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 release_atomic_write_cnt(inode); 196 clear_inode_flag(inode, FI_ATOMIC_COMMITTED); 197 clear_inode_flag(inode, FI_ATOMIC_REPLACE); 198 clear_inode_flag(inode, FI_ATOMIC_FILE); 199 stat_dec_atomic_inode(inode); 200 201 F2FS_I(inode)->atomic_write_task = NULL; 202 203 if (clean) { 204 truncate_inode_pages_final(inode->i_mapping); 205 f2fs_i_size_write(inode, fi->original_i_size); 206 fi->original_i_size = 0; 207 } 208 } 209 210 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index, 211 block_t new_addr, block_t *old_addr, bool recover) 212 { 213 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 214 struct dnode_of_data dn; 215 struct node_info ni; 216 int err; 217 218 retry: 219 set_new_dnode(&dn, inode, NULL, NULL, 0); 220 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE); 221 if (err) { 222 if (err == -ENOMEM) { 223 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 224 goto retry; 225 } 226 return err; 227 } 228 229 err = f2fs_get_node_info(sbi, dn.nid, &ni, false); 230 if (err) { 231 f2fs_put_dnode(&dn); 232 return err; 233 } 234 235 if (recover) { 236 /* dn.data_blkaddr is always valid */ 237 if (!__is_valid_data_blkaddr(new_addr)) { 238 if (new_addr == NULL_ADDR) 239 dec_valid_block_count(sbi, inode, 1); 240 f2fs_invalidate_blocks(sbi, dn.data_blkaddr); 241 f2fs_update_data_blkaddr(&dn, new_addr); 242 } else { 243 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 244 new_addr, ni.version, true, true); 245 } 246 } else { 247 blkcnt_t count = 1; 248 249 err = inc_valid_block_count(sbi, inode, &count); 250 if (err) { 251 f2fs_put_dnode(&dn); 252 return err; 253 } 254 255 *old_addr = dn.data_blkaddr; 256 f2fs_truncate_data_blocks_range(&dn, 1); 257 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count); 258 259 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr, 260 ni.version, true, false); 261 } 262 263 f2fs_put_dnode(&dn); 264 265 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode, 266 index, old_addr ? *old_addr : 0, new_addr, recover); 267 return 0; 268 } 269 270 static void __complete_revoke_list(struct inode *inode, struct list_head *head, 271 bool revoke) 272 { 273 struct revoke_entry *cur, *tmp; 274 pgoff_t start_index = 0; 275 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE); 276 277 list_for_each_entry_safe(cur, tmp, head, list) { 278 if (revoke) { 279 __replace_atomic_write_block(inode, cur->index, 280 cur->old_addr, NULL, true); 281 } else if (truncate) { 282 f2fs_truncate_hole(inode, start_index, cur->index); 283 start_index = cur->index + 1; 284 } 285 286 list_del(&cur->list); 287 kmem_cache_free(revoke_entry_slab, cur); 288 } 289 290 if (!revoke && truncate) 291 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false); 292 } 293 294 static int __f2fs_commit_atomic_write(struct inode *inode) 295 { 296 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 297 struct f2fs_inode_info *fi = F2FS_I(inode); 298 struct inode *cow_inode = fi->cow_inode; 299 struct revoke_entry *new; 300 struct list_head revoke_list; 301 block_t blkaddr; 302 struct dnode_of_data dn; 303 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 304 pgoff_t off = 0, blen, index; 305 int ret = 0, i; 306 307 INIT_LIST_HEAD(&revoke_list); 308 309 while (len) { 310 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len); 311 312 set_new_dnode(&dn, cow_inode, NULL, NULL, 0); 313 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA); 314 if (ret && ret != -ENOENT) { 315 goto out; 316 } else if (ret == -ENOENT) { 317 ret = 0; 318 if (dn.max_level == 0) 319 goto out; 320 goto next; 321 } 322 323 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode), 324 len); 325 index = off; 326 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) { 327 blkaddr = f2fs_data_blkaddr(&dn); 328 329 if (!__is_valid_data_blkaddr(blkaddr)) { 330 continue; 331 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 332 DATA_GENERIC_ENHANCE)) { 333 f2fs_put_dnode(&dn); 334 ret = -EFSCORRUPTED; 335 f2fs_handle_error(sbi, 336 ERROR_INVALID_BLKADDR); 337 goto out; 338 } 339 340 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS, 341 true, NULL); 342 343 ret = __replace_atomic_write_block(inode, index, blkaddr, 344 &new->old_addr, false); 345 if (ret) { 346 f2fs_put_dnode(&dn); 347 kmem_cache_free(revoke_entry_slab, new); 348 goto out; 349 } 350 351 f2fs_update_data_blkaddr(&dn, NULL_ADDR); 352 new->index = index; 353 list_add_tail(&new->list, &revoke_list); 354 } 355 f2fs_put_dnode(&dn); 356 next: 357 off += blen; 358 len -= blen; 359 } 360 361 out: 362 if (ret) { 363 sbi->revoked_atomic_block += fi->atomic_write_cnt; 364 } else { 365 sbi->committed_atomic_block += fi->atomic_write_cnt; 366 set_inode_flag(inode, FI_ATOMIC_COMMITTED); 367 } 368 369 __complete_revoke_list(inode, &revoke_list, ret ? true : false); 370 371 return ret; 372 } 373 374 int f2fs_commit_atomic_write(struct inode *inode) 375 { 376 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 377 struct f2fs_inode_info *fi = F2FS_I(inode); 378 int err; 379 380 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 381 if (err) 382 return err; 383 384 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 385 f2fs_lock_op(sbi); 386 387 err = __f2fs_commit_atomic_write(inode); 388 389 f2fs_unlock_op(sbi); 390 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 391 392 return err; 393 } 394 395 /* 396 * This function balances dirty node and dentry pages. 397 * In addition, it controls garbage collection. 398 */ 399 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 400 { 401 if (time_to_inject(sbi, FAULT_CHECKPOINT)) 402 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT); 403 404 /* balance_fs_bg is able to be pending */ 405 if (need && excess_cached_nats(sbi)) 406 f2fs_balance_fs_bg(sbi, false); 407 408 if (!f2fs_is_checkpoint_ready(sbi)) 409 return; 410 411 /* 412 * We should do GC or end up with checkpoint, if there are so many dirty 413 * dir/node pages without enough free segments. 414 */ 415 if (has_enough_free_secs(sbi, 0, 0)) 416 return; 417 418 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread && 419 sbi->gc_thread->f2fs_gc_task) { 420 DEFINE_WAIT(wait); 421 422 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait, 423 TASK_UNINTERRUPTIBLE); 424 wake_up(&sbi->gc_thread->gc_wait_queue_head); 425 io_schedule(); 426 finish_wait(&sbi->gc_thread->fggc_wq, &wait); 427 } else { 428 struct f2fs_gc_control gc_control = { 429 .victim_segno = NULL_SEGNO, 430 .init_gc_type = BG_GC, 431 .no_bg_gc = true, 432 .should_migrate_blocks = false, 433 .err_gc_skipped = false, 434 .nr_free_secs = 1 }; 435 f2fs_down_write(&sbi->gc_lock); 436 f2fs_gc(sbi, &gc_control); 437 } 438 } 439 440 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi) 441 { 442 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2; 443 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS); 444 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA); 445 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES); 446 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META); 447 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA); 448 unsigned int threshold = sbi->blocks_per_seg * factor * 449 DEFAULT_DIRTY_THRESHOLD; 450 unsigned int global_threshold = threshold * 3 / 2; 451 452 if (dents >= threshold || qdata >= threshold || 453 nodes >= threshold || meta >= threshold || 454 imeta >= threshold) 455 return true; 456 return dents + qdata + nodes + meta + imeta > global_threshold; 457 } 458 459 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 460 { 461 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 462 return; 463 464 /* try to shrink extent cache when there is no enough memory */ 465 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE)) 466 f2fs_shrink_read_extent_tree(sbi, 467 READ_EXTENT_CACHE_SHRINK_NUMBER); 468 469 /* try to shrink age extent cache when there is no enough memory */ 470 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE)) 471 f2fs_shrink_age_extent_tree(sbi, 472 AGE_EXTENT_CACHE_SHRINK_NUMBER); 473 474 /* check the # of cached NAT entries */ 475 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 476 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 477 478 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 479 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 480 else 481 f2fs_build_free_nids(sbi, false, false); 482 483 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) || 484 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi)) 485 goto do_sync; 486 487 /* there is background inflight IO or foreground operation recently */ 488 if (is_inflight_io(sbi, REQ_TIME) || 489 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem))) 490 return; 491 492 /* exceed periodical checkpoint timeout threshold */ 493 if (f2fs_time_over(sbi, CP_TIME)) 494 goto do_sync; 495 496 /* checkpoint is the only way to shrink partial cached entries */ 497 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) && 498 f2fs_available_free_memory(sbi, INO_ENTRIES)) 499 return; 500 501 do_sync: 502 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 503 struct blk_plug plug; 504 505 mutex_lock(&sbi->flush_lock); 506 507 blk_start_plug(&plug); 508 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false); 509 blk_finish_plug(&plug); 510 511 mutex_unlock(&sbi->flush_lock); 512 } 513 f2fs_sync_fs(sbi->sb, 1); 514 stat_inc_bg_cp_count(sbi->stat_info); 515 } 516 517 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 518 struct block_device *bdev) 519 { 520 int ret = blkdev_issue_flush(bdev); 521 522 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 523 test_opt(sbi, FLUSH_MERGE), ret); 524 if (!ret) 525 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0); 526 return ret; 527 } 528 529 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 530 { 531 int ret = 0; 532 int i; 533 534 if (!f2fs_is_multi_device(sbi)) 535 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 536 537 for (i = 0; i < sbi->s_ndevs; i++) { 538 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 539 continue; 540 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 541 if (ret) 542 break; 543 } 544 return ret; 545 } 546 547 static int issue_flush_thread(void *data) 548 { 549 struct f2fs_sb_info *sbi = data; 550 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 551 wait_queue_head_t *q = &fcc->flush_wait_queue; 552 repeat: 553 if (kthread_should_stop()) 554 return 0; 555 556 if (!llist_empty(&fcc->issue_list)) { 557 struct flush_cmd *cmd, *next; 558 int ret; 559 560 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 561 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 562 563 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 564 565 ret = submit_flush_wait(sbi, cmd->ino); 566 atomic_inc(&fcc->issued_flush); 567 568 llist_for_each_entry_safe(cmd, next, 569 fcc->dispatch_list, llnode) { 570 cmd->ret = ret; 571 complete(&cmd->wait); 572 } 573 fcc->dispatch_list = NULL; 574 } 575 576 wait_event_interruptible(*q, 577 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 578 goto repeat; 579 } 580 581 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 582 { 583 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 584 struct flush_cmd cmd; 585 int ret; 586 587 if (test_opt(sbi, NOBARRIER)) 588 return 0; 589 590 if (!test_opt(sbi, FLUSH_MERGE)) { 591 atomic_inc(&fcc->queued_flush); 592 ret = submit_flush_wait(sbi, ino); 593 atomic_dec(&fcc->queued_flush); 594 atomic_inc(&fcc->issued_flush); 595 return ret; 596 } 597 598 if (atomic_inc_return(&fcc->queued_flush) == 1 || 599 f2fs_is_multi_device(sbi)) { 600 ret = submit_flush_wait(sbi, ino); 601 atomic_dec(&fcc->queued_flush); 602 603 atomic_inc(&fcc->issued_flush); 604 return ret; 605 } 606 607 cmd.ino = ino; 608 init_completion(&cmd.wait); 609 610 llist_add(&cmd.llnode, &fcc->issue_list); 611 612 /* 613 * update issue_list before we wake up issue_flush thread, this 614 * smp_mb() pairs with another barrier in ___wait_event(), see 615 * more details in comments of waitqueue_active(). 616 */ 617 smp_mb(); 618 619 if (waitqueue_active(&fcc->flush_wait_queue)) 620 wake_up(&fcc->flush_wait_queue); 621 622 if (fcc->f2fs_issue_flush) { 623 wait_for_completion(&cmd.wait); 624 atomic_dec(&fcc->queued_flush); 625 } else { 626 struct llist_node *list; 627 628 list = llist_del_all(&fcc->issue_list); 629 if (!list) { 630 wait_for_completion(&cmd.wait); 631 atomic_dec(&fcc->queued_flush); 632 } else { 633 struct flush_cmd *tmp, *next; 634 635 ret = submit_flush_wait(sbi, ino); 636 637 llist_for_each_entry_safe(tmp, next, list, llnode) { 638 if (tmp == &cmd) { 639 cmd.ret = ret; 640 atomic_dec(&fcc->queued_flush); 641 continue; 642 } 643 tmp->ret = ret; 644 complete(&tmp->wait); 645 } 646 } 647 } 648 649 return cmd.ret; 650 } 651 652 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 653 { 654 dev_t dev = sbi->sb->s_bdev->bd_dev; 655 struct flush_cmd_control *fcc; 656 657 if (SM_I(sbi)->fcc_info) { 658 fcc = SM_I(sbi)->fcc_info; 659 if (fcc->f2fs_issue_flush) 660 return 0; 661 goto init_thread; 662 } 663 664 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 665 if (!fcc) 666 return -ENOMEM; 667 atomic_set(&fcc->issued_flush, 0); 668 atomic_set(&fcc->queued_flush, 0); 669 init_waitqueue_head(&fcc->flush_wait_queue); 670 init_llist_head(&fcc->issue_list); 671 SM_I(sbi)->fcc_info = fcc; 672 if (!test_opt(sbi, FLUSH_MERGE)) 673 return 0; 674 675 init_thread: 676 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 677 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 678 if (IS_ERR(fcc->f2fs_issue_flush)) { 679 int err = PTR_ERR(fcc->f2fs_issue_flush); 680 681 fcc->f2fs_issue_flush = NULL; 682 return err; 683 } 684 685 return 0; 686 } 687 688 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 689 { 690 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 691 692 if (fcc && fcc->f2fs_issue_flush) { 693 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 694 695 fcc->f2fs_issue_flush = NULL; 696 kthread_stop(flush_thread); 697 } 698 if (free) { 699 kfree(fcc); 700 SM_I(sbi)->fcc_info = NULL; 701 } 702 } 703 704 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 705 { 706 int ret = 0, i; 707 708 if (!f2fs_is_multi_device(sbi)) 709 return 0; 710 711 if (test_opt(sbi, NOBARRIER)) 712 return 0; 713 714 for (i = 1; i < sbi->s_ndevs; i++) { 715 int count = DEFAULT_RETRY_IO_COUNT; 716 717 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 718 continue; 719 720 do { 721 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 722 if (ret) 723 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 724 } while (ret && --count); 725 726 if (ret) { 727 f2fs_stop_checkpoint(sbi, false, 728 STOP_CP_REASON_FLUSH_FAIL); 729 break; 730 } 731 732 spin_lock(&sbi->dev_lock); 733 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 734 spin_unlock(&sbi->dev_lock); 735 } 736 737 return ret; 738 } 739 740 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 741 enum dirty_type dirty_type) 742 { 743 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 744 745 /* need not be added */ 746 if (IS_CURSEG(sbi, segno)) 747 return; 748 749 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 750 dirty_i->nr_dirty[dirty_type]++; 751 752 if (dirty_type == DIRTY) { 753 struct seg_entry *sentry = get_seg_entry(sbi, segno); 754 enum dirty_type t = sentry->type; 755 756 if (unlikely(t >= DIRTY)) { 757 f2fs_bug_on(sbi, 1); 758 return; 759 } 760 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 761 dirty_i->nr_dirty[t]++; 762 763 if (__is_large_section(sbi)) { 764 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 765 block_t valid_blocks = 766 get_valid_blocks(sbi, segno, true); 767 768 f2fs_bug_on(sbi, unlikely(!valid_blocks || 769 valid_blocks == CAP_BLKS_PER_SEC(sbi))); 770 771 if (!IS_CURSEC(sbi, secno)) 772 set_bit(secno, dirty_i->dirty_secmap); 773 } 774 } 775 } 776 777 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 778 enum dirty_type dirty_type) 779 { 780 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 781 block_t valid_blocks; 782 783 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 784 dirty_i->nr_dirty[dirty_type]--; 785 786 if (dirty_type == DIRTY) { 787 struct seg_entry *sentry = get_seg_entry(sbi, segno); 788 enum dirty_type t = sentry->type; 789 790 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 791 dirty_i->nr_dirty[t]--; 792 793 valid_blocks = get_valid_blocks(sbi, segno, true); 794 if (valid_blocks == 0) { 795 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 796 dirty_i->victim_secmap); 797 #ifdef CONFIG_F2FS_CHECK_FS 798 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 799 #endif 800 } 801 if (__is_large_section(sbi)) { 802 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 803 804 if (!valid_blocks || 805 valid_blocks == CAP_BLKS_PER_SEC(sbi)) { 806 clear_bit(secno, dirty_i->dirty_secmap); 807 return; 808 } 809 810 if (!IS_CURSEC(sbi, secno)) 811 set_bit(secno, dirty_i->dirty_secmap); 812 } 813 } 814 } 815 816 /* 817 * Should not occur error such as -ENOMEM. 818 * Adding dirty entry into seglist is not critical operation. 819 * If a given segment is one of current working segments, it won't be added. 820 */ 821 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 822 { 823 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 824 unsigned short valid_blocks, ckpt_valid_blocks; 825 unsigned int usable_blocks; 826 827 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 828 return; 829 830 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno); 831 mutex_lock(&dirty_i->seglist_lock); 832 833 valid_blocks = get_valid_blocks(sbi, segno, false); 834 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false); 835 836 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 837 ckpt_valid_blocks == usable_blocks)) { 838 __locate_dirty_segment(sbi, segno, PRE); 839 __remove_dirty_segment(sbi, segno, DIRTY); 840 } else if (valid_blocks < usable_blocks) { 841 __locate_dirty_segment(sbi, segno, DIRTY); 842 } else { 843 /* Recovery routine with SSR needs this */ 844 __remove_dirty_segment(sbi, segno, DIRTY); 845 } 846 847 mutex_unlock(&dirty_i->seglist_lock); 848 } 849 850 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 851 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 852 { 853 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 854 unsigned int segno; 855 856 mutex_lock(&dirty_i->seglist_lock); 857 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 858 if (get_valid_blocks(sbi, segno, false)) 859 continue; 860 if (IS_CURSEG(sbi, segno)) 861 continue; 862 __locate_dirty_segment(sbi, segno, PRE); 863 __remove_dirty_segment(sbi, segno, DIRTY); 864 } 865 mutex_unlock(&dirty_i->seglist_lock); 866 } 867 868 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 869 { 870 int ovp_hole_segs = 871 (overprovision_segments(sbi) - reserved_segments(sbi)); 872 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg; 873 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 874 block_t holes[2] = {0, 0}; /* DATA and NODE */ 875 block_t unusable; 876 struct seg_entry *se; 877 unsigned int segno; 878 879 mutex_lock(&dirty_i->seglist_lock); 880 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 881 se = get_seg_entry(sbi, segno); 882 if (IS_NODESEG(se->type)) 883 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) - 884 se->valid_blocks; 885 else 886 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) - 887 se->valid_blocks; 888 } 889 mutex_unlock(&dirty_i->seglist_lock); 890 891 unusable = max(holes[DATA], holes[NODE]); 892 if (unusable > ovp_holes) 893 return unusable - ovp_holes; 894 return 0; 895 } 896 897 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 898 { 899 int ovp_hole_segs = 900 (overprovision_segments(sbi) - reserved_segments(sbi)); 901 if (unusable > F2FS_OPTION(sbi).unusable_cap) 902 return -EAGAIN; 903 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 904 dirty_segments(sbi) > ovp_hole_segs) 905 return -EAGAIN; 906 return 0; 907 } 908 909 /* This is only used by SBI_CP_DISABLED */ 910 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 911 { 912 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 913 unsigned int segno = 0; 914 915 mutex_lock(&dirty_i->seglist_lock); 916 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 917 if (get_valid_blocks(sbi, segno, false)) 918 continue; 919 if (get_ckpt_valid_blocks(sbi, segno, false)) 920 continue; 921 mutex_unlock(&dirty_i->seglist_lock); 922 return segno; 923 } 924 mutex_unlock(&dirty_i->seglist_lock); 925 return NULL_SEGNO; 926 } 927 928 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 929 struct block_device *bdev, block_t lstart, 930 block_t start, block_t len) 931 { 932 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 933 struct list_head *pend_list; 934 struct discard_cmd *dc; 935 936 f2fs_bug_on(sbi, !len); 937 938 pend_list = &dcc->pend_list[plist_idx(len)]; 939 940 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL); 941 INIT_LIST_HEAD(&dc->list); 942 dc->bdev = bdev; 943 dc->di.lstart = lstart; 944 dc->di.start = start; 945 dc->di.len = len; 946 dc->ref = 0; 947 dc->state = D_PREP; 948 dc->queued = 0; 949 dc->error = 0; 950 init_completion(&dc->wait); 951 list_add_tail(&dc->list, pend_list); 952 spin_lock_init(&dc->lock); 953 dc->bio_ref = 0; 954 atomic_inc(&dcc->discard_cmd_cnt); 955 dcc->undiscard_blks += len; 956 957 return dc; 958 } 959 960 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi) 961 { 962 #ifdef CONFIG_F2FS_CHECK_FS 963 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 964 struct rb_node *cur = rb_first_cached(&dcc->root), *next; 965 struct discard_cmd *cur_dc, *next_dc; 966 967 while (cur) { 968 next = rb_next(cur); 969 if (!next) 970 return true; 971 972 cur_dc = rb_entry(cur, struct discard_cmd, rb_node); 973 next_dc = rb_entry(next, struct discard_cmd, rb_node); 974 975 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) { 976 f2fs_info(sbi, "broken discard_rbtree, " 977 "cur(%u, %u) next(%u, %u)", 978 cur_dc->di.lstart, cur_dc->di.len, 979 next_dc->di.lstart, next_dc->di.len); 980 return false; 981 } 982 cur = next; 983 } 984 #endif 985 return true; 986 } 987 988 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi, 989 block_t blkaddr) 990 { 991 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 992 struct rb_node *node = dcc->root.rb_root.rb_node; 993 struct discard_cmd *dc; 994 995 while (node) { 996 dc = rb_entry(node, struct discard_cmd, rb_node); 997 998 if (blkaddr < dc->di.lstart) 999 node = node->rb_left; 1000 else if (blkaddr >= dc->di.lstart + dc->di.len) 1001 node = node->rb_right; 1002 else 1003 return dc; 1004 } 1005 return NULL; 1006 } 1007 1008 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root, 1009 block_t blkaddr, 1010 struct discard_cmd **prev_entry, 1011 struct discard_cmd **next_entry, 1012 struct rb_node ***insert_p, 1013 struct rb_node **insert_parent) 1014 { 1015 struct rb_node **pnode = &root->rb_root.rb_node; 1016 struct rb_node *parent = NULL, *tmp_node; 1017 struct discard_cmd *dc; 1018 1019 *insert_p = NULL; 1020 *insert_parent = NULL; 1021 *prev_entry = NULL; 1022 *next_entry = NULL; 1023 1024 if (RB_EMPTY_ROOT(&root->rb_root)) 1025 return NULL; 1026 1027 while (*pnode) { 1028 parent = *pnode; 1029 dc = rb_entry(*pnode, struct discard_cmd, rb_node); 1030 1031 if (blkaddr < dc->di.lstart) 1032 pnode = &(*pnode)->rb_left; 1033 else if (blkaddr >= dc->di.lstart + dc->di.len) 1034 pnode = &(*pnode)->rb_right; 1035 else 1036 goto lookup_neighbors; 1037 } 1038 1039 *insert_p = pnode; 1040 *insert_parent = parent; 1041 1042 dc = rb_entry(parent, struct discard_cmd, rb_node); 1043 tmp_node = parent; 1044 if (parent && blkaddr > dc->di.lstart) 1045 tmp_node = rb_next(parent); 1046 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1047 1048 tmp_node = parent; 1049 if (parent && blkaddr < dc->di.lstart) 1050 tmp_node = rb_prev(parent); 1051 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1052 return NULL; 1053 1054 lookup_neighbors: 1055 /* lookup prev node for merging backward later */ 1056 tmp_node = rb_prev(&dc->rb_node); 1057 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1058 1059 /* lookup next node for merging frontward later */ 1060 tmp_node = rb_next(&dc->rb_node); 1061 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1062 return dc; 1063 } 1064 1065 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 1066 struct discard_cmd *dc) 1067 { 1068 if (dc->state == D_DONE) 1069 atomic_sub(dc->queued, &dcc->queued_discard); 1070 1071 list_del(&dc->list); 1072 rb_erase_cached(&dc->rb_node, &dcc->root); 1073 dcc->undiscard_blks -= dc->di.len; 1074 1075 kmem_cache_free(discard_cmd_slab, dc); 1076 1077 atomic_dec(&dcc->discard_cmd_cnt); 1078 } 1079 1080 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 1081 struct discard_cmd *dc) 1082 { 1083 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1084 unsigned long flags; 1085 1086 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len); 1087 1088 spin_lock_irqsave(&dc->lock, flags); 1089 if (dc->bio_ref) { 1090 spin_unlock_irqrestore(&dc->lock, flags); 1091 return; 1092 } 1093 spin_unlock_irqrestore(&dc->lock, flags); 1094 1095 f2fs_bug_on(sbi, dc->ref); 1096 1097 if (dc->error == -EOPNOTSUPP) 1098 dc->error = 0; 1099 1100 if (dc->error) 1101 printk_ratelimited( 1102 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d", 1103 KERN_INFO, sbi->sb->s_id, 1104 dc->di.lstart, dc->di.start, dc->di.len, dc->error); 1105 __detach_discard_cmd(dcc, dc); 1106 } 1107 1108 static void f2fs_submit_discard_endio(struct bio *bio) 1109 { 1110 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1111 unsigned long flags; 1112 1113 spin_lock_irqsave(&dc->lock, flags); 1114 if (!dc->error) 1115 dc->error = blk_status_to_errno(bio->bi_status); 1116 dc->bio_ref--; 1117 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1118 dc->state = D_DONE; 1119 complete_all(&dc->wait); 1120 } 1121 spin_unlock_irqrestore(&dc->lock, flags); 1122 bio_put(bio); 1123 } 1124 1125 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1126 block_t start, block_t end) 1127 { 1128 #ifdef CONFIG_F2FS_CHECK_FS 1129 struct seg_entry *sentry; 1130 unsigned int segno; 1131 block_t blk = start; 1132 unsigned long offset, size, max_blocks = sbi->blocks_per_seg; 1133 unsigned long *map; 1134 1135 while (blk < end) { 1136 segno = GET_SEGNO(sbi, blk); 1137 sentry = get_seg_entry(sbi, segno); 1138 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1139 1140 if (end < START_BLOCK(sbi, segno + 1)) 1141 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1142 else 1143 size = max_blocks; 1144 map = (unsigned long *)(sentry->cur_valid_map); 1145 offset = __find_rev_next_bit(map, size, offset); 1146 f2fs_bug_on(sbi, offset != size); 1147 blk = START_BLOCK(sbi, segno + 1); 1148 } 1149 #endif 1150 } 1151 1152 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1153 struct discard_policy *dpolicy, 1154 int discard_type, unsigned int granularity) 1155 { 1156 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1157 1158 /* common policy */ 1159 dpolicy->type = discard_type; 1160 dpolicy->sync = true; 1161 dpolicy->ordered = false; 1162 dpolicy->granularity = granularity; 1163 1164 dpolicy->max_requests = dcc->max_discard_request; 1165 dpolicy->io_aware_gran = dcc->discard_io_aware_gran; 1166 dpolicy->timeout = false; 1167 1168 if (discard_type == DPOLICY_BG) { 1169 dpolicy->min_interval = dcc->min_discard_issue_time; 1170 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1171 dpolicy->max_interval = dcc->max_discard_issue_time; 1172 dpolicy->io_aware = true; 1173 dpolicy->sync = false; 1174 dpolicy->ordered = true; 1175 if (utilization(sbi) > dcc->discard_urgent_util) { 1176 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1177 if (atomic_read(&dcc->discard_cmd_cnt)) 1178 dpolicy->max_interval = 1179 dcc->min_discard_issue_time; 1180 } 1181 } else if (discard_type == DPOLICY_FORCE) { 1182 dpolicy->min_interval = dcc->min_discard_issue_time; 1183 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1184 dpolicy->max_interval = dcc->max_discard_issue_time; 1185 dpolicy->io_aware = false; 1186 } else if (discard_type == DPOLICY_FSTRIM) { 1187 dpolicy->io_aware = false; 1188 } else if (discard_type == DPOLICY_UMOUNT) { 1189 dpolicy->io_aware = false; 1190 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1191 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1192 dpolicy->timeout = true; 1193 } 1194 } 1195 1196 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1197 struct block_device *bdev, block_t lstart, 1198 block_t start, block_t len); 1199 1200 #ifdef CONFIG_BLK_DEV_ZONED 1201 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi, 1202 struct discard_cmd *dc, blk_opf_t flag, 1203 struct list_head *wait_list, 1204 unsigned int *issued) 1205 { 1206 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1207 struct block_device *bdev = dc->bdev; 1208 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS); 1209 unsigned long flags; 1210 1211 trace_f2fs_issue_reset_zone(bdev, dc->di.start); 1212 1213 spin_lock_irqsave(&dc->lock, flags); 1214 dc->state = D_SUBMIT; 1215 dc->bio_ref++; 1216 spin_unlock_irqrestore(&dc->lock, flags); 1217 1218 if (issued) 1219 (*issued)++; 1220 1221 atomic_inc(&dcc->queued_discard); 1222 dc->queued++; 1223 list_move_tail(&dc->list, wait_list); 1224 1225 /* sanity check on discard range */ 1226 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len); 1227 1228 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start); 1229 bio->bi_private = dc; 1230 bio->bi_end_io = f2fs_submit_discard_endio; 1231 submit_bio(bio); 1232 1233 atomic_inc(&dcc->issued_discard); 1234 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE); 1235 } 1236 #endif 1237 1238 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1239 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1240 struct discard_policy *dpolicy, 1241 struct discard_cmd *dc, int *issued) 1242 { 1243 struct block_device *bdev = dc->bdev; 1244 unsigned int max_discard_blocks = 1245 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1246 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1247 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1248 &(dcc->fstrim_list) : &(dcc->wait_list); 1249 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0; 1250 block_t lstart, start, len, total_len; 1251 int err = 0; 1252 1253 if (dc->state != D_PREP) 1254 return 0; 1255 1256 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1257 return 0; 1258 1259 #ifdef CONFIG_BLK_DEV_ZONED 1260 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) { 1261 __submit_zone_reset_cmd(sbi, dc, flag, wait_list, issued); 1262 return 0; 1263 } 1264 #endif 1265 1266 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len); 1267 1268 lstart = dc->di.lstart; 1269 start = dc->di.start; 1270 len = dc->di.len; 1271 total_len = len; 1272 1273 dc->di.len = 0; 1274 1275 while (total_len && *issued < dpolicy->max_requests && !err) { 1276 struct bio *bio = NULL; 1277 unsigned long flags; 1278 bool last = true; 1279 1280 if (len > max_discard_blocks) { 1281 len = max_discard_blocks; 1282 last = false; 1283 } 1284 1285 (*issued)++; 1286 if (*issued == dpolicy->max_requests) 1287 last = true; 1288 1289 dc->di.len += len; 1290 1291 if (time_to_inject(sbi, FAULT_DISCARD)) { 1292 err = -EIO; 1293 } else { 1294 err = __blkdev_issue_discard(bdev, 1295 SECTOR_FROM_BLOCK(start), 1296 SECTOR_FROM_BLOCK(len), 1297 GFP_NOFS, &bio); 1298 } 1299 if (err) { 1300 spin_lock_irqsave(&dc->lock, flags); 1301 if (dc->state == D_PARTIAL) 1302 dc->state = D_SUBMIT; 1303 spin_unlock_irqrestore(&dc->lock, flags); 1304 1305 break; 1306 } 1307 1308 f2fs_bug_on(sbi, !bio); 1309 1310 /* 1311 * should keep before submission to avoid D_DONE 1312 * right away 1313 */ 1314 spin_lock_irqsave(&dc->lock, flags); 1315 if (last) 1316 dc->state = D_SUBMIT; 1317 else 1318 dc->state = D_PARTIAL; 1319 dc->bio_ref++; 1320 spin_unlock_irqrestore(&dc->lock, flags); 1321 1322 atomic_inc(&dcc->queued_discard); 1323 dc->queued++; 1324 list_move_tail(&dc->list, wait_list); 1325 1326 /* sanity check on discard range */ 1327 __check_sit_bitmap(sbi, lstart, lstart + len); 1328 1329 bio->bi_private = dc; 1330 bio->bi_end_io = f2fs_submit_discard_endio; 1331 bio->bi_opf |= flag; 1332 submit_bio(bio); 1333 1334 atomic_inc(&dcc->issued_discard); 1335 1336 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE); 1337 1338 lstart += len; 1339 start += len; 1340 total_len -= len; 1341 len = total_len; 1342 } 1343 1344 if (!err && len) { 1345 dcc->undiscard_blks -= len; 1346 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1347 } 1348 return err; 1349 } 1350 1351 static void __insert_discard_cmd(struct f2fs_sb_info *sbi, 1352 struct block_device *bdev, block_t lstart, 1353 block_t start, block_t len) 1354 { 1355 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1356 struct rb_node **p = &dcc->root.rb_root.rb_node; 1357 struct rb_node *parent = NULL; 1358 struct discard_cmd *dc; 1359 bool leftmost = true; 1360 1361 /* look up rb tree to find parent node */ 1362 while (*p) { 1363 parent = *p; 1364 dc = rb_entry(parent, struct discard_cmd, rb_node); 1365 1366 if (lstart < dc->di.lstart) { 1367 p = &(*p)->rb_left; 1368 } else if (lstart >= dc->di.lstart + dc->di.len) { 1369 p = &(*p)->rb_right; 1370 leftmost = false; 1371 } else { 1372 f2fs_bug_on(sbi, 1); 1373 } 1374 } 1375 1376 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 1377 1378 rb_link_node(&dc->rb_node, parent, p); 1379 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 1380 } 1381 1382 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1383 struct discard_cmd *dc) 1384 { 1385 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]); 1386 } 1387 1388 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1389 struct discard_cmd *dc, block_t blkaddr) 1390 { 1391 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1392 struct discard_info di = dc->di; 1393 bool modified = false; 1394 1395 if (dc->state == D_DONE || dc->di.len == 1) { 1396 __remove_discard_cmd(sbi, dc); 1397 return; 1398 } 1399 1400 dcc->undiscard_blks -= di.len; 1401 1402 if (blkaddr > di.lstart) { 1403 dc->di.len = blkaddr - dc->di.lstart; 1404 dcc->undiscard_blks += dc->di.len; 1405 __relocate_discard_cmd(dcc, dc); 1406 modified = true; 1407 } 1408 1409 if (blkaddr < di.lstart + di.len - 1) { 1410 if (modified) { 1411 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1, 1412 di.start + blkaddr + 1 - di.lstart, 1413 di.lstart + di.len - 1 - blkaddr); 1414 } else { 1415 dc->di.lstart++; 1416 dc->di.len--; 1417 dc->di.start++; 1418 dcc->undiscard_blks += dc->di.len; 1419 __relocate_discard_cmd(dcc, dc); 1420 } 1421 } 1422 } 1423 1424 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1425 struct block_device *bdev, block_t lstart, 1426 block_t start, block_t len) 1427 { 1428 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1429 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1430 struct discard_cmd *dc; 1431 struct discard_info di = {0}; 1432 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1433 unsigned int max_discard_blocks = 1434 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1435 block_t end = lstart + len; 1436 1437 dc = __lookup_discard_cmd_ret(&dcc->root, lstart, 1438 &prev_dc, &next_dc, &insert_p, &insert_parent); 1439 if (dc) 1440 prev_dc = dc; 1441 1442 if (!prev_dc) { 1443 di.lstart = lstart; 1444 di.len = next_dc ? next_dc->di.lstart - lstart : len; 1445 di.len = min(di.len, len); 1446 di.start = start; 1447 } 1448 1449 while (1) { 1450 struct rb_node *node; 1451 bool merged = false; 1452 struct discard_cmd *tdc = NULL; 1453 1454 if (prev_dc) { 1455 di.lstart = prev_dc->di.lstart + prev_dc->di.len; 1456 if (di.lstart < lstart) 1457 di.lstart = lstart; 1458 if (di.lstart >= end) 1459 break; 1460 1461 if (!next_dc || next_dc->di.lstart > end) 1462 di.len = end - di.lstart; 1463 else 1464 di.len = next_dc->di.lstart - di.lstart; 1465 di.start = start + di.lstart - lstart; 1466 } 1467 1468 if (!di.len) 1469 goto next; 1470 1471 if (prev_dc && prev_dc->state == D_PREP && 1472 prev_dc->bdev == bdev && 1473 __is_discard_back_mergeable(&di, &prev_dc->di, 1474 max_discard_blocks)) { 1475 prev_dc->di.len += di.len; 1476 dcc->undiscard_blks += di.len; 1477 __relocate_discard_cmd(dcc, prev_dc); 1478 di = prev_dc->di; 1479 tdc = prev_dc; 1480 merged = true; 1481 } 1482 1483 if (next_dc && next_dc->state == D_PREP && 1484 next_dc->bdev == bdev && 1485 __is_discard_front_mergeable(&di, &next_dc->di, 1486 max_discard_blocks)) { 1487 next_dc->di.lstart = di.lstart; 1488 next_dc->di.len += di.len; 1489 next_dc->di.start = di.start; 1490 dcc->undiscard_blks += di.len; 1491 __relocate_discard_cmd(dcc, next_dc); 1492 if (tdc) 1493 __remove_discard_cmd(sbi, tdc); 1494 merged = true; 1495 } 1496 1497 if (!merged) 1498 __insert_discard_cmd(sbi, bdev, 1499 di.lstart, di.start, di.len); 1500 next: 1501 prev_dc = next_dc; 1502 if (!prev_dc) 1503 break; 1504 1505 node = rb_next(&prev_dc->rb_node); 1506 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1507 } 1508 } 1509 1510 #ifdef CONFIG_BLK_DEV_ZONED 1511 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi, 1512 struct block_device *bdev, block_t blkstart, block_t lblkstart, 1513 block_t blklen) 1514 { 1515 trace_f2fs_queue_reset_zone(bdev, blkstart); 1516 1517 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1518 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen); 1519 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1520 } 1521 #endif 1522 1523 static void __queue_discard_cmd(struct f2fs_sb_info *sbi, 1524 struct block_device *bdev, block_t blkstart, block_t blklen) 1525 { 1526 block_t lblkstart = blkstart; 1527 1528 if (!f2fs_bdev_support_discard(bdev)) 1529 return; 1530 1531 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1532 1533 if (f2fs_is_multi_device(sbi)) { 1534 int devi = f2fs_target_device_index(sbi, blkstart); 1535 1536 blkstart -= FDEV(devi).start_blk; 1537 } 1538 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1539 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1540 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1541 } 1542 1543 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1544 struct discard_policy *dpolicy, int *issued) 1545 { 1546 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1547 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1548 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1549 struct discard_cmd *dc; 1550 struct blk_plug plug; 1551 bool io_interrupted = false; 1552 1553 mutex_lock(&dcc->cmd_lock); 1554 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos, 1555 &prev_dc, &next_dc, &insert_p, &insert_parent); 1556 if (!dc) 1557 dc = next_dc; 1558 1559 blk_start_plug(&plug); 1560 1561 while (dc) { 1562 struct rb_node *node; 1563 int err = 0; 1564 1565 if (dc->state != D_PREP) 1566 goto next; 1567 1568 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1569 io_interrupted = true; 1570 break; 1571 } 1572 1573 dcc->next_pos = dc->di.lstart + dc->di.len; 1574 err = __submit_discard_cmd(sbi, dpolicy, dc, issued); 1575 1576 if (*issued >= dpolicy->max_requests) 1577 break; 1578 next: 1579 node = rb_next(&dc->rb_node); 1580 if (err) 1581 __remove_discard_cmd(sbi, dc); 1582 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1583 } 1584 1585 blk_finish_plug(&plug); 1586 1587 if (!dc) 1588 dcc->next_pos = 0; 1589 1590 mutex_unlock(&dcc->cmd_lock); 1591 1592 if (!(*issued) && io_interrupted) 1593 *issued = -1; 1594 } 1595 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1596 struct discard_policy *dpolicy); 1597 1598 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1599 struct discard_policy *dpolicy) 1600 { 1601 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1602 struct list_head *pend_list; 1603 struct discard_cmd *dc, *tmp; 1604 struct blk_plug plug; 1605 int i, issued; 1606 bool io_interrupted = false; 1607 1608 if (dpolicy->timeout) 1609 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1610 1611 retry: 1612 issued = 0; 1613 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1614 if (dpolicy->timeout && 1615 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1616 break; 1617 1618 if (i + 1 < dpolicy->granularity) 1619 break; 1620 1621 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) { 1622 __issue_discard_cmd_orderly(sbi, dpolicy, &issued); 1623 return issued; 1624 } 1625 1626 pend_list = &dcc->pend_list[i]; 1627 1628 mutex_lock(&dcc->cmd_lock); 1629 if (list_empty(pend_list)) 1630 goto next; 1631 if (unlikely(dcc->rbtree_check)) 1632 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 1633 blk_start_plug(&plug); 1634 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1635 f2fs_bug_on(sbi, dc->state != D_PREP); 1636 1637 if (dpolicy->timeout && 1638 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1639 break; 1640 1641 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1642 !is_idle(sbi, DISCARD_TIME)) { 1643 io_interrupted = true; 1644 break; 1645 } 1646 1647 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1648 1649 if (issued >= dpolicy->max_requests) 1650 break; 1651 } 1652 blk_finish_plug(&plug); 1653 next: 1654 mutex_unlock(&dcc->cmd_lock); 1655 1656 if (issued >= dpolicy->max_requests || io_interrupted) 1657 break; 1658 } 1659 1660 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1661 __wait_all_discard_cmd(sbi, dpolicy); 1662 goto retry; 1663 } 1664 1665 if (!issued && io_interrupted) 1666 issued = -1; 1667 1668 return issued; 1669 } 1670 1671 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1672 { 1673 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1674 struct list_head *pend_list; 1675 struct discard_cmd *dc, *tmp; 1676 int i; 1677 bool dropped = false; 1678 1679 mutex_lock(&dcc->cmd_lock); 1680 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1681 pend_list = &dcc->pend_list[i]; 1682 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1683 f2fs_bug_on(sbi, dc->state != D_PREP); 1684 __remove_discard_cmd(sbi, dc); 1685 dropped = true; 1686 } 1687 } 1688 mutex_unlock(&dcc->cmd_lock); 1689 1690 return dropped; 1691 } 1692 1693 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1694 { 1695 __drop_discard_cmd(sbi); 1696 } 1697 1698 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1699 struct discard_cmd *dc) 1700 { 1701 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1702 unsigned int len = 0; 1703 1704 wait_for_completion_io(&dc->wait); 1705 mutex_lock(&dcc->cmd_lock); 1706 f2fs_bug_on(sbi, dc->state != D_DONE); 1707 dc->ref--; 1708 if (!dc->ref) { 1709 if (!dc->error) 1710 len = dc->di.len; 1711 __remove_discard_cmd(sbi, dc); 1712 } 1713 mutex_unlock(&dcc->cmd_lock); 1714 1715 return len; 1716 } 1717 1718 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1719 struct discard_policy *dpolicy, 1720 block_t start, block_t end) 1721 { 1722 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1723 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1724 &(dcc->fstrim_list) : &(dcc->wait_list); 1725 struct discard_cmd *dc = NULL, *iter, *tmp; 1726 unsigned int trimmed = 0; 1727 1728 next: 1729 dc = NULL; 1730 1731 mutex_lock(&dcc->cmd_lock); 1732 list_for_each_entry_safe(iter, tmp, wait_list, list) { 1733 if (iter->di.lstart + iter->di.len <= start || 1734 end <= iter->di.lstart) 1735 continue; 1736 if (iter->di.len < dpolicy->granularity) 1737 continue; 1738 if (iter->state == D_DONE && !iter->ref) { 1739 wait_for_completion_io(&iter->wait); 1740 if (!iter->error) 1741 trimmed += iter->di.len; 1742 __remove_discard_cmd(sbi, iter); 1743 } else { 1744 iter->ref++; 1745 dc = iter; 1746 break; 1747 } 1748 } 1749 mutex_unlock(&dcc->cmd_lock); 1750 1751 if (dc) { 1752 trimmed += __wait_one_discard_bio(sbi, dc); 1753 goto next; 1754 } 1755 1756 return trimmed; 1757 } 1758 1759 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1760 struct discard_policy *dpolicy) 1761 { 1762 struct discard_policy dp; 1763 unsigned int discard_blks; 1764 1765 if (dpolicy) 1766 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1767 1768 /* wait all */ 1769 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY); 1770 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1771 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY); 1772 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1773 1774 return discard_blks; 1775 } 1776 1777 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1778 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1779 { 1780 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1781 struct discard_cmd *dc; 1782 bool need_wait = false; 1783 1784 mutex_lock(&dcc->cmd_lock); 1785 dc = __lookup_discard_cmd(sbi, blkaddr); 1786 #ifdef CONFIG_BLK_DEV_ZONED 1787 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) { 1788 /* force submit zone reset */ 1789 if (dc->state == D_PREP) 1790 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC, 1791 &dcc->wait_list, NULL); 1792 dc->ref++; 1793 mutex_unlock(&dcc->cmd_lock); 1794 /* wait zone reset */ 1795 __wait_one_discard_bio(sbi, dc); 1796 return; 1797 } 1798 #endif 1799 if (dc) { 1800 if (dc->state == D_PREP) { 1801 __punch_discard_cmd(sbi, dc, blkaddr); 1802 } else { 1803 dc->ref++; 1804 need_wait = true; 1805 } 1806 } 1807 mutex_unlock(&dcc->cmd_lock); 1808 1809 if (need_wait) 1810 __wait_one_discard_bio(sbi, dc); 1811 } 1812 1813 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1814 { 1815 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1816 1817 if (dcc && dcc->f2fs_issue_discard) { 1818 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1819 1820 dcc->f2fs_issue_discard = NULL; 1821 kthread_stop(discard_thread); 1822 } 1823 } 1824 1825 /** 1826 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT 1827 * @sbi: the f2fs_sb_info data for discard cmd to issue 1828 * 1829 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped 1830 * 1831 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false. 1832 */ 1833 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1834 { 1835 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1836 struct discard_policy dpolicy; 1837 bool dropped; 1838 1839 if (!atomic_read(&dcc->discard_cmd_cnt)) 1840 return true; 1841 1842 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1843 dcc->discard_granularity); 1844 __issue_discard_cmd(sbi, &dpolicy); 1845 dropped = __drop_discard_cmd(sbi); 1846 1847 /* just to make sure there is no pending discard commands */ 1848 __wait_all_discard_cmd(sbi, NULL); 1849 1850 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1851 return !dropped; 1852 } 1853 1854 static int issue_discard_thread(void *data) 1855 { 1856 struct f2fs_sb_info *sbi = data; 1857 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1858 wait_queue_head_t *q = &dcc->discard_wait_queue; 1859 struct discard_policy dpolicy; 1860 unsigned int wait_ms = dcc->min_discard_issue_time; 1861 int issued; 1862 1863 set_freezable(); 1864 1865 do { 1866 wait_event_interruptible_timeout(*q, 1867 kthread_should_stop() || freezing(current) || 1868 dcc->discard_wake, 1869 msecs_to_jiffies(wait_ms)); 1870 1871 if (sbi->gc_mode == GC_URGENT_HIGH || 1872 !f2fs_available_free_memory(sbi, DISCARD_CACHE)) 1873 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1874 MIN_DISCARD_GRANULARITY); 1875 else 1876 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1877 dcc->discard_granularity); 1878 1879 if (dcc->discard_wake) 1880 dcc->discard_wake = false; 1881 1882 /* clean up pending candidates before going to sleep */ 1883 if (atomic_read(&dcc->queued_discard)) 1884 __wait_all_discard_cmd(sbi, NULL); 1885 1886 if (try_to_freeze()) 1887 continue; 1888 if (f2fs_readonly(sbi->sb)) 1889 continue; 1890 if (kthread_should_stop()) 1891 return 0; 1892 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) || 1893 !atomic_read(&dcc->discard_cmd_cnt)) { 1894 wait_ms = dpolicy.max_interval; 1895 continue; 1896 } 1897 1898 sb_start_intwrite(sbi->sb); 1899 1900 issued = __issue_discard_cmd(sbi, &dpolicy); 1901 if (issued > 0) { 1902 __wait_all_discard_cmd(sbi, &dpolicy); 1903 wait_ms = dpolicy.min_interval; 1904 } else if (issued == -1) { 1905 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1906 if (!wait_ms) 1907 wait_ms = dpolicy.mid_interval; 1908 } else { 1909 wait_ms = dpolicy.max_interval; 1910 } 1911 if (!atomic_read(&dcc->discard_cmd_cnt)) 1912 wait_ms = dpolicy.max_interval; 1913 1914 sb_end_intwrite(sbi->sb); 1915 1916 } while (!kthread_should_stop()); 1917 return 0; 1918 } 1919 1920 #ifdef CONFIG_BLK_DEV_ZONED 1921 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1922 struct block_device *bdev, block_t blkstart, block_t blklen) 1923 { 1924 sector_t sector, nr_sects; 1925 block_t lblkstart = blkstart; 1926 int devi = 0; 1927 u64 remainder = 0; 1928 1929 if (f2fs_is_multi_device(sbi)) { 1930 devi = f2fs_target_device_index(sbi, blkstart); 1931 if (blkstart < FDEV(devi).start_blk || 1932 blkstart > FDEV(devi).end_blk) { 1933 f2fs_err(sbi, "Invalid block %x", blkstart); 1934 return -EIO; 1935 } 1936 blkstart -= FDEV(devi).start_blk; 1937 } 1938 1939 /* For sequential zones, reset the zone write pointer */ 1940 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1941 sector = SECTOR_FROM_BLOCK(blkstart); 1942 nr_sects = SECTOR_FROM_BLOCK(blklen); 1943 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder); 1944 1945 if (remainder || nr_sects != bdev_zone_sectors(bdev)) { 1946 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1947 devi, sbi->s_ndevs ? FDEV(devi).path : "", 1948 blkstart, blklen); 1949 return -EIO; 1950 } 1951 1952 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) { 1953 trace_f2fs_issue_reset_zone(bdev, blkstart); 1954 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 1955 sector, nr_sects, GFP_NOFS); 1956 } 1957 1958 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen); 1959 return 0; 1960 } 1961 1962 /* For conventional zones, use regular discard if supported */ 1963 __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1964 return 0; 1965 } 1966 #endif 1967 1968 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1969 struct block_device *bdev, block_t blkstart, block_t blklen) 1970 { 1971 #ifdef CONFIG_BLK_DEV_ZONED 1972 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 1973 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1974 #endif 1975 __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1976 return 0; 1977 } 1978 1979 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1980 block_t blkstart, block_t blklen) 1981 { 1982 sector_t start = blkstart, len = 0; 1983 struct block_device *bdev; 1984 struct seg_entry *se; 1985 unsigned int offset; 1986 block_t i; 1987 int err = 0; 1988 1989 bdev = f2fs_target_device(sbi, blkstart, NULL); 1990 1991 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1992 if (i != start) { 1993 struct block_device *bdev2 = 1994 f2fs_target_device(sbi, i, NULL); 1995 1996 if (bdev2 != bdev) { 1997 err = __issue_discard_async(sbi, bdev, 1998 start, len); 1999 if (err) 2000 return err; 2001 bdev = bdev2; 2002 start = i; 2003 len = 0; 2004 } 2005 } 2006 2007 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 2008 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 2009 2010 if (f2fs_block_unit_discard(sbi) && 2011 !f2fs_test_and_set_bit(offset, se->discard_map)) 2012 sbi->discard_blks--; 2013 } 2014 2015 if (len) 2016 err = __issue_discard_async(sbi, bdev, start, len); 2017 return err; 2018 } 2019 2020 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 2021 bool check_only) 2022 { 2023 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2024 int max_blocks = sbi->blocks_per_seg; 2025 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 2026 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2027 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2028 unsigned long *discard_map = (unsigned long *)se->discard_map; 2029 unsigned long *dmap = SIT_I(sbi)->tmp_map; 2030 unsigned int start = 0, end = -1; 2031 bool force = (cpc->reason & CP_DISCARD); 2032 struct discard_entry *de = NULL; 2033 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 2034 int i; 2035 2036 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) || 2037 !f2fs_block_unit_discard(sbi)) 2038 return false; 2039 2040 if (!force) { 2041 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks || 2042 SM_I(sbi)->dcc_info->nr_discards >= 2043 SM_I(sbi)->dcc_info->max_discards) 2044 return false; 2045 } 2046 2047 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 2048 for (i = 0; i < entries; i++) 2049 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 2050 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 2051 2052 while (force || SM_I(sbi)->dcc_info->nr_discards <= 2053 SM_I(sbi)->dcc_info->max_discards) { 2054 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 2055 if (start >= max_blocks) 2056 break; 2057 2058 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 2059 if (force && start && end != max_blocks 2060 && (end - start) < cpc->trim_minlen) 2061 continue; 2062 2063 if (check_only) 2064 return true; 2065 2066 if (!de) { 2067 de = f2fs_kmem_cache_alloc(discard_entry_slab, 2068 GFP_F2FS_ZERO, true, NULL); 2069 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 2070 list_add_tail(&de->list, head); 2071 } 2072 2073 for (i = start; i < end; i++) 2074 __set_bit_le(i, (void *)de->discard_map); 2075 2076 SM_I(sbi)->dcc_info->nr_discards += end - start; 2077 } 2078 return false; 2079 } 2080 2081 static void release_discard_addr(struct discard_entry *entry) 2082 { 2083 list_del(&entry->list); 2084 kmem_cache_free(discard_entry_slab, entry); 2085 } 2086 2087 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 2088 { 2089 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 2090 struct discard_entry *entry, *this; 2091 2092 /* drop caches */ 2093 list_for_each_entry_safe(entry, this, head, list) 2094 release_discard_addr(entry); 2095 } 2096 2097 /* 2098 * Should call f2fs_clear_prefree_segments after checkpoint is done. 2099 */ 2100 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 2101 { 2102 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2103 unsigned int segno; 2104 2105 mutex_lock(&dirty_i->seglist_lock); 2106 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 2107 __set_test_and_free(sbi, segno, false); 2108 mutex_unlock(&dirty_i->seglist_lock); 2109 } 2110 2111 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 2112 struct cp_control *cpc) 2113 { 2114 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2115 struct list_head *head = &dcc->entry_list; 2116 struct discard_entry *entry, *this; 2117 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2118 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 2119 unsigned int start = 0, end = -1; 2120 unsigned int secno, start_segno; 2121 bool force = (cpc->reason & CP_DISCARD); 2122 bool section_alignment = F2FS_OPTION(sbi).discard_unit == 2123 DISCARD_UNIT_SECTION; 2124 2125 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi)) 2126 section_alignment = true; 2127 2128 mutex_lock(&dirty_i->seglist_lock); 2129 2130 while (1) { 2131 int i; 2132 2133 if (section_alignment && end != -1) 2134 end--; 2135 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 2136 if (start >= MAIN_SEGS(sbi)) 2137 break; 2138 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 2139 start + 1); 2140 2141 if (section_alignment) { 2142 start = rounddown(start, sbi->segs_per_sec); 2143 end = roundup(end, sbi->segs_per_sec); 2144 } 2145 2146 for (i = start; i < end; i++) { 2147 if (test_and_clear_bit(i, prefree_map)) 2148 dirty_i->nr_dirty[PRE]--; 2149 } 2150 2151 if (!f2fs_realtime_discard_enable(sbi)) 2152 continue; 2153 2154 if (force && start >= cpc->trim_start && 2155 (end - 1) <= cpc->trim_end) 2156 continue; 2157 2158 /* Should cover 2MB zoned device for zone-based reset */ 2159 if (!f2fs_sb_has_blkzoned(sbi) && 2160 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) { 2161 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 2162 (end - start) << sbi->log_blocks_per_seg); 2163 continue; 2164 } 2165 next: 2166 secno = GET_SEC_FROM_SEG(sbi, start); 2167 start_segno = GET_SEG_FROM_SEC(sbi, secno); 2168 if (!IS_CURSEC(sbi, secno) && 2169 !get_valid_blocks(sbi, start, true)) 2170 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 2171 sbi->segs_per_sec << sbi->log_blocks_per_seg); 2172 2173 start = start_segno + sbi->segs_per_sec; 2174 if (start < end) 2175 goto next; 2176 else 2177 end = start - 1; 2178 } 2179 mutex_unlock(&dirty_i->seglist_lock); 2180 2181 if (!f2fs_block_unit_discard(sbi)) 2182 goto wakeup; 2183 2184 /* send small discards */ 2185 list_for_each_entry_safe(entry, this, head, list) { 2186 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2187 bool is_valid = test_bit_le(0, entry->discard_map); 2188 2189 find_next: 2190 if (is_valid) { 2191 next_pos = find_next_zero_bit_le(entry->discard_map, 2192 sbi->blocks_per_seg, cur_pos); 2193 len = next_pos - cur_pos; 2194 2195 if (f2fs_sb_has_blkzoned(sbi) || 2196 !force || len < cpc->trim_minlen) 2197 goto skip; 2198 2199 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2200 len); 2201 total_len += len; 2202 } else { 2203 next_pos = find_next_bit_le(entry->discard_map, 2204 sbi->blocks_per_seg, cur_pos); 2205 } 2206 skip: 2207 cur_pos = next_pos; 2208 is_valid = !is_valid; 2209 2210 if (cur_pos < sbi->blocks_per_seg) 2211 goto find_next; 2212 2213 release_discard_addr(entry); 2214 dcc->nr_discards -= total_len; 2215 } 2216 2217 wakeup: 2218 wake_up_discard_thread(sbi, false); 2219 } 2220 2221 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi) 2222 { 2223 dev_t dev = sbi->sb->s_bdev->bd_dev; 2224 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2225 int err = 0; 2226 2227 if (!f2fs_realtime_discard_enable(sbi)) 2228 return 0; 2229 2230 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2231 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2232 if (IS_ERR(dcc->f2fs_issue_discard)) { 2233 err = PTR_ERR(dcc->f2fs_issue_discard); 2234 dcc->f2fs_issue_discard = NULL; 2235 } 2236 2237 return err; 2238 } 2239 2240 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2241 { 2242 struct discard_cmd_control *dcc; 2243 int err = 0, i; 2244 2245 if (SM_I(sbi)->dcc_info) { 2246 dcc = SM_I(sbi)->dcc_info; 2247 goto init_thread; 2248 } 2249 2250 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2251 if (!dcc) 2252 return -ENOMEM; 2253 2254 dcc->discard_io_aware_gran = MAX_PLIST_NUM; 2255 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2256 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY; 2257 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT) 2258 dcc->discard_granularity = sbi->blocks_per_seg; 2259 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION) 2260 dcc->discard_granularity = BLKS_PER_SEC(sbi); 2261 2262 INIT_LIST_HEAD(&dcc->entry_list); 2263 for (i = 0; i < MAX_PLIST_NUM; i++) 2264 INIT_LIST_HEAD(&dcc->pend_list[i]); 2265 INIT_LIST_HEAD(&dcc->wait_list); 2266 INIT_LIST_HEAD(&dcc->fstrim_list); 2267 mutex_init(&dcc->cmd_lock); 2268 atomic_set(&dcc->issued_discard, 0); 2269 atomic_set(&dcc->queued_discard, 0); 2270 atomic_set(&dcc->discard_cmd_cnt, 0); 2271 dcc->nr_discards = 0; 2272 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 2273 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST; 2274 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME; 2275 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME; 2276 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME; 2277 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL; 2278 dcc->undiscard_blks = 0; 2279 dcc->next_pos = 0; 2280 dcc->root = RB_ROOT_CACHED; 2281 dcc->rbtree_check = false; 2282 2283 init_waitqueue_head(&dcc->discard_wait_queue); 2284 SM_I(sbi)->dcc_info = dcc; 2285 init_thread: 2286 err = f2fs_start_discard_thread(sbi); 2287 if (err) { 2288 kfree(dcc); 2289 SM_I(sbi)->dcc_info = NULL; 2290 } 2291 2292 return err; 2293 } 2294 2295 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2296 { 2297 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2298 2299 if (!dcc) 2300 return; 2301 2302 f2fs_stop_discard_thread(sbi); 2303 2304 /* 2305 * Recovery can cache discard commands, so in error path of 2306 * fill_super(), it needs to give a chance to handle them. 2307 */ 2308 f2fs_issue_discard_timeout(sbi); 2309 2310 kfree(dcc); 2311 SM_I(sbi)->dcc_info = NULL; 2312 } 2313 2314 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2315 { 2316 struct sit_info *sit_i = SIT_I(sbi); 2317 2318 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2319 sit_i->dirty_sentries++; 2320 return false; 2321 } 2322 2323 return true; 2324 } 2325 2326 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2327 unsigned int segno, int modified) 2328 { 2329 struct seg_entry *se = get_seg_entry(sbi, segno); 2330 2331 se->type = type; 2332 if (modified) 2333 __mark_sit_entry_dirty(sbi, segno); 2334 } 2335 2336 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi, 2337 block_t blkaddr) 2338 { 2339 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2340 2341 if (segno == NULL_SEGNO) 2342 return 0; 2343 return get_seg_entry(sbi, segno)->mtime; 2344 } 2345 2346 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr, 2347 unsigned long long old_mtime) 2348 { 2349 struct seg_entry *se; 2350 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2351 unsigned long long ctime = get_mtime(sbi, false); 2352 unsigned long long mtime = old_mtime ? old_mtime : ctime; 2353 2354 if (segno == NULL_SEGNO) 2355 return; 2356 2357 se = get_seg_entry(sbi, segno); 2358 2359 if (!se->mtime) 2360 se->mtime = mtime; 2361 else 2362 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime, 2363 se->valid_blocks + 1); 2364 2365 if (ctime > SIT_I(sbi)->max_mtime) 2366 SIT_I(sbi)->max_mtime = ctime; 2367 } 2368 2369 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2370 { 2371 struct seg_entry *se; 2372 unsigned int segno, offset; 2373 long int new_vblocks; 2374 bool exist; 2375 #ifdef CONFIG_F2FS_CHECK_FS 2376 bool mir_exist; 2377 #endif 2378 2379 segno = GET_SEGNO(sbi, blkaddr); 2380 2381 se = get_seg_entry(sbi, segno); 2382 new_vblocks = se->valid_blocks + del; 2383 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2384 2385 f2fs_bug_on(sbi, (new_vblocks < 0 || 2386 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno)))); 2387 2388 se->valid_blocks = new_vblocks; 2389 2390 /* Update valid block bitmap */ 2391 if (del > 0) { 2392 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2393 #ifdef CONFIG_F2FS_CHECK_FS 2394 mir_exist = f2fs_test_and_set_bit(offset, 2395 se->cur_valid_map_mir); 2396 if (unlikely(exist != mir_exist)) { 2397 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2398 blkaddr, exist); 2399 f2fs_bug_on(sbi, 1); 2400 } 2401 #endif 2402 if (unlikely(exist)) { 2403 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", 2404 blkaddr); 2405 f2fs_bug_on(sbi, 1); 2406 se->valid_blocks--; 2407 del = 0; 2408 } 2409 2410 if (f2fs_block_unit_discard(sbi) && 2411 !f2fs_test_and_set_bit(offset, se->discard_map)) 2412 sbi->discard_blks--; 2413 2414 /* 2415 * SSR should never reuse block which is checkpointed 2416 * or newly invalidated. 2417 */ 2418 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2419 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 2420 se->ckpt_valid_blocks++; 2421 } 2422 } else { 2423 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 2424 #ifdef CONFIG_F2FS_CHECK_FS 2425 mir_exist = f2fs_test_and_clear_bit(offset, 2426 se->cur_valid_map_mir); 2427 if (unlikely(exist != mir_exist)) { 2428 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2429 blkaddr, exist); 2430 f2fs_bug_on(sbi, 1); 2431 } 2432 #endif 2433 if (unlikely(!exist)) { 2434 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", 2435 blkaddr); 2436 f2fs_bug_on(sbi, 1); 2437 se->valid_blocks++; 2438 del = 0; 2439 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2440 /* 2441 * If checkpoints are off, we must not reuse data that 2442 * was used in the previous checkpoint. If it was used 2443 * before, we must track that to know how much space we 2444 * really have. 2445 */ 2446 if (f2fs_test_bit(offset, se->ckpt_valid_map)) { 2447 spin_lock(&sbi->stat_lock); 2448 sbi->unusable_block_count++; 2449 spin_unlock(&sbi->stat_lock); 2450 } 2451 } 2452 2453 if (f2fs_block_unit_discard(sbi) && 2454 f2fs_test_and_clear_bit(offset, se->discard_map)) 2455 sbi->discard_blks++; 2456 } 2457 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 2458 se->ckpt_valid_blocks += del; 2459 2460 __mark_sit_entry_dirty(sbi, segno); 2461 2462 /* update total number of valid blocks to be written in ckpt area */ 2463 SIT_I(sbi)->written_valid_blocks += del; 2464 2465 if (__is_large_section(sbi)) 2466 get_sec_entry(sbi, segno)->valid_blocks += del; 2467 } 2468 2469 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 2470 { 2471 unsigned int segno = GET_SEGNO(sbi, addr); 2472 struct sit_info *sit_i = SIT_I(sbi); 2473 2474 f2fs_bug_on(sbi, addr == NULL_ADDR); 2475 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2476 return; 2477 2478 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr); 2479 f2fs_invalidate_compress_page(sbi, addr); 2480 2481 /* add it into sit main buffer */ 2482 down_write(&sit_i->sentry_lock); 2483 2484 update_segment_mtime(sbi, addr, 0); 2485 update_sit_entry(sbi, addr, -1); 2486 2487 /* add it into dirty seglist */ 2488 locate_dirty_segment(sbi, segno); 2489 2490 up_write(&sit_i->sentry_lock); 2491 } 2492 2493 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2494 { 2495 struct sit_info *sit_i = SIT_I(sbi); 2496 unsigned int segno, offset; 2497 struct seg_entry *se; 2498 bool is_cp = false; 2499 2500 if (!__is_valid_data_blkaddr(blkaddr)) 2501 return true; 2502 2503 down_read(&sit_i->sentry_lock); 2504 2505 segno = GET_SEGNO(sbi, blkaddr); 2506 se = get_seg_entry(sbi, segno); 2507 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2508 2509 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2510 is_cp = true; 2511 2512 up_read(&sit_i->sentry_lock); 2513 2514 return is_cp; 2515 } 2516 2517 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type) 2518 { 2519 struct curseg_info *curseg = CURSEG_I(sbi, type); 2520 2521 if (sbi->ckpt->alloc_type[type] == SSR) 2522 return sbi->blocks_per_seg; 2523 return curseg->next_blkoff; 2524 } 2525 2526 /* 2527 * Calculate the number of current summary pages for writing 2528 */ 2529 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2530 { 2531 int valid_sum_count = 0; 2532 int i, sum_in_page; 2533 2534 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2535 if (sbi->ckpt->alloc_type[i] != SSR && for_ra) 2536 valid_sum_count += 2537 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2538 else 2539 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i); 2540 } 2541 2542 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 2543 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2544 if (valid_sum_count <= sum_in_page) 2545 return 1; 2546 else if ((valid_sum_count - sum_in_page) <= 2547 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2548 return 2; 2549 return 3; 2550 } 2551 2552 /* 2553 * Caller should put this summary page 2554 */ 2555 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 2556 { 2557 if (unlikely(f2fs_cp_error(sbi))) 2558 return ERR_PTR(-EIO); 2559 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno)); 2560 } 2561 2562 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2563 void *src, block_t blk_addr) 2564 { 2565 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2566 2567 memcpy(page_address(page), src, PAGE_SIZE); 2568 set_page_dirty(page); 2569 f2fs_put_page(page, 1); 2570 } 2571 2572 static void write_sum_page(struct f2fs_sb_info *sbi, 2573 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2574 { 2575 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2576 } 2577 2578 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2579 int type, block_t blk_addr) 2580 { 2581 struct curseg_info *curseg = CURSEG_I(sbi, type); 2582 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2583 struct f2fs_summary_block *src = curseg->sum_blk; 2584 struct f2fs_summary_block *dst; 2585 2586 dst = (struct f2fs_summary_block *)page_address(page); 2587 memset(dst, 0, PAGE_SIZE); 2588 2589 mutex_lock(&curseg->curseg_mutex); 2590 2591 down_read(&curseg->journal_rwsem); 2592 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2593 up_read(&curseg->journal_rwsem); 2594 2595 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2596 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2597 2598 mutex_unlock(&curseg->curseg_mutex); 2599 2600 set_page_dirty(page); 2601 f2fs_put_page(page, 1); 2602 } 2603 2604 static int is_next_segment_free(struct f2fs_sb_info *sbi, 2605 struct curseg_info *curseg, int type) 2606 { 2607 unsigned int segno = curseg->segno + 1; 2608 struct free_segmap_info *free_i = FREE_I(sbi); 2609 2610 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 2611 return !test_bit(segno, free_i->free_segmap); 2612 return 0; 2613 } 2614 2615 /* 2616 * Find a new segment from the free segments bitmap to right order 2617 * This function should be returned with success, otherwise BUG 2618 */ 2619 static void get_new_segment(struct f2fs_sb_info *sbi, 2620 unsigned int *newseg, bool new_sec, int dir) 2621 { 2622 struct free_segmap_info *free_i = FREE_I(sbi); 2623 unsigned int segno, secno, zoneno; 2624 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2625 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2626 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2627 unsigned int left_start = hint; 2628 bool init = true; 2629 int go_left = 0; 2630 int i; 2631 2632 spin_lock(&free_i->segmap_lock); 2633 2634 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 2635 segno = find_next_zero_bit(free_i->free_segmap, 2636 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2637 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2638 goto got_it; 2639 } 2640 find_other_zone: 2641 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2642 if (secno >= MAIN_SECS(sbi)) { 2643 if (dir == ALLOC_RIGHT) { 2644 secno = find_first_zero_bit(free_i->free_secmap, 2645 MAIN_SECS(sbi)); 2646 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 2647 } else { 2648 go_left = 1; 2649 left_start = hint - 1; 2650 } 2651 } 2652 if (go_left == 0) 2653 goto skip_left; 2654 2655 while (test_bit(left_start, free_i->free_secmap)) { 2656 if (left_start > 0) { 2657 left_start--; 2658 continue; 2659 } 2660 left_start = find_first_zero_bit(free_i->free_secmap, 2661 MAIN_SECS(sbi)); 2662 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 2663 break; 2664 } 2665 secno = left_start; 2666 skip_left: 2667 segno = GET_SEG_FROM_SEC(sbi, secno); 2668 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2669 2670 /* give up on finding another zone */ 2671 if (!init) 2672 goto got_it; 2673 if (sbi->secs_per_zone == 1) 2674 goto got_it; 2675 if (zoneno == old_zoneno) 2676 goto got_it; 2677 if (dir == ALLOC_LEFT) { 2678 if (!go_left && zoneno + 1 >= total_zones) 2679 goto got_it; 2680 if (go_left && zoneno == 0) 2681 goto got_it; 2682 } 2683 for (i = 0; i < NR_CURSEG_TYPE; i++) 2684 if (CURSEG_I(sbi, i)->zone == zoneno) 2685 break; 2686 2687 if (i < NR_CURSEG_TYPE) { 2688 /* zone is in user, try another */ 2689 if (go_left) 2690 hint = zoneno * sbi->secs_per_zone - 1; 2691 else if (zoneno + 1 >= total_zones) 2692 hint = 0; 2693 else 2694 hint = (zoneno + 1) * sbi->secs_per_zone; 2695 init = false; 2696 goto find_other_zone; 2697 } 2698 got_it: 2699 /* set it as dirty segment in free segmap */ 2700 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2701 __set_inuse(sbi, segno); 2702 *newseg = segno; 2703 spin_unlock(&free_i->segmap_lock); 2704 } 2705 2706 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2707 { 2708 struct curseg_info *curseg = CURSEG_I(sbi, type); 2709 struct summary_footer *sum_footer; 2710 unsigned short seg_type = curseg->seg_type; 2711 2712 curseg->inited = true; 2713 curseg->segno = curseg->next_segno; 2714 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2715 curseg->next_blkoff = 0; 2716 curseg->next_segno = NULL_SEGNO; 2717 2718 sum_footer = &(curseg->sum_blk->footer); 2719 memset(sum_footer, 0, sizeof(struct summary_footer)); 2720 2721 sanity_check_seg_type(sbi, seg_type); 2722 2723 if (IS_DATASEG(seg_type)) 2724 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2725 if (IS_NODESEG(seg_type)) 2726 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2727 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified); 2728 } 2729 2730 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2731 { 2732 struct curseg_info *curseg = CURSEG_I(sbi, type); 2733 unsigned short seg_type = curseg->seg_type; 2734 2735 sanity_check_seg_type(sbi, seg_type); 2736 if (f2fs_need_rand_seg(sbi)) 2737 return get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec); 2738 2739 /* if segs_per_sec is large than 1, we need to keep original policy. */ 2740 if (__is_large_section(sbi)) 2741 return curseg->segno; 2742 2743 /* inmem log may not locate on any segment after mount */ 2744 if (!curseg->inited) 2745 return 0; 2746 2747 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2748 return 0; 2749 2750 if (test_opt(sbi, NOHEAP) && 2751 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))) 2752 return 0; 2753 2754 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2755 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2756 2757 /* find segments from 0 to reuse freed segments */ 2758 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2759 return 0; 2760 2761 return curseg->segno; 2762 } 2763 2764 /* 2765 * Allocate a current working segment. 2766 * This function always allocates a free segment in LFS manner. 2767 */ 2768 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2769 { 2770 struct curseg_info *curseg = CURSEG_I(sbi, type); 2771 unsigned short seg_type = curseg->seg_type; 2772 unsigned int segno = curseg->segno; 2773 int dir = ALLOC_LEFT; 2774 2775 if (curseg->inited) 2776 write_sum_page(sbi, curseg->sum_blk, 2777 GET_SUM_BLOCK(sbi, segno)); 2778 if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA) 2779 dir = ALLOC_RIGHT; 2780 2781 if (test_opt(sbi, NOHEAP)) 2782 dir = ALLOC_RIGHT; 2783 2784 segno = __get_next_segno(sbi, type); 2785 get_new_segment(sbi, &segno, new_sec, dir); 2786 curseg->next_segno = segno; 2787 reset_curseg(sbi, type, 1); 2788 curseg->alloc_type = LFS; 2789 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 2790 curseg->fragment_remained_chunk = 2791 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 2792 } 2793 2794 static int __next_free_blkoff(struct f2fs_sb_info *sbi, 2795 int segno, block_t start) 2796 { 2797 struct seg_entry *se = get_seg_entry(sbi, segno); 2798 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2799 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2800 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2801 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2802 int i; 2803 2804 for (i = 0; i < entries; i++) 2805 target_map[i] = ckpt_map[i] | cur_map[i]; 2806 2807 return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 2808 } 2809 2810 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi, 2811 struct curseg_info *seg) 2812 { 2813 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1); 2814 } 2815 2816 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno) 2817 { 2818 return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg; 2819 } 2820 2821 /* 2822 * This function always allocates a used segment(from dirty seglist) by SSR 2823 * manner, so it should recover the existing segment information of valid blocks 2824 */ 2825 static void change_curseg(struct f2fs_sb_info *sbi, int type) 2826 { 2827 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2828 struct curseg_info *curseg = CURSEG_I(sbi, type); 2829 unsigned int new_segno = curseg->next_segno; 2830 struct f2fs_summary_block *sum_node; 2831 struct page *sum_page; 2832 2833 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno)); 2834 2835 __set_test_and_inuse(sbi, new_segno); 2836 2837 mutex_lock(&dirty_i->seglist_lock); 2838 __remove_dirty_segment(sbi, new_segno, PRE); 2839 __remove_dirty_segment(sbi, new_segno, DIRTY); 2840 mutex_unlock(&dirty_i->seglist_lock); 2841 2842 reset_curseg(sbi, type, 1); 2843 curseg->alloc_type = SSR; 2844 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0); 2845 2846 sum_page = f2fs_get_sum_page(sbi, new_segno); 2847 if (IS_ERR(sum_page)) { 2848 /* GC won't be able to use stale summary pages by cp_error */ 2849 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE); 2850 return; 2851 } 2852 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2853 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2854 f2fs_put_page(sum_page, 1); 2855 } 2856 2857 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2858 int alloc_mode, unsigned long long age); 2859 2860 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type, 2861 int target_type, int alloc_mode, 2862 unsigned long long age) 2863 { 2864 struct curseg_info *curseg = CURSEG_I(sbi, type); 2865 2866 curseg->seg_type = target_type; 2867 2868 if (get_ssr_segment(sbi, type, alloc_mode, age)) { 2869 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno); 2870 2871 curseg->seg_type = se->type; 2872 change_curseg(sbi, type); 2873 } else { 2874 /* allocate cold segment by default */ 2875 curseg->seg_type = CURSEG_COLD_DATA; 2876 new_curseg(sbi, type, true); 2877 } 2878 stat_inc_seg_type(sbi, curseg); 2879 } 2880 2881 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi) 2882 { 2883 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC); 2884 2885 if (!sbi->am.atgc_enabled) 2886 return; 2887 2888 f2fs_down_read(&SM_I(sbi)->curseg_lock); 2889 2890 mutex_lock(&curseg->curseg_mutex); 2891 down_write(&SIT_I(sbi)->sentry_lock); 2892 2893 get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0); 2894 2895 up_write(&SIT_I(sbi)->sentry_lock); 2896 mutex_unlock(&curseg->curseg_mutex); 2897 2898 f2fs_up_read(&SM_I(sbi)->curseg_lock); 2899 2900 } 2901 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi) 2902 { 2903 __f2fs_init_atgc_curseg(sbi); 2904 } 2905 2906 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2907 { 2908 struct curseg_info *curseg = CURSEG_I(sbi, type); 2909 2910 mutex_lock(&curseg->curseg_mutex); 2911 if (!curseg->inited) 2912 goto out; 2913 2914 if (get_valid_blocks(sbi, curseg->segno, false)) { 2915 write_sum_page(sbi, curseg->sum_blk, 2916 GET_SUM_BLOCK(sbi, curseg->segno)); 2917 } else { 2918 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2919 __set_test_and_free(sbi, curseg->segno, true); 2920 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2921 } 2922 out: 2923 mutex_unlock(&curseg->curseg_mutex); 2924 } 2925 2926 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi) 2927 { 2928 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2929 2930 if (sbi->am.atgc_enabled) 2931 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2932 } 2933 2934 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2935 { 2936 struct curseg_info *curseg = CURSEG_I(sbi, type); 2937 2938 mutex_lock(&curseg->curseg_mutex); 2939 if (!curseg->inited) 2940 goto out; 2941 if (get_valid_blocks(sbi, curseg->segno, false)) 2942 goto out; 2943 2944 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2945 __set_test_and_inuse(sbi, curseg->segno); 2946 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2947 out: 2948 mutex_unlock(&curseg->curseg_mutex); 2949 } 2950 2951 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi) 2952 { 2953 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2954 2955 if (sbi->am.atgc_enabled) 2956 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2957 } 2958 2959 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2960 int alloc_mode, unsigned long long age) 2961 { 2962 struct curseg_info *curseg = CURSEG_I(sbi, type); 2963 unsigned segno = NULL_SEGNO; 2964 unsigned short seg_type = curseg->seg_type; 2965 int i, cnt; 2966 bool reversed = false; 2967 2968 sanity_check_seg_type(sbi, seg_type); 2969 2970 /* f2fs_need_SSR() already forces to do this */ 2971 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) { 2972 curseg->next_segno = segno; 2973 return 1; 2974 } 2975 2976 /* For node segments, let's do SSR more intensively */ 2977 if (IS_NODESEG(seg_type)) { 2978 if (seg_type >= CURSEG_WARM_NODE) { 2979 reversed = true; 2980 i = CURSEG_COLD_NODE; 2981 } else { 2982 i = CURSEG_HOT_NODE; 2983 } 2984 cnt = NR_CURSEG_NODE_TYPE; 2985 } else { 2986 if (seg_type >= CURSEG_WARM_DATA) { 2987 reversed = true; 2988 i = CURSEG_COLD_DATA; 2989 } else { 2990 i = CURSEG_HOT_DATA; 2991 } 2992 cnt = NR_CURSEG_DATA_TYPE; 2993 } 2994 2995 for (; cnt-- > 0; reversed ? i-- : i++) { 2996 if (i == seg_type) 2997 continue; 2998 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) { 2999 curseg->next_segno = segno; 3000 return 1; 3001 } 3002 } 3003 3004 /* find valid_blocks=0 in dirty list */ 3005 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 3006 segno = get_free_segment(sbi); 3007 if (segno != NULL_SEGNO) { 3008 curseg->next_segno = segno; 3009 return 1; 3010 } 3011 } 3012 return 0; 3013 } 3014 3015 static bool need_new_seg(struct f2fs_sb_info *sbi, int type) 3016 { 3017 struct curseg_info *curseg = CURSEG_I(sbi, type); 3018 3019 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 3020 curseg->seg_type == CURSEG_WARM_NODE) 3021 return true; 3022 if (curseg->alloc_type == LFS && 3023 is_next_segment_free(sbi, curseg, type) && 3024 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 3025 return true; 3026 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0)) 3027 return true; 3028 return false; 3029 } 3030 3031 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 3032 unsigned int start, unsigned int end) 3033 { 3034 struct curseg_info *curseg = CURSEG_I(sbi, type); 3035 unsigned int segno; 3036 3037 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3038 mutex_lock(&curseg->curseg_mutex); 3039 down_write(&SIT_I(sbi)->sentry_lock); 3040 3041 segno = CURSEG_I(sbi, type)->segno; 3042 if (segno < start || segno > end) 3043 goto unlock; 3044 3045 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0)) 3046 change_curseg(sbi, type); 3047 else 3048 new_curseg(sbi, type, true); 3049 3050 stat_inc_seg_type(sbi, curseg); 3051 3052 locate_dirty_segment(sbi, segno); 3053 unlock: 3054 up_write(&SIT_I(sbi)->sentry_lock); 3055 3056 if (segno != curseg->segno) 3057 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 3058 type, segno, curseg->segno); 3059 3060 mutex_unlock(&curseg->curseg_mutex); 3061 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3062 } 3063 3064 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type, 3065 bool new_sec, bool force) 3066 { 3067 struct curseg_info *curseg = CURSEG_I(sbi, type); 3068 unsigned int old_segno; 3069 3070 if (!force && curseg->inited && 3071 !curseg->next_blkoff && 3072 !get_valid_blocks(sbi, curseg->segno, new_sec) && 3073 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec)) 3074 return; 3075 3076 old_segno = curseg->segno; 3077 new_curseg(sbi, type, true); 3078 stat_inc_seg_type(sbi, curseg); 3079 locate_dirty_segment(sbi, old_segno); 3080 } 3081 3082 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force) 3083 { 3084 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3085 down_write(&SIT_I(sbi)->sentry_lock); 3086 __allocate_new_segment(sbi, type, true, force); 3087 up_write(&SIT_I(sbi)->sentry_lock); 3088 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3089 } 3090 3091 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 3092 { 3093 int i; 3094 3095 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3096 down_write(&SIT_I(sbi)->sentry_lock); 3097 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 3098 __allocate_new_segment(sbi, i, false, false); 3099 up_write(&SIT_I(sbi)->sentry_lock); 3100 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3101 } 3102 3103 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 3104 struct cp_control *cpc) 3105 { 3106 __u64 trim_start = cpc->trim_start; 3107 bool has_candidate = false; 3108 3109 down_write(&SIT_I(sbi)->sentry_lock); 3110 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 3111 if (add_discard_addrs(sbi, cpc, true)) { 3112 has_candidate = true; 3113 break; 3114 } 3115 } 3116 up_write(&SIT_I(sbi)->sentry_lock); 3117 3118 cpc->trim_start = trim_start; 3119 return has_candidate; 3120 } 3121 3122 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 3123 struct discard_policy *dpolicy, 3124 unsigned int start, unsigned int end) 3125 { 3126 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 3127 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 3128 struct rb_node **insert_p = NULL, *insert_parent = NULL; 3129 struct discard_cmd *dc; 3130 struct blk_plug plug; 3131 int issued; 3132 unsigned int trimmed = 0; 3133 3134 next: 3135 issued = 0; 3136 3137 mutex_lock(&dcc->cmd_lock); 3138 if (unlikely(dcc->rbtree_check)) 3139 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 3140 3141 dc = __lookup_discard_cmd_ret(&dcc->root, start, 3142 &prev_dc, &next_dc, &insert_p, &insert_parent); 3143 if (!dc) 3144 dc = next_dc; 3145 3146 blk_start_plug(&plug); 3147 3148 while (dc && dc->di.lstart <= end) { 3149 struct rb_node *node; 3150 int err = 0; 3151 3152 if (dc->di.len < dpolicy->granularity) 3153 goto skip; 3154 3155 if (dc->state != D_PREP) { 3156 list_move_tail(&dc->list, &dcc->fstrim_list); 3157 goto skip; 3158 } 3159 3160 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 3161 3162 if (issued >= dpolicy->max_requests) { 3163 start = dc->di.lstart + dc->di.len; 3164 3165 if (err) 3166 __remove_discard_cmd(sbi, dc); 3167 3168 blk_finish_plug(&plug); 3169 mutex_unlock(&dcc->cmd_lock); 3170 trimmed += __wait_all_discard_cmd(sbi, NULL); 3171 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 3172 goto next; 3173 } 3174 skip: 3175 node = rb_next(&dc->rb_node); 3176 if (err) 3177 __remove_discard_cmd(sbi, dc); 3178 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 3179 3180 if (fatal_signal_pending(current)) 3181 break; 3182 } 3183 3184 blk_finish_plug(&plug); 3185 mutex_unlock(&dcc->cmd_lock); 3186 3187 return trimmed; 3188 } 3189 3190 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 3191 { 3192 __u64 start = F2FS_BYTES_TO_BLK(range->start); 3193 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 3194 unsigned int start_segno, end_segno; 3195 block_t start_block, end_block; 3196 struct cp_control cpc; 3197 struct discard_policy dpolicy; 3198 unsigned long long trimmed = 0; 3199 int err = 0; 3200 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 3201 3202 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 3203 return -EINVAL; 3204 3205 if (end < MAIN_BLKADDR(sbi)) 3206 goto out; 3207 3208 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 3209 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 3210 return -EFSCORRUPTED; 3211 } 3212 3213 /* start/end segment number in main_area */ 3214 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 3215 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 3216 GET_SEGNO(sbi, end); 3217 if (need_align) { 3218 start_segno = rounddown(start_segno, sbi->segs_per_sec); 3219 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1; 3220 } 3221 3222 cpc.reason = CP_DISCARD; 3223 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 3224 cpc.trim_start = start_segno; 3225 cpc.trim_end = end_segno; 3226 3227 if (sbi->discard_blks == 0) 3228 goto out; 3229 3230 f2fs_down_write(&sbi->gc_lock); 3231 err = f2fs_write_checkpoint(sbi, &cpc); 3232 f2fs_up_write(&sbi->gc_lock); 3233 if (err) 3234 goto out; 3235 3236 /* 3237 * We filed discard candidates, but actually we don't need to wait for 3238 * all of them, since they'll be issued in idle time along with runtime 3239 * discard option. User configuration looks like using runtime discard 3240 * or periodic fstrim instead of it. 3241 */ 3242 if (f2fs_realtime_discard_enable(sbi)) 3243 goto out; 3244 3245 start_block = START_BLOCK(sbi, start_segno); 3246 end_block = START_BLOCK(sbi, end_segno + 1); 3247 3248 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 3249 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 3250 start_block, end_block); 3251 3252 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 3253 start_block, end_block); 3254 out: 3255 if (!err) 3256 range->len = F2FS_BLK_TO_BYTES(trimmed); 3257 return err; 3258 } 3259 3260 int f2fs_rw_hint_to_seg_type(enum rw_hint hint) 3261 { 3262 switch (hint) { 3263 case WRITE_LIFE_SHORT: 3264 return CURSEG_HOT_DATA; 3265 case WRITE_LIFE_EXTREME: 3266 return CURSEG_COLD_DATA; 3267 default: 3268 return CURSEG_WARM_DATA; 3269 } 3270 } 3271 3272 static int __get_segment_type_2(struct f2fs_io_info *fio) 3273 { 3274 if (fio->type == DATA) 3275 return CURSEG_HOT_DATA; 3276 else 3277 return CURSEG_HOT_NODE; 3278 } 3279 3280 static int __get_segment_type_4(struct f2fs_io_info *fio) 3281 { 3282 if (fio->type == DATA) { 3283 struct inode *inode = fio->page->mapping->host; 3284 3285 if (S_ISDIR(inode->i_mode)) 3286 return CURSEG_HOT_DATA; 3287 else 3288 return CURSEG_COLD_DATA; 3289 } else { 3290 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3291 return CURSEG_WARM_NODE; 3292 else 3293 return CURSEG_COLD_NODE; 3294 } 3295 } 3296 3297 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs) 3298 { 3299 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3300 struct extent_info ei = {}; 3301 3302 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) { 3303 if (!ei.age) 3304 return NO_CHECK_TYPE; 3305 if (ei.age <= sbi->hot_data_age_threshold) 3306 return CURSEG_HOT_DATA; 3307 if (ei.age <= sbi->warm_data_age_threshold) 3308 return CURSEG_WARM_DATA; 3309 return CURSEG_COLD_DATA; 3310 } 3311 return NO_CHECK_TYPE; 3312 } 3313 3314 static int __get_segment_type_6(struct f2fs_io_info *fio) 3315 { 3316 if (fio->type == DATA) { 3317 struct inode *inode = fio->page->mapping->host; 3318 int type; 3319 3320 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 3321 return CURSEG_COLD_DATA_PINNED; 3322 3323 if (page_private_gcing(fio->page)) { 3324 if (fio->sbi->am.atgc_enabled && 3325 (fio->io_type == FS_DATA_IO) && 3326 (fio->sbi->gc_mode != GC_URGENT_HIGH)) 3327 return CURSEG_ALL_DATA_ATGC; 3328 else 3329 return CURSEG_COLD_DATA; 3330 } 3331 if (file_is_cold(inode) || f2fs_need_compress_data(inode)) 3332 return CURSEG_COLD_DATA; 3333 3334 type = __get_age_segment_type(inode, fio->page->index); 3335 if (type != NO_CHECK_TYPE) 3336 return type; 3337 3338 if (file_is_hot(inode) || 3339 is_inode_flag_set(inode, FI_HOT_DATA) || 3340 f2fs_is_cow_file(inode)) 3341 return CURSEG_HOT_DATA; 3342 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 3343 } else { 3344 if (IS_DNODE(fio->page)) 3345 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3346 CURSEG_HOT_NODE; 3347 return CURSEG_COLD_NODE; 3348 } 3349 } 3350 3351 static int __get_segment_type(struct f2fs_io_info *fio) 3352 { 3353 int type = 0; 3354 3355 switch (F2FS_OPTION(fio->sbi).active_logs) { 3356 case 2: 3357 type = __get_segment_type_2(fio); 3358 break; 3359 case 4: 3360 type = __get_segment_type_4(fio); 3361 break; 3362 case 6: 3363 type = __get_segment_type_6(fio); 3364 break; 3365 default: 3366 f2fs_bug_on(fio->sbi, true); 3367 } 3368 3369 if (IS_HOT(type)) 3370 fio->temp = HOT; 3371 else if (IS_WARM(type)) 3372 fio->temp = WARM; 3373 else 3374 fio->temp = COLD; 3375 return type; 3376 } 3377 3378 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi, 3379 struct curseg_info *seg) 3380 { 3381 /* To allocate block chunks in different sizes, use random number */ 3382 if (--seg->fragment_remained_chunk > 0) 3383 return; 3384 3385 seg->fragment_remained_chunk = 3386 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 3387 seg->next_blkoff += 3388 get_random_u32_inclusive(1, sbi->max_fragment_hole); 3389 } 3390 3391 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3392 block_t old_blkaddr, block_t *new_blkaddr, 3393 struct f2fs_summary *sum, int type, 3394 struct f2fs_io_info *fio) 3395 { 3396 struct sit_info *sit_i = SIT_I(sbi); 3397 struct curseg_info *curseg = CURSEG_I(sbi, type); 3398 unsigned long long old_mtime; 3399 bool from_gc = (type == CURSEG_ALL_DATA_ATGC); 3400 struct seg_entry *se = NULL; 3401 bool segment_full = false; 3402 3403 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3404 3405 mutex_lock(&curseg->curseg_mutex); 3406 down_write(&sit_i->sentry_lock); 3407 3408 if (from_gc) { 3409 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO); 3410 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr)); 3411 sanity_check_seg_type(sbi, se->type); 3412 f2fs_bug_on(sbi, IS_NODESEG(se->type)); 3413 } 3414 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3415 3416 f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg); 3417 3418 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3419 3420 curseg->sum_blk->entries[curseg->next_blkoff] = *sum; 3421 if (curseg->alloc_type == SSR) { 3422 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg); 3423 } else { 3424 curseg->next_blkoff++; 3425 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 3426 f2fs_randomize_chunk(sbi, curseg); 3427 } 3428 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno)) 3429 segment_full = true; 3430 stat_inc_block_count(sbi, curseg); 3431 3432 if (from_gc) { 3433 old_mtime = get_segment_mtime(sbi, old_blkaddr); 3434 } else { 3435 update_segment_mtime(sbi, old_blkaddr, 0); 3436 old_mtime = 0; 3437 } 3438 update_segment_mtime(sbi, *new_blkaddr, old_mtime); 3439 3440 /* 3441 * SIT information should be updated before segment allocation, 3442 * since SSR needs latest valid block information. 3443 */ 3444 update_sit_entry(sbi, *new_blkaddr, 1); 3445 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 3446 update_sit_entry(sbi, old_blkaddr, -1); 3447 3448 /* 3449 * If the current segment is full, flush it out and replace it with a 3450 * new segment. 3451 */ 3452 if (segment_full) { 3453 if (from_gc) { 3454 get_atssr_segment(sbi, type, se->type, 3455 AT_SSR, se->mtime); 3456 } else { 3457 if (need_new_seg(sbi, type)) 3458 new_curseg(sbi, type, false); 3459 else 3460 change_curseg(sbi, type); 3461 stat_inc_seg_type(sbi, curseg); 3462 } 3463 } 3464 /* 3465 * segment dirty status should be updated after segment allocation, 3466 * so we just need to update status only one time after previous 3467 * segment being closed. 3468 */ 3469 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3470 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3471 3472 if (IS_DATASEG(type)) 3473 atomic64_inc(&sbi->allocated_data_blocks); 3474 3475 up_write(&sit_i->sentry_lock); 3476 3477 if (page && IS_NODESEG(type)) { 3478 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3479 3480 f2fs_inode_chksum_set(sbi, page); 3481 } 3482 3483 if (fio) { 3484 struct f2fs_bio_info *io; 3485 3486 if (F2FS_IO_ALIGNED(sbi)) 3487 fio->retry = 0; 3488 3489 INIT_LIST_HEAD(&fio->list); 3490 fio->in_list = 1; 3491 io = sbi->write_io[fio->type] + fio->temp; 3492 spin_lock(&io->io_lock); 3493 list_add_tail(&fio->list, &io->io_list); 3494 spin_unlock(&io->io_lock); 3495 } 3496 3497 mutex_unlock(&curseg->curseg_mutex); 3498 3499 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3500 } 3501 3502 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino, 3503 block_t blkaddr, unsigned int blkcnt) 3504 { 3505 if (!f2fs_is_multi_device(sbi)) 3506 return; 3507 3508 while (1) { 3509 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr); 3510 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1; 3511 3512 /* update device state for fsync */ 3513 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO); 3514 3515 /* update device state for checkpoint */ 3516 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3517 spin_lock(&sbi->dev_lock); 3518 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3519 spin_unlock(&sbi->dev_lock); 3520 } 3521 3522 if (blkcnt <= blks) 3523 break; 3524 blkcnt -= blks; 3525 blkaddr += blks; 3526 } 3527 } 3528 3529 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3530 { 3531 int type = __get_segment_type(fio); 3532 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA); 3533 3534 if (keep_order) 3535 f2fs_down_read(&fio->sbi->io_order_lock); 3536 reallocate: 3537 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3538 &fio->new_blkaddr, sum, type, fio); 3539 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) { 3540 invalidate_mapping_pages(META_MAPPING(fio->sbi), 3541 fio->old_blkaddr, fio->old_blkaddr); 3542 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr); 3543 } 3544 3545 /* writeout dirty page into bdev */ 3546 f2fs_submit_page_write(fio); 3547 if (fio->retry) { 3548 fio->old_blkaddr = fio->new_blkaddr; 3549 goto reallocate; 3550 } 3551 3552 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1); 3553 3554 if (keep_order) 3555 f2fs_up_read(&fio->sbi->io_order_lock); 3556 } 3557 3558 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 3559 enum iostat_type io_type) 3560 { 3561 struct f2fs_io_info fio = { 3562 .sbi = sbi, 3563 .type = META, 3564 .temp = HOT, 3565 .op = REQ_OP_WRITE, 3566 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3567 .old_blkaddr = page->index, 3568 .new_blkaddr = page->index, 3569 .page = page, 3570 .encrypted_page = NULL, 3571 .in_list = 0, 3572 }; 3573 3574 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 3575 fio.op_flags &= ~REQ_META; 3576 3577 set_page_writeback(page); 3578 f2fs_submit_page_write(&fio); 3579 3580 stat_inc_meta_count(sbi, page->index); 3581 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE); 3582 } 3583 3584 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3585 { 3586 struct f2fs_summary sum; 3587 3588 set_summary(&sum, nid, 0, 0); 3589 do_write_page(&sum, fio); 3590 3591 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE); 3592 } 3593 3594 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3595 struct f2fs_io_info *fio) 3596 { 3597 struct f2fs_sb_info *sbi = fio->sbi; 3598 struct f2fs_summary sum; 3599 3600 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3601 if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO) 3602 f2fs_update_age_extent_cache(dn); 3603 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3604 do_write_page(&sum, fio); 3605 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3606 3607 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE); 3608 } 3609 3610 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3611 { 3612 int err; 3613 struct f2fs_sb_info *sbi = fio->sbi; 3614 unsigned int segno; 3615 3616 fio->new_blkaddr = fio->old_blkaddr; 3617 /* i/o temperature is needed for passing down write hints */ 3618 __get_segment_type(fio); 3619 3620 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3621 3622 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 3623 set_sbi_flag(sbi, SBI_NEED_FSCK); 3624 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 3625 __func__, segno); 3626 err = -EFSCORRUPTED; 3627 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 3628 goto drop_bio; 3629 } 3630 3631 if (f2fs_cp_error(sbi)) { 3632 err = -EIO; 3633 goto drop_bio; 3634 } 3635 3636 if (fio->post_read) 3637 invalidate_mapping_pages(META_MAPPING(sbi), 3638 fio->new_blkaddr, fio->new_blkaddr); 3639 3640 stat_inc_inplace_blocks(fio->sbi); 3641 3642 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi)) 3643 err = f2fs_merge_page_bio(fio); 3644 else 3645 err = f2fs_submit_page_bio(fio); 3646 if (!err) { 3647 f2fs_update_device_state(fio->sbi, fio->ino, 3648 fio->new_blkaddr, 1); 3649 f2fs_update_iostat(fio->sbi, fio->page->mapping->host, 3650 fio->io_type, F2FS_BLKSIZE); 3651 } 3652 3653 return err; 3654 drop_bio: 3655 if (fio->bio && *(fio->bio)) { 3656 struct bio *bio = *(fio->bio); 3657 3658 bio->bi_status = BLK_STS_IOERR; 3659 bio_endio(bio); 3660 *(fio->bio) = NULL; 3661 } 3662 return err; 3663 } 3664 3665 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 3666 unsigned int segno) 3667 { 3668 int i; 3669 3670 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 3671 if (CURSEG_I(sbi, i)->segno == segno) 3672 break; 3673 } 3674 return i; 3675 } 3676 3677 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 3678 block_t old_blkaddr, block_t new_blkaddr, 3679 bool recover_curseg, bool recover_newaddr, 3680 bool from_gc) 3681 { 3682 struct sit_info *sit_i = SIT_I(sbi); 3683 struct curseg_info *curseg; 3684 unsigned int segno, old_cursegno; 3685 struct seg_entry *se; 3686 int type; 3687 unsigned short old_blkoff; 3688 unsigned char old_alloc_type; 3689 3690 segno = GET_SEGNO(sbi, new_blkaddr); 3691 se = get_seg_entry(sbi, segno); 3692 type = se->type; 3693 3694 f2fs_down_write(&SM_I(sbi)->curseg_lock); 3695 3696 if (!recover_curseg) { 3697 /* for recovery flow */ 3698 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 3699 if (old_blkaddr == NULL_ADDR) 3700 type = CURSEG_COLD_DATA; 3701 else 3702 type = CURSEG_WARM_DATA; 3703 } 3704 } else { 3705 if (IS_CURSEG(sbi, segno)) { 3706 /* se->type is volatile as SSR allocation */ 3707 type = __f2fs_get_curseg(sbi, segno); 3708 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 3709 } else { 3710 type = CURSEG_WARM_DATA; 3711 } 3712 } 3713 3714 f2fs_bug_on(sbi, !IS_DATASEG(type)); 3715 curseg = CURSEG_I(sbi, type); 3716 3717 mutex_lock(&curseg->curseg_mutex); 3718 down_write(&sit_i->sentry_lock); 3719 3720 old_cursegno = curseg->segno; 3721 old_blkoff = curseg->next_blkoff; 3722 old_alloc_type = curseg->alloc_type; 3723 3724 /* change the current segment */ 3725 if (segno != curseg->segno) { 3726 curseg->next_segno = segno; 3727 change_curseg(sbi, type); 3728 } 3729 3730 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 3731 curseg->sum_blk->entries[curseg->next_blkoff] = *sum; 3732 3733 if (!recover_curseg || recover_newaddr) { 3734 if (!from_gc) 3735 update_segment_mtime(sbi, new_blkaddr, 0); 3736 update_sit_entry(sbi, new_blkaddr, 1); 3737 } 3738 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 3739 invalidate_mapping_pages(META_MAPPING(sbi), 3740 old_blkaddr, old_blkaddr); 3741 f2fs_invalidate_compress_page(sbi, old_blkaddr); 3742 if (!from_gc) 3743 update_segment_mtime(sbi, old_blkaddr, 0); 3744 update_sit_entry(sbi, old_blkaddr, -1); 3745 } 3746 3747 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3748 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 3749 3750 locate_dirty_segment(sbi, old_cursegno); 3751 3752 if (recover_curseg) { 3753 if (old_cursegno != curseg->segno) { 3754 curseg->next_segno = old_cursegno; 3755 change_curseg(sbi, type); 3756 } 3757 curseg->next_blkoff = old_blkoff; 3758 curseg->alloc_type = old_alloc_type; 3759 } 3760 3761 up_write(&sit_i->sentry_lock); 3762 mutex_unlock(&curseg->curseg_mutex); 3763 f2fs_up_write(&SM_I(sbi)->curseg_lock); 3764 } 3765 3766 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 3767 block_t old_addr, block_t new_addr, 3768 unsigned char version, bool recover_curseg, 3769 bool recover_newaddr) 3770 { 3771 struct f2fs_summary sum; 3772 3773 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 3774 3775 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 3776 recover_curseg, recover_newaddr, false); 3777 3778 f2fs_update_data_blkaddr(dn, new_addr); 3779 } 3780 3781 void f2fs_wait_on_page_writeback(struct page *page, 3782 enum page_type type, bool ordered, bool locked) 3783 { 3784 if (PageWriteback(page)) { 3785 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 3786 3787 /* submit cached LFS IO */ 3788 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type); 3789 /* submit cached IPU IO */ 3790 f2fs_submit_merged_ipu_write(sbi, NULL, page); 3791 if (ordered) { 3792 wait_on_page_writeback(page); 3793 f2fs_bug_on(sbi, locked && PageWriteback(page)); 3794 } else { 3795 wait_for_stable_page(page); 3796 } 3797 } 3798 } 3799 3800 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 3801 { 3802 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3803 struct page *cpage; 3804 3805 if (!f2fs_post_read_required(inode)) 3806 return; 3807 3808 if (!__is_valid_data_blkaddr(blkaddr)) 3809 return; 3810 3811 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3812 if (cpage) { 3813 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 3814 f2fs_put_page(cpage, 1); 3815 } 3816 } 3817 3818 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 3819 block_t len) 3820 { 3821 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3822 block_t i; 3823 3824 if (!f2fs_post_read_required(inode)) 3825 return; 3826 3827 for (i = 0; i < len; i++) 3828 f2fs_wait_on_block_writeback(inode, blkaddr + i); 3829 3830 invalidate_mapping_pages(META_MAPPING(sbi), blkaddr, blkaddr + len - 1); 3831 } 3832 3833 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 3834 { 3835 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3836 struct curseg_info *seg_i; 3837 unsigned char *kaddr; 3838 struct page *page; 3839 block_t start; 3840 int i, j, offset; 3841 3842 start = start_sum_block(sbi); 3843 3844 page = f2fs_get_meta_page(sbi, start++); 3845 if (IS_ERR(page)) 3846 return PTR_ERR(page); 3847 kaddr = (unsigned char *)page_address(page); 3848 3849 /* Step 1: restore nat cache */ 3850 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3851 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3852 3853 /* Step 2: restore sit cache */ 3854 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3855 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3856 offset = 2 * SUM_JOURNAL_SIZE; 3857 3858 /* Step 3: restore summary entries */ 3859 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3860 unsigned short blk_off; 3861 unsigned int segno; 3862 3863 seg_i = CURSEG_I(sbi, i); 3864 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3865 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3866 seg_i->next_segno = segno; 3867 reset_curseg(sbi, i, 0); 3868 seg_i->alloc_type = ckpt->alloc_type[i]; 3869 seg_i->next_blkoff = blk_off; 3870 3871 if (seg_i->alloc_type == SSR) 3872 blk_off = sbi->blocks_per_seg; 3873 3874 for (j = 0; j < blk_off; j++) { 3875 struct f2fs_summary *s; 3876 3877 s = (struct f2fs_summary *)(kaddr + offset); 3878 seg_i->sum_blk->entries[j] = *s; 3879 offset += SUMMARY_SIZE; 3880 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3881 SUM_FOOTER_SIZE) 3882 continue; 3883 3884 f2fs_put_page(page, 1); 3885 page = NULL; 3886 3887 page = f2fs_get_meta_page(sbi, start++); 3888 if (IS_ERR(page)) 3889 return PTR_ERR(page); 3890 kaddr = (unsigned char *)page_address(page); 3891 offset = 0; 3892 } 3893 } 3894 f2fs_put_page(page, 1); 3895 return 0; 3896 } 3897 3898 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3899 { 3900 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3901 struct f2fs_summary_block *sum; 3902 struct curseg_info *curseg; 3903 struct page *new; 3904 unsigned short blk_off; 3905 unsigned int segno = 0; 3906 block_t blk_addr = 0; 3907 int err = 0; 3908 3909 /* get segment number and block addr */ 3910 if (IS_DATASEG(type)) { 3911 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3912 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3913 CURSEG_HOT_DATA]); 3914 if (__exist_node_summaries(sbi)) 3915 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type); 3916 else 3917 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3918 } else { 3919 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3920 CURSEG_HOT_NODE]); 3921 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3922 CURSEG_HOT_NODE]); 3923 if (__exist_node_summaries(sbi)) 3924 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3925 type - CURSEG_HOT_NODE); 3926 else 3927 blk_addr = GET_SUM_BLOCK(sbi, segno); 3928 } 3929 3930 new = f2fs_get_meta_page(sbi, blk_addr); 3931 if (IS_ERR(new)) 3932 return PTR_ERR(new); 3933 sum = (struct f2fs_summary_block *)page_address(new); 3934 3935 if (IS_NODESEG(type)) { 3936 if (__exist_node_summaries(sbi)) { 3937 struct f2fs_summary *ns = &sum->entries[0]; 3938 int i; 3939 3940 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3941 ns->version = 0; 3942 ns->ofs_in_node = 0; 3943 } 3944 } else { 3945 err = f2fs_restore_node_summary(sbi, segno, sum); 3946 if (err) 3947 goto out; 3948 } 3949 } 3950 3951 /* set uncompleted segment to curseg */ 3952 curseg = CURSEG_I(sbi, type); 3953 mutex_lock(&curseg->curseg_mutex); 3954 3955 /* update journal info */ 3956 down_write(&curseg->journal_rwsem); 3957 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3958 up_write(&curseg->journal_rwsem); 3959 3960 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3961 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3962 curseg->next_segno = segno; 3963 reset_curseg(sbi, type, 0); 3964 curseg->alloc_type = ckpt->alloc_type[type]; 3965 curseg->next_blkoff = blk_off; 3966 mutex_unlock(&curseg->curseg_mutex); 3967 out: 3968 f2fs_put_page(new, 1); 3969 return err; 3970 } 3971 3972 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3973 { 3974 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3975 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3976 int type = CURSEG_HOT_DATA; 3977 int err; 3978 3979 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 3980 int npages = f2fs_npages_for_summary_flush(sbi, true); 3981 3982 if (npages >= 2) 3983 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 3984 META_CP, true); 3985 3986 /* restore for compacted data summary */ 3987 err = read_compacted_summaries(sbi); 3988 if (err) 3989 return err; 3990 type = CURSEG_HOT_NODE; 3991 } 3992 3993 if (__exist_node_summaries(sbi)) 3994 f2fs_ra_meta_pages(sbi, 3995 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type), 3996 NR_CURSEG_PERSIST_TYPE - type, META_CP, true); 3997 3998 for (; type <= CURSEG_COLD_NODE; type++) { 3999 err = read_normal_summaries(sbi, type); 4000 if (err) 4001 return err; 4002 } 4003 4004 /* sanity check for summary blocks */ 4005 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 4006 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 4007 f2fs_err(sbi, "invalid journal entries nats %u sits %u", 4008 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 4009 return -EINVAL; 4010 } 4011 4012 return 0; 4013 } 4014 4015 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 4016 { 4017 struct page *page; 4018 unsigned char *kaddr; 4019 struct f2fs_summary *summary; 4020 struct curseg_info *seg_i; 4021 int written_size = 0; 4022 int i, j; 4023 4024 page = f2fs_grab_meta_page(sbi, blkaddr++); 4025 kaddr = (unsigned char *)page_address(page); 4026 memset(kaddr, 0, PAGE_SIZE); 4027 4028 /* Step 1: write nat cache */ 4029 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4030 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 4031 written_size += SUM_JOURNAL_SIZE; 4032 4033 /* Step 2: write sit cache */ 4034 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4035 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 4036 written_size += SUM_JOURNAL_SIZE; 4037 4038 /* Step 3: write summary entries */ 4039 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4040 seg_i = CURSEG_I(sbi, i); 4041 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) { 4042 if (!page) { 4043 page = f2fs_grab_meta_page(sbi, blkaddr++); 4044 kaddr = (unsigned char *)page_address(page); 4045 memset(kaddr, 0, PAGE_SIZE); 4046 written_size = 0; 4047 } 4048 summary = (struct f2fs_summary *)(kaddr + written_size); 4049 *summary = seg_i->sum_blk->entries[j]; 4050 written_size += SUMMARY_SIZE; 4051 4052 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 4053 SUM_FOOTER_SIZE) 4054 continue; 4055 4056 set_page_dirty(page); 4057 f2fs_put_page(page, 1); 4058 page = NULL; 4059 } 4060 } 4061 if (page) { 4062 set_page_dirty(page); 4063 f2fs_put_page(page, 1); 4064 } 4065 } 4066 4067 static void write_normal_summaries(struct f2fs_sb_info *sbi, 4068 block_t blkaddr, int type) 4069 { 4070 int i, end; 4071 4072 if (IS_DATASEG(type)) 4073 end = type + NR_CURSEG_DATA_TYPE; 4074 else 4075 end = type + NR_CURSEG_NODE_TYPE; 4076 4077 for (i = type; i < end; i++) 4078 write_current_sum_page(sbi, i, blkaddr + (i - type)); 4079 } 4080 4081 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4082 { 4083 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 4084 write_compacted_summaries(sbi, start_blk); 4085 else 4086 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 4087 } 4088 4089 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4090 { 4091 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 4092 } 4093 4094 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 4095 unsigned int val, int alloc) 4096 { 4097 int i; 4098 4099 if (type == NAT_JOURNAL) { 4100 for (i = 0; i < nats_in_cursum(journal); i++) { 4101 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 4102 return i; 4103 } 4104 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 4105 return update_nats_in_cursum(journal, 1); 4106 } else if (type == SIT_JOURNAL) { 4107 for (i = 0; i < sits_in_cursum(journal); i++) 4108 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 4109 return i; 4110 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 4111 return update_sits_in_cursum(journal, 1); 4112 } 4113 return -1; 4114 } 4115 4116 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 4117 unsigned int segno) 4118 { 4119 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno)); 4120 } 4121 4122 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 4123 unsigned int start) 4124 { 4125 struct sit_info *sit_i = SIT_I(sbi); 4126 struct page *page; 4127 pgoff_t src_off, dst_off; 4128 4129 src_off = current_sit_addr(sbi, start); 4130 dst_off = next_sit_addr(sbi, src_off); 4131 4132 page = f2fs_grab_meta_page(sbi, dst_off); 4133 seg_info_to_sit_page(sbi, page, start); 4134 4135 set_page_dirty(page); 4136 set_to_next_sit(sit_i, start); 4137 4138 return page; 4139 } 4140 4141 static struct sit_entry_set *grab_sit_entry_set(void) 4142 { 4143 struct sit_entry_set *ses = 4144 f2fs_kmem_cache_alloc(sit_entry_set_slab, 4145 GFP_NOFS, true, NULL); 4146 4147 ses->entry_cnt = 0; 4148 INIT_LIST_HEAD(&ses->set_list); 4149 return ses; 4150 } 4151 4152 static void release_sit_entry_set(struct sit_entry_set *ses) 4153 { 4154 list_del(&ses->set_list); 4155 kmem_cache_free(sit_entry_set_slab, ses); 4156 } 4157 4158 static void adjust_sit_entry_set(struct sit_entry_set *ses, 4159 struct list_head *head) 4160 { 4161 struct sit_entry_set *next = ses; 4162 4163 if (list_is_last(&ses->set_list, head)) 4164 return; 4165 4166 list_for_each_entry_continue(next, head, set_list) 4167 if (ses->entry_cnt <= next->entry_cnt) { 4168 list_move_tail(&ses->set_list, &next->set_list); 4169 return; 4170 } 4171 4172 list_move_tail(&ses->set_list, head); 4173 } 4174 4175 static void add_sit_entry(unsigned int segno, struct list_head *head) 4176 { 4177 struct sit_entry_set *ses; 4178 unsigned int start_segno = START_SEGNO(segno); 4179 4180 list_for_each_entry(ses, head, set_list) { 4181 if (ses->start_segno == start_segno) { 4182 ses->entry_cnt++; 4183 adjust_sit_entry_set(ses, head); 4184 return; 4185 } 4186 } 4187 4188 ses = grab_sit_entry_set(); 4189 4190 ses->start_segno = start_segno; 4191 ses->entry_cnt++; 4192 list_add(&ses->set_list, head); 4193 } 4194 4195 static void add_sits_in_set(struct f2fs_sb_info *sbi) 4196 { 4197 struct f2fs_sm_info *sm_info = SM_I(sbi); 4198 struct list_head *set_list = &sm_info->sit_entry_set; 4199 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 4200 unsigned int segno; 4201 4202 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 4203 add_sit_entry(segno, set_list); 4204 } 4205 4206 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 4207 { 4208 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4209 struct f2fs_journal *journal = curseg->journal; 4210 int i; 4211 4212 down_write(&curseg->journal_rwsem); 4213 for (i = 0; i < sits_in_cursum(journal); i++) { 4214 unsigned int segno; 4215 bool dirtied; 4216 4217 segno = le32_to_cpu(segno_in_journal(journal, i)); 4218 dirtied = __mark_sit_entry_dirty(sbi, segno); 4219 4220 if (!dirtied) 4221 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 4222 } 4223 update_sits_in_cursum(journal, -i); 4224 up_write(&curseg->journal_rwsem); 4225 } 4226 4227 /* 4228 * CP calls this function, which flushes SIT entries including sit_journal, 4229 * and moves prefree segs to free segs. 4230 */ 4231 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 4232 { 4233 struct sit_info *sit_i = SIT_I(sbi); 4234 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 4235 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4236 struct f2fs_journal *journal = curseg->journal; 4237 struct sit_entry_set *ses, *tmp; 4238 struct list_head *head = &SM_I(sbi)->sit_entry_set; 4239 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 4240 struct seg_entry *se; 4241 4242 down_write(&sit_i->sentry_lock); 4243 4244 if (!sit_i->dirty_sentries) 4245 goto out; 4246 4247 /* 4248 * add and account sit entries of dirty bitmap in sit entry 4249 * set temporarily 4250 */ 4251 add_sits_in_set(sbi); 4252 4253 /* 4254 * if there are no enough space in journal to store dirty sit 4255 * entries, remove all entries from journal and add and account 4256 * them in sit entry set. 4257 */ 4258 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 4259 !to_journal) 4260 remove_sits_in_journal(sbi); 4261 4262 /* 4263 * there are two steps to flush sit entries: 4264 * #1, flush sit entries to journal in current cold data summary block. 4265 * #2, flush sit entries to sit page. 4266 */ 4267 list_for_each_entry_safe(ses, tmp, head, set_list) { 4268 struct page *page = NULL; 4269 struct f2fs_sit_block *raw_sit = NULL; 4270 unsigned int start_segno = ses->start_segno; 4271 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 4272 (unsigned long)MAIN_SEGS(sbi)); 4273 unsigned int segno = start_segno; 4274 4275 if (to_journal && 4276 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 4277 to_journal = false; 4278 4279 if (to_journal) { 4280 down_write(&curseg->journal_rwsem); 4281 } else { 4282 page = get_next_sit_page(sbi, start_segno); 4283 raw_sit = page_address(page); 4284 } 4285 4286 /* flush dirty sit entries in region of current sit set */ 4287 for_each_set_bit_from(segno, bitmap, end) { 4288 int offset, sit_offset; 4289 4290 se = get_seg_entry(sbi, segno); 4291 #ifdef CONFIG_F2FS_CHECK_FS 4292 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 4293 SIT_VBLOCK_MAP_SIZE)) 4294 f2fs_bug_on(sbi, 1); 4295 #endif 4296 4297 /* add discard candidates */ 4298 if (!(cpc->reason & CP_DISCARD)) { 4299 cpc->trim_start = segno; 4300 add_discard_addrs(sbi, cpc, false); 4301 } 4302 4303 if (to_journal) { 4304 offset = f2fs_lookup_journal_in_cursum(journal, 4305 SIT_JOURNAL, segno, 1); 4306 f2fs_bug_on(sbi, offset < 0); 4307 segno_in_journal(journal, offset) = 4308 cpu_to_le32(segno); 4309 seg_info_to_raw_sit(se, 4310 &sit_in_journal(journal, offset)); 4311 check_block_count(sbi, segno, 4312 &sit_in_journal(journal, offset)); 4313 } else { 4314 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 4315 seg_info_to_raw_sit(se, 4316 &raw_sit->entries[sit_offset]); 4317 check_block_count(sbi, segno, 4318 &raw_sit->entries[sit_offset]); 4319 } 4320 4321 __clear_bit(segno, bitmap); 4322 sit_i->dirty_sentries--; 4323 ses->entry_cnt--; 4324 } 4325 4326 if (to_journal) 4327 up_write(&curseg->journal_rwsem); 4328 else 4329 f2fs_put_page(page, 1); 4330 4331 f2fs_bug_on(sbi, ses->entry_cnt); 4332 release_sit_entry_set(ses); 4333 } 4334 4335 f2fs_bug_on(sbi, !list_empty(head)); 4336 f2fs_bug_on(sbi, sit_i->dirty_sentries); 4337 out: 4338 if (cpc->reason & CP_DISCARD) { 4339 __u64 trim_start = cpc->trim_start; 4340 4341 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 4342 add_discard_addrs(sbi, cpc, false); 4343 4344 cpc->trim_start = trim_start; 4345 } 4346 up_write(&sit_i->sentry_lock); 4347 4348 set_prefree_as_free_segments(sbi); 4349 } 4350 4351 static int build_sit_info(struct f2fs_sb_info *sbi) 4352 { 4353 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4354 struct sit_info *sit_i; 4355 unsigned int sit_segs, start; 4356 char *src_bitmap, *bitmap; 4357 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 4358 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0; 4359 4360 /* allocate memory for SIT information */ 4361 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4362 if (!sit_i) 4363 return -ENOMEM; 4364 4365 SM_I(sbi)->sit_info = sit_i; 4366 4367 sit_i->sentries = 4368 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4369 MAIN_SEGS(sbi)), 4370 GFP_KERNEL); 4371 if (!sit_i->sentries) 4372 return -ENOMEM; 4373 4374 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4375 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4376 GFP_KERNEL); 4377 if (!sit_i->dirty_sentries_bitmap) 4378 return -ENOMEM; 4379 4380 #ifdef CONFIG_F2FS_CHECK_FS 4381 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map); 4382 #else 4383 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map); 4384 #endif 4385 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4386 if (!sit_i->bitmap) 4387 return -ENOMEM; 4388 4389 bitmap = sit_i->bitmap; 4390 4391 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4392 sit_i->sentries[start].cur_valid_map = bitmap; 4393 bitmap += SIT_VBLOCK_MAP_SIZE; 4394 4395 sit_i->sentries[start].ckpt_valid_map = bitmap; 4396 bitmap += SIT_VBLOCK_MAP_SIZE; 4397 4398 #ifdef CONFIG_F2FS_CHECK_FS 4399 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4400 bitmap += SIT_VBLOCK_MAP_SIZE; 4401 #endif 4402 4403 if (discard_map) { 4404 sit_i->sentries[start].discard_map = bitmap; 4405 bitmap += SIT_VBLOCK_MAP_SIZE; 4406 } 4407 } 4408 4409 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4410 if (!sit_i->tmp_map) 4411 return -ENOMEM; 4412 4413 if (__is_large_section(sbi)) { 4414 sit_i->sec_entries = 4415 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4416 MAIN_SECS(sbi)), 4417 GFP_KERNEL); 4418 if (!sit_i->sec_entries) 4419 return -ENOMEM; 4420 } 4421 4422 /* get information related with SIT */ 4423 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4424 4425 /* setup SIT bitmap from ckeckpoint pack */ 4426 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4427 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4428 4429 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4430 if (!sit_i->sit_bitmap) 4431 return -ENOMEM; 4432 4433 #ifdef CONFIG_F2FS_CHECK_FS 4434 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4435 sit_bitmap_size, GFP_KERNEL); 4436 if (!sit_i->sit_bitmap_mir) 4437 return -ENOMEM; 4438 4439 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4440 main_bitmap_size, GFP_KERNEL); 4441 if (!sit_i->invalid_segmap) 4442 return -ENOMEM; 4443 #endif 4444 4445 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4446 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 4447 sit_i->written_valid_blocks = 0; 4448 sit_i->bitmap_size = sit_bitmap_size; 4449 sit_i->dirty_sentries = 0; 4450 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4451 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4452 sit_i->mounted_time = ktime_get_boottime_seconds(); 4453 init_rwsem(&sit_i->sentry_lock); 4454 return 0; 4455 } 4456 4457 static int build_free_segmap(struct f2fs_sb_info *sbi) 4458 { 4459 struct free_segmap_info *free_i; 4460 unsigned int bitmap_size, sec_bitmap_size; 4461 4462 /* allocate memory for free segmap information */ 4463 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4464 if (!free_i) 4465 return -ENOMEM; 4466 4467 SM_I(sbi)->free_info = free_i; 4468 4469 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4470 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4471 if (!free_i->free_segmap) 4472 return -ENOMEM; 4473 4474 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4475 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4476 if (!free_i->free_secmap) 4477 return -ENOMEM; 4478 4479 /* set all segments as dirty temporarily */ 4480 memset(free_i->free_segmap, 0xff, bitmap_size); 4481 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4482 4483 /* init free segmap information */ 4484 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4485 free_i->free_segments = 0; 4486 free_i->free_sections = 0; 4487 spin_lock_init(&free_i->segmap_lock); 4488 return 0; 4489 } 4490 4491 static int build_curseg(struct f2fs_sb_info *sbi) 4492 { 4493 struct curseg_info *array; 4494 int i; 4495 4496 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, 4497 sizeof(*array)), GFP_KERNEL); 4498 if (!array) 4499 return -ENOMEM; 4500 4501 SM_I(sbi)->curseg_array = array; 4502 4503 for (i = 0; i < NO_CHECK_TYPE; i++) { 4504 mutex_init(&array[i].curseg_mutex); 4505 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4506 if (!array[i].sum_blk) 4507 return -ENOMEM; 4508 init_rwsem(&array[i].journal_rwsem); 4509 array[i].journal = f2fs_kzalloc(sbi, 4510 sizeof(struct f2fs_journal), GFP_KERNEL); 4511 if (!array[i].journal) 4512 return -ENOMEM; 4513 if (i < NR_PERSISTENT_LOG) 4514 array[i].seg_type = CURSEG_HOT_DATA + i; 4515 else if (i == CURSEG_COLD_DATA_PINNED) 4516 array[i].seg_type = CURSEG_COLD_DATA; 4517 else if (i == CURSEG_ALL_DATA_ATGC) 4518 array[i].seg_type = CURSEG_COLD_DATA; 4519 array[i].segno = NULL_SEGNO; 4520 array[i].next_blkoff = 0; 4521 array[i].inited = false; 4522 } 4523 return restore_curseg_summaries(sbi); 4524 } 4525 4526 static int build_sit_entries(struct f2fs_sb_info *sbi) 4527 { 4528 struct sit_info *sit_i = SIT_I(sbi); 4529 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4530 struct f2fs_journal *journal = curseg->journal; 4531 struct seg_entry *se; 4532 struct f2fs_sit_entry sit; 4533 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4534 unsigned int i, start, end; 4535 unsigned int readed, start_blk = 0; 4536 int err = 0; 4537 block_t sit_valid_blocks[2] = {0, 0}; 4538 4539 do { 4540 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS, 4541 META_SIT, true); 4542 4543 start = start_blk * sit_i->sents_per_block; 4544 end = (start_blk + readed) * sit_i->sents_per_block; 4545 4546 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4547 struct f2fs_sit_block *sit_blk; 4548 struct page *page; 4549 4550 se = &sit_i->sentries[start]; 4551 page = get_current_sit_page(sbi, start); 4552 if (IS_ERR(page)) 4553 return PTR_ERR(page); 4554 sit_blk = (struct f2fs_sit_block *)page_address(page); 4555 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4556 f2fs_put_page(page, 1); 4557 4558 err = check_block_count(sbi, start, &sit); 4559 if (err) 4560 return err; 4561 seg_info_from_raw_sit(se, &sit); 4562 4563 if (se->type >= NR_PERSISTENT_LOG) { 4564 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4565 se->type, start); 4566 f2fs_handle_error(sbi, 4567 ERROR_INCONSISTENT_SUM_TYPE); 4568 return -EFSCORRUPTED; 4569 } 4570 4571 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4572 4573 if (f2fs_block_unit_discard(sbi)) { 4574 /* build discard map only one time */ 4575 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4576 memset(se->discard_map, 0xff, 4577 SIT_VBLOCK_MAP_SIZE); 4578 } else { 4579 memcpy(se->discard_map, 4580 se->cur_valid_map, 4581 SIT_VBLOCK_MAP_SIZE); 4582 sbi->discard_blks += 4583 sbi->blocks_per_seg - 4584 se->valid_blocks; 4585 } 4586 } 4587 4588 if (__is_large_section(sbi)) 4589 get_sec_entry(sbi, start)->valid_blocks += 4590 se->valid_blocks; 4591 } 4592 start_blk += readed; 4593 } while (start_blk < sit_blk_cnt); 4594 4595 down_read(&curseg->journal_rwsem); 4596 for (i = 0; i < sits_in_cursum(journal); i++) { 4597 unsigned int old_valid_blocks; 4598 4599 start = le32_to_cpu(segno_in_journal(journal, i)); 4600 if (start >= MAIN_SEGS(sbi)) { 4601 f2fs_err(sbi, "Wrong journal entry on segno %u", 4602 start); 4603 err = -EFSCORRUPTED; 4604 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL); 4605 break; 4606 } 4607 4608 se = &sit_i->sentries[start]; 4609 sit = sit_in_journal(journal, i); 4610 4611 old_valid_blocks = se->valid_blocks; 4612 4613 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks; 4614 4615 err = check_block_count(sbi, start, &sit); 4616 if (err) 4617 break; 4618 seg_info_from_raw_sit(se, &sit); 4619 4620 if (se->type >= NR_PERSISTENT_LOG) { 4621 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4622 se->type, start); 4623 err = -EFSCORRUPTED; 4624 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 4625 break; 4626 } 4627 4628 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4629 4630 if (f2fs_block_unit_discard(sbi)) { 4631 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4632 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 4633 } else { 4634 memcpy(se->discard_map, se->cur_valid_map, 4635 SIT_VBLOCK_MAP_SIZE); 4636 sbi->discard_blks += old_valid_blocks; 4637 sbi->discard_blks -= se->valid_blocks; 4638 } 4639 } 4640 4641 if (__is_large_section(sbi)) { 4642 get_sec_entry(sbi, start)->valid_blocks += 4643 se->valid_blocks; 4644 get_sec_entry(sbi, start)->valid_blocks -= 4645 old_valid_blocks; 4646 } 4647 } 4648 up_read(&curseg->journal_rwsem); 4649 4650 if (err) 4651 return err; 4652 4653 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) { 4654 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 4655 sit_valid_blocks[NODE], valid_node_count(sbi)); 4656 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT); 4657 return -EFSCORRUPTED; 4658 } 4659 4660 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] > 4661 valid_user_blocks(sbi)) { 4662 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u", 4663 sit_valid_blocks[DATA], sit_valid_blocks[NODE], 4664 valid_user_blocks(sbi)); 4665 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT); 4666 return -EFSCORRUPTED; 4667 } 4668 4669 return 0; 4670 } 4671 4672 static void init_free_segmap(struct f2fs_sb_info *sbi) 4673 { 4674 unsigned int start; 4675 int type; 4676 struct seg_entry *sentry; 4677 4678 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4679 if (f2fs_usable_blks_in_seg(sbi, start) == 0) 4680 continue; 4681 sentry = get_seg_entry(sbi, start); 4682 if (!sentry->valid_blocks) 4683 __set_free(sbi, start); 4684 else 4685 SIT_I(sbi)->written_valid_blocks += 4686 sentry->valid_blocks; 4687 } 4688 4689 /* set use the current segments */ 4690 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 4691 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 4692 4693 __set_test_and_inuse(sbi, curseg_t->segno); 4694 } 4695 } 4696 4697 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 4698 { 4699 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4700 struct free_segmap_info *free_i = FREE_I(sbi); 4701 unsigned int segno = 0, offset = 0, secno; 4702 block_t valid_blocks, usable_blks_in_seg; 4703 4704 while (1) { 4705 /* find dirty segment based on free segmap */ 4706 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 4707 if (segno >= MAIN_SEGS(sbi)) 4708 break; 4709 offset = segno + 1; 4710 valid_blocks = get_valid_blocks(sbi, segno, false); 4711 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 4712 if (valid_blocks == usable_blks_in_seg || !valid_blocks) 4713 continue; 4714 if (valid_blocks > usable_blks_in_seg) { 4715 f2fs_bug_on(sbi, 1); 4716 continue; 4717 } 4718 mutex_lock(&dirty_i->seglist_lock); 4719 __locate_dirty_segment(sbi, segno, DIRTY); 4720 mutex_unlock(&dirty_i->seglist_lock); 4721 } 4722 4723 if (!__is_large_section(sbi)) 4724 return; 4725 4726 mutex_lock(&dirty_i->seglist_lock); 4727 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 4728 valid_blocks = get_valid_blocks(sbi, segno, true); 4729 secno = GET_SEC_FROM_SEG(sbi, segno); 4730 4731 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi)) 4732 continue; 4733 if (IS_CURSEC(sbi, secno)) 4734 continue; 4735 set_bit(secno, dirty_i->dirty_secmap); 4736 } 4737 mutex_unlock(&dirty_i->seglist_lock); 4738 } 4739 4740 static int init_victim_secmap(struct f2fs_sb_info *sbi) 4741 { 4742 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4743 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4744 4745 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4746 if (!dirty_i->victim_secmap) 4747 return -ENOMEM; 4748 4749 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4750 if (!dirty_i->pinned_secmap) 4751 return -ENOMEM; 4752 4753 dirty_i->pinned_secmap_cnt = 0; 4754 dirty_i->enable_pin_section = true; 4755 return 0; 4756 } 4757 4758 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 4759 { 4760 struct dirty_seglist_info *dirty_i; 4761 unsigned int bitmap_size, i; 4762 4763 /* allocate memory for dirty segments list information */ 4764 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 4765 GFP_KERNEL); 4766 if (!dirty_i) 4767 return -ENOMEM; 4768 4769 SM_I(sbi)->dirty_info = dirty_i; 4770 mutex_init(&dirty_i->seglist_lock); 4771 4772 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4773 4774 for (i = 0; i < NR_DIRTY_TYPE; i++) { 4775 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 4776 GFP_KERNEL); 4777 if (!dirty_i->dirty_segmap[i]) 4778 return -ENOMEM; 4779 } 4780 4781 if (__is_large_section(sbi)) { 4782 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4783 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi, 4784 bitmap_size, GFP_KERNEL); 4785 if (!dirty_i->dirty_secmap) 4786 return -ENOMEM; 4787 } 4788 4789 init_dirty_segmap(sbi); 4790 return init_victim_secmap(sbi); 4791 } 4792 4793 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 4794 { 4795 int i; 4796 4797 /* 4798 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 4799 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 4800 */ 4801 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 4802 struct curseg_info *curseg = CURSEG_I(sbi, i); 4803 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 4804 unsigned int blkofs = curseg->next_blkoff; 4805 4806 if (f2fs_sb_has_readonly(sbi) && 4807 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE) 4808 continue; 4809 4810 sanity_check_seg_type(sbi, curseg->seg_type); 4811 4812 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) { 4813 f2fs_err(sbi, 4814 "Current segment has invalid alloc_type:%d", 4815 curseg->alloc_type); 4816 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 4817 return -EFSCORRUPTED; 4818 } 4819 4820 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 4821 goto out; 4822 4823 if (curseg->alloc_type == SSR) 4824 continue; 4825 4826 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) { 4827 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 4828 continue; 4829 out: 4830 f2fs_err(sbi, 4831 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 4832 i, curseg->segno, curseg->alloc_type, 4833 curseg->next_blkoff, blkofs); 4834 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 4835 return -EFSCORRUPTED; 4836 } 4837 } 4838 return 0; 4839 } 4840 4841 #ifdef CONFIG_BLK_DEV_ZONED 4842 4843 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 4844 struct f2fs_dev_info *fdev, 4845 struct blk_zone *zone) 4846 { 4847 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno; 4848 block_t zone_block, wp_block, last_valid_block; 4849 int i, s, b, ret; 4850 struct seg_entry *se; 4851 4852 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4853 return 0; 4854 4855 wp_block = fdev->start_blk + (zone->wp >> sbi->log_sectors_per_block); 4856 wp_segno = GET_SEGNO(sbi, wp_block); 4857 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4858 zone_block = fdev->start_blk + (zone->start >> 4859 sbi->log_sectors_per_block); 4860 zone_segno = GET_SEGNO(sbi, zone_block); 4861 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno); 4862 4863 if (zone_segno >= MAIN_SEGS(sbi)) 4864 return 0; 4865 4866 /* 4867 * Skip check of zones cursegs point to, since 4868 * fix_curseg_write_pointer() checks them. 4869 */ 4870 for (i = 0; i < NO_CHECK_TYPE; i++) 4871 if (zone_secno == GET_SEC_FROM_SEG(sbi, 4872 CURSEG_I(sbi, i)->segno)) 4873 return 0; 4874 4875 /* 4876 * Get last valid block of the zone. 4877 */ 4878 last_valid_block = zone_block - 1; 4879 for (s = sbi->segs_per_sec - 1; s >= 0; s--) { 4880 segno = zone_segno + s; 4881 se = get_seg_entry(sbi, segno); 4882 for (b = sbi->blocks_per_seg - 1; b >= 0; b--) 4883 if (f2fs_test_bit(b, se->cur_valid_map)) { 4884 last_valid_block = START_BLOCK(sbi, segno) + b; 4885 break; 4886 } 4887 if (last_valid_block >= zone_block) 4888 break; 4889 } 4890 4891 /* 4892 * The write pointer matches with the valid blocks or 4893 * already points to the end of the zone. 4894 */ 4895 if ((last_valid_block + 1 == wp_block) || 4896 (zone->wp == zone->start + zone->len)) 4897 return 0; 4898 4899 if (last_valid_block + 1 == zone_block) { 4900 /* 4901 * If there is no valid block in the zone and if write pointer 4902 * is not at zone start, reset the write pointer. 4903 */ 4904 f2fs_notice(sbi, 4905 "Zone without valid block has non-zero write " 4906 "pointer. Reset the write pointer: wp[0x%x,0x%x]", 4907 wp_segno, wp_blkoff); 4908 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 4909 zone->len >> sbi->log_sectors_per_block); 4910 if (ret) 4911 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4912 fdev->path, ret); 4913 4914 return ret; 4915 } 4916 4917 /* 4918 * If there are valid blocks and the write pointer doesn't 4919 * match with them, we need to report the inconsistency and 4920 * fill the zone till the end to close the zone. This inconsistency 4921 * does not cause write error because the zone will not be selected 4922 * for write operation until it get discarded. 4923 */ 4924 f2fs_notice(sbi, "Valid blocks are not aligned with write pointer: " 4925 "valid block[0x%x,0x%x] wp[0x%x,0x%x]", 4926 GET_SEGNO(sbi, last_valid_block), 4927 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block), 4928 wp_segno, wp_blkoff); 4929 4930 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp, 4931 zone->len - (zone->wp - zone->start), 4932 GFP_NOFS, 0); 4933 if (ret) 4934 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)", 4935 fdev->path, ret); 4936 4937 return ret; 4938 } 4939 4940 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 4941 block_t zone_blkaddr) 4942 { 4943 int i; 4944 4945 for (i = 0; i < sbi->s_ndevs; i++) { 4946 if (!bdev_is_zoned(FDEV(i).bdev)) 4947 continue; 4948 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 4949 zone_blkaddr <= FDEV(i).end_blk)) 4950 return &FDEV(i); 4951 } 4952 4953 return NULL; 4954 } 4955 4956 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 4957 void *data) 4958 { 4959 memcpy(data, zone, sizeof(struct blk_zone)); 4960 return 0; 4961 } 4962 4963 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 4964 { 4965 struct curseg_info *cs = CURSEG_I(sbi, type); 4966 struct f2fs_dev_info *zbd; 4967 struct blk_zone zone; 4968 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 4969 block_t cs_zone_block, wp_block; 4970 sector_t zone_sector; 4971 int err; 4972 4973 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4974 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4975 4976 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4977 if (!zbd) 4978 return 0; 4979 4980 /* report zone for the sector the curseg points to */ 4981 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) << 4982 sbi->log_sectors_per_block; 4983 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4984 report_one_zone_cb, &zone); 4985 if (err != 1) { 4986 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4987 zbd->path, err); 4988 return err; 4989 } 4990 4991 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4992 return 0; 4993 4994 wp_block = zbd->start_blk + (zone.wp >> sbi->log_sectors_per_block); 4995 wp_segno = GET_SEGNO(sbi, wp_block); 4996 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4997 wp_sector_off = zone.wp & GENMASK(sbi->log_sectors_per_block - 1, 0); 4998 4999 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 5000 wp_sector_off == 0) 5001 return 0; 5002 5003 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 5004 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", 5005 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff); 5006 5007 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 5008 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff); 5009 5010 f2fs_allocate_new_section(sbi, type, true); 5011 5012 /* check consistency of the zone curseg pointed to */ 5013 if (check_zone_write_pointer(sbi, zbd, &zone)) 5014 return -EIO; 5015 5016 /* check newly assigned zone */ 5017 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5018 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5019 5020 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5021 if (!zbd) 5022 return 0; 5023 5024 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) << 5025 sbi->log_sectors_per_block; 5026 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5027 report_one_zone_cb, &zone); 5028 if (err != 1) { 5029 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5030 zbd->path, err); 5031 return err; 5032 } 5033 5034 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5035 return 0; 5036 5037 if (zone.wp != zone.start) { 5038 f2fs_notice(sbi, 5039 "New zone for curseg[%d] is not yet discarded. " 5040 "Reset the zone: curseg[0x%x,0x%x]", 5041 type, cs->segno, cs->next_blkoff); 5042 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block, 5043 zone.len >> sbi->log_sectors_per_block); 5044 if (err) { 5045 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5046 zbd->path, err); 5047 return err; 5048 } 5049 } 5050 5051 return 0; 5052 } 5053 5054 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5055 { 5056 int i, ret; 5057 5058 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5059 ret = fix_curseg_write_pointer(sbi, i); 5060 if (ret) 5061 return ret; 5062 } 5063 5064 return 0; 5065 } 5066 5067 struct check_zone_write_pointer_args { 5068 struct f2fs_sb_info *sbi; 5069 struct f2fs_dev_info *fdev; 5070 }; 5071 5072 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 5073 void *data) 5074 { 5075 struct check_zone_write_pointer_args *args; 5076 5077 args = (struct check_zone_write_pointer_args *)data; 5078 5079 return check_zone_write_pointer(args->sbi, args->fdev, zone); 5080 } 5081 5082 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5083 { 5084 int i, ret; 5085 struct check_zone_write_pointer_args args; 5086 5087 for (i = 0; i < sbi->s_ndevs; i++) { 5088 if (!bdev_is_zoned(FDEV(i).bdev)) 5089 continue; 5090 5091 args.sbi = sbi; 5092 args.fdev = &FDEV(i); 5093 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 5094 check_zone_write_pointer_cb, &args); 5095 if (ret < 0) 5096 return ret; 5097 } 5098 5099 return 0; 5100 } 5101 5102 /* 5103 * Return the number of usable blocks in a segment. The number of blocks 5104 * returned is always equal to the number of blocks in a segment for 5105 * segments fully contained within a sequential zone capacity or a 5106 * conventional zone. For segments partially contained in a sequential 5107 * zone capacity, the number of usable blocks up to the zone capacity 5108 * is returned. 0 is returned in all other cases. 5109 */ 5110 static inline unsigned int f2fs_usable_zone_blks_in_seg( 5111 struct f2fs_sb_info *sbi, unsigned int segno) 5112 { 5113 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr; 5114 unsigned int secno; 5115 5116 if (!sbi->unusable_blocks_per_sec) 5117 return sbi->blocks_per_seg; 5118 5119 secno = GET_SEC_FROM_SEG(sbi, segno); 5120 seg_start = START_BLOCK(sbi, segno); 5121 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5122 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi); 5123 5124 /* 5125 * If segment starts before zone capacity and spans beyond 5126 * zone capacity, then usable blocks are from seg start to 5127 * zone capacity. If the segment starts after the zone capacity, 5128 * then there are no usable blocks. 5129 */ 5130 if (seg_start >= sec_cap_blkaddr) 5131 return 0; 5132 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr) 5133 return sec_cap_blkaddr - seg_start; 5134 5135 return sbi->blocks_per_seg; 5136 } 5137 #else 5138 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5139 { 5140 return 0; 5141 } 5142 5143 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5144 { 5145 return 0; 5146 } 5147 5148 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi, 5149 unsigned int segno) 5150 { 5151 return 0; 5152 } 5153 5154 #endif 5155 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi, 5156 unsigned int segno) 5157 { 5158 if (f2fs_sb_has_blkzoned(sbi)) 5159 return f2fs_usable_zone_blks_in_seg(sbi, segno); 5160 5161 return sbi->blocks_per_seg; 5162 } 5163 5164 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi, 5165 unsigned int segno) 5166 { 5167 if (f2fs_sb_has_blkzoned(sbi)) 5168 return CAP_SEGS_PER_SEC(sbi); 5169 5170 return sbi->segs_per_sec; 5171 } 5172 5173 /* 5174 * Update min, max modified time for cost-benefit GC algorithm 5175 */ 5176 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 5177 { 5178 struct sit_info *sit_i = SIT_I(sbi); 5179 unsigned int segno; 5180 5181 down_write(&sit_i->sentry_lock); 5182 5183 sit_i->min_mtime = ULLONG_MAX; 5184 5185 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 5186 unsigned int i; 5187 unsigned long long mtime = 0; 5188 5189 for (i = 0; i < sbi->segs_per_sec; i++) 5190 mtime += get_seg_entry(sbi, segno + i)->mtime; 5191 5192 mtime = div_u64(mtime, sbi->segs_per_sec); 5193 5194 if (sit_i->min_mtime > mtime) 5195 sit_i->min_mtime = mtime; 5196 } 5197 sit_i->max_mtime = get_mtime(sbi, false); 5198 sit_i->dirty_max_mtime = 0; 5199 up_write(&sit_i->sentry_lock); 5200 } 5201 5202 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 5203 { 5204 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 5205 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 5206 struct f2fs_sm_info *sm_info; 5207 int err; 5208 5209 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 5210 if (!sm_info) 5211 return -ENOMEM; 5212 5213 /* init sm info */ 5214 sbi->sm_info = sm_info; 5215 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 5216 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 5217 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 5218 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 5219 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 5220 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 5221 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 5222 sm_info->rec_prefree_segments = sm_info->main_segments * 5223 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 5224 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 5225 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 5226 5227 if (!f2fs_lfs_mode(sbi)) 5228 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC); 5229 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 5230 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 5231 sm_info->min_seq_blocks = sbi->blocks_per_seg; 5232 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 5233 sm_info->min_ssr_sections = reserved_sections(sbi); 5234 5235 INIT_LIST_HEAD(&sm_info->sit_entry_set); 5236 5237 init_f2fs_rwsem(&sm_info->curseg_lock); 5238 5239 err = f2fs_create_flush_cmd_control(sbi); 5240 if (err) 5241 return err; 5242 5243 err = create_discard_cmd_control(sbi); 5244 if (err) 5245 return err; 5246 5247 err = build_sit_info(sbi); 5248 if (err) 5249 return err; 5250 err = build_free_segmap(sbi); 5251 if (err) 5252 return err; 5253 err = build_curseg(sbi); 5254 if (err) 5255 return err; 5256 5257 /* reinit free segmap based on SIT */ 5258 err = build_sit_entries(sbi); 5259 if (err) 5260 return err; 5261 5262 init_free_segmap(sbi); 5263 err = build_dirty_segmap(sbi); 5264 if (err) 5265 return err; 5266 5267 err = sanity_check_curseg(sbi); 5268 if (err) 5269 return err; 5270 5271 init_min_max_mtime(sbi); 5272 return 0; 5273 } 5274 5275 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 5276 enum dirty_type dirty_type) 5277 { 5278 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5279 5280 mutex_lock(&dirty_i->seglist_lock); 5281 kvfree(dirty_i->dirty_segmap[dirty_type]); 5282 dirty_i->nr_dirty[dirty_type] = 0; 5283 mutex_unlock(&dirty_i->seglist_lock); 5284 } 5285 5286 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 5287 { 5288 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5289 5290 kvfree(dirty_i->pinned_secmap); 5291 kvfree(dirty_i->victim_secmap); 5292 } 5293 5294 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 5295 { 5296 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5297 int i; 5298 5299 if (!dirty_i) 5300 return; 5301 5302 /* discard pre-free/dirty segments list */ 5303 for (i = 0; i < NR_DIRTY_TYPE; i++) 5304 discard_dirty_segmap(sbi, i); 5305 5306 if (__is_large_section(sbi)) { 5307 mutex_lock(&dirty_i->seglist_lock); 5308 kvfree(dirty_i->dirty_secmap); 5309 mutex_unlock(&dirty_i->seglist_lock); 5310 } 5311 5312 destroy_victim_secmap(sbi); 5313 SM_I(sbi)->dirty_info = NULL; 5314 kfree(dirty_i); 5315 } 5316 5317 static void destroy_curseg(struct f2fs_sb_info *sbi) 5318 { 5319 struct curseg_info *array = SM_I(sbi)->curseg_array; 5320 int i; 5321 5322 if (!array) 5323 return; 5324 SM_I(sbi)->curseg_array = NULL; 5325 for (i = 0; i < NR_CURSEG_TYPE; i++) { 5326 kfree(array[i].sum_blk); 5327 kfree(array[i].journal); 5328 } 5329 kfree(array); 5330 } 5331 5332 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 5333 { 5334 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 5335 5336 if (!free_i) 5337 return; 5338 SM_I(sbi)->free_info = NULL; 5339 kvfree(free_i->free_segmap); 5340 kvfree(free_i->free_secmap); 5341 kfree(free_i); 5342 } 5343 5344 static void destroy_sit_info(struct f2fs_sb_info *sbi) 5345 { 5346 struct sit_info *sit_i = SIT_I(sbi); 5347 5348 if (!sit_i) 5349 return; 5350 5351 if (sit_i->sentries) 5352 kvfree(sit_i->bitmap); 5353 kfree(sit_i->tmp_map); 5354 5355 kvfree(sit_i->sentries); 5356 kvfree(sit_i->sec_entries); 5357 kvfree(sit_i->dirty_sentries_bitmap); 5358 5359 SM_I(sbi)->sit_info = NULL; 5360 kvfree(sit_i->sit_bitmap); 5361 #ifdef CONFIG_F2FS_CHECK_FS 5362 kvfree(sit_i->sit_bitmap_mir); 5363 kvfree(sit_i->invalid_segmap); 5364 #endif 5365 kfree(sit_i); 5366 } 5367 5368 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 5369 { 5370 struct f2fs_sm_info *sm_info = SM_I(sbi); 5371 5372 if (!sm_info) 5373 return; 5374 f2fs_destroy_flush_cmd_control(sbi, true); 5375 destroy_discard_cmd_control(sbi); 5376 destroy_dirty_segmap(sbi); 5377 destroy_curseg(sbi); 5378 destroy_free_segmap(sbi); 5379 destroy_sit_info(sbi); 5380 sbi->sm_info = NULL; 5381 kfree(sm_info); 5382 } 5383 5384 int __init f2fs_create_segment_manager_caches(void) 5385 { 5386 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 5387 sizeof(struct discard_entry)); 5388 if (!discard_entry_slab) 5389 goto fail; 5390 5391 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 5392 sizeof(struct discard_cmd)); 5393 if (!discard_cmd_slab) 5394 goto destroy_discard_entry; 5395 5396 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 5397 sizeof(struct sit_entry_set)); 5398 if (!sit_entry_set_slab) 5399 goto destroy_discard_cmd; 5400 5401 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry", 5402 sizeof(struct revoke_entry)); 5403 if (!revoke_entry_slab) 5404 goto destroy_sit_entry_set; 5405 return 0; 5406 5407 destroy_sit_entry_set: 5408 kmem_cache_destroy(sit_entry_set_slab); 5409 destroy_discard_cmd: 5410 kmem_cache_destroy(discard_cmd_slab); 5411 destroy_discard_entry: 5412 kmem_cache_destroy(discard_entry_slab); 5413 fail: 5414 return -ENOMEM; 5415 } 5416 5417 void f2fs_destroy_segment_manager_caches(void) 5418 { 5419 kmem_cache_destroy(sit_entry_set_slab); 5420 kmem_cache_destroy(discard_cmd_slab); 5421 kmem_cache_destroy(discard_entry_slab); 5422 kmem_cache_destroy(revoke_entry_slab); 5423 } 5424