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 /* avoid stale dirty inode during eviction */ 209 sync_inode_metadata(inode, 0); 210 } 211 212 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index, 213 block_t new_addr, block_t *old_addr, bool recover) 214 { 215 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 216 struct dnode_of_data dn; 217 struct node_info ni; 218 int err; 219 220 retry: 221 set_new_dnode(&dn, inode, NULL, NULL, 0); 222 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE); 223 if (err) { 224 if (err == -ENOMEM) { 225 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 226 goto retry; 227 } 228 return err; 229 } 230 231 err = f2fs_get_node_info(sbi, dn.nid, &ni, false); 232 if (err) { 233 f2fs_put_dnode(&dn); 234 return err; 235 } 236 237 if (recover) { 238 /* dn.data_blkaddr is always valid */ 239 if (!__is_valid_data_blkaddr(new_addr)) { 240 if (new_addr == NULL_ADDR) 241 dec_valid_block_count(sbi, inode, 1); 242 f2fs_invalidate_blocks(sbi, dn.data_blkaddr); 243 f2fs_update_data_blkaddr(&dn, new_addr); 244 } else { 245 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 246 new_addr, ni.version, true, true); 247 } 248 } else { 249 blkcnt_t count = 1; 250 251 err = inc_valid_block_count(sbi, inode, &count); 252 if (err) { 253 f2fs_put_dnode(&dn); 254 return err; 255 } 256 257 *old_addr = dn.data_blkaddr; 258 f2fs_truncate_data_blocks_range(&dn, 1); 259 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count); 260 261 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr, 262 ni.version, true, false); 263 } 264 265 f2fs_put_dnode(&dn); 266 267 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode, 268 index, old_addr ? *old_addr : 0, new_addr, recover); 269 return 0; 270 } 271 272 static void __complete_revoke_list(struct inode *inode, struct list_head *head, 273 bool revoke) 274 { 275 struct revoke_entry *cur, *tmp; 276 pgoff_t start_index = 0; 277 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE); 278 279 list_for_each_entry_safe(cur, tmp, head, list) { 280 if (revoke) { 281 __replace_atomic_write_block(inode, cur->index, 282 cur->old_addr, NULL, true); 283 } else if (truncate) { 284 f2fs_truncate_hole(inode, start_index, cur->index); 285 start_index = cur->index + 1; 286 } 287 288 list_del(&cur->list); 289 kmem_cache_free(revoke_entry_slab, cur); 290 } 291 292 if (!revoke && truncate) 293 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false); 294 } 295 296 static int __f2fs_commit_atomic_write(struct inode *inode) 297 { 298 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 299 struct f2fs_inode_info *fi = F2FS_I(inode); 300 struct inode *cow_inode = fi->cow_inode; 301 struct revoke_entry *new; 302 struct list_head revoke_list; 303 block_t blkaddr; 304 struct dnode_of_data dn; 305 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 306 pgoff_t off = 0, blen, index; 307 int ret = 0, i; 308 309 INIT_LIST_HEAD(&revoke_list); 310 311 while (len) { 312 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len); 313 314 set_new_dnode(&dn, cow_inode, NULL, NULL, 0); 315 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA); 316 if (ret && ret != -ENOENT) { 317 goto out; 318 } else if (ret == -ENOENT) { 319 ret = 0; 320 if (dn.max_level == 0) 321 goto out; 322 goto next; 323 } 324 325 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode), 326 len); 327 index = off; 328 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) { 329 blkaddr = f2fs_data_blkaddr(&dn); 330 331 if (!__is_valid_data_blkaddr(blkaddr)) { 332 continue; 333 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 334 DATA_GENERIC_ENHANCE)) { 335 f2fs_put_dnode(&dn); 336 ret = -EFSCORRUPTED; 337 f2fs_handle_error(sbi, 338 ERROR_INVALID_BLKADDR); 339 goto out; 340 } 341 342 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS, 343 true, NULL); 344 345 ret = __replace_atomic_write_block(inode, index, blkaddr, 346 &new->old_addr, false); 347 if (ret) { 348 f2fs_put_dnode(&dn); 349 kmem_cache_free(revoke_entry_slab, new); 350 goto out; 351 } 352 353 f2fs_update_data_blkaddr(&dn, NULL_ADDR); 354 new->index = index; 355 list_add_tail(&new->list, &revoke_list); 356 } 357 f2fs_put_dnode(&dn); 358 next: 359 off += blen; 360 len -= blen; 361 } 362 363 out: 364 if (ret) { 365 sbi->revoked_atomic_block += fi->atomic_write_cnt; 366 } else { 367 sbi->committed_atomic_block += fi->atomic_write_cnt; 368 set_inode_flag(inode, FI_ATOMIC_COMMITTED); 369 } 370 371 __complete_revoke_list(inode, &revoke_list, ret ? true : false); 372 373 return ret; 374 } 375 376 int f2fs_commit_atomic_write(struct inode *inode) 377 { 378 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 379 struct f2fs_inode_info *fi = F2FS_I(inode); 380 int err; 381 382 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 383 if (err) 384 return err; 385 386 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 387 f2fs_lock_op(sbi); 388 389 err = __f2fs_commit_atomic_write(inode); 390 391 f2fs_unlock_op(sbi); 392 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 393 394 return err; 395 } 396 397 /* 398 * This function balances dirty node and dentry pages. 399 * In addition, it controls garbage collection. 400 */ 401 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 402 { 403 if (time_to_inject(sbi, FAULT_CHECKPOINT)) 404 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT); 405 406 /* balance_fs_bg is able to be pending */ 407 if (need && excess_cached_nats(sbi)) 408 f2fs_balance_fs_bg(sbi, false); 409 410 if (!f2fs_is_checkpoint_ready(sbi)) 411 return; 412 413 /* 414 * We should do GC or end up with checkpoint, if there are so many dirty 415 * dir/node pages without enough free segments. 416 */ 417 if (has_enough_free_secs(sbi, 0, 0)) 418 return; 419 420 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread && 421 sbi->gc_thread->f2fs_gc_task) { 422 DEFINE_WAIT(wait); 423 424 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait, 425 TASK_UNINTERRUPTIBLE); 426 wake_up(&sbi->gc_thread->gc_wait_queue_head); 427 io_schedule(); 428 finish_wait(&sbi->gc_thread->fggc_wq, &wait); 429 } else { 430 struct f2fs_gc_control gc_control = { 431 .victim_segno = NULL_SEGNO, 432 .init_gc_type = BG_GC, 433 .no_bg_gc = true, 434 .should_migrate_blocks = false, 435 .err_gc_skipped = false, 436 .nr_free_secs = 1 }; 437 f2fs_down_write(&sbi->gc_lock); 438 stat_inc_gc_call_count(sbi, FOREGROUND); 439 f2fs_gc(sbi, &gc_control); 440 } 441 } 442 443 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi) 444 { 445 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2; 446 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS); 447 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA); 448 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES); 449 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META); 450 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA); 451 unsigned int threshold = sbi->blocks_per_seg * factor * 452 DEFAULT_DIRTY_THRESHOLD; 453 unsigned int global_threshold = threshold * 3 / 2; 454 455 if (dents >= threshold || qdata >= threshold || 456 nodes >= threshold || meta >= threshold || 457 imeta >= threshold) 458 return true; 459 return dents + qdata + nodes + meta + imeta > global_threshold; 460 } 461 462 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 463 { 464 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 465 return; 466 467 /* try to shrink extent cache when there is no enough memory */ 468 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE)) 469 f2fs_shrink_read_extent_tree(sbi, 470 READ_EXTENT_CACHE_SHRINK_NUMBER); 471 472 /* try to shrink age extent cache when there is no enough memory */ 473 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE)) 474 f2fs_shrink_age_extent_tree(sbi, 475 AGE_EXTENT_CACHE_SHRINK_NUMBER); 476 477 /* check the # of cached NAT entries */ 478 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 479 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 480 481 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 482 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 483 else 484 f2fs_build_free_nids(sbi, false, false); 485 486 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) || 487 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi)) 488 goto do_sync; 489 490 /* there is background inflight IO or foreground operation recently */ 491 if (is_inflight_io(sbi, REQ_TIME) || 492 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem))) 493 return; 494 495 /* exceed periodical checkpoint timeout threshold */ 496 if (f2fs_time_over(sbi, CP_TIME)) 497 goto do_sync; 498 499 /* checkpoint is the only way to shrink partial cached entries */ 500 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) && 501 f2fs_available_free_memory(sbi, INO_ENTRIES)) 502 return; 503 504 do_sync: 505 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 506 struct blk_plug plug; 507 508 mutex_lock(&sbi->flush_lock); 509 510 blk_start_plug(&plug); 511 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false); 512 blk_finish_plug(&plug); 513 514 mutex_unlock(&sbi->flush_lock); 515 } 516 stat_inc_cp_call_count(sbi, BACKGROUND); 517 f2fs_sync_fs(sbi->sb, 1); 518 } 519 520 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 521 struct block_device *bdev) 522 { 523 int ret = blkdev_issue_flush(bdev); 524 525 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 526 test_opt(sbi, FLUSH_MERGE), ret); 527 if (!ret) 528 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0); 529 return ret; 530 } 531 532 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 533 { 534 int ret = 0; 535 int i; 536 537 if (!f2fs_is_multi_device(sbi)) 538 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 539 540 for (i = 0; i < sbi->s_ndevs; i++) { 541 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 542 continue; 543 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 544 if (ret) 545 break; 546 } 547 return ret; 548 } 549 550 static int issue_flush_thread(void *data) 551 { 552 struct f2fs_sb_info *sbi = data; 553 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 554 wait_queue_head_t *q = &fcc->flush_wait_queue; 555 repeat: 556 if (kthread_should_stop()) 557 return 0; 558 559 if (!llist_empty(&fcc->issue_list)) { 560 struct flush_cmd *cmd, *next; 561 int ret; 562 563 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 564 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 565 566 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 567 568 ret = submit_flush_wait(sbi, cmd->ino); 569 atomic_inc(&fcc->issued_flush); 570 571 llist_for_each_entry_safe(cmd, next, 572 fcc->dispatch_list, llnode) { 573 cmd->ret = ret; 574 complete(&cmd->wait); 575 } 576 fcc->dispatch_list = NULL; 577 } 578 579 wait_event_interruptible(*q, 580 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 581 goto repeat; 582 } 583 584 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 585 { 586 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 587 struct flush_cmd cmd; 588 int ret; 589 590 if (test_opt(sbi, NOBARRIER)) 591 return 0; 592 593 if (!test_opt(sbi, FLUSH_MERGE)) { 594 atomic_inc(&fcc->queued_flush); 595 ret = submit_flush_wait(sbi, ino); 596 atomic_dec(&fcc->queued_flush); 597 atomic_inc(&fcc->issued_flush); 598 return ret; 599 } 600 601 if (atomic_inc_return(&fcc->queued_flush) == 1 || 602 f2fs_is_multi_device(sbi)) { 603 ret = submit_flush_wait(sbi, ino); 604 atomic_dec(&fcc->queued_flush); 605 606 atomic_inc(&fcc->issued_flush); 607 return ret; 608 } 609 610 cmd.ino = ino; 611 init_completion(&cmd.wait); 612 613 llist_add(&cmd.llnode, &fcc->issue_list); 614 615 /* 616 * update issue_list before we wake up issue_flush thread, this 617 * smp_mb() pairs with another barrier in ___wait_event(), see 618 * more details in comments of waitqueue_active(). 619 */ 620 smp_mb(); 621 622 if (waitqueue_active(&fcc->flush_wait_queue)) 623 wake_up(&fcc->flush_wait_queue); 624 625 if (fcc->f2fs_issue_flush) { 626 wait_for_completion(&cmd.wait); 627 atomic_dec(&fcc->queued_flush); 628 } else { 629 struct llist_node *list; 630 631 list = llist_del_all(&fcc->issue_list); 632 if (!list) { 633 wait_for_completion(&cmd.wait); 634 atomic_dec(&fcc->queued_flush); 635 } else { 636 struct flush_cmd *tmp, *next; 637 638 ret = submit_flush_wait(sbi, ino); 639 640 llist_for_each_entry_safe(tmp, next, list, llnode) { 641 if (tmp == &cmd) { 642 cmd.ret = ret; 643 atomic_dec(&fcc->queued_flush); 644 continue; 645 } 646 tmp->ret = ret; 647 complete(&tmp->wait); 648 } 649 } 650 } 651 652 return cmd.ret; 653 } 654 655 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 656 { 657 dev_t dev = sbi->sb->s_bdev->bd_dev; 658 struct flush_cmd_control *fcc; 659 660 if (SM_I(sbi)->fcc_info) { 661 fcc = SM_I(sbi)->fcc_info; 662 if (fcc->f2fs_issue_flush) 663 return 0; 664 goto init_thread; 665 } 666 667 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 668 if (!fcc) 669 return -ENOMEM; 670 atomic_set(&fcc->issued_flush, 0); 671 atomic_set(&fcc->queued_flush, 0); 672 init_waitqueue_head(&fcc->flush_wait_queue); 673 init_llist_head(&fcc->issue_list); 674 SM_I(sbi)->fcc_info = fcc; 675 if (!test_opt(sbi, FLUSH_MERGE)) 676 return 0; 677 678 init_thread: 679 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 680 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 681 if (IS_ERR(fcc->f2fs_issue_flush)) { 682 int err = PTR_ERR(fcc->f2fs_issue_flush); 683 684 fcc->f2fs_issue_flush = NULL; 685 return err; 686 } 687 688 return 0; 689 } 690 691 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 692 { 693 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 694 695 if (fcc && fcc->f2fs_issue_flush) { 696 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 697 698 fcc->f2fs_issue_flush = NULL; 699 kthread_stop(flush_thread); 700 } 701 if (free) { 702 kfree(fcc); 703 SM_I(sbi)->fcc_info = NULL; 704 } 705 } 706 707 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 708 { 709 int ret = 0, i; 710 711 if (!f2fs_is_multi_device(sbi)) 712 return 0; 713 714 if (test_opt(sbi, NOBARRIER)) 715 return 0; 716 717 for (i = 1; i < sbi->s_ndevs; i++) { 718 int count = DEFAULT_RETRY_IO_COUNT; 719 720 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 721 continue; 722 723 do { 724 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 725 if (ret) 726 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 727 } while (ret && --count); 728 729 if (ret) { 730 f2fs_stop_checkpoint(sbi, false, 731 STOP_CP_REASON_FLUSH_FAIL); 732 break; 733 } 734 735 spin_lock(&sbi->dev_lock); 736 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 737 spin_unlock(&sbi->dev_lock); 738 } 739 740 return ret; 741 } 742 743 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 744 enum dirty_type dirty_type) 745 { 746 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 747 748 /* need not be added */ 749 if (IS_CURSEG(sbi, segno)) 750 return; 751 752 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 753 dirty_i->nr_dirty[dirty_type]++; 754 755 if (dirty_type == DIRTY) { 756 struct seg_entry *sentry = get_seg_entry(sbi, segno); 757 enum dirty_type t = sentry->type; 758 759 if (unlikely(t >= DIRTY)) { 760 f2fs_bug_on(sbi, 1); 761 return; 762 } 763 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 764 dirty_i->nr_dirty[t]++; 765 766 if (__is_large_section(sbi)) { 767 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 768 block_t valid_blocks = 769 get_valid_blocks(sbi, segno, true); 770 771 f2fs_bug_on(sbi, unlikely(!valid_blocks || 772 valid_blocks == CAP_BLKS_PER_SEC(sbi))); 773 774 if (!IS_CURSEC(sbi, secno)) 775 set_bit(secno, dirty_i->dirty_secmap); 776 } 777 } 778 } 779 780 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 781 enum dirty_type dirty_type) 782 { 783 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 784 block_t valid_blocks; 785 786 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 787 dirty_i->nr_dirty[dirty_type]--; 788 789 if (dirty_type == DIRTY) { 790 struct seg_entry *sentry = get_seg_entry(sbi, segno); 791 enum dirty_type t = sentry->type; 792 793 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 794 dirty_i->nr_dirty[t]--; 795 796 valid_blocks = get_valid_blocks(sbi, segno, true); 797 if (valid_blocks == 0) { 798 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 799 dirty_i->victim_secmap); 800 #ifdef CONFIG_F2FS_CHECK_FS 801 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 802 #endif 803 } 804 if (__is_large_section(sbi)) { 805 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 806 807 if (!valid_blocks || 808 valid_blocks == CAP_BLKS_PER_SEC(sbi)) { 809 clear_bit(secno, dirty_i->dirty_secmap); 810 return; 811 } 812 813 if (!IS_CURSEC(sbi, secno)) 814 set_bit(secno, dirty_i->dirty_secmap); 815 } 816 } 817 } 818 819 /* 820 * Should not occur error such as -ENOMEM. 821 * Adding dirty entry into seglist is not critical operation. 822 * If a given segment is one of current working segments, it won't be added. 823 */ 824 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 825 { 826 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 827 unsigned short valid_blocks, ckpt_valid_blocks; 828 unsigned int usable_blocks; 829 830 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 831 return; 832 833 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno); 834 mutex_lock(&dirty_i->seglist_lock); 835 836 valid_blocks = get_valid_blocks(sbi, segno, false); 837 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false); 838 839 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 840 ckpt_valid_blocks == usable_blocks)) { 841 __locate_dirty_segment(sbi, segno, PRE); 842 __remove_dirty_segment(sbi, segno, DIRTY); 843 } else if (valid_blocks < usable_blocks) { 844 __locate_dirty_segment(sbi, segno, DIRTY); 845 } else { 846 /* Recovery routine with SSR needs this */ 847 __remove_dirty_segment(sbi, segno, DIRTY); 848 } 849 850 mutex_unlock(&dirty_i->seglist_lock); 851 } 852 853 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 854 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 855 { 856 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 857 unsigned int segno; 858 859 mutex_lock(&dirty_i->seglist_lock); 860 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 861 if (get_valid_blocks(sbi, segno, false)) 862 continue; 863 if (IS_CURSEG(sbi, segno)) 864 continue; 865 __locate_dirty_segment(sbi, segno, PRE); 866 __remove_dirty_segment(sbi, segno, DIRTY); 867 } 868 mutex_unlock(&dirty_i->seglist_lock); 869 } 870 871 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 872 { 873 int ovp_hole_segs = 874 (overprovision_segments(sbi) - reserved_segments(sbi)); 875 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg; 876 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 877 block_t holes[2] = {0, 0}; /* DATA and NODE */ 878 block_t unusable; 879 struct seg_entry *se; 880 unsigned int segno; 881 882 mutex_lock(&dirty_i->seglist_lock); 883 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 884 se = get_seg_entry(sbi, segno); 885 if (IS_NODESEG(se->type)) 886 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) - 887 se->valid_blocks; 888 else 889 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) - 890 se->valid_blocks; 891 } 892 mutex_unlock(&dirty_i->seglist_lock); 893 894 unusable = max(holes[DATA], holes[NODE]); 895 if (unusable > ovp_holes) 896 return unusable - ovp_holes; 897 return 0; 898 } 899 900 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 901 { 902 int ovp_hole_segs = 903 (overprovision_segments(sbi) - reserved_segments(sbi)); 904 if (unusable > F2FS_OPTION(sbi).unusable_cap) 905 return -EAGAIN; 906 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 907 dirty_segments(sbi) > ovp_hole_segs) 908 return -EAGAIN; 909 return 0; 910 } 911 912 /* This is only used by SBI_CP_DISABLED */ 913 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 914 { 915 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 916 unsigned int segno = 0; 917 918 mutex_lock(&dirty_i->seglist_lock); 919 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 920 if (get_valid_blocks(sbi, segno, false)) 921 continue; 922 if (get_ckpt_valid_blocks(sbi, segno, false)) 923 continue; 924 mutex_unlock(&dirty_i->seglist_lock); 925 return segno; 926 } 927 mutex_unlock(&dirty_i->seglist_lock); 928 return NULL_SEGNO; 929 } 930 931 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 932 struct block_device *bdev, block_t lstart, 933 block_t start, block_t len) 934 { 935 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 936 struct list_head *pend_list; 937 struct discard_cmd *dc; 938 939 f2fs_bug_on(sbi, !len); 940 941 pend_list = &dcc->pend_list[plist_idx(len)]; 942 943 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL); 944 INIT_LIST_HEAD(&dc->list); 945 dc->bdev = bdev; 946 dc->di.lstart = lstart; 947 dc->di.start = start; 948 dc->di.len = len; 949 dc->ref = 0; 950 dc->state = D_PREP; 951 dc->queued = 0; 952 dc->error = 0; 953 init_completion(&dc->wait); 954 list_add_tail(&dc->list, pend_list); 955 spin_lock_init(&dc->lock); 956 dc->bio_ref = 0; 957 atomic_inc(&dcc->discard_cmd_cnt); 958 dcc->undiscard_blks += len; 959 960 return dc; 961 } 962 963 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi) 964 { 965 #ifdef CONFIG_F2FS_CHECK_FS 966 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 967 struct rb_node *cur = rb_first_cached(&dcc->root), *next; 968 struct discard_cmd *cur_dc, *next_dc; 969 970 while (cur) { 971 next = rb_next(cur); 972 if (!next) 973 return true; 974 975 cur_dc = rb_entry(cur, struct discard_cmd, rb_node); 976 next_dc = rb_entry(next, struct discard_cmd, rb_node); 977 978 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) { 979 f2fs_info(sbi, "broken discard_rbtree, " 980 "cur(%u, %u) next(%u, %u)", 981 cur_dc->di.lstart, cur_dc->di.len, 982 next_dc->di.lstart, next_dc->di.len); 983 return false; 984 } 985 cur = next; 986 } 987 #endif 988 return true; 989 } 990 991 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi, 992 block_t blkaddr) 993 { 994 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 995 struct rb_node *node = dcc->root.rb_root.rb_node; 996 struct discard_cmd *dc; 997 998 while (node) { 999 dc = rb_entry(node, struct discard_cmd, rb_node); 1000 1001 if (blkaddr < dc->di.lstart) 1002 node = node->rb_left; 1003 else if (blkaddr >= dc->di.lstart + dc->di.len) 1004 node = node->rb_right; 1005 else 1006 return dc; 1007 } 1008 return NULL; 1009 } 1010 1011 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root, 1012 block_t blkaddr, 1013 struct discard_cmd **prev_entry, 1014 struct discard_cmd **next_entry, 1015 struct rb_node ***insert_p, 1016 struct rb_node **insert_parent) 1017 { 1018 struct rb_node **pnode = &root->rb_root.rb_node; 1019 struct rb_node *parent = NULL, *tmp_node; 1020 struct discard_cmd *dc; 1021 1022 *insert_p = NULL; 1023 *insert_parent = NULL; 1024 *prev_entry = NULL; 1025 *next_entry = NULL; 1026 1027 if (RB_EMPTY_ROOT(&root->rb_root)) 1028 return NULL; 1029 1030 while (*pnode) { 1031 parent = *pnode; 1032 dc = rb_entry(*pnode, struct discard_cmd, rb_node); 1033 1034 if (blkaddr < dc->di.lstart) 1035 pnode = &(*pnode)->rb_left; 1036 else if (blkaddr >= dc->di.lstart + dc->di.len) 1037 pnode = &(*pnode)->rb_right; 1038 else 1039 goto lookup_neighbors; 1040 } 1041 1042 *insert_p = pnode; 1043 *insert_parent = parent; 1044 1045 dc = rb_entry(parent, struct discard_cmd, rb_node); 1046 tmp_node = parent; 1047 if (parent && blkaddr > dc->di.lstart) 1048 tmp_node = rb_next(parent); 1049 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1050 1051 tmp_node = parent; 1052 if (parent && blkaddr < dc->di.lstart) 1053 tmp_node = rb_prev(parent); 1054 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1055 return NULL; 1056 1057 lookup_neighbors: 1058 /* lookup prev node for merging backward later */ 1059 tmp_node = rb_prev(&dc->rb_node); 1060 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1061 1062 /* lookup next node for merging frontward later */ 1063 tmp_node = rb_next(&dc->rb_node); 1064 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1065 return dc; 1066 } 1067 1068 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 1069 struct discard_cmd *dc) 1070 { 1071 if (dc->state == D_DONE) 1072 atomic_sub(dc->queued, &dcc->queued_discard); 1073 1074 list_del(&dc->list); 1075 rb_erase_cached(&dc->rb_node, &dcc->root); 1076 dcc->undiscard_blks -= dc->di.len; 1077 1078 kmem_cache_free(discard_cmd_slab, dc); 1079 1080 atomic_dec(&dcc->discard_cmd_cnt); 1081 } 1082 1083 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 1084 struct discard_cmd *dc) 1085 { 1086 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1087 unsigned long flags; 1088 1089 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len); 1090 1091 spin_lock_irqsave(&dc->lock, flags); 1092 if (dc->bio_ref) { 1093 spin_unlock_irqrestore(&dc->lock, flags); 1094 return; 1095 } 1096 spin_unlock_irqrestore(&dc->lock, flags); 1097 1098 f2fs_bug_on(sbi, dc->ref); 1099 1100 if (dc->error == -EOPNOTSUPP) 1101 dc->error = 0; 1102 1103 if (dc->error) 1104 printk_ratelimited( 1105 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d", 1106 KERN_INFO, sbi->sb->s_id, 1107 dc->di.lstart, dc->di.start, dc->di.len, dc->error); 1108 __detach_discard_cmd(dcc, dc); 1109 } 1110 1111 static void f2fs_submit_discard_endio(struct bio *bio) 1112 { 1113 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1114 unsigned long flags; 1115 1116 spin_lock_irqsave(&dc->lock, flags); 1117 if (!dc->error) 1118 dc->error = blk_status_to_errno(bio->bi_status); 1119 dc->bio_ref--; 1120 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1121 dc->state = D_DONE; 1122 complete_all(&dc->wait); 1123 } 1124 spin_unlock_irqrestore(&dc->lock, flags); 1125 bio_put(bio); 1126 } 1127 1128 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1129 block_t start, block_t end) 1130 { 1131 #ifdef CONFIG_F2FS_CHECK_FS 1132 struct seg_entry *sentry; 1133 unsigned int segno; 1134 block_t blk = start; 1135 unsigned long offset, size, max_blocks = sbi->blocks_per_seg; 1136 unsigned long *map; 1137 1138 while (blk < end) { 1139 segno = GET_SEGNO(sbi, blk); 1140 sentry = get_seg_entry(sbi, segno); 1141 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1142 1143 if (end < START_BLOCK(sbi, segno + 1)) 1144 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1145 else 1146 size = max_blocks; 1147 map = (unsigned long *)(sentry->cur_valid_map); 1148 offset = __find_rev_next_bit(map, size, offset); 1149 f2fs_bug_on(sbi, offset != size); 1150 blk = START_BLOCK(sbi, segno + 1); 1151 } 1152 #endif 1153 } 1154 1155 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1156 struct discard_policy *dpolicy, 1157 int discard_type, unsigned int granularity) 1158 { 1159 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1160 1161 /* common policy */ 1162 dpolicy->type = discard_type; 1163 dpolicy->sync = true; 1164 dpolicy->ordered = false; 1165 dpolicy->granularity = granularity; 1166 1167 dpolicy->max_requests = dcc->max_discard_request; 1168 dpolicy->io_aware_gran = dcc->discard_io_aware_gran; 1169 dpolicy->timeout = false; 1170 1171 if (discard_type == DPOLICY_BG) { 1172 dpolicy->min_interval = dcc->min_discard_issue_time; 1173 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1174 dpolicy->max_interval = dcc->max_discard_issue_time; 1175 if (dcc->discard_io_aware == DPOLICY_IO_AWARE_ENABLE) 1176 dpolicy->io_aware = true; 1177 else if (dcc->discard_io_aware == DPOLICY_IO_AWARE_DISABLE) 1178 dpolicy->io_aware = false; 1179 dpolicy->sync = false; 1180 dpolicy->ordered = true; 1181 if (utilization(sbi) > dcc->discard_urgent_util) { 1182 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1183 if (atomic_read(&dcc->discard_cmd_cnt)) 1184 dpolicy->max_interval = 1185 dcc->min_discard_issue_time; 1186 } 1187 } else if (discard_type == DPOLICY_FORCE) { 1188 dpolicy->min_interval = dcc->min_discard_issue_time; 1189 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1190 dpolicy->max_interval = dcc->max_discard_issue_time; 1191 dpolicy->io_aware = false; 1192 } else if (discard_type == DPOLICY_FSTRIM) { 1193 dpolicy->io_aware = false; 1194 } else if (discard_type == DPOLICY_UMOUNT) { 1195 dpolicy->io_aware = false; 1196 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1197 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1198 dpolicy->timeout = true; 1199 } 1200 } 1201 1202 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1203 struct block_device *bdev, block_t lstart, 1204 block_t start, block_t len); 1205 1206 #ifdef CONFIG_BLK_DEV_ZONED 1207 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi, 1208 struct discard_cmd *dc, blk_opf_t flag, 1209 struct list_head *wait_list, 1210 unsigned int *issued) 1211 { 1212 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1213 struct block_device *bdev = dc->bdev; 1214 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS); 1215 unsigned long flags; 1216 1217 trace_f2fs_issue_reset_zone(bdev, dc->di.start); 1218 1219 spin_lock_irqsave(&dc->lock, flags); 1220 dc->state = D_SUBMIT; 1221 dc->bio_ref++; 1222 spin_unlock_irqrestore(&dc->lock, flags); 1223 1224 if (issued) 1225 (*issued)++; 1226 1227 atomic_inc(&dcc->queued_discard); 1228 dc->queued++; 1229 list_move_tail(&dc->list, wait_list); 1230 1231 /* sanity check on discard range */ 1232 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len); 1233 1234 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start); 1235 bio->bi_private = dc; 1236 bio->bi_end_io = f2fs_submit_discard_endio; 1237 submit_bio(bio); 1238 1239 atomic_inc(&dcc->issued_discard); 1240 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE); 1241 } 1242 #endif 1243 1244 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1245 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1246 struct discard_policy *dpolicy, 1247 struct discard_cmd *dc, int *issued) 1248 { 1249 struct block_device *bdev = dc->bdev; 1250 unsigned int max_discard_blocks = 1251 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1252 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1253 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1254 &(dcc->fstrim_list) : &(dcc->wait_list); 1255 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0; 1256 block_t lstart, start, len, total_len; 1257 int err = 0; 1258 1259 if (dc->state != D_PREP) 1260 return 0; 1261 1262 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1263 return 0; 1264 1265 #ifdef CONFIG_BLK_DEV_ZONED 1266 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) { 1267 int devi = f2fs_bdev_index(sbi, bdev); 1268 1269 if (devi < 0) 1270 return -EINVAL; 1271 1272 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1273 __submit_zone_reset_cmd(sbi, dc, flag, 1274 wait_list, issued); 1275 return 0; 1276 } 1277 } 1278 #endif 1279 1280 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len); 1281 1282 lstart = dc->di.lstart; 1283 start = dc->di.start; 1284 len = dc->di.len; 1285 total_len = len; 1286 1287 dc->di.len = 0; 1288 1289 while (total_len && *issued < dpolicy->max_requests && !err) { 1290 struct bio *bio = NULL; 1291 unsigned long flags; 1292 bool last = true; 1293 1294 if (len > max_discard_blocks) { 1295 len = max_discard_blocks; 1296 last = false; 1297 } 1298 1299 (*issued)++; 1300 if (*issued == dpolicy->max_requests) 1301 last = true; 1302 1303 dc->di.len += len; 1304 1305 if (time_to_inject(sbi, FAULT_DISCARD)) { 1306 err = -EIO; 1307 } else { 1308 err = __blkdev_issue_discard(bdev, 1309 SECTOR_FROM_BLOCK(start), 1310 SECTOR_FROM_BLOCK(len), 1311 GFP_NOFS, &bio); 1312 } 1313 if (err) { 1314 spin_lock_irqsave(&dc->lock, flags); 1315 if (dc->state == D_PARTIAL) 1316 dc->state = D_SUBMIT; 1317 spin_unlock_irqrestore(&dc->lock, flags); 1318 1319 break; 1320 } 1321 1322 f2fs_bug_on(sbi, !bio); 1323 1324 /* 1325 * should keep before submission to avoid D_DONE 1326 * right away 1327 */ 1328 spin_lock_irqsave(&dc->lock, flags); 1329 if (last) 1330 dc->state = D_SUBMIT; 1331 else 1332 dc->state = D_PARTIAL; 1333 dc->bio_ref++; 1334 spin_unlock_irqrestore(&dc->lock, flags); 1335 1336 atomic_inc(&dcc->queued_discard); 1337 dc->queued++; 1338 list_move_tail(&dc->list, wait_list); 1339 1340 /* sanity check on discard range */ 1341 __check_sit_bitmap(sbi, lstart, lstart + len); 1342 1343 bio->bi_private = dc; 1344 bio->bi_end_io = f2fs_submit_discard_endio; 1345 bio->bi_opf |= flag; 1346 submit_bio(bio); 1347 1348 atomic_inc(&dcc->issued_discard); 1349 1350 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE); 1351 1352 lstart += len; 1353 start += len; 1354 total_len -= len; 1355 len = total_len; 1356 } 1357 1358 if (!err && len) { 1359 dcc->undiscard_blks -= len; 1360 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1361 } 1362 return err; 1363 } 1364 1365 static void __insert_discard_cmd(struct f2fs_sb_info *sbi, 1366 struct block_device *bdev, block_t lstart, 1367 block_t start, block_t len) 1368 { 1369 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1370 struct rb_node **p = &dcc->root.rb_root.rb_node; 1371 struct rb_node *parent = NULL; 1372 struct discard_cmd *dc; 1373 bool leftmost = true; 1374 1375 /* look up rb tree to find parent node */ 1376 while (*p) { 1377 parent = *p; 1378 dc = rb_entry(parent, struct discard_cmd, rb_node); 1379 1380 if (lstart < dc->di.lstart) { 1381 p = &(*p)->rb_left; 1382 } else if (lstart >= dc->di.lstart + dc->di.len) { 1383 p = &(*p)->rb_right; 1384 leftmost = false; 1385 } else { 1386 /* Let's skip to add, if exists */ 1387 return; 1388 } 1389 } 1390 1391 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 1392 1393 rb_link_node(&dc->rb_node, parent, p); 1394 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 1395 } 1396 1397 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1398 struct discard_cmd *dc) 1399 { 1400 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]); 1401 } 1402 1403 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1404 struct discard_cmd *dc, block_t blkaddr) 1405 { 1406 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1407 struct discard_info di = dc->di; 1408 bool modified = false; 1409 1410 if (dc->state == D_DONE || dc->di.len == 1) { 1411 __remove_discard_cmd(sbi, dc); 1412 return; 1413 } 1414 1415 dcc->undiscard_blks -= di.len; 1416 1417 if (blkaddr > di.lstart) { 1418 dc->di.len = blkaddr - dc->di.lstart; 1419 dcc->undiscard_blks += dc->di.len; 1420 __relocate_discard_cmd(dcc, dc); 1421 modified = true; 1422 } 1423 1424 if (blkaddr < di.lstart + di.len - 1) { 1425 if (modified) { 1426 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1, 1427 di.start + blkaddr + 1 - di.lstart, 1428 di.lstart + di.len - 1 - blkaddr); 1429 } else { 1430 dc->di.lstart++; 1431 dc->di.len--; 1432 dc->di.start++; 1433 dcc->undiscard_blks += dc->di.len; 1434 __relocate_discard_cmd(dcc, dc); 1435 } 1436 } 1437 } 1438 1439 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1440 struct block_device *bdev, block_t lstart, 1441 block_t start, block_t len) 1442 { 1443 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1444 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1445 struct discard_cmd *dc; 1446 struct discard_info di = {0}; 1447 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1448 unsigned int max_discard_blocks = 1449 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1450 block_t end = lstart + len; 1451 1452 dc = __lookup_discard_cmd_ret(&dcc->root, lstart, 1453 &prev_dc, &next_dc, &insert_p, &insert_parent); 1454 if (dc) 1455 prev_dc = dc; 1456 1457 if (!prev_dc) { 1458 di.lstart = lstart; 1459 di.len = next_dc ? next_dc->di.lstart - lstart : len; 1460 di.len = min(di.len, len); 1461 di.start = start; 1462 } 1463 1464 while (1) { 1465 struct rb_node *node; 1466 bool merged = false; 1467 struct discard_cmd *tdc = NULL; 1468 1469 if (prev_dc) { 1470 di.lstart = prev_dc->di.lstart + prev_dc->di.len; 1471 if (di.lstart < lstart) 1472 di.lstart = lstart; 1473 if (di.lstart >= end) 1474 break; 1475 1476 if (!next_dc || next_dc->di.lstart > end) 1477 di.len = end - di.lstart; 1478 else 1479 di.len = next_dc->di.lstart - di.lstart; 1480 di.start = start + di.lstart - lstart; 1481 } 1482 1483 if (!di.len) 1484 goto next; 1485 1486 if (prev_dc && prev_dc->state == D_PREP && 1487 prev_dc->bdev == bdev && 1488 __is_discard_back_mergeable(&di, &prev_dc->di, 1489 max_discard_blocks)) { 1490 prev_dc->di.len += di.len; 1491 dcc->undiscard_blks += di.len; 1492 __relocate_discard_cmd(dcc, prev_dc); 1493 di = prev_dc->di; 1494 tdc = prev_dc; 1495 merged = true; 1496 } 1497 1498 if (next_dc && next_dc->state == D_PREP && 1499 next_dc->bdev == bdev && 1500 __is_discard_front_mergeable(&di, &next_dc->di, 1501 max_discard_blocks)) { 1502 next_dc->di.lstart = di.lstart; 1503 next_dc->di.len += di.len; 1504 next_dc->di.start = di.start; 1505 dcc->undiscard_blks += di.len; 1506 __relocate_discard_cmd(dcc, next_dc); 1507 if (tdc) 1508 __remove_discard_cmd(sbi, tdc); 1509 merged = true; 1510 } 1511 1512 if (!merged) 1513 __insert_discard_cmd(sbi, bdev, 1514 di.lstart, di.start, di.len); 1515 next: 1516 prev_dc = next_dc; 1517 if (!prev_dc) 1518 break; 1519 1520 node = rb_next(&prev_dc->rb_node); 1521 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1522 } 1523 } 1524 1525 #ifdef CONFIG_BLK_DEV_ZONED 1526 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi, 1527 struct block_device *bdev, block_t blkstart, block_t lblkstart, 1528 block_t blklen) 1529 { 1530 trace_f2fs_queue_reset_zone(bdev, blkstart); 1531 1532 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1533 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen); 1534 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1535 } 1536 #endif 1537 1538 static void __queue_discard_cmd(struct f2fs_sb_info *sbi, 1539 struct block_device *bdev, block_t blkstart, block_t blklen) 1540 { 1541 block_t lblkstart = blkstart; 1542 1543 if (!f2fs_bdev_support_discard(bdev)) 1544 return; 1545 1546 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1547 1548 if (f2fs_is_multi_device(sbi)) { 1549 int devi = f2fs_target_device_index(sbi, blkstart); 1550 1551 blkstart -= FDEV(devi).start_blk; 1552 } 1553 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1554 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1555 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1556 } 1557 1558 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1559 struct discard_policy *dpolicy, int *issued) 1560 { 1561 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1562 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1563 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1564 struct discard_cmd *dc; 1565 struct blk_plug plug; 1566 bool io_interrupted = false; 1567 1568 mutex_lock(&dcc->cmd_lock); 1569 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos, 1570 &prev_dc, &next_dc, &insert_p, &insert_parent); 1571 if (!dc) 1572 dc = next_dc; 1573 1574 blk_start_plug(&plug); 1575 1576 while (dc) { 1577 struct rb_node *node; 1578 int err = 0; 1579 1580 if (dc->state != D_PREP) 1581 goto next; 1582 1583 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1584 io_interrupted = true; 1585 break; 1586 } 1587 1588 dcc->next_pos = dc->di.lstart + dc->di.len; 1589 err = __submit_discard_cmd(sbi, dpolicy, dc, issued); 1590 1591 if (*issued >= dpolicy->max_requests) 1592 break; 1593 next: 1594 node = rb_next(&dc->rb_node); 1595 if (err) 1596 __remove_discard_cmd(sbi, dc); 1597 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1598 } 1599 1600 blk_finish_plug(&plug); 1601 1602 if (!dc) 1603 dcc->next_pos = 0; 1604 1605 mutex_unlock(&dcc->cmd_lock); 1606 1607 if (!(*issued) && io_interrupted) 1608 *issued = -1; 1609 } 1610 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1611 struct discard_policy *dpolicy); 1612 1613 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1614 struct discard_policy *dpolicy) 1615 { 1616 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1617 struct list_head *pend_list; 1618 struct discard_cmd *dc, *tmp; 1619 struct blk_plug plug; 1620 int i, issued; 1621 bool io_interrupted = false; 1622 1623 if (dpolicy->timeout) 1624 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1625 1626 retry: 1627 issued = 0; 1628 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1629 if (dpolicy->timeout && 1630 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1631 break; 1632 1633 if (i + 1 < dpolicy->granularity) 1634 break; 1635 1636 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) { 1637 __issue_discard_cmd_orderly(sbi, dpolicy, &issued); 1638 return issued; 1639 } 1640 1641 pend_list = &dcc->pend_list[i]; 1642 1643 mutex_lock(&dcc->cmd_lock); 1644 if (list_empty(pend_list)) 1645 goto next; 1646 if (unlikely(dcc->rbtree_check)) 1647 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 1648 blk_start_plug(&plug); 1649 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1650 f2fs_bug_on(sbi, dc->state != D_PREP); 1651 1652 if (dpolicy->timeout && 1653 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1654 break; 1655 1656 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1657 !is_idle(sbi, DISCARD_TIME)) { 1658 io_interrupted = true; 1659 break; 1660 } 1661 1662 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1663 1664 if (issued >= dpolicy->max_requests) 1665 break; 1666 } 1667 blk_finish_plug(&plug); 1668 next: 1669 mutex_unlock(&dcc->cmd_lock); 1670 1671 if (issued >= dpolicy->max_requests || io_interrupted) 1672 break; 1673 } 1674 1675 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1676 __wait_all_discard_cmd(sbi, dpolicy); 1677 goto retry; 1678 } 1679 1680 if (!issued && io_interrupted) 1681 issued = -1; 1682 1683 return issued; 1684 } 1685 1686 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1687 { 1688 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1689 struct list_head *pend_list; 1690 struct discard_cmd *dc, *tmp; 1691 int i; 1692 bool dropped = false; 1693 1694 mutex_lock(&dcc->cmd_lock); 1695 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1696 pend_list = &dcc->pend_list[i]; 1697 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1698 f2fs_bug_on(sbi, dc->state != D_PREP); 1699 __remove_discard_cmd(sbi, dc); 1700 dropped = true; 1701 } 1702 } 1703 mutex_unlock(&dcc->cmd_lock); 1704 1705 return dropped; 1706 } 1707 1708 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1709 { 1710 __drop_discard_cmd(sbi); 1711 } 1712 1713 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1714 struct discard_cmd *dc) 1715 { 1716 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1717 unsigned int len = 0; 1718 1719 wait_for_completion_io(&dc->wait); 1720 mutex_lock(&dcc->cmd_lock); 1721 f2fs_bug_on(sbi, dc->state != D_DONE); 1722 dc->ref--; 1723 if (!dc->ref) { 1724 if (!dc->error) 1725 len = dc->di.len; 1726 __remove_discard_cmd(sbi, dc); 1727 } 1728 mutex_unlock(&dcc->cmd_lock); 1729 1730 return len; 1731 } 1732 1733 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1734 struct discard_policy *dpolicy, 1735 block_t start, block_t end) 1736 { 1737 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1738 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1739 &(dcc->fstrim_list) : &(dcc->wait_list); 1740 struct discard_cmd *dc = NULL, *iter, *tmp; 1741 unsigned int trimmed = 0; 1742 1743 next: 1744 dc = NULL; 1745 1746 mutex_lock(&dcc->cmd_lock); 1747 list_for_each_entry_safe(iter, tmp, wait_list, list) { 1748 if (iter->di.lstart + iter->di.len <= start || 1749 end <= iter->di.lstart) 1750 continue; 1751 if (iter->di.len < dpolicy->granularity) 1752 continue; 1753 if (iter->state == D_DONE && !iter->ref) { 1754 wait_for_completion_io(&iter->wait); 1755 if (!iter->error) 1756 trimmed += iter->di.len; 1757 __remove_discard_cmd(sbi, iter); 1758 } else { 1759 iter->ref++; 1760 dc = iter; 1761 break; 1762 } 1763 } 1764 mutex_unlock(&dcc->cmd_lock); 1765 1766 if (dc) { 1767 trimmed += __wait_one_discard_bio(sbi, dc); 1768 goto next; 1769 } 1770 1771 return trimmed; 1772 } 1773 1774 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1775 struct discard_policy *dpolicy) 1776 { 1777 struct discard_policy dp; 1778 unsigned int discard_blks; 1779 1780 if (dpolicy) 1781 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1782 1783 /* wait all */ 1784 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY); 1785 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1786 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY); 1787 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1788 1789 return discard_blks; 1790 } 1791 1792 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1793 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1794 { 1795 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1796 struct discard_cmd *dc; 1797 bool need_wait = false; 1798 1799 mutex_lock(&dcc->cmd_lock); 1800 dc = __lookup_discard_cmd(sbi, blkaddr); 1801 #ifdef CONFIG_BLK_DEV_ZONED 1802 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) { 1803 int devi = f2fs_bdev_index(sbi, dc->bdev); 1804 1805 if (devi < 0) { 1806 mutex_unlock(&dcc->cmd_lock); 1807 return; 1808 } 1809 1810 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1811 /* force submit zone reset */ 1812 if (dc->state == D_PREP) 1813 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC, 1814 &dcc->wait_list, NULL); 1815 dc->ref++; 1816 mutex_unlock(&dcc->cmd_lock); 1817 /* wait zone reset */ 1818 __wait_one_discard_bio(sbi, dc); 1819 return; 1820 } 1821 } 1822 #endif 1823 if (dc) { 1824 if (dc->state == D_PREP) { 1825 __punch_discard_cmd(sbi, dc, blkaddr); 1826 } else { 1827 dc->ref++; 1828 need_wait = true; 1829 } 1830 } 1831 mutex_unlock(&dcc->cmd_lock); 1832 1833 if (need_wait) 1834 __wait_one_discard_bio(sbi, dc); 1835 } 1836 1837 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1838 { 1839 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1840 1841 if (dcc && dcc->f2fs_issue_discard) { 1842 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1843 1844 dcc->f2fs_issue_discard = NULL; 1845 kthread_stop(discard_thread); 1846 } 1847 } 1848 1849 /** 1850 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT 1851 * @sbi: the f2fs_sb_info data for discard cmd to issue 1852 * 1853 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped 1854 * 1855 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false. 1856 */ 1857 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1858 { 1859 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1860 struct discard_policy dpolicy; 1861 bool dropped; 1862 1863 if (!atomic_read(&dcc->discard_cmd_cnt)) 1864 return true; 1865 1866 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1867 dcc->discard_granularity); 1868 __issue_discard_cmd(sbi, &dpolicy); 1869 dropped = __drop_discard_cmd(sbi); 1870 1871 /* just to make sure there is no pending discard commands */ 1872 __wait_all_discard_cmd(sbi, NULL); 1873 1874 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1875 return !dropped; 1876 } 1877 1878 static int issue_discard_thread(void *data) 1879 { 1880 struct f2fs_sb_info *sbi = data; 1881 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1882 wait_queue_head_t *q = &dcc->discard_wait_queue; 1883 struct discard_policy dpolicy; 1884 unsigned int wait_ms = dcc->min_discard_issue_time; 1885 int issued; 1886 1887 set_freezable(); 1888 1889 do { 1890 wait_event_freezable_timeout(*q, 1891 kthread_should_stop() || dcc->discard_wake, 1892 msecs_to_jiffies(wait_ms)); 1893 1894 if (sbi->gc_mode == GC_URGENT_HIGH || 1895 !f2fs_available_free_memory(sbi, DISCARD_CACHE)) 1896 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1897 MIN_DISCARD_GRANULARITY); 1898 else 1899 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1900 dcc->discard_granularity); 1901 1902 if (dcc->discard_wake) 1903 dcc->discard_wake = false; 1904 1905 /* clean up pending candidates before going to sleep */ 1906 if (atomic_read(&dcc->queued_discard)) 1907 __wait_all_discard_cmd(sbi, NULL); 1908 1909 if (f2fs_readonly(sbi->sb)) 1910 continue; 1911 if (kthread_should_stop()) 1912 return 0; 1913 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) || 1914 !atomic_read(&dcc->discard_cmd_cnt)) { 1915 wait_ms = dpolicy.max_interval; 1916 continue; 1917 } 1918 1919 sb_start_intwrite(sbi->sb); 1920 1921 issued = __issue_discard_cmd(sbi, &dpolicy); 1922 if (issued > 0) { 1923 __wait_all_discard_cmd(sbi, &dpolicy); 1924 wait_ms = dpolicy.min_interval; 1925 } else if (issued == -1) { 1926 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1927 if (!wait_ms) 1928 wait_ms = dpolicy.mid_interval; 1929 } else { 1930 wait_ms = dpolicy.max_interval; 1931 } 1932 if (!atomic_read(&dcc->discard_cmd_cnt)) 1933 wait_ms = dpolicy.max_interval; 1934 1935 sb_end_intwrite(sbi->sb); 1936 1937 } while (!kthread_should_stop()); 1938 return 0; 1939 } 1940 1941 #ifdef CONFIG_BLK_DEV_ZONED 1942 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1943 struct block_device *bdev, block_t blkstart, block_t blklen) 1944 { 1945 sector_t sector, nr_sects; 1946 block_t lblkstart = blkstart; 1947 int devi = 0; 1948 u64 remainder = 0; 1949 1950 if (f2fs_is_multi_device(sbi)) { 1951 devi = f2fs_target_device_index(sbi, blkstart); 1952 if (blkstart < FDEV(devi).start_blk || 1953 blkstart > FDEV(devi).end_blk) { 1954 f2fs_err(sbi, "Invalid block %x", blkstart); 1955 return -EIO; 1956 } 1957 blkstart -= FDEV(devi).start_blk; 1958 } 1959 1960 /* For sequential zones, reset the zone write pointer */ 1961 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1962 sector = SECTOR_FROM_BLOCK(blkstart); 1963 nr_sects = SECTOR_FROM_BLOCK(blklen); 1964 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder); 1965 1966 if (remainder || nr_sects != bdev_zone_sectors(bdev)) { 1967 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1968 devi, sbi->s_ndevs ? FDEV(devi).path : "", 1969 blkstart, blklen); 1970 return -EIO; 1971 } 1972 1973 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) { 1974 unsigned int nofs_flags; 1975 int ret; 1976 1977 trace_f2fs_issue_reset_zone(bdev, blkstart); 1978 nofs_flags = memalloc_nofs_save(); 1979 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 1980 sector, nr_sects); 1981 memalloc_nofs_restore(nofs_flags); 1982 return ret; 1983 } 1984 1985 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen); 1986 return 0; 1987 } 1988 1989 /* For conventional zones, use regular discard if supported */ 1990 __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1991 return 0; 1992 } 1993 #endif 1994 1995 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1996 struct block_device *bdev, block_t blkstart, block_t blklen) 1997 { 1998 #ifdef CONFIG_BLK_DEV_ZONED 1999 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 2000 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 2001 #endif 2002 __queue_discard_cmd(sbi, bdev, blkstart, blklen); 2003 return 0; 2004 } 2005 2006 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 2007 block_t blkstart, block_t blklen) 2008 { 2009 sector_t start = blkstart, len = 0; 2010 struct block_device *bdev; 2011 struct seg_entry *se; 2012 unsigned int offset; 2013 block_t i; 2014 int err = 0; 2015 2016 bdev = f2fs_target_device(sbi, blkstart, NULL); 2017 2018 for (i = blkstart; i < blkstart + blklen; i++, len++) { 2019 if (i != start) { 2020 struct block_device *bdev2 = 2021 f2fs_target_device(sbi, i, NULL); 2022 2023 if (bdev2 != bdev) { 2024 err = __issue_discard_async(sbi, bdev, 2025 start, len); 2026 if (err) 2027 return err; 2028 bdev = bdev2; 2029 start = i; 2030 len = 0; 2031 } 2032 } 2033 2034 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 2035 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 2036 2037 if (f2fs_block_unit_discard(sbi) && 2038 !f2fs_test_and_set_bit(offset, se->discard_map)) 2039 sbi->discard_blks--; 2040 } 2041 2042 if (len) 2043 err = __issue_discard_async(sbi, bdev, start, len); 2044 return err; 2045 } 2046 2047 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 2048 bool check_only) 2049 { 2050 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2051 int max_blocks = sbi->blocks_per_seg; 2052 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 2053 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2054 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2055 unsigned long *discard_map = (unsigned long *)se->discard_map; 2056 unsigned long *dmap = SIT_I(sbi)->tmp_map; 2057 unsigned int start = 0, end = -1; 2058 bool force = (cpc->reason & CP_DISCARD); 2059 struct discard_entry *de = NULL; 2060 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 2061 int i; 2062 2063 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) || 2064 !f2fs_block_unit_discard(sbi)) 2065 return false; 2066 2067 if (!force) { 2068 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks || 2069 SM_I(sbi)->dcc_info->nr_discards >= 2070 SM_I(sbi)->dcc_info->max_discards) 2071 return false; 2072 } 2073 2074 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 2075 for (i = 0; i < entries; i++) 2076 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 2077 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 2078 2079 while (force || SM_I(sbi)->dcc_info->nr_discards <= 2080 SM_I(sbi)->dcc_info->max_discards) { 2081 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 2082 if (start >= max_blocks) 2083 break; 2084 2085 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 2086 if (force && start && end != max_blocks 2087 && (end - start) < cpc->trim_minlen) 2088 continue; 2089 2090 if (check_only) 2091 return true; 2092 2093 if (!de) { 2094 de = f2fs_kmem_cache_alloc(discard_entry_slab, 2095 GFP_F2FS_ZERO, true, NULL); 2096 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 2097 list_add_tail(&de->list, head); 2098 } 2099 2100 for (i = start; i < end; i++) 2101 __set_bit_le(i, (void *)de->discard_map); 2102 2103 SM_I(sbi)->dcc_info->nr_discards += end - start; 2104 } 2105 return false; 2106 } 2107 2108 static void release_discard_addr(struct discard_entry *entry) 2109 { 2110 list_del(&entry->list); 2111 kmem_cache_free(discard_entry_slab, entry); 2112 } 2113 2114 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 2115 { 2116 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 2117 struct discard_entry *entry, *this; 2118 2119 /* drop caches */ 2120 list_for_each_entry_safe(entry, this, head, list) 2121 release_discard_addr(entry); 2122 } 2123 2124 /* 2125 * Should call f2fs_clear_prefree_segments after checkpoint is done. 2126 */ 2127 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 2128 { 2129 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2130 unsigned int segno; 2131 2132 mutex_lock(&dirty_i->seglist_lock); 2133 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 2134 __set_test_and_free(sbi, segno, false); 2135 mutex_unlock(&dirty_i->seglist_lock); 2136 } 2137 2138 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 2139 struct cp_control *cpc) 2140 { 2141 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2142 struct list_head *head = &dcc->entry_list; 2143 struct discard_entry *entry, *this; 2144 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2145 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 2146 unsigned int start = 0, end = -1; 2147 unsigned int secno, start_segno; 2148 bool force = (cpc->reason & CP_DISCARD); 2149 bool section_alignment = F2FS_OPTION(sbi).discard_unit == 2150 DISCARD_UNIT_SECTION; 2151 2152 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi)) 2153 section_alignment = true; 2154 2155 mutex_lock(&dirty_i->seglist_lock); 2156 2157 while (1) { 2158 int i; 2159 2160 if (section_alignment && end != -1) 2161 end--; 2162 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 2163 if (start >= MAIN_SEGS(sbi)) 2164 break; 2165 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 2166 start + 1); 2167 2168 if (section_alignment) { 2169 start = rounddown(start, sbi->segs_per_sec); 2170 end = roundup(end, sbi->segs_per_sec); 2171 } 2172 2173 for (i = start; i < end; i++) { 2174 if (test_and_clear_bit(i, prefree_map)) 2175 dirty_i->nr_dirty[PRE]--; 2176 } 2177 2178 if (!f2fs_realtime_discard_enable(sbi)) 2179 continue; 2180 2181 if (force && start >= cpc->trim_start && 2182 (end - 1) <= cpc->trim_end) 2183 continue; 2184 2185 /* Should cover 2MB zoned device for zone-based reset */ 2186 if (!f2fs_sb_has_blkzoned(sbi) && 2187 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) { 2188 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 2189 (end - start) << sbi->log_blocks_per_seg); 2190 continue; 2191 } 2192 next: 2193 secno = GET_SEC_FROM_SEG(sbi, start); 2194 start_segno = GET_SEG_FROM_SEC(sbi, secno); 2195 if (!IS_CURSEC(sbi, secno) && 2196 !get_valid_blocks(sbi, start, true)) 2197 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 2198 sbi->segs_per_sec << sbi->log_blocks_per_seg); 2199 2200 start = start_segno + sbi->segs_per_sec; 2201 if (start < end) 2202 goto next; 2203 else 2204 end = start - 1; 2205 } 2206 mutex_unlock(&dirty_i->seglist_lock); 2207 2208 if (!f2fs_block_unit_discard(sbi)) 2209 goto wakeup; 2210 2211 /* send small discards */ 2212 list_for_each_entry_safe(entry, this, head, list) { 2213 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2214 bool is_valid = test_bit_le(0, entry->discard_map); 2215 2216 find_next: 2217 if (is_valid) { 2218 next_pos = find_next_zero_bit_le(entry->discard_map, 2219 sbi->blocks_per_seg, cur_pos); 2220 len = next_pos - cur_pos; 2221 2222 if (f2fs_sb_has_blkzoned(sbi) || 2223 (force && len < cpc->trim_minlen)) 2224 goto skip; 2225 2226 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2227 len); 2228 total_len += len; 2229 } else { 2230 next_pos = find_next_bit_le(entry->discard_map, 2231 sbi->blocks_per_seg, cur_pos); 2232 } 2233 skip: 2234 cur_pos = next_pos; 2235 is_valid = !is_valid; 2236 2237 if (cur_pos < sbi->blocks_per_seg) 2238 goto find_next; 2239 2240 release_discard_addr(entry); 2241 dcc->nr_discards -= total_len; 2242 } 2243 2244 wakeup: 2245 wake_up_discard_thread(sbi, false); 2246 } 2247 2248 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi) 2249 { 2250 dev_t dev = sbi->sb->s_bdev->bd_dev; 2251 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2252 int err = 0; 2253 2254 if (!f2fs_realtime_discard_enable(sbi)) 2255 return 0; 2256 2257 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2258 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2259 if (IS_ERR(dcc->f2fs_issue_discard)) { 2260 err = PTR_ERR(dcc->f2fs_issue_discard); 2261 dcc->f2fs_issue_discard = NULL; 2262 } 2263 2264 return err; 2265 } 2266 2267 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2268 { 2269 struct discard_cmd_control *dcc; 2270 int err = 0, i; 2271 2272 if (SM_I(sbi)->dcc_info) { 2273 dcc = SM_I(sbi)->dcc_info; 2274 goto init_thread; 2275 } 2276 2277 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2278 if (!dcc) 2279 return -ENOMEM; 2280 2281 dcc->discard_io_aware_gran = MAX_PLIST_NUM; 2282 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2283 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY; 2284 dcc->discard_io_aware = DPOLICY_IO_AWARE_ENABLE; 2285 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT) 2286 dcc->discard_granularity = sbi->blocks_per_seg; 2287 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION) 2288 dcc->discard_granularity = BLKS_PER_SEC(sbi); 2289 2290 INIT_LIST_HEAD(&dcc->entry_list); 2291 for (i = 0; i < MAX_PLIST_NUM; i++) 2292 INIT_LIST_HEAD(&dcc->pend_list[i]); 2293 INIT_LIST_HEAD(&dcc->wait_list); 2294 INIT_LIST_HEAD(&dcc->fstrim_list); 2295 mutex_init(&dcc->cmd_lock); 2296 atomic_set(&dcc->issued_discard, 0); 2297 atomic_set(&dcc->queued_discard, 0); 2298 atomic_set(&dcc->discard_cmd_cnt, 0); 2299 dcc->nr_discards = 0; 2300 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 2301 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST; 2302 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME; 2303 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME; 2304 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME; 2305 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL; 2306 dcc->undiscard_blks = 0; 2307 dcc->next_pos = 0; 2308 dcc->root = RB_ROOT_CACHED; 2309 dcc->rbtree_check = false; 2310 2311 init_waitqueue_head(&dcc->discard_wait_queue); 2312 SM_I(sbi)->dcc_info = dcc; 2313 init_thread: 2314 err = f2fs_start_discard_thread(sbi); 2315 if (err) { 2316 kfree(dcc); 2317 SM_I(sbi)->dcc_info = NULL; 2318 } 2319 2320 return err; 2321 } 2322 2323 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2324 { 2325 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2326 2327 if (!dcc) 2328 return; 2329 2330 f2fs_stop_discard_thread(sbi); 2331 2332 /* 2333 * Recovery can cache discard commands, so in error path of 2334 * fill_super(), it needs to give a chance to handle them. 2335 */ 2336 f2fs_issue_discard_timeout(sbi); 2337 2338 kfree(dcc); 2339 SM_I(sbi)->dcc_info = NULL; 2340 } 2341 2342 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2343 { 2344 struct sit_info *sit_i = SIT_I(sbi); 2345 2346 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2347 sit_i->dirty_sentries++; 2348 return false; 2349 } 2350 2351 return true; 2352 } 2353 2354 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2355 unsigned int segno, int modified) 2356 { 2357 struct seg_entry *se = get_seg_entry(sbi, segno); 2358 2359 se->type = type; 2360 if (modified) 2361 __mark_sit_entry_dirty(sbi, segno); 2362 } 2363 2364 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi, 2365 block_t blkaddr) 2366 { 2367 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2368 2369 if (segno == NULL_SEGNO) 2370 return 0; 2371 return get_seg_entry(sbi, segno)->mtime; 2372 } 2373 2374 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr, 2375 unsigned long long old_mtime) 2376 { 2377 struct seg_entry *se; 2378 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2379 unsigned long long ctime = get_mtime(sbi, false); 2380 unsigned long long mtime = old_mtime ? old_mtime : ctime; 2381 2382 if (segno == NULL_SEGNO) 2383 return; 2384 2385 se = get_seg_entry(sbi, segno); 2386 2387 if (!se->mtime) 2388 se->mtime = mtime; 2389 else 2390 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime, 2391 se->valid_blocks + 1); 2392 2393 if (ctime > SIT_I(sbi)->max_mtime) 2394 SIT_I(sbi)->max_mtime = ctime; 2395 } 2396 2397 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2398 { 2399 struct seg_entry *se; 2400 unsigned int segno, offset; 2401 long int new_vblocks; 2402 bool exist; 2403 #ifdef CONFIG_F2FS_CHECK_FS 2404 bool mir_exist; 2405 #endif 2406 2407 segno = GET_SEGNO(sbi, blkaddr); 2408 2409 se = get_seg_entry(sbi, segno); 2410 new_vblocks = se->valid_blocks + del; 2411 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2412 2413 f2fs_bug_on(sbi, (new_vblocks < 0 || 2414 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno)))); 2415 2416 se->valid_blocks = new_vblocks; 2417 2418 /* Update valid block bitmap */ 2419 if (del > 0) { 2420 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2421 #ifdef CONFIG_F2FS_CHECK_FS 2422 mir_exist = f2fs_test_and_set_bit(offset, 2423 se->cur_valid_map_mir); 2424 if (unlikely(exist != mir_exist)) { 2425 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2426 blkaddr, exist); 2427 f2fs_bug_on(sbi, 1); 2428 } 2429 #endif 2430 if (unlikely(exist)) { 2431 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", 2432 blkaddr); 2433 f2fs_bug_on(sbi, 1); 2434 se->valid_blocks--; 2435 del = 0; 2436 } 2437 2438 if (f2fs_block_unit_discard(sbi) && 2439 !f2fs_test_and_set_bit(offset, se->discard_map)) 2440 sbi->discard_blks--; 2441 2442 /* 2443 * SSR should never reuse block which is checkpointed 2444 * or newly invalidated. 2445 */ 2446 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2447 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 2448 se->ckpt_valid_blocks++; 2449 } 2450 } else { 2451 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 2452 #ifdef CONFIG_F2FS_CHECK_FS 2453 mir_exist = f2fs_test_and_clear_bit(offset, 2454 se->cur_valid_map_mir); 2455 if (unlikely(exist != mir_exist)) { 2456 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2457 blkaddr, exist); 2458 f2fs_bug_on(sbi, 1); 2459 } 2460 #endif 2461 if (unlikely(!exist)) { 2462 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", 2463 blkaddr); 2464 f2fs_bug_on(sbi, 1); 2465 se->valid_blocks++; 2466 del = 0; 2467 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2468 /* 2469 * If checkpoints are off, we must not reuse data that 2470 * was used in the previous checkpoint. If it was used 2471 * before, we must track that to know how much space we 2472 * really have. 2473 */ 2474 if (f2fs_test_bit(offset, se->ckpt_valid_map)) { 2475 spin_lock(&sbi->stat_lock); 2476 sbi->unusable_block_count++; 2477 spin_unlock(&sbi->stat_lock); 2478 } 2479 } 2480 2481 if (f2fs_block_unit_discard(sbi) && 2482 f2fs_test_and_clear_bit(offset, se->discard_map)) 2483 sbi->discard_blks++; 2484 } 2485 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 2486 se->ckpt_valid_blocks += del; 2487 2488 __mark_sit_entry_dirty(sbi, segno); 2489 2490 /* update total number of valid blocks to be written in ckpt area */ 2491 SIT_I(sbi)->written_valid_blocks += del; 2492 2493 if (__is_large_section(sbi)) 2494 get_sec_entry(sbi, segno)->valid_blocks += del; 2495 } 2496 2497 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 2498 { 2499 unsigned int segno = GET_SEGNO(sbi, addr); 2500 struct sit_info *sit_i = SIT_I(sbi); 2501 2502 f2fs_bug_on(sbi, addr == NULL_ADDR); 2503 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2504 return; 2505 2506 f2fs_invalidate_internal_cache(sbi, addr); 2507 2508 /* add it into sit main buffer */ 2509 down_write(&sit_i->sentry_lock); 2510 2511 update_segment_mtime(sbi, addr, 0); 2512 update_sit_entry(sbi, addr, -1); 2513 2514 /* add it into dirty seglist */ 2515 locate_dirty_segment(sbi, segno); 2516 2517 up_write(&sit_i->sentry_lock); 2518 } 2519 2520 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2521 { 2522 struct sit_info *sit_i = SIT_I(sbi); 2523 unsigned int segno, offset; 2524 struct seg_entry *se; 2525 bool is_cp = false; 2526 2527 if (!__is_valid_data_blkaddr(blkaddr)) 2528 return true; 2529 2530 down_read(&sit_i->sentry_lock); 2531 2532 segno = GET_SEGNO(sbi, blkaddr); 2533 se = get_seg_entry(sbi, segno); 2534 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2535 2536 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2537 is_cp = true; 2538 2539 up_read(&sit_i->sentry_lock); 2540 2541 return is_cp; 2542 } 2543 2544 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type) 2545 { 2546 struct curseg_info *curseg = CURSEG_I(sbi, type); 2547 2548 if (sbi->ckpt->alloc_type[type] == SSR) 2549 return sbi->blocks_per_seg; 2550 return curseg->next_blkoff; 2551 } 2552 2553 /* 2554 * Calculate the number of current summary pages for writing 2555 */ 2556 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2557 { 2558 int valid_sum_count = 0; 2559 int i, sum_in_page; 2560 2561 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2562 if (sbi->ckpt->alloc_type[i] != SSR && for_ra) 2563 valid_sum_count += 2564 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2565 else 2566 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i); 2567 } 2568 2569 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 2570 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2571 if (valid_sum_count <= sum_in_page) 2572 return 1; 2573 else if ((valid_sum_count - sum_in_page) <= 2574 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2575 return 2; 2576 return 3; 2577 } 2578 2579 /* 2580 * Caller should put this summary page 2581 */ 2582 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 2583 { 2584 if (unlikely(f2fs_cp_error(sbi))) 2585 return ERR_PTR(-EIO); 2586 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno)); 2587 } 2588 2589 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2590 void *src, block_t blk_addr) 2591 { 2592 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2593 2594 memcpy(page_address(page), src, PAGE_SIZE); 2595 set_page_dirty(page); 2596 f2fs_put_page(page, 1); 2597 } 2598 2599 static void write_sum_page(struct f2fs_sb_info *sbi, 2600 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2601 { 2602 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2603 } 2604 2605 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2606 int type, block_t blk_addr) 2607 { 2608 struct curseg_info *curseg = CURSEG_I(sbi, type); 2609 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2610 struct f2fs_summary_block *src = curseg->sum_blk; 2611 struct f2fs_summary_block *dst; 2612 2613 dst = (struct f2fs_summary_block *)page_address(page); 2614 memset(dst, 0, PAGE_SIZE); 2615 2616 mutex_lock(&curseg->curseg_mutex); 2617 2618 down_read(&curseg->journal_rwsem); 2619 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2620 up_read(&curseg->journal_rwsem); 2621 2622 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2623 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2624 2625 mutex_unlock(&curseg->curseg_mutex); 2626 2627 set_page_dirty(page); 2628 f2fs_put_page(page, 1); 2629 } 2630 2631 static int is_next_segment_free(struct f2fs_sb_info *sbi, 2632 struct curseg_info *curseg, int type) 2633 { 2634 unsigned int segno = curseg->segno + 1; 2635 struct free_segmap_info *free_i = FREE_I(sbi); 2636 2637 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 2638 return !test_bit(segno, free_i->free_segmap); 2639 return 0; 2640 } 2641 2642 /* 2643 * Find a new segment from the free segments bitmap to right order 2644 * This function should be returned with success, otherwise BUG 2645 */ 2646 static void get_new_segment(struct f2fs_sb_info *sbi, 2647 unsigned int *newseg, bool new_sec, int dir) 2648 { 2649 struct free_segmap_info *free_i = FREE_I(sbi); 2650 unsigned int segno, secno, zoneno; 2651 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2652 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2653 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2654 unsigned int left_start = hint; 2655 bool init = true; 2656 int go_left = 0; 2657 int i; 2658 2659 spin_lock(&free_i->segmap_lock); 2660 2661 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 2662 segno = find_next_zero_bit(free_i->free_segmap, 2663 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2664 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2665 goto got_it; 2666 } 2667 find_other_zone: 2668 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2669 if (secno >= MAIN_SECS(sbi)) { 2670 if (dir == ALLOC_RIGHT) { 2671 secno = find_first_zero_bit(free_i->free_secmap, 2672 MAIN_SECS(sbi)); 2673 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 2674 } else { 2675 go_left = 1; 2676 left_start = hint - 1; 2677 } 2678 } 2679 if (go_left == 0) 2680 goto skip_left; 2681 2682 while (test_bit(left_start, free_i->free_secmap)) { 2683 if (left_start > 0) { 2684 left_start--; 2685 continue; 2686 } 2687 left_start = find_first_zero_bit(free_i->free_secmap, 2688 MAIN_SECS(sbi)); 2689 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 2690 break; 2691 } 2692 secno = left_start; 2693 skip_left: 2694 segno = GET_SEG_FROM_SEC(sbi, secno); 2695 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2696 2697 /* give up on finding another zone */ 2698 if (!init) 2699 goto got_it; 2700 if (sbi->secs_per_zone == 1) 2701 goto got_it; 2702 if (zoneno == old_zoneno) 2703 goto got_it; 2704 if (dir == ALLOC_LEFT) { 2705 if (!go_left && zoneno + 1 >= total_zones) 2706 goto got_it; 2707 if (go_left && zoneno == 0) 2708 goto got_it; 2709 } 2710 for (i = 0; i < NR_CURSEG_TYPE; i++) 2711 if (CURSEG_I(sbi, i)->zone == zoneno) 2712 break; 2713 2714 if (i < NR_CURSEG_TYPE) { 2715 /* zone is in user, try another */ 2716 if (go_left) 2717 hint = zoneno * sbi->secs_per_zone - 1; 2718 else if (zoneno + 1 >= total_zones) 2719 hint = 0; 2720 else 2721 hint = (zoneno + 1) * sbi->secs_per_zone; 2722 init = false; 2723 goto find_other_zone; 2724 } 2725 got_it: 2726 /* set it as dirty segment in free segmap */ 2727 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2728 __set_inuse(sbi, segno); 2729 *newseg = segno; 2730 spin_unlock(&free_i->segmap_lock); 2731 } 2732 2733 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2734 { 2735 struct curseg_info *curseg = CURSEG_I(sbi, type); 2736 struct summary_footer *sum_footer; 2737 unsigned short seg_type = curseg->seg_type; 2738 2739 curseg->inited = true; 2740 curseg->segno = curseg->next_segno; 2741 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2742 curseg->next_blkoff = 0; 2743 curseg->next_segno = NULL_SEGNO; 2744 2745 sum_footer = &(curseg->sum_blk->footer); 2746 memset(sum_footer, 0, sizeof(struct summary_footer)); 2747 2748 sanity_check_seg_type(sbi, seg_type); 2749 2750 if (IS_DATASEG(seg_type)) 2751 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2752 if (IS_NODESEG(seg_type)) 2753 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2754 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified); 2755 } 2756 2757 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2758 { 2759 struct curseg_info *curseg = CURSEG_I(sbi, type); 2760 unsigned short seg_type = curseg->seg_type; 2761 2762 sanity_check_seg_type(sbi, seg_type); 2763 if (f2fs_need_rand_seg(sbi)) 2764 return get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec); 2765 2766 /* if segs_per_sec is large than 1, we need to keep original policy. */ 2767 if (__is_large_section(sbi)) 2768 return curseg->segno; 2769 2770 /* inmem log may not locate on any segment after mount */ 2771 if (!curseg->inited) 2772 return 0; 2773 2774 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2775 return 0; 2776 2777 if (test_opt(sbi, NOHEAP) && 2778 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))) 2779 return 0; 2780 2781 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2782 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2783 2784 /* find segments from 0 to reuse freed segments */ 2785 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2786 return 0; 2787 2788 return curseg->segno; 2789 } 2790 2791 /* 2792 * Allocate a current working segment. 2793 * This function always allocates a free segment in LFS manner. 2794 */ 2795 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2796 { 2797 struct curseg_info *curseg = CURSEG_I(sbi, type); 2798 unsigned short seg_type = curseg->seg_type; 2799 unsigned int segno = curseg->segno; 2800 int dir = ALLOC_LEFT; 2801 2802 if (curseg->inited) 2803 write_sum_page(sbi, curseg->sum_blk, 2804 GET_SUM_BLOCK(sbi, segno)); 2805 if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA) 2806 dir = ALLOC_RIGHT; 2807 2808 if (test_opt(sbi, NOHEAP)) 2809 dir = ALLOC_RIGHT; 2810 2811 segno = __get_next_segno(sbi, type); 2812 get_new_segment(sbi, &segno, new_sec, dir); 2813 curseg->next_segno = segno; 2814 reset_curseg(sbi, type, 1); 2815 curseg->alloc_type = LFS; 2816 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 2817 curseg->fragment_remained_chunk = 2818 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 2819 } 2820 2821 static int __next_free_blkoff(struct f2fs_sb_info *sbi, 2822 int segno, block_t start) 2823 { 2824 struct seg_entry *se = get_seg_entry(sbi, segno); 2825 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2826 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2827 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2828 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2829 int i; 2830 2831 for (i = 0; i < entries; i++) 2832 target_map[i] = ckpt_map[i] | cur_map[i]; 2833 2834 return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 2835 } 2836 2837 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi, 2838 struct curseg_info *seg) 2839 { 2840 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1); 2841 } 2842 2843 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno) 2844 { 2845 return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg; 2846 } 2847 2848 /* 2849 * This function always allocates a used segment(from dirty seglist) by SSR 2850 * manner, so it should recover the existing segment information of valid blocks 2851 */ 2852 static void change_curseg(struct f2fs_sb_info *sbi, int type) 2853 { 2854 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2855 struct curseg_info *curseg = CURSEG_I(sbi, type); 2856 unsigned int new_segno = curseg->next_segno; 2857 struct f2fs_summary_block *sum_node; 2858 struct page *sum_page; 2859 2860 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno)); 2861 2862 __set_test_and_inuse(sbi, new_segno); 2863 2864 mutex_lock(&dirty_i->seglist_lock); 2865 __remove_dirty_segment(sbi, new_segno, PRE); 2866 __remove_dirty_segment(sbi, new_segno, DIRTY); 2867 mutex_unlock(&dirty_i->seglist_lock); 2868 2869 reset_curseg(sbi, type, 1); 2870 curseg->alloc_type = SSR; 2871 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0); 2872 2873 sum_page = f2fs_get_sum_page(sbi, new_segno); 2874 if (IS_ERR(sum_page)) { 2875 /* GC won't be able to use stale summary pages by cp_error */ 2876 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE); 2877 return; 2878 } 2879 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2880 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2881 f2fs_put_page(sum_page, 1); 2882 } 2883 2884 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2885 int alloc_mode, unsigned long long age); 2886 2887 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type, 2888 int target_type, int alloc_mode, 2889 unsigned long long age) 2890 { 2891 struct curseg_info *curseg = CURSEG_I(sbi, type); 2892 2893 curseg->seg_type = target_type; 2894 2895 if (get_ssr_segment(sbi, type, alloc_mode, age)) { 2896 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno); 2897 2898 curseg->seg_type = se->type; 2899 change_curseg(sbi, type); 2900 } else { 2901 /* allocate cold segment by default */ 2902 curseg->seg_type = CURSEG_COLD_DATA; 2903 new_curseg(sbi, type, true); 2904 } 2905 stat_inc_seg_type(sbi, curseg); 2906 } 2907 2908 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi) 2909 { 2910 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC); 2911 2912 if (!sbi->am.atgc_enabled) 2913 return; 2914 2915 f2fs_down_read(&SM_I(sbi)->curseg_lock); 2916 2917 mutex_lock(&curseg->curseg_mutex); 2918 down_write(&SIT_I(sbi)->sentry_lock); 2919 2920 get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0); 2921 2922 up_write(&SIT_I(sbi)->sentry_lock); 2923 mutex_unlock(&curseg->curseg_mutex); 2924 2925 f2fs_up_read(&SM_I(sbi)->curseg_lock); 2926 2927 } 2928 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi) 2929 { 2930 __f2fs_init_atgc_curseg(sbi); 2931 } 2932 2933 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2934 { 2935 struct curseg_info *curseg = CURSEG_I(sbi, type); 2936 2937 mutex_lock(&curseg->curseg_mutex); 2938 if (!curseg->inited) 2939 goto out; 2940 2941 if (get_valid_blocks(sbi, curseg->segno, false)) { 2942 write_sum_page(sbi, curseg->sum_blk, 2943 GET_SUM_BLOCK(sbi, curseg->segno)); 2944 } else { 2945 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2946 __set_test_and_free(sbi, curseg->segno, true); 2947 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2948 } 2949 out: 2950 mutex_unlock(&curseg->curseg_mutex); 2951 } 2952 2953 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi) 2954 { 2955 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2956 2957 if (sbi->am.atgc_enabled) 2958 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2959 } 2960 2961 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2962 { 2963 struct curseg_info *curseg = CURSEG_I(sbi, type); 2964 2965 mutex_lock(&curseg->curseg_mutex); 2966 if (!curseg->inited) 2967 goto out; 2968 if (get_valid_blocks(sbi, curseg->segno, false)) 2969 goto out; 2970 2971 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2972 __set_test_and_inuse(sbi, curseg->segno); 2973 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2974 out: 2975 mutex_unlock(&curseg->curseg_mutex); 2976 } 2977 2978 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi) 2979 { 2980 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2981 2982 if (sbi->am.atgc_enabled) 2983 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2984 } 2985 2986 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2987 int alloc_mode, unsigned long long age) 2988 { 2989 struct curseg_info *curseg = CURSEG_I(sbi, type); 2990 unsigned segno = NULL_SEGNO; 2991 unsigned short seg_type = curseg->seg_type; 2992 int i, cnt; 2993 bool reversed = false; 2994 2995 sanity_check_seg_type(sbi, seg_type); 2996 2997 /* f2fs_need_SSR() already forces to do this */ 2998 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) { 2999 curseg->next_segno = segno; 3000 return 1; 3001 } 3002 3003 /* For node segments, let's do SSR more intensively */ 3004 if (IS_NODESEG(seg_type)) { 3005 if (seg_type >= CURSEG_WARM_NODE) { 3006 reversed = true; 3007 i = CURSEG_COLD_NODE; 3008 } else { 3009 i = CURSEG_HOT_NODE; 3010 } 3011 cnt = NR_CURSEG_NODE_TYPE; 3012 } else { 3013 if (seg_type >= CURSEG_WARM_DATA) { 3014 reversed = true; 3015 i = CURSEG_COLD_DATA; 3016 } else { 3017 i = CURSEG_HOT_DATA; 3018 } 3019 cnt = NR_CURSEG_DATA_TYPE; 3020 } 3021 3022 for (; cnt-- > 0; reversed ? i-- : i++) { 3023 if (i == seg_type) 3024 continue; 3025 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) { 3026 curseg->next_segno = segno; 3027 return 1; 3028 } 3029 } 3030 3031 /* find valid_blocks=0 in dirty list */ 3032 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 3033 segno = get_free_segment(sbi); 3034 if (segno != NULL_SEGNO) { 3035 curseg->next_segno = segno; 3036 return 1; 3037 } 3038 } 3039 return 0; 3040 } 3041 3042 static bool need_new_seg(struct f2fs_sb_info *sbi, int type) 3043 { 3044 struct curseg_info *curseg = CURSEG_I(sbi, type); 3045 3046 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 3047 curseg->seg_type == CURSEG_WARM_NODE) 3048 return true; 3049 if (curseg->alloc_type == LFS && 3050 is_next_segment_free(sbi, curseg, type) && 3051 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 3052 return true; 3053 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0)) 3054 return true; 3055 return false; 3056 } 3057 3058 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 3059 unsigned int start, unsigned int end) 3060 { 3061 struct curseg_info *curseg = CURSEG_I(sbi, type); 3062 unsigned int segno; 3063 3064 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3065 mutex_lock(&curseg->curseg_mutex); 3066 down_write(&SIT_I(sbi)->sentry_lock); 3067 3068 segno = CURSEG_I(sbi, type)->segno; 3069 if (segno < start || segno > end) 3070 goto unlock; 3071 3072 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0)) 3073 change_curseg(sbi, type); 3074 else 3075 new_curseg(sbi, type, true); 3076 3077 stat_inc_seg_type(sbi, curseg); 3078 3079 locate_dirty_segment(sbi, segno); 3080 unlock: 3081 up_write(&SIT_I(sbi)->sentry_lock); 3082 3083 if (segno != curseg->segno) 3084 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 3085 type, segno, curseg->segno); 3086 3087 mutex_unlock(&curseg->curseg_mutex); 3088 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3089 } 3090 3091 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type, 3092 bool new_sec, bool force) 3093 { 3094 struct curseg_info *curseg = CURSEG_I(sbi, type); 3095 unsigned int old_segno; 3096 3097 if (!force && curseg->inited && 3098 !curseg->next_blkoff && 3099 !get_valid_blocks(sbi, curseg->segno, new_sec) && 3100 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec)) 3101 return; 3102 3103 old_segno = curseg->segno; 3104 new_curseg(sbi, type, true); 3105 stat_inc_seg_type(sbi, curseg); 3106 locate_dirty_segment(sbi, old_segno); 3107 } 3108 3109 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force) 3110 { 3111 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3112 down_write(&SIT_I(sbi)->sentry_lock); 3113 __allocate_new_segment(sbi, type, true, force); 3114 up_write(&SIT_I(sbi)->sentry_lock); 3115 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3116 } 3117 3118 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 3119 { 3120 int i; 3121 3122 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3123 down_write(&SIT_I(sbi)->sentry_lock); 3124 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 3125 __allocate_new_segment(sbi, i, false, false); 3126 up_write(&SIT_I(sbi)->sentry_lock); 3127 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3128 } 3129 3130 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 3131 struct cp_control *cpc) 3132 { 3133 __u64 trim_start = cpc->trim_start; 3134 bool has_candidate = false; 3135 3136 down_write(&SIT_I(sbi)->sentry_lock); 3137 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 3138 if (add_discard_addrs(sbi, cpc, true)) { 3139 has_candidate = true; 3140 break; 3141 } 3142 } 3143 up_write(&SIT_I(sbi)->sentry_lock); 3144 3145 cpc->trim_start = trim_start; 3146 return has_candidate; 3147 } 3148 3149 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 3150 struct discard_policy *dpolicy, 3151 unsigned int start, unsigned int end) 3152 { 3153 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 3154 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 3155 struct rb_node **insert_p = NULL, *insert_parent = NULL; 3156 struct discard_cmd *dc; 3157 struct blk_plug plug; 3158 int issued; 3159 unsigned int trimmed = 0; 3160 3161 next: 3162 issued = 0; 3163 3164 mutex_lock(&dcc->cmd_lock); 3165 if (unlikely(dcc->rbtree_check)) 3166 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 3167 3168 dc = __lookup_discard_cmd_ret(&dcc->root, start, 3169 &prev_dc, &next_dc, &insert_p, &insert_parent); 3170 if (!dc) 3171 dc = next_dc; 3172 3173 blk_start_plug(&plug); 3174 3175 while (dc && dc->di.lstart <= end) { 3176 struct rb_node *node; 3177 int err = 0; 3178 3179 if (dc->di.len < dpolicy->granularity) 3180 goto skip; 3181 3182 if (dc->state != D_PREP) { 3183 list_move_tail(&dc->list, &dcc->fstrim_list); 3184 goto skip; 3185 } 3186 3187 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 3188 3189 if (issued >= dpolicy->max_requests) { 3190 start = dc->di.lstart + dc->di.len; 3191 3192 if (err) 3193 __remove_discard_cmd(sbi, dc); 3194 3195 blk_finish_plug(&plug); 3196 mutex_unlock(&dcc->cmd_lock); 3197 trimmed += __wait_all_discard_cmd(sbi, NULL); 3198 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 3199 goto next; 3200 } 3201 skip: 3202 node = rb_next(&dc->rb_node); 3203 if (err) 3204 __remove_discard_cmd(sbi, dc); 3205 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 3206 3207 if (fatal_signal_pending(current)) 3208 break; 3209 } 3210 3211 blk_finish_plug(&plug); 3212 mutex_unlock(&dcc->cmd_lock); 3213 3214 return trimmed; 3215 } 3216 3217 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 3218 { 3219 __u64 start = F2FS_BYTES_TO_BLK(range->start); 3220 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 3221 unsigned int start_segno, end_segno; 3222 block_t start_block, end_block; 3223 struct cp_control cpc; 3224 struct discard_policy dpolicy; 3225 unsigned long long trimmed = 0; 3226 int err = 0; 3227 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 3228 3229 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 3230 return -EINVAL; 3231 3232 if (end < MAIN_BLKADDR(sbi)) 3233 goto out; 3234 3235 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 3236 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 3237 return -EFSCORRUPTED; 3238 } 3239 3240 /* start/end segment number in main_area */ 3241 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 3242 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 3243 GET_SEGNO(sbi, end); 3244 if (need_align) { 3245 start_segno = rounddown(start_segno, sbi->segs_per_sec); 3246 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1; 3247 } 3248 3249 cpc.reason = CP_DISCARD; 3250 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 3251 cpc.trim_start = start_segno; 3252 cpc.trim_end = end_segno; 3253 3254 if (sbi->discard_blks == 0) 3255 goto out; 3256 3257 f2fs_down_write(&sbi->gc_lock); 3258 stat_inc_cp_call_count(sbi, TOTAL_CALL); 3259 err = f2fs_write_checkpoint(sbi, &cpc); 3260 f2fs_up_write(&sbi->gc_lock); 3261 if (err) 3262 goto out; 3263 3264 /* 3265 * We filed discard candidates, but actually we don't need to wait for 3266 * all of them, since they'll be issued in idle time along with runtime 3267 * discard option. User configuration looks like using runtime discard 3268 * or periodic fstrim instead of it. 3269 */ 3270 if (f2fs_realtime_discard_enable(sbi)) 3271 goto out; 3272 3273 start_block = START_BLOCK(sbi, start_segno); 3274 end_block = START_BLOCK(sbi, end_segno + 1); 3275 3276 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 3277 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 3278 start_block, end_block); 3279 3280 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 3281 start_block, end_block); 3282 out: 3283 if (!err) 3284 range->len = F2FS_BLK_TO_BYTES(trimmed); 3285 return err; 3286 } 3287 3288 int f2fs_rw_hint_to_seg_type(enum rw_hint hint) 3289 { 3290 switch (hint) { 3291 case WRITE_LIFE_SHORT: 3292 return CURSEG_HOT_DATA; 3293 case WRITE_LIFE_EXTREME: 3294 return CURSEG_COLD_DATA; 3295 default: 3296 return CURSEG_WARM_DATA; 3297 } 3298 } 3299 3300 static int __get_segment_type_2(struct f2fs_io_info *fio) 3301 { 3302 if (fio->type == DATA) 3303 return CURSEG_HOT_DATA; 3304 else 3305 return CURSEG_HOT_NODE; 3306 } 3307 3308 static int __get_segment_type_4(struct f2fs_io_info *fio) 3309 { 3310 if (fio->type == DATA) { 3311 struct inode *inode = fio->page->mapping->host; 3312 3313 if (S_ISDIR(inode->i_mode)) 3314 return CURSEG_HOT_DATA; 3315 else 3316 return CURSEG_COLD_DATA; 3317 } else { 3318 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3319 return CURSEG_WARM_NODE; 3320 else 3321 return CURSEG_COLD_NODE; 3322 } 3323 } 3324 3325 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs) 3326 { 3327 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3328 struct extent_info ei = {}; 3329 3330 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) { 3331 if (!ei.age) 3332 return NO_CHECK_TYPE; 3333 if (ei.age <= sbi->hot_data_age_threshold) 3334 return CURSEG_HOT_DATA; 3335 if (ei.age <= sbi->warm_data_age_threshold) 3336 return CURSEG_WARM_DATA; 3337 return CURSEG_COLD_DATA; 3338 } 3339 return NO_CHECK_TYPE; 3340 } 3341 3342 static int __get_segment_type_6(struct f2fs_io_info *fio) 3343 { 3344 if (fio->type == DATA) { 3345 struct inode *inode = fio->page->mapping->host; 3346 int type; 3347 3348 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 3349 return CURSEG_COLD_DATA_PINNED; 3350 3351 if (page_private_gcing(fio->page)) { 3352 if (fio->sbi->am.atgc_enabled && 3353 (fio->io_type == FS_DATA_IO) && 3354 (fio->sbi->gc_mode != GC_URGENT_HIGH)) 3355 return CURSEG_ALL_DATA_ATGC; 3356 else 3357 return CURSEG_COLD_DATA; 3358 } 3359 if (file_is_cold(inode) || f2fs_need_compress_data(inode)) 3360 return CURSEG_COLD_DATA; 3361 3362 type = __get_age_segment_type(inode, fio->page->index); 3363 if (type != NO_CHECK_TYPE) 3364 return type; 3365 3366 if (file_is_hot(inode) || 3367 is_inode_flag_set(inode, FI_HOT_DATA) || 3368 f2fs_is_cow_file(inode)) 3369 return CURSEG_HOT_DATA; 3370 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 3371 } else { 3372 if (IS_DNODE(fio->page)) 3373 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3374 CURSEG_HOT_NODE; 3375 return CURSEG_COLD_NODE; 3376 } 3377 } 3378 3379 static int __get_segment_type(struct f2fs_io_info *fio) 3380 { 3381 int type = 0; 3382 3383 switch (F2FS_OPTION(fio->sbi).active_logs) { 3384 case 2: 3385 type = __get_segment_type_2(fio); 3386 break; 3387 case 4: 3388 type = __get_segment_type_4(fio); 3389 break; 3390 case 6: 3391 type = __get_segment_type_6(fio); 3392 break; 3393 default: 3394 f2fs_bug_on(fio->sbi, true); 3395 } 3396 3397 if (IS_HOT(type)) 3398 fio->temp = HOT; 3399 else if (IS_WARM(type)) 3400 fio->temp = WARM; 3401 else 3402 fio->temp = COLD; 3403 return type; 3404 } 3405 3406 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi, 3407 struct curseg_info *seg) 3408 { 3409 /* To allocate block chunks in different sizes, use random number */ 3410 if (--seg->fragment_remained_chunk > 0) 3411 return; 3412 3413 seg->fragment_remained_chunk = 3414 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 3415 seg->next_blkoff += 3416 get_random_u32_inclusive(1, sbi->max_fragment_hole); 3417 } 3418 3419 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3420 block_t old_blkaddr, block_t *new_blkaddr, 3421 struct f2fs_summary *sum, int type, 3422 struct f2fs_io_info *fio) 3423 { 3424 struct sit_info *sit_i = SIT_I(sbi); 3425 struct curseg_info *curseg = CURSEG_I(sbi, type); 3426 unsigned long long old_mtime; 3427 bool from_gc = (type == CURSEG_ALL_DATA_ATGC); 3428 struct seg_entry *se = NULL; 3429 bool segment_full = false; 3430 3431 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3432 3433 mutex_lock(&curseg->curseg_mutex); 3434 down_write(&sit_i->sentry_lock); 3435 3436 if (from_gc) { 3437 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO); 3438 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr)); 3439 sanity_check_seg_type(sbi, se->type); 3440 f2fs_bug_on(sbi, IS_NODESEG(se->type)); 3441 } 3442 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3443 3444 f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg); 3445 3446 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3447 3448 curseg->sum_blk->entries[curseg->next_blkoff] = *sum; 3449 if (curseg->alloc_type == SSR) { 3450 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg); 3451 } else { 3452 curseg->next_blkoff++; 3453 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 3454 f2fs_randomize_chunk(sbi, curseg); 3455 } 3456 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno)) 3457 segment_full = true; 3458 stat_inc_block_count(sbi, curseg); 3459 3460 if (from_gc) { 3461 old_mtime = get_segment_mtime(sbi, old_blkaddr); 3462 } else { 3463 update_segment_mtime(sbi, old_blkaddr, 0); 3464 old_mtime = 0; 3465 } 3466 update_segment_mtime(sbi, *new_blkaddr, old_mtime); 3467 3468 /* 3469 * SIT information should be updated before segment allocation, 3470 * since SSR needs latest valid block information. 3471 */ 3472 update_sit_entry(sbi, *new_blkaddr, 1); 3473 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 3474 update_sit_entry(sbi, old_blkaddr, -1); 3475 3476 /* 3477 * If the current segment is full, flush it out and replace it with a 3478 * new segment. 3479 */ 3480 if (segment_full) { 3481 if (from_gc) { 3482 get_atssr_segment(sbi, type, se->type, 3483 AT_SSR, se->mtime); 3484 } else { 3485 if (need_new_seg(sbi, type)) 3486 new_curseg(sbi, type, false); 3487 else 3488 change_curseg(sbi, type); 3489 stat_inc_seg_type(sbi, curseg); 3490 } 3491 } 3492 /* 3493 * segment dirty status should be updated after segment allocation, 3494 * so we just need to update status only one time after previous 3495 * segment being closed. 3496 */ 3497 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3498 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3499 3500 if (IS_DATASEG(type)) 3501 atomic64_inc(&sbi->allocated_data_blocks); 3502 3503 up_write(&sit_i->sentry_lock); 3504 3505 if (page && IS_NODESEG(type)) { 3506 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3507 3508 f2fs_inode_chksum_set(sbi, page); 3509 } 3510 3511 if (fio) { 3512 struct f2fs_bio_info *io; 3513 3514 if (F2FS_IO_ALIGNED(sbi)) 3515 fio->retry = 0; 3516 3517 INIT_LIST_HEAD(&fio->list); 3518 fio->in_list = 1; 3519 io = sbi->write_io[fio->type] + fio->temp; 3520 spin_lock(&io->io_lock); 3521 list_add_tail(&fio->list, &io->io_list); 3522 spin_unlock(&io->io_lock); 3523 } 3524 3525 mutex_unlock(&curseg->curseg_mutex); 3526 3527 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3528 } 3529 3530 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino, 3531 block_t blkaddr, unsigned int blkcnt) 3532 { 3533 if (!f2fs_is_multi_device(sbi)) 3534 return; 3535 3536 while (1) { 3537 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr); 3538 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1; 3539 3540 /* update device state for fsync */ 3541 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO); 3542 3543 /* update device state for checkpoint */ 3544 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3545 spin_lock(&sbi->dev_lock); 3546 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3547 spin_unlock(&sbi->dev_lock); 3548 } 3549 3550 if (blkcnt <= blks) 3551 break; 3552 blkcnt -= blks; 3553 blkaddr += blks; 3554 } 3555 } 3556 3557 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3558 { 3559 int type = __get_segment_type(fio); 3560 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA); 3561 3562 if (keep_order) 3563 f2fs_down_read(&fio->sbi->io_order_lock); 3564 reallocate: 3565 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3566 &fio->new_blkaddr, sum, type, fio); 3567 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) 3568 f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr); 3569 3570 /* writeout dirty page into bdev */ 3571 f2fs_submit_page_write(fio); 3572 if (fio->retry) { 3573 fio->old_blkaddr = fio->new_blkaddr; 3574 goto reallocate; 3575 } 3576 3577 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1); 3578 3579 if (keep_order) 3580 f2fs_up_read(&fio->sbi->io_order_lock); 3581 } 3582 3583 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 3584 enum iostat_type io_type) 3585 { 3586 struct f2fs_io_info fio = { 3587 .sbi = sbi, 3588 .type = META, 3589 .temp = HOT, 3590 .op = REQ_OP_WRITE, 3591 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3592 .old_blkaddr = page->index, 3593 .new_blkaddr = page->index, 3594 .page = page, 3595 .encrypted_page = NULL, 3596 .in_list = 0, 3597 }; 3598 3599 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 3600 fio.op_flags &= ~REQ_META; 3601 3602 set_page_writeback(page); 3603 f2fs_submit_page_write(&fio); 3604 3605 stat_inc_meta_count(sbi, page->index); 3606 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE); 3607 } 3608 3609 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3610 { 3611 struct f2fs_summary sum; 3612 3613 set_summary(&sum, nid, 0, 0); 3614 do_write_page(&sum, fio); 3615 3616 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE); 3617 } 3618 3619 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3620 struct f2fs_io_info *fio) 3621 { 3622 struct f2fs_sb_info *sbi = fio->sbi; 3623 struct f2fs_summary sum; 3624 3625 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3626 if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO) 3627 f2fs_update_age_extent_cache(dn); 3628 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3629 do_write_page(&sum, fio); 3630 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3631 3632 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE); 3633 } 3634 3635 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3636 { 3637 int err; 3638 struct f2fs_sb_info *sbi = fio->sbi; 3639 unsigned int segno; 3640 3641 fio->new_blkaddr = fio->old_blkaddr; 3642 /* i/o temperature is needed for passing down write hints */ 3643 __get_segment_type(fio); 3644 3645 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3646 3647 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 3648 set_sbi_flag(sbi, SBI_NEED_FSCK); 3649 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 3650 __func__, segno); 3651 err = -EFSCORRUPTED; 3652 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 3653 goto drop_bio; 3654 } 3655 3656 if (f2fs_cp_error(sbi)) { 3657 err = -EIO; 3658 goto drop_bio; 3659 } 3660 3661 if (fio->post_read) 3662 invalidate_mapping_pages(META_MAPPING(sbi), 3663 fio->new_blkaddr, fio->new_blkaddr); 3664 3665 stat_inc_inplace_blocks(fio->sbi); 3666 3667 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi)) 3668 err = f2fs_merge_page_bio(fio); 3669 else 3670 err = f2fs_submit_page_bio(fio); 3671 if (!err) { 3672 f2fs_update_device_state(fio->sbi, fio->ino, 3673 fio->new_blkaddr, 1); 3674 f2fs_update_iostat(fio->sbi, fio->page->mapping->host, 3675 fio->io_type, F2FS_BLKSIZE); 3676 } 3677 3678 return err; 3679 drop_bio: 3680 if (fio->bio && *(fio->bio)) { 3681 struct bio *bio = *(fio->bio); 3682 3683 bio->bi_status = BLK_STS_IOERR; 3684 bio_endio(bio); 3685 *(fio->bio) = NULL; 3686 } 3687 return err; 3688 } 3689 3690 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 3691 unsigned int segno) 3692 { 3693 int i; 3694 3695 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 3696 if (CURSEG_I(sbi, i)->segno == segno) 3697 break; 3698 } 3699 return i; 3700 } 3701 3702 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 3703 block_t old_blkaddr, block_t new_blkaddr, 3704 bool recover_curseg, bool recover_newaddr, 3705 bool from_gc) 3706 { 3707 struct sit_info *sit_i = SIT_I(sbi); 3708 struct curseg_info *curseg; 3709 unsigned int segno, old_cursegno; 3710 struct seg_entry *se; 3711 int type; 3712 unsigned short old_blkoff; 3713 unsigned char old_alloc_type; 3714 3715 segno = GET_SEGNO(sbi, new_blkaddr); 3716 se = get_seg_entry(sbi, segno); 3717 type = se->type; 3718 3719 f2fs_down_write(&SM_I(sbi)->curseg_lock); 3720 3721 if (!recover_curseg) { 3722 /* for recovery flow */ 3723 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 3724 if (old_blkaddr == NULL_ADDR) 3725 type = CURSEG_COLD_DATA; 3726 else 3727 type = CURSEG_WARM_DATA; 3728 } 3729 } else { 3730 if (IS_CURSEG(sbi, segno)) { 3731 /* se->type is volatile as SSR allocation */ 3732 type = __f2fs_get_curseg(sbi, segno); 3733 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 3734 } else { 3735 type = CURSEG_WARM_DATA; 3736 } 3737 } 3738 3739 f2fs_bug_on(sbi, !IS_DATASEG(type)); 3740 curseg = CURSEG_I(sbi, type); 3741 3742 mutex_lock(&curseg->curseg_mutex); 3743 down_write(&sit_i->sentry_lock); 3744 3745 old_cursegno = curseg->segno; 3746 old_blkoff = curseg->next_blkoff; 3747 old_alloc_type = curseg->alloc_type; 3748 3749 /* change the current segment */ 3750 if (segno != curseg->segno) { 3751 curseg->next_segno = segno; 3752 change_curseg(sbi, type); 3753 } 3754 3755 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 3756 curseg->sum_blk->entries[curseg->next_blkoff] = *sum; 3757 3758 if (!recover_curseg || recover_newaddr) { 3759 if (!from_gc) 3760 update_segment_mtime(sbi, new_blkaddr, 0); 3761 update_sit_entry(sbi, new_blkaddr, 1); 3762 } 3763 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 3764 f2fs_invalidate_internal_cache(sbi, old_blkaddr); 3765 if (!from_gc) 3766 update_segment_mtime(sbi, old_blkaddr, 0); 3767 update_sit_entry(sbi, old_blkaddr, -1); 3768 } 3769 3770 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3771 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 3772 3773 locate_dirty_segment(sbi, old_cursegno); 3774 3775 if (recover_curseg) { 3776 if (old_cursegno != curseg->segno) { 3777 curseg->next_segno = old_cursegno; 3778 change_curseg(sbi, type); 3779 } 3780 curseg->next_blkoff = old_blkoff; 3781 curseg->alloc_type = old_alloc_type; 3782 } 3783 3784 up_write(&sit_i->sentry_lock); 3785 mutex_unlock(&curseg->curseg_mutex); 3786 f2fs_up_write(&SM_I(sbi)->curseg_lock); 3787 } 3788 3789 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 3790 block_t old_addr, block_t new_addr, 3791 unsigned char version, bool recover_curseg, 3792 bool recover_newaddr) 3793 { 3794 struct f2fs_summary sum; 3795 3796 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 3797 3798 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 3799 recover_curseg, recover_newaddr, false); 3800 3801 f2fs_update_data_blkaddr(dn, new_addr); 3802 } 3803 3804 void f2fs_wait_on_page_writeback(struct page *page, 3805 enum page_type type, bool ordered, bool locked) 3806 { 3807 if (PageWriteback(page)) { 3808 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 3809 3810 /* submit cached LFS IO */ 3811 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type); 3812 /* submit cached IPU IO */ 3813 f2fs_submit_merged_ipu_write(sbi, NULL, page); 3814 if (ordered) { 3815 wait_on_page_writeback(page); 3816 f2fs_bug_on(sbi, locked && PageWriteback(page)); 3817 } else { 3818 wait_for_stable_page(page); 3819 } 3820 } 3821 } 3822 3823 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 3824 { 3825 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3826 struct page *cpage; 3827 3828 if (!f2fs_post_read_required(inode)) 3829 return; 3830 3831 if (!__is_valid_data_blkaddr(blkaddr)) 3832 return; 3833 3834 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3835 if (cpage) { 3836 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 3837 f2fs_put_page(cpage, 1); 3838 } 3839 } 3840 3841 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 3842 block_t len) 3843 { 3844 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3845 block_t i; 3846 3847 if (!f2fs_post_read_required(inode)) 3848 return; 3849 3850 for (i = 0; i < len; i++) 3851 f2fs_wait_on_block_writeback(inode, blkaddr + i); 3852 3853 invalidate_mapping_pages(META_MAPPING(sbi), blkaddr, blkaddr + len - 1); 3854 } 3855 3856 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 3857 { 3858 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3859 struct curseg_info *seg_i; 3860 unsigned char *kaddr; 3861 struct page *page; 3862 block_t start; 3863 int i, j, offset; 3864 3865 start = start_sum_block(sbi); 3866 3867 page = f2fs_get_meta_page(sbi, start++); 3868 if (IS_ERR(page)) 3869 return PTR_ERR(page); 3870 kaddr = (unsigned char *)page_address(page); 3871 3872 /* Step 1: restore nat cache */ 3873 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3874 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3875 3876 /* Step 2: restore sit cache */ 3877 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3878 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3879 offset = 2 * SUM_JOURNAL_SIZE; 3880 3881 /* Step 3: restore summary entries */ 3882 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3883 unsigned short blk_off; 3884 unsigned int segno; 3885 3886 seg_i = CURSEG_I(sbi, i); 3887 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3888 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3889 seg_i->next_segno = segno; 3890 reset_curseg(sbi, i, 0); 3891 seg_i->alloc_type = ckpt->alloc_type[i]; 3892 seg_i->next_blkoff = blk_off; 3893 3894 if (seg_i->alloc_type == SSR) 3895 blk_off = sbi->blocks_per_seg; 3896 3897 for (j = 0; j < blk_off; j++) { 3898 struct f2fs_summary *s; 3899 3900 s = (struct f2fs_summary *)(kaddr + offset); 3901 seg_i->sum_blk->entries[j] = *s; 3902 offset += SUMMARY_SIZE; 3903 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3904 SUM_FOOTER_SIZE) 3905 continue; 3906 3907 f2fs_put_page(page, 1); 3908 page = NULL; 3909 3910 page = f2fs_get_meta_page(sbi, start++); 3911 if (IS_ERR(page)) 3912 return PTR_ERR(page); 3913 kaddr = (unsigned char *)page_address(page); 3914 offset = 0; 3915 } 3916 } 3917 f2fs_put_page(page, 1); 3918 return 0; 3919 } 3920 3921 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3922 { 3923 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3924 struct f2fs_summary_block *sum; 3925 struct curseg_info *curseg; 3926 struct page *new; 3927 unsigned short blk_off; 3928 unsigned int segno = 0; 3929 block_t blk_addr = 0; 3930 int err = 0; 3931 3932 /* get segment number and block addr */ 3933 if (IS_DATASEG(type)) { 3934 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3935 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3936 CURSEG_HOT_DATA]); 3937 if (__exist_node_summaries(sbi)) 3938 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type); 3939 else 3940 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3941 } else { 3942 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3943 CURSEG_HOT_NODE]); 3944 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3945 CURSEG_HOT_NODE]); 3946 if (__exist_node_summaries(sbi)) 3947 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3948 type - CURSEG_HOT_NODE); 3949 else 3950 blk_addr = GET_SUM_BLOCK(sbi, segno); 3951 } 3952 3953 new = f2fs_get_meta_page(sbi, blk_addr); 3954 if (IS_ERR(new)) 3955 return PTR_ERR(new); 3956 sum = (struct f2fs_summary_block *)page_address(new); 3957 3958 if (IS_NODESEG(type)) { 3959 if (__exist_node_summaries(sbi)) { 3960 struct f2fs_summary *ns = &sum->entries[0]; 3961 int i; 3962 3963 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3964 ns->version = 0; 3965 ns->ofs_in_node = 0; 3966 } 3967 } else { 3968 err = f2fs_restore_node_summary(sbi, segno, sum); 3969 if (err) 3970 goto out; 3971 } 3972 } 3973 3974 /* set uncompleted segment to curseg */ 3975 curseg = CURSEG_I(sbi, type); 3976 mutex_lock(&curseg->curseg_mutex); 3977 3978 /* update journal info */ 3979 down_write(&curseg->journal_rwsem); 3980 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3981 up_write(&curseg->journal_rwsem); 3982 3983 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3984 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3985 curseg->next_segno = segno; 3986 reset_curseg(sbi, type, 0); 3987 curseg->alloc_type = ckpt->alloc_type[type]; 3988 curseg->next_blkoff = blk_off; 3989 mutex_unlock(&curseg->curseg_mutex); 3990 out: 3991 f2fs_put_page(new, 1); 3992 return err; 3993 } 3994 3995 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3996 { 3997 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3998 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3999 int type = CURSEG_HOT_DATA; 4000 int err; 4001 4002 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 4003 int npages = f2fs_npages_for_summary_flush(sbi, true); 4004 4005 if (npages >= 2) 4006 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 4007 META_CP, true); 4008 4009 /* restore for compacted data summary */ 4010 err = read_compacted_summaries(sbi); 4011 if (err) 4012 return err; 4013 type = CURSEG_HOT_NODE; 4014 } 4015 4016 if (__exist_node_summaries(sbi)) 4017 f2fs_ra_meta_pages(sbi, 4018 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type), 4019 NR_CURSEG_PERSIST_TYPE - type, META_CP, true); 4020 4021 for (; type <= CURSEG_COLD_NODE; type++) { 4022 err = read_normal_summaries(sbi, type); 4023 if (err) 4024 return err; 4025 } 4026 4027 /* sanity check for summary blocks */ 4028 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 4029 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 4030 f2fs_err(sbi, "invalid journal entries nats %u sits %u", 4031 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 4032 return -EINVAL; 4033 } 4034 4035 return 0; 4036 } 4037 4038 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 4039 { 4040 struct page *page; 4041 unsigned char *kaddr; 4042 struct f2fs_summary *summary; 4043 struct curseg_info *seg_i; 4044 int written_size = 0; 4045 int i, j; 4046 4047 page = f2fs_grab_meta_page(sbi, blkaddr++); 4048 kaddr = (unsigned char *)page_address(page); 4049 memset(kaddr, 0, PAGE_SIZE); 4050 4051 /* Step 1: write nat cache */ 4052 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4053 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 4054 written_size += SUM_JOURNAL_SIZE; 4055 4056 /* Step 2: write sit cache */ 4057 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4058 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 4059 written_size += SUM_JOURNAL_SIZE; 4060 4061 /* Step 3: write summary entries */ 4062 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4063 seg_i = CURSEG_I(sbi, i); 4064 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) { 4065 if (!page) { 4066 page = f2fs_grab_meta_page(sbi, blkaddr++); 4067 kaddr = (unsigned char *)page_address(page); 4068 memset(kaddr, 0, PAGE_SIZE); 4069 written_size = 0; 4070 } 4071 summary = (struct f2fs_summary *)(kaddr + written_size); 4072 *summary = seg_i->sum_blk->entries[j]; 4073 written_size += SUMMARY_SIZE; 4074 4075 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 4076 SUM_FOOTER_SIZE) 4077 continue; 4078 4079 set_page_dirty(page); 4080 f2fs_put_page(page, 1); 4081 page = NULL; 4082 } 4083 } 4084 if (page) { 4085 set_page_dirty(page); 4086 f2fs_put_page(page, 1); 4087 } 4088 } 4089 4090 static void write_normal_summaries(struct f2fs_sb_info *sbi, 4091 block_t blkaddr, int type) 4092 { 4093 int i, end; 4094 4095 if (IS_DATASEG(type)) 4096 end = type + NR_CURSEG_DATA_TYPE; 4097 else 4098 end = type + NR_CURSEG_NODE_TYPE; 4099 4100 for (i = type; i < end; i++) 4101 write_current_sum_page(sbi, i, blkaddr + (i - type)); 4102 } 4103 4104 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4105 { 4106 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 4107 write_compacted_summaries(sbi, start_blk); 4108 else 4109 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 4110 } 4111 4112 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4113 { 4114 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 4115 } 4116 4117 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 4118 unsigned int val, int alloc) 4119 { 4120 int i; 4121 4122 if (type == NAT_JOURNAL) { 4123 for (i = 0; i < nats_in_cursum(journal); i++) { 4124 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 4125 return i; 4126 } 4127 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 4128 return update_nats_in_cursum(journal, 1); 4129 } else if (type == SIT_JOURNAL) { 4130 for (i = 0; i < sits_in_cursum(journal); i++) 4131 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 4132 return i; 4133 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 4134 return update_sits_in_cursum(journal, 1); 4135 } 4136 return -1; 4137 } 4138 4139 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 4140 unsigned int segno) 4141 { 4142 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno)); 4143 } 4144 4145 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 4146 unsigned int start) 4147 { 4148 struct sit_info *sit_i = SIT_I(sbi); 4149 struct page *page; 4150 pgoff_t src_off, dst_off; 4151 4152 src_off = current_sit_addr(sbi, start); 4153 dst_off = next_sit_addr(sbi, src_off); 4154 4155 page = f2fs_grab_meta_page(sbi, dst_off); 4156 seg_info_to_sit_page(sbi, page, start); 4157 4158 set_page_dirty(page); 4159 set_to_next_sit(sit_i, start); 4160 4161 return page; 4162 } 4163 4164 static struct sit_entry_set *grab_sit_entry_set(void) 4165 { 4166 struct sit_entry_set *ses = 4167 f2fs_kmem_cache_alloc(sit_entry_set_slab, 4168 GFP_NOFS, true, NULL); 4169 4170 ses->entry_cnt = 0; 4171 INIT_LIST_HEAD(&ses->set_list); 4172 return ses; 4173 } 4174 4175 static void release_sit_entry_set(struct sit_entry_set *ses) 4176 { 4177 list_del(&ses->set_list); 4178 kmem_cache_free(sit_entry_set_slab, ses); 4179 } 4180 4181 static void adjust_sit_entry_set(struct sit_entry_set *ses, 4182 struct list_head *head) 4183 { 4184 struct sit_entry_set *next = ses; 4185 4186 if (list_is_last(&ses->set_list, head)) 4187 return; 4188 4189 list_for_each_entry_continue(next, head, set_list) 4190 if (ses->entry_cnt <= next->entry_cnt) { 4191 list_move_tail(&ses->set_list, &next->set_list); 4192 return; 4193 } 4194 4195 list_move_tail(&ses->set_list, head); 4196 } 4197 4198 static void add_sit_entry(unsigned int segno, struct list_head *head) 4199 { 4200 struct sit_entry_set *ses; 4201 unsigned int start_segno = START_SEGNO(segno); 4202 4203 list_for_each_entry(ses, head, set_list) { 4204 if (ses->start_segno == start_segno) { 4205 ses->entry_cnt++; 4206 adjust_sit_entry_set(ses, head); 4207 return; 4208 } 4209 } 4210 4211 ses = grab_sit_entry_set(); 4212 4213 ses->start_segno = start_segno; 4214 ses->entry_cnt++; 4215 list_add(&ses->set_list, head); 4216 } 4217 4218 static void add_sits_in_set(struct f2fs_sb_info *sbi) 4219 { 4220 struct f2fs_sm_info *sm_info = SM_I(sbi); 4221 struct list_head *set_list = &sm_info->sit_entry_set; 4222 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 4223 unsigned int segno; 4224 4225 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 4226 add_sit_entry(segno, set_list); 4227 } 4228 4229 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 4230 { 4231 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4232 struct f2fs_journal *journal = curseg->journal; 4233 int i; 4234 4235 down_write(&curseg->journal_rwsem); 4236 for (i = 0; i < sits_in_cursum(journal); i++) { 4237 unsigned int segno; 4238 bool dirtied; 4239 4240 segno = le32_to_cpu(segno_in_journal(journal, i)); 4241 dirtied = __mark_sit_entry_dirty(sbi, segno); 4242 4243 if (!dirtied) 4244 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 4245 } 4246 update_sits_in_cursum(journal, -i); 4247 up_write(&curseg->journal_rwsem); 4248 } 4249 4250 /* 4251 * CP calls this function, which flushes SIT entries including sit_journal, 4252 * and moves prefree segs to free segs. 4253 */ 4254 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 4255 { 4256 struct sit_info *sit_i = SIT_I(sbi); 4257 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 4258 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4259 struct f2fs_journal *journal = curseg->journal; 4260 struct sit_entry_set *ses, *tmp; 4261 struct list_head *head = &SM_I(sbi)->sit_entry_set; 4262 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 4263 struct seg_entry *se; 4264 4265 down_write(&sit_i->sentry_lock); 4266 4267 if (!sit_i->dirty_sentries) 4268 goto out; 4269 4270 /* 4271 * add and account sit entries of dirty bitmap in sit entry 4272 * set temporarily 4273 */ 4274 add_sits_in_set(sbi); 4275 4276 /* 4277 * if there are no enough space in journal to store dirty sit 4278 * entries, remove all entries from journal and add and account 4279 * them in sit entry set. 4280 */ 4281 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 4282 !to_journal) 4283 remove_sits_in_journal(sbi); 4284 4285 /* 4286 * there are two steps to flush sit entries: 4287 * #1, flush sit entries to journal in current cold data summary block. 4288 * #2, flush sit entries to sit page. 4289 */ 4290 list_for_each_entry_safe(ses, tmp, head, set_list) { 4291 struct page *page = NULL; 4292 struct f2fs_sit_block *raw_sit = NULL; 4293 unsigned int start_segno = ses->start_segno; 4294 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 4295 (unsigned long)MAIN_SEGS(sbi)); 4296 unsigned int segno = start_segno; 4297 4298 if (to_journal && 4299 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 4300 to_journal = false; 4301 4302 if (to_journal) { 4303 down_write(&curseg->journal_rwsem); 4304 } else { 4305 page = get_next_sit_page(sbi, start_segno); 4306 raw_sit = page_address(page); 4307 } 4308 4309 /* flush dirty sit entries in region of current sit set */ 4310 for_each_set_bit_from(segno, bitmap, end) { 4311 int offset, sit_offset; 4312 4313 se = get_seg_entry(sbi, segno); 4314 #ifdef CONFIG_F2FS_CHECK_FS 4315 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 4316 SIT_VBLOCK_MAP_SIZE)) 4317 f2fs_bug_on(sbi, 1); 4318 #endif 4319 4320 /* add discard candidates */ 4321 if (!(cpc->reason & CP_DISCARD)) { 4322 cpc->trim_start = segno; 4323 add_discard_addrs(sbi, cpc, false); 4324 } 4325 4326 if (to_journal) { 4327 offset = f2fs_lookup_journal_in_cursum(journal, 4328 SIT_JOURNAL, segno, 1); 4329 f2fs_bug_on(sbi, offset < 0); 4330 segno_in_journal(journal, offset) = 4331 cpu_to_le32(segno); 4332 seg_info_to_raw_sit(se, 4333 &sit_in_journal(journal, offset)); 4334 check_block_count(sbi, segno, 4335 &sit_in_journal(journal, offset)); 4336 } else { 4337 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 4338 seg_info_to_raw_sit(se, 4339 &raw_sit->entries[sit_offset]); 4340 check_block_count(sbi, segno, 4341 &raw_sit->entries[sit_offset]); 4342 } 4343 4344 __clear_bit(segno, bitmap); 4345 sit_i->dirty_sentries--; 4346 ses->entry_cnt--; 4347 } 4348 4349 if (to_journal) 4350 up_write(&curseg->journal_rwsem); 4351 else 4352 f2fs_put_page(page, 1); 4353 4354 f2fs_bug_on(sbi, ses->entry_cnt); 4355 release_sit_entry_set(ses); 4356 } 4357 4358 f2fs_bug_on(sbi, !list_empty(head)); 4359 f2fs_bug_on(sbi, sit_i->dirty_sentries); 4360 out: 4361 if (cpc->reason & CP_DISCARD) { 4362 __u64 trim_start = cpc->trim_start; 4363 4364 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 4365 add_discard_addrs(sbi, cpc, false); 4366 4367 cpc->trim_start = trim_start; 4368 } 4369 up_write(&sit_i->sentry_lock); 4370 4371 set_prefree_as_free_segments(sbi); 4372 } 4373 4374 static int build_sit_info(struct f2fs_sb_info *sbi) 4375 { 4376 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4377 struct sit_info *sit_i; 4378 unsigned int sit_segs, start; 4379 char *src_bitmap, *bitmap; 4380 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 4381 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0; 4382 4383 /* allocate memory for SIT information */ 4384 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4385 if (!sit_i) 4386 return -ENOMEM; 4387 4388 SM_I(sbi)->sit_info = sit_i; 4389 4390 sit_i->sentries = 4391 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4392 MAIN_SEGS(sbi)), 4393 GFP_KERNEL); 4394 if (!sit_i->sentries) 4395 return -ENOMEM; 4396 4397 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4398 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4399 GFP_KERNEL); 4400 if (!sit_i->dirty_sentries_bitmap) 4401 return -ENOMEM; 4402 4403 #ifdef CONFIG_F2FS_CHECK_FS 4404 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map); 4405 #else 4406 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map); 4407 #endif 4408 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4409 if (!sit_i->bitmap) 4410 return -ENOMEM; 4411 4412 bitmap = sit_i->bitmap; 4413 4414 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4415 sit_i->sentries[start].cur_valid_map = bitmap; 4416 bitmap += SIT_VBLOCK_MAP_SIZE; 4417 4418 sit_i->sentries[start].ckpt_valid_map = bitmap; 4419 bitmap += SIT_VBLOCK_MAP_SIZE; 4420 4421 #ifdef CONFIG_F2FS_CHECK_FS 4422 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4423 bitmap += SIT_VBLOCK_MAP_SIZE; 4424 #endif 4425 4426 if (discard_map) { 4427 sit_i->sentries[start].discard_map = bitmap; 4428 bitmap += SIT_VBLOCK_MAP_SIZE; 4429 } 4430 } 4431 4432 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4433 if (!sit_i->tmp_map) 4434 return -ENOMEM; 4435 4436 if (__is_large_section(sbi)) { 4437 sit_i->sec_entries = 4438 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4439 MAIN_SECS(sbi)), 4440 GFP_KERNEL); 4441 if (!sit_i->sec_entries) 4442 return -ENOMEM; 4443 } 4444 4445 /* get information related with SIT */ 4446 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4447 4448 /* setup SIT bitmap from ckeckpoint pack */ 4449 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4450 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4451 4452 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4453 if (!sit_i->sit_bitmap) 4454 return -ENOMEM; 4455 4456 #ifdef CONFIG_F2FS_CHECK_FS 4457 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4458 sit_bitmap_size, GFP_KERNEL); 4459 if (!sit_i->sit_bitmap_mir) 4460 return -ENOMEM; 4461 4462 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4463 main_bitmap_size, GFP_KERNEL); 4464 if (!sit_i->invalid_segmap) 4465 return -ENOMEM; 4466 #endif 4467 4468 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4469 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 4470 sit_i->written_valid_blocks = 0; 4471 sit_i->bitmap_size = sit_bitmap_size; 4472 sit_i->dirty_sentries = 0; 4473 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4474 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4475 sit_i->mounted_time = ktime_get_boottime_seconds(); 4476 init_rwsem(&sit_i->sentry_lock); 4477 return 0; 4478 } 4479 4480 static int build_free_segmap(struct f2fs_sb_info *sbi) 4481 { 4482 struct free_segmap_info *free_i; 4483 unsigned int bitmap_size, sec_bitmap_size; 4484 4485 /* allocate memory for free segmap information */ 4486 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4487 if (!free_i) 4488 return -ENOMEM; 4489 4490 SM_I(sbi)->free_info = free_i; 4491 4492 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4493 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4494 if (!free_i->free_segmap) 4495 return -ENOMEM; 4496 4497 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4498 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4499 if (!free_i->free_secmap) 4500 return -ENOMEM; 4501 4502 /* set all segments as dirty temporarily */ 4503 memset(free_i->free_segmap, 0xff, bitmap_size); 4504 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4505 4506 /* init free segmap information */ 4507 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4508 free_i->free_segments = 0; 4509 free_i->free_sections = 0; 4510 spin_lock_init(&free_i->segmap_lock); 4511 return 0; 4512 } 4513 4514 static int build_curseg(struct f2fs_sb_info *sbi) 4515 { 4516 struct curseg_info *array; 4517 int i; 4518 4519 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, 4520 sizeof(*array)), GFP_KERNEL); 4521 if (!array) 4522 return -ENOMEM; 4523 4524 SM_I(sbi)->curseg_array = array; 4525 4526 for (i = 0; i < NO_CHECK_TYPE; i++) { 4527 mutex_init(&array[i].curseg_mutex); 4528 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4529 if (!array[i].sum_blk) 4530 return -ENOMEM; 4531 init_rwsem(&array[i].journal_rwsem); 4532 array[i].journal = f2fs_kzalloc(sbi, 4533 sizeof(struct f2fs_journal), GFP_KERNEL); 4534 if (!array[i].journal) 4535 return -ENOMEM; 4536 if (i < NR_PERSISTENT_LOG) 4537 array[i].seg_type = CURSEG_HOT_DATA + i; 4538 else if (i == CURSEG_COLD_DATA_PINNED) 4539 array[i].seg_type = CURSEG_COLD_DATA; 4540 else if (i == CURSEG_ALL_DATA_ATGC) 4541 array[i].seg_type = CURSEG_COLD_DATA; 4542 array[i].segno = NULL_SEGNO; 4543 array[i].next_blkoff = 0; 4544 array[i].inited = false; 4545 } 4546 return restore_curseg_summaries(sbi); 4547 } 4548 4549 static int build_sit_entries(struct f2fs_sb_info *sbi) 4550 { 4551 struct sit_info *sit_i = SIT_I(sbi); 4552 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4553 struct f2fs_journal *journal = curseg->journal; 4554 struct seg_entry *se; 4555 struct f2fs_sit_entry sit; 4556 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4557 unsigned int i, start, end; 4558 unsigned int readed, start_blk = 0; 4559 int err = 0; 4560 block_t sit_valid_blocks[2] = {0, 0}; 4561 4562 do { 4563 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS, 4564 META_SIT, true); 4565 4566 start = start_blk * sit_i->sents_per_block; 4567 end = (start_blk + readed) * sit_i->sents_per_block; 4568 4569 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4570 struct f2fs_sit_block *sit_blk; 4571 struct page *page; 4572 4573 se = &sit_i->sentries[start]; 4574 page = get_current_sit_page(sbi, start); 4575 if (IS_ERR(page)) 4576 return PTR_ERR(page); 4577 sit_blk = (struct f2fs_sit_block *)page_address(page); 4578 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4579 f2fs_put_page(page, 1); 4580 4581 err = check_block_count(sbi, start, &sit); 4582 if (err) 4583 return err; 4584 seg_info_from_raw_sit(se, &sit); 4585 4586 if (se->type >= NR_PERSISTENT_LOG) { 4587 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4588 se->type, start); 4589 f2fs_handle_error(sbi, 4590 ERROR_INCONSISTENT_SUM_TYPE); 4591 return -EFSCORRUPTED; 4592 } 4593 4594 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4595 4596 if (f2fs_block_unit_discard(sbi)) { 4597 /* build discard map only one time */ 4598 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4599 memset(se->discard_map, 0xff, 4600 SIT_VBLOCK_MAP_SIZE); 4601 } else { 4602 memcpy(se->discard_map, 4603 se->cur_valid_map, 4604 SIT_VBLOCK_MAP_SIZE); 4605 sbi->discard_blks += 4606 sbi->blocks_per_seg - 4607 se->valid_blocks; 4608 } 4609 } 4610 4611 if (__is_large_section(sbi)) 4612 get_sec_entry(sbi, start)->valid_blocks += 4613 se->valid_blocks; 4614 } 4615 start_blk += readed; 4616 } while (start_blk < sit_blk_cnt); 4617 4618 down_read(&curseg->journal_rwsem); 4619 for (i = 0; i < sits_in_cursum(journal); i++) { 4620 unsigned int old_valid_blocks; 4621 4622 start = le32_to_cpu(segno_in_journal(journal, i)); 4623 if (start >= MAIN_SEGS(sbi)) { 4624 f2fs_err(sbi, "Wrong journal entry on segno %u", 4625 start); 4626 err = -EFSCORRUPTED; 4627 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL); 4628 break; 4629 } 4630 4631 se = &sit_i->sentries[start]; 4632 sit = sit_in_journal(journal, i); 4633 4634 old_valid_blocks = se->valid_blocks; 4635 4636 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks; 4637 4638 err = check_block_count(sbi, start, &sit); 4639 if (err) 4640 break; 4641 seg_info_from_raw_sit(se, &sit); 4642 4643 if (se->type >= NR_PERSISTENT_LOG) { 4644 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4645 se->type, start); 4646 err = -EFSCORRUPTED; 4647 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 4648 break; 4649 } 4650 4651 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4652 4653 if (f2fs_block_unit_discard(sbi)) { 4654 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4655 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 4656 } else { 4657 memcpy(se->discard_map, se->cur_valid_map, 4658 SIT_VBLOCK_MAP_SIZE); 4659 sbi->discard_blks += old_valid_blocks; 4660 sbi->discard_blks -= se->valid_blocks; 4661 } 4662 } 4663 4664 if (__is_large_section(sbi)) { 4665 get_sec_entry(sbi, start)->valid_blocks += 4666 se->valid_blocks; 4667 get_sec_entry(sbi, start)->valid_blocks -= 4668 old_valid_blocks; 4669 } 4670 } 4671 up_read(&curseg->journal_rwsem); 4672 4673 if (err) 4674 return err; 4675 4676 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) { 4677 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 4678 sit_valid_blocks[NODE], valid_node_count(sbi)); 4679 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT); 4680 return -EFSCORRUPTED; 4681 } 4682 4683 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] > 4684 valid_user_blocks(sbi)) { 4685 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u", 4686 sit_valid_blocks[DATA], sit_valid_blocks[NODE], 4687 valid_user_blocks(sbi)); 4688 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT); 4689 return -EFSCORRUPTED; 4690 } 4691 4692 return 0; 4693 } 4694 4695 static void init_free_segmap(struct f2fs_sb_info *sbi) 4696 { 4697 unsigned int start; 4698 int type; 4699 struct seg_entry *sentry; 4700 4701 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4702 if (f2fs_usable_blks_in_seg(sbi, start) == 0) 4703 continue; 4704 sentry = get_seg_entry(sbi, start); 4705 if (!sentry->valid_blocks) 4706 __set_free(sbi, start); 4707 else 4708 SIT_I(sbi)->written_valid_blocks += 4709 sentry->valid_blocks; 4710 } 4711 4712 /* set use the current segments */ 4713 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 4714 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 4715 4716 __set_test_and_inuse(sbi, curseg_t->segno); 4717 } 4718 } 4719 4720 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 4721 { 4722 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4723 struct free_segmap_info *free_i = FREE_I(sbi); 4724 unsigned int segno = 0, offset = 0, secno; 4725 block_t valid_blocks, usable_blks_in_seg; 4726 4727 while (1) { 4728 /* find dirty segment based on free segmap */ 4729 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 4730 if (segno >= MAIN_SEGS(sbi)) 4731 break; 4732 offset = segno + 1; 4733 valid_blocks = get_valid_blocks(sbi, segno, false); 4734 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 4735 if (valid_blocks == usable_blks_in_seg || !valid_blocks) 4736 continue; 4737 if (valid_blocks > usable_blks_in_seg) { 4738 f2fs_bug_on(sbi, 1); 4739 continue; 4740 } 4741 mutex_lock(&dirty_i->seglist_lock); 4742 __locate_dirty_segment(sbi, segno, DIRTY); 4743 mutex_unlock(&dirty_i->seglist_lock); 4744 } 4745 4746 if (!__is_large_section(sbi)) 4747 return; 4748 4749 mutex_lock(&dirty_i->seglist_lock); 4750 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 4751 valid_blocks = get_valid_blocks(sbi, segno, true); 4752 secno = GET_SEC_FROM_SEG(sbi, segno); 4753 4754 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi)) 4755 continue; 4756 if (IS_CURSEC(sbi, secno)) 4757 continue; 4758 set_bit(secno, dirty_i->dirty_secmap); 4759 } 4760 mutex_unlock(&dirty_i->seglist_lock); 4761 } 4762 4763 static int init_victim_secmap(struct f2fs_sb_info *sbi) 4764 { 4765 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4766 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4767 4768 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4769 if (!dirty_i->victim_secmap) 4770 return -ENOMEM; 4771 4772 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4773 if (!dirty_i->pinned_secmap) 4774 return -ENOMEM; 4775 4776 dirty_i->pinned_secmap_cnt = 0; 4777 dirty_i->enable_pin_section = true; 4778 return 0; 4779 } 4780 4781 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 4782 { 4783 struct dirty_seglist_info *dirty_i; 4784 unsigned int bitmap_size, i; 4785 4786 /* allocate memory for dirty segments list information */ 4787 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 4788 GFP_KERNEL); 4789 if (!dirty_i) 4790 return -ENOMEM; 4791 4792 SM_I(sbi)->dirty_info = dirty_i; 4793 mutex_init(&dirty_i->seglist_lock); 4794 4795 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4796 4797 for (i = 0; i < NR_DIRTY_TYPE; i++) { 4798 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 4799 GFP_KERNEL); 4800 if (!dirty_i->dirty_segmap[i]) 4801 return -ENOMEM; 4802 } 4803 4804 if (__is_large_section(sbi)) { 4805 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4806 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi, 4807 bitmap_size, GFP_KERNEL); 4808 if (!dirty_i->dirty_secmap) 4809 return -ENOMEM; 4810 } 4811 4812 init_dirty_segmap(sbi); 4813 return init_victim_secmap(sbi); 4814 } 4815 4816 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 4817 { 4818 int i; 4819 4820 /* 4821 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 4822 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 4823 */ 4824 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 4825 struct curseg_info *curseg = CURSEG_I(sbi, i); 4826 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 4827 unsigned int blkofs = curseg->next_blkoff; 4828 4829 if (f2fs_sb_has_readonly(sbi) && 4830 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE) 4831 continue; 4832 4833 sanity_check_seg_type(sbi, curseg->seg_type); 4834 4835 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) { 4836 f2fs_err(sbi, 4837 "Current segment has invalid alloc_type:%d", 4838 curseg->alloc_type); 4839 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 4840 return -EFSCORRUPTED; 4841 } 4842 4843 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 4844 goto out; 4845 4846 if (curseg->alloc_type == SSR) 4847 continue; 4848 4849 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) { 4850 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 4851 continue; 4852 out: 4853 f2fs_err(sbi, 4854 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 4855 i, curseg->segno, curseg->alloc_type, 4856 curseg->next_blkoff, blkofs); 4857 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 4858 return -EFSCORRUPTED; 4859 } 4860 } 4861 return 0; 4862 } 4863 4864 #ifdef CONFIG_BLK_DEV_ZONED 4865 4866 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 4867 struct f2fs_dev_info *fdev, 4868 struct blk_zone *zone) 4869 { 4870 unsigned int zone_segno; 4871 block_t zone_block, valid_block_cnt; 4872 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4873 int ret; 4874 unsigned int nofs_flags; 4875 4876 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4877 return 0; 4878 4879 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 4880 zone_segno = GET_SEGNO(sbi, zone_block); 4881 4882 /* 4883 * Skip check of zones cursegs point to, since 4884 * fix_curseg_write_pointer() checks them. 4885 */ 4886 if (zone_segno >= MAIN_SEGS(sbi) || 4887 IS_CURSEC(sbi, GET_SEC_FROM_SEG(sbi, zone_segno))) 4888 return 0; 4889 4890 /* 4891 * Get # of valid block of the zone. 4892 */ 4893 valid_block_cnt = get_valid_blocks(sbi, zone_segno, true); 4894 4895 if ((!valid_block_cnt && zone->cond == BLK_ZONE_COND_EMPTY) || 4896 (valid_block_cnt && zone->cond == BLK_ZONE_COND_FULL)) 4897 return 0; 4898 4899 if (!valid_block_cnt) { 4900 f2fs_notice(sbi, "Zone without valid block has non-zero write " 4901 "pointer. Reset the write pointer: cond[0x%x]", 4902 zone->cond); 4903 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 4904 zone->len >> log_sectors_per_block); 4905 if (ret) 4906 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4907 fdev->path, ret); 4908 return ret; 4909 } 4910 4911 /* 4912 * If there are valid blocks and the write pointer doesn't match 4913 * with them, we need to report the inconsistency and fill 4914 * the zone till the end to close the zone. This inconsistency 4915 * does not cause write error because the zone will not be 4916 * selected for write operation until it get discarded. 4917 */ 4918 f2fs_notice(sbi, "Valid blocks are not aligned with write " 4919 "pointer: valid block[0x%x,0x%x] cond[0x%x]", 4920 zone_segno, valid_block_cnt, zone->cond); 4921 4922 nofs_flags = memalloc_nofs_save(); 4923 ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH, 4924 zone->start, zone->len); 4925 memalloc_nofs_restore(nofs_flags); 4926 if (ret == -EOPNOTSUPP) { 4927 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp, 4928 zone->len - (zone->wp - zone->start), 4929 GFP_NOFS, 0); 4930 if (ret) 4931 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)", 4932 fdev->path, ret); 4933 } else if (ret) { 4934 f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)", 4935 fdev->path, ret); 4936 } 4937 4938 return ret; 4939 } 4940 4941 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 4942 block_t zone_blkaddr) 4943 { 4944 int i; 4945 4946 for (i = 0; i < sbi->s_ndevs; i++) { 4947 if (!bdev_is_zoned(FDEV(i).bdev)) 4948 continue; 4949 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 4950 zone_blkaddr <= FDEV(i).end_blk)) 4951 return &FDEV(i); 4952 } 4953 4954 return NULL; 4955 } 4956 4957 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 4958 void *data) 4959 { 4960 memcpy(data, zone, sizeof(struct blk_zone)); 4961 return 0; 4962 } 4963 4964 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 4965 { 4966 struct curseg_info *cs = CURSEG_I(sbi, type); 4967 struct f2fs_dev_info *zbd; 4968 struct blk_zone zone; 4969 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 4970 block_t cs_zone_block, wp_block; 4971 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4972 sector_t zone_sector; 4973 int err; 4974 4975 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4976 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4977 4978 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4979 if (!zbd) 4980 return 0; 4981 4982 /* report zone for the sector the curseg points to */ 4983 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4984 << log_sectors_per_block; 4985 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4986 report_one_zone_cb, &zone); 4987 if (err != 1) { 4988 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4989 zbd->path, err); 4990 return err; 4991 } 4992 4993 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4994 return 0; 4995 4996 /* 4997 * When safely unmounted in the previous mount, we could use current 4998 * segments. Otherwise, allocate new sections. 4999 */ 5000 if (is_set_ckpt_flags(sbi, CP_UMOUNT_FLAG)) { 5001 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 5002 wp_segno = GET_SEGNO(sbi, wp_block); 5003 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 5004 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 5005 5006 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 5007 wp_sector_off == 0) 5008 return 0; 5009 5010 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 5011 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", type, cs->segno, 5012 cs->next_blkoff, wp_segno, wp_blkoff); 5013 } 5014 5015 /* Allocate a new section if it's not new. */ 5016 if (cs->next_blkoff) { 5017 unsigned int old_segno = cs->segno, old_blkoff = cs->next_blkoff; 5018 5019 f2fs_allocate_new_section(sbi, type, true); 5020 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 5021 "[0x%x,0x%x] -> [0x%x,0x%x]", 5022 type, old_segno, old_blkoff, 5023 cs->segno, cs->next_blkoff); 5024 } 5025 5026 /* check consistency of the zone curseg pointed to */ 5027 if (check_zone_write_pointer(sbi, zbd, &zone)) 5028 return -EIO; 5029 5030 /* check newly assigned zone */ 5031 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5032 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5033 5034 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5035 if (!zbd) 5036 return 0; 5037 5038 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5039 << log_sectors_per_block; 5040 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5041 report_one_zone_cb, &zone); 5042 if (err != 1) { 5043 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5044 zbd->path, err); 5045 return err; 5046 } 5047 5048 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5049 return 0; 5050 5051 if (zone.wp != zone.start) { 5052 f2fs_notice(sbi, 5053 "New zone for curseg[%d] is not yet discarded. " 5054 "Reset the zone: curseg[0x%x,0x%x]", 5055 type, cs->segno, cs->next_blkoff); 5056 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block, 5057 zone.len >> log_sectors_per_block); 5058 if (err) { 5059 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5060 zbd->path, err); 5061 return err; 5062 } 5063 } 5064 5065 return 0; 5066 } 5067 5068 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5069 { 5070 int i, ret; 5071 5072 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5073 ret = fix_curseg_write_pointer(sbi, i); 5074 if (ret) 5075 return ret; 5076 } 5077 5078 return 0; 5079 } 5080 5081 struct check_zone_write_pointer_args { 5082 struct f2fs_sb_info *sbi; 5083 struct f2fs_dev_info *fdev; 5084 }; 5085 5086 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 5087 void *data) 5088 { 5089 struct check_zone_write_pointer_args *args; 5090 5091 args = (struct check_zone_write_pointer_args *)data; 5092 5093 return check_zone_write_pointer(args->sbi, args->fdev, zone); 5094 } 5095 5096 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5097 { 5098 int i, ret; 5099 struct check_zone_write_pointer_args args; 5100 5101 for (i = 0; i < sbi->s_ndevs; i++) { 5102 if (!bdev_is_zoned(FDEV(i).bdev)) 5103 continue; 5104 5105 args.sbi = sbi; 5106 args.fdev = &FDEV(i); 5107 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 5108 check_zone_write_pointer_cb, &args); 5109 if (ret < 0) 5110 return ret; 5111 } 5112 5113 return 0; 5114 } 5115 5116 /* 5117 * Return the number of usable blocks in a segment. The number of blocks 5118 * returned is always equal to the number of blocks in a segment for 5119 * segments fully contained within a sequential zone capacity or a 5120 * conventional zone. For segments partially contained in a sequential 5121 * zone capacity, the number of usable blocks up to the zone capacity 5122 * is returned. 0 is returned in all other cases. 5123 */ 5124 static inline unsigned int f2fs_usable_zone_blks_in_seg( 5125 struct f2fs_sb_info *sbi, unsigned int segno) 5126 { 5127 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr; 5128 unsigned int secno; 5129 5130 if (!sbi->unusable_blocks_per_sec) 5131 return sbi->blocks_per_seg; 5132 5133 secno = GET_SEC_FROM_SEG(sbi, segno); 5134 seg_start = START_BLOCK(sbi, segno); 5135 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5136 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi); 5137 5138 /* 5139 * If segment starts before zone capacity and spans beyond 5140 * zone capacity, then usable blocks are from seg start to 5141 * zone capacity. If the segment starts after the zone capacity, 5142 * then there are no usable blocks. 5143 */ 5144 if (seg_start >= sec_cap_blkaddr) 5145 return 0; 5146 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr) 5147 return sec_cap_blkaddr - seg_start; 5148 5149 return sbi->blocks_per_seg; 5150 } 5151 #else 5152 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5153 { 5154 return 0; 5155 } 5156 5157 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5158 { 5159 return 0; 5160 } 5161 5162 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi, 5163 unsigned int segno) 5164 { 5165 return 0; 5166 } 5167 5168 #endif 5169 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi, 5170 unsigned int segno) 5171 { 5172 if (f2fs_sb_has_blkzoned(sbi)) 5173 return f2fs_usable_zone_blks_in_seg(sbi, segno); 5174 5175 return sbi->blocks_per_seg; 5176 } 5177 5178 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi, 5179 unsigned int segno) 5180 { 5181 if (f2fs_sb_has_blkzoned(sbi)) 5182 return CAP_SEGS_PER_SEC(sbi); 5183 5184 return sbi->segs_per_sec; 5185 } 5186 5187 /* 5188 * Update min, max modified time for cost-benefit GC algorithm 5189 */ 5190 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 5191 { 5192 struct sit_info *sit_i = SIT_I(sbi); 5193 unsigned int segno; 5194 5195 down_write(&sit_i->sentry_lock); 5196 5197 sit_i->min_mtime = ULLONG_MAX; 5198 5199 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 5200 unsigned int i; 5201 unsigned long long mtime = 0; 5202 5203 for (i = 0; i < sbi->segs_per_sec; i++) 5204 mtime += get_seg_entry(sbi, segno + i)->mtime; 5205 5206 mtime = div_u64(mtime, sbi->segs_per_sec); 5207 5208 if (sit_i->min_mtime > mtime) 5209 sit_i->min_mtime = mtime; 5210 } 5211 sit_i->max_mtime = get_mtime(sbi, false); 5212 sit_i->dirty_max_mtime = 0; 5213 up_write(&sit_i->sentry_lock); 5214 } 5215 5216 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 5217 { 5218 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 5219 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 5220 struct f2fs_sm_info *sm_info; 5221 int err; 5222 5223 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 5224 if (!sm_info) 5225 return -ENOMEM; 5226 5227 /* init sm info */ 5228 sbi->sm_info = sm_info; 5229 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 5230 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 5231 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 5232 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 5233 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 5234 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 5235 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 5236 sm_info->rec_prefree_segments = sm_info->main_segments * 5237 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 5238 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 5239 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 5240 5241 if (!f2fs_lfs_mode(sbi)) 5242 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC); 5243 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 5244 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 5245 sm_info->min_seq_blocks = sbi->blocks_per_seg; 5246 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 5247 sm_info->min_ssr_sections = reserved_sections(sbi); 5248 5249 INIT_LIST_HEAD(&sm_info->sit_entry_set); 5250 5251 init_f2fs_rwsem(&sm_info->curseg_lock); 5252 5253 err = f2fs_create_flush_cmd_control(sbi); 5254 if (err) 5255 return err; 5256 5257 err = create_discard_cmd_control(sbi); 5258 if (err) 5259 return err; 5260 5261 err = build_sit_info(sbi); 5262 if (err) 5263 return err; 5264 err = build_free_segmap(sbi); 5265 if (err) 5266 return err; 5267 err = build_curseg(sbi); 5268 if (err) 5269 return err; 5270 5271 /* reinit free segmap based on SIT */ 5272 err = build_sit_entries(sbi); 5273 if (err) 5274 return err; 5275 5276 init_free_segmap(sbi); 5277 err = build_dirty_segmap(sbi); 5278 if (err) 5279 return err; 5280 5281 err = sanity_check_curseg(sbi); 5282 if (err) 5283 return err; 5284 5285 init_min_max_mtime(sbi); 5286 return 0; 5287 } 5288 5289 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 5290 enum dirty_type dirty_type) 5291 { 5292 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5293 5294 mutex_lock(&dirty_i->seglist_lock); 5295 kvfree(dirty_i->dirty_segmap[dirty_type]); 5296 dirty_i->nr_dirty[dirty_type] = 0; 5297 mutex_unlock(&dirty_i->seglist_lock); 5298 } 5299 5300 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 5301 { 5302 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5303 5304 kvfree(dirty_i->pinned_secmap); 5305 kvfree(dirty_i->victim_secmap); 5306 } 5307 5308 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 5309 { 5310 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5311 int i; 5312 5313 if (!dirty_i) 5314 return; 5315 5316 /* discard pre-free/dirty segments list */ 5317 for (i = 0; i < NR_DIRTY_TYPE; i++) 5318 discard_dirty_segmap(sbi, i); 5319 5320 if (__is_large_section(sbi)) { 5321 mutex_lock(&dirty_i->seglist_lock); 5322 kvfree(dirty_i->dirty_secmap); 5323 mutex_unlock(&dirty_i->seglist_lock); 5324 } 5325 5326 destroy_victim_secmap(sbi); 5327 SM_I(sbi)->dirty_info = NULL; 5328 kfree(dirty_i); 5329 } 5330 5331 static void destroy_curseg(struct f2fs_sb_info *sbi) 5332 { 5333 struct curseg_info *array = SM_I(sbi)->curseg_array; 5334 int i; 5335 5336 if (!array) 5337 return; 5338 SM_I(sbi)->curseg_array = NULL; 5339 for (i = 0; i < NR_CURSEG_TYPE; i++) { 5340 kfree(array[i].sum_blk); 5341 kfree(array[i].journal); 5342 } 5343 kfree(array); 5344 } 5345 5346 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 5347 { 5348 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 5349 5350 if (!free_i) 5351 return; 5352 SM_I(sbi)->free_info = NULL; 5353 kvfree(free_i->free_segmap); 5354 kvfree(free_i->free_secmap); 5355 kfree(free_i); 5356 } 5357 5358 static void destroy_sit_info(struct f2fs_sb_info *sbi) 5359 { 5360 struct sit_info *sit_i = SIT_I(sbi); 5361 5362 if (!sit_i) 5363 return; 5364 5365 if (sit_i->sentries) 5366 kvfree(sit_i->bitmap); 5367 kfree(sit_i->tmp_map); 5368 5369 kvfree(sit_i->sentries); 5370 kvfree(sit_i->sec_entries); 5371 kvfree(sit_i->dirty_sentries_bitmap); 5372 5373 SM_I(sbi)->sit_info = NULL; 5374 kvfree(sit_i->sit_bitmap); 5375 #ifdef CONFIG_F2FS_CHECK_FS 5376 kvfree(sit_i->sit_bitmap_mir); 5377 kvfree(sit_i->invalid_segmap); 5378 #endif 5379 kfree(sit_i); 5380 } 5381 5382 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 5383 { 5384 struct f2fs_sm_info *sm_info = SM_I(sbi); 5385 5386 if (!sm_info) 5387 return; 5388 f2fs_destroy_flush_cmd_control(sbi, true); 5389 destroy_discard_cmd_control(sbi); 5390 destroy_dirty_segmap(sbi); 5391 destroy_curseg(sbi); 5392 destroy_free_segmap(sbi); 5393 destroy_sit_info(sbi); 5394 sbi->sm_info = NULL; 5395 kfree(sm_info); 5396 } 5397 5398 int __init f2fs_create_segment_manager_caches(void) 5399 { 5400 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 5401 sizeof(struct discard_entry)); 5402 if (!discard_entry_slab) 5403 goto fail; 5404 5405 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 5406 sizeof(struct discard_cmd)); 5407 if (!discard_cmd_slab) 5408 goto destroy_discard_entry; 5409 5410 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 5411 sizeof(struct sit_entry_set)); 5412 if (!sit_entry_set_slab) 5413 goto destroy_discard_cmd; 5414 5415 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry", 5416 sizeof(struct revoke_entry)); 5417 if (!revoke_entry_slab) 5418 goto destroy_sit_entry_set; 5419 return 0; 5420 5421 destroy_sit_entry_set: 5422 kmem_cache_destroy(sit_entry_set_slab); 5423 destroy_discard_cmd: 5424 kmem_cache_destroy(discard_cmd_slab); 5425 destroy_discard_entry: 5426 kmem_cache_destroy(discard_entry_slab); 5427 fail: 5428 return -ENOMEM; 5429 } 5430 5431 void f2fs_destroy_segment_manager_caches(void) 5432 { 5433 kmem_cache_destroy(sit_entry_set_slab); 5434 kmem_cache_destroy(discard_cmd_slab); 5435 kmem_cache_destroy(discard_entry_slab); 5436 kmem_cache_destroy(revoke_entry_slab); 5437 } 5438