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 *revoke_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_abort_atomic_write(struct inode *inode, bool clean) 189 { 190 struct f2fs_inode_info *fi = F2FS_I(inode); 191 192 if (!f2fs_is_atomic_file(inode)) 193 return; 194 195 if (clean) 196 truncate_inode_pages_final(inode->i_mapping); 197 198 release_atomic_write_cnt(inode); 199 clear_inode_flag(inode, FI_ATOMIC_COMMITTED); 200 clear_inode_flag(inode, FI_ATOMIC_REPLACE); 201 clear_inode_flag(inode, FI_ATOMIC_FILE); 202 if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) { 203 clear_inode_flag(inode, FI_ATOMIC_DIRTIED); 204 /* 205 * The vfs inode keeps clean during commit, but the f2fs inode 206 * doesn't. So clear the dirty state after commit and let 207 * f2fs_mark_inode_dirty_sync ensure a consistent dirty state. 208 */ 209 f2fs_inode_synced(inode); 210 f2fs_mark_inode_dirty_sync(inode, true); 211 } 212 stat_dec_atomic_inode(inode); 213 214 F2FS_I(inode)->atomic_write_task = NULL; 215 216 if (clean) { 217 f2fs_i_size_write(inode, fi->original_i_size); 218 fi->original_i_size = 0; 219 } 220 /* avoid stale dirty inode during eviction */ 221 sync_inode_metadata(inode, 0); 222 } 223 224 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index, 225 block_t new_addr, block_t *old_addr, bool recover) 226 { 227 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 228 struct dnode_of_data dn; 229 struct node_info ni; 230 int err; 231 232 retry: 233 set_new_dnode(&dn, inode, NULL, NULL, 0); 234 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE); 235 if (err) { 236 if (err == -ENOMEM) { 237 memalloc_retry_wait(GFP_NOFS); 238 goto retry; 239 } 240 return err; 241 } 242 243 err = f2fs_get_node_info(sbi, dn.nid, &ni, false); 244 if (err) { 245 f2fs_put_dnode(&dn); 246 return err; 247 } 248 249 if (recover) { 250 /* dn.data_blkaddr is always valid */ 251 if (!__is_valid_data_blkaddr(new_addr)) { 252 if (new_addr == NULL_ADDR) 253 dec_valid_block_count(sbi, inode, 1); 254 f2fs_invalidate_blocks(sbi, dn.data_blkaddr, 1); 255 f2fs_update_data_blkaddr(&dn, new_addr); 256 } else { 257 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 258 new_addr, ni.version, true, true); 259 } 260 } else { 261 blkcnt_t count = 1; 262 263 err = inc_valid_block_count(sbi, inode, &count, true); 264 if (err) { 265 f2fs_put_dnode(&dn); 266 return err; 267 } 268 269 *old_addr = dn.data_blkaddr; 270 f2fs_truncate_data_blocks_range(&dn, 1); 271 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count); 272 273 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr, 274 ni.version, true, false); 275 } 276 277 f2fs_put_dnode(&dn); 278 279 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode, 280 index, old_addr ? *old_addr : 0, new_addr, recover); 281 return 0; 282 } 283 284 static void __complete_revoke_list(struct inode *inode, struct list_head *head, 285 bool revoke) 286 { 287 struct revoke_entry *cur, *tmp; 288 pgoff_t start_index = 0; 289 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE); 290 291 list_for_each_entry_safe(cur, tmp, head, list) { 292 if (revoke) { 293 __replace_atomic_write_block(inode, cur->index, 294 cur->old_addr, NULL, true); 295 } else if (truncate) { 296 f2fs_truncate_hole(inode, start_index, cur->index); 297 start_index = cur->index + 1; 298 } 299 300 list_del(&cur->list); 301 kmem_cache_free(revoke_entry_slab, cur); 302 } 303 304 if (!revoke && truncate) 305 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false); 306 } 307 308 static int __f2fs_commit_atomic_write(struct inode *inode) 309 { 310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 311 struct f2fs_inode_info *fi = F2FS_I(inode); 312 struct inode *cow_inode = fi->cow_inode; 313 struct revoke_entry *new; 314 struct list_head revoke_list; 315 block_t blkaddr; 316 struct dnode_of_data dn; 317 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 318 pgoff_t off = 0, blen, index; 319 int ret = 0, i; 320 321 INIT_LIST_HEAD(&revoke_list); 322 323 while (len) { 324 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len); 325 326 set_new_dnode(&dn, cow_inode, NULL, NULL, 0); 327 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA); 328 if (ret && ret != -ENOENT) { 329 goto out; 330 } else if (ret == -ENOENT) { 331 ret = 0; 332 if (dn.max_level == 0) 333 goto out; 334 goto next; 335 } 336 337 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_folio, cow_inode), 338 len); 339 index = off; 340 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) { 341 blkaddr = f2fs_data_blkaddr(&dn); 342 343 if (!__is_valid_data_blkaddr(blkaddr)) { 344 continue; 345 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 346 DATA_GENERIC_ENHANCE)) { 347 f2fs_put_dnode(&dn); 348 ret = -EFSCORRUPTED; 349 goto out; 350 } 351 352 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS, 353 true, NULL); 354 355 ret = __replace_atomic_write_block(inode, index, blkaddr, 356 &new->old_addr, false); 357 if (ret) { 358 f2fs_put_dnode(&dn); 359 kmem_cache_free(revoke_entry_slab, new); 360 goto out; 361 } 362 363 f2fs_update_data_blkaddr(&dn, NULL_ADDR); 364 new->index = index; 365 list_add_tail(&new->list, &revoke_list); 366 } 367 f2fs_put_dnode(&dn); 368 next: 369 off += blen; 370 len -= blen; 371 } 372 373 out: 374 if (time_to_inject(sbi, FAULT_ATOMIC_TIMEOUT)) 375 f2fs_schedule_timeout_killable(DEFAULT_FAULT_TIMEOUT, true); 376 377 if (ret) { 378 sbi->revoked_atomic_block += fi->atomic_write_cnt; 379 } else { 380 sbi->committed_atomic_block += fi->atomic_write_cnt; 381 set_inode_flag(inode, FI_ATOMIC_COMMITTED); 382 383 /* 384 * inode may has no FI_ATOMIC_DIRTIED flag due to no write 385 * before commit. 386 */ 387 if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) { 388 /* clear atomic dirty status and set vfs dirty status */ 389 clear_inode_flag(inode, FI_ATOMIC_DIRTIED); 390 f2fs_mark_inode_dirty_sync(inode, true); 391 } 392 } 393 394 __complete_revoke_list(inode, &revoke_list, ret ? true : false); 395 396 return ret; 397 } 398 399 int f2fs_commit_atomic_write(struct inode *inode) 400 { 401 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 402 struct f2fs_inode_info *fi = F2FS_I(inode); 403 struct f2fs_lock_context lc; 404 int err; 405 406 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 407 if (err) 408 return err; 409 410 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 411 f2fs_lock_op(sbi, &lc); 412 413 err = __f2fs_commit_atomic_write(inode); 414 415 f2fs_unlock_op(sbi, &lc); 416 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 417 418 return err; 419 } 420 421 /* 422 * This function balances dirty node and dentry pages. 423 * In addition, it controls garbage collection. 424 */ 425 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 426 { 427 if (f2fs_cp_error(sbi)) 428 return; 429 430 if (time_to_inject(sbi, FAULT_CHECKPOINT)) 431 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT); 432 433 /* balance_fs_bg is able to be pending */ 434 if (need && excess_cached_nats(sbi)) 435 f2fs_balance_fs_bg(sbi, false); 436 437 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 438 return; 439 440 /* 441 * We should do GC or end up with checkpoint, if there are so many dirty 442 * dir/node pages without enough free segments. 443 */ 444 if (has_enough_free_secs(sbi, 0, 0)) 445 return; 446 447 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread && 448 sbi->gc_thread->f2fs_gc_task) { 449 DEFINE_WAIT(wait); 450 451 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait, 452 TASK_UNINTERRUPTIBLE); 453 wake_up(&sbi->gc_thread->gc_wait_queue_head); 454 io_schedule(); 455 finish_wait(&sbi->gc_thread->fggc_wq, &wait); 456 } else { 457 struct f2fs_gc_control gc_control = { 458 .victim_segno = NULL_SEGNO, 459 .init_gc_type = f2fs_sb_has_blkzoned(sbi) ? 460 FG_GC : BG_GC, 461 .no_bg_gc = true, 462 .should_migrate_blocks = false, 463 .err_gc_skipped = false, 464 .nr_free_secs = 1 }; 465 f2fs_down_write_trace(&sbi->gc_lock, &gc_control.lc); 466 stat_inc_gc_call_count(sbi, FOREGROUND); 467 f2fs_gc(sbi, &gc_control); 468 } 469 } 470 471 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi) 472 { 473 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2; 474 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS); 475 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA); 476 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES); 477 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META); 478 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA); 479 unsigned int threshold = 480 SEGS_TO_BLKS(sbi, (factor * DEFAULT_DIRTY_THRESHOLD)); 481 unsigned int global_threshold = threshold * 3 / 2; 482 483 if (dents >= threshold || qdata >= threshold || 484 nodes >= threshold || meta >= threshold || 485 imeta >= threshold) 486 return true; 487 return dents + qdata + nodes + meta + imeta > global_threshold; 488 } 489 490 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 491 { 492 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 493 return; 494 495 /* try to shrink extent cache when there is no enough memory */ 496 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE)) 497 f2fs_shrink_read_extent_tree(sbi, 498 READ_EXTENT_CACHE_SHRINK_NUMBER); 499 500 /* try to shrink age extent cache when there is no enough memory */ 501 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE)) 502 f2fs_shrink_age_extent_tree(sbi, 503 AGE_EXTENT_CACHE_SHRINK_NUMBER); 504 505 /* check the # of cached NAT entries */ 506 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 507 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 508 509 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 510 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 511 else 512 f2fs_build_free_nids(sbi, false, false); 513 514 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) || 515 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi)) 516 goto do_sync; 517 518 /* there is background inflight IO or foreground operation recently */ 519 if (is_inflight_io(sbi, REQ_TIME) || 520 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem))) 521 return; 522 523 /* exceed periodical checkpoint timeout threshold */ 524 if (f2fs_time_over(sbi, CP_TIME)) 525 goto do_sync; 526 527 /* checkpoint is the only way to shrink partial cached entries */ 528 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) && 529 f2fs_available_free_memory(sbi, INO_ENTRIES)) 530 return; 531 532 do_sync: 533 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 534 struct blk_plug plug; 535 536 mutex_lock(&sbi->flush_lock); 537 538 blk_start_plug(&plug); 539 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false); 540 blk_finish_plug(&plug); 541 542 mutex_unlock(&sbi->flush_lock); 543 } 544 stat_inc_cp_call_count(sbi, BACKGROUND); 545 f2fs_sync_fs(sbi->sb, 1); 546 } 547 548 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 549 struct block_device *bdev) 550 { 551 int ret = blkdev_issue_flush(bdev); 552 553 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 554 test_opt(sbi, FLUSH_MERGE), ret); 555 if (!ret) 556 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0); 557 return ret; 558 } 559 560 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 561 { 562 int ret = 0; 563 int i; 564 565 if (!f2fs_is_multi_device(sbi)) 566 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 567 568 for (i = 0; i < sbi->s_ndevs; i++) { 569 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 570 continue; 571 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 572 if (ret) 573 break; 574 } 575 return ret; 576 } 577 578 static int issue_flush_thread(void *data) 579 { 580 struct f2fs_sb_info *sbi = data; 581 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 582 wait_queue_head_t *q = &fcc->flush_wait_queue; 583 repeat: 584 if (kthread_should_stop()) 585 return 0; 586 587 if (!llist_empty(&fcc->issue_list)) { 588 struct flush_cmd *cmd, *next; 589 int ret; 590 591 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 592 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 593 594 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 595 596 ret = submit_flush_wait(sbi, cmd->ino); 597 atomic_inc(&fcc->issued_flush); 598 599 llist_for_each_entry_safe(cmd, next, 600 fcc->dispatch_list, llnode) { 601 cmd->ret = ret; 602 complete(&cmd->wait); 603 } 604 fcc->dispatch_list = NULL; 605 } 606 607 wait_event_interruptible(*q, 608 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 609 goto repeat; 610 } 611 612 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 613 { 614 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 615 struct flush_cmd cmd; 616 int ret; 617 618 if (test_opt(sbi, NOBARRIER)) 619 return 0; 620 621 if (!test_opt(sbi, FLUSH_MERGE)) { 622 atomic_inc(&fcc->queued_flush); 623 ret = submit_flush_wait(sbi, ino); 624 atomic_dec(&fcc->queued_flush); 625 atomic_inc(&fcc->issued_flush); 626 return ret; 627 } 628 629 if (atomic_inc_return(&fcc->queued_flush) == 1 || 630 f2fs_is_multi_device(sbi)) { 631 ret = submit_flush_wait(sbi, ino); 632 atomic_dec(&fcc->queued_flush); 633 634 atomic_inc(&fcc->issued_flush); 635 return ret; 636 } 637 638 cmd.ino = ino; 639 init_completion(&cmd.wait); 640 641 llist_add(&cmd.llnode, &fcc->issue_list); 642 643 /* 644 * update issue_list before we wake up issue_flush thread, this 645 * smp_mb() pairs with another barrier in ___wait_event(), see 646 * more details in comments of waitqueue_active(). 647 */ 648 smp_mb(); 649 650 if (waitqueue_active(&fcc->flush_wait_queue)) 651 wake_up(&fcc->flush_wait_queue); 652 653 if (fcc->f2fs_issue_flush) { 654 wait_for_completion(&cmd.wait); 655 atomic_dec(&fcc->queued_flush); 656 } else { 657 struct llist_node *list; 658 659 list = llist_del_all(&fcc->issue_list); 660 if (!list) { 661 wait_for_completion(&cmd.wait); 662 atomic_dec(&fcc->queued_flush); 663 } else { 664 struct flush_cmd *tmp, *next; 665 666 ret = submit_flush_wait(sbi, ino); 667 668 llist_for_each_entry_safe(tmp, next, list, llnode) { 669 if (tmp == &cmd) { 670 cmd.ret = ret; 671 atomic_dec(&fcc->queued_flush); 672 continue; 673 } 674 tmp->ret = ret; 675 complete(&tmp->wait); 676 } 677 } 678 } 679 680 return cmd.ret; 681 } 682 683 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 684 { 685 dev_t dev = sbi->sb->s_bdev->bd_dev; 686 struct flush_cmd_control *fcc; 687 688 if (SM_I(sbi)->fcc_info) { 689 fcc = SM_I(sbi)->fcc_info; 690 if (fcc->f2fs_issue_flush) 691 return 0; 692 goto init_thread; 693 } 694 695 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 696 if (!fcc) 697 return -ENOMEM; 698 atomic_set(&fcc->issued_flush, 0); 699 atomic_set(&fcc->queued_flush, 0); 700 init_waitqueue_head(&fcc->flush_wait_queue); 701 init_llist_head(&fcc->issue_list); 702 SM_I(sbi)->fcc_info = fcc; 703 if (!test_opt(sbi, FLUSH_MERGE)) 704 return 0; 705 706 init_thread: 707 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 708 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 709 if (IS_ERR(fcc->f2fs_issue_flush)) { 710 int err = PTR_ERR(fcc->f2fs_issue_flush); 711 712 fcc->f2fs_issue_flush = NULL; 713 return err; 714 } 715 716 return 0; 717 } 718 719 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 720 { 721 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 722 723 if (fcc && fcc->f2fs_issue_flush) { 724 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 725 726 fcc->f2fs_issue_flush = NULL; 727 kthread_stop(flush_thread); 728 } 729 if (free) { 730 kfree(fcc); 731 SM_I(sbi)->fcc_info = NULL; 732 } 733 } 734 735 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 736 { 737 int ret = 0, i; 738 739 if (!f2fs_is_multi_device(sbi)) 740 return 0; 741 742 if (test_opt(sbi, NOBARRIER)) 743 return 0; 744 745 for (i = 1; i < sbi->s_ndevs; i++) { 746 int count = DEFAULT_RETRY_IO_COUNT; 747 748 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 749 continue; 750 751 do { 752 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 753 if (ret) 754 f2fs_schedule_timeout(DEFAULT_SCHEDULE_TIMEOUT); 755 } while (ret && --count); 756 757 if (ret) { 758 f2fs_stop_checkpoint(sbi, false, 759 STOP_CP_REASON_FLUSH_FAIL); 760 break; 761 } 762 763 spin_lock(&sbi->dev_lock); 764 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 765 spin_unlock(&sbi->dev_lock); 766 } 767 768 return ret; 769 } 770 771 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 772 enum dirty_type dirty_type) 773 { 774 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 775 776 /* need not be added */ 777 if (is_curseg(sbi, segno)) 778 return; 779 780 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 781 dirty_i->nr_dirty[dirty_type]++; 782 783 if (dirty_type == DIRTY) { 784 struct seg_entry *sentry = get_seg_entry(sbi, segno); 785 enum dirty_type t = sentry->type; 786 787 if (unlikely(t >= DIRTY)) { 788 f2fs_bug_on(sbi, 1); 789 return; 790 } 791 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 792 dirty_i->nr_dirty[t]++; 793 794 if (__is_large_section(sbi)) { 795 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 796 block_t valid_blocks = 797 get_valid_blocks(sbi, segno, true); 798 799 f2fs_bug_on(sbi, 800 (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 801 !valid_blocks) || 802 valid_blocks == CAP_BLKS_PER_SEC(sbi)); 803 804 if (!is_cursec(sbi, secno)) 805 set_bit(secno, dirty_i->dirty_secmap); 806 } 807 } 808 } 809 810 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 811 enum dirty_type dirty_type) 812 { 813 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 814 block_t valid_blocks; 815 816 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 817 dirty_i->nr_dirty[dirty_type]--; 818 819 if (dirty_type == DIRTY) { 820 struct seg_entry *sentry = get_seg_entry(sbi, segno); 821 enum dirty_type t = sentry->type; 822 823 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 824 dirty_i->nr_dirty[t]--; 825 826 valid_blocks = get_valid_blocks(sbi, segno, true); 827 if (valid_blocks == 0) { 828 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 829 dirty_i->victim_secmap); 830 #ifdef CONFIG_F2FS_CHECK_FS 831 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 832 #endif 833 } 834 if (__is_large_section(sbi)) { 835 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 836 837 if (!valid_blocks || 838 valid_blocks == CAP_BLKS_PER_SEC(sbi)) { 839 clear_bit(secno, dirty_i->dirty_secmap); 840 return; 841 } 842 843 if (!is_cursec(sbi, secno)) 844 set_bit(secno, dirty_i->dirty_secmap); 845 } 846 } 847 } 848 849 /* 850 * Should not occur error such as -ENOMEM. 851 * Adding dirty entry into seglist is not critical operation. 852 * If a given segment is one of current working segments, it won't be added. 853 */ 854 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 855 { 856 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 857 unsigned short valid_blocks, ckpt_valid_blocks; 858 unsigned int usable_blocks; 859 860 if (segno == NULL_SEGNO || is_curseg(sbi, segno)) 861 return; 862 863 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno); 864 mutex_lock(&dirty_i->seglist_lock); 865 866 valid_blocks = get_valid_blocks(sbi, segno, false); 867 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false); 868 869 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 870 ckpt_valid_blocks == usable_blocks)) { 871 __locate_dirty_segment(sbi, segno, PRE); 872 __remove_dirty_segment(sbi, segno, DIRTY); 873 } else if (valid_blocks < usable_blocks) { 874 __locate_dirty_segment(sbi, segno, DIRTY); 875 } else { 876 /* Recovery routine with SSR needs this */ 877 __remove_dirty_segment(sbi, segno, DIRTY); 878 } 879 880 mutex_unlock(&dirty_i->seglist_lock); 881 } 882 883 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 884 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 885 { 886 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 887 unsigned int segno; 888 889 mutex_lock(&dirty_i->seglist_lock); 890 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 891 if (get_valid_blocks(sbi, segno, false)) 892 continue; 893 if (is_curseg(sbi, segno)) 894 continue; 895 __locate_dirty_segment(sbi, segno, PRE); 896 __remove_dirty_segment(sbi, segno, DIRTY); 897 } 898 mutex_unlock(&dirty_i->seglist_lock); 899 } 900 901 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 902 { 903 int ovp_hole_segs = 904 (overprovision_segments(sbi) - reserved_segments(sbi)); 905 block_t ovp_holes = SEGS_TO_BLKS(sbi, ovp_hole_segs); 906 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 907 block_t holes[2] = {0, 0}; /* DATA and NODE */ 908 block_t unusable; 909 struct seg_entry *se; 910 unsigned int segno; 911 912 mutex_lock(&dirty_i->seglist_lock); 913 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 914 se = get_seg_entry(sbi, segno); 915 if (IS_NODESEG(se->type)) 916 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) - 917 se->valid_blocks; 918 else 919 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) - 920 se->valid_blocks; 921 } 922 mutex_unlock(&dirty_i->seglist_lock); 923 924 unusable = max(holes[DATA], holes[NODE]); 925 if (unusable > ovp_holes) 926 return unusable - ovp_holes; 927 return 0; 928 } 929 930 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 931 { 932 int ovp_hole_segs = 933 (overprovision_segments(sbi) - reserved_segments(sbi)); 934 935 if (F2FS_OPTION(sbi).unusable_cap_perc == 100) 936 return 0; 937 if (unusable > F2FS_OPTION(sbi).unusable_cap) 938 return -EAGAIN; 939 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 940 dirty_segments(sbi) > ovp_hole_segs) 941 return -EAGAIN; 942 if (has_not_enough_free_secs(sbi, 0, 0)) 943 return -EAGAIN; 944 return 0; 945 } 946 947 /* This is only used by SBI_CP_DISABLED */ 948 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 949 { 950 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 951 unsigned int segno = 0; 952 953 mutex_lock(&dirty_i->seglist_lock); 954 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 955 if (get_valid_blocks(sbi, segno, false)) 956 continue; 957 if (get_ckpt_valid_blocks(sbi, segno, false)) 958 continue; 959 mutex_unlock(&dirty_i->seglist_lock); 960 return segno; 961 } 962 mutex_unlock(&dirty_i->seglist_lock); 963 return NULL_SEGNO; 964 } 965 966 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 967 struct block_device *bdev, block_t lstart, 968 block_t start, block_t len) 969 { 970 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 971 struct list_head *pend_list; 972 struct discard_cmd *dc; 973 974 f2fs_bug_on(sbi, !len); 975 976 pend_list = &dcc->pend_list[plist_idx(len)]; 977 978 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL); 979 INIT_LIST_HEAD(&dc->list); 980 dc->bdev = bdev; 981 dc->di.lstart = lstart; 982 dc->di.start = start; 983 dc->di.len = len; 984 dc->ref = 0; 985 dc->state = D_PREP; 986 dc->queued = 0; 987 dc->error = 0; 988 init_completion(&dc->wait); 989 list_add_tail(&dc->list, pend_list); 990 spin_lock_init(&dc->lock); 991 dc->bio_ref = 0; 992 atomic_inc(&dcc->discard_cmd_cnt); 993 dcc->undiscard_blks += len; 994 995 return dc; 996 } 997 998 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi) 999 { 1000 #ifdef CONFIG_F2FS_CHECK_FS 1001 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1002 struct rb_node *cur = rb_first_cached(&dcc->root), *next; 1003 struct discard_cmd *cur_dc, *next_dc; 1004 1005 while (cur) { 1006 next = rb_next(cur); 1007 if (!next) 1008 return true; 1009 1010 cur_dc = rb_entry(cur, struct discard_cmd, rb_node); 1011 next_dc = rb_entry(next, struct discard_cmd, rb_node); 1012 1013 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) { 1014 f2fs_info(sbi, "broken discard_rbtree, " 1015 "cur(%u, %u) next(%u, %u)", 1016 cur_dc->di.lstart, cur_dc->di.len, 1017 next_dc->di.lstart, next_dc->di.len); 1018 return false; 1019 } 1020 cur = next; 1021 } 1022 #endif 1023 return true; 1024 } 1025 1026 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi, 1027 block_t blkaddr) 1028 { 1029 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1030 struct rb_node *node = dcc->root.rb_root.rb_node; 1031 struct discard_cmd *dc; 1032 1033 while (node) { 1034 dc = rb_entry(node, struct discard_cmd, rb_node); 1035 1036 if (blkaddr < dc->di.lstart) 1037 node = node->rb_left; 1038 else if (blkaddr >= dc->di.lstart + dc->di.len) 1039 node = node->rb_right; 1040 else 1041 return dc; 1042 } 1043 return NULL; 1044 } 1045 1046 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root, 1047 block_t blkaddr, 1048 struct discard_cmd **prev_entry, 1049 struct discard_cmd **next_entry, 1050 struct rb_node ***insert_p, 1051 struct rb_node **insert_parent) 1052 { 1053 struct rb_node **pnode = &root->rb_root.rb_node; 1054 struct rb_node *parent = NULL, *tmp_node; 1055 struct discard_cmd *dc; 1056 1057 *insert_p = NULL; 1058 *insert_parent = NULL; 1059 *prev_entry = NULL; 1060 *next_entry = NULL; 1061 1062 if (RB_EMPTY_ROOT(&root->rb_root)) 1063 return NULL; 1064 1065 while (*pnode) { 1066 parent = *pnode; 1067 dc = rb_entry(*pnode, struct discard_cmd, rb_node); 1068 1069 if (blkaddr < dc->di.lstart) 1070 pnode = &(*pnode)->rb_left; 1071 else if (blkaddr >= dc->di.lstart + dc->di.len) 1072 pnode = &(*pnode)->rb_right; 1073 else 1074 goto lookup_neighbors; 1075 } 1076 1077 *insert_p = pnode; 1078 *insert_parent = parent; 1079 1080 dc = rb_entry(parent, struct discard_cmd, rb_node); 1081 tmp_node = parent; 1082 if (parent && blkaddr > dc->di.lstart) 1083 tmp_node = rb_next(parent); 1084 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1085 1086 tmp_node = parent; 1087 if (parent && blkaddr < dc->di.lstart) 1088 tmp_node = rb_prev(parent); 1089 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1090 return NULL; 1091 1092 lookup_neighbors: 1093 /* lookup prev node for merging backward later */ 1094 tmp_node = rb_prev(&dc->rb_node); 1095 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1096 1097 /* lookup next node for merging frontward later */ 1098 tmp_node = rb_next(&dc->rb_node); 1099 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1100 return dc; 1101 } 1102 1103 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 1104 struct discard_cmd *dc) 1105 { 1106 if (dc->state == D_DONE) 1107 atomic_sub(dc->queued, &dcc->queued_discard); 1108 1109 list_del(&dc->list); 1110 rb_erase_cached(&dc->rb_node, &dcc->root); 1111 dcc->undiscard_blks -= dc->di.len; 1112 1113 kmem_cache_free(discard_cmd_slab, dc); 1114 1115 atomic_dec(&dcc->discard_cmd_cnt); 1116 } 1117 1118 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 1119 struct discard_cmd *dc) 1120 { 1121 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1122 unsigned long flags; 1123 1124 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len); 1125 1126 spin_lock_irqsave(&dc->lock, flags); 1127 if (dc->bio_ref) { 1128 spin_unlock_irqrestore(&dc->lock, flags); 1129 return; 1130 } 1131 spin_unlock_irqrestore(&dc->lock, flags); 1132 1133 f2fs_bug_on(sbi, dc->ref); 1134 1135 if (dc->error == -EOPNOTSUPP) 1136 dc->error = 0; 1137 1138 if (dc->error) 1139 f2fs_info_ratelimited(sbi, 1140 "Issue discard(%u, %u, %u) failed, ret: %d", 1141 dc->di.lstart, dc->di.start, dc->di.len, dc->error); 1142 __detach_discard_cmd(dcc, dc); 1143 } 1144 1145 static void f2fs_submit_discard_endio(struct bio *bio) 1146 { 1147 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1148 unsigned long flags; 1149 1150 spin_lock_irqsave(&dc->lock, flags); 1151 if (!dc->error) 1152 dc->error = blk_status_to_errno(bio->bi_status); 1153 dc->bio_ref--; 1154 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1155 dc->state = D_DONE; 1156 complete_all(&dc->wait); 1157 } 1158 spin_unlock_irqrestore(&dc->lock, flags); 1159 bio_put(bio); 1160 } 1161 1162 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1163 block_t start, block_t end) 1164 { 1165 #ifdef CONFIG_F2FS_CHECK_FS 1166 struct seg_entry *sentry; 1167 unsigned int segno; 1168 block_t blk = start; 1169 unsigned long offset, size, *map; 1170 1171 while (blk < end) { 1172 segno = GET_SEGNO(sbi, blk); 1173 sentry = get_seg_entry(sbi, segno); 1174 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1175 1176 if (end < START_BLOCK(sbi, segno + 1)) 1177 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1178 else 1179 size = BLKS_PER_SEG(sbi); 1180 map = (unsigned long *)(sentry->cur_valid_map); 1181 offset = __find_rev_next_bit(map, size, offset); 1182 f2fs_bug_on(sbi, offset != size); 1183 blk = START_BLOCK(sbi, segno + 1); 1184 } 1185 #endif 1186 } 1187 1188 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1189 struct discard_policy *dpolicy, 1190 int discard_type, unsigned int granularity) 1191 { 1192 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1193 1194 /* common policy */ 1195 dpolicy->type = discard_type; 1196 dpolicy->sync = true; 1197 dpolicy->ordered = false; 1198 dpolicy->granularity = granularity; 1199 1200 dpolicy->max_requests = dcc->max_discard_request; 1201 dpolicy->io_aware_gran = dcc->discard_io_aware_gran; 1202 dpolicy->timeout = false; 1203 1204 if (discard_type == DPOLICY_BG) { 1205 dpolicy->min_interval = dcc->min_discard_issue_time; 1206 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1207 dpolicy->max_interval = dcc->max_discard_issue_time; 1208 if (dcc->discard_io_aware == DPOLICY_IO_AWARE_ENABLE) 1209 dpolicy->io_aware = true; 1210 else if (dcc->discard_io_aware == DPOLICY_IO_AWARE_DISABLE) 1211 dpolicy->io_aware = false; 1212 dpolicy->sync = false; 1213 dpolicy->ordered = true; 1214 if (utilization(sbi) > dcc->discard_urgent_util) { 1215 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1216 if (atomic_read(&dcc->discard_cmd_cnt)) 1217 dpolicy->max_interval = 1218 dcc->min_discard_issue_time; 1219 } 1220 } else if (discard_type == DPOLICY_FORCE) { 1221 dpolicy->min_interval = dcc->min_discard_issue_time; 1222 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1223 dpolicy->max_interval = dcc->max_discard_issue_time; 1224 dpolicy->io_aware = false; 1225 } else if (discard_type == DPOLICY_FSTRIM) { 1226 dpolicy->io_aware = false; 1227 } else if (discard_type == DPOLICY_UMOUNT) { 1228 dpolicy->io_aware = false; 1229 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1230 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1231 dpolicy->timeout = true; 1232 } 1233 } 1234 1235 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1236 struct block_device *bdev, block_t lstart, 1237 block_t start, block_t len); 1238 1239 #ifdef CONFIG_BLK_DEV_ZONED 1240 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi, 1241 struct discard_cmd *dc, blk_opf_t flag, 1242 struct list_head *wait_list, 1243 unsigned int *issued) 1244 { 1245 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1246 struct block_device *bdev = dc->bdev; 1247 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS); 1248 unsigned long flags; 1249 1250 trace_f2fs_issue_reset_zone(bdev, dc->di.start); 1251 1252 spin_lock_irqsave(&dc->lock, flags); 1253 dc->state = D_SUBMIT; 1254 dc->bio_ref++; 1255 spin_unlock_irqrestore(&dc->lock, flags); 1256 1257 if (issued) 1258 (*issued)++; 1259 1260 atomic_inc(&dcc->queued_discard); 1261 dc->queued++; 1262 list_move_tail(&dc->list, wait_list); 1263 1264 /* sanity check on discard range */ 1265 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len); 1266 1267 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start); 1268 bio->bi_private = dc; 1269 bio->bi_end_io = f2fs_submit_discard_endio; 1270 submit_bio(bio); 1271 1272 atomic_inc(&dcc->issued_discard); 1273 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE); 1274 } 1275 #endif 1276 1277 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1278 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1279 struct discard_policy *dpolicy, 1280 struct discard_cmd *dc, int *issued) 1281 { 1282 struct block_device *bdev = dc->bdev; 1283 unsigned int max_discard_blocks = 1284 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1285 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1286 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1287 &(dcc->fstrim_list) : &(dcc->wait_list); 1288 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0; 1289 block_t lstart, start, len, total_len; 1290 1291 if (dc->state != D_PREP) 1292 return 0; 1293 1294 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1295 return 0; 1296 1297 #ifdef CONFIG_BLK_DEV_ZONED 1298 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) { 1299 int devi = f2fs_bdev_index(sbi, bdev); 1300 1301 if (devi < 0) 1302 return -EINVAL; 1303 1304 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1305 __submit_zone_reset_cmd(sbi, dc, flag, 1306 wait_list, issued); 1307 return 0; 1308 } 1309 } 1310 #endif 1311 1312 /* 1313 * stop issuing discard for any of below cases: 1314 * 1. device is conventional zone, but it doesn't support discard. 1315 * 2. device is regulare device, after snapshot it doesn't support 1316 * discard. 1317 */ 1318 if (!bdev_max_discard_sectors(bdev)) 1319 return -EOPNOTSUPP; 1320 1321 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len); 1322 1323 lstart = dc->di.lstart; 1324 start = dc->di.start; 1325 len = dc->di.len; 1326 total_len = len; 1327 1328 dc->di.len = 0; 1329 1330 while (total_len && *issued < dpolicy->max_requests) { 1331 struct bio *bio = NULL; 1332 unsigned long flags; 1333 bool last = true; 1334 1335 if (len > max_discard_blocks) { 1336 len = max_discard_blocks; 1337 last = false; 1338 } 1339 1340 (*issued)++; 1341 if (*issued == dpolicy->max_requests) 1342 last = true; 1343 1344 dc->di.len += len; 1345 1346 __blkdev_issue_discard(bdev, SECTOR_FROM_BLOCK(start), 1347 SECTOR_FROM_BLOCK(len), GFP_NOFS, &bio); 1348 f2fs_bug_on(sbi, !bio); 1349 1350 /* 1351 * should keep before submission to avoid D_DONE 1352 * right away 1353 */ 1354 spin_lock_irqsave(&dc->lock, flags); 1355 if (last) 1356 dc->state = D_SUBMIT; 1357 else 1358 dc->state = D_PARTIAL; 1359 dc->bio_ref++; 1360 spin_unlock_irqrestore(&dc->lock, flags); 1361 1362 atomic_inc(&dcc->queued_discard); 1363 dc->queued++; 1364 list_move_tail(&dc->list, wait_list); 1365 1366 /* sanity check on discard range */ 1367 __check_sit_bitmap(sbi, lstart, lstart + len); 1368 1369 bio->bi_private = dc; 1370 bio->bi_end_io = f2fs_submit_discard_endio; 1371 bio->bi_opf |= flag; 1372 submit_bio(bio); 1373 1374 atomic_inc(&dcc->issued_discard); 1375 1376 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE); 1377 1378 lstart += len; 1379 start += len; 1380 total_len -= len; 1381 len = total_len; 1382 } 1383 1384 if (len) { 1385 dcc->undiscard_blks -= len; 1386 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1387 } 1388 return 0; 1389 } 1390 1391 static void __insert_discard_cmd(struct f2fs_sb_info *sbi, 1392 struct block_device *bdev, block_t lstart, 1393 block_t start, block_t len) 1394 { 1395 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1396 struct rb_node **p = &dcc->root.rb_root.rb_node; 1397 struct rb_node *parent = NULL; 1398 struct discard_cmd *dc; 1399 bool leftmost = true; 1400 1401 /* look up rb tree to find parent node */ 1402 while (*p) { 1403 parent = *p; 1404 dc = rb_entry(parent, struct discard_cmd, rb_node); 1405 1406 if (lstart < dc->di.lstart) { 1407 p = &(*p)->rb_left; 1408 } else if (lstart >= dc->di.lstart + dc->di.len) { 1409 p = &(*p)->rb_right; 1410 leftmost = false; 1411 } else { 1412 /* Let's skip to add, if exists */ 1413 return; 1414 } 1415 } 1416 1417 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 1418 1419 rb_link_node(&dc->rb_node, parent, p); 1420 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 1421 } 1422 1423 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1424 struct discard_cmd *dc) 1425 { 1426 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]); 1427 } 1428 1429 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1430 struct discard_cmd *dc, block_t blkaddr) 1431 { 1432 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1433 struct discard_info di = dc->di; 1434 bool modified = false; 1435 1436 if (dc->state == D_DONE || dc->di.len == 1) { 1437 __remove_discard_cmd(sbi, dc); 1438 return; 1439 } 1440 1441 dcc->undiscard_blks -= di.len; 1442 1443 if (blkaddr > di.lstart) { 1444 dc->di.len = blkaddr - dc->di.lstart; 1445 dcc->undiscard_blks += dc->di.len; 1446 __relocate_discard_cmd(dcc, dc); 1447 modified = true; 1448 } 1449 1450 if (blkaddr < di.lstart + di.len - 1) { 1451 if (modified) { 1452 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1, 1453 di.start + blkaddr + 1 - di.lstart, 1454 di.lstart + di.len - 1 - blkaddr); 1455 } else { 1456 dc->di.lstart++; 1457 dc->di.len--; 1458 dc->di.start++; 1459 dcc->undiscard_blks += dc->di.len; 1460 __relocate_discard_cmd(dcc, dc); 1461 } 1462 } 1463 } 1464 1465 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1466 struct block_device *bdev, block_t lstart, 1467 block_t start, block_t len) 1468 { 1469 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1470 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1471 struct discard_cmd *dc; 1472 struct discard_info di = {0}; 1473 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1474 unsigned int max_discard_blocks = 1475 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1476 block_t end = lstart + len; 1477 1478 dc = __lookup_discard_cmd_ret(&dcc->root, lstart, 1479 &prev_dc, &next_dc, &insert_p, &insert_parent); 1480 if (dc) 1481 prev_dc = dc; 1482 1483 if (!prev_dc) { 1484 di.lstart = lstart; 1485 di.len = next_dc ? next_dc->di.lstart - lstart : len; 1486 di.len = min(di.len, len); 1487 di.start = start; 1488 } 1489 1490 while (1) { 1491 struct rb_node *node; 1492 bool merged = false; 1493 struct discard_cmd *tdc = NULL; 1494 1495 if (prev_dc) { 1496 di.lstart = prev_dc->di.lstart + prev_dc->di.len; 1497 if (di.lstart < lstart) 1498 di.lstart = lstart; 1499 if (di.lstart >= end) 1500 break; 1501 1502 if (!next_dc || next_dc->di.lstart > end) 1503 di.len = end - di.lstart; 1504 else 1505 di.len = next_dc->di.lstart - di.lstart; 1506 di.start = start + di.lstart - lstart; 1507 } 1508 1509 if (!di.len) 1510 goto next; 1511 1512 if (prev_dc && prev_dc->state == D_PREP && 1513 prev_dc->bdev == bdev && 1514 __is_discard_back_mergeable(&di, &prev_dc->di, 1515 max_discard_blocks)) { 1516 prev_dc->di.len += di.len; 1517 dcc->undiscard_blks += di.len; 1518 __relocate_discard_cmd(dcc, prev_dc); 1519 di = prev_dc->di; 1520 tdc = prev_dc; 1521 merged = true; 1522 } 1523 1524 if (next_dc && next_dc->state == D_PREP && 1525 next_dc->bdev == bdev && 1526 __is_discard_front_mergeable(&di, &next_dc->di, 1527 max_discard_blocks)) { 1528 next_dc->di.lstart = di.lstart; 1529 next_dc->di.len += di.len; 1530 next_dc->di.start = di.start; 1531 dcc->undiscard_blks += di.len; 1532 __relocate_discard_cmd(dcc, next_dc); 1533 if (tdc) 1534 __remove_discard_cmd(sbi, tdc); 1535 merged = true; 1536 } 1537 1538 if (!merged) 1539 __insert_discard_cmd(sbi, bdev, 1540 di.lstart, di.start, di.len); 1541 next: 1542 prev_dc = next_dc; 1543 if (!prev_dc) 1544 break; 1545 1546 node = rb_next(&prev_dc->rb_node); 1547 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1548 } 1549 } 1550 1551 #ifdef CONFIG_BLK_DEV_ZONED 1552 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi, 1553 struct block_device *bdev, block_t blkstart, block_t lblkstart, 1554 block_t blklen) 1555 { 1556 trace_f2fs_queue_reset_zone(bdev, blkstart); 1557 1558 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1559 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen); 1560 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1561 } 1562 #endif 1563 1564 static void __queue_discard_cmd(struct f2fs_sb_info *sbi, 1565 struct block_device *bdev, block_t blkstart, block_t blklen) 1566 { 1567 block_t lblkstart = blkstart; 1568 1569 if (!f2fs_bdev_support_discard(bdev)) 1570 return; 1571 1572 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1573 1574 if (f2fs_is_multi_device(sbi)) { 1575 int devi = f2fs_target_device_index(sbi, blkstart); 1576 1577 blkstart -= FDEV(devi).start_blk; 1578 } 1579 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1580 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1581 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1582 } 1583 1584 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1585 struct discard_policy *dpolicy, int *issued) 1586 { 1587 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1588 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1589 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1590 struct discard_cmd *dc; 1591 struct blk_plug plug; 1592 bool io_interrupted = false; 1593 1594 mutex_lock(&dcc->cmd_lock); 1595 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos, 1596 &prev_dc, &next_dc, &insert_p, &insert_parent); 1597 if (!dc) 1598 dc = next_dc; 1599 1600 blk_start_plug(&plug); 1601 1602 while (dc) { 1603 struct rb_node *node; 1604 int err = 0; 1605 1606 if (dc->state != D_PREP) 1607 goto next; 1608 1609 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1610 io_interrupted = true; 1611 break; 1612 } 1613 1614 dcc->next_pos = dc->di.lstart + dc->di.len; 1615 err = __submit_discard_cmd(sbi, dpolicy, dc, issued); 1616 1617 if (*issued >= dpolicy->max_requests) 1618 break; 1619 next: 1620 node = rb_next(&dc->rb_node); 1621 if (err) 1622 __remove_discard_cmd(sbi, dc); 1623 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1624 } 1625 1626 blk_finish_plug(&plug); 1627 1628 if (!dc) 1629 dcc->next_pos = 0; 1630 1631 mutex_unlock(&dcc->cmd_lock); 1632 1633 if (!(*issued) && io_interrupted) 1634 *issued = -1; 1635 } 1636 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1637 struct discard_policy *dpolicy); 1638 1639 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1640 struct discard_policy *dpolicy) 1641 { 1642 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1643 struct list_head *pend_list; 1644 struct discard_cmd *dc, *tmp; 1645 struct blk_plug plug; 1646 int i, issued; 1647 bool io_interrupted = false; 1648 1649 if (dpolicy->timeout) 1650 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1651 1652 retry: 1653 issued = 0; 1654 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1655 if (dpolicy->timeout && 1656 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1657 break; 1658 1659 if (i + 1 < dpolicy->granularity) 1660 break; 1661 1662 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) { 1663 __issue_discard_cmd_orderly(sbi, dpolicy, &issued); 1664 return issued; 1665 } 1666 1667 pend_list = &dcc->pend_list[i]; 1668 1669 mutex_lock(&dcc->cmd_lock); 1670 if (list_empty(pend_list)) 1671 goto next; 1672 if (unlikely(dcc->rbtree_check)) 1673 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 1674 blk_start_plug(&plug); 1675 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1676 f2fs_bug_on(sbi, dc->state != D_PREP); 1677 1678 if (dpolicy->timeout && 1679 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1680 break; 1681 1682 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1683 !is_idle(sbi, DISCARD_TIME)) { 1684 io_interrupted = true; 1685 break; 1686 } 1687 1688 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1689 1690 if (issued >= dpolicy->max_requests) 1691 break; 1692 } 1693 blk_finish_plug(&plug); 1694 next: 1695 mutex_unlock(&dcc->cmd_lock); 1696 1697 if (issued >= dpolicy->max_requests || io_interrupted) 1698 break; 1699 } 1700 1701 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1702 __wait_all_discard_cmd(sbi, dpolicy); 1703 goto retry; 1704 } 1705 1706 if (!issued && io_interrupted) 1707 issued = -1; 1708 1709 return issued; 1710 } 1711 1712 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1713 { 1714 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1715 struct list_head *pend_list; 1716 struct discard_cmd *dc, *tmp; 1717 int i; 1718 bool dropped = false; 1719 1720 mutex_lock(&dcc->cmd_lock); 1721 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1722 pend_list = &dcc->pend_list[i]; 1723 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1724 f2fs_bug_on(sbi, dc->state != D_PREP); 1725 __remove_discard_cmd(sbi, dc); 1726 dropped = true; 1727 } 1728 } 1729 mutex_unlock(&dcc->cmd_lock); 1730 1731 return dropped; 1732 } 1733 1734 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1735 { 1736 __drop_discard_cmd(sbi); 1737 } 1738 1739 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1740 struct discard_cmd *dc) 1741 { 1742 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1743 unsigned int len = 0; 1744 1745 wait_for_completion_io(&dc->wait); 1746 mutex_lock(&dcc->cmd_lock); 1747 f2fs_bug_on(sbi, dc->state != D_DONE); 1748 dc->ref--; 1749 if (!dc->ref) { 1750 if (!dc->error) 1751 len = dc->di.len; 1752 __remove_discard_cmd(sbi, dc); 1753 } 1754 mutex_unlock(&dcc->cmd_lock); 1755 1756 return len; 1757 } 1758 1759 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1760 struct discard_policy *dpolicy, 1761 block_t start, block_t end) 1762 { 1763 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1764 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1765 &(dcc->fstrim_list) : &(dcc->wait_list); 1766 struct discard_cmd *dc = NULL, *iter, *tmp; 1767 unsigned int trimmed = 0; 1768 1769 next: 1770 dc = NULL; 1771 1772 mutex_lock(&dcc->cmd_lock); 1773 list_for_each_entry_safe(iter, tmp, wait_list, list) { 1774 if (iter->di.lstart + iter->di.len <= start || 1775 end <= iter->di.lstart) 1776 continue; 1777 if (iter->di.len < dpolicy->granularity) 1778 continue; 1779 if (iter->state == D_DONE && !iter->ref) { 1780 wait_for_completion_io(&iter->wait); 1781 if (!iter->error) 1782 trimmed += iter->di.len; 1783 __remove_discard_cmd(sbi, iter); 1784 } else { 1785 iter->ref++; 1786 dc = iter; 1787 break; 1788 } 1789 } 1790 mutex_unlock(&dcc->cmd_lock); 1791 1792 if (dc) { 1793 trimmed += __wait_one_discard_bio(sbi, dc); 1794 goto next; 1795 } 1796 1797 return trimmed; 1798 } 1799 1800 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1801 struct discard_policy *dpolicy) 1802 { 1803 struct discard_policy dp; 1804 unsigned int discard_blks; 1805 1806 if (dpolicy) 1807 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1808 1809 /* wait all */ 1810 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY); 1811 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1812 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY); 1813 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1814 1815 return discard_blks; 1816 } 1817 1818 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1819 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1820 { 1821 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1822 struct discard_cmd *dc; 1823 bool need_wait = false; 1824 1825 mutex_lock(&dcc->cmd_lock); 1826 dc = __lookup_discard_cmd(sbi, blkaddr); 1827 #ifdef CONFIG_BLK_DEV_ZONED 1828 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) { 1829 int devi = f2fs_bdev_index(sbi, dc->bdev); 1830 1831 if (devi < 0) { 1832 mutex_unlock(&dcc->cmd_lock); 1833 return; 1834 } 1835 1836 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1837 /* force submit zone reset */ 1838 if (dc->state == D_PREP) 1839 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC, 1840 &dcc->wait_list, NULL); 1841 dc->ref++; 1842 mutex_unlock(&dcc->cmd_lock); 1843 /* wait zone reset */ 1844 __wait_one_discard_bio(sbi, dc); 1845 return; 1846 } 1847 } 1848 #endif 1849 if (dc) { 1850 if (dc->state == D_PREP) { 1851 __punch_discard_cmd(sbi, dc, blkaddr); 1852 } else { 1853 dc->ref++; 1854 need_wait = true; 1855 } 1856 } 1857 mutex_unlock(&dcc->cmd_lock); 1858 1859 if (need_wait) 1860 __wait_one_discard_bio(sbi, dc); 1861 } 1862 1863 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1864 { 1865 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1866 1867 if (dcc && dcc->f2fs_issue_discard) { 1868 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1869 1870 dcc->f2fs_issue_discard = NULL; 1871 kthread_stop(discard_thread); 1872 } 1873 } 1874 1875 /** 1876 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT 1877 * @sbi: the f2fs_sb_info data for discard cmd to issue 1878 * 1879 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped 1880 * 1881 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false. 1882 */ 1883 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1884 { 1885 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1886 struct discard_policy dpolicy; 1887 bool dropped; 1888 1889 if (!atomic_read(&dcc->discard_cmd_cnt)) 1890 return true; 1891 1892 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1893 dcc->discard_granularity); 1894 __issue_discard_cmd(sbi, &dpolicy); 1895 dropped = __drop_discard_cmd(sbi); 1896 1897 /* just to make sure there is no pending discard commands */ 1898 __wait_all_discard_cmd(sbi, NULL); 1899 1900 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1901 return !dropped; 1902 } 1903 1904 static int issue_discard_thread(void *data) 1905 { 1906 struct f2fs_sb_info *sbi = data; 1907 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1908 wait_queue_head_t *q = &dcc->discard_wait_queue; 1909 struct discard_policy dpolicy; 1910 unsigned int wait_ms = dcc->min_discard_issue_time; 1911 int issued; 1912 1913 set_freezable(); 1914 1915 do { 1916 wait_event_freezable_timeout(*q, 1917 kthread_should_stop() || dcc->discard_wake, 1918 msecs_to_jiffies(wait_ms)); 1919 1920 if (sbi->gc_mode == GC_URGENT_HIGH || 1921 !f2fs_available_free_memory(sbi, DISCARD_CACHE)) 1922 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1923 MIN_DISCARD_GRANULARITY); 1924 else 1925 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1926 dcc->discard_granularity); 1927 1928 if (dcc->discard_wake) 1929 dcc->discard_wake = false; 1930 1931 /* clean up pending candidates before going to sleep */ 1932 if (atomic_read(&dcc->queued_discard)) 1933 __wait_all_discard_cmd(sbi, NULL); 1934 1935 if (f2fs_readonly(sbi->sb)) 1936 continue; 1937 if (kthread_should_stop()) 1938 return 0; 1939 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) || 1940 !atomic_read(&dcc->discard_cmd_cnt)) { 1941 wait_ms = dpolicy.max_interval; 1942 continue; 1943 } 1944 1945 sb_start_intwrite(sbi->sb); 1946 1947 issued = __issue_discard_cmd(sbi, &dpolicy); 1948 if (issued > 0) { 1949 __wait_all_discard_cmd(sbi, &dpolicy); 1950 wait_ms = dpolicy.min_interval; 1951 } else if (issued == -1) { 1952 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1953 if (!wait_ms) 1954 wait_ms = dpolicy.mid_interval; 1955 } else { 1956 wait_ms = dpolicy.max_interval; 1957 } 1958 if (!atomic_read(&dcc->discard_cmd_cnt)) 1959 wait_ms = dpolicy.max_interval; 1960 1961 sb_end_intwrite(sbi->sb); 1962 1963 } while (!kthread_should_stop()); 1964 return 0; 1965 } 1966 1967 #ifdef CONFIG_BLK_DEV_ZONED 1968 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1969 struct block_device *bdev, block_t blkstart, block_t blklen) 1970 { 1971 sector_t sector, nr_sects; 1972 block_t lblkstart = blkstart; 1973 int devi = 0; 1974 u64 remainder = 0; 1975 1976 if (f2fs_is_multi_device(sbi)) { 1977 devi = f2fs_target_device_index(sbi, blkstart); 1978 if (blkstart < FDEV(devi).start_blk || 1979 blkstart > FDEV(devi).end_blk) { 1980 f2fs_err(sbi, "Invalid block %x", blkstart); 1981 return -EIO; 1982 } 1983 blkstart -= FDEV(devi).start_blk; 1984 } 1985 1986 /* For sequential zones, reset the zone write pointer */ 1987 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1988 sector = SECTOR_FROM_BLOCK(blkstart); 1989 nr_sects = SECTOR_FROM_BLOCK(blklen); 1990 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder); 1991 1992 if (remainder || nr_sects != bdev_zone_sectors(bdev)) { 1993 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1994 devi, sbi->s_ndevs ? FDEV(devi).path : "", 1995 blkstart, blklen); 1996 return -EIO; 1997 } 1998 1999 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) { 2000 unsigned int nofs_flags; 2001 int ret; 2002 2003 trace_f2fs_issue_reset_zone(bdev, blkstart); 2004 nofs_flags = memalloc_nofs_save(); 2005 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 2006 sector, nr_sects); 2007 memalloc_nofs_restore(nofs_flags); 2008 return ret; 2009 } 2010 2011 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen); 2012 return 0; 2013 } 2014 2015 /* For conventional zones, use regular discard if supported */ 2016 __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 2017 return 0; 2018 } 2019 #endif 2020 2021 static int __issue_discard_async(struct f2fs_sb_info *sbi, 2022 struct block_device *bdev, block_t blkstart, block_t blklen) 2023 { 2024 #ifdef CONFIG_BLK_DEV_ZONED 2025 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 2026 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 2027 #endif 2028 __queue_discard_cmd(sbi, bdev, blkstart, blklen); 2029 return 0; 2030 } 2031 2032 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 2033 block_t blkstart, block_t blklen) 2034 { 2035 sector_t start = blkstart, len = 0; 2036 struct block_device *bdev; 2037 struct seg_entry *se; 2038 unsigned int offset; 2039 block_t i; 2040 int err = 0; 2041 2042 bdev = f2fs_target_device(sbi, blkstart, NULL); 2043 2044 for (i = blkstart; i < blkstart + blklen; i++, len++) { 2045 if (i != start) { 2046 struct block_device *bdev2 = 2047 f2fs_target_device(sbi, i, NULL); 2048 2049 if (bdev2 != bdev) { 2050 err = __issue_discard_async(sbi, bdev, 2051 start, len); 2052 if (err) 2053 return err; 2054 bdev = bdev2; 2055 start = i; 2056 len = 0; 2057 } 2058 } 2059 2060 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 2061 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 2062 2063 if (f2fs_block_unit_discard(sbi) && 2064 !f2fs_test_and_set_bit(offset, se->discard_map)) 2065 sbi->discard_blks--; 2066 } 2067 2068 if (len) 2069 err = __issue_discard_async(sbi, bdev, start, len); 2070 return err; 2071 } 2072 2073 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 2074 bool check_only) 2075 { 2076 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2077 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 2078 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2079 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2080 unsigned long *discard_map = (unsigned long *)se->discard_map; 2081 unsigned long *dmap = SIT_I(sbi)->tmp_map; 2082 unsigned int start = 0, end = -1; 2083 bool force = (cpc->reason & CP_DISCARD); 2084 struct discard_entry *de = NULL; 2085 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 2086 int i; 2087 2088 if (se->valid_blocks == BLKS_PER_SEG(sbi) || 2089 !f2fs_hw_support_discard(sbi) || 2090 !f2fs_block_unit_discard(sbi)) 2091 return false; 2092 2093 if (!force) { 2094 if (!f2fs_realtime_discard_enable(sbi) || 2095 (!se->valid_blocks && 2096 !is_curseg(sbi, cpc->trim_start)) || 2097 SM_I(sbi)->dcc_info->nr_discards >= 2098 SM_I(sbi)->dcc_info->max_discards) 2099 return false; 2100 } 2101 2102 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 2103 for (i = 0; i < entries; i++) 2104 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 2105 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 2106 2107 while (force || SM_I(sbi)->dcc_info->nr_discards <= 2108 SM_I(sbi)->dcc_info->max_discards) { 2109 start = __find_rev_next_bit(dmap, BLKS_PER_SEG(sbi), end + 1); 2110 if (start >= BLKS_PER_SEG(sbi)) 2111 break; 2112 2113 end = __find_rev_next_zero_bit(dmap, 2114 BLKS_PER_SEG(sbi), start + 1); 2115 if (force && start && end != BLKS_PER_SEG(sbi) && 2116 (end - start) < cpc->trim_minlen) 2117 continue; 2118 2119 if (check_only) 2120 return true; 2121 2122 if (!de) { 2123 de = f2fs_kmem_cache_alloc(discard_entry_slab, 2124 GFP_F2FS_ZERO, true, NULL); 2125 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 2126 list_add_tail(&de->list, head); 2127 } 2128 2129 for (i = start; i < end; i++) 2130 __set_bit_le(i, (void *)de->discard_map); 2131 2132 SM_I(sbi)->dcc_info->nr_discards += end - start; 2133 } 2134 return false; 2135 } 2136 2137 static void release_discard_addr(struct discard_entry *entry) 2138 { 2139 list_del(&entry->list); 2140 kmem_cache_free(discard_entry_slab, entry); 2141 } 2142 2143 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 2144 { 2145 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 2146 struct discard_entry *entry, *this; 2147 2148 /* drop caches */ 2149 list_for_each_entry_safe(entry, this, head, list) 2150 release_discard_addr(entry); 2151 } 2152 2153 /* 2154 * Should call f2fs_clear_prefree_segments after checkpoint is done. 2155 */ 2156 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 2157 { 2158 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2159 unsigned int segno; 2160 2161 mutex_lock(&dirty_i->seglist_lock); 2162 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 2163 __set_test_and_free(sbi, segno, false); 2164 mutex_unlock(&dirty_i->seglist_lock); 2165 } 2166 2167 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 2168 struct cp_control *cpc) 2169 { 2170 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2171 struct list_head *head = &dcc->entry_list; 2172 struct discard_entry *entry, *this; 2173 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2174 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 2175 unsigned int start = 0, end = -1; 2176 unsigned int secno, start_segno; 2177 bool force = (cpc->reason & CP_DISCARD); 2178 bool section_alignment = F2FS_OPTION(sbi).discard_unit == 2179 DISCARD_UNIT_SECTION; 2180 2181 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi)) 2182 section_alignment = true; 2183 2184 mutex_lock(&dirty_i->seglist_lock); 2185 2186 while (1) { 2187 int i; 2188 2189 if (section_alignment && end != -1) 2190 end--; 2191 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 2192 if (start >= MAIN_SEGS(sbi)) 2193 break; 2194 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 2195 start + 1); 2196 2197 if (section_alignment) { 2198 start = rounddown(start, SEGS_PER_SEC(sbi)); 2199 end = roundup(end, SEGS_PER_SEC(sbi)); 2200 } 2201 2202 for (i = start; i < end; i++) { 2203 if (test_and_clear_bit(i, prefree_map)) 2204 dirty_i->nr_dirty[PRE]--; 2205 } 2206 2207 if (!f2fs_realtime_discard_enable(sbi)) 2208 continue; 2209 2210 if (force && start >= cpc->trim_start && 2211 (end - 1) <= cpc->trim_end) 2212 continue; 2213 2214 /* Should cover 2MB zoned device for zone-based reset */ 2215 if (!f2fs_sb_has_blkzoned(sbi) && 2216 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) { 2217 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 2218 SEGS_TO_BLKS(sbi, end - start)); 2219 continue; 2220 } 2221 next: 2222 secno = GET_SEC_FROM_SEG(sbi, start); 2223 start_segno = GET_SEG_FROM_SEC(sbi, secno); 2224 if (!is_cursec(sbi, secno) && 2225 !get_valid_blocks(sbi, start, true)) 2226 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 2227 BLKS_PER_SEC(sbi)); 2228 2229 start = start_segno + SEGS_PER_SEC(sbi); 2230 if (start < end) 2231 goto next; 2232 else 2233 end = start - 1; 2234 } 2235 mutex_unlock(&dirty_i->seglist_lock); 2236 2237 if (!f2fs_block_unit_discard(sbi)) 2238 goto wakeup; 2239 2240 /* send small discards */ 2241 list_for_each_entry_safe(entry, this, head, list) { 2242 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2243 bool is_valid = test_bit_le(0, entry->discard_map); 2244 2245 find_next: 2246 if (is_valid) { 2247 next_pos = find_next_zero_bit_le(entry->discard_map, 2248 BLKS_PER_SEG(sbi), cur_pos); 2249 len = next_pos - cur_pos; 2250 2251 if (f2fs_sb_has_blkzoned(sbi) || 2252 (force && len < cpc->trim_minlen)) 2253 goto skip; 2254 2255 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2256 len); 2257 total_len += len; 2258 } else { 2259 next_pos = find_next_bit_le(entry->discard_map, 2260 BLKS_PER_SEG(sbi), cur_pos); 2261 } 2262 skip: 2263 cur_pos = next_pos; 2264 is_valid = !is_valid; 2265 2266 if (cur_pos < BLKS_PER_SEG(sbi)) 2267 goto find_next; 2268 2269 release_discard_addr(entry); 2270 dcc->nr_discards -= total_len; 2271 } 2272 2273 wakeup: 2274 wake_up_discard_thread(sbi, false); 2275 } 2276 2277 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi) 2278 { 2279 dev_t dev = sbi->sb->s_bdev->bd_dev; 2280 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2281 int err = 0; 2282 2283 if (f2fs_sb_has_readonly(sbi)) { 2284 f2fs_info(sbi, 2285 "Skip to start discard thread for readonly image"); 2286 return 0; 2287 } 2288 2289 if (!f2fs_realtime_discard_enable(sbi)) 2290 return 0; 2291 2292 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2293 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2294 if (IS_ERR(dcc->f2fs_issue_discard)) { 2295 err = PTR_ERR(dcc->f2fs_issue_discard); 2296 dcc->f2fs_issue_discard = NULL; 2297 } 2298 2299 return err; 2300 } 2301 2302 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2303 { 2304 struct discard_cmd_control *dcc; 2305 int err = 0, i; 2306 2307 if (SM_I(sbi)->dcc_info) { 2308 dcc = SM_I(sbi)->dcc_info; 2309 goto init_thread; 2310 } 2311 2312 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2313 if (!dcc) 2314 return -ENOMEM; 2315 2316 dcc->discard_io_aware_gran = MAX_PLIST_NUM; 2317 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2318 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY; 2319 dcc->discard_io_aware = DPOLICY_IO_AWARE_ENABLE; 2320 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT || 2321 F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION) 2322 dcc->discard_granularity = BLKS_PER_SEG(sbi); 2323 2324 INIT_LIST_HEAD(&dcc->entry_list); 2325 for (i = 0; i < MAX_PLIST_NUM; i++) 2326 INIT_LIST_HEAD(&dcc->pend_list[i]); 2327 INIT_LIST_HEAD(&dcc->wait_list); 2328 INIT_LIST_HEAD(&dcc->fstrim_list); 2329 mutex_init(&dcc->cmd_lock); 2330 atomic_set(&dcc->issued_discard, 0); 2331 atomic_set(&dcc->queued_discard, 0); 2332 atomic_set(&dcc->discard_cmd_cnt, 0); 2333 dcc->nr_discards = 0; 2334 dcc->max_discards = SEGS_TO_BLKS(sbi, MAIN_SEGS(sbi)); 2335 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST; 2336 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME; 2337 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME; 2338 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME; 2339 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL; 2340 dcc->undiscard_blks = 0; 2341 dcc->next_pos = 0; 2342 dcc->root = RB_ROOT_CACHED; 2343 dcc->rbtree_check = false; 2344 2345 init_waitqueue_head(&dcc->discard_wait_queue); 2346 SM_I(sbi)->dcc_info = dcc; 2347 init_thread: 2348 err = f2fs_start_discard_thread(sbi); 2349 if (err) { 2350 kfree(dcc); 2351 SM_I(sbi)->dcc_info = NULL; 2352 } 2353 2354 return err; 2355 } 2356 2357 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2358 { 2359 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2360 2361 if (!dcc) 2362 return; 2363 2364 f2fs_stop_discard_thread(sbi); 2365 2366 /* 2367 * Recovery can cache discard commands, so in error path of 2368 * fill_super(), it needs to give a chance to handle them. 2369 */ 2370 f2fs_issue_discard_timeout(sbi); 2371 2372 kfree(dcc); 2373 SM_I(sbi)->dcc_info = NULL; 2374 } 2375 2376 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2377 { 2378 struct sit_info *sit_i = SIT_I(sbi); 2379 2380 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2381 sit_i->dirty_sentries++; 2382 return false; 2383 } 2384 2385 return true; 2386 } 2387 2388 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2389 unsigned int segno, int modified) 2390 { 2391 struct seg_entry *se = get_seg_entry(sbi, segno); 2392 2393 se->type = type; 2394 if (modified) 2395 __mark_sit_entry_dirty(sbi, segno); 2396 } 2397 2398 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi, 2399 block_t blkaddr) 2400 { 2401 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2402 2403 if (segno == NULL_SEGNO) 2404 return 0; 2405 return get_seg_entry(sbi, segno)->mtime; 2406 } 2407 2408 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr, 2409 unsigned long long old_mtime) 2410 { 2411 struct seg_entry *se; 2412 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2413 unsigned long long ctime = get_mtime(sbi, false); 2414 unsigned long long mtime = old_mtime ? old_mtime : ctime; 2415 2416 if (segno == NULL_SEGNO) 2417 return; 2418 2419 se = get_seg_entry(sbi, segno); 2420 2421 if (!se->mtime) 2422 se->mtime = mtime; 2423 else 2424 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime, 2425 se->valid_blocks + 1); 2426 2427 if (ctime > SIT_I(sbi)->max_mtime) 2428 SIT_I(sbi)->max_mtime = ctime; 2429 } 2430 2431 /* 2432 * NOTE: when updating multiple blocks at the same time, please ensure 2433 * that the consecutive input blocks belong to the same segment. 2434 */ 2435 static int update_sit_entry_for_release(struct f2fs_sb_info *sbi, struct seg_entry *se, 2436 unsigned int segno, block_t blkaddr, unsigned int offset, int del) 2437 { 2438 bool exist; 2439 #ifdef CONFIG_F2FS_CHECK_FS 2440 bool mir_exist; 2441 #endif 2442 int i; 2443 int del_count = -del; 2444 2445 f2fs_bug_on(sbi, GET_SEGNO(sbi, blkaddr) != GET_SEGNO(sbi, blkaddr + del_count - 1)); 2446 2447 for (i = 0; i < del_count; i++) { 2448 exist = f2fs_test_and_clear_bit(offset + i, se->cur_valid_map); 2449 #ifdef CONFIG_F2FS_CHECK_FS 2450 mir_exist = f2fs_test_and_clear_bit(offset + i, 2451 se->cur_valid_map_mir); 2452 if (unlikely(exist != mir_exist)) { 2453 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2454 blkaddr + i, exist); 2455 f2fs_bug_on(sbi, 1); 2456 } 2457 #endif 2458 if (unlikely(!exist)) { 2459 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", blkaddr + i); 2460 f2fs_bug_on(sbi, 1); 2461 se->valid_blocks++; 2462 del += 1; 2463 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2464 /* 2465 * If checkpoints are off, we must not reuse data that 2466 * was used in the previous checkpoint. If it was used 2467 * before, we must track that to know how much space we 2468 * really have. 2469 */ 2470 if (f2fs_test_bit(offset + i, se->ckpt_valid_map)) { 2471 spin_lock(&sbi->stat_lock); 2472 sbi->unusable_block_count++; 2473 spin_unlock(&sbi->stat_lock); 2474 } 2475 } 2476 2477 if (f2fs_block_unit_discard(sbi) && 2478 f2fs_test_and_clear_bit(offset + i, se->discard_map)) 2479 sbi->discard_blks++; 2480 2481 if (!f2fs_test_bit(offset + i, se->ckpt_valid_map)) { 2482 se->ckpt_valid_blocks -= 1; 2483 if (__is_large_section(sbi)) 2484 get_sec_entry(sbi, segno)->ckpt_valid_blocks -= 1; 2485 } 2486 } 2487 2488 if (__is_large_section(sbi)) 2489 sanity_check_valid_blocks(sbi, segno); 2490 2491 return del; 2492 } 2493 2494 static int update_sit_entry_for_alloc(struct f2fs_sb_info *sbi, struct seg_entry *se, 2495 unsigned int segno, block_t blkaddr, unsigned int offset, int del) 2496 { 2497 bool exist; 2498 #ifdef CONFIG_F2FS_CHECK_FS 2499 bool mir_exist; 2500 #endif 2501 2502 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2503 #ifdef CONFIG_F2FS_CHECK_FS 2504 mir_exist = f2fs_test_and_set_bit(offset, 2505 se->cur_valid_map_mir); 2506 if (unlikely(exist != mir_exist)) { 2507 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2508 blkaddr, exist); 2509 f2fs_bug_on(sbi, 1); 2510 } 2511 #endif 2512 if (unlikely(exist)) { 2513 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", blkaddr); 2514 f2fs_bug_on(sbi, 1); 2515 se->valid_blocks--; 2516 del = 0; 2517 } 2518 2519 if (f2fs_block_unit_discard(sbi) && 2520 !f2fs_test_and_set_bit(offset, se->discard_map)) 2521 sbi->discard_blks--; 2522 2523 /* 2524 * SSR should never reuse block which is checkpointed 2525 * or newly invalidated. 2526 */ 2527 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2528 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) { 2529 se->ckpt_valid_blocks++; 2530 if (__is_large_section(sbi)) 2531 get_sec_entry(sbi, segno)->ckpt_valid_blocks++; 2532 } 2533 } 2534 2535 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) { 2536 se->ckpt_valid_blocks += del; 2537 if (__is_large_section(sbi)) 2538 get_sec_entry(sbi, segno)->ckpt_valid_blocks += del; 2539 } 2540 2541 if (__is_large_section(sbi)) 2542 sanity_check_valid_blocks(sbi, segno); 2543 2544 return del; 2545 } 2546 2547 /* 2548 * If releasing blocks, this function supports updating multiple consecutive blocks 2549 * at one time, but please note that these consecutive blocks need to belong to the 2550 * same segment. 2551 */ 2552 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2553 { 2554 struct seg_entry *se; 2555 unsigned int segno, offset; 2556 long int new_vblocks; 2557 2558 segno = GET_SEGNO(sbi, blkaddr); 2559 if (segno == NULL_SEGNO) 2560 return; 2561 2562 se = get_seg_entry(sbi, segno); 2563 new_vblocks = se->valid_blocks + del; 2564 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2565 2566 f2fs_bug_on(sbi, (new_vblocks < 0 || 2567 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno)))); 2568 2569 se->valid_blocks = new_vblocks; 2570 2571 /* Update valid block bitmap */ 2572 if (del > 0) { 2573 del = update_sit_entry_for_alloc(sbi, se, segno, blkaddr, offset, del); 2574 } else { 2575 del = update_sit_entry_for_release(sbi, se, segno, blkaddr, offset, del); 2576 } 2577 2578 __mark_sit_entry_dirty(sbi, segno); 2579 2580 /* update total number of valid blocks to be written in ckpt area */ 2581 SIT_I(sbi)->written_valid_blocks += del; 2582 2583 if (__is_large_section(sbi)) 2584 get_sec_entry(sbi, segno)->valid_blocks += del; 2585 } 2586 2587 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr, 2588 unsigned int len) 2589 { 2590 unsigned int segno = GET_SEGNO(sbi, addr); 2591 struct sit_info *sit_i = SIT_I(sbi); 2592 block_t addr_start = addr, addr_end = addr + len - 1; 2593 unsigned int seg_num = GET_SEGNO(sbi, addr_end) - segno + 1; 2594 unsigned int i = 1, max_blocks = sbi->blocks_per_seg, cnt; 2595 2596 f2fs_bug_on(sbi, addr == NULL_ADDR); 2597 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2598 return; 2599 2600 f2fs_invalidate_internal_cache(sbi, addr, len); 2601 2602 /* add it into sit main buffer */ 2603 down_write(&sit_i->sentry_lock); 2604 2605 if (seg_num == 1) 2606 cnt = len; 2607 else 2608 cnt = max_blocks - GET_BLKOFF_FROM_SEG0(sbi, addr); 2609 2610 do { 2611 update_segment_mtime(sbi, addr_start, 0); 2612 update_sit_entry(sbi, addr_start, -cnt); 2613 2614 /* add it into dirty seglist */ 2615 locate_dirty_segment(sbi, segno); 2616 2617 /* update @addr_start and @cnt and @segno */ 2618 addr_start = START_BLOCK(sbi, ++segno); 2619 if (++i == seg_num) 2620 cnt = GET_BLKOFF_FROM_SEG0(sbi, addr_end) + 1; 2621 else 2622 cnt = max_blocks; 2623 } while (i <= seg_num); 2624 2625 up_write(&sit_i->sentry_lock); 2626 } 2627 2628 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2629 { 2630 struct sit_info *sit_i = SIT_I(sbi); 2631 unsigned int segno, offset; 2632 struct seg_entry *se; 2633 bool is_cp = false; 2634 2635 if (!__is_valid_data_blkaddr(blkaddr)) 2636 return true; 2637 2638 down_read(&sit_i->sentry_lock); 2639 2640 segno = GET_SEGNO(sbi, blkaddr); 2641 se = get_seg_entry(sbi, segno); 2642 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2643 2644 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2645 is_cp = true; 2646 2647 up_read(&sit_i->sentry_lock); 2648 2649 return is_cp; 2650 } 2651 2652 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type) 2653 { 2654 struct curseg_info *curseg = CURSEG_I(sbi, type); 2655 2656 if (sbi->ckpt->alloc_type[type] == SSR) 2657 return BLKS_PER_SEG(sbi); 2658 return curseg->next_blkoff; 2659 } 2660 2661 /* 2662 * Calculate the number of current summary pages for writing 2663 */ 2664 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2665 { 2666 int valid_sum_count = 0; 2667 int i, sum_in_page; 2668 2669 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2670 if (sbi->ckpt->alloc_type[i] != SSR && for_ra) 2671 valid_sum_count += 2672 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2673 else 2674 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i); 2675 } 2676 2677 sum_in_page = (sbi->blocksize - 2 * sbi->sum_journal_size - 2678 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2679 if (valid_sum_count <= sum_in_page) 2680 return 1; 2681 else if ((valid_sum_count - sum_in_page) <= 2682 (sbi->blocksize - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2683 return 2; 2684 return 3; 2685 } 2686 2687 /* 2688 * Caller should put this summary folio 2689 */ 2690 struct folio *f2fs_get_sum_folio(struct f2fs_sb_info *sbi, unsigned int segno) 2691 { 2692 if (unlikely(f2fs_cp_error(sbi))) 2693 return ERR_PTR(-EIO); 2694 return f2fs_get_meta_folio_retry(sbi, GET_SUM_BLOCK(sbi, segno)); 2695 } 2696 2697 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2698 void *src, block_t blk_addr) 2699 { 2700 struct folio *folio; 2701 2702 if (!f2fs_sb_has_packed_ssa(sbi)) 2703 folio = f2fs_grab_meta_folio(sbi, blk_addr); 2704 else 2705 folio = f2fs_get_meta_folio_retry(sbi, blk_addr); 2706 2707 if (IS_ERR(folio)) 2708 return; 2709 2710 memcpy(folio_address(folio), src, PAGE_SIZE); 2711 folio_mark_dirty(folio); 2712 f2fs_folio_put(folio, true); 2713 } 2714 2715 static void write_sum_page(struct f2fs_sb_info *sbi, 2716 struct f2fs_summary_block *sum_blk, unsigned int segno) 2717 { 2718 struct folio *folio; 2719 2720 if (!f2fs_sb_has_packed_ssa(sbi)) 2721 return f2fs_update_meta_page(sbi, (void *)sum_blk, 2722 GET_SUM_BLOCK(sbi, segno)); 2723 2724 folio = f2fs_get_sum_folio(sbi, segno); 2725 if (IS_ERR(folio)) 2726 return; 2727 2728 memcpy(SUM_BLK_PAGE_ADDR(sbi, folio, segno), sum_blk, 2729 sbi->sum_blocksize); 2730 folio_mark_dirty(folio); 2731 f2fs_folio_put(folio, true); 2732 } 2733 2734 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2735 int type, block_t blk_addr) 2736 { 2737 struct curseg_info *curseg = CURSEG_I(sbi, type); 2738 struct folio *folio = f2fs_grab_meta_folio(sbi, blk_addr); 2739 struct f2fs_summary_block *src = curseg->sum_blk; 2740 struct f2fs_summary_block *dst; 2741 2742 dst = folio_address(folio); 2743 memset(dst, 0, PAGE_SIZE); 2744 2745 mutex_lock(&curseg->curseg_mutex); 2746 2747 down_read(&curseg->journal_rwsem); 2748 memcpy(sum_journal(sbi, dst), curseg->journal, sbi->sum_journal_size); 2749 up_read(&curseg->journal_rwsem); 2750 2751 memcpy(sum_entries(dst), sum_entries(src), sbi->sum_entry_size); 2752 memcpy(sum_footer(sbi, dst), sum_footer(sbi, src), SUM_FOOTER_SIZE); 2753 2754 mutex_unlock(&curseg->curseg_mutex); 2755 2756 folio_mark_dirty(folio); 2757 f2fs_folio_put(folio, true); 2758 } 2759 2760 static int is_next_segment_free(struct f2fs_sb_info *sbi, 2761 struct curseg_info *curseg) 2762 { 2763 unsigned int segno = curseg->segno + 1; 2764 struct free_segmap_info *free_i = FREE_I(sbi); 2765 2766 if (segno < MAIN_SEGS(sbi) && segno % SEGS_PER_SEC(sbi)) 2767 return !test_bit(segno, free_i->free_segmap); 2768 return 0; 2769 } 2770 2771 /* 2772 * Find a new segment from the free segments bitmap to right order 2773 * This function should be returned with success, otherwise BUG 2774 */ 2775 static int get_new_segment(struct f2fs_sb_info *sbi, 2776 unsigned int *newseg, bool new_sec, bool pinning) 2777 { 2778 struct free_segmap_info *free_i = FREE_I(sbi); 2779 unsigned int segno, secno, zoneno; 2780 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2781 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2782 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2783 unsigned int alloc_policy = sbi->allocate_section_policy; 2784 unsigned int alloc_hint = sbi->allocate_section_hint; 2785 bool init = true; 2786 int i; 2787 int ret = 0; 2788 2789 spin_lock(&free_i->segmap_lock); 2790 2791 if (time_to_inject(sbi, FAULT_NO_SEGMENT)) { 2792 ret = -ENOSPC; 2793 goto out_unlock; 2794 } 2795 2796 if (!new_sec && ((*newseg + 1) % SEGS_PER_SEC(sbi))) { 2797 segno = find_next_zero_bit(free_i->free_segmap, 2798 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2799 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2800 goto got_it; 2801 } 2802 2803 #ifdef CONFIG_BLK_DEV_ZONED 2804 /* 2805 * If we format f2fs on zoned storage, let's try to get pinned sections 2806 * from beginning of the storage, which should be a conventional one. 2807 */ 2808 if (f2fs_sb_has_blkzoned(sbi)) { 2809 /* Prioritize writing to conventional zones */ 2810 if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_PRIOR_CONV || pinning) 2811 segno = 0; 2812 else 2813 segno = max(sbi->first_seq_zone_segno, *newseg); 2814 hint = GET_SEC_FROM_SEG(sbi, segno); 2815 } 2816 #endif 2817 2818 /* 2819 * Prevent allocate_section_hint from exceeding MAIN_SECS() 2820 * due to desynchronization. 2821 */ 2822 if (alloc_policy != ALLOCATE_FORWARD_NOHINT && 2823 alloc_hint > MAIN_SECS(sbi)) 2824 alloc_hint = MAIN_SECS(sbi); 2825 2826 if (alloc_policy == ALLOCATE_FORWARD_FROM_HINT && 2827 hint < alloc_hint) 2828 hint = alloc_hint; 2829 else if (alloc_policy == ALLOCATE_FORWARD_WITHIN_HINT && 2830 hint >= alloc_hint) 2831 hint = 0; 2832 2833 find_other_zone: 2834 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2835 2836 #ifdef CONFIG_BLK_DEV_ZONED 2837 if (secno >= MAIN_SECS(sbi) && f2fs_sb_has_blkzoned(sbi)) { 2838 /* Write only to sequential zones */ 2839 if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_ONLY_SEQ) { 2840 hint = GET_SEC_FROM_SEG(sbi, sbi->first_seq_zone_segno); 2841 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2842 } else 2843 secno = find_first_zero_bit(free_i->free_secmap, 2844 MAIN_SECS(sbi)); 2845 if (secno >= MAIN_SECS(sbi)) { 2846 ret = -ENOSPC; 2847 f2fs_bug_on(sbi, 1); 2848 goto out_unlock; 2849 } 2850 } 2851 #endif 2852 2853 if (secno >= MAIN_SECS(sbi)) { 2854 secno = find_first_zero_bit(free_i->free_secmap, 2855 MAIN_SECS(sbi)); 2856 if (secno >= MAIN_SECS(sbi)) { 2857 ret = -ENOSPC; 2858 f2fs_bug_on(sbi, !pinning); 2859 goto out_unlock; 2860 } 2861 } 2862 segno = GET_SEG_FROM_SEC(sbi, secno); 2863 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2864 2865 /* give up on finding another zone */ 2866 if (!init) 2867 goto got_it; 2868 if (sbi->secs_per_zone == 1) 2869 goto got_it; 2870 if (zoneno == old_zoneno) 2871 goto got_it; 2872 for (i = 0; i < NR_CURSEG_TYPE; i++) 2873 if (CURSEG_I(sbi, i)->zone == zoneno) 2874 break; 2875 2876 if (i < NR_CURSEG_TYPE) { 2877 /* zone is in user, try another */ 2878 if (zoneno + 1 >= total_zones) 2879 hint = 0; 2880 else 2881 hint = (zoneno + 1) * sbi->secs_per_zone; 2882 init = false; 2883 goto find_other_zone; 2884 } 2885 got_it: 2886 /* set it as dirty segment in free segmap */ 2887 if (test_bit(segno, free_i->free_segmap)) { 2888 ret = -EFSCORRUPTED; 2889 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_CORRUPTED_FREE_BITMAP); 2890 goto out_unlock; 2891 } 2892 2893 /* no free section in conventional device or conventional zone */ 2894 if (new_sec && pinning && 2895 f2fs_is_sequential_zone_area(sbi, START_BLOCK(sbi, segno))) { 2896 ret = -EAGAIN; 2897 goto out_unlock; 2898 } 2899 __set_inuse(sbi, segno); 2900 *newseg = segno; 2901 out_unlock: 2902 spin_unlock(&free_i->segmap_lock); 2903 2904 if (ret == -ENOSPC && !pinning) 2905 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_NO_SEGMENT); 2906 return ret; 2907 } 2908 2909 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2910 { 2911 struct curseg_info *curseg = CURSEG_I(sbi, type); 2912 struct summary_footer *sum_footer; 2913 unsigned short seg_type = curseg->seg_type; 2914 2915 /* only happen when get_new_segment() fails */ 2916 if (curseg->next_segno == NULL_SEGNO) 2917 return; 2918 2919 curseg->inited = true; 2920 curseg->segno = curseg->next_segno; 2921 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2922 curseg->next_blkoff = 0; 2923 curseg->next_segno = NULL_SEGNO; 2924 2925 sum_footer = sum_footer(sbi, curseg->sum_blk); 2926 memset(sum_footer, 0, sizeof(struct summary_footer)); 2927 2928 sanity_check_seg_type(sbi, seg_type); 2929 2930 if (IS_DATASEG(seg_type)) 2931 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2932 if (IS_NODESEG(seg_type)) 2933 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2934 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified); 2935 } 2936 2937 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2938 { 2939 struct curseg_info *curseg = CURSEG_I(sbi, type); 2940 unsigned short seg_type = curseg->seg_type; 2941 2942 sanity_check_seg_type(sbi, seg_type); 2943 if (__is_large_section(sbi)) { 2944 if (f2fs_need_rand_seg(sbi)) { 2945 unsigned int hint = GET_SEC_FROM_SEG(sbi, curseg->segno); 2946 2947 if (GET_SEC_FROM_SEG(sbi, curseg->segno + 1) != hint) 2948 return curseg->segno; 2949 return get_random_u32_inclusive(curseg->segno + 1, 2950 GET_SEG_FROM_SEC(sbi, hint + 1) - 1); 2951 } 2952 return curseg->segno; 2953 } else if (f2fs_need_rand_seg(sbi)) { 2954 return get_random_u32_below(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi)); 2955 } 2956 2957 /* inmem log may not locate on any segment after mount */ 2958 if (!curseg->inited) 2959 return 0; 2960 2961 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2962 return 0; 2963 2964 if (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)) 2965 return 0; 2966 2967 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2968 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2969 2970 /* find segments from 0 to reuse freed segments */ 2971 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2972 return 0; 2973 2974 return curseg->segno; 2975 } 2976 2977 static void reset_curseg_fields(struct curseg_info *curseg) 2978 { 2979 curseg->inited = false; 2980 curseg->segno = NULL_SEGNO; 2981 curseg->next_segno = 0; 2982 } 2983 2984 /* 2985 * Allocate a current working segment. 2986 * This function always allocates a free segment in LFS manner. 2987 */ 2988 static int new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2989 { 2990 struct curseg_info *curseg = CURSEG_I(sbi, type); 2991 unsigned int segno = curseg->segno; 2992 bool pinning = type == CURSEG_COLD_DATA_PINNED; 2993 int ret; 2994 2995 if (curseg->inited) 2996 write_sum_page(sbi, curseg->sum_blk, segno); 2997 2998 segno = __get_next_segno(sbi, type); 2999 ret = get_new_segment(sbi, &segno, new_sec, pinning); 3000 if (ret) { 3001 if (ret == -ENOSPC) 3002 reset_curseg_fields(curseg); 3003 return ret; 3004 } 3005 3006 curseg->next_segno = segno; 3007 reset_curseg(sbi, type, 1); 3008 curseg->alloc_type = LFS; 3009 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 3010 curseg->fragment_remained_chunk = 3011 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 3012 return 0; 3013 } 3014 3015 static int __next_free_blkoff(struct f2fs_sb_info *sbi, 3016 int segno, block_t start) 3017 { 3018 struct seg_entry *se = get_seg_entry(sbi, segno); 3019 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 3020 unsigned long *target_map = SIT_I(sbi)->tmp_map; 3021 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 3022 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 3023 int i; 3024 3025 for (i = 0; i < entries; i++) 3026 target_map[i] = ckpt_map[i] | cur_map[i]; 3027 3028 return __find_rev_next_zero_bit(target_map, BLKS_PER_SEG(sbi), start); 3029 } 3030 3031 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi, 3032 struct curseg_info *seg) 3033 { 3034 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1); 3035 } 3036 3037 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno) 3038 { 3039 return __next_free_blkoff(sbi, segno, 0) < BLKS_PER_SEG(sbi); 3040 } 3041 3042 /* 3043 * This function always allocates a used segment(from dirty seglist) by SSR 3044 * manner, so it should recover the existing segment information of valid blocks 3045 */ 3046 static int change_curseg(struct f2fs_sb_info *sbi, int type) 3047 { 3048 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3049 struct curseg_info *curseg = CURSEG_I(sbi, type); 3050 unsigned int new_segno = curseg->next_segno; 3051 struct f2fs_summary_block *sum_node; 3052 struct folio *sum_folio; 3053 3054 if (curseg->inited) 3055 write_sum_page(sbi, curseg->sum_blk, curseg->segno); 3056 3057 __set_test_and_inuse(sbi, new_segno); 3058 3059 mutex_lock(&dirty_i->seglist_lock); 3060 __remove_dirty_segment(sbi, new_segno, PRE); 3061 __remove_dirty_segment(sbi, new_segno, DIRTY); 3062 mutex_unlock(&dirty_i->seglist_lock); 3063 3064 reset_curseg(sbi, type, 1); 3065 curseg->alloc_type = SSR; 3066 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0); 3067 3068 sum_folio = f2fs_get_sum_folio(sbi, new_segno); 3069 if (IS_ERR(sum_folio)) { 3070 /* GC won't be able to use stale summary pages by cp_error */ 3071 memset(curseg->sum_blk, 0, sbi->sum_entry_size); 3072 return PTR_ERR(sum_folio); 3073 } 3074 sum_node = SUM_BLK_PAGE_ADDR(sbi, sum_folio, new_segno); 3075 memcpy(curseg->sum_blk, sum_node, sbi->sum_entry_size); 3076 f2fs_folio_put(sum_folio, true); 3077 return 0; 3078 } 3079 3080 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 3081 int alloc_mode, unsigned long long age); 3082 3083 static int get_atssr_segment(struct f2fs_sb_info *sbi, int type, 3084 int target_type, int alloc_mode, 3085 unsigned long long age) 3086 { 3087 struct curseg_info *curseg = CURSEG_I(sbi, type); 3088 int ret = 0; 3089 3090 curseg->seg_type = target_type; 3091 3092 if (get_ssr_segment(sbi, type, alloc_mode, age)) { 3093 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno); 3094 3095 curseg->seg_type = se->type; 3096 ret = change_curseg(sbi, type); 3097 } else { 3098 /* allocate cold segment by default */ 3099 curseg->seg_type = CURSEG_COLD_DATA; 3100 ret = new_curseg(sbi, type, true); 3101 } 3102 stat_inc_seg_type(sbi, curseg); 3103 return ret; 3104 } 3105 3106 static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi, bool force) 3107 { 3108 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC); 3109 int ret = 0; 3110 3111 if (!sbi->am.atgc_enabled && !force) 3112 return 0; 3113 3114 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3115 3116 mutex_lock(&curseg->curseg_mutex); 3117 down_write(&SIT_I(sbi)->sentry_lock); 3118 3119 ret = get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, 3120 CURSEG_COLD_DATA, SSR, 0); 3121 3122 up_write(&SIT_I(sbi)->sentry_lock); 3123 mutex_unlock(&curseg->curseg_mutex); 3124 3125 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3126 return ret; 3127 } 3128 3129 int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi) 3130 { 3131 return __f2fs_init_atgc_curseg(sbi, false); 3132 } 3133 3134 int f2fs_reinit_atgc_curseg(struct f2fs_sb_info *sbi) 3135 { 3136 int ret; 3137 3138 if (!test_opt(sbi, ATGC)) 3139 return 0; 3140 if (sbi->am.atgc_enabled) 3141 return 0; 3142 if (le64_to_cpu(F2FS_CKPT(sbi)->elapsed_time) < 3143 sbi->am.age_threshold) 3144 return 0; 3145 3146 ret = __f2fs_init_atgc_curseg(sbi, true); 3147 if (!ret) { 3148 sbi->am.atgc_enabled = true; 3149 f2fs_info(sbi, "reenabled age threshold GC"); 3150 } 3151 return ret; 3152 } 3153 3154 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type) 3155 { 3156 struct curseg_info *curseg = CURSEG_I(sbi, type); 3157 3158 mutex_lock(&curseg->curseg_mutex); 3159 if (!curseg->inited) 3160 goto out; 3161 3162 if (get_valid_blocks(sbi, curseg->segno, false)) { 3163 write_sum_page(sbi, curseg->sum_blk, curseg->segno); 3164 } else { 3165 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 3166 __set_test_and_free(sbi, curseg->segno, true); 3167 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 3168 } 3169 out: 3170 mutex_unlock(&curseg->curseg_mutex); 3171 } 3172 3173 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi) 3174 { 3175 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 3176 3177 if (sbi->am.atgc_enabled) 3178 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 3179 } 3180 3181 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type) 3182 { 3183 struct curseg_info *curseg = CURSEG_I(sbi, type); 3184 3185 mutex_lock(&curseg->curseg_mutex); 3186 if (!curseg->inited) 3187 goto out; 3188 if (get_valid_blocks(sbi, curseg->segno, false)) 3189 goto out; 3190 3191 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 3192 __set_test_and_inuse(sbi, curseg->segno); 3193 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 3194 out: 3195 mutex_unlock(&curseg->curseg_mutex); 3196 } 3197 3198 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi) 3199 { 3200 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 3201 3202 if (sbi->am.atgc_enabled) 3203 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 3204 } 3205 3206 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 3207 int alloc_mode, unsigned long long age) 3208 { 3209 struct curseg_info *curseg = CURSEG_I(sbi, type); 3210 unsigned segno = NULL_SEGNO; 3211 unsigned short seg_type = curseg->seg_type; 3212 int i, cnt; 3213 bool reversed = false; 3214 3215 sanity_check_seg_type(sbi, seg_type); 3216 3217 /* f2fs_need_SSR() already forces to do this */ 3218 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, 3219 alloc_mode, age, false)) { 3220 curseg->next_segno = segno; 3221 return 1; 3222 } 3223 3224 /* For node segments, let's do SSR more intensively */ 3225 if (IS_NODESEG(seg_type)) { 3226 if (seg_type >= CURSEG_WARM_NODE) { 3227 reversed = true; 3228 i = CURSEG_COLD_NODE; 3229 } else { 3230 i = CURSEG_HOT_NODE; 3231 } 3232 cnt = NR_CURSEG_NODE_TYPE; 3233 } else { 3234 if (seg_type >= CURSEG_WARM_DATA) { 3235 reversed = true; 3236 i = CURSEG_COLD_DATA; 3237 } else { 3238 i = CURSEG_HOT_DATA; 3239 } 3240 cnt = NR_CURSEG_DATA_TYPE; 3241 } 3242 3243 for (; cnt-- > 0; reversed ? i-- : i++) { 3244 if (i == seg_type) 3245 continue; 3246 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, 3247 alloc_mode, age, false)) { 3248 curseg->next_segno = segno; 3249 return 1; 3250 } 3251 } 3252 3253 /* find valid_blocks=0 in dirty list */ 3254 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 3255 segno = get_free_segment(sbi); 3256 if (segno != NULL_SEGNO) { 3257 curseg->next_segno = segno; 3258 return 1; 3259 } 3260 } 3261 return 0; 3262 } 3263 3264 static bool need_new_seg(struct f2fs_sb_info *sbi, int type) 3265 { 3266 struct curseg_info *curseg = CURSEG_I(sbi, type); 3267 3268 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 3269 curseg->seg_type == CURSEG_WARM_NODE) 3270 return true; 3271 if (curseg->alloc_type == LFS && is_next_segment_free(sbi, curseg) && 3272 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 3273 return true; 3274 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0)) 3275 return true; 3276 return false; 3277 } 3278 3279 int f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 3280 unsigned int start, unsigned int end) 3281 { 3282 struct curseg_info *curseg = CURSEG_I(sbi, type); 3283 unsigned int segno; 3284 int ret = 0; 3285 3286 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3287 mutex_lock(&curseg->curseg_mutex); 3288 down_write(&SIT_I(sbi)->sentry_lock); 3289 3290 segno = CURSEG_I(sbi, type)->segno; 3291 if (segno < start || segno > end) 3292 goto unlock; 3293 3294 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0)) 3295 ret = change_curseg(sbi, type); 3296 else 3297 ret = new_curseg(sbi, type, true); 3298 3299 stat_inc_seg_type(sbi, curseg); 3300 3301 locate_dirty_segment(sbi, segno); 3302 unlock: 3303 up_write(&SIT_I(sbi)->sentry_lock); 3304 3305 if (segno != curseg->segno) 3306 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 3307 type, segno, curseg->segno); 3308 3309 mutex_unlock(&curseg->curseg_mutex); 3310 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3311 return ret; 3312 } 3313 3314 static int __allocate_new_segment(struct f2fs_sb_info *sbi, int type, 3315 bool new_sec, bool force) 3316 { 3317 struct curseg_info *curseg = CURSEG_I(sbi, type); 3318 unsigned int old_segno; 3319 int err = 0; 3320 3321 if (type == CURSEG_COLD_DATA_PINNED && !curseg->inited) 3322 goto allocate; 3323 3324 if (!force && curseg->inited && 3325 !curseg->next_blkoff && 3326 !get_valid_blocks(sbi, curseg->segno, new_sec) && 3327 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec)) 3328 return 0; 3329 3330 allocate: 3331 old_segno = curseg->segno; 3332 err = new_curseg(sbi, type, true); 3333 if (err) 3334 return err; 3335 stat_inc_seg_type(sbi, curseg); 3336 locate_dirty_segment(sbi, old_segno); 3337 return 0; 3338 } 3339 3340 int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force) 3341 { 3342 int ret; 3343 3344 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3345 down_write(&SIT_I(sbi)->sentry_lock); 3346 ret = __allocate_new_segment(sbi, type, true, force); 3347 up_write(&SIT_I(sbi)->sentry_lock); 3348 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3349 3350 return ret; 3351 } 3352 3353 int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi) 3354 { 3355 struct f2fs_lock_context lc; 3356 int err; 3357 bool gc_required = true; 3358 3359 retry: 3360 f2fs_lock_op(sbi, &lc); 3361 err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false); 3362 f2fs_unlock_op(sbi, &lc); 3363 3364 if (f2fs_sb_has_blkzoned(sbi) && err == -EAGAIN && gc_required) { 3365 f2fs_down_write_trace(&sbi->gc_lock, &lc); 3366 err = f2fs_gc_range(sbi, 0, sbi->first_seq_zone_segno - 1, 3367 true, ZONED_PIN_SEC_REQUIRED_COUNT); 3368 f2fs_up_write_trace(&sbi->gc_lock, &lc); 3369 3370 gc_required = false; 3371 if (!err) 3372 goto retry; 3373 } 3374 3375 return err; 3376 } 3377 3378 int f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 3379 { 3380 int i; 3381 int err = 0; 3382 3383 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3384 down_write(&SIT_I(sbi)->sentry_lock); 3385 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 3386 err += __allocate_new_segment(sbi, i, false, false); 3387 up_write(&SIT_I(sbi)->sentry_lock); 3388 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3389 3390 return err; 3391 } 3392 3393 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 3394 struct cp_control *cpc) 3395 { 3396 __u64 trim_start = cpc->trim_start; 3397 bool has_candidate = false; 3398 3399 down_write(&SIT_I(sbi)->sentry_lock); 3400 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 3401 if (add_discard_addrs(sbi, cpc, true)) { 3402 has_candidate = true; 3403 break; 3404 } 3405 } 3406 up_write(&SIT_I(sbi)->sentry_lock); 3407 3408 cpc->trim_start = trim_start; 3409 return has_candidate; 3410 } 3411 3412 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 3413 struct discard_policy *dpolicy, 3414 unsigned int start, unsigned int end) 3415 { 3416 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 3417 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 3418 struct rb_node **insert_p = NULL, *insert_parent = NULL; 3419 struct discard_cmd *dc; 3420 struct blk_plug plug; 3421 int issued; 3422 unsigned int trimmed = 0; 3423 3424 next: 3425 issued = 0; 3426 3427 mutex_lock(&dcc->cmd_lock); 3428 if (unlikely(dcc->rbtree_check)) 3429 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 3430 3431 dc = __lookup_discard_cmd_ret(&dcc->root, start, 3432 &prev_dc, &next_dc, &insert_p, &insert_parent); 3433 if (!dc) 3434 dc = next_dc; 3435 3436 blk_start_plug(&plug); 3437 3438 while (dc && dc->di.lstart <= end) { 3439 struct rb_node *node; 3440 int err = 0; 3441 3442 if (dc->di.len < dpolicy->granularity) 3443 goto skip; 3444 3445 if (dc->state != D_PREP) { 3446 list_move_tail(&dc->list, &dcc->fstrim_list); 3447 goto skip; 3448 } 3449 3450 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 3451 3452 if (issued >= dpolicy->max_requests) { 3453 start = dc->di.lstart + dc->di.len; 3454 3455 if (err) 3456 __remove_discard_cmd(sbi, dc); 3457 3458 blk_finish_plug(&plug); 3459 mutex_unlock(&dcc->cmd_lock); 3460 trimmed += __wait_all_discard_cmd(sbi, NULL); 3461 f2fs_schedule_timeout(DEFAULT_DISCARD_INTERVAL); 3462 goto next; 3463 } 3464 skip: 3465 node = rb_next(&dc->rb_node); 3466 if (err) 3467 __remove_discard_cmd(sbi, dc); 3468 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 3469 3470 if (fatal_signal_pending(current)) 3471 break; 3472 } 3473 3474 blk_finish_plug(&plug); 3475 mutex_unlock(&dcc->cmd_lock); 3476 3477 return trimmed; 3478 } 3479 3480 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 3481 { 3482 __u64 start = F2FS_BYTES_TO_BLK(range->start); 3483 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 3484 unsigned int start_segno, end_segno; 3485 block_t start_block, end_block; 3486 struct cp_control cpc; 3487 struct discard_policy dpolicy; 3488 struct f2fs_lock_context lc; 3489 unsigned long long trimmed = 0; 3490 int err = 0; 3491 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 3492 3493 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 3494 return -EINVAL; 3495 3496 if (end < MAIN_BLKADDR(sbi)) 3497 goto out; 3498 3499 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 3500 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 3501 return -EFSCORRUPTED; 3502 } 3503 3504 /* start/end segment number in main_area */ 3505 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 3506 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 3507 GET_SEGNO(sbi, end); 3508 if (need_align) { 3509 start_segno = rounddown(start_segno, SEGS_PER_SEC(sbi)); 3510 end_segno = roundup(end_segno + 1, SEGS_PER_SEC(sbi)) - 1; 3511 } 3512 3513 cpc.reason = CP_DISCARD; 3514 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 3515 cpc.trim_start = start_segno; 3516 cpc.trim_end = end_segno; 3517 3518 if (sbi->discard_blks == 0) 3519 goto out; 3520 3521 f2fs_down_write_trace(&sbi->gc_lock, &lc); 3522 stat_inc_cp_call_count(sbi, TOTAL_CALL); 3523 err = f2fs_write_checkpoint(sbi, &cpc); 3524 f2fs_up_write_trace(&sbi->gc_lock, &lc); 3525 if (err) 3526 goto out; 3527 3528 /* 3529 * We filed discard candidates, but actually we don't need to wait for 3530 * all of them, since they'll be issued in idle time along with runtime 3531 * discard option. User configuration looks like using runtime discard 3532 * or periodic fstrim instead of it. 3533 */ 3534 if (f2fs_realtime_discard_enable(sbi)) 3535 goto out; 3536 3537 start_block = START_BLOCK(sbi, start_segno); 3538 end_block = START_BLOCK(sbi, end_segno + 1); 3539 3540 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 3541 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 3542 start_block, end_block); 3543 3544 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 3545 start_block, end_block); 3546 out: 3547 if (!err) 3548 range->len = F2FS_BLK_TO_BYTES(trimmed); 3549 return err; 3550 } 3551 3552 int f2fs_rw_hint_to_seg_type(struct f2fs_sb_info *sbi, enum rw_hint hint) 3553 { 3554 if (F2FS_OPTION(sbi).active_logs == 2) 3555 return CURSEG_HOT_DATA; 3556 else if (F2FS_OPTION(sbi).active_logs == 4) 3557 return CURSEG_COLD_DATA; 3558 3559 /* active_log == 6 */ 3560 switch (hint) { 3561 case WRITE_LIFE_SHORT: 3562 return CURSEG_HOT_DATA; 3563 case WRITE_LIFE_EXTREME: 3564 return CURSEG_COLD_DATA; 3565 default: 3566 return CURSEG_WARM_DATA; 3567 } 3568 } 3569 3570 /* 3571 * This returns write hints for each segment type. This hints will be 3572 * passed down to block layer as below by default. 3573 * 3574 * User F2FS Block 3575 * ---- ---- ----- 3576 * META WRITE_LIFE_NONE|REQ_META 3577 * HOT_NODE WRITE_LIFE_NONE 3578 * WARM_NODE WRITE_LIFE_MEDIUM 3579 * COLD_NODE WRITE_LIFE_LONG 3580 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 3581 * extension list " " 3582 * 3583 * -- buffered io 3584 * COLD_DATA WRITE_LIFE_EXTREME 3585 * HOT_DATA WRITE_LIFE_SHORT 3586 * WARM_DATA WRITE_LIFE_NOT_SET 3587 * 3588 * -- direct io 3589 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 3590 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 3591 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 3592 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 3593 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 3594 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 3595 */ 3596 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi, 3597 enum page_type type, enum temp_type temp) 3598 { 3599 switch (type) { 3600 case DATA: 3601 switch (temp) { 3602 case WARM: 3603 return WRITE_LIFE_NOT_SET; 3604 case HOT: 3605 return WRITE_LIFE_SHORT; 3606 case COLD: 3607 return WRITE_LIFE_EXTREME; 3608 default: 3609 return WRITE_LIFE_NONE; 3610 } 3611 case NODE: 3612 switch (temp) { 3613 case WARM: 3614 return WRITE_LIFE_MEDIUM; 3615 case HOT: 3616 return WRITE_LIFE_NONE; 3617 case COLD: 3618 return WRITE_LIFE_LONG; 3619 default: 3620 return WRITE_LIFE_NONE; 3621 } 3622 case META: 3623 return WRITE_LIFE_NONE; 3624 default: 3625 return WRITE_LIFE_NONE; 3626 } 3627 } 3628 3629 static int __get_segment_type_2(struct f2fs_io_info *fio) 3630 { 3631 if (fio->type == DATA) 3632 return CURSEG_HOT_DATA; 3633 else 3634 return CURSEG_HOT_NODE; 3635 } 3636 3637 static int __get_segment_type_4(struct f2fs_io_info *fio) 3638 { 3639 if (fio->type == DATA) { 3640 struct inode *inode = fio_inode(fio); 3641 3642 if (S_ISDIR(inode->i_mode)) 3643 return CURSEG_HOT_DATA; 3644 else 3645 return CURSEG_COLD_DATA; 3646 } else { 3647 if (IS_DNODE(fio->folio) && is_cold_node(fio->folio)) 3648 return CURSEG_WARM_NODE; 3649 else 3650 return CURSEG_COLD_NODE; 3651 } 3652 } 3653 3654 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs) 3655 { 3656 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3657 struct extent_info ei = {}; 3658 3659 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) { 3660 if (!ei.age) 3661 return NO_CHECK_TYPE; 3662 if (ei.age <= sbi->hot_data_age_threshold) 3663 return CURSEG_HOT_DATA; 3664 if (ei.age <= sbi->warm_data_age_threshold) 3665 return CURSEG_WARM_DATA; 3666 return CURSEG_COLD_DATA; 3667 } 3668 return NO_CHECK_TYPE; 3669 } 3670 3671 static int __get_segment_type_6(struct f2fs_io_info *fio) 3672 { 3673 if (fio->type == DATA) { 3674 struct inode *inode = fio_inode(fio); 3675 int type; 3676 3677 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 3678 return CURSEG_COLD_DATA_PINNED; 3679 3680 if (page_private_gcing(fio->page)) { 3681 if (fio->sbi->am.atgc_enabled && 3682 (fio->io_type == FS_DATA_IO) && 3683 (fio->sbi->gc_mode != GC_URGENT_HIGH) && 3684 __is_valid_data_blkaddr(fio->old_blkaddr) && 3685 !is_inode_flag_set(inode, FI_OPU_WRITE)) 3686 return CURSEG_ALL_DATA_ATGC; 3687 else 3688 return CURSEG_COLD_DATA; 3689 } 3690 if (file_is_cold(inode) || f2fs_need_compress_data(inode)) 3691 return CURSEG_COLD_DATA; 3692 3693 type = __get_age_segment_type(inode, fio->folio->index); 3694 if (type != NO_CHECK_TYPE) 3695 return type; 3696 3697 if (file_is_hot(inode) || 3698 is_inode_flag_set(inode, FI_HOT_DATA) || 3699 f2fs_is_cow_file(inode) || 3700 is_inode_flag_set(inode, FI_NEED_IPU)) 3701 return CURSEG_HOT_DATA; 3702 return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode), 3703 inode->i_write_hint); 3704 } else { 3705 if (IS_DNODE(fio->folio)) 3706 return is_cold_node(fio->folio) ? CURSEG_WARM_NODE : 3707 CURSEG_HOT_NODE; 3708 return CURSEG_COLD_NODE; 3709 } 3710 } 3711 3712 enum temp_type f2fs_get_segment_temp(struct f2fs_sb_info *sbi, 3713 enum log_type type) 3714 { 3715 struct curseg_info *curseg = CURSEG_I(sbi, type); 3716 enum temp_type temp = COLD; 3717 3718 switch (curseg->seg_type) { 3719 case CURSEG_HOT_NODE: 3720 case CURSEG_HOT_DATA: 3721 temp = HOT; 3722 break; 3723 case CURSEG_WARM_NODE: 3724 case CURSEG_WARM_DATA: 3725 temp = WARM; 3726 break; 3727 case CURSEG_COLD_NODE: 3728 case CURSEG_COLD_DATA: 3729 temp = COLD; 3730 break; 3731 default: 3732 f2fs_bug_on(sbi, 1); 3733 } 3734 3735 return temp; 3736 } 3737 3738 static int __get_segment_type(struct f2fs_io_info *fio) 3739 { 3740 enum log_type type = CURSEG_HOT_DATA; 3741 3742 switch (F2FS_OPTION(fio->sbi).active_logs) { 3743 case 2: 3744 type = __get_segment_type_2(fio); 3745 break; 3746 case 4: 3747 type = __get_segment_type_4(fio); 3748 break; 3749 case 6: 3750 type = __get_segment_type_6(fio); 3751 break; 3752 default: 3753 f2fs_bug_on(fio->sbi, true); 3754 } 3755 3756 fio->temp = f2fs_get_segment_temp(fio->sbi, type); 3757 3758 return type; 3759 } 3760 3761 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi, 3762 struct curseg_info *seg) 3763 { 3764 /* To allocate block chunks in different sizes, use random number */ 3765 if (--seg->fragment_remained_chunk > 0) 3766 return; 3767 3768 seg->fragment_remained_chunk = 3769 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 3770 seg->next_blkoff += 3771 get_random_u32_inclusive(1, sbi->max_fragment_hole); 3772 } 3773 3774 int f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct folio *folio, 3775 block_t old_blkaddr, block_t *new_blkaddr, 3776 struct f2fs_summary *sum, int type, 3777 struct f2fs_io_info *fio) 3778 { 3779 struct sit_info *sit_i = SIT_I(sbi); 3780 struct curseg_info *curseg = CURSEG_I(sbi, type); 3781 unsigned long long old_mtime; 3782 bool from_gc = (type == CURSEG_ALL_DATA_ATGC); 3783 struct seg_entry *se = NULL; 3784 bool segment_full = false; 3785 int ret = 0; 3786 3787 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3788 3789 mutex_lock(&curseg->curseg_mutex); 3790 down_write(&sit_i->sentry_lock); 3791 3792 if (curseg->segno == NULL_SEGNO) { 3793 ret = -ENOSPC; 3794 goto out_err; 3795 } 3796 3797 if (from_gc) { 3798 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO); 3799 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr)); 3800 sanity_check_seg_type(sbi, se->type); 3801 f2fs_bug_on(sbi, IS_NODESEG(se->type)); 3802 } 3803 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3804 3805 f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi)); 3806 3807 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3808 3809 sum_entries(curseg->sum_blk)[curseg->next_blkoff] = *sum; 3810 if (curseg->alloc_type == SSR) { 3811 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg); 3812 } else { 3813 curseg->next_blkoff++; 3814 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 3815 f2fs_randomize_chunk(sbi, curseg); 3816 } 3817 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno)) 3818 segment_full = true; 3819 stat_inc_block_count(sbi, curseg); 3820 3821 if (from_gc) { 3822 old_mtime = get_segment_mtime(sbi, old_blkaddr); 3823 } else { 3824 update_segment_mtime(sbi, old_blkaddr, 0); 3825 old_mtime = 0; 3826 } 3827 update_segment_mtime(sbi, *new_blkaddr, old_mtime); 3828 3829 /* 3830 * SIT information should be updated before segment allocation, 3831 * since SSR needs latest valid block information. 3832 */ 3833 update_sit_entry(sbi, *new_blkaddr, 1); 3834 update_sit_entry(sbi, old_blkaddr, -1); 3835 3836 /* 3837 * If the current segment is full, flush it out and replace it with a 3838 * new segment. 3839 */ 3840 if (segment_full) { 3841 if (type == CURSEG_COLD_DATA_PINNED && 3842 !((curseg->segno + 1) % sbi->segs_per_sec)) { 3843 write_sum_page(sbi, curseg->sum_blk, curseg->segno); 3844 reset_curseg_fields(curseg); 3845 goto skip_new_segment; 3846 } 3847 3848 if (from_gc) { 3849 ret = get_atssr_segment(sbi, type, se->type, 3850 AT_SSR, se->mtime); 3851 } else { 3852 if (need_new_seg(sbi, type)) 3853 ret = new_curseg(sbi, type, false); 3854 else 3855 ret = change_curseg(sbi, type); 3856 stat_inc_seg_type(sbi, curseg); 3857 } 3858 3859 if (ret) 3860 goto out_err; 3861 } 3862 3863 skip_new_segment: 3864 /* 3865 * segment dirty status should be updated after segment allocation, 3866 * so we just need to update status only one time after previous 3867 * segment being closed. 3868 */ 3869 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3870 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3871 3872 if (IS_DATASEG(curseg->seg_type)) { 3873 unsigned long long new_val; 3874 3875 new_val = atomic64_inc_return(&sbi->allocated_data_blocks); 3876 if (unlikely(new_val == ULLONG_MAX)) 3877 atomic64_set(&sbi->allocated_data_blocks, 0); 3878 } 3879 3880 up_write(&sit_i->sentry_lock); 3881 3882 if (folio && IS_NODESEG(curseg->seg_type)) { 3883 fill_node_footer_blkaddr(folio, NEXT_FREE_BLKADDR(sbi, curseg)); 3884 3885 f2fs_inode_chksum_set(sbi, folio); 3886 } 3887 3888 if (fio) { 3889 struct f2fs_bio_info *io; 3890 3891 INIT_LIST_HEAD(&fio->list); 3892 fio->in_list = 1; 3893 io = sbi->write_io[fio->type] + fio->temp; 3894 spin_lock(&io->io_lock); 3895 list_add_tail(&fio->list, &io->io_list); 3896 spin_unlock(&io->io_lock); 3897 } 3898 3899 mutex_unlock(&curseg->curseg_mutex); 3900 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3901 return 0; 3902 3903 out_err: 3904 *new_blkaddr = NULL_ADDR; 3905 up_write(&sit_i->sentry_lock); 3906 mutex_unlock(&curseg->curseg_mutex); 3907 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3908 return ret; 3909 } 3910 3911 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino, 3912 block_t blkaddr, unsigned int blkcnt) 3913 { 3914 if (!f2fs_is_multi_device(sbi)) 3915 return; 3916 3917 while (1) { 3918 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr); 3919 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1; 3920 3921 /* update device state for fsync */ 3922 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO); 3923 3924 /* update device state for checkpoint */ 3925 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3926 spin_lock(&sbi->dev_lock); 3927 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3928 spin_unlock(&sbi->dev_lock); 3929 } 3930 3931 if (blkcnt <= blks) 3932 break; 3933 blkcnt -= blks; 3934 blkaddr += blks; 3935 } 3936 } 3937 3938 static int log_type_to_seg_type(enum log_type type) 3939 { 3940 int seg_type = CURSEG_COLD_DATA; 3941 3942 switch (type) { 3943 case CURSEG_HOT_DATA: 3944 case CURSEG_WARM_DATA: 3945 case CURSEG_COLD_DATA: 3946 case CURSEG_HOT_NODE: 3947 case CURSEG_WARM_NODE: 3948 case CURSEG_COLD_NODE: 3949 seg_type = (int)type; 3950 break; 3951 case CURSEG_COLD_DATA_PINNED: 3952 case CURSEG_ALL_DATA_ATGC: 3953 seg_type = CURSEG_COLD_DATA; 3954 break; 3955 default: 3956 break; 3957 } 3958 return seg_type; 3959 } 3960 3961 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3962 { 3963 struct folio *folio = fio->folio; 3964 enum log_type type = __get_segment_type(fio); 3965 int seg_type = log_type_to_seg_type(type); 3966 bool keep_order = (f2fs_lfs_mode(fio->sbi) && 3967 seg_type == CURSEG_COLD_DATA); 3968 int err; 3969 3970 if (keep_order) 3971 f2fs_down_read(&fio->sbi->io_order_lock); 3972 3973 err = f2fs_allocate_data_block(fio->sbi, folio, fio->old_blkaddr, 3974 &fio->new_blkaddr, sum, type, fio); 3975 if (unlikely(err)) { 3976 f2fs_err_ratelimited(fio->sbi, 3977 "%s Failed to allocate data block, ino:%u, index:%lu, type:%d, old_blkaddr:0x%x, new_blkaddr:0x%x, err:%d", 3978 __func__, fio->ino, folio->index, type, 3979 fio->old_blkaddr, fio->new_blkaddr, err); 3980 if (fscrypt_inode_uses_fs_layer_crypto(folio->mapping->host)) 3981 fscrypt_finalize_bounce_page(&fio->encrypted_page); 3982 folio_end_writeback(folio); 3983 if (f2fs_in_warm_node_list(fio->sbi, folio)) 3984 f2fs_del_fsync_node_entry(fio->sbi, folio); 3985 f2fs_bug_on(fio->sbi, !is_set_ckpt_flags(fio->sbi, 3986 CP_ERROR_FLAG)); 3987 goto out; 3988 } 3989 3990 f2fs_bug_on(fio->sbi, !f2fs_is_valid_blkaddr_raw(fio->sbi, 3991 fio->new_blkaddr, DATA_GENERIC_ENHANCE)); 3992 3993 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) 3994 f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr, 1); 3995 3996 /* writeout dirty page into bdev */ 3997 f2fs_submit_page_write(fio); 3998 3999 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1); 4000 out: 4001 if (keep_order) 4002 f2fs_up_read(&fio->sbi->io_order_lock); 4003 } 4004 4005 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct folio *folio, 4006 enum iostat_type io_type) 4007 { 4008 struct f2fs_io_info fio = { 4009 .sbi = sbi, 4010 .type = META, 4011 .temp = HOT, 4012 .op = REQ_OP_WRITE, 4013 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 4014 .old_blkaddr = folio->index, 4015 .new_blkaddr = folio->index, 4016 .folio = folio, 4017 .encrypted_page = NULL, 4018 .in_list = 0, 4019 }; 4020 4021 if (unlikely(folio->index >= MAIN_BLKADDR(sbi))) 4022 fio.op_flags &= ~REQ_META; 4023 4024 folio_start_writeback(folio); 4025 f2fs_submit_page_write(&fio); 4026 4027 stat_inc_meta_count(sbi, folio->index); 4028 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE); 4029 } 4030 4031 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 4032 { 4033 struct f2fs_summary sum; 4034 4035 set_summary(&sum, nid, 0, 0); 4036 do_write_page(&sum, fio); 4037 4038 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE); 4039 } 4040 4041 void f2fs_outplace_write_data(struct dnode_of_data *dn, 4042 struct f2fs_io_info *fio) 4043 { 4044 struct f2fs_sb_info *sbi = fio->sbi; 4045 struct f2fs_summary sum; 4046 4047 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 4048 if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO) 4049 f2fs_update_age_extent_cache(dn); 4050 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 4051 do_write_page(&sum, fio); 4052 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 4053 4054 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE); 4055 } 4056 4057 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 4058 { 4059 int err; 4060 struct f2fs_sb_info *sbi = fio->sbi; 4061 unsigned int segno; 4062 4063 fio->new_blkaddr = fio->old_blkaddr; 4064 /* i/o temperature is needed for passing down write hints */ 4065 __get_segment_type(fio); 4066 4067 segno = GET_SEGNO(sbi, fio->new_blkaddr); 4068 4069 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 4070 set_sbi_flag(sbi, SBI_NEED_FSCK); 4071 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 4072 __func__, segno); 4073 err = -EFSCORRUPTED; 4074 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 4075 goto drop_bio; 4076 } 4077 4078 if (f2fs_cp_error(sbi)) { 4079 err = -EIO; 4080 goto drop_bio; 4081 } 4082 4083 if (fio->meta_gc) 4084 f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1); 4085 4086 stat_inc_inplace_blocks(fio->sbi); 4087 4088 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi)) 4089 err = f2fs_merge_page_bio(fio); 4090 else 4091 err = f2fs_submit_page_bio(fio); 4092 if (!err) { 4093 f2fs_update_device_state(fio->sbi, fio->ino, 4094 fio->new_blkaddr, 1); 4095 f2fs_update_iostat(fio->sbi, fio_inode(fio), 4096 fio->io_type, F2FS_BLKSIZE); 4097 } 4098 4099 return err; 4100 drop_bio: 4101 if (fio->bio && *(fio->bio)) { 4102 struct bio *bio = *(fio->bio); 4103 4104 bio->bi_status = BLK_STS_IOERR; 4105 bio_endio(bio); 4106 *(fio->bio) = NULL; 4107 } 4108 return err; 4109 } 4110 4111 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 4112 unsigned int segno) 4113 { 4114 int i; 4115 4116 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 4117 if (CURSEG_I(sbi, i)->segno == segno) 4118 break; 4119 } 4120 return i; 4121 } 4122 4123 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 4124 block_t old_blkaddr, block_t new_blkaddr, 4125 bool recover_curseg, bool recover_newaddr, 4126 bool from_gc) 4127 { 4128 struct sit_info *sit_i = SIT_I(sbi); 4129 struct curseg_info *curseg; 4130 unsigned int segno, old_cursegno; 4131 struct seg_entry *se; 4132 int type; 4133 unsigned short old_blkoff; 4134 unsigned char old_alloc_type; 4135 4136 segno = GET_SEGNO(sbi, new_blkaddr); 4137 se = get_seg_entry(sbi, segno); 4138 type = se->type; 4139 4140 f2fs_down_write(&SM_I(sbi)->curseg_lock); 4141 4142 if (!recover_curseg) { 4143 /* for recovery flow */ 4144 if (se->valid_blocks == 0 && !is_curseg(sbi, segno)) { 4145 if (old_blkaddr == NULL_ADDR) 4146 type = CURSEG_COLD_DATA; 4147 else 4148 type = CURSEG_WARM_DATA; 4149 } 4150 } else { 4151 if (is_curseg(sbi, segno)) { 4152 /* se->type is volatile as SSR allocation */ 4153 type = __f2fs_get_curseg(sbi, segno); 4154 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 4155 } else { 4156 type = CURSEG_WARM_DATA; 4157 } 4158 } 4159 4160 curseg = CURSEG_I(sbi, type); 4161 f2fs_bug_on(sbi, !IS_DATASEG(curseg->seg_type)); 4162 4163 mutex_lock(&curseg->curseg_mutex); 4164 down_write(&sit_i->sentry_lock); 4165 4166 old_cursegno = curseg->segno; 4167 old_blkoff = curseg->next_blkoff; 4168 old_alloc_type = curseg->alloc_type; 4169 4170 /* change the current segment */ 4171 if (segno != curseg->segno) { 4172 curseg->next_segno = segno; 4173 if (change_curseg(sbi, type)) 4174 goto out_unlock; 4175 } 4176 4177 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 4178 sum_entries(curseg->sum_blk)[curseg->next_blkoff] = *sum; 4179 4180 if (!recover_curseg || recover_newaddr) { 4181 if (!from_gc) 4182 update_segment_mtime(sbi, new_blkaddr, 0); 4183 update_sit_entry(sbi, new_blkaddr, 1); 4184 } 4185 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 4186 f2fs_invalidate_internal_cache(sbi, old_blkaddr, 1); 4187 if (!from_gc) 4188 update_segment_mtime(sbi, old_blkaddr, 0); 4189 update_sit_entry(sbi, old_blkaddr, -1); 4190 } 4191 4192 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 4193 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 4194 4195 locate_dirty_segment(sbi, old_cursegno); 4196 4197 if (recover_curseg) { 4198 if (old_cursegno != curseg->segno) { 4199 curseg->next_segno = old_cursegno; 4200 if (change_curseg(sbi, type)) 4201 goto out_unlock; 4202 } 4203 curseg->next_blkoff = old_blkoff; 4204 curseg->alloc_type = old_alloc_type; 4205 } 4206 4207 out_unlock: 4208 up_write(&sit_i->sentry_lock); 4209 mutex_unlock(&curseg->curseg_mutex); 4210 f2fs_up_write(&SM_I(sbi)->curseg_lock); 4211 } 4212 4213 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 4214 block_t old_addr, block_t new_addr, 4215 unsigned char version, bool recover_curseg, 4216 bool recover_newaddr) 4217 { 4218 struct f2fs_summary sum; 4219 4220 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 4221 4222 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 4223 recover_curseg, recover_newaddr, false); 4224 4225 f2fs_update_data_blkaddr(dn, new_addr); 4226 } 4227 4228 void f2fs_folio_wait_writeback(struct folio *folio, enum page_type type, 4229 bool ordered, bool locked) 4230 { 4231 if (folio_test_writeback(folio)) { 4232 struct f2fs_sb_info *sbi = F2FS_F_SB(folio); 4233 4234 /* submit cached LFS IO */ 4235 f2fs_submit_merged_write_folio(sbi, folio, type); 4236 /* submit cached IPU IO */ 4237 f2fs_submit_merged_ipu_write(sbi, NULL, folio); 4238 if (ordered) { 4239 folio_wait_writeback(folio); 4240 f2fs_bug_on(sbi, locked && folio_test_writeback(folio)); 4241 } else { 4242 folio_wait_stable(folio); 4243 } 4244 } 4245 } 4246 4247 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 4248 { 4249 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4250 struct folio *cfolio; 4251 4252 if (!f2fs_meta_inode_gc_required(inode)) 4253 return; 4254 4255 if (!__is_valid_data_blkaddr(blkaddr)) 4256 return; 4257 4258 cfolio = filemap_lock_folio(META_MAPPING(sbi), blkaddr); 4259 if (!IS_ERR(cfolio)) { 4260 f2fs_folio_wait_writeback(cfolio, DATA, true, true); 4261 f2fs_folio_put(cfolio, true); 4262 } 4263 } 4264 4265 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 4266 block_t len) 4267 { 4268 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4269 block_t i; 4270 4271 if (!f2fs_meta_inode_gc_required(inode)) 4272 return; 4273 4274 for (i = 0; i < len; i++) 4275 f2fs_wait_on_block_writeback(inode, blkaddr + i); 4276 4277 f2fs_truncate_meta_inode_pages(sbi, blkaddr, len); 4278 } 4279 4280 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 4281 { 4282 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 4283 struct curseg_info *seg_i; 4284 unsigned char *kaddr; 4285 struct folio *folio; 4286 block_t start; 4287 int i, j, offset; 4288 4289 start = start_sum_block(sbi); 4290 4291 folio = f2fs_get_meta_folio(sbi, start++); 4292 if (IS_ERR(folio)) 4293 return PTR_ERR(folio); 4294 kaddr = folio_address(folio); 4295 4296 /* Step 1: restore nat cache */ 4297 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4298 memcpy(seg_i->journal, kaddr, sbi->sum_journal_size); 4299 4300 /* Step 2: restore sit cache */ 4301 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4302 memcpy(seg_i->journal, kaddr + sbi->sum_journal_size, sbi->sum_journal_size); 4303 offset = 2 * sbi->sum_journal_size; 4304 4305 /* Step 3: restore summary entries */ 4306 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4307 unsigned short blk_off; 4308 unsigned int segno; 4309 4310 seg_i = CURSEG_I(sbi, i); 4311 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 4312 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 4313 seg_i->next_segno = segno; 4314 reset_curseg(sbi, i, 0); 4315 seg_i->alloc_type = ckpt->alloc_type[i]; 4316 seg_i->next_blkoff = blk_off; 4317 4318 if (seg_i->alloc_type == SSR) 4319 blk_off = BLKS_PER_SEG(sbi); 4320 4321 for (j = 0; j < blk_off; j++) { 4322 struct f2fs_summary *s; 4323 4324 s = (struct f2fs_summary *)(kaddr + offset); 4325 sum_entries(seg_i->sum_blk)[j] = *s; 4326 offset += SUMMARY_SIZE; 4327 if (offset + SUMMARY_SIZE <= sbi->blocksize - 4328 SUM_FOOTER_SIZE) 4329 continue; 4330 4331 f2fs_folio_put(folio, true); 4332 4333 folio = f2fs_get_meta_folio(sbi, start++); 4334 if (IS_ERR(folio)) 4335 return PTR_ERR(folio); 4336 kaddr = folio_address(folio); 4337 offset = 0; 4338 } 4339 } 4340 f2fs_folio_put(folio, true); 4341 return 0; 4342 } 4343 4344 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 4345 { 4346 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 4347 struct f2fs_summary_block *sum; 4348 struct curseg_info *curseg; 4349 struct folio *new; 4350 unsigned short blk_off; 4351 unsigned int segno = 0; 4352 block_t blk_addr = 0; 4353 int err = 0; 4354 4355 /* get segment number and block addr */ 4356 if (IS_DATASEG(type)) { 4357 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 4358 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 4359 CURSEG_HOT_DATA]); 4360 if (__exist_node_summaries(sbi)) 4361 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type); 4362 else 4363 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 4364 } else { 4365 segno = le32_to_cpu(ckpt->cur_node_segno[type - 4366 CURSEG_HOT_NODE]); 4367 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 4368 CURSEG_HOT_NODE]); 4369 if (__exist_node_summaries(sbi)) 4370 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 4371 type - CURSEG_HOT_NODE); 4372 else 4373 blk_addr = GET_SUM_BLOCK(sbi, segno); 4374 } 4375 4376 new = f2fs_get_meta_folio(sbi, blk_addr); 4377 if (IS_ERR(new)) 4378 return PTR_ERR(new); 4379 sum = folio_address(new); 4380 4381 if (IS_NODESEG(type)) { 4382 if (__exist_node_summaries(sbi)) { 4383 struct f2fs_summary *ns = sum_entries(sum); 4384 int i; 4385 4386 for (i = 0; i < BLKS_PER_SEG(sbi); i++, ns++) { 4387 ns->version = 0; 4388 ns->ofs_in_node = 0; 4389 } 4390 } else { 4391 err = f2fs_restore_node_summary(sbi, segno, sum); 4392 if (err) 4393 goto out; 4394 } 4395 } 4396 4397 /* set uncompleted segment to curseg */ 4398 curseg = CURSEG_I(sbi, type); 4399 mutex_lock(&curseg->curseg_mutex); 4400 4401 /* update journal info */ 4402 down_write(&curseg->journal_rwsem); 4403 memcpy(curseg->journal, sum_journal(sbi, sum), sbi->sum_journal_size); 4404 up_write(&curseg->journal_rwsem); 4405 4406 memcpy(sum_entries(curseg->sum_blk), sum_entries(sum), 4407 sbi->sum_entry_size); 4408 memcpy(sum_footer(sbi, curseg->sum_blk), sum_footer(sbi, sum), 4409 SUM_FOOTER_SIZE); 4410 curseg->next_segno = segno; 4411 reset_curseg(sbi, type, 0); 4412 curseg->alloc_type = ckpt->alloc_type[type]; 4413 curseg->next_blkoff = blk_off; 4414 mutex_unlock(&curseg->curseg_mutex); 4415 out: 4416 f2fs_folio_put(new, true); 4417 return err; 4418 } 4419 4420 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 4421 { 4422 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 4423 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 4424 int type = CURSEG_HOT_DATA; 4425 int err; 4426 4427 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 4428 int npages = f2fs_npages_for_summary_flush(sbi, true); 4429 4430 if (npages >= 2) 4431 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 4432 META_CP, true); 4433 4434 /* restore for compacted data summary */ 4435 err = read_compacted_summaries(sbi); 4436 if (err) 4437 return err; 4438 type = CURSEG_HOT_NODE; 4439 } 4440 4441 if (__exist_node_summaries(sbi)) 4442 f2fs_ra_meta_pages(sbi, 4443 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type), 4444 NR_CURSEG_PERSIST_TYPE - type, META_CP, true); 4445 4446 for (; type <= CURSEG_COLD_NODE; type++) { 4447 err = read_normal_summaries(sbi, type); 4448 if (err) 4449 return err; 4450 } 4451 4452 /* sanity check for summary blocks */ 4453 if (nats_in_cursum(nat_j) > sbi->nat_journal_entries || 4454 sits_in_cursum(sit_j) > sbi->sit_journal_entries) { 4455 f2fs_err(sbi, "invalid journal entries nats %u sits %u", 4456 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 4457 return -EINVAL; 4458 } 4459 4460 return 0; 4461 } 4462 4463 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 4464 { 4465 struct folio *folio; 4466 unsigned char *kaddr; 4467 struct f2fs_summary *summary; 4468 struct curseg_info *seg_i; 4469 int written_size = 0; 4470 int i, j; 4471 4472 folio = f2fs_grab_meta_folio(sbi, blkaddr++); 4473 kaddr = folio_address(folio); 4474 memset(kaddr, 0, PAGE_SIZE); 4475 4476 /* Step 1: write nat cache */ 4477 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4478 memcpy(kaddr, seg_i->journal, sbi->sum_journal_size); 4479 written_size += sbi->sum_journal_size; 4480 4481 /* Step 2: write sit cache */ 4482 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4483 memcpy(kaddr + written_size, seg_i->journal, sbi->sum_journal_size); 4484 written_size += sbi->sum_journal_size; 4485 4486 /* Step 3: write summary entries */ 4487 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4488 seg_i = CURSEG_I(sbi, i); 4489 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) { 4490 if (!folio) { 4491 folio = f2fs_grab_meta_folio(sbi, blkaddr++); 4492 kaddr = folio_address(folio); 4493 memset(kaddr, 0, PAGE_SIZE); 4494 written_size = 0; 4495 } 4496 summary = (struct f2fs_summary *)(kaddr + written_size); 4497 *summary = sum_entries(seg_i->sum_blk)[j]; 4498 written_size += SUMMARY_SIZE; 4499 4500 if (written_size + SUMMARY_SIZE <= sbi->blocksize - 4501 SUM_FOOTER_SIZE) 4502 continue; 4503 4504 folio_mark_dirty(folio); 4505 f2fs_folio_put(folio, true); 4506 folio = NULL; 4507 } 4508 } 4509 if (folio) { 4510 folio_mark_dirty(folio); 4511 f2fs_folio_put(folio, true); 4512 } 4513 } 4514 4515 static void write_normal_summaries(struct f2fs_sb_info *sbi, 4516 block_t blkaddr, int type) 4517 { 4518 int i, end; 4519 4520 if (IS_DATASEG(type)) 4521 end = type + NR_CURSEG_DATA_TYPE; 4522 else 4523 end = type + NR_CURSEG_NODE_TYPE; 4524 4525 for (i = type; i < end; i++) 4526 write_current_sum_page(sbi, i, blkaddr + (i - type)); 4527 } 4528 4529 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4530 { 4531 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 4532 write_compacted_summaries(sbi, start_blk); 4533 else 4534 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 4535 } 4536 4537 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4538 { 4539 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 4540 } 4541 4542 int f2fs_lookup_journal_in_cursum(struct f2fs_sb_info *sbi, 4543 struct f2fs_journal *journal, int type, 4544 unsigned int val, int alloc) 4545 { 4546 int i; 4547 4548 if (type == NAT_JOURNAL) { 4549 for (i = 0; i < nats_in_cursum(journal); i++) { 4550 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 4551 return i; 4552 } 4553 if (alloc && __has_cursum_space(sbi, journal, 1, NAT_JOURNAL)) 4554 return update_nats_in_cursum(journal, 1); 4555 } else if (type == SIT_JOURNAL) { 4556 for (i = 0; i < sits_in_cursum(journal); i++) 4557 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 4558 return i; 4559 if (alloc && __has_cursum_space(sbi, journal, 1, SIT_JOURNAL)) 4560 return update_sits_in_cursum(journal, 1); 4561 } 4562 return -1; 4563 } 4564 4565 static struct folio *get_current_sit_folio(struct f2fs_sb_info *sbi, 4566 unsigned int segno) 4567 { 4568 return f2fs_get_meta_folio(sbi, current_sit_addr(sbi, segno)); 4569 } 4570 4571 static struct folio *get_next_sit_folio(struct f2fs_sb_info *sbi, 4572 unsigned int start) 4573 { 4574 struct sit_info *sit_i = SIT_I(sbi); 4575 struct folio *folio; 4576 pgoff_t src_off, dst_off; 4577 4578 src_off = current_sit_addr(sbi, start); 4579 dst_off = next_sit_addr(sbi, src_off); 4580 4581 folio = f2fs_grab_meta_folio(sbi, dst_off); 4582 seg_info_to_sit_folio(sbi, folio, start); 4583 4584 folio_mark_dirty(folio); 4585 set_to_next_sit(sit_i, start); 4586 4587 return folio; 4588 } 4589 4590 static struct sit_entry_set *grab_sit_entry_set(void) 4591 { 4592 struct sit_entry_set *ses = 4593 f2fs_kmem_cache_alloc(sit_entry_set_slab, 4594 GFP_NOFS, true, NULL); 4595 4596 ses->entry_cnt = 0; 4597 INIT_LIST_HEAD(&ses->set_list); 4598 return ses; 4599 } 4600 4601 static void release_sit_entry_set(struct sit_entry_set *ses) 4602 { 4603 list_del(&ses->set_list); 4604 kmem_cache_free(sit_entry_set_slab, ses); 4605 } 4606 4607 static void adjust_sit_entry_set(struct sit_entry_set *ses, 4608 struct list_head *head) 4609 { 4610 struct sit_entry_set *next = ses; 4611 4612 if (list_is_last(&ses->set_list, head)) 4613 return; 4614 4615 list_for_each_entry_continue(next, head, set_list) 4616 if (ses->entry_cnt <= next->entry_cnt) { 4617 list_move_tail(&ses->set_list, &next->set_list); 4618 return; 4619 } 4620 4621 list_move_tail(&ses->set_list, head); 4622 } 4623 4624 static void add_sit_entry(unsigned int segno, struct list_head *head) 4625 { 4626 struct sit_entry_set *ses; 4627 unsigned int start_segno = START_SEGNO(segno); 4628 4629 list_for_each_entry(ses, head, set_list) { 4630 if (ses->start_segno == start_segno) { 4631 ses->entry_cnt++; 4632 adjust_sit_entry_set(ses, head); 4633 return; 4634 } 4635 } 4636 4637 ses = grab_sit_entry_set(); 4638 4639 ses->start_segno = start_segno; 4640 ses->entry_cnt++; 4641 list_add(&ses->set_list, head); 4642 } 4643 4644 static void add_sits_in_set(struct f2fs_sb_info *sbi) 4645 { 4646 struct f2fs_sm_info *sm_info = SM_I(sbi); 4647 struct list_head *set_list = &sm_info->sit_entry_set; 4648 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 4649 unsigned int segno; 4650 4651 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 4652 add_sit_entry(segno, set_list); 4653 } 4654 4655 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 4656 { 4657 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4658 struct f2fs_journal *journal = curseg->journal; 4659 int i; 4660 4661 down_write(&curseg->journal_rwsem); 4662 for (i = 0; i < sits_in_cursum(journal); i++) { 4663 unsigned int segno; 4664 bool dirtied; 4665 4666 segno = le32_to_cpu(segno_in_journal(journal, i)); 4667 dirtied = __mark_sit_entry_dirty(sbi, segno); 4668 4669 if (!dirtied) 4670 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 4671 } 4672 update_sits_in_cursum(journal, -i); 4673 up_write(&curseg->journal_rwsem); 4674 } 4675 4676 /* 4677 * CP calls this function, which flushes SIT entries including sit_journal, 4678 * and moves prefree segs to free segs. 4679 */ 4680 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 4681 { 4682 struct sit_info *sit_i = SIT_I(sbi); 4683 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 4684 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4685 struct f2fs_journal *journal = curseg->journal; 4686 struct sit_entry_set *ses, *tmp; 4687 struct list_head *head = &SM_I(sbi)->sit_entry_set; 4688 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 4689 struct seg_entry *se; 4690 4691 down_write(&sit_i->sentry_lock); 4692 4693 if (!sit_i->dirty_sentries) 4694 goto out; 4695 4696 /* 4697 * add and account sit entries of dirty bitmap in sit entry 4698 * set temporarily 4699 */ 4700 add_sits_in_set(sbi); 4701 4702 /* 4703 * if there are no enough space in journal to store dirty sit 4704 * entries, remove all entries from journal and add and account 4705 * them in sit entry set. 4706 */ 4707 if (!__has_cursum_space(sbi, journal, 4708 sit_i->dirty_sentries, SIT_JOURNAL) || !to_journal) 4709 remove_sits_in_journal(sbi); 4710 4711 /* 4712 * there are two steps to flush sit entries: 4713 * #1, flush sit entries to journal in current cold data summary block. 4714 * #2, flush sit entries to sit page. 4715 */ 4716 list_for_each_entry_safe(ses, tmp, head, set_list) { 4717 struct folio *folio = NULL; 4718 struct f2fs_sit_block *raw_sit = NULL; 4719 unsigned int start_segno = ses->start_segno; 4720 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 4721 (unsigned long)MAIN_SEGS(sbi)); 4722 unsigned int segno = start_segno; 4723 4724 if (to_journal && 4725 !__has_cursum_space(sbi, journal, ses->entry_cnt, 4726 SIT_JOURNAL)) 4727 to_journal = false; 4728 4729 if (to_journal) { 4730 down_write(&curseg->journal_rwsem); 4731 } else { 4732 folio = get_next_sit_folio(sbi, start_segno); 4733 raw_sit = folio_address(folio); 4734 } 4735 4736 /* flush dirty sit entries in region of current sit set */ 4737 for_each_set_bit_from(segno, bitmap, end) { 4738 int offset, sit_offset; 4739 4740 se = get_seg_entry(sbi, segno); 4741 #ifdef CONFIG_F2FS_CHECK_FS 4742 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 4743 SIT_VBLOCK_MAP_SIZE)) 4744 f2fs_bug_on(sbi, 1); 4745 #endif 4746 4747 /* add discard candidates */ 4748 if (!(cpc->reason & CP_DISCARD)) { 4749 cpc->trim_start = segno; 4750 add_discard_addrs(sbi, cpc, false); 4751 } 4752 4753 if (to_journal) { 4754 offset = f2fs_lookup_journal_in_cursum(sbi, journal, 4755 SIT_JOURNAL, segno, 1); 4756 f2fs_bug_on(sbi, offset < 0); 4757 segno_in_journal(journal, offset) = 4758 cpu_to_le32(segno); 4759 seg_info_to_raw_sit(se, 4760 &sit_in_journal(journal, offset)); 4761 check_block_count(sbi, segno, 4762 &sit_in_journal(journal, offset)); 4763 } else { 4764 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 4765 seg_info_to_raw_sit(se, 4766 &raw_sit->entries[sit_offset]); 4767 check_block_count(sbi, segno, 4768 &raw_sit->entries[sit_offset]); 4769 } 4770 4771 /* update ckpt_valid_block */ 4772 if (__is_large_section(sbi)) { 4773 set_ckpt_valid_blocks(sbi, segno); 4774 sanity_check_valid_blocks(sbi, segno); 4775 } 4776 4777 __clear_bit(segno, bitmap); 4778 sit_i->dirty_sentries--; 4779 ses->entry_cnt--; 4780 } 4781 4782 if (to_journal) 4783 up_write(&curseg->journal_rwsem); 4784 else 4785 f2fs_folio_put(folio, true); 4786 4787 f2fs_bug_on(sbi, ses->entry_cnt); 4788 release_sit_entry_set(ses); 4789 } 4790 4791 f2fs_bug_on(sbi, !list_empty(head)); 4792 f2fs_bug_on(sbi, sit_i->dirty_sentries); 4793 out: 4794 if (cpc->reason & CP_DISCARD) { 4795 __u64 trim_start = cpc->trim_start; 4796 4797 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 4798 add_discard_addrs(sbi, cpc, false); 4799 4800 cpc->trim_start = trim_start; 4801 } 4802 up_write(&sit_i->sentry_lock); 4803 4804 set_prefree_as_free_segments(sbi); 4805 } 4806 4807 static int build_sit_info(struct f2fs_sb_info *sbi) 4808 { 4809 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4810 struct sit_info *sit_i; 4811 unsigned int sit_segs, start; 4812 char *src_bitmap, *bitmap; 4813 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 4814 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0; 4815 4816 /* allocate memory for SIT information */ 4817 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4818 if (!sit_i) 4819 return -ENOMEM; 4820 4821 SM_I(sbi)->sit_info = sit_i; 4822 4823 sit_i->sentries = 4824 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4825 MAIN_SEGS(sbi)), 4826 GFP_KERNEL); 4827 if (!sit_i->sentries) 4828 return -ENOMEM; 4829 4830 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4831 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4832 GFP_KERNEL); 4833 if (!sit_i->dirty_sentries_bitmap) 4834 return -ENOMEM; 4835 4836 #ifdef CONFIG_F2FS_CHECK_FS 4837 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map); 4838 #else 4839 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map); 4840 #endif 4841 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4842 if (!sit_i->bitmap) 4843 return -ENOMEM; 4844 4845 bitmap = sit_i->bitmap; 4846 4847 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4848 sit_i->sentries[start].cur_valid_map = bitmap; 4849 bitmap += SIT_VBLOCK_MAP_SIZE; 4850 4851 sit_i->sentries[start].ckpt_valid_map = bitmap; 4852 bitmap += SIT_VBLOCK_MAP_SIZE; 4853 4854 #ifdef CONFIG_F2FS_CHECK_FS 4855 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4856 bitmap += SIT_VBLOCK_MAP_SIZE; 4857 #endif 4858 4859 if (discard_map) { 4860 sit_i->sentries[start].discard_map = bitmap; 4861 bitmap += SIT_VBLOCK_MAP_SIZE; 4862 } 4863 } 4864 4865 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4866 if (!sit_i->tmp_map) 4867 return -ENOMEM; 4868 4869 if (__is_large_section(sbi)) { 4870 sit_i->sec_entries = 4871 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4872 MAIN_SECS(sbi)), 4873 GFP_KERNEL); 4874 if (!sit_i->sec_entries) 4875 return -ENOMEM; 4876 } 4877 4878 /* get information related with SIT */ 4879 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4880 4881 /* setup SIT bitmap from ckeckpoint pack */ 4882 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4883 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4884 4885 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4886 if (!sit_i->sit_bitmap) 4887 return -ENOMEM; 4888 4889 #ifdef CONFIG_F2FS_CHECK_FS 4890 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4891 sit_bitmap_size, GFP_KERNEL); 4892 if (!sit_i->sit_bitmap_mir) 4893 return -ENOMEM; 4894 4895 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4896 main_bitmap_size, GFP_KERNEL); 4897 if (!sit_i->invalid_segmap) 4898 return -ENOMEM; 4899 #endif 4900 4901 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4902 sit_i->sit_blocks = SEGS_TO_BLKS(sbi, sit_segs); 4903 sit_i->written_valid_blocks = 0; 4904 sit_i->bitmap_size = sit_bitmap_size; 4905 sit_i->dirty_sentries = 0; 4906 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4907 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4908 sit_i->mounted_time = ktime_get_boottime_seconds(); 4909 init_rwsem(&sit_i->sentry_lock); 4910 return 0; 4911 } 4912 4913 static int build_free_segmap(struct f2fs_sb_info *sbi) 4914 { 4915 struct free_segmap_info *free_i; 4916 unsigned int bitmap_size, sec_bitmap_size; 4917 4918 /* allocate memory for free segmap information */ 4919 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4920 if (!free_i) 4921 return -ENOMEM; 4922 4923 SM_I(sbi)->free_info = free_i; 4924 4925 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4926 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4927 if (!free_i->free_segmap) 4928 return -ENOMEM; 4929 4930 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4931 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4932 if (!free_i->free_secmap) 4933 return -ENOMEM; 4934 4935 /* set all segments as dirty temporarily */ 4936 memset(free_i->free_segmap, 0xff, bitmap_size); 4937 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4938 4939 /* init free segmap information */ 4940 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4941 free_i->free_segments = 0; 4942 free_i->free_sections = 0; 4943 spin_lock_init(&free_i->segmap_lock); 4944 return 0; 4945 } 4946 4947 static int build_curseg(struct f2fs_sb_info *sbi) 4948 { 4949 struct curseg_info *array; 4950 int i; 4951 4952 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, 4953 sizeof(*array)), GFP_KERNEL); 4954 if (!array) 4955 return -ENOMEM; 4956 4957 SM_I(sbi)->curseg_array = array; 4958 4959 for (i = 0; i < NO_CHECK_TYPE; i++) { 4960 mutex_init(&array[i].curseg_mutex); 4961 array[i].sum_blk = f2fs_kzalloc(sbi, sbi->sum_blocksize, 4962 GFP_KERNEL); 4963 if (!array[i].sum_blk) 4964 return -ENOMEM; 4965 init_rwsem(&array[i].journal_rwsem); 4966 array[i].journal = f2fs_kzalloc(sbi, 4967 sbi->sum_journal_size, GFP_KERNEL); 4968 if (!array[i].journal) 4969 return -ENOMEM; 4970 array[i].seg_type = log_type_to_seg_type(i); 4971 reset_curseg_fields(&array[i]); 4972 } 4973 return restore_curseg_summaries(sbi); 4974 } 4975 4976 static int build_sit_entries(struct f2fs_sb_info *sbi) 4977 { 4978 struct sit_info *sit_i = SIT_I(sbi); 4979 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4980 struct f2fs_journal *journal = curseg->journal; 4981 struct seg_entry *se; 4982 struct f2fs_sit_entry sit; 4983 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4984 unsigned int i, start, end; 4985 unsigned int readed, start_blk = 0; 4986 int err = 0; 4987 block_t sit_valid_blocks[2] = {0, 0}; 4988 4989 do { 4990 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS, 4991 META_SIT, true); 4992 4993 start = start_blk * sit_i->sents_per_block; 4994 end = (start_blk + readed) * sit_i->sents_per_block; 4995 4996 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4997 struct f2fs_sit_block *sit_blk; 4998 struct folio *folio; 4999 5000 se = &sit_i->sentries[start]; 5001 folio = get_current_sit_folio(sbi, start); 5002 if (IS_ERR(folio)) 5003 return PTR_ERR(folio); 5004 sit_blk = folio_address(folio); 5005 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 5006 f2fs_folio_put(folio, true); 5007 5008 err = check_block_count(sbi, start, &sit); 5009 if (err) 5010 return err; 5011 seg_info_from_raw_sit(se, &sit); 5012 5013 if (se->type >= NR_PERSISTENT_LOG) { 5014 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 5015 se->type, start); 5016 f2fs_handle_error(sbi, 5017 ERROR_INCONSISTENT_SUM_TYPE); 5018 return -EFSCORRUPTED; 5019 } 5020 5021 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 5022 5023 if (!f2fs_block_unit_discard(sbi)) 5024 goto init_discard_map_done; 5025 5026 /* build discard map only one time */ 5027 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 5028 memset(se->discard_map, 0xff, 5029 SIT_VBLOCK_MAP_SIZE); 5030 goto init_discard_map_done; 5031 } 5032 memcpy(se->discard_map, se->cur_valid_map, 5033 SIT_VBLOCK_MAP_SIZE); 5034 sbi->discard_blks += BLKS_PER_SEG(sbi) - 5035 se->valid_blocks; 5036 init_discard_map_done: 5037 if (__is_large_section(sbi)) 5038 get_sec_entry(sbi, start)->valid_blocks += 5039 se->valid_blocks; 5040 } 5041 start_blk += readed; 5042 } while (start_blk < sit_blk_cnt); 5043 5044 down_read(&curseg->journal_rwsem); 5045 for (i = 0; i < sits_in_cursum(journal); i++) { 5046 unsigned int old_valid_blocks; 5047 5048 start = le32_to_cpu(segno_in_journal(journal, i)); 5049 if (start >= MAIN_SEGS(sbi)) { 5050 f2fs_err(sbi, "Wrong journal entry on segno %u", 5051 start); 5052 err = -EFSCORRUPTED; 5053 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL); 5054 break; 5055 } 5056 5057 se = &sit_i->sentries[start]; 5058 sit = sit_in_journal(journal, i); 5059 5060 old_valid_blocks = se->valid_blocks; 5061 5062 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks; 5063 5064 err = check_block_count(sbi, start, &sit); 5065 if (err) 5066 break; 5067 seg_info_from_raw_sit(se, &sit); 5068 5069 if (se->type >= NR_PERSISTENT_LOG) { 5070 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 5071 se->type, start); 5072 err = -EFSCORRUPTED; 5073 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 5074 break; 5075 } 5076 5077 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 5078 5079 if (f2fs_block_unit_discard(sbi)) { 5080 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 5081 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 5082 } else { 5083 memcpy(se->discard_map, se->cur_valid_map, 5084 SIT_VBLOCK_MAP_SIZE); 5085 sbi->discard_blks += old_valid_blocks; 5086 sbi->discard_blks -= se->valid_blocks; 5087 } 5088 } 5089 5090 if (__is_large_section(sbi)) { 5091 get_sec_entry(sbi, start)->valid_blocks += 5092 se->valid_blocks; 5093 get_sec_entry(sbi, start)->valid_blocks -= 5094 old_valid_blocks; 5095 } 5096 } 5097 up_read(&curseg->journal_rwsem); 5098 5099 /* update ckpt_valid_block */ 5100 if (__is_large_section(sbi)) { 5101 unsigned int segno; 5102 5103 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) { 5104 set_ckpt_valid_blocks(sbi, segno); 5105 sanity_check_valid_blocks(sbi, segno); 5106 } 5107 } 5108 5109 if (err) 5110 return err; 5111 5112 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) { 5113 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 5114 sit_valid_blocks[NODE], valid_node_count(sbi)); 5115 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT); 5116 return -EFSCORRUPTED; 5117 } 5118 5119 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] > 5120 valid_user_blocks(sbi)) { 5121 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u", 5122 sit_valid_blocks[DATA], sit_valid_blocks[NODE], 5123 valid_user_blocks(sbi)); 5124 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT); 5125 return -EFSCORRUPTED; 5126 } 5127 5128 return 0; 5129 } 5130 5131 static void init_free_segmap(struct f2fs_sb_info *sbi) 5132 { 5133 unsigned int start; 5134 int type; 5135 struct seg_entry *sentry; 5136 5137 for (start = 0; start < MAIN_SEGS(sbi); start++) { 5138 if (f2fs_usable_blks_in_seg(sbi, start) == 0) 5139 continue; 5140 sentry = get_seg_entry(sbi, start); 5141 if (!sentry->valid_blocks) 5142 __set_free(sbi, start); 5143 else 5144 SIT_I(sbi)->written_valid_blocks += 5145 sentry->valid_blocks; 5146 } 5147 5148 /* set use the current segments */ 5149 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 5150 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 5151 5152 __set_test_and_inuse(sbi, curseg_t->segno); 5153 } 5154 } 5155 5156 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 5157 { 5158 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5159 struct free_segmap_info *free_i = FREE_I(sbi); 5160 unsigned int segno = 0, offset = 0, secno; 5161 block_t valid_blocks, usable_blks_in_seg; 5162 5163 while (1) { 5164 /* find dirty segment based on free segmap */ 5165 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 5166 if (segno >= MAIN_SEGS(sbi)) 5167 break; 5168 offset = segno + 1; 5169 valid_blocks = get_valid_blocks(sbi, segno, false); 5170 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 5171 if (valid_blocks == usable_blks_in_seg || !valid_blocks) 5172 continue; 5173 if (valid_blocks > usable_blks_in_seg) { 5174 f2fs_bug_on(sbi, 1); 5175 continue; 5176 } 5177 mutex_lock(&dirty_i->seglist_lock); 5178 __locate_dirty_segment(sbi, segno, DIRTY); 5179 mutex_unlock(&dirty_i->seglist_lock); 5180 } 5181 5182 if (!__is_large_section(sbi)) 5183 return; 5184 5185 mutex_lock(&dirty_i->seglist_lock); 5186 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) { 5187 valid_blocks = get_valid_blocks(sbi, segno, true); 5188 secno = GET_SEC_FROM_SEG(sbi, segno); 5189 5190 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi)) 5191 continue; 5192 if (is_cursec(sbi, secno)) 5193 continue; 5194 set_bit(secno, dirty_i->dirty_secmap); 5195 } 5196 mutex_unlock(&dirty_i->seglist_lock); 5197 } 5198 5199 static int init_victim_secmap(struct f2fs_sb_info *sbi) 5200 { 5201 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5202 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 5203 5204 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 5205 if (!dirty_i->victim_secmap) 5206 return -ENOMEM; 5207 5208 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 5209 if (!dirty_i->pinned_secmap) 5210 return -ENOMEM; 5211 5212 dirty_i->pinned_secmap_cnt = 0; 5213 dirty_i->enable_pin_section = true; 5214 return 0; 5215 } 5216 5217 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 5218 { 5219 struct dirty_seglist_info *dirty_i; 5220 unsigned int bitmap_size, i; 5221 5222 /* allocate memory for dirty segments list information */ 5223 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 5224 GFP_KERNEL); 5225 if (!dirty_i) 5226 return -ENOMEM; 5227 5228 SM_I(sbi)->dirty_info = dirty_i; 5229 mutex_init(&dirty_i->seglist_lock); 5230 5231 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 5232 5233 for (i = 0; i < NR_DIRTY_TYPE; i++) { 5234 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 5235 GFP_KERNEL); 5236 if (!dirty_i->dirty_segmap[i]) 5237 return -ENOMEM; 5238 } 5239 5240 if (__is_large_section(sbi)) { 5241 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 5242 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi, 5243 bitmap_size, GFP_KERNEL); 5244 if (!dirty_i->dirty_secmap) 5245 return -ENOMEM; 5246 } 5247 5248 init_dirty_segmap(sbi); 5249 return init_victim_secmap(sbi); 5250 } 5251 5252 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 5253 { 5254 int i; 5255 5256 /* 5257 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 5258 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 5259 */ 5260 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5261 struct curseg_info *curseg = CURSEG_I(sbi, i); 5262 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 5263 unsigned int blkofs = curseg->next_blkoff; 5264 5265 if (f2fs_sb_has_readonly(sbi) && 5266 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE) 5267 continue; 5268 5269 sanity_check_seg_type(sbi, curseg->seg_type); 5270 5271 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) { 5272 f2fs_err(sbi, 5273 "Current segment has invalid alloc_type:%d", 5274 curseg->alloc_type); 5275 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 5276 return -EFSCORRUPTED; 5277 } 5278 5279 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 5280 goto out; 5281 5282 if (curseg->alloc_type == SSR) 5283 continue; 5284 5285 for (blkofs += 1; blkofs < BLKS_PER_SEG(sbi); blkofs++) { 5286 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 5287 continue; 5288 out: 5289 f2fs_err(sbi, 5290 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 5291 i, curseg->segno, curseg->alloc_type, 5292 curseg->next_blkoff, blkofs); 5293 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 5294 return -EFSCORRUPTED; 5295 } 5296 } 5297 return 0; 5298 } 5299 5300 #ifdef CONFIG_BLK_DEV_ZONED 5301 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 5302 struct f2fs_dev_info *fdev, 5303 struct blk_zone *zone) 5304 { 5305 unsigned int zone_segno; 5306 block_t zone_block, valid_block_cnt; 5307 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 5308 int ret; 5309 unsigned int nofs_flags; 5310 5311 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5312 return 0; 5313 5314 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 5315 zone_segno = GET_SEGNO(sbi, zone_block); 5316 5317 /* 5318 * Skip check of zones cursegs point to, since 5319 * fix_curseg_write_pointer() checks them. 5320 */ 5321 if (zone_segno >= MAIN_SEGS(sbi)) 5322 return 0; 5323 5324 /* 5325 * Get # of valid block of the zone. 5326 */ 5327 valid_block_cnt = get_valid_blocks(sbi, zone_segno, true); 5328 if (is_cursec(sbi, GET_SEC_FROM_SEG(sbi, zone_segno))) { 5329 f2fs_notice(sbi, "Open zones: valid block[0x%x,0x%x] cond[%s]", 5330 zone_segno, valid_block_cnt, 5331 blk_zone_cond_str(zone->cond)); 5332 return 0; 5333 } 5334 5335 if ((!valid_block_cnt && zone->cond == BLK_ZONE_COND_EMPTY) || 5336 (valid_block_cnt && zone->cond == BLK_ZONE_COND_FULL)) 5337 return 0; 5338 5339 if (!valid_block_cnt) { 5340 f2fs_notice(sbi, "Zone without valid block has non-zero write " 5341 "pointer. Reset the write pointer: cond[%s]", 5342 blk_zone_cond_str(zone->cond)); 5343 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 5344 zone->len >> log_sectors_per_block); 5345 if (ret) 5346 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5347 fdev->path, ret); 5348 return ret; 5349 } 5350 5351 /* 5352 * If there are valid blocks and the write pointer doesn't match 5353 * with them, we need to report the inconsistency and fill 5354 * the zone till the end to close the zone. This inconsistency 5355 * does not cause write error because the zone will not be 5356 * selected for write operation until it get discarded. 5357 */ 5358 f2fs_notice(sbi, "Valid blocks are not aligned with write " 5359 "pointer: valid block[0x%x,0x%x] cond[%s]", 5360 zone_segno, valid_block_cnt, blk_zone_cond_str(zone->cond)); 5361 5362 nofs_flags = memalloc_nofs_save(); 5363 ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH, 5364 zone->start, zone->len); 5365 memalloc_nofs_restore(nofs_flags); 5366 if (ret == -EOPNOTSUPP) { 5367 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp, 5368 zone->len - (zone->wp - zone->start), 5369 GFP_NOFS, 0); 5370 if (ret) 5371 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)", 5372 fdev->path, ret); 5373 } else if (ret) { 5374 f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)", 5375 fdev->path, ret); 5376 } 5377 5378 return ret; 5379 } 5380 5381 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 5382 block_t zone_blkaddr) 5383 { 5384 int i; 5385 5386 for (i = 0; i < sbi->s_ndevs; i++) { 5387 if (!bdev_is_zoned(FDEV(i).bdev)) 5388 continue; 5389 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 5390 zone_blkaddr <= FDEV(i).end_blk)) 5391 return &FDEV(i); 5392 } 5393 5394 return NULL; 5395 } 5396 5397 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 5398 void *data) 5399 { 5400 memcpy(data, zone, sizeof(struct blk_zone)); 5401 return 0; 5402 } 5403 5404 static int do_fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 5405 { 5406 struct curseg_info *cs = CURSEG_I(sbi, type); 5407 struct f2fs_dev_info *zbd; 5408 struct blk_zone zone; 5409 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 5410 block_t cs_zone_block, wp_block; 5411 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 5412 sector_t zone_sector; 5413 int err; 5414 5415 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5416 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5417 5418 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5419 if (!zbd) 5420 return 0; 5421 5422 /* report zone for the sector the curseg points to */ 5423 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5424 << log_sectors_per_block; 5425 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5426 report_one_zone_cb, &zone); 5427 if (err != 1) { 5428 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5429 zbd->path, err); 5430 return err; 5431 } 5432 5433 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5434 return 0; 5435 5436 /* 5437 * When safely unmounted in the previous mount, we could use current 5438 * segments. Otherwise, allocate new sections. 5439 */ 5440 if (is_set_ckpt_flags(sbi, CP_UMOUNT_FLAG)) { 5441 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 5442 wp_segno = GET_SEGNO(sbi, wp_block); 5443 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 5444 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 5445 5446 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 5447 wp_sector_off == 0) 5448 return 0; 5449 5450 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 5451 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", type, cs->segno, 5452 cs->next_blkoff, wp_segno, wp_blkoff); 5453 } 5454 5455 /* Allocate a new section if it's not new. */ 5456 if (cs->next_blkoff || 5457 cs->segno != GET_SEG_FROM_SEC(sbi, GET_ZONE_FROM_SEC(sbi, cs_section))) { 5458 unsigned int old_segno = cs->segno, old_blkoff = cs->next_blkoff; 5459 5460 f2fs_allocate_new_section(sbi, type, true); 5461 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 5462 "[0x%x,0x%x] -> [0x%x,0x%x]", 5463 type, old_segno, old_blkoff, 5464 cs->segno, cs->next_blkoff); 5465 } 5466 5467 /* check consistency of the zone curseg pointed to */ 5468 if (check_zone_write_pointer(sbi, zbd, &zone)) 5469 return -EIO; 5470 5471 /* check newly assigned zone */ 5472 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5473 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5474 5475 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5476 if (!zbd) 5477 return 0; 5478 5479 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5480 << log_sectors_per_block; 5481 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5482 report_one_zone_cb, &zone); 5483 if (err != 1) { 5484 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5485 zbd->path, err); 5486 return err; 5487 } 5488 5489 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5490 return 0; 5491 5492 if (zone.wp != zone.start) { 5493 f2fs_notice(sbi, 5494 "New zone for curseg[%d] is not yet discarded. " 5495 "Reset the zone: curseg[0x%x,0x%x]", 5496 type, cs->segno, cs->next_blkoff); 5497 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block, 5498 zone.len >> log_sectors_per_block); 5499 if (err) { 5500 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5501 zbd->path, err); 5502 return err; 5503 } 5504 } 5505 5506 return 0; 5507 } 5508 5509 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5510 { 5511 int i, ret; 5512 5513 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5514 ret = do_fix_curseg_write_pointer(sbi, i); 5515 if (ret) 5516 return ret; 5517 } 5518 5519 return 0; 5520 } 5521 5522 struct check_zone_write_pointer_args { 5523 struct f2fs_sb_info *sbi; 5524 struct f2fs_dev_info *fdev; 5525 }; 5526 5527 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 5528 void *data) 5529 { 5530 struct check_zone_write_pointer_args *args; 5531 5532 args = (struct check_zone_write_pointer_args *)data; 5533 5534 return check_zone_write_pointer(args->sbi, args->fdev, zone); 5535 } 5536 5537 static int check_write_pointer(struct f2fs_sb_info *sbi) 5538 { 5539 int i, ret; 5540 struct check_zone_write_pointer_args args; 5541 5542 for (i = 0; i < sbi->s_ndevs; i++) { 5543 if (!bdev_is_zoned(FDEV(i).bdev)) 5544 continue; 5545 5546 args.sbi = sbi; 5547 args.fdev = &FDEV(i); 5548 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 5549 check_zone_write_pointer_cb, &args); 5550 if (ret < 0) 5551 return ret; 5552 } 5553 5554 return 0; 5555 } 5556 5557 int f2fs_check_and_fix_write_pointer(struct f2fs_sb_info *sbi) 5558 { 5559 int ret; 5560 5561 if (!f2fs_sb_has_blkzoned(sbi) || f2fs_readonly(sbi->sb) || 5562 f2fs_hw_is_readonly(sbi)) 5563 return 0; 5564 5565 f2fs_notice(sbi, "Checking entire write pointers"); 5566 ret = fix_curseg_write_pointer(sbi); 5567 if (!ret) 5568 ret = check_write_pointer(sbi); 5569 return ret; 5570 } 5571 5572 /* 5573 * Return the number of usable blocks in a segment. The number of blocks 5574 * returned is always equal to the number of blocks in a segment for 5575 * segments fully contained within a sequential zone capacity or a 5576 * conventional zone. For segments partially contained in a sequential 5577 * zone capacity, the number of usable blocks up to the zone capacity 5578 * is returned. 0 is returned in all other cases. 5579 */ 5580 static inline unsigned int f2fs_usable_zone_blks_in_seg( 5581 struct f2fs_sb_info *sbi, unsigned int segno) 5582 { 5583 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr; 5584 unsigned int secno; 5585 5586 if (!sbi->unusable_blocks_per_sec) 5587 return BLKS_PER_SEG(sbi); 5588 5589 secno = GET_SEC_FROM_SEG(sbi, segno); 5590 seg_start = START_BLOCK(sbi, segno); 5591 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5592 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi); 5593 5594 /* 5595 * If segment starts before zone capacity and spans beyond 5596 * zone capacity, then usable blocks are from seg start to 5597 * zone capacity. If the segment starts after the zone capacity, 5598 * then there are no usable blocks. 5599 */ 5600 if (seg_start >= sec_cap_blkaddr) 5601 return 0; 5602 if (seg_start + BLKS_PER_SEG(sbi) > sec_cap_blkaddr) 5603 return sec_cap_blkaddr - seg_start; 5604 5605 return BLKS_PER_SEG(sbi); 5606 } 5607 #else 5608 int f2fs_check_and_fix_write_pointer(struct f2fs_sb_info *sbi) 5609 { 5610 return 0; 5611 } 5612 5613 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi, 5614 unsigned int segno) 5615 { 5616 return 0; 5617 } 5618 5619 #endif 5620 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi, 5621 unsigned int segno) 5622 { 5623 if (f2fs_sb_has_blkzoned(sbi)) 5624 return f2fs_usable_zone_blks_in_seg(sbi, segno); 5625 5626 return BLKS_PER_SEG(sbi); 5627 } 5628 5629 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi) 5630 { 5631 if (f2fs_sb_has_blkzoned(sbi)) 5632 return CAP_SEGS_PER_SEC(sbi); 5633 5634 return SEGS_PER_SEC(sbi); 5635 } 5636 5637 unsigned long long f2fs_get_section_mtime(struct f2fs_sb_info *sbi, 5638 unsigned int segno) 5639 { 5640 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi); 5641 unsigned int secno = 0, start = 0; 5642 unsigned int total_valid_blocks = 0; 5643 unsigned long long mtime = 0; 5644 unsigned int i = 0; 5645 5646 secno = GET_SEC_FROM_SEG(sbi, segno); 5647 start = GET_SEG_FROM_SEC(sbi, secno); 5648 5649 if (!__is_large_section(sbi)) { 5650 mtime = get_seg_entry(sbi, start + i)->mtime; 5651 goto out; 5652 } 5653 5654 for (i = 0; i < usable_segs_per_sec; i++) { 5655 /* for large section, only check the mtime of valid segments */ 5656 struct seg_entry *se = get_seg_entry(sbi, start+i); 5657 5658 mtime += se->mtime * se->valid_blocks; 5659 total_valid_blocks += se->valid_blocks; 5660 } 5661 5662 if (total_valid_blocks == 0) 5663 return INVALID_MTIME; 5664 5665 mtime = div_u64(mtime, total_valid_blocks); 5666 out: 5667 if (unlikely(mtime == INVALID_MTIME)) 5668 mtime -= 1; 5669 return mtime; 5670 } 5671 5672 /* 5673 * Update min, max modified time for cost-benefit GC algorithm 5674 */ 5675 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 5676 { 5677 struct sit_info *sit_i = SIT_I(sbi); 5678 unsigned int segno; 5679 5680 down_write(&sit_i->sentry_lock); 5681 5682 sit_i->min_mtime = ULLONG_MAX; 5683 5684 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) { 5685 unsigned long long mtime = 0; 5686 5687 mtime = f2fs_get_section_mtime(sbi, segno); 5688 5689 if (sit_i->min_mtime > mtime) 5690 sit_i->min_mtime = mtime; 5691 } 5692 sit_i->max_mtime = get_mtime(sbi, false); 5693 sit_i->dirty_max_mtime = 0; 5694 up_write(&sit_i->sentry_lock); 5695 } 5696 5697 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 5698 { 5699 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 5700 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 5701 struct f2fs_sm_info *sm_info; 5702 int err; 5703 5704 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 5705 if (!sm_info) 5706 return -ENOMEM; 5707 5708 /* init sm info */ 5709 sbi->sm_info = sm_info; 5710 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 5711 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 5712 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 5713 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 5714 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 5715 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 5716 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 5717 sm_info->rec_prefree_segments = sm_info->main_segments * 5718 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 5719 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 5720 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 5721 5722 if (!f2fs_lfs_mode(sbi)) 5723 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC); 5724 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 5725 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 5726 sm_info->min_seq_blocks = BLKS_PER_SEG(sbi); 5727 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 5728 sm_info->min_ssr_sections = reserved_sections(sbi); 5729 5730 INIT_LIST_HEAD(&sm_info->sit_entry_set); 5731 5732 init_f2fs_rwsem(&sm_info->curseg_lock); 5733 5734 err = f2fs_create_flush_cmd_control(sbi); 5735 if (err) 5736 return err; 5737 5738 err = create_discard_cmd_control(sbi); 5739 if (err) 5740 return err; 5741 5742 err = build_sit_info(sbi); 5743 if (err) 5744 return err; 5745 err = build_free_segmap(sbi); 5746 if (err) 5747 return err; 5748 err = build_curseg(sbi); 5749 if (err) 5750 return err; 5751 5752 /* reinit free segmap based on SIT */ 5753 err = build_sit_entries(sbi); 5754 if (err) 5755 return err; 5756 5757 init_free_segmap(sbi); 5758 err = build_dirty_segmap(sbi); 5759 if (err) 5760 return err; 5761 5762 err = sanity_check_curseg(sbi); 5763 if (err) 5764 return err; 5765 5766 init_min_max_mtime(sbi); 5767 return 0; 5768 } 5769 5770 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 5771 enum dirty_type dirty_type) 5772 { 5773 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5774 5775 mutex_lock(&dirty_i->seglist_lock); 5776 kvfree(dirty_i->dirty_segmap[dirty_type]); 5777 dirty_i->nr_dirty[dirty_type] = 0; 5778 mutex_unlock(&dirty_i->seglist_lock); 5779 } 5780 5781 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 5782 { 5783 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5784 5785 kvfree(dirty_i->pinned_secmap); 5786 kvfree(dirty_i->victim_secmap); 5787 } 5788 5789 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 5790 { 5791 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5792 int i; 5793 5794 if (!dirty_i) 5795 return; 5796 5797 /* discard pre-free/dirty segments list */ 5798 for (i = 0; i < NR_DIRTY_TYPE; i++) 5799 discard_dirty_segmap(sbi, i); 5800 5801 if (__is_large_section(sbi)) { 5802 mutex_lock(&dirty_i->seglist_lock); 5803 kvfree(dirty_i->dirty_secmap); 5804 mutex_unlock(&dirty_i->seglist_lock); 5805 } 5806 5807 destroy_victim_secmap(sbi); 5808 SM_I(sbi)->dirty_info = NULL; 5809 kfree(dirty_i); 5810 } 5811 5812 static void destroy_curseg(struct f2fs_sb_info *sbi) 5813 { 5814 struct curseg_info *array = SM_I(sbi)->curseg_array; 5815 int i; 5816 5817 if (!array) 5818 return; 5819 SM_I(sbi)->curseg_array = NULL; 5820 for (i = 0; i < NR_CURSEG_TYPE; i++) { 5821 kfree(array[i].sum_blk); 5822 kfree(array[i].journal); 5823 } 5824 kfree(array); 5825 } 5826 5827 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 5828 { 5829 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 5830 5831 if (!free_i) 5832 return; 5833 SM_I(sbi)->free_info = NULL; 5834 kvfree(free_i->free_segmap); 5835 kvfree(free_i->free_secmap); 5836 kfree(free_i); 5837 } 5838 5839 static void destroy_sit_info(struct f2fs_sb_info *sbi) 5840 { 5841 struct sit_info *sit_i = SIT_I(sbi); 5842 5843 if (!sit_i) 5844 return; 5845 5846 if (sit_i->sentries) 5847 kvfree(sit_i->bitmap); 5848 kfree(sit_i->tmp_map); 5849 5850 kvfree(sit_i->sentries); 5851 kvfree(sit_i->sec_entries); 5852 kvfree(sit_i->dirty_sentries_bitmap); 5853 5854 SM_I(sbi)->sit_info = NULL; 5855 kfree(sit_i->sit_bitmap); 5856 #ifdef CONFIG_F2FS_CHECK_FS 5857 kfree(sit_i->sit_bitmap_mir); 5858 kvfree(sit_i->invalid_segmap); 5859 #endif 5860 kfree(sit_i); 5861 } 5862 5863 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 5864 { 5865 struct f2fs_sm_info *sm_info = SM_I(sbi); 5866 5867 if (!sm_info) 5868 return; 5869 f2fs_destroy_flush_cmd_control(sbi, true); 5870 destroy_discard_cmd_control(sbi); 5871 destroy_dirty_segmap(sbi); 5872 destroy_curseg(sbi); 5873 destroy_free_segmap(sbi); 5874 destroy_sit_info(sbi); 5875 sbi->sm_info = NULL; 5876 kfree(sm_info); 5877 } 5878 5879 int __init f2fs_create_segment_manager_caches(void) 5880 { 5881 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 5882 sizeof(struct discard_entry)); 5883 if (!discard_entry_slab) 5884 goto fail; 5885 5886 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 5887 sizeof(struct discard_cmd)); 5888 if (!discard_cmd_slab) 5889 goto destroy_discard_entry; 5890 5891 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 5892 sizeof(struct sit_entry_set)); 5893 if (!sit_entry_set_slab) 5894 goto destroy_discard_cmd; 5895 5896 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry", 5897 sizeof(struct revoke_entry)); 5898 if (!revoke_entry_slab) 5899 goto destroy_sit_entry_set; 5900 return 0; 5901 5902 destroy_sit_entry_set: 5903 kmem_cache_destroy(sit_entry_set_slab); 5904 destroy_discard_cmd: 5905 kmem_cache_destroy(discard_cmd_slab); 5906 destroy_discard_entry: 5907 kmem_cache_destroy(discard_entry_slab); 5908 fail: 5909 return -ENOMEM; 5910 } 5911 5912 void f2fs_destroy_segment_manager_caches(void) 5913 { 5914 kmem_cache_destroy(sit_entry_set_slab); 5915 kmem_cache_destroy(discard_cmd_slab); 5916 kmem_cache_destroy(discard_entry_slab); 5917 kmem_cache_destroy(revoke_entry_slab); 5918 } 5919