkaddr++); 3938 kaddr = (unsigned char *)page_address(page); 3939 memset(kaddr, 0, PAGE_SIZE); 3940 3941 /* Step 1: write nat cache */ 3942 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3943 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 3944 written_size += SUM_JOURNAL_SIZE; 3945 3946 /* Step 2: write sit cache */ 3947 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3948 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 3949 written_size += SUM_JOURNAL_SIZE; 3950 3951 /* Step 3: write summary entries */ 3952 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3953 unsigned short blkoff; 3954 3955 seg_i = CURSEG_I(sbi, i); 3956 if (sbi->ckpt->alloc_type[i] == SSR) 3957 blkoff = sbi->blocks_per_seg; 3958 else 3959 blkoff = curseg_blkoff(sbi, i); 3960 3961 for (j = 0; j < blkoff; j++) { 3962 if (!page) { 3963 page = f2fs_grab_meta_page(sbi, blkaddr++); 3964 kaddr = (unsigned char *)page_address(page); 3965 memset(kaddr, 0, PAGE_SIZE); 3966 written_size = 0; 3967 } 3968 summary = (struct f2fs_summary *)(kaddr + written_size); 3969 *summary = seg_i->sum_blk->entries[j]; 3970 written_size += SUMMARY_SIZE; 3971 3972 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 3973 SUM_FOOTER_SIZE) 3974 continue; 3975 3976 set_page_dirty(page); 3977 f2fs_put_page(page, 1); 3978 page = NULL; 3979 } 3980 } 3981 if (page) { 3982 set_page_dirty(page); 3983 f2fs_put_page(page, 1); 3984 } 3985 } 3986 3987 static void write_normal_summaries(struct f2fs_sb_info *sbi, 3988 block_t blkaddr, int type) 3989 { 3990 int i, end; 3991 3992 if (IS_DATASEG(type)) 3993 end = type + NR_CURSEG_DATA_TYPE; 3994 else 3995 end = type + NR_CURSEG_NODE_TYPE; 3996 3997 for (i = type; i < end; i++) 3998 write_current_sum_page(sbi, i, blkaddr + (i - type)); 3999 } 4000 4001 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4002 { 4003 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 4004 write_compacted_summaries(sbi, start_blk); 4005 else 4006 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 4007 } 4008 4009 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4010 { 4011 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 4012 } 4013 4014 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 4015 unsigned int val, int alloc) 4016 { 4017 int i; 4018 4019 if (type == NAT_JOURNAL) { 4020 for (i = 0; i < nats_in_cursum(journal); i++) { 4021 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 4022 return i; 4023 } 4024 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 4025 return update_nats_in_cursum(journal, 1); 4026 } else if (type == SIT_JOURNAL) { 4027 for (i = 0; i < sits_in_cursum(journal); i++) 4028 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 4029 return i; 4030 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 4031 return update_sits_in_cursum(journal, 1); 4032 } 4033 return -1; 4034 } 4035 4036 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 4037 unsigned int segno) 4038 { 4039 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno)); 4040 } 4041 4042 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 4043 unsigned int start) 4044 { 4045 struct sit_info *sit_i = SIT_I(sbi); 4046 struct page *page; 4047 pgoff_t src_off, dst_off; 4048 4049 src_off = current_sit_addr(sbi, start); 4050 dst_off = next_sit_addr(sbi, src_off); 4051 4052 page = f2fs_grab_meta_page(sbi, dst_off); 4053 seg_info_to_sit_page(sbi, page, start); 4054 4055 set_page_dirty(page); 4056 set_to_next_sit(sit_i, start); 4057 4058 return page; 4059 } 4060 4061 static struct sit_entry_set *grab_sit_entry_set(void) 4062 { 4063 struct sit_entry_set *ses = 4064 f2fs_kmem_cache_alloc(sit_entry_set_slab, 4065 GFP_NOFS, true, NULL); 4066 4067 ses->entry_cnt = 0; 4068 INIT_LIST_HEAD(&ses->set_list); 4069 return ses; 4070 } 4071 4072 static void release_sit_entry_set(struct sit_entry_set *ses) 4073 { 4074 list_del(&ses->set_list); 4075 kmem_cache_free(sit_entry_set_slab, ses); 4076 } 4077 4078 static void adjust_sit_entry_set(struct sit_entry_set *ses, 4079 struct list_head *head) 4080 { 4081 struct sit_entry_set *next = ses; 4082 4083 if (list_is_last(&ses->set_list, head)) 4084 return; 4085 4086 list_for_each_entry_continue(next, head, set_list) 4087 if (ses->entry_cnt <= next->entry_cnt) 4088 break; 4089 4090 list_move_tail(&ses->set_list, &next->set_list); 4091 } 4092 4093 static void add_sit_entry(unsigned int segno, struct list_head *head) 4094 { 4095 struct sit_entry_set *ses; 4096 unsigned int start_segno = START_SEGNO(segno); 4097 4098 list_for_each_entry(ses, head, set_list) { 4099 if (ses->start_segno == start_segno) { 4100 ses->entry_cnt++; 4101 adjust_sit_entry_set(ses, head); 4102 return; 4103 } 4104 } 4105 4106 ses = grab_sit_entry_set(); 4107 4108 ses->start_segno = start_segno; 4109 ses->entry_cnt++; 4110 list_add(&ses->set_list, head); 4111 } 4112 4113 static void add_sits_in_set(struct f2fs_sb_info *sbi) 4114 { 4115 struct f2fs_sm_info *sm_info = SM_I(sbi); 4116 struct list_head *set_list = &sm_info->sit_entry_set; 4117 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 4118 unsigned int segno; 4119 4120 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 4121 add_sit_entry(segno, set_list); 4122 } 4123 4124 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 4125 { 4126 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4127 struct f2fs_journal *journal = curseg->journal; 4128 int i; 4129 4130 down_write(&curseg->journal_rwsem); 4131 for (i = 0; i < sits_in_cursum(journal); i++) { 4132 unsigned int segno; 4133 bool dirtied; 4134 4135 segno = le32_to_cpu(segno_in_journal(journal, i)); 4136 dirtied = __mark_sit_entry_dirty(sbi, segno); 4137 4138 if (!dirtied) 4139 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 4140 } 4141 update_sits_in_cursum(journal, -i); 4142 up_write(&curseg->journal_rwsem); 4143 } 4144 4145 /* 4146 * CP calls this function, which flushes SIT entries including sit_journal, 4147 * and moves prefree segs to free segs. 4148 */ 4149 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 4150 { 4151 struct sit_info *sit_i = SIT_I(sbi); 4152 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 4153 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4154 struct f2fs_journal *journal = curseg->journal; 4155 struct sit_entry_set *ses, *tmp; 4156 struct list_head *head = &SM_I(sbi)->sit_entry_set; 4157 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 4158 struct seg_entry *se; 4159 4160 down_write(&sit_i->sentry_lock); 4161 4162 if (!sit_i->dirty_sentries) 4163 goto out; 4164 4165 /* 4166 * add and account sit entries of dirty bitmap in sit entry 4167 * set temporarily 4168 */ 4169 add_sits_in_set(sbi); 4170 4171 /* 4172 * if there are no enough space in journal to store dirty sit 4173 * entries, remove all entries from journal and add and account 4174 * them in sit entry set. 4175 */ 4176 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 4177 !to_journal) 4178 remove_sits_in_journal(sbi); 4179 4180 /* 4181 * there are two steps to flush sit entries: 4182 * #1, flush sit entries to journal in current cold data summary block. 4183 * #2, flush sit entries to sit page. 4184 */ 4185 list_for_each_entry_safe(ses, tmp, head, set_list) { 4186 struct page *page = NULL; 4187 struct f2fs_sit_block *raw_sit = NULL; 4188 unsigned int start_segno = ses->start_segno; 4189 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 4190 (unsigned long)MAIN_SEGS(sbi)); 4191 unsigned int segno = start_segno; 4192 4193 if (to_journal && 4194 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 4195 to_journal = false; 4196 4197 if (to_journal) { 4198 down_write(&curseg->journal_rwsem); 4199 } else { 4200 page = get_next_sit_page(sbi, start_segno); 4201 raw_sit = page_address(page); 4202 } 4203 4204 /* flush dirty sit entries in region of current sit set */ 4205 for_each_set_bit_from(segno, bitmap, end) { 4206 int offset, sit_offset; 4207 4208 se = get_seg_entry(sbi, segno); 4209 #ifdef CONFIG_F2FS_CHECK_FS 4210 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 4211 SIT_VBLOCK_MAP_SIZE)) 4212 f2fs_bug_on(sbi, 1); 4213 #endif 4214 4215 /* add discard candidates */ 4216 if (!(cpc->reason & CP_DISCARD)) { 4217 cpc->trim_start = segno; 4218 add_discard_addrs(sbi, cpc, false); 4219 } 4220 4221 if (to_journal) { 4222 offset = f2fs_lookup_journal_in_cursum(journal, 4223 SIT_JOURNAL, segno, 1); 4224 f2fs_bug_on(sbi, offset < 0); 4225 segno_in_journal(journal, offset) = 4226 cpu_to_le32(segno); 4227 seg_info_to_raw_sit(se, 4228 &sit_in_journal(journal, offset)); 4229 check_block_count(sbi, segno, 4230 &sit_in_journal(journal, offset)); 4231 } else { 4232 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 4233 seg_info_to_raw_sit(se, 4234 &raw_sit->entries[sit_offset]); 4235 check_block_count(sbi, segno, 4236 &raw_sit->entries[sit_offset]); 4237 } 4238 4239 __clear_bit(segno, bitmap); 4240 sit_i->dirty_sentries--; 4241 ses->entry_cnt--; 4242 } 4243 4244 if (to_journal) 4245 up_write(&curseg->journal_rwsem); 4246 else 4247 f2fs_put_page(page, 1); 4248 4249 f2fs_bug_on(sbi, ses->entry_cnt); 4250 release_sit_entry_set(ses); 4251 } 4252 4253 f2fs_bug_on(sbi, !list_empty(head)); 4254 f2fs_bug_on(sbi, sit_i->dirty_sentries); 4255 out: 4256 if (cpc->reason & CP_DISCARD) { 4257 __u64 trim_start = cpc->trim_start; 4258 4259 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 4260 add_discard_addrs(sbi, cpc, false); 4261 4262 cpc->trim_start = trim_start; 4263 } 4264 up_write(&sit_i->sentry_lock); 4265 4266 set_prefree_as_free_segments(sbi); 4267 } 4268 4269 static int build_sit_info(struct f2fs_sb_info *sbi) 4270 { 4271 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4272 struct sit_info *sit_i; 4273 unsigned int sit_segs, start; 4274 char *src_bitmap, *bitmap; 4275 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 4276 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0; 4277 4278 /* allocate memory for SIT information */ 4279 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4280 if (!sit_i) 4281 return -ENOMEM; 4282 4283 SM_I(sbi)->sit_info = sit_i; 4284 4285 sit_i->sentries = 4286 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4287 MAIN_SEGS(sbi)), 4288 GFP_KERNEL); 4289 if (!sit_i->sentries) 4290 return -ENOMEM; 4291 4292 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4293 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4294 GFP_KERNEL); 4295 if (!sit_i->dirty_sentries_bitmap) 4296 return -ENOMEM; 4297 4298 #ifdef CONFIG_F2FS_CHECK_FS 4299 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map); 4300 #else 4301 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map); 4302 #endif 4303 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4304 if (!sit_i->bitmap) 4305 return -ENOMEM; 4306 4307 bitmap = sit_i->bitmap; 4308 4309 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4310 sit_i->sentries[start].cur_valid_map = bitmap; 4311 bitmap += SIT_VBLOCK_MAP_SIZE; 4312 4313 sit_i->sentries[start].ckpt_valid_map = bitmap; 4314 bitmap += SIT_VBLOCK_MAP_SIZE; 4315 4316 #ifdef CONFIG_F2FS_CHECK_FS 4317 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4318 bitmap += SIT_VBLOCK_MAP_SIZE; 4319 #endif 4320 4321 if (discard_map) { 4322 sit_i->sentries[start].discard_map = bitmap; 4323 bitmap += SIT_VBLOCK_MAP_SIZE; 4324 } 4325 } 4326 4327 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4328 if (!sit_i->tmp_map) 4329 return -ENOMEM; 4330 4331 if (__is_large_section(sbi)) { 4332 sit_i->sec_entries = 4333 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4334 MAIN_SECS(sbi)), 4335 GFP_KERNEL); 4336 if (!sit_i->sec_entries) 4337 return -ENOMEM; 4338 } 4339 4340 /* get information related with SIT */ 4341 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4342 4343 /* setup SIT bitmap from ckeckpoint pack */ 4344 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4345 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4346 4347 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4348 if (!sit_i->sit_bitmap) 4349 return -ENOMEM; 4350 4351 #ifdef CONFIG_F2FS_CHECK_FS 4352 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4353 sit_bitmap_size, GFP_KERNEL); 4354 if (!sit_i->sit_bitmap_mir) 4355 return -ENOMEM; 4356 4357 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4358 main_bitmap_size, GFP_KERNEL); 4359 if (!sit_i->invalid_segmap) 4360 return -ENOMEM; 4361 #endif 4362 4363 /* init SIT information */ 4364 sit_i->s_ops = &default_salloc_ops; 4365 4366 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4367 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 4368 sit_i->written_valid_blocks = 0; 4369 sit_i->bitmap_size = sit_bitmap_size; 4370 sit_i->dirty_sentries = 0; 4371 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4372 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4373 sit_i->mounted_time = ktime_get_boottime_seconds(); 4374 init_rwsem(&sit_i->sentry_lock); 4375 return 0; 4376 } 4377 4378 static int build_free_segmap(struct f2fs_sb_info *sbi) 4379 { 4380 struct free_segmap_info *free_i; 4381 unsigned int bitmap_size, sec_bitmap_size; 4382 4383 /* allocate memory for free segmap information */ 4384 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4385 if (!free_i) 4386 return -ENOMEM; 4387 4388 SM_I(sbi)->free_info = free_i; 4389 4390 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4391 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4392 if (!free_i->free_segmap) 4393 return -ENOMEM; 4394 4395 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4396 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4397 if (!free_i->free_secmap) 4398 return -ENOMEM; 4399 4400 /* set all segments as dirty temporarily */ 4401 memset(free_i->free_segmap, 0xff, bitmap_size); 4402 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4403 4404 /* init free segmap information */ 4405 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4406 free_i->free_segments = 0; 4407 free_i->free_sections = 0; 4408 spin_lock_init(&free_i->segmap_lock); 4409 return 0; 4410 } 4411 4412 static int build_curseg(struct f2fs_sb_info *sbi) 4413 { 4414 struct curseg_info *array; 4415 int i; 4416 4417 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, 4418 sizeof(*array)), GFP_KERNEL); 4419 if (!array) 4420 return -ENOMEM; 4421 4422 SM_I(sbi)->curseg_array = array; 4423 4424 for (i = 0; i < NO_CHECK_TYPE; i++) { 4425 mutex_init(&array[i].curseg_mutex); 4426 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4427 if (!array[i].sum_blk) 4428 return -ENOMEM; 4429 init_rwsem(&array[i].journal_rwsem); 4430 array[i].journal = f2fs_kzalloc(sbi, 4431 sizeof(struct f2fs_journal), GFP_KERNEL); 4432 if (!array[i].journal) 4433 return -ENOMEM; 4434 if (i < NR_PERSISTENT_LOG) 4435 array[i].seg_type = CURSEG_HOT_DATA + i; 4436 else if (i == CURSEG_COLD_DATA_PINNED) 4437 array[i].seg_type = CURSEG_COLD_DATA; 4438 else if (i == CURSEG_ALL_DATA_ATGC) 4439 array[i].seg_type = CURSEG_COLD_DATA; 4440 array[i].segno = NULL_SEGNO; 4441 array[i].next_blkoff = 0; 4442 array[i].inited = false; 4443 } 4444 return restore_curseg_summaries(sbi); 4445 } 4446 4447 static int build_sit_entries(struct f2fs_sb_info *sbi) 4448 { 4449 struct sit_info *sit_i = SIT_I(sbi); 4450 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4451 struct f2fs_journal *journal = curseg->journal; 4452 struct seg_entry *se; 4453 struct f2fs_sit_entry sit; 4454 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4455 unsigned int i, start, end; 4456 unsigned int readed, start_blk = 0; 4457 int err = 0; 4458 block_t total_node_blocks = 0; 4459 4460 do { 4461 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS, 4462 META_SIT, true); 4463 4464 start = start_blk * sit_i->sents_per_block; 4465 end = (start_blk + readed) * sit_i->sents_per_block; 4466 4467 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4468 struct f2fs_sit_block *sit_blk; 4469 struct page *page; 4470 4471 se = &sit_i->sentries[start]; 4472 page = get_current_sit_page(sbi, start); 4473 if (IS_ERR(page)) 4474 return PTR_ERR(page); 4475 sit_blk = (struct f2fs_sit_block *)page_address(page); 4476 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4477 f2fs_put_page(page, 1); 4478 4479 err = check_block_count(sbi, start, &sit); 4480 if (err) 4481 return err; 4482 seg_info_from_raw_sit(se, &sit); 4483 if (IS_NODESEG(se->type)) 4484 total_node_blocks += se->valid_blocks; 4485 4486 if (f2fs_block_unit_discard(sbi)) { 4487 /* build discard map only one time */ 4488 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4489 memset(se->discard_map, 0xff, 4490 SIT_VBLOCK_MAP_SIZE); 4491 } else { 4492 memcpy(se->discard_map, 4493 se->cur_valid_map, 4494 SIT_VBLOCK_MAP_SIZE); 4495 sbi->discard_blks += 4496 sbi->blocks_per_seg - 4497 se->valid_blocks; 4498 } 4499 } 4500 4501 if (__is_large_section(sbi)) 4502 get_sec_entry(sbi, start)->valid_blocks += 4503 se->valid_blocks; 4504 } 4505 start_blk += readed; 4506 } while (start_blk < sit_blk_cnt); 4507 4508 down_read(&curseg->journal_rwsem); 4509 for (i = 0; i < sits_in_cursum(journal); i++) { 4510 unsigned int old_valid_blocks; 4511 4512 start = le32_to_cpu(segno_in_journal(journal, i)); 4513 if (start >= MAIN_SEGS(sbi)) { 4514 f2fs_err(sbi, "Wrong journal entry on segno %u", 4515 start); 4516 err = -EFSCORRUPTED; 4517 break; 4518 } 4519 4520 se = &sit_i->sentries[start]; 4521 sit = sit_in_journal(journal, i); 4522 4523 old_valid_blocks = se->valid_blocks; 4524 if (IS_NODESEG(se->type)) 4525 total_node_blocks -= old_valid_blocks; 4526 4527 err = check_block_count(sbi, start, &sit); 4528 if (err) 4529 break; 4530 seg_info_from_raw_sit(se, &sit); 4531 if (IS_NODESEG(se->type)) 4532 total_node_blocks += se->valid_blocks; 4533 4534 if (f2fs_block_unit_discard(sbi)) { 4535 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4536 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 4537 } else { 4538 memcpy(se->discard_map, se->cur_valid_map, 4539 SIT_VBLOCK_MAP_SIZE); 4540 sbi->discard_blks += old_valid_blocks; 4541 sbi->discard_blks -= se->valid_blocks; 4542 } 4543 } 4544 4545 if (__is_large_section(sbi)) { 4546 get_sec_entry(sbi, start)->valid_blocks += 4547 se->valid_blocks; 4548 get_sec_entry(sbi, start)->valid_blocks -= 4549 old_valid_blocks; 4550 } 4551 } 4552 up_read(&curseg->journal_rwsem); 4553 4554 if (!err && total_node_blocks != valid_node_count(sbi)) { 4555 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 4556 total_node_blocks, valid_node_count(sbi)); 4557 err = -EFSCORRUPTED; 4558 } 4559 4560 return err; 4561 } 4562 4563 static void init_free_segmap(struct f2fs_sb_info *sbi) 4564 { 4565 unsigned int start; 4566 int type; 4567 struct seg_entry *sentry; 4568 4569 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4570 if (f2fs_usable_blks_in_seg(sbi, start) == 0) 4571 continue; 4572 sentry = get_seg_entry(sbi, start); 4573 if (!sentry->valid_blocks) 4574 __set_free(sbi, start); 4575 else 4576 SIT_I(sbi)->written_valid_blocks += 4577 sentry->valid_blocks; 4578 } 4579 4580 /* set use the current segments */ 4581 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 4582 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 4583 4584 __set_test_and_inuse(sbi, curseg_t->segno); 4585 } 4586 } 4587 4588 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 4589 { 4590 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4591 struct free_segmap_info *free_i = FREE_I(sbi); 4592 unsigned int segno = 0, offset = 0, secno; 4593 block_t valid_blocks, usable_blks_in_seg; 4594 block_t blks_per_sec = BLKS_PER_SEC(sbi); 4595 4596 while (1) { 4597 /* find dirty segment based on free segmap */ 4598 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 4599 if (segno >= MAIN_SEGS(sbi)) 4600 break; 4601 offset = segno + 1; 4602 valid_blocks = get_valid_blocks(sbi, segno, false); 4603 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 4604 if (valid_blocks == usable_blks_in_seg || !valid_blocks) 4605 continue; 4606 if (valid_blocks > usable_blks_in_seg) { 4607 f2fs_bug_on(sbi, 1); 4608 continue; 4609 } 4610 mutex_lock(&dirty_i->seglist_lock); 4611 __locate_dirty_segment(sbi, segno, DIRTY); 4612 mutex_unlock(&dirty_i->seglist_lock); 4613 } 4614 4615 if (!__is_large_section(sbi)) 4616 return; 4617 4618 mutex_lock(&dirty_i->seglist_lock); 4619 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 4620 valid_blocks = get_valid_blocks(sbi, segno, true); 4621 secno = GET_SEC_FROM_SEG(sbi, segno); 4622 4623 if (!valid_blocks || valid_blocks == blks_per_sec) 4624 continue; 4625 if (IS_CURSEC(sbi, secno)) 4626 continue; 4627 set_bit(secno, dirty_i->dirty_secmap); 4628 } 4629 mutex_unlock(&dirty_i->seglist_lock); 4630 } 4631 4632 static int init_victim_secmap(struct f2fs_sb_info *sbi) 4633 { 4634 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4635 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4636 4637 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4638 if (!dirty_i->victim_secmap) 4639 return -ENOMEM; 4640 return 0; 4641 } 4642 4643 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 4644 { 4645 struct dirty_seglist_info *dirty_i; 4646 unsigned int bitmap_size, i; 4647 4648 /* allocate memory for dirty segments list information */ 4649 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 4650 GFP_KERNEL); 4651 if (!dirty_i) 4652 return -ENOMEM; 4653 4654 SM_I(sbi)->dirty_info = dirty_i; 4655 mutex_init(&dirty_i->seglist_lock); 4656 4657 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4658 4659 for (i = 0; i < NR_DIRTY_TYPE; i++) { 4660 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 4661 GFP_KERNEL); 4662 if (!dirty_i->dirty_segmap[i]) 4663 return -ENOMEM; 4664 } 4665 4666 if (__is_large_section(sbi)) { 4667 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4668 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi, 4669 bitmap_size, GFP_KERNEL); 4670 if (!dirty_i->dirty_secmap) 4671 return -ENOMEM; 4672 } 4673 4674 init_dirty_segmap(sbi); 4675 return init_victim_secmap(sbi); 4676 } 4677 4678 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 4679 { 4680 int i; 4681 4682 /* 4683 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 4684 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 4685 */ 4686 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 4687 struct curseg_info *curseg = CURSEG_I(sbi, i); 4688 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 4689 unsigned int blkofs = curseg->next_blkoff; 4690 4691 if (f2fs_sb_has_readonly(sbi) && 4692 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE) 4693 continue; 4694 4695 sanity_check_seg_type(sbi, curseg->seg_type); 4696 4697 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) { 4698 f2fs_err(sbi, 4699 "Current segment has invalid alloc_type:%d", 4700 curseg->alloc_type); 4701 return -EFSCORRUPTED; 4702 } 4703 4704 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 4705 goto out; 4706 4707 if (curseg->alloc_type == SSR) 4708 continue; 4709 4710 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) { 4711 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 4712 continue; 4713 out: 4714 f2fs_err(sbi, 4715 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 4716 i, curseg->segno, curseg->alloc_type, 4717 curseg->next_blkoff, blkofs); 4718 return -EFSCORRUPTED; 4719 } 4720 } 4721 return 0; 4722 } 4723 4724 #ifdef CONFIG_BLK_DEV_ZONED 4725 4726 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 4727 struct f2fs_dev_info *fdev, 4728 struct blk_zone *zone) 4729 { 4730 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno; 4731 block_t zone_block, wp_block, last_valid_block; 4732 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4733 int i, s, b, ret; 4734 struct seg_entry *se; 4735 4736 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4737 return 0; 4738 4739 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block); 4740 wp_segno = GET_SEGNO(sbi, wp_block); 4741 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4742 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 4743 zone_segno = GET_SEGNO(sbi, zone_block); 4744 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno); 4745 4746 if (zone_segno >= MAIN_SEGS(sbi)) 4747 return 0; 4748 4749 /* 4750 * Skip check of zones cursegs point to, since 4751 * fix_curseg_write_pointer() checks them. 4752 */ 4753 for (i = 0; i < NO_CHECK_TYPE; i++) 4754 if (zone_secno == GET_SEC_FROM_SEG(sbi, 4755 CURSEG_I(sbi, i)->segno)) 4756 return 0; 4757 4758 /* 4759 * Get last valid block of the zone. 4760 */ 4761 last_valid_block = zone_block - 1; 4762 for (s = sbi->segs_per_sec - 1; s >= 0; s--) { 4763 segno = zone_segno + s; 4764 se = get_seg_entry(sbi, segno); 4765 for (b = sbi->blocks_per_seg - 1; b >= 0; b--) 4766 if (f2fs_test_bit(b, se->cur_valid_map)) { 4767 last_valid_block = START_BLOCK(sbi, segno) + b; 4768 break; 4769 } 4770 if (last_valid_block >= zone_block) 4771 break; 4772 } 4773 4774 /* 4775 * If last valid block is beyond the write pointer, report the 4776 * inconsistency. This inconsistency does not cause write error 4777 * because the zone will not be selected for write operation until 4778 * it get discarded. Just report it. 4779 */ 4780 if (last_valid_block >= wp_block) { 4781 f2fs_notice(sbi, "Valid block beyond write pointer: " 4782 "valid block[0x%x,0x%x] wp[0x%x,0x%x]", 4783 GET_SEGNO(sbi, last_valid_block), 4784 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block), 4785 wp_segno, wp_blkoff); 4786 return 0; 4787 } 4788 4789 /* 4790 * If there is no valid block in the zone and if write pointer is 4791 * not at zone start, reset the write pointer. 4792 */ 4793 if (last_valid_block + 1 == zone_block && zone->wp != zone->start) { 4794 f2fs_notice(sbi, 4795 "Zone without valid block has non-zero write " 4796 "pointer. Reset the write pointer: wp[0x%x,0x%x]", 4797 wp_segno, wp_blkoff); 4798 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 4799 zone->len >> log_sectors_per_block); 4800 if (ret) { 4801 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4802 fdev->path, ret); 4803 return ret; 4804 } 4805 } 4806 4807 return 0; 4808 } 4809 4810 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 4811 block_t zone_blkaddr) 4812 { 4813 int i; 4814 4815 for (i = 0; i < sbi->s_ndevs; i++) { 4816 if (!bdev_is_zoned(FDEV(i).bdev)) 4817 continue; 4818 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 4819 zone_blkaddr <= FDEV(i).end_blk)) 4820 return &FDEV(i); 4821 } 4822 4823 return NULL; 4824 } 4825 4826 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 4827 void *data) 4828 { 4829 memcpy(data, zone, sizeof(struct blk_zone)); 4830 return 0; 4831 } 4832 4833 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 4834 { 4835 struct curseg_info *cs = CURSEG_I(sbi, type); 4836 struct f2fs_dev_info *zbd; 4837 struct blk_zone zone; 4838 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 4839 block_t cs_zone_block, wp_block; 4840 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4841 sector_t zone_sector; 4842 int err; 4843 4844 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4845 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4846 4847 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4848 if (!zbd) 4849 return 0; 4850 4851 /* report zone for the sector the curseg points to */ 4852 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4853 << log_sectors_per_block; 4854 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4855 report_one_zone_cb, &zone); 4856 if (err != 1) { 4857 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4858 zbd->path, err); 4859 return err; 4860 } 4861 4862 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4863 return 0; 4864 4865 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 4866 wp_segno = GET_SEGNO(sbi, wp_block); 4867 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4868 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 4869 4870 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 4871 wp_sector_off == 0) 4872 return 0; 4873 4874 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 4875 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", 4876 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff); 4877 4878 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 4879 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff); 4880 4881 f2fs_allocate_new_section(sbi, type, true); 4882 4883 /* check consistency of the zone curseg pointed to */ 4884 if (check_zone_write_pointer(sbi, zbd, &zone)) 4885 return -EIO; 4886 4887 /* check newly assigned zone */ 4888 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4889 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4890 4891 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4892 if (!zbd) 4893 return 0; 4894 4895 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4896 << log_sectors_per_block; 4897 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4898 report_one_zone_cb, &zone); 4899 if (err != 1) { 4900 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4901 zbd->path, err); 4902 return err; 4903 } 4904 4905 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4906 return 0; 4907 4908 if (zone.wp != zone.start) { 4909 f2fs_notice(sbi, 4910 "New zone for curseg[%d] is not yet discarded. " 4911 "Reset the zone: curseg[0x%x,0x%x]", 4912 type, cs->segno, cs->next_blkoff); 4913 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, 4914 zone_sector >> log_sectors_per_block, 4915 zone.len >> log_sectors_per_block); 4916 if (err) { 4917 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4918 zbd->path, err); 4919 return err; 4920 } 4921 } 4922 4923 return 0; 4924 } 4925 4926 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 4927 { 4928 int i, ret; 4929 4930 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 4931 ret = fix_curseg_write_pointer(sbi, i); 4932 if (ret) 4933 return ret; 4934 } 4935 4936 return 0; 4937 } 4938 4939 struct check_zone_write_pointer_args { 4940 struct f2fs_sb_info *sbi; 4941 struct f2fs_dev_info *fdev; 4942 }; 4943 4944 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 4945 void *data) 4946 { 4947 struct check_zone_write_pointer_args *args; 4948 4949 args = (struct check_zone_write_pointer_args *)data; 4950 4951 return check_zone_write_pointer(args->sbi, args->fdev, zone); 4952 } 4953 4954 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 4955 { 4956 int i, ret; 4957 struct check_zone_write_pointer_args args; 4958 4959 for (i = 0; i < sbi->s_ndevs; i++) { 4960 if (!bdev_is_zoned(FDEV(i).bdev)) 4961 continue; 4962 4963 args.sbi = sbi; 4964 args.fdev = &FDEV(i); 4965 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 4966 check_zone_write_pointer_cb, &args); 4967 if (ret < 0) 4968 return ret; 4969 } 4970 4971 return 0; 4972 } 4973 4974 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx, 4975 unsigned int dev_idx) 4976 { 4977 if (!bdev_is_zoned(FDEV(dev_idx).bdev)) 4978 return true; 4979 return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq); 4980 } 4981 4982 /* Return the zone index in the given device */ 4983 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno, 4984 int dev_idx) 4985 { 4986 block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 4987 4988 return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >> 4989 sbi->log_blocks_per_blkz; 4990 } 4991 4992 /* 4993 * Return the usable segments in a section based on the zone's 4994 * corresponding zone capacity. Zone is equal to a section. 4995 */ 4996 static inline unsigned int f2fs_usable_zone_segs_in_sec( 4997 struct f2fs_sb_info *sbi, unsigned int segno) 4998 { 4999 unsigned int dev_idx, zone_idx, unusable_segs_in_sec; 5000 5001 dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno)); 5002 zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx); 5003 5004 /* Conventional zone's capacity is always equal to zone size */ 5005 if (is_conv_zone(sbi, zone_idx, dev_idx)) 5006 return sbi->segs_per_sec; 5007 5008 /* 5009 * If the zone_capacity_blocks array is NULL, then zone capacity 5010 * is equal to the zone size for all zones 5011 */ 5012 if (!FDEV(dev_idx).zone_capacity_blocks) 5013 return sbi->segs_per_sec; 5014 5015 /* Get the segment count beyond zone capacity block */ 5016 unusable_segs_in_sec = (sbi->blocks_per_blkz - 5017 FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >> 5018 sbi->log_blocks_per_seg; 5019 return sbi->segs_per_sec - unusable_segs_in_sec; 5020 } 5021 5022 /* 5023 * Return the number of usable blocks in a segment. The number of blocks 5024 * returned is always equal to the number of blocks in a segment for 5025 * segments fully contained within a sequential zone capacity or a 5026 * conventional zone. For segments partially contained in a sequential 5027 * zone capacity, the number of usable blocks up to the zone capacity 5028 * is returned. 0 is returned in all other cases. 5029 */ 5030 static inline unsigned int f2fs_usable_zone_blks_in_seg( 5031 struct f2fs_sb_info *sbi, unsigned int segno) 5032 { 5033 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr; 5034 unsigned int zone_idx, dev_idx, secno; 5035 5036 secno = GET_SEC_FROM_SEG(sbi, segno); 5037 seg_start = START_BLOCK(sbi, segno); 5038 dev_idx = f2fs_target_device_index(sbi, seg_start); 5039 zone_idx = get_zone_idx(sbi, secno, dev_idx); 5040 5041 /* 5042 * Conventional zone's capacity is always equal to zone size, 5043 * so, blocks per segment is unchanged. 5044 */ 5045 if (is_conv_zone(sbi, zone_idx, dev_idx)) 5046 return sbi->blocks_per_seg; 5047 5048 if (!FDEV(dev_idx).zone_capacity_blocks) 5049 return sbi->blocks_per_seg; 5050 5051 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5052 sec_cap_blkaddr = sec_start_blkaddr + 5053 FDEV(dev_idx).zone_capacity_blocks[zone_idx]; 5054 5055 /* 5056 * If segment starts before zone capacity and spans beyond 5057 * zone capacity, then usable blocks are from seg start to 5058 * zone capacity. If the segment starts after the zone capacity, 5059 * then there are no usable blocks. 5060 */ 5061 if (seg_start >= sec_cap_blkaddr) 5062 return 0; 5063 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr) 5064 return sec_cap_blkaddr - seg_start; 5065 5066 return sbi->blocks_per_seg; 5067 } 5068 #else 5069 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5070 { 5071 return 0; 5072 } 5073 5074 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5075 { 5076 return 0; 5077 } 5078 5079 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi, 5080 unsigned int segno) 5081 { 5082 return 0; 5083 } 5084 5085 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi, 5086 unsigned int segno) 5087 { 5088 return 0; 5089 } 5090 #endif 5091 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi, 5092 unsigned int segno) 5093 { 5094 if (f2fs_sb_has_blkzoned(sbi)) 5095 return f2fs_usable_zone_blks_in_seg(sbi, segno); 5096 5097 return sbi->blocks_per_seg; 5098 } 5099 5100 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi, 5101 unsigned int segno) 5102 { 5103 if (f2fs_sb_has_blkzoned(sbi)) 5104 return f2fs_usable_zone_segs_in_sec(sbi, segno); 5105 5106 return sbi->segs_per_sec; 5107 } 5108 5109 /* 5110 * Update min, max modified time for cost-benefit GC algorithm 5111 */ 5112 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 5113 { 5114 struct sit_info *sit_i = SIT_I(sbi); 5115 unsigned int segno; 5116 5117 down_write(&sit_i->sentry_lock); 5118 5119 sit_i->min_mtime = ULLONG_MAX; 5120 5121 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 5122 unsigned int i; 5123 unsigned long long mtime = 0; 5124 5125 for (i = 0; i < sbi->segs_per_sec; i++) 5126 mtime += get_seg_entry(sbi, segno + i)->mtime; 5127 5128 mtime = div_u64(mtime, sbi->segs_per_sec); 5129 5130 if (sit_i->min_mtime > mtime) 5131 sit_i->min_mtime = mtime; 5132 } 5133 sit_i->max_mtime = get_mtime(sbi, false); 5134 sit_i->dirty_max_mtime = 0; 5135 up_write(&sit_i->sentry_lock); 5136 } 5137 5138 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 5139 { 5140 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 5141 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 5142 struct f2fs_sm_info *sm_info; 5143 int err; 5144 5145 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 5146 if (!sm_info) 5147 return -ENOMEM; 5148 5149 /* init sm info */ 5150 sbi->sm_info = sm_info; 5151 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 5152 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 5153 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 5154 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 5155 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 5156 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 5157 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 5158 sm_info->rec_prefree_segments = sm_info->main_segments * 5159 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 5160 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 5161 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 5162 5163 if (!f2fs_lfs_mode(sbi)) 5164 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 5165 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 5166 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 5167 sm_info->min_seq_blocks = sbi->blocks_per_seg; 5168 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 5169 sm_info->min_ssr_sections = reserved_sections(sbi); 5170 5171 INIT_LIST_HEAD(&sm_info->sit_entry_set); 5172 5173 init_f2fs_rwsem(&sm_info->curseg_lock); 5174 5175 if (!f2fs_readonly(sbi->sb)) { 5176 err = f2fs_create_flush_cmd_control(sbi); 5177 if (err) 5178 return err; 5179 } 5180 5181 err = create_discard_cmd_control(sbi); 5182 if (err) 5183 return err; 5184 5185 err = build_sit_info(sbi); 5186 if (err) 5187 return err; 5188 err = build_free_segmap(sbi); 5189 if (err) 5190 return err; 5191 err = build_curseg(sbi); 5192 if (err) 5193 return err; 5194 5195 /* reinit free segmap based on SIT */ 5196 err = build_sit_entries(sbi); 5197 if (err) 5198 return err; 5199 5200 init_free_segmap(sbi); 5201 err = build_dirty_segmap(sbi); 5202 if (err) 5203 return err; 5204 5205 err = sanity_check_curseg(sbi); 5206 if (err) 5207 return err; 5208 5209 init_min_max_mtime(sbi); 5210 return 0; 5211 } 5212 5213 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 5214 enum dirty_type dirty_type) 5215 { 5216 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5217 5218 mutex_lock(&dirty_i->seglist_lock); 5219 kvfree(dirty_i->dirty_segmap[dirty_type]); 5220 dirty_i->nr_dirty[dirty_type] = 0; 5221 mutex_unlock(&dirty_i->seglist_lock); 5222 } 5223 5224 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 5225 { 5226 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5227 5228 kvfree(dirty_i->victim_secmap); 5229 } 5230 5231 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 5232 { 5233 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5234 int i; 5235 5236 if (!dirty_i) 5237 return; 5238 5239 /* discard pre-free/dirty segments list */ 5240 for (i = 0; i < NR_DIRTY_TYPE; i++) 5241 discard_dirty_segmap(sbi, i); 5242 5243 if (__is_large_section(sbi)) { 5244 mutex_lock(&dirty_i->seglist_lock); 5245 kvfree(dirty_i->dirty_secmap); 5246 mutex_unlock(&dirty_i->seglist_lock); 5247 } 5248 5249 destroy_victim_secmap(sbi); 5250 SM_I(sbi)->dirty_info = NULL; 5251 kfree(dirty_i); 5252 } 5253 5254 static void destroy_curseg(struct f2fs_sb_info *sbi) 5255 { 5256 struct curseg_info *array = SM_I(sbi)->curseg_array; 5257 int i; 5258 5259 if (!array) 5260 return; 5261 SM_I(sbi)->curseg_array = NULL; 5262 for (i = 0; i < NR_CURSEG_TYPE; i++) { 5263 kfree(array[i].sum_blk); 5264 kfree(array[i].journal); 5265 } 5266 kfree(array); 5267 } 5268 5269 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 5270 { 5271 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 5272 5273 if (!free_i) 5274 return; 5275 SM_I(sbi)->free_info = NULL; 5276 kvfree(free_i->free_segmap); 5277 kvfree(free_i->free_secmap); 5278 kfree(free_i); 5279 } 5280 5281 static void destroy_sit_info(struct f2fs_sb_info *sbi) 5282 { 5283 struct sit_info *sit_i = SIT_I(sbi); 5284 5285 if (!sit_i) 5286 return; 5287 5288 if (sit_i->sentries) 5289 kvfree(sit_i->bitmap); 5290 kfree(sit_i->tmp_map); 5291 5292 kvfree(sit_i->sentries); 5293 kvfree(sit_i->sec_entries); 5294 kvfree(sit_i->dirty_sentries_bitmap); 5295 5296 SM_I(sbi)->sit_info = NULL; 5297 kvfree(sit_i->sit_bitmap); 5298 #ifdef CONFIG_F2FS_CHECK_FS 5299 kvfree(sit_i->sit_bitmap_mir); 5300 kvfree(sit_i->invalid_segmap); 5301 #endif 5302 kfree(sit_i); 5303 } 5304 5305 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 5306 { 5307 struct f2fs_sm_info *sm_info = SM_I(sbi); 5308 5309 if (!sm_info) 5310 return; 5311 f2fs_destroy_flush_cmd_control(sbi, true); 5312 destroy_discard_cmd_control(sbi); 5313 destroy_dirty_segmap(sbi); 5314 destroy_curseg(sbi); 5315 destroy_free_segmap(sbi); 5316 destroy_sit_info(sbi); 5317 sbi->sm_info = NULL; 5318 kfree(sm_info); 5319 } 5320 5321 int __init f2fs_create_segment_manager_caches(void) 5322 { 5323 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 5324 sizeof(struct discard_entry)); 5325 if (!discard_entry_slab) 5326 goto fail; 5327 5328 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 5329 sizeof(struct discard_cmd)); 5330 if (!discard_cmd_slab) 5331 goto destroy_discard_entry; 5332 5333 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 5334 sizeof(struct sit_entry_set)); 5335 if (!sit_entry_set_slab) 5336 goto destroy_discard_cmd; 5337 5338 inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry", 5339 sizeof(struct inmem_pages)); 5340 if (!inmem_entry_slab) 5341 goto destroy_sit_entry_set; 5342 return 0; 5343 5344 destroy_sit_entry_set: 5345 kmem_cache_destroy(sit_entry_set_slab); 5346 destroy_discard_cmd: 5347 kmem_cache_destroy(discard_cmd_slab); 5348 destroy_discard_entry: 5349 kmem_cache_destroy(discard_entry_slab); 5350 fail: 5351 return -ENOMEM; 5352 } 5353 5354 void f2fs_destroy_segment_manager_caches(void) 5355 { 5356 kmem_cache_destroy(sit_entry_set_slab); 5357 kmem_cache_destroy(discard_cmd_slab); 5358 kmem_cache_destroy(discard_entry_slab); 5359 kmem_cache_destroy(inmem_entry_slab); 5360 } 5361