FILE
segment.c
File preview with syntax highlighting.
linux-6.19.14/fs/f2fs/segment.c
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 30static struct kmem_cache *discard_entry_slab; 31static struct kmem_cache *discard_cmd_slab; 32static struct kmem_cache *sit_entry_set_slab; 33static struct kmem_cache *revoke_entry_slab; 34 35static 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 */ 54static 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 */ 98static 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; 122pass: 123 if (size <= BITS_PER_LONG) 124 break; 125 size -= BITS_PER_LONG; 126 offset = 0; 127 p++; 128 } 129 return result; 130found: 131 return result - size + __reverse_ffs(tmp); 132} 133 134static 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; 159pass: 160 if (size <= BITS_PER_LONG) 161 break; 162 size -= BITS_PER_LONG; 163 offset = 0; 164 p++; 165 } 166 return result; 167found: 168 return result - size + __reverse_ffz(tmp); 169} 170 171bool 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 188void f2fs_abort_atomic_write(struct inode *inode, bool clean) 189{ 190 struct f2fs_inode_info *fi = F2FS_I(inode); 191 192 if (!f2fs_is_atomic_file(inode)) 193 return; 194 195 if (clean) 196 truncate_inode_pages_final(inode->i_mapping); 197 198 release_atomic_write_cnt(inode); 199 clear_inode_flag(inode, FI_ATOMIC_COMMITTED); 200 clear_inode_flag(inode, FI_ATOMIC_REPLACE); 201 clear_inode_flag(inode, FI_ATOMIC_FILE); 202 if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) { 203 clear_inode_flag(inode, FI_ATOMIC_DIRTIED); 204 /* 205 * The vfs inode keeps clean during commit, but the f2fs inode 206 * doesn't. So clear the dirty state after commit and let 207 * f2fs_mark_inode_dirty_sync ensure a consistent dirty state. 208 */ 209 f2fs_inode_synced(inode); 210 f2fs_mark_inode_dirty_sync(inode, true); 211 } 212 stat_dec_atomic_inode(inode); 213 214 F2FS_I(inode)->atomic_write_task = NULL; 215 216 if (clean) { 217 f2fs_i_size_write(inode, fi->original_i_size); 218 fi->original_i_size = 0; 219 } 220 /* avoid stale dirty inode during eviction */ 221 sync_inode_metadata(inode, 0); 222} 223 224static int __replace_atomic_write_block(struct inode *inode, pgoff_t index, 225 block_t new_addr, block_t *old_addr, bool recover) 226{ 227 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 228 struct dnode_of_data dn; 229 struct node_info ni; 230 int err; 231 232retry: 233 set_new_dnode(&dn, inode, NULL, NULL, 0); 234 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE); 235 if (err) { 236 if (err == -ENOMEM) { 237 memalloc_retry_wait(GFP_NOFS); 238 goto retry; 239 } 240 return err; 241 } 242 243 err = f2fs_get_node_info(sbi, dn.nid, &ni, false); 244 if (err) { 245 f2fs_put_dnode(&dn); 246 return err; 247 } 248 249 if (recover) { 250 /* dn.data_blkaddr is always valid */ 251 if (!__is_valid_data_blkaddr(new_addr)) { 252 if (new_addr == NULL_ADDR) 253 dec_valid_block_count(sbi, inode, 1); 254 f2fs_invalidate_blocks(sbi, dn.data_blkaddr, 1); 255 f2fs_update_data_blkaddr(&dn, new_addr); 256 } else { 257 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 258 new_addr, ni.version, true, true); 259 } 260 } else { 261 blkcnt_t count = 1; 262 263 err = inc_valid_block_count(sbi, inode, &count, true); 264 if (err) { 265 f2fs_put_dnode(&dn); 266 return err; 267 } 268 269 *old_addr = dn.data_blkaddr; 270 f2fs_truncate_data_blocks_range(&dn, 1); 271 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count); 272 273 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr, 274 ni.version, true, false); 275 } 276 277 f2fs_put_dnode(&dn); 278 279 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode, 280 index, old_addr ? *old_addr : 0, new_addr, recover); 281 return 0; 282} 283 284static void __complete_revoke_list(struct inode *inode, struct list_head *head, 285 bool revoke) 286{ 287 struct revoke_entry *cur, *tmp; 288 pgoff_t start_index = 0; 289 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE); 290 291 list_for_each_entry_safe(cur, tmp, head, list) { 292 if (revoke) { 293 __replace_atomic_write_block(inode, cur->index, 294 cur->old_addr, NULL, true); 295 } else if (truncate) { 296 f2fs_truncate_hole(inode, start_index, cur->index); 297 start_index = cur->index + 1; 298 } 299 300 list_del(&cur->list); 301 kmem_cache_free(revoke_entry_slab, cur); 302 } 303 304 if (!revoke && truncate) 305 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false); 306} 307 308static int __f2fs_commit_atomic_write(struct inode *inode) 309{ 310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 311 struct f2fs_inode_info *fi = F2FS_I(inode); 312 struct inode *cow_inode = fi->cow_inode; 313 struct revoke_entry *new; 314 struct list_head revoke_list; 315 block_t blkaddr; 316 struct dnode_of_data dn; 317 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 318 pgoff_t off = 0, blen, index; 319 int ret = 0, i; 320 321 INIT_LIST_HEAD(&revoke_list); 322 323 while (len) { 324 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len); 325 326 set_new_dnode(&dn, cow_inode, NULL, NULL, 0); 327 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA); 328 if (ret && ret != -ENOENT) { 329 goto out; 330 } else if (ret == -ENOENT) { 331 ret = 0; 332 if (dn.max_level == 0) 333 goto out; 334 goto next; 335 } 336 337 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_folio, cow_inode), 338 len); 339 index = off; 340 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) { 341 blkaddr = f2fs_data_blkaddr(&dn); 342 343 if (!__is_valid_data_blkaddr(blkaddr)) { 344 continue; 345 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 346 DATA_GENERIC_ENHANCE)) { 347 f2fs_put_dnode(&dn); 348 ret = -EFSCORRUPTED; 349 goto out; 350 } 351 352 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS, 353 true, NULL); 354 355 ret = __replace_atomic_write_block(inode, index, blkaddr, 356 &new->old_addr, false); 357 if (ret) { 358 f2fs_put_dnode(&dn); 359 kmem_cache_free(revoke_entry_slab, new); 360 goto out; 361 } 362 363 f2fs_update_data_blkaddr(&dn, NULL_ADDR); 364 new->index = index; 365 list_add_tail(&new->list, &revoke_list); 366 } 367 f2fs_put_dnode(&dn); 368next: 369 off += blen; 370 len -= blen; 371 } 372 373out: 374 if (time_to_inject(sbi, FAULT_TIMEOUT)) 375 f2fs_io_schedule_timeout_killable(DEFAULT_FAULT_TIMEOUT); 376 377 if (ret) { 378 sbi->revoked_atomic_block += fi->atomic_write_cnt; 379 } else { 380 sbi->committed_atomic_block += fi->atomic_write_cnt; 381 set_inode_flag(inode, FI_ATOMIC_COMMITTED); 382 383 /* 384 * inode may has no FI_ATOMIC_DIRTIED flag due to no write 385 * before commit. 386 */ 387 if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) { 388 /* clear atomic dirty status and set vfs dirty status */ 389 clear_inode_flag(inode, FI_ATOMIC_DIRTIED); 390 f2fs_mark_inode_dirty_sync(inode, true); 391 } 392 } 393 394 __complete_revoke_list(inode, &revoke_list, ret ? true : false); 395 396 return ret; 397} 398 399int f2fs_commit_atomic_write(struct inode *inode) 400{ 401 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 402 struct f2fs_inode_info *fi = F2FS_I(inode); 403 int err; 404 405 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 406 if (err) 407 return err; 408 409 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 410 f2fs_lock_op(sbi); 411 412 err = __f2fs_commit_atomic_write(inode); 413 414 f2fs_unlock_op(sbi); 415 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 416 417 return err; 418} 419 420/* 421 * This function balances dirty node and dentry pages. 422 * In addition, it controls garbage collection. 423 */ 424void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 425{ 426 if (f2fs_cp_error(sbi)) 427 return; 428 429 if (time_to_inject(sbi, FAULT_CHECKPOINT)) 430 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT); 431 432 /* balance_fs_bg is able to be pending */ 433 if (need && excess_cached_nats(sbi)) 434 f2fs_balance_fs_bg(sbi, false); 435 436 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 437 return; 438 439 /* 440 * We should do GC or end up with checkpoint, if there are so many dirty 441 * dir/node pages without enough free segments. 442 */ 443 if (has_enough_free_secs(sbi, 0, 0)) 444 return; 445 446 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread && 447 sbi->gc_thread->f2fs_gc_task) { 448 DEFINE_WAIT(wait); 449 450 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait, 451 TASK_UNINTERRUPTIBLE); 452 wake_up(&sbi->gc_thread->gc_wait_queue_head); 453 io_schedule(); 454 finish_wait(&sbi->gc_thread->fggc_wq, &wait); 455 } else { 456 struct f2fs_gc_control gc_control = { 457 .victim_segno = NULL_SEGNO, 458 .init_gc_type = f2fs_sb_has_blkzoned(sbi) ? 459 FG_GC : BG_GC, 460 .no_bg_gc = true, 461 .should_migrate_blocks = false, 462 .err_gc_skipped = false, 463 .nr_free_secs = 1 }; 464 f2fs_down_write(&sbi->gc_lock); 465 stat_inc_gc_call_count(sbi, FOREGROUND); 466 f2fs_gc(sbi, &gc_control); 467 } 468} 469 470static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi) 471{ 472 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2; 473 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS); 474 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA); 475 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES); 476 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META); 477 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA); 478 unsigned int threshold = 479 SEGS_TO_BLKS(sbi, (factor * DEFAULT_DIRTY_THRESHOLD)); 480 unsigned int global_threshold = threshold * 3 / 2; 481 482 if (dents >= threshold || qdata >= threshold || 483 nodes >= threshold || meta >= threshold || 484 imeta >= threshold) 485 return true; 486 return dents + qdata + nodes + meta + imeta > global_threshold; 487} 488 489void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 490{ 491 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 492 return; 493 494 /* try to shrink extent cache when there is no enough memory */ 495 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE)) 496 f2fs_shrink_read_extent_tree(sbi, 497 READ_EXTENT_CACHE_SHRINK_NUMBER); 498 499 /* try to shrink age extent cache when there is no enough memory */ 500 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE)) 501 f2fs_shrink_age_extent_tree(sbi, 502 AGE_EXTENT_CACHE_SHRINK_NUMBER); 503 504 /* check the # of cached NAT entries */ 505 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 506 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 507 508 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 509 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 510 else 511 f2fs_build_free_nids(sbi, false, false); 512 513 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) || 514 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi)) 515 goto do_sync; 516 517 /* there is background inflight IO or foreground operation recently */ 518 if (is_inflight_io(sbi, REQ_TIME) || 519 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem))) 520 return; 521 522 /* exceed periodical checkpoint timeout threshold */ 523 if (f2fs_time_over(sbi, CP_TIME)) 524 goto do_sync; 525 526 /* checkpoint is the only way to shrink partial cached entries */ 527 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) && 528 f2fs_available_free_memory(sbi, INO_ENTRIES)) 529 return; 530 531do_sync: 532 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 533 struct blk_plug plug; 534 535 mutex_lock(&sbi->flush_lock); 536 537 blk_start_plug(&plug); 538 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false); 539 blk_finish_plug(&plug); 540 541 mutex_unlock(&sbi->flush_lock); 542 } 543 stat_inc_cp_call_count(sbi, BACKGROUND); 544 f2fs_sync_fs(sbi->sb, 1); 545} 546 547static int __submit_flush_wait(struct f2fs_sb_info *sbi, 548 struct block_device *bdev) 549{ 550 int ret = blkdev_issue_flush(bdev); 551 552 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 553 test_opt(sbi, FLUSH_MERGE), ret); 554 if (!ret) 555 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0); 556 return ret; 557} 558 559static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 560{ 561 int ret = 0; 562 int i; 563 564 if (!f2fs_is_multi_device(sbi)) 565 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 566 567 for (i = 0; i < sbi->s_ndevs; i++) { 568 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 569 continue; 570 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 571 if (ret) 572 break; 573 } 574 return ret; 575} 576 577static int issue_flush_thread(void *data) 578{ 579 struct f2fs_sb_info *sbi = data; 580 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 581 wait_queue_head_t *q = &fcc->flush_wait_queue; 582repeat: 583 if (kthread_should_stop()) 584 return 0; 585 586 if (!llist_empty(&fcc->issue_list)) { 587 struct flush_cmd *cmd, *next; 588 int ret; 589 590 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 591 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 592 593 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 594 595 ret = submit_flush_wait(sbi, cmd->ino); 596 atomic_inc(&fcc->issued_flush); 597 598 llist_for_each_entry_safe(cmd, next, 599 fcc->dispatch_list, llnode) { 600 cmd->ret = ret; 601 complete(&cmd->wait); 602 } 603 fcc->dispatch_list = NULL; 604 } 605 606 wait_event_interruptible(*q, 607 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 608 goto repeat; 609} 610 611int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 612{ 613 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 614 struct flush_cmd cmd; 615 int ret; 616 617 if (test_opt(sbi, NOBARRIER)) 618 return 0; 619 620 if (!test_opt(sbi, FLUSH_MERGE)) { 621 atomic_inc(&fcc->queued_flush); 622 ret = submit_flush_wait(sbi, ino); 623 atomic_dec(&fcc->queued_flush); 624 atomic_inc(&fcc->issued_flush); 625 return ret; 626 } 627 628 if (atomic_inc_return(&fcc->queued_flush) == 1 || 629 f2fs_is_multi_device(sbi)) { 630 ret = submit_flush_wait(sbi, ino); 631 atomic_dec(&fcc->queued_flush); 632 633 atomic_inc(&fcc->issued_flush); 634 return ret; 635 } 636 637 cmd.ino = ino; 638 init_completion(&cmd.wait); 639 640 llist_add(&cmd.llnode, &fcc->issue_list); 641 642 /* 643 * update issue_list before we wake up issue_flush thread, this 644 * smp_mb() pairs with another barrier in ___wait_event(), see 645 * more details in comments of waitqueue_active(). 646 */ 647 smp_mb(); 648 649 if (waitqueue_active(&fcc->flush_wait_queue)) 650 wake_up(&fcc->flush_wait_queue); 651 652 if (fcc->f2fs_issue_flush) { 653 wait_for_completion(&cmd.wait); 654 atomic_dec(&fcc->queued_flush); 655 } else { 656 struct llist_node *list; 657 658 list = llist_del_all(&fcc->issue_list); 659 if (!list) { 660 wait_for_completion(&cmd.wait); 661 atomic_dec(&fcc->queued_flush); 662 } else { 663 struct flush_cmd *tmp, *next; 664 665 ret = submit_flush_wait(sbi, ino); 666 667 llist_for_each_entry_safe(tmp, next, list, llnode) { 668 if (tmp == &cmd) { 669 cmd.ret = ret; 670 atomic_dec(&fcc->queued_flush); 671 continue; 672 } 673 tmp->ret = ret; 674 complete(&tmp->wait); 675 } 676 } 677 } 678 679 return cmd.ret; 680} 681 682int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 683{ 684 dev_t dev = sbi->sb->s_bdev->bd_dev; 685 struct flush_cmd_control *fcc; 686 687 if (SM_I(sbi)->fcc_info) { 688 fcc = SM_I(sbi)->fcc_info; 689 if (fcc->f2fs_issue_flush) 690 return 0; 691 goto init_thread; 692 } 693 694 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 695 if (!fcc) 696 return -ENOMEM; 697 atomic_set(&fcc->issued_flush, 0); 698 atomic_set(&fcc->queued_flush, 0); 699 init_waitqueue_head(&fcc->flush_wait_queue); 700 init_llist_head(&fcc->issue_list); 701 SM_I(sbi)->fcc_info = fcc; 702 if (!test_opt(sbi, FLUSH_MERGE)) 703 return 0; 704 705init_thread: 706 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 707 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 708 if (IS_ERR(fcc->f2fs_issue_flush)) { 709 int err = PTR_ERR(fcc->f2fs_issue_flush); 710 711 fcc->f2fs_issue_flush = NULL; 712 return err; 713 } 714 715 return 0; 716} 717 718void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 719{ 720 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 721 722 if (fcc && fcc->f2fs_issue_flush) { 723 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 724 725 fcc->f2fs_issue_flush = NULL; 726 kthread_stop(flush_thread); 727 } 728 if (free) { 729 kfree(fcc); 730 SM_I(sbi)->fcc_info = NULL; 731 } 732} 733 734int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 735{ 736 int ret = 0, i; 737 738 if (!f2fs_is_multi_device(sbi)) 739 return 0; 740 741 if (test_opt(sbi, NOBARRIER)) 742 return 0; 743 744 for (i = 1; i < sbi->s_ndevs; i++) { 745 int count = DEFAULT_RETRY_IO_COUNT; 746 747 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 748 continue; 749 750 do { 751 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 752 if (ret) 753 f2fs_schedule_timeout(DEFAULT_SCHEDULE_TIMEOUT); 754 } while (ret && --count); 755 756 if (ret) { 757 f2fs_stop_checkpoint(sbi, false, 758 STOP_CP_REASON_FLUSH_FAIL); 759 break; 760 } 761 762 spin_lock(&sbi->dev_lock); 763 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 764 spin_unlock(&sbi->dev_lock); 765 } 766 767 return ret; 768} 769 770static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 771 enum dirty_type dirty_type) 772{ 773 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 774 775 /* need not be added */ 776 if (is_curseg(sbi, segno)) 777 return; 778 779 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 780 dirty_i->nr_dirty[dirty_type]++; 781 782 if (dirty_type == DIRTY) { 783 struct seg_entry *sentry = get_seg_entry(sbi, segno); 784 enum dirty_type t = sentry->type; 785 786 if (unlikely(t >= DIRTY)) { 787 f2fs_bug_on(sbi, 1); 788 return; 789 } 790 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 791 dirty_i->nr_dirty[t]++; 792 793 if (__is_large_section(sbi)) { 794 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 795 block_t valid_blocks = 796 get_valid_blocks(sbi, segno, true); 797 798 f2fs_bug_on(sbi, 799 (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 800 !valid_blocks) || 801 valid_blocks == CAP_BLKS_PER_SEC(sbi)); 802 803 if (!is_cursec(sbi, secno)) 804 set_bit(secno, dirty_i->dirty_secmap); 805 } 806 } 807} 808 809static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 810 enum dirty_type dirty_type) 811{ 812 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 813 block_t valid_blocks; 814 815 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 816 dirty_i->nr_dirty[dirty_type]--; 817 818 if (dirty_type == DIRTY) { 819 struct seg_entry *sentry = get_seg_entry(sbi, segno); 820 enum dirty_type t = sentry->type; 821 822 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 823 dirty_i->nr_dirty[t]--; 824 825 valid_blocks = get_valid_blocks(sbi, segno, true); 826 if (valid_blocks == 0) { 827 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 828 dirty_i->victim_secmap); 829#ifdef CONFIG_F2FS_CHECK_FS 830 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 831#endif 832 } 833 if (__is_large_section(sbi)) { 834 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 835 836 if (!valid_blocks || 837 valid_blocks == CAP_BLKS_PER_SEC(sbi)) { 838 clear_bit(secno, dirty_i->dirty_secmap); 839 return; 840 } 841 842 if (!is_cursec(sbi, secno)) 843 set_bit(secno, dirty_i->dirty_secmap); 844 } 845 } 846} 847 848/* 849 * Should not occur error such as -ENOMEM. 850 * Adding dirty entry into seglist is not critical operation. 851 * If a given segment is one of current working segments, it won't be added. 852 */ 853static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 854{ 855 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 856 unsigned short valid_blocks, ckpt_valid_blocks; 857 unsigned int usable_blocks; 858 859 if (segno == NULL_SEGNO || is_curseg(sbi, segno)) 860 return; 861 862 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno); 863 mutex_lock(&dirty_i->seglist_lock); 864 865 valid_blocks = get_valid_blocks(sbi, segno, false); 866 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false); 867 868 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 869 ckpt_valid_blocks == usable_blocks)) { 870 __locate_dirty_segment(sbi, segno, PRE); 871 __remove_dirty_segment(sbi, segno, DIRTY); 872 } else if (valid_blocks < usable_blocks) { 873 __locate_dirty_segment(sbi, segno, DIRTY); 874 } else { 875 /* Recovery routine with SSR needs this */ 876 __remove_dirty_segment(sbi, segno, DIRTY); 877 } 878 879 mutex_unlock(&dirty_i->seglist_lock); 880} 881 882/* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 883void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 884{ 885 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 886 unsigned int segno; 887 888 mutex_lock(&dirty_i->seglist_lock); 889 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 890 if (get_valid_blocks(sbi, segno, false)) 891 continue; 892 if (is_curseg(sbi, segno)) 893 continue; 894 __locate_dirty_segment(sbi, segno, PRE); 895 __remove_dirty_segment(sbi, segno, DIRTY); 896 } 897 mutex_unlock(&dirty_i->seglist_lock); 898} 899 900block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 901{ 902 int ovp_hole_segs = 903 (overprovision_segments(sbi) - reserved_segments(sbi)); 904 block_t ovp_holes = SEGS_TO_BLKS(sbi, ovp_hole_segs); 905 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 906 block_t holes[2] = {0, 0}; /* DATA and NODE */ 907 block_t unusable; 908 struct seg_entry *se; 909 unsigned int segno; 910 911 mutex_lock(&dirty_i->seglist_lock); 912 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 913 se = get_seg_entry(sbi, segno); 914 if (IS_NODESEG(se->type)) 915 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) - 916 se->valid_blocks; 917 else 918 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) - 919 se->valid_blocks; 920 } 921 mutex_unlock(&dirty_i->seglist_lock); 922 923 unusable = max(holes[DATA], holes[NODE]); 924 if (unusable > ovp_holes) 925 return unusable - ovp_holes; 926 return 0; 927} 928 929int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 930{ 931 int ovp_hole_segs = 932 (overprovision_segments(sbi) - reserved_segments(sbi)); 933 934 if (F2FS_OPTION(sbi).unusable_cap_perc == 100) 935 return 0; 936 if (unusable > F2FS_OPTION(sbi).unusable_cap) 937 return -EAGAIN; 938 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 939 dirty_segments(sbi) > ovp_hole_segs) 940 return -EAGAIN; 941 if (has_not_enough_free_secs(sbi, 0, 0)) 942 return -EAGAIN; 943 return 0; 944} 945 946/* This is only used by SBI_CP_DISABLED */ 947static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 948{ 949 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 950 unsigned int segno = 0; 951 952 mutex_lock(&dirty_i->seglist_lock); 953 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 954 if (get_valid_blocks(sbi, segno, false)) 955 continue; 956 if (get_ckpt_valid_blocks(sbi, segno, false)) 957 continue; 958 mutex_unlock(&dirty_i->seglist_lock); 959 return segno; 960 } 961 mutex_unlock(&dirty_i->seglist_lock); 962 return NULL_SEGNO; 963} 964 965static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 966 struct block_device *bdev, block_t lstart, 967 block_t start, block_t len) 968{ 969 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 970 struct list_head *pend_list; 971 struct discard_cmd *dc; 972 973 f2fs_bug_on(sbi, !len); 974 975 pend_list = &dcc->pend_list[plist_idx(len)]; 976 977 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL); 978 INIT_LIST_HEAD(&dc->list); 979 dc->bdev = bdev; 980 dc->di.lstart = lstart; 981 dc->di.start = start; 982 dc->di.len = len; 983 dc->ref = 0; 984 dc->state = D_PREP; 985 dc->queued = 0; 986 dc->error = 0; 987 init_completion(&dc->wait); 988 list_add_tail(&dc->list, pend_list); 989 spin_lock_init(&dc->lock); 990 dc->bio_ref = 0; 991 atomic_inc(&dcc->discard_cmd_cnt); 992 dcc->undiscard_blks += len; 993 994 return dc; 995} 996 997static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi) 998{ 999#ifdef CONFIG_F2FS_CHECK_FS 1000 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1001 struct rb_node *cur = rb_first_cached(&dcc->root), *next; 1002 struct discard_cmd *cur_dc, *next_dc; 1003 1004 while (cur) { 1005 next = rb_next(cur); 1006 if (!next) 1007 return true; 1008 1009 cur_dc = rb_entry(cur, struct discard_cmd, rb_node); 1010 next_dc = rb_entry(next, struct discard_cmd, rb_node); 1011 1012 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) { 1013 f2fs_info(sbi, "broken discard_rbtree, " 1014 "cur(%u, %u) next(%u, %u)", 1015 cur_dc->di.lstart, cur_dc->di.len, 1016 next_dc->di.lstart, next_dc->di.len); 1017 return false; 1018 } 1019 cur = next; 1020 } 1021#endif 1022 return true; 1023} 1024 1025static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi, 1026 block_t blkaddr) 1027{ 1028 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1029 struct rb_node *node = dcc->root.rb_root.rb_node; 1030 struct discard_cmd *dc; 1031 1032 while (node) { 1033 dc = rb_entry(node, struct discard_cmd, rb_node); 1034 1035 if (blkaddr < dc->di.lstart) 1036 node = node->rb_left; 1037 else if (blkaddr >= dc->di.lstart + dc->di.len) 1038 node = node->rb_right; 1039 else 1040 return dc; 1041 } 1042 return NULL; 1043} 1044 1045static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root, 1046 block_t blkaddr, 1047 struct discard_cmd **prev_entry, 1048 struct discard_cmd **next_entry, 1049 struct rb_node ***insert_p, 1050 struct rb_node **insert_parent) 1051{ 1052 struct rb_node **pnode = &root->rb_root.rb_node; 1053 struct rb_node *parent = NULL, *tmp_node; 1054 struct discard_cmd *dc; 1055 1056 *insert_p = NULL; 1057 *insert_parent = NULL; 1058 *prev_entry = NULL; 1059 *next_entry = NULL; 1060 1061 if (RB_EMPTY_ROOT(&root->rb_root)) 1062 return NULL; 1063 1064 while (*pnode) { 1065 parent = *pnode; 1066 dc = rb_entry(*pnode, struct discard_cmd, rb_node); 1067 1068 if (blkaddr < dc->di.lstart) 1069 pnode = &(*pnode)->rb_left; 1070 else if (blkaddr >= dc->di.lstart + dc->di.len) 1071 pnode = &(*pnode)->rb_right; 1072 else 1073 goto lookup_neighbors; 1074 } 1075 1076 *insert_p = pnode; 1077 *insert_parent = parent; 1078 1079 dc = rb_entry(parent, struct discard_cmd, rb_node); 1080 tmp_node = parent; 1081 if (parent && blkaddr > dc->di.lstart) 1082 tmp_node = rb_next(parent); 1083 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1084 1085 tmp_node = parent; 1086 if (parent && blkaddr < dc->di.lstart) 1087 tmp_node = rb_prev(parent); 1088 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1089 return NULL; 1090 1091lookup_neighbors: 1092 /* lookup prev node for merging backward later */ 1093 tmp_node = rb_prev(&dc->rb_node); 1094 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1095 1096 /* lookup next node for merging frontward later */ 1097 tmp_node = rb_next(&dc->rb_node); 1098 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1099 return dc; 1100} 1101 1102static void __detach_discard_cmd(struct discard_cmd_control *dcc, 1103 struct discard_cmd *dc) 1104{ 1105 if (dc->state == D_DONE) 1106 atomic_sub(dc->queued, &dcc->queued_discard); 1107 1108 list_del(&dc->list); 1109 rb_erase_cached(&dc->rb_node, &dcc->root); 1110 dcc->undiscard_blks -= dc->di.len; 1111 1112 kmem_cache_free(discard_cmd_slab, dc); 1113 1114 atomic_dec(&dcc->discard_cmd_cnt); 1115} 1116 1117static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 1118 struct discard_cmd *dc) 1119{ 1120 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1121 unsigned long flags; 1122 1123 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len); 1124 1125 spin_lock_irqsave(&dc->lock, flags); 1126 if (dc->bio_ref) { 1127 spin_unlock_irqrestore(&dc->lock, flags); 1128 return; 1129 } 1130 spin_unlock_irqrestore(&dc->lock, flags); 1131 1132 f2fs_bug_on(sbi, dc->ref); 1133 1134 if (dc->error == -EOPNOTSUPP) 1135 dc->error = 0; 1136 1137 if (dc->error) 1138 f2fs_info_ratelimited(sbi, 1139 "Issue discard(%u, %u, %u) failed, ret: %d", 1140 dc->di.lstart, dc->di.start, dc->di.len, dc->error); 1141 __detach_discard_cmd(dcc, dc); 1142} 1143 1144static void f2fs_submit_discard_endio(struct bio *bio) 1145{ 1146 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1147 unsigned long flags; 1148 1149 spin_lock_irqsave(&dc->lock, flags); 1150 if (!dc->error) 1151 dc->error = blk_status_to_errno(bio->bi_status); 1152 dc->bio_ref--; 1153 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1154 dc->state = D_DONE; 1155 complete_all(&dc->wait); 1156 } 1157 spin_unlock_irqrestore(&dc->lock, flags); 1158 bio_put(bio); 1159} 1160 1161static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1162 block_t start, block_t end) 1163{ 1164#ifdef CONFIG_F2FS_CHECK_FS 1165 struct seg_entry *sentry; 1166 unsigned int segno; 1167 block_t blk = start; 1168 unsigned long offset, size, *map; 1169 1170 while (blk < end) { 1171 segno = GET_SEGNO(sbi, blk); 1172 sentry = get_seg_entry(sbi, segno); 1173 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1174 1175 if (end < START_BLOCK(sbi, segno + 1)) 1176 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1177 else 1178 size = BLKS_PER_SEG(sbi); 1179 map = (unsigned long *)(sentry->cur_valid_map); 1180 offset = __find_rev_next_bit(map, size, offset); 1181 f2fs_bug_on(sbi, offset != size); 1182 blk = START_BLOCK(sbi, segno + 1); 1183 } 1184#endif 1185} 1186 1187static void __init_discard_policy(struct f2fs_sb_info *sbi, 1188 struct discard_policy *dpolicy, 1189 int discard_type, unsigned int granularity) 1190{ 1191 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1192 1193 /* common policy */ 1194 dpolicy->type = discard_type; 1195 dpolicy->sync = true; 1196 dpolicy->ordered = false; 1197 dpolicy->granularity = granularity; 1198 1199 dpolicy->max_requests = dcc->max_discard_request; 1200 dpolicy->io_aware_gran = dcc->discard_io_aware_gran; 1201 dpolicy->timeout = false; 1202 1203 if (discard_type == DPOLICY_BG) { 1204 dpolicy->min_interval = dcc->min_discard_issue_time; 1205 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1206 dpolicy->max_interval = dcc->max_discard_issue_time; 1207 if (dcc->discard_io_aware == DPOLICY_IO_AWARE_ENABLE) 1208 dpolicy->io_aware = true; 1209 else if (dcc->discard_io_aware == DPOLICY_IO_AWARE_DISABLE) 1210 dpolicy->io_aware = false; 1211 dpolicy->sync = false; 1212 dpolicy->ordered = true; 1213 if (utilization(sbi) > dcc->discard_urgent_util) { 1214 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1215 if (atomic_read(&dcc->discard_cmd_cnt)) 1216 dpolicy->max_interval = 1217 dcc->min_discard_issue_time; 1218 } 1219 } else if (discard_type == DPOLICY_FORCE) { 1220 dpolicy->min_interval = dcc->min_discard_issue_time; 1221 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1222 dpolicy->max_interval = dcc->max_discard_issue_time; 1223 dpolicy->io_aware = false; 1224 } else if (discard_type == DPOLICY_FSTRIM) { 1225 dpolicy->io_aware = false; 1226 } else if (discard_type == DPOLICY_UMOUNT) { 1227 dpolicy->io_aware = false; 1228 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1229 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1230 dpolicy->timeout = true; 1231 } 1232} 1233 1234static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1235 struct block_device *bdev, block_t lstart, 1236 block_t start, block_t len); 1237 1238#ifdef CONFIG_BLK_DEV_ZONED 1239static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi, 1240 struct discard_cmd *dc, blk_opf_t flag, 1241 struct list_head *wait_list, 1242 unsigned int *issued) 1243{ 1244 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1245 struct block_device *bdev = dc->bdev; 1246 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS); 1247 unsigned long flags; 1248 1249 trace_f2fs_issue_reset_zone(bdev, dc->di.start); 1250 1251 spin_lock_irqsave(&dc->lock, flags); 1252 dc->state = D_SUBMIT; 1253 dc->bio_ref++; 1254 spin_unlock_irqrestore(&dc->lock, flags); 1255 1256 if (issued) 1257 (*issued)++; 1258 1259 atomic_inc(&dcc->queued_discard); 1260 dc->queued++; 1261 list_move_tail(&dc->list, wait_list); 1262 1263 /* sanity check on discard range */ 1264 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len); 1265 1266 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start); 1267 bio->bi_private = dc; 1268 bio->bi_end_io = f2fs_submit_discard_endio; 1269 submit_bio(bio); 1270 1271 atomic_inc(&dcc->issued_discard); 1272 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE); 1273} 1274#endif 1275 1276/* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1277static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1278 struct discard_policy *dpolicy, 1279 struct discard_cmd *dc, int *issued) 1280{ 1281 struct block_device *bdev = dc->bdev; 1282 unsigned int max_discard_blocks = 1283 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1284 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1285 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1286 &(dcc->fstrim_list) : &(dcc->wait_list); 1287 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0; 1288 block_t lstart, start, len, total_len; 1289 int err = 0; 1290 1291 if (dc->state != D_PREP) 1292 return 0; 1293 1294 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1295 return 0; 1296 1297#ifdef CONFIG_BLK_DEV_ZONED 1298 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) { 1299 int devi = f2fs_bdev_index(sbi, bdev); 1300 1301 if (devi < 0) 1302 return -EINVAL; 1303 1304 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1305 __submit_zone_reset_cmd(sbi, dc, flag, 1306 wait_list, issued); 1307 return 0; 1308 } 1309 } 1310#endif 1311 1312 /* 1313 * stop issuing discard for any of below cases: 1314 * 1. device is conventional zone, but it doesn't support discard. 1315 * 2. device is regulare device, after snapshot it doesn't support 1316 * discard. 1317 */ 1318 if (!bdev_max_discard_sectors(bdev)) 1319 return -EOPNOTSUPP; 1320 1321 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len); 1322 1323 lstart = dc->di.lstart; 1324 start = dc->di.start; 1325 len = dc->di.len; 1326 total_len = len; 1327 1328 dc->di.len = 0; 1329 1330 while (total_len && *issued < dpolicy->max_requests && !err) { 1331 struct bio *bio = NULL; 1332 unsigned long flags; 1333 bool last = true; 1334 1335 if (len > max_discard_blocks) { 1336 len = max_discard_blocks; 1337 last = false; 1338 } 1339 1340 (*issued)++; 1341 if (*issued == dpolicy->max_requests) 1342 last = true; 1343 1344 dc->di.len += len; 1345 1346 err = 0; 1347 if (time_to_inject(sbi, FAULT_DISCARD)) { 1348 err = -EIO; 1349 spin_lock_irqsave(&dc->lock, flags); 1350 if (dc->state == D_PARTIAL) 1351 dc->state = D_SUBMIT; 1352 spin_unlock_irqrestore(&dc->lock, flags); 1353 1354 break; 1355 } 1356 1357 __blkdev_issue_discard(bdev, SECTOR_FROM_BLOCK(start), 1358 SECTOR_FROM_BLOCK(len), GFP_NOFS, &bio); 1359 f2fs_bug_on(sbi, !bio); 1360 1361 /* 1362 * should keep before submission to avoid D_DONE 1363 * right away 1364 */ 1365 spin_lock_irqsave(&dc->lock, flags); 1366 if (last) 1367 dc->state = D_SUBMIT; 1368 else 1369 dc->state = D_PARTIAL; 1370 dc->bio_ref++; 1371 spin_unlock_irqrestore(&dc->lock, flags); 1372 1373 atomic_inc(&dcc->queued_discard); 1374 dc->queued++; 1375 list_move_tail(&dc->list, wait_list); 1376 1377 /* sanity check on discard range */ 1378 __check_sit_bitmap(sbi, lstart, lstart + len); 1379 1380 bio->bi_private = dc; 1381 bio->bi_end_io = f2fs_submit_discard_endio; 1382 bio->bi_opf |= flag; 1383 submit_bio(bio); 1384 1385 atomic_inc(&dcc->issued_discard); 1386 1387 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE); 1388 1389 lstart += len; 1390 start += len; 1391 total_len -= len; 1392 len = total_len; 1393 } 1394 1395 if (!err && len) { 1396 dcc->undiscard_blks -= len; 1397 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1398 } 1399 return err; 1400} 1401 1402static void __insert_discard_cmd(struct f2fs_sb_info *sbi, 1403 struct block_device *bdev, block_t lstart, 1404 block_t start, block_t len) 1405{ 1406 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1407 struct rb_node **p = &dcc->root.rb_root.rb_node; 1408 struct rb_node *parent = NULL; 1409 struct discard_cmd *dc; 1410 bool leftmost = true; 1411 1412 /* look up rb tree to find parent node */ 1413 while (*p) { 1414 parent = *p; 1415 dc = rb_entry(parent, struct discard_cmd, rb_node); 1416 1417 if (lstart < dc->di.lstart) { 1418 p = &(*p)->rb_left; 1419 } else if (lstart >= dc->di.lstart + dc->di.len) { 1420 p = &(*p)->rb_right; 1421 leftmost = false; 1422 } else { 1423 /* Let's skip to add, if exists */ 1424 return; 1425 } 1426 } 1427 1428 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 1429 1430 rb_link_node(&dc->rb_node, parent, p); 1431 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 1432} 1433 1434static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1435 struct discard_cmd *dc) 1436{ 1437 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]); 1438} 1439 1440static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1441 struct discard_cmd *dc, block_t blkaddr) 1442{ 1443 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1444 struct discard_info di = dc->di; 1445 bool modified = false; 1446 1447 if (dc->state == D_DONE || dc->di.len == 1) { 1448 __remove_discard_cmd(sbi, dc); 1449 return; 1450 } 1451 1452 dcc->undiscard_blks -= di.len; 1453 1454 if (blkaddr > di.lstart) { 1455 dc->di.len = blkaddr - dc->di.lstart; 1456 dcc->undiscard_blks += dc->di.len; 1457 __relocate_discard_cmd(dcc, dc); 1458 modified = true; 1459 } 1460 1461 if (blkaddr < di.lstart + di.len - 1) { 1462 if (modified) { 1463 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1, 1464 di.start + blkaddr + 1 - di.lstart, 1465 di.lstart + di.len - 1 - blkaddr); 1466 } else { 1467 dc->di.lstart++; 1468 dc->di.len--; 1469 dc->di.start++; 1470 dcc->undiscard_blks += dc->di.len; 1471 __relocate_discard_cmd(dcc, dc); 1472 } 1473 } 1474} 1475 1476static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1477 struct block_device *bdev, block_t lstart, 1478 block_t start, block_t len) 1479{ 1480 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1481 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1482 struct discard_cmd *dc; 1483 struct discard_info di = {0}; 1484 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1485 unsigned int max_discard_blocks = 1486 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1487 block_t end = lstart + len; 1488 1489 dc = __lookup_discard_cmd_ret(&dcc->root, lstart, 1490 &prev_dc, &next_dc, &insert_p, &insert_parent); 1491 if (dc) 1492 prev_dc = dc; 1493 1494 if (!prev_dc) { 1495 di.lstart = lstart; 1496 di.len = next_dc ? next_dc->di.lstart - lstart : len; 1497 di.len = min(di.len, len); 1498 di.start = start; 1499 } 1500 1501 while (1) { 1502 struct rb_node *node; 1503 bool merged = false; 1504 struct discard_cmd *tdc = NULL; 1505 1506 if (prev_dc) { 1507 di.lstart = prev_dc->di.lstart + prev_dc->di.len; 1508 if (di.lstart < lstart) 1509 di.lstart = lstart; 1510 if (di.lstart >= end) 1511 break; 1512 1513 if (!next_dc || next_dc->di.lstart > end) 1514 di.len = end - di.lstart; 1515 else 1516 di.len = next_dc->di.lstart - di.lstart; 1517 di.start = start + di.lstart - lstart; 1518 } 1519 1520 if (!di.len) 1521 goto next; 1522 1523 if (prev_dc && prev_dc->state == D_PREP && 1524 prev_dc->bdev == bdev && 1525 __is_discard_back_mergeable(&di, &prev_dc->di, 1526 max_discard_blocks)) { 1527 prev_dc->di.len += di.len; 1528 dcc->undiscard_blks += di.len; 1529 __relocate_discard_cmd(dcc, prev_dc); 1530 di = prev_dc->di; 1531 tdc = prev_dc; 1532 merged = true; 1533 } 1534 1535 if (next_dc && next_dc->state == D_PREP && 1536 next_dc->bdev == bdev && 1537 __is_discard_front_mergeable(&di, &next_dc->di, 1538 max_discard_blocks)) { 1539 next_dc->di.lstart = di.lstart; 1540 next_dc->di.len += di.len; 1541 next_dc->di.start = di.start; 1542 dcc->undiscard_blks += di.len; 1543 __relocate_discard_cmd(dcc, next_dc); 1544 if (tdc) 1545 __remove_discard_cmd(sbi, tdc); 1546 merged = true; 1547 } 1548 1549 if (!merged) 1550 __insert_discard_cmd(sbi, bdev, 1551 di.lstart, di.start, di.len); 1552 next: 1553 prev_dc = next_dc; 1554 if (!prev_dc) 1555 break; 1556 1557 node = rb_next(&prev_dc->rb_node); 1558 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1559 } 1560} 1561 1562#ifdef CONFIG_BLK_DEV_ZONED 1563static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi, 1564 struct block_device *bdev, block_t blkstart, block_t lblkstart, 1565 block_t blklen) 1566{ 1567 trace_f2fs_queue_reset_zone(bdev, blkstart); 1568 1569 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1570 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen); 1571 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1572} 1573#endif 1574 1575static void __queue_discard_cmd(struct f2fs_sb_info *sbi, 1576 struct block_device *bdev, block_t blkstart, block_t blklen) 1577{ 1578 block_t lblkstart = blkstart; 1579 1580 if (!f2fs_bdev_support_discard(bdev)) 1581 return; 1582 1583 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1584 1585 if (f2fs_is_multi_device(sbi)) { 1586 int devi = f2fs_target_device_index(sbi, blkstart); 1587 1588 blkstart -= FDEV(devi).start_blk; 1589 } 1590 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1591 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1592 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1593} 1594 1595static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1596 struct discard_policy *dpolicy, int *issued) 1597{ 1598 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1599 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1600 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1601 struct discard_cmd *dc; 1602 struct blk_plug plug; 1603 bool io_interrupted = false; 1604 1605 mutex_lock(&dcc->cmd_lock); 1606 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos, 1607 &prev_dc, &next_dc, &insert_p, &insert_parent); 1608 if (!dc) 1609 dc = next_dc; 1610 1611 blk_start_plug(&plug); 1612 1613 while (dc) { 1614 struct rb_node *node; 1615 int err = 0; 1616 1617 if (dc->state != D_PREP) 1618 goto next; 1619 1620 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1621 io_interrupted = true; 1622 break; 1623 } 1624 1625 dcc->next_pos = dc->di.lstart + dc->di.len; 1626 err = __submit_discard_cmd(sbi, dpolicy, dc, issued); 1627 1628 if (*issued >= dpolicy->max_requests) 1629 break; 1630next: 1631 node = rb_next(&dc->rb_node); 1632 if (err) 1633 __remove_discard_cmd(sbi, dc); 1634 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1635 } 1636 1637 blk_finish_plug(&plug); 1638 1639 if (!dc) 1640 dcc->next_pos = 0; 1641 1642 mutex_unlock(&dcc->cmd_lock); 1643 1644 if (!(*issued) && io_interrupted) 1645 *issued = -1; 1646} 1647static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1648 struct discard_policy *dpolicy); 1649 1650static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1651 struct discard_policy *dpolicy) 1652{ 1653 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1654 struct list_head *pend_list; 1655 struct discard_cmd *dc, *tmp; 1656 struct blk_plug plug; 1657 int i, issued; 1658 bool io_interrupted = false; 1659 1660 if (dpolicy->timeout) 1661 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1662 1663retry: 1664 issued = 0; 1665 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1666 if (dpolicy->timeout && 1667 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1668 break; 1669 1670 if (i + 1 < dpolicy->granularity) 1671 break; 1672 1673 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) { 1674 __issue_discard_cmd_orderly(sbi, dpolicy, &issued); 1675 return issued; 1676 } 1677 1678 pend_list = &dcc->pend_list[i]; 1679 1680 mutex_lock(&dcc->cmd_lock); 1681 if (list_empty(pend_list)) 1682 goto next; 1683 if (unlikely(dcc->rbtree_check)) 1684 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 1685 blk_start_plug(&plug); 1686 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1687 f2fs_bug_on(sbi, dc->state != D_PREP); 1688 1689 if (dpolicy->timeout && 1690 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1691 break; 1692 1693 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1694 !is_idle(sbi, DISCARD_TIME)) { 1695 io_interrupted = true; 1696 break; 1697 } 1698 1699 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1700 1701 if (issued >= dpolicy->max_requests) 1702 break; 1703 } 1704 blk_finish_plug(&plug); 1705next: 1706 mutex_unlock(&dcc->cmd_lock); 1707 1708 if (issued >= dpolicy->max_requests || io_interrupted) 1709 break; 1710 } 1711 1712 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1713 __wait_all_discard_cmd(sbi, dpolicy); 1714 goto retry; 1715 } 1716 1717 if (!issued && io_interrupted) 1718 issued = -1; 1719 1720 return issued; 1721} 1722 1723static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1724{ 1725 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1726 struct list_head *pend_list; 1727 struct discard_cmd *dc, *tmp; 1728 int i; 1729 bool dropped = false; 1730 1731 mutex_lock(&dcc->cmd_lock); 1732 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1733 pend_list = &dcc->pend_list[i]; 1734 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1735 f2fs_bug_on(sbi, dc->state != D_PREP); 1736 __remove_discard_cmd(sbi, dc); 1737 dropped = true; 1738 } 1739 } 1740 mutex_unlock(&dcc->cmd_lock); 1741 1742 return dropped; 1743} 1744 1745void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1746{ 1747 __drop_discard_cmd(sbi); 1748} 1749 1750static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1751 struct discard_cmd *dc) 1752{ 1753 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1754 unsigned int len = 0; 1755 1756 wait_for_completion_io(&dc->wait); 1757 mutex_lock(&dcc->cmd_lock); 1758 f2fs_bug_on(sbi, dc->state != D_DONE); 1759 dc->ref--; 1760 if (!dc->ref) { 1761 if (!dc->error) 1762 len = dc->di.len; 1763 __remove_discard_cmd(sbi, dc); 1764 } 1765 mutex_unlock(&dcc->cmd_lock); 1766 1767 return len; 1768} 1769 1770static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1771 struct discard_policy *dpolicy, 1772 block_t start, block_t end) 1773{ 1774 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1775 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1776 &(dcc->fstrim_list) : &(dcc->wait_list); 1777 struct discard_cmd *dc = NULL, *iter, *tmp; 1778 unsigned int trimmed = 0; 1779 1780next: 1781 dc = NULL; 1782 1783 mutex_lock(&dcc->cmd_lock); 1784 list_for_each_entry_safe(iter, tmp, wait_list, list) { 1785 if (iter->di.lstart + iter->di.len <= start || 1786 end <= iter->di.lstart) 1787 continue; 1788 if (iter->di.len < dpolicy->granularity) 1789 continue; 1790 if (iter->state == D_DONE && !iter->ref) { 1791 wait_for_completion_io(&iter->wait); 1792 if (!iter->error) 1793 trimmed += iter->di.len; 1794 __remove_discard_cmd(sbi, iter); 1795 } else { 1796 iter->ref++; 1797 dc = iter; 1798 break; 1799 } 1800 } 1801 mutex_unlock(&dcc->cmd_lock); 1802 1803 if (dc) { 1804 trimmed += __wait_one_discard_bio(sbi, dc); 1805 goto next; 1806 } 1807 1808 return trimmed; 1809} 1810 1811static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1812 struct discard_policy *dpolicy) 1813{ 1814 struct discard_policy dp; 1815 unsigned int discard_blks; 1816 1817 if (dpolicy) 1818 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1819 1820 /* wait all */ 1821 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY); 1822 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1823 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY); 1824 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1825 1826 return discard_blks; 1827} 1828 1829/* This should be covered by global mutex, &sit_i->sentry_lock */ 1830static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1831{ 1832 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1833 struct discard_cmd *dc; 1834 bool need_wait = false; 1835 1836 mutex_lock(&dcc->cmd_lock); 1837 dc = __lookup_discard_cmd(sbi, blkaddr); 1838#ifdef CONFIG_BLK_DEV_ZONED 1839 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) { 1840 int devi = f2fs_bdev_index(sbi, dc->bdev); 1841 1842 if (devi < 0) { 1843 mutex_unlock(&dcc->cmd_lock); 1844 return; 1845 } 1846 1847 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1848 /* force submit zone reset */ 1849 if (dc->state == D_PREP) 1850 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC, 1851 &dcc->wait_list, NULL); 1852 dc->ref++; 1853 mutex_unlock(&dcc->cmd_lock); 1854 /* wait zone reset */ 1855 __wait_one_discard_bio(sbi, dc); 1856 return; 1857 } 1858 } 1859#endif 1860 if (dc) { 1861 if (dc->state == D_PREP) { 1862 __punch_discard_cmd(sbi, dc, blkaddr); 1863 } else { 1864 dc->ref++; 1865 need_wait = true; 1866 } 1867 } 1868 mutex_unlock(&dcc->cmd_lock); 1869 1870 if (need_wait) 1871 __wait_one_discard_bio(sbi, dc); 1872} 1873 1874void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1875{ 1876 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1877 1878 if (dcc && dcc->f2fs_issue_discard) { 1879 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1880 1881 dcc->f2fs_issue_discard = NULL; 1882 kthread_stop(discard_thread); 1883 } 1884} 1885 1886/** 1887 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT 1888 * @sbi: the f2fs_sb_info data for discard cmd to issue 1889 * 1890 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped 1891 * 1892 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false. 1893 */ 1894bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1895{ 1896 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1897 struct discard_policy dpolicy; 1898 bool dropped; 1899 1900 if (!atomic_read(&dcc->discard_cmd_cnt)) 1901 return true; 1902 1903 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1904 dcc->discard_granularity); 1905 __issue_discard_cmd(sbi, &dpolicy); 1906 dropped = __drop_discard_cmd(sbi); 1907 1908 /* just to make sure there is no pending discard commands */ 1909 __wait_all_discard_cmd(sbi, NULL); 1910 1911 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1912 return !dropped; 1913} 1914 1915static int issue_discard_thread(void *data) 1916{ 1917 struct f2fs_sb_info *sbi = data; 1918 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1919 wait_queue_head_t *q = &dcc->discard_wait_queue; 1920 struct discard_policy dpolicy; 1921 unsigned int wait_ms = dcc->min_discard_issue_time; 1922 int issued; 1923 1924 set_freezable(); 1925 1926 do { 1927 wait_event_freezable_timeout(*q, 1928 kthread_should_stop() || dcc->discard_wake, 1929 msecs_to_jiffies(wait_ms)); 1930 1931 if (sbi->gc_mode == GC_URGENT_HIGH || 1932 !f2fs_available_free_memory(sbi, DISCARD_CACHE)) 1933 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1934 MIN_DISCARD_GRANULARITY); 1935 else 1936 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1937 dcc->discard_granularity); 1938 1939 if (dcc->discard_wake) 1940 dcc->discard_wake = false; 1941 1942 /* clean up pending candidates before going to sleep */ 1943 if (atomic_read(&dcc->queued_discard)) 1944 __wait_all_discard_cmd(sbi, NULL); 1945 1946 if (f2fs_readonly(sbi->sb)) 1947 continue; 1948 if (kthread_should_stop()) 1949 return 0; 1950 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) || 1951 !atomic_read(&dcc->discard_cmd_cnt)) { 1952 wait_ms = dpolicy.max_interval; 1953 continue; 1954 } 1955 1956 sb_start_intwrite(sbi->sb); 1957 1958 issued = __issue_discard_cmd(sbi, &dpolicy); 1959 if (issued > 0) { 1960 __wait_all_discard_cmd(sbi, &dpolicy); 1961 wait_ms = dpolicy.min_interval; 1962 } else if (issued == -1) { 1963 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1964 if (!wait_ms) 1965 wait_ms = dpolicy.mid_interval; 1966 } else { 1967 wait_ms = dpolicy.max_interval; 1968 } 1969 if (!atomic_read(&dcc->discard_cmd_cnt)) 1970 wait_ms = dpolicy.max_interval; 1971 1972 sb_end_intwrite(sbi->sb); 1973 1974 } while (!kthread_should_stop()); 1975 return 0; 1976} 1977 1978#ifdef CONFIG_BLK_DEV_ZONED 1979static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1980 struct block_device *bdev, block_t blkstart, block_t blklen) 1981{ 1982 sector_t sector, nr_sects; 1983 block_t lblkstart = blkstart; 1984 int devi = 0; 1985 u64 remainder = 0; 1986 1987 if (f2fs_is_multi_device(sbi)) { 1988 devi = f2fs_target_device_index(sbi, blkstart); 1989 if (blkstart < FDEV(devi).start_blk || 1990 blkstart > FDEV(devi).end_blk) { 1991 f2fs_err(sbi, "Invalid block %x", blkstart); 1992 return -EIO; 1993 } 1994 blkstart -= FDEV(devi).start_blk; 1995 } 1996 1997 /* For sequential zones, reset the zone write pointer */ 1998 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1999 sector = SECTOR_FROM_BLOCK(blkstart); 2000 nr_sects = SECTOR_FROM_BLOCK(blklen); 2001 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder); 2002 2003 if (remainder || nr_sects != bdev_zone_sectors(bdev)) { 2004 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 2005 devi, sbi->s_ndevs ? FDEV(devi).path : "", 2006 blkstart, blklen); 2007 return -EIO; 2008 } 2009 2010 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) { 2011 unsigned int nofs_flags; 2012 int ret; 2013 2014 trace_f2fs_issue_reset_zone(bdev, blkstart); 2015 nofs_flags = memalloc_nofs_save(); 2016 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 2017 sector, nr_sects); 2018 memalloc_nofs_restore(nofs_flags); 2019 return ret; 2020 } 2021 2022 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen); 2023 return 0; 2024 } 2025 2026 /* For conventional zones, use regular discard if supported */ 2027 __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 2028 return 0; 2029} 2030#endif 2031 2032static int __issue_discard_async(struct f2fs_sb_info *sbi, 2033 struct block_device *bdev, block_t blkstart, block_t blklen) 2034{ 2035#ifdef CONFIG_BLK_DEV_ZONED 2036 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 2037 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 2038#endif 2039 __queue_discard_cmd(sbi, bdev, blkstart, blklen); 2040 return 0; 2041} 2042 2043static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 2044 block_t blkstart, block_t blklen) 2045{ 2046 sector_t start = blkstart, len = 0; 2047 struct block_device *bdev; 2048 struct seg_entry *se; 2049 unsigned int offset; 2050 block_t i; 2051 int err = 0; 2052 2053 bdev = f2fs_target_device(sbi, blkstart, NULL); 2054 2055 for (i = blkstart; i < blkstart + blklen; i++, len++) { 2056 if (i != start) { 2057 struct block_device *bdev2 = 2058 f2fs_target_device(sbi, i, NULL); 2059 2060 if (bdev2 != bdev) { 2061 err = __issue_discard_async(sbi, bdev, 2062 start, len); 2063 if (err) 2064 return err; 2065 bdev = bdev2; 2066 start = i; 2067 len = 0; 2068 } 2069 } 2070 2071 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 2072 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 2073 2074 if (f2fs_block_unit_discard(sbi) && 2075 !f2fs_test_and_set_bit(offset, se->discard_map)) 2076 sbi->discard_blks--; 2077 } 2078 2079 if (len) 2080 err = __issue_discard_async(sbi, bdev, start, len); 2081 return err; 2082} 2083 2084static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 2085 bool check_only) 2086{ 2087 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2088 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 2089 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2090 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2091 unsigned long *discard_map = (unsigned long *)se->discard_map; 2092 unsigned long *dmap = SIT_I(sbi)->tmp_map; 2093 unsigned int start = 0, end = -1; 2094 bool force = (cpc->reason & CP_DISCARD); 2095 struct discard_entry *de = NULL; 2096 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 2097 int i; 2098 2099 if (se->valid_blocks == BLKS_PER_SEG(sbi) || 2100 !f2fs_hw_support_discard(sbi) || 2101 !f2fs_block_unit_discard(sbi)) 2102 return false; 2103 2104 if (!force) { 2105 if (!f2fs_realtime_discard_enable(sbi) || 2106 (!se->valid_blocks && 2107 !is_curseg(sbi, cpc->trim_start)) || 2108 SM_I(sbi)->dcc_info->nr_discards >= 2109 SM_I(sbi)->dcc_info->max_discards) 2110 return false; 2111 } 2112 2113 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 2114 for (i = 0; i < entries; i++) 2115 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 2116 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 2117 2118 while (force || SM_I(sbi)->dcc_info->nr_discards <= 2119 SM_I(sbi)->dcc_info->max_discards) { 2120 start = __find_rev_next_bit(dmap, BLKS_PER_SEG(sbi), end + 1); 2121 if (start >= BLKS_PER_SEG(sbi)) 2122 break; 2123 2124 end = __find_rev_next_zero_bit(dmap, 2125 BLKS_PER_SEG(sbi), start + 1); 2126 if (force && start && end != BLKS_PER_SEG(sbi) && 2127 (end - start) < cpc->trim_minlen) 2128 continue; 2129 2130 if (check_only) 2131 return true; 2132 2133 if (!de) { 2134 de = f2fs_kmem_cache_alloc(discard_entry_slab, 2135 GFP_F2FS_ZERO, true, NULL); 2136 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 2137 list_add_tail(&de->list, head); 2138 } 2139 2140 for (i = start; i < end; i++) 2141 __set_bit_le(i, (void *)de->discard_map); 2142 2143 SM_I(sbi)->dcc_info->nr_discards += end - start; 2144 } 2145 return false; 2146} 2147 2148static void release_discard_addr(struct discard_entry *entry) 2149{ 2150 list_del(&entry->list); 2151 kmem_cache_free(discard_entry_slab, entry); 2152} 2153 2154void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 2155{ 2156 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 2157 struct discard_entry *entry, *this; 2158 2159 /* drop caches */ 2160 list_for_each_entry_safe(entry, this, head, list) 2161 release_discard_addr(entry); 2162} 2163 2164/* 2165 * Should call f2fs_clear_prefree_segments after checkpoint is done. 2166 */ 2167static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 2168{ 2169 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2170 unsigned int segno; 2171 2172 mutex_lock(&dirty_i->seglist_lock); 2173 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 2174 __set_test_and_free(sbi, segno, false); 2175 mutex_unlock(&dirty_i->seglist_lock); 2176} 2177 2178void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 2179 struct cp_control *cpc) 2180{ 2181 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2182 struct list_head *head = &dcc->entry_list; 2183 struct discard_entry *entry, *this; 2184 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2185 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 2186 unsigned int start = 0, end = -1; 2187 unsigned int secno, start_segno; 2188 bool force = (cpc->reason & CP_DISCARD); 2189 bool section_alignment = F2FS_OPTION(sbi).discard_unit == 2190 DISCARD_UNIT_SECTION; 2191 2192 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi)) 2193 section_alignment = true; 2194 2195 mutex_lock(&dirty_i->seglist_lock); 2196 2197 while (1) { 2198 int i; 2199 2200 if (section_alignment && end != -1) 2201 end--; 2202 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 2203 if (start >= MAIN_SEGS(sbi)) 2204 break; 2205 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 2206 start + 1); 2207 2208 if (section_alignment) { 2209 start = rounddown(start, SEGS_PER_SEC(sbi)); 2210 end = roundup(end, SEGS_PER_SEC(sbi)); 2211 } 2212 2213 for (i = start; i < end; i++) { 2214 if (test_and_clear_bit(i, prefree_map)) 2215 dirty_i->nr_dirty[PRE]--; 2216 } 2217 2218 if (!f2fs_realtime_discard_enable(sbi)) 2219 continue; 2220 2221 if (force && start >= cpc->trim_start && 2222 (end - 1) <= cpc->trim_end) 2223 continue; 2224 2225 /* Should cover 2MB zoned device for zone-based reset */ 2226 if (!f2fs_sb_has_blkzoned(sbi) && 2227 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) { 2228 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 2229 SEGS_TO_BLKS(sbi, end - start)); 2230 continue; 2231 } 2232next: 2233 secno = GET_SEC_FROM_SEG(sbi, start); 2234 start_segno = GET_SEG_FROM_SEC(sbi, secno); 2235 if (!is_cursec(sbi, secno) && 2236 !get_valid_blocks(sbi, start, true)) 2237 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 2238 BLKS_PER_SEC(sbi)); 2239 2240 start = start_segno + SEGS_PER_SEC(sbi); 2241 if (start < end) 2242 goto next; 2243 else 2244 end = start - 1; 2245 } 2246 mutex_unlock(&dirty_i->seglist_lock); 2247 2248 if (!f2fs_block_unit_discard(sbi)) 2249 goto wakeup; 2250 2251 /* send small discards */ 2252 list_for_each_entry_safe(entry, this, head, list) { 2253 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2254 bool is_valid = test_bit_le(0, entry->discard_map); 2255 2256find_next: 2257 if (is_valid) { 2258 next_pos = find_next_zero_bit_le(entry->discard_map, 2259 BLKS_PER_SEG(sbi), cur_pos); 2260 len = next_pos - cur_pos; 2261 2262 if (f2fs_sb_has_blkzoned(sbi) || 2263 (force && len < cpc->trim_minlen)) 2264 goto skip; 2265 2266 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2267 len); 2268 total_len += len; 2269 } else { 2270 next_pos = find_next_bit_le(entry->discard_map, 2271 BLKS_PER_SEG(sbi), cur_pos); 2272 } 2273skip: 2274 cur_pos = next_pos; 2275 is_valid = !is_valid; 2276 2277 if (cur_pos < BLKS_PER_SEG(sbi)) 2278 goto find_next; 2279 2280 release_discard_addr(entry); 2281 dcc->nr_discards -= total_len; 2282 } 2283 2284wakeup: 2285 wake_up_discard_thread(sbi, false); 2286} 2287 2288int f2fs_start_discard_thread(struct f2fs_sb_info *sbi) 2289{ 2290 dev_t dev = sbi->sb->s_bdev->bd_dev; 2291 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2292 int err = 0; 2293 2294 if (f2fs_sb_has_readonly(sbi)) { 2295 f2fs_info(sbi, 2296 "Skip to start discard thread for readonly image"); 2297 return 0; 2298 } 2299 2300 if (!f2fs_realtime_discard_enable(sbi)) 2301 return 0; 2302 2303 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2304 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2305 if (IS_ERR(dcc->f2fs_issue_discard)) { 2306 err = PTR_ERR(dcc->f2fs_issue_discard); 2307 dcc->f2fs_issue_discard = NULL; 2308 } 2309 2310 return err; 2311} 2312 2313static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2314{ 2315 struct discard_cmd_control *dcc; 2316 int err = 0, i; 2317 2318 if (SM_I(sbi)->dcc_info) { 2319 dcc = SM_I(sbi)->dcc_info; 2320 goto init_thread; 2321 } 2322 2323 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2324 if (!dcc) 2325 return -ENOMEM; 2326 2327 dcc->discard_io_aware_gran = MAX_PLIST_NUM; 2328 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2329 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY; 2330 dcc->discard_io_aware = DPOLICY_IO_AWARE_ENABLE; 2331 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT || 2332 F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION) 2333 dcc->discard_granularity = BLKS_PER_SEG(sbi); 2334 2335 INIT_LIST_HEAD(&dcc->entry_list); 2336 for (i = 0; i < MAX_PLIST_NUM; i++) 2337 INIT_LIST_HEAD(&dcc->pend_list[i]); 2338 INIT_LIST_HEAD(&dcc->wait_list); 2339 INIT_LIST_HEAD(&dcc->fstrim_list); 2340 mutex_init(&dcc->cmd_lock); 2341 atomic_set(&dcc->issued_discard, 0); 2342 atomic_set(&dcc->queued_discard, 0); 2343 atomic_set(&dcc->discard_cmd_cnt, 0); 2344 dcc->nr_discards = 0; 2345 dcc->max_discards = SEGS_TO_BLKS(sbi, MAIN_SEGS(sbi)); 2346 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST; 2347 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME; 2348 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME; 2349 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME; 2350 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL; 2351 dcc->undiscard_blks = 0; 2352 dcc->next_pos = 0; 2353 dcc->root = RB_ROOT_CACHED; 2354 dcc->rbtree_check = false; 2355 2356 init_waitqueue_head(&dcc->discard_wait_queue); 2357 SM_I(sbi)->dcc_info = dcc; 2358init_thread: 2359 err = f2fs_start_discard_thread(sbi); 2360 if (err) { 2361 kfree(dcc); 2362 SM_I(sbi)->dcc_info = NULL; 2363 } 2364 2365 return err; 2366} 2367 2368static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2369{ 2370 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2371 2372 if (!dcc) 2373 return; 2374 2375 f2fs_stop_discard_thread(sbi); 2376 2377 /* 2378 * Recovery can cache discard commands, so in error path of 2379 * fill_super(), it needs to give a chance to handle them. 2380 */ 2381 f2fs_issue_discard_timeout(sbi); 2382 2383 kfree(dcc); 2384 SM_I(sbi)->dcc_info = NULL; 2385} 2386 2387static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2388{ 2389 struct sit_info *sit_i = SIT_I(sbi); 2390 2391 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2392 sit_i->dirty_sentries++; 2393 return false; 2394 } 2395 2396 return true; 2397} 2398 2399static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2400 unsigned int segno, int modified) 2401{ 2402 struct seg_entry *se = get_seg_entry(sbi, segno); 2403 2404 se->type = type; 2405 if (modified) 2406 __mark_sit_entry_dirty(sbi, segno); 2407} 2408 2409static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi, 2410 block_t blkaddr) 2411{ 2412 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2413 2414 if (segno == NULL_SEGNO) 2415 return 0; 2416 return get_seg_entry(sbi, segno)->mtime; 2417} 2418 2419static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr, 2420 unsigned long long old_mtime) 2421{ 2422 struct seg_entry *se; 2423 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2424 unsigned long long ctime = get_mtime(sbi, false); 2425 unsigned long long mtime = old_mtime ? old_mtime : ctime; 2426 2427 if (segno == NULL_SEGNO) 2428 return; 2429 2430 se = get_seg_entry(sbi, segno); 2431 2432 if (!se->mtime) 2433 se->mtime = mtime; 2434 else 2435 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime, 2436 se->valid_blocks + 1); 2437 2438 if (ctime > SIT_I(sbi)->max_mtime) 2439 SIT_I(sbi)->max_mtime = ctime; 2440} 2441 2442/* 2443 * NOTE: when updating multiple blocks at the same time, please ensure 2444 * that the consecutive input blocks belong to the same segment. 2445 */ 2446static int update_sit_entry_for_release(struct f2fs_sb_info *sbi, struct seg_entry *se, 2447 unsigned int segno, block_t blkaddr, unsigned int offset, int del) 2448{ 2449 bool exist; 2450#ifdef CONFIG_F2FS_CHECK_FS 2451 bool mir_exist; 2452#endif 2453 int i; 2454 int del_count = -del; 2455 2456 f2fs_bug_on(sbi, GET_SEGNO(sbi, blkaddr) != GET_SEGNO(sbi, blkaddr + del_count - 1)); 2457 2458 for (i = 0; i < del_count; i++) { 2459 exist = f2fs_test_and_clear_bit(offset + i, se->cur_valid_map); 2460#ifdef CONFIG_F2FS_CHECK_FS 2461 mir_exist = f2fs_test_and_clear_bit(offset + i, 2462 se->cur_valid_map_mir); 2463 if (unlikely(exist != mir_exist)) { 2464 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2465 blkaddr + i, exist); 2466 f2fs_bug_on(sbi, 1); 2467 } 2468#endif 2469 if (unlikely(!exist)) { 2470 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", blkaddr + i); 2471 f2fs_bug_on(sbi, 1); 2472 se->valid_blocks++; 2473 del += 1; 2474 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2475 /* 2476 * If checkpoints are off, we must not reuse data that 2477 * was used in the previous checkpoint. If it was used 2478 * before, we must track that to know how much space we 2479 * really have. 2480 */ 2481 if (f2fs_test_bit(offset + i, se->ckpt_valid_map)) { 2482 spin_lock(&sbi->stat_lock); 2483 sbi->unusable_block_count++; 2484 spin_unlock(&sbi->stat_lock); 2485 } 2486 } 2487 2488 if (f2fs_block_unit_discard(sbi) && 2489 f2fs_test_and_clear_bit(offset + i, se->discard_map)) 2490 sbi->discard_blks++; 2491 2492 if (!f2fs_test_bit(offset + i, se->ckpt_valid_map)) { 2493 se->ckpt_valid_blocks -= 1; 2494 if (__is_large_section(sbi)) 2495 get_sec_entry(sbi, segno)->ckpt_valid_blocks -= 1; 2496 } 2497 } 2498 2499 if (__is_large_section(sbi)) 2500 sanity_check_valid_blocks(sbi, segno); 2501 2502 return del; 2503} 2504 2505static int update_sit_entry_for_alloc(struct f2fs_sb_info *sbi, struct seg_entry *se, 2506 unsigned int segno, block_t blkaddr, unsigned int offset, int del) 2507{ 2508 bool exist; 2509#ifdef CONFIG_F2FS_CHECK_FS 2510 bool mir_exist; 2511#endif 2512 2513 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2514#ifdef CONFIG_F2FS_CHECK_FS 2515 mir_exist = f2fs_test_and_set_bit(offset, 2516 se->cur_valid_map_mir); 2517 if (unlikely(exist != mir_exist)) { 2518 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2519 blkaddr, exist); 2520 f2fs_bug_on(sbi, 1); 2521 } 2522#endif 2523 if (unlikely(exist)) { 2524 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", blkaddr); 2525 f2fs_bug_on(sbi, 1); 2526 se->valid_blocks--; 2527 del = 0; 2528 } 2529 2530 if (f2fs_block_unit_discard(sbi) && 2531 !f2fs_test_and_set_bit(offset, se->discard_map)) 2532 sbi->discard_blks--; 2533 2534 /* 2535 * SSR should never reuse block which is checkpointed 2536 * or newly invalidated. 2537 */ 2538 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2539 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) { 2540 se->ckpt_valid_blocks++; 2541 if (__is_large_section(sbi)) 2542 get_sec_entry(sbi, segno)->ckpt_valid_blocks++; 2543 } 2544 } 2545 2546 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) { 2547 se->ckpt_valid_blocks += del; 2548 if (__is_large_section(sbi)) 2549 get_sec_entry(sbi, segno)->ckpt_valid_blocks += del; 2550 } 2551 2552 if (__is_large_section(sbi)) 2553 sanity_check_valid_blocks(sbi, segno); 2554 2555 return del; 2556} 2557 2558/* 2559 * If releasing blocks, this function supports updating multiple consecutive blocks 2560 * at one time, but please note that these consecutive blocks need to belong to the 2561 * same segment. 2562 */ 2563static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2564{ 2565 struct seg_entry *se; 2566 unsigned int segno, offset; 2567 long int new_vblocks; 2568 2569 segno = GET_SEGNO(sbi, blkaddr); 2570 if (segno == NULL_SEGNO) 2571 return; 2572 2573 se = get_seg_entry(sbi, segno); 2574 new_vblocks = se->valid_blocks + del; 2575 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2576 2577 f2fs_bug_on(sbi, (new_vblocks < 0 || 2578 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno)))); 2579 2580 se->valid_blocks = new_vblocks; 2581 2582 /* Update valid block bitmap */ 2583 if (del > 0) { 2584 del = update_sit_entry_for_alloc(sbi, se, segno, blkaddr, offset, del); 2585 } else { 2586 del = update_sit_entry_for_release(sbi, se, segno, blkaddr, offset, del); 2587 } 2588 2589 __mark_sit_entry_dirty(sbi, segno); 2590 2591 /* update total number of valid blocks to be written in ckpt area */ 2592 SIT_I(sbi)->written_valid_blocks += del; 2593 2594 if (__is_large_section(sbi)) 2595 get_sec_entry(sbi, segno)->valid_blocks += del; 2596} 2597 2598void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr, 2599 unsigned int len) 2600{ 2601 unsigned int segno = GET_SEGNO(sbi, addr); 2602 struct sit_info *sit_i = SIT_I(sbi); 2603 block_t addr_start = addr, addr_end = addr + len - 1; 2604 unsigned int seg_num = GET_SEGNO(sbi, addr_end) - segno + 1; 2605 unsigned int i = 1, max_blocks = sbi->blocks_per_seg, cnt; 2606 2607 f2fs_bug_on(sbi, addr == NULL_ADDR); 2608 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2609 return; 2610 2611 f2fs_invalidate_internal_cache(sbi, addr, len); 2612 2613 /* add it into sit main buffer */ 2614 down_write(&sit_i->sentry_lock); 2615 2616 if (seg_num == 1) 2617 cnt = len; 2618 else 2619 cnt = max_blocks - GET_BLKOFF_FROM_SEG0(sbi, addr); 2620 2621 do { 2622 update_segment_mtime(sbi, addr_start, 0); 2623 update_sit_entry(sbi, addr_start, -cnt); 2624 2625 /* add it into dirty seglist */ 2626 locate_dirty_segment(sbi, segno); 2627 2628 /* update @addr_start and @cnt and @segno */ 2629 addr_start = START_BLOCK(sbi, ++segno); 2630 if (++i == seg_num) 2631 cnt = GET_BLKOFF_FROM_SEG0(sbi, addr_end) + 1; 2632 else 2633 cnt = max_blocks; 2634 } while (i <= seg_num); 2635 2636 up_write(&sit_i->sentry_lock); 2637} 2638 2639bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2640{ 2641 struct sit_info *sit_i = SIT_I(sbi); 2642 unsigned int segno, offset; 2643 struct seg_entry *se; 2644 bool is_cp = false; 2645 2646 if (!__is_valid_data_blkaddr(blkaddr)) 2647 return true; 2648 2649 down_read(&sit_i->sentry_lock); 2650 2651 segno = GET_SEGNO(sbi, blkaddr); 2652 se = get_seg_entry(sbi, segno); 2653 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2654 2655 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2656 is_cp = true; 2657 2658 up_read(&sit_i->sentry_lock); 2659 2660 return is_cp; 2661} 2662 2663static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type) 2664{ 2665 struct curseg_info *curseg = CURSEG_I(sbi, type); 2666 2667 if (sbi->ckpt->alloc_type[type] == SSR) 2668 return BLKS_PER_SEG(sbi); 2669 return curseg->next_blkoff; 2670} 2671 2672/* 2673 * Calculate the number of current summary pages for writing 2674 */ 2675int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2676{ 2677 int valid_sum_count = 0; 2678 int i, sum_in_page; 2679 2680 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2681 if (sbi->ckpt->alloc_type[i] != SSR && for_ra) 2682 valid_sum_count += 2683 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2684 else 2685 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i); 2686 } 2687 2688 sum_in_page = (sbi->blocksize - 2 * sbi->sum_journal_size - 2689 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2690 if (valid_sum_count <= sum_in_page) 2691 return 1; 2692 else if ((valid_sum_count - sum_in_page) <= 2693 (sbi->blocksize - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2694 return 2; 2695 return 3; 2696} 2697 2698/* 2699 * Caller should put this summary folio 2700 */ 2701struct folio *f2fs_get_sum_folio(struct f2fs_sb_info *sbi, unsigned int segno) 2702{ 2703 if (unlikely(f2fs_cp_error(sbi))) 2704 return ERR_PTR(-EIO); 2705 return f2fs_get_meta_folio_retry(sbi, GET_SUM_BLOCK(sbi, segno)); 2706} 2707 2708void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2709 void *src, block_t blk_addr) 2710{ 2711 struct folio *folio; 2712 2713 if (!f2fs_sb_has_packed_ssa(sbi)) 2714 folio = f2fs_grab_meta_folio(sbi, blk_addr); 2715 else 2716 folio = f2fs_get_meta_folio_retry(sbi, blk_addr); 2717 2718 if (IS_ERR(folio)) 2719 return; 2720 2721 memcpy(folio_address(folio), src, PAGE_SIZE); 2722 folio_mark_dirty(folio); 2723 f2fs_folio_put(folio, true); 2724} 2725 2726static void write_sum_page(struct f2fs_sb_info *sbi, 2727 struct f2fs_summary_block *sum_blk, unsigned int segno) 2728{ 2729 struct folio *folio; 2730 2731 if (!f2fs_sb_has_packed_ssa(sbi)) 2732 return f2fs_update_meta_page(sbi, (void *)sum_blk, 2733 GET_SUM_BLOCK(sbi, segno)); 2734 2735 folio = f2fs_get_sum_folio(sbi, segno); 2736 if (IS_ERR(folio)) 2737 return; 2738 2739 memcpy(SUM_BLK_PAGE_ADDR(sbi, folio, segno), sum_blk, 2740 sbi->sum_blocksize); 2741 folio_mark_dirty(folio); 2742 f2fs_folio_put(folio, true); 2743} 2744 2745static void write_current_sum_page(struct f2fs_sb_info *sbi, 2746 int type, block_t blk_addr) 2747{ 2748 struct curseg_info *curseg = CURSEG_I(sbi, type); 2749 struct folio *folio = f2fs_grab_meta_folio(sbi, blk_addr); 2750 struct f2fs_summary_block *src = curseg->sum_blk; 2751 struct f2fs_summary_block *dst; 2752 2753 dst = folio_address(folio); 2754 memset(dst, 0, PAGE_SIZE); 2755 2756 mutex_lock(&curseg->curseg_mutex); 2757 2758 down_read(&curseg->journal_rwsem); 2759 memcpy(sum_journal(sbi, dst), curseg->journal, sbi->sum_journal_size); 2760 up_read(&curseg->journal_rwsem); 2761 2762 memcpy(sum_entries(dst), sum_entries(src), sbi->sum_entry_size); 2763 memcpy(sum_footer(sbi, dst), sum_footer(sbi, src), SUM_FOOTER_SIZE); 2764 2765 mutex_unlock(&curseg->curseg_mutex); 2766 2767 folio_mark_dirty(folio); 2768 f2fs_folio_put(folio, true); 2769} 2770 2771static int is_next_segment_free(struct f2fs_sb_info *sbi, 2772 struct curseg_info *curseg) 2773{ 2774 unsigned int segno = curseg->segno + 1; 2775 struct free_segmap_info *free_i = FREE_I(sbi); 2776 2777 if (segno < MAIN_SEGS(sbi) && segno % SEGS_PER_SEC(sbi)) 2778 return !test_bit(segno, free_i->free_segmap); 2779 return 0; 2780} 2781 2782/* 2783 * Find a new segment from the free segments bitmap to right order 2784 * This function should be returned with success, otherwise BUG 2785 */ 2786static int get_new_segment(struct f2fs_sb_info *sbi, 2787 unsigned int *newseg, bool new_sec, bool pinning) 2788{ 2789 struct free_segmap_info *free_i = FREE_I(sbi); 2790 unsigned int segno, secno, zoneno; 2791 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2792 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2793 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2794 unsigned int alloc_policy = sbi->allocate_section_policy; 2795 unsigned int alloc_hint = sbi->allocate_section_hint; 2796 bool init = true; 2797 int i; 2798 int ret = 0; 2799 2800 spin_lock(&free_i->segmap_lock); 2801 2802 if (time_to_inject(sbi, FAULT_NO_SEGMENT)) { 2803 ret = -ENOSPC; 2804 goto out_unlock; 2805 } 2806 2807 if (!new_sec && ((*newseg + 1) % SEGS_PER_SEC(sbi))) { 2808 segno = find_next_zero_bit(free_i->free_segmap, 2809 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2810 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2811 goto got_it; 2812 } 2813 2814#ifdef CONFIG_BLK_DEV_ZONED 2815 /* 2816 * If we format f2fs on zoned storage, let's try to get pinned sections 2817 * from beginning of the storage, which should be a conventional one. 2818 */ 2819 if (f2fs_sb_has_blkzoned(sbi)) { 2820 /* Prioritize writing to conventional zones */ 2821 if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_PRIOR_CONV || pinning) 2822 segno = 0; 2823 else 2824 segno = max(sbi->first_seq_zone_segno, *newseg); 2825 hint = GET_SEC_FROM_SEG(sbi, segno); 2826 } 2827#endif 2828 2829 /* 2830 * Prevent allocate_section_hint from exceeding MAIN_SECS() 2831 * due to desynchronization. 2832 */ 2833 if (alloc_policy != ALLOCATE_FORWARD_NOHINT && 2834 alloc_hint > MAIN_SECS(sbi)) 2835 alloc_hint = MAIN_SECS(sbi); 2836 2837 if (alloc_policy == ALLOCATE_FORWARD_FROM_HINT && 2838 hint < alloc_hint) 2839 hint = alloc_hint; 2840 else if (alloc_policy == ALLOCATE_FORWARD_WITHIN_HINT && 2841 hint >= alloc_hint) 2842 hint = 0; 2843 2844find_other_zone: 2845 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2846 2847#ifdef CONFIG_BLK_DEV_ZONED 2848 if (secno >= MAIN_SECS(sbi) && f2fs_sb_has_blkzoned(sbi)) { 2849 /* Write only to sequential zones */ 2850 if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_ONLY_SEQ) { 2851 hint = GET_SEC_FROM_SEG(sbi, sbi->first_seq_zone_segno); 2852 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2853 } else 2854 secno = find_first_zero_bit(free_i->free_secmap, 2855 MAIN_SECS(sbi)); 2856 if (secno >= MAIN_SECS(sbi)) { 2857 ret = -ENOSPC; 2858 f2fs_bug_on(sbi, 1); 2859 goto out_unlock; 2860 } 2861 } 2862#endif 2863 2864 if (secno >= MAIN_SECS(sbi)) { 2865 secno = find_first_zero_bit(free_i->free_secmap, 2866 MAIN_SECS(sbi)); 2867 if (secno >= MAIN_SECS(sbi)) { 2868 ret = -ENOSPC; 2869 f2fs_bug_on(sbi, !pinning); 2870 goto out_unlock; 2871 } 2872 } 2873 segno = GET_SEG_FROM_SEC(sbi, secno); 2874 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2875 2876 /* give up on finding another zone */ 2877 if (!init) 2878 goto got_it; 2879 if (sbi->secs_per_zone == 1) 2880 goto got_it; 2881 if (zoneno == old_zoneno) 2882 goto got_it; 2883 for (i = 0; i < NR_CURSEG_TYPE; i++) 2884 if (CURSEG_I(sbi, i)->zone == zoneno) 2885 break; 2886 2887 if (i < NR_CURSEG_TYPE) { 2888 /* zone is in user, try another */ 2889 if (zoneno + 1 >= total_zones) 2890 hint = 0; 2891 else 2892 hint = (zoneno + 1) * sbi->secs_per_zone; 2893 init = false; 2894 goto find_other_zone; 2895 } 2896got_it: 2897 /* set it as dirty segment in free segmap */ 2898 if (test_bit(segno, free_i->free_segmap)) { 2899 ret = -EFSCORRUPTED; 2900 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_CORRUPTED_FREE_BITMAP); 2901 goto out_unlock; 2902 } 2903 2904 /* no free section in conventional device or conventional zone */ 2905 if (new_sec && pinning && 2906 f2fs_is_sequential_zone_area(sbi, START_BLOCK(sbi, segno))) { 2907 ret = -EAGAIN; 2908 goto out_unlock; 2909 } 2910 __set_inuse(sbi, segno); 2911 *newseg = segno; 2912out_unlock: 2913 spin_unlock(&free_i->segmap_lock); 2914 2915 if (ret == -ENOSPC && !pinning) 2916 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_NO_SEGMENT); 2917 return ret; 2918} 2919 2920static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2921{ 2922 struct curseg_info *curseg = CURSEG_I(sbi, type); 2923 struct summary_footer *sum_footer; 2924 unsigned short seg_type = curseg->seg_type; 2925 2926 /* only happen when get_new_segment() fails */ 2927 if (curseg->next_segno == NULL_SEGNO) 2928 return; 2929 2930 curseg->inited = true; 2931 curseg->segno = curseg->next_segno; 2932 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2933 curseg->next_blkoff = 0; 2934 curseg->next_segno = NULL_SEGNO; 2935 2936 sum_footer = sum_footer(sbi, curseg->sum_blk); 2937 memset(sum_footer, 0, sizeof(struct summary_footer)); 2938 2939 sanity_check_seg_type(sbi, seg_type); 2940 2941 if (IS_DATASEG(seg_type)) 2942 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2943 if (IS_NODESEG(seg_type)) 2944 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2945 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified); 2946} 2947 2948static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2949{ 2950 struct curseg_info *curseg = CURSEG_I(sbi, type); 2951 unsigned short seg_type = curseg->seg_type; 2952 2953 sanity_check_seg_type(sbi, seg_type); 2954 if (__is_large_section(sbi)) { 2955 if (f2fs_need_rand_seg(sbi)) { 2956 unsigned int hint = GET_SEC_FROM_SEG(sbi, curseg->segno); 2957 2958 if (GET_SEC_FROM_SEG(sbi, curseg->segno + 1) != hint) 2959 return curseg->segno; 2960 return get_random_u32_inclusive(curseg->segno + 1, 2961 GET_SEG_FROM_SEC(sbi, hint + 1) - 1); 2962 } 2963 return curseg->segno; 2964 } else if (f2fs_need_rand_seg(sbi)) { 2965 return get_random_u32_below(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi)); 2966 } 2967 2968 /* inmem log may not locate on any segment after mount */ 2969 if (!curseg->inited) 2970 return 0; 2971 2972 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2973 return 0; 2974 2975 if (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)) 2976 return 0; 2977 2978 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2979 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2980 2981 /* find segments from 0 to reuse freed segments */ 2982 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2983 return 0; 2984 2985 return curseg->segno; 2986} 2987 2988static void reset_curseg_fields(struct curseg_info *curseg) 2989{ 2990 curseg->inited = false; 2991 curseg->segno = NULL_SEGNO; 2992 curseg->next_segno = 0; 2993} 2994 2995/* 2996 * Allocate a current working segment. 2997 * This function always allocates a free segment in LFS manner. 2998 */ 2999static int new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 3000{ 3001 struct curseg_info *curseg = CURSEG_I(sbi, type); 3002 unsigned int segno = curseg->segno; 3003 bool pinning = type == CURSEG_COLD_DATA_PINNED; 3004 int ret; 3005 3006 if (curseg->inited) 3007 write_sum_page(sbi, curseg->sum_blk, segno); 3008 3009 segno = __get_next_segno(sbi, type); 3010 ret = get_new_segment(sbi, &segno, new_sec, pinning); 3011 if (ret) { 3012 if (ret == -ENOSPC) 3013 reset_curseg_fields(curseg); 3014 return ret; 3015 } 3016 3017 curseg->next_segno = segno; 3018 reset_curseg(sbi, type, 1); 3019 curseg->alloc_type = LFS; 3020 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 3021 curseg->fragment_remained_chunk = 3022 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 3023 return 0; 3024} 3025 3026static int __next_free_blkoff(struct f2fs_sb_info *sbi, 3027 int segno, block_t start) 3028{ 3029 struct seg_entry *se = get_seg_entry(sbi, segno); 3030 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 3031 unsigned long *target_map = SIT_I(sbi)->tmp_map; 3032 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 3033 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 3034 int i; 3035 3036 for (i = 0; i < entries; i++) 3037 target_map[i] = ckpt_map[i] | cur_map[i]; 3038 3039 return __find_rev_next_zero_bit(target_map, BLKS_PER_SEG(sbi), start); 3040} 3041 3042static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi, 3043 struct curseg_info *seg) 3044{ 3045 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1); 3046} 3047 3048bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno) 3049{ 3050 return __next_free_blkoff(sbi, segno, 0) < BLKS_PER_SEG(sbi); 3051} 3052 3053/* 3054 * This function always allocates a used segment(from dirty seglist) by SSR 3055 * manner, so it should recover the existing segment information of valid blocks 3056 */ 3057static int change_curseg(struct f2fs_sb_info *sbi, int type) 3058{ 3059 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3060 struct curseg_info *curseg = CURSEG_I(sbi, type); 3061 unsigned int new_segno = curseg->next_segno; 3062 struct f2fs_summary_block *sum_node; 3063 struct folio *sum_folio; 3064 3065 if (curseg->inited) 3066 write_sum_page(sbi, curseg->sum_blk, curseg->segno); 3067 3068 __set_test_and_inuse(sbi, new_segno); 3069 3070 mutex_lock(&dirty_i->seglist_lock); 3071 __remove_dirty_segment(sbi, new_segno, PRE); 3072 __remove_dirty_segment(sbi, new_segno, DIRTY); 3073 mutex_unlock(&dirty_i->seglist_lock); 3074 3075 reset_curseg(sbi, type, 1); 3076 curseg->alloc_type = SSR; 3077 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0); 3078 3079 sum_folio = f2fs_get_sum_folio(sbi, new_segno); 3080 if (IS_ERR(sum_folio)) { 3081 /* GC won't be able to use stale summary pages by cp_error */ 3082 memset(curseg->sum_blk, 0, sbi->sum_entry_size); 3083 return PTR_ERR(sum_folio); 3084 } 3085 sum_node = SUM_BLK_PAGE_ADDR(sbi, sum_folio, new_segno); 3086 memcpy(curseg->sum_blk, sum_node, sbi->sum_entry_size); 3087 f2fs_folio_put(sum_folio, true); 3088 return 0; 3089} 3090 3091static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 3092 int alloc_mode, unsigned long long age); 3093 3094static int get_atssr_segment(struct f2fs_sb_info *sbi, int type, 3095 int target_type, int alloc_mode, 3096 unsigned long long age) 3097{ 3098 struct curseg_info *curseg = CURSEG_I(sbi, type); 3099 int ret = 0; 3100 3101 curseg->seg_type = target_type; 3102 3103 if (get_ssr_segment(sbi, type, alloc_mode, age)) { 3104 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno); 3105 3106 curseg->seg_type = se->type; 3107 ret = change_curseg(sbi, type); 3108 } else { 3109 /* allocate cold segment by default */ 3110 curseg->seg_type = CURSEG_COLD_DATA; 3111 ret = new_curseg(sbi, type, true); 3112 } 3113 stat_inc_seg_type(sbi, curseg); 3114 return ret; 3115} 3116 3117static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi, bool force) 3118{ 3119 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC); 3120 int ret = 0; 3121 3122 if (!sbi->am.atgc_enabled && !force) 3123 return 0; 3124 3125 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3126 3127 mutex_lock(&curseg->curseg_mutex); 3128 down_write(&SIT_I(sbi)->sentry_lock); 3129 3130 ret = get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, 3131 CURSEG_COLD_DATA, SSR, 0); 3132 3133 up_write(&SIT_I(sbi)->sentry_lock); 3134 mutex_unlock(&curseg->curseg_mutex); 3135 3136 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3137 return ret; 3138} 3139 3140int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi) 3141{ 3142 return __f2fs_init_atgc_curseg(sbi, false); 3143} 3144 3145int f2fs_reinit_atgc_curseg(struct f2fs_sb_info *sbi) 3146{ 3147 int ret; 3148 3149 if (!test_opt(sbi, ATGC)) 3150 return 0; 3151 if (sbi->am.atgc_enabled) 3152 return 0; 3153 if (le64_to_cpu(F2FS_CKPT(sbi)->elapsed_time) < 3154 sbi->am.age_threshold) 3155 return 0; 3156 3157 ret = __f2fs_init_atgc_curseg(sbi, true); 3158 if (!ret) { 3159 sbi->am.atgc_enabled = true; 3160 f2fs_info(sbi, "reenabled age threshold GC"); 3161 } 3162 return ret; 3163} 3164 3165static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type) 3166{ 3167 struct curseg_info *curseg = CURSEG_I(sbi, type); 3168 3169 mutex_lock(&curseg->curseg_mutex); 3170 if (!curseg->inited) 3171 goto out; 3172 3173 if (get_valid_blocks(sbi, curseg->segno, false)) { 3174 write_sum_page(sbi, curseg->sum_blk, curseg->segno); 3175 } else { 3176 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 3177 __set_test_and_free(sbi, curseg->segno, true); 3178 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 3179 } 3180out: 3181 mutex_unlock(&curseg->curseg_mutex); 3182} 3183 3184void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi) 3185{ 3186 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 3187 3188 if (sbi->am.atgc_enabled) 3189 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 3190} 3191 3192static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type) 3193{ 3194 struct curseg_info *curseg = CURSEG_I(sbi, type); 3195 3196 mutex_lock(&curseg->curseg_mutex); 3197 if (!curseg->inited) 3198 goto out; 3199 if (get_valid_blocks(sbi, curseg->segno, false)) 3200 goto out; 3201 3202 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 3203 __set_test_and_inuse(sbi, curseg->segno); 3204 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 3205out: 3206 mutex_unlock(&curseg->curseg_mutex); 3207} 3208 3209void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi) 3210{ 3211 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 3212 3213 if (sbi->am.atgc_enabled) 3214 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 3215} 3216 3217static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 3218 int alloc_mode, unsigned long long age) 3219{ 3220 struct curseg_info *curseg = CURSEG_I(sbi, type); 3221 unsigned segno = NULL_SEGNO; 3222 unsigned short seg_type = curseg->seg_type; 3223 int i, cnt; 3224 bool reversed = false; 3225 3226 sanity_check_seg_type(sbi, seg_type); 3227 3228 /* f2fs_need_SSR() already forces to do this */ 3229 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, 3230 alloc_mode, age, false)) { 3231 curseg->next_segno = segno; 3232 return 1; 3233 } 3234 3235 /* For node segments, let's do SSR more intensively */ 3236 if (IS_NODESEG(seg_type)) { 3237 if (seg_type >= CURSEG_WARM_NODE) { 3238 reversed = true; 3239 i = CURSEG_COLD_NODE; 3240 } else { 3241 i = CURSEG_HOT_NODE; 3242 } 3243 cnt = NR_CURSEG_NODE_TYPE; 3244 } else { 3245 if (seg_type >= CURSEG_WARM_DATA) { 3246 reversed = true; 3247 i = CURSEG_COLD_DATA; 3248 } else { 3249 i = CURSEG_HOT_DATA; 3250 } 3251 cnt = NR_CURSEG_DATA_TYPE; 3252 } 3253 3254 for (; cnt-- > 0; reversed ? i-- : i++) { 3255 if (i == seg_type) 3256 continue; 3257 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, 3258 alloc_mode, age, false)) { 3259 curseg->next_segno = segno; 3260 return 1; 3261 } 3262 } 3263 3264 /* find valid_blocks=0 in dirty list */ 3265 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 3266 segno = get_free_segment(sbi); 3267 if (segno != NULL_SEGNO) { 3268 curseg->next_segno = segno; 3269 return 1; 3270 } 3271 } 3272 return 0; 3273} 3274 3275static bool need_new_seg(struct f2fs_sb_info *sbi, int type) 3276{ 3277 struct curseg_info *curseg = CURSEG_I(sbi, type); 3278 3279 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 3280 curseg->seg_type == CURSEG_WARM_NODE) 3281 return true; 3282 if (curseg->alloc_type == LFS && is_next_segment_free(sbi, curseg) && 3283 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 3284 return true; 3285 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0)) 3286 return true; 3287 return false; 3288} 3289 3290int f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 3291 unsigned int start, unsigned int end) 3292{ 3293 struct curseg_info *curseg = CURSEG_I(sbi, type); 3294 unsigned int segno; 3295 int ret = 0; 3296 3297 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3298 mutex_lock(&curseg->curseg_mutex); 3299 down_write(&SIT_I(sbi)->sentry_lock); 3300 3301 segno = CURSEG_I(sbi, type)->segno; 3302 if (segno < start || segno > end) 3303 goto unlock; 3304 3305 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0)) 3306 ret = change_curseg(sbi, type); 3307 else 3308 ret = new_curseg(sbi, type, true); 3309 3310 stat_inc_seg_type(sbi, curseg); 3311 3312 locate_dirty_segment(sbi, segno); 3313unlock: 3314 up_write(&SIT_I(sbi)->sentry_lock); 3315 3316 if (segno != curseg->segno) 3317 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 3318 type, segno, curseg->segno); 3319 3320 mutex_unlock(&curseg->curseg_mutex); 3321 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3322 return ret; 3323} 3324 3325static int __allocate_new_segment(struct f2fs_sb_info *sbi, int type, 3326 bool new_sec, bool force) 3327{ 3328 struct curseg_info *curseg = CURSEG_I(sbi, type); 3329 unsigned int old_segno; 3330 int err = 0; 3331 3332 if (type == CURSEG_COLD_DATA_PINNED && !curseg->inited) 3333 goto allocate; 3334 3335 if (!force && curseg->inited && 3336 !curseg->next_blkoff && 3337 !get_valid_blocks(sbi, curseg->segno, new_sec) && 3338 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec)) 3339 return 0; 3340 3341allocate: 3342 old_segno = curseg->segno; 3343 err = new_curseg(sbi, type, true); 3344 if (err) 3345 return err; 3346 stat_inc_seg_type(sbi, curseg); 3347 locate_dirty_segment(sbi, old_segno); 3348 return 0; 3349} 3350 3351int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force) 3352{ 3353 int ret; 3354 3355 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3356 down_write(&SIT_I(sbi)->sentry_lock); 3357 ret = __allocate_new_segment(sbi, type, true, force); 3358 up_write(&SIT_I(sbi)->sentry_lock); 3359 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3360 3361 return ret; 3362} 3363 3364int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi) 3365{ 3366 int err; 3367 bool gc_required = true; 3368 3369retry: 3370 f2fs_lock_op(sbi); 3371 err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false); 3372 f2fs_unlock_op(sbi); 3373 3374 if (f2fs_sb_has_blkzoned(sbi) && err == -EAGAIN && gc_required) { 3375 f2fs_down_write(&sbi->gc_lock); 3376 err = f2fs_gc_range(sbi, 0, sbi->first_seq_zone_segno - 1, 3377 true, ZONED_PIN_SEC_REQUIRED_COUNT); 3378 f2fs_up_write(&sbi->gc_lock); 3379 3380 gc_required = false; 3381 if (!err) 3382 goto retry; 3383 } 3384 3385 return err; 3386} 3387 3388int f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 3389{ 3390 int i; 3391 int err = 0; 3392 3393 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3394 down_write(&SIT_I(sbi)->sentry_lock); 3395 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 3396 err += __allocate_new_segment(sbi, i, false, false); 3397 up_write(&SIT_I(sbi)->sentry_lock); 3398 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3399 3400 return err; 3401} 3402 3403bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 3404 struct cp_control *cpc) 3405{ 3406 __u64 trim_start = cpc->trim_start; 3407 bool has_candidate = false; 3408 3409 down_write(&SIT_I(sbi)->sentry_lock); 3410 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 3411 if (add_discard_addrs(sbi, cpc, true)) { 3412 has_candidate = true; 3413 break; 3414 } 3415 } 3416 up_write(&SIT_I(sbi)->sentry_lock); 3417 3418 cpc->trim_start = trim_start; 3419 return has_candidate; 3420} 3421 3422static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 3423 struct discard_policy *dpolicy, 3424 unsigned int start, unsigned int end) 3425{ 3426 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 3427 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 3428 struct rb_node **insert_p = NULL, *insert_parent = NULL; 3429 struct discard_cmd *dc; 3430 struct blk_plug plug; 3431 int issued; 3432 unsigned int trimmed = 0; 3433 3434next: 3435 issued = 0; 3436 3437 mutex_lock(&dcc->cmd_lock); 3438 if (unlikely(dcc->rbtree_check)) 3439 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 3440 3441 dc = __lookup_discard_cmd_ret(&dcc->root, start, 3442 &prev_dc, &next_dc, &insert_p, &insert_parent); 3443 if (!dc) 3444 dc = next_dc; 3445 3446 blk_start_plug(&plug); 3447 3448 while (dc && dc->di.lstart <= end) { 3449 struct rb_node *node; 3450 int err = 0; 3451 3452 if (dc->di.len < dpolicy->granularity) 3453 goto skip; 3454 3455 if (dc->state != D_PREP) { 3456 list_move_tail(&dc->list, &dcc->fstrim_list); 3457 goto skip; 3458 } 3459 3460 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 3461 3462 if (issued >= dpolicy->max_requests) { 3463 start = dc->di.lstart + dc->di.len; 3464 3465 if (err) 3466 __remove_discard_cmd(sbi, dc); 3467 3468 blk_finish_plug(&plug); 3469 mutex_unlock(&dcc->cmd_lock); 3470 trimmed += __wait_all_discard_cmd(sbi, NULL); 3471 f2fs_schedule_timeout(DEFAULT_DISCARD_INTERVAL); 3472 goto next; 3473 } 3474skip: 3475 node = rb_next(&dc->rb_node); 3476 if (err) 3477 __remove_discard_cmd(sbi, dc); 3478 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 3479 3480 if (fatal_signal_pending(current)) 3481 break; 3482 } 3483 3484 blk_finish_plug(&plug); 3485 mutex_unlock(&dcc->cmd_lock); 3486 3487 return trimmed; 3488} 3489 3490int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 3491{ 3492 __u64 start = F2FS_BYTES_TO_BLK(range->start); 3493 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 3494 unsigned int start_segno, end_segno; 3495 block_t start_block, end_block; 3496 struct cp_control cpc; 3497 struct discard_policy dpolicy; 3498 unsigned long long trimmed = 0; 3499 int err = 0; 3500 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 3501 3502 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 3503 return -EINVAL; 3504 3505 if (end < MAIN_BLKADDR(sbi)) 3506 goto out; 3507 3508 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 3509 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 3510 return -EFSCORRUPTED; 3511 } 3512 3513 /* start/end segment number in main_area */ 3514 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 3515 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 3516 GET_SEGNO(sbi, end); 3517 if (need_align) { 3518 start_segno = rounddown(start_segno, SEGS_PER_SEC(sbi)); 3519 end_segno = roundup(end_segno + 1, SEGS_PER_SEC(sbi)) - 1; 3520 } 3521 3522 cpc.reason = CP_DISCARD; 3523 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 3524 cpc.trim_start = start_segno; 3525 cpc.trim_end = end_segno; 3526 3527 if (sbi->discard_blks == 0) 3528 goto out; 3529 3530 f2fs_down_write(&sbi->gc_lock); 3531 stat_inc_cp_call_count(sbi, TOTAL_CALL); 3532 err = f2fs_write_checkpoint(sbi, &cpc); 3533 f2fs_up_write(&sbi->gc_lock); 3534 if (err) 3535 goto out; 3536 3537 /* 3538 * We filed discard candidates, but actually we don't need to wait for 3539 * all of them, since they'll be issued in idle time along with runtime 3540 * discard option. User configuration looks like using runtime discard 3541 * or periodic fstrim instead of it. 3542 */ 3543 if (f2fs_realtime_discard_enable(sbi)) 3544 goto out; 3545 3546 start_block = START_BLOCK(sbi, start_segno); 3547 end_block = START_BLOCK(sbi, end_segno + 1); 3548 3549 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 3550 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 3551 start_block, end_block); 3552 3553 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 3554 start_block, end_block); 3555out: 3556 if (!err) 3557 range->len = F2FS_BLK_TO_BYTES(trimmed); 3558 return err; 3559} 3560 3561int f2fs_rw_hint_to_seg_type(struct f2fs_sb_info *sbi, enum rw_hint hint) 3562{ 3563 if (F2FS_OPTION(sbi).active_logs == 2) 3564 return CURSEG_HOT_DATA; 3565 else if (F2FS_OPTION(sbi).active_logs == 4) 3566 return CURSEG_COLD_DATA; 3567 3568 /* active_log == 6 */ 3569 switch (hint) { 3570 case WRITE_LIFE_SHORT: 3571 return CURSEG_HOT_DATA; 3572 case WRITE_LIFE_EXTREME: 3573 return CURSEG_COLD_DATA; 3574 default: 3575 return CURSEG_WARM_DATA; 3576 } 3577} 3578 3579/* 3580 * This returns write hints for each segment type. This hints will be 3581 * passed down to block layer as below by default. 3582 * 3583 * User F2FS Block 3584 * ---- ---- ----- 3585 * META WRITE_LIFE_NONE|REQ_META 3586 * HOT_NODE WRITE_LIFE_NONE 3587 * WARM_NODE WRITE_LIFE_MEDIUM 3588 * COLD_NODE WRITE_LIFE_LONG 3589 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 3590 * extension list " " 3591 * 3592 * -- buffered io 3593 * COLD_DATA WRITE_LIFE_EXTREME 3594 * HOT_DATA WRITE_LIFE_SHORT 3595 * WARM_DATA WRITE_LIFE_NOT_SET 3596 * 3597 * -- direct io 3598 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 3599 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 3600 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 3601 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 3602 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 3603 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 3604 */ 3605enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi, 3606 enum page_type type, enum temp_type temp) 3607{ 3608 switch (type) { 3609 case DATA: 3610 switch (temp) { 3611 case WARM: 3612 return WRITE_LIFE_NOT_SET; 3613 case HOT: 3614 return WRITE_LIFE_SHORT; 3615 case COLD: 3616 return WRITE_LIFE_EXTREME; 3617 default: 3618 return WRITE_LIFE_NONE; 3619 } 3620 case NODE: 3621 switch (temp) { 3622 case WARM: 3623 return WRITE_LIFE_MEDIUM; 3624 case HOT: 3625 return WRITE_LIFE_NONE; 3626 case COLD: 3627 return WRITE_LIFE_LONG; 3628 default: 3629 return WRITE_LIFE_NONE; 3630 } 3631 case META: 3632 return WRITE_LIFE_NONE; 3633 default: 3634 return WRITE_LIFE_NONE; 3635 } 3636} 3637 3638static int __get_segment_type_2(struct f2fs_io_info *fio) 3639{ 3640 if (fio->type == DATA) 3641 return CURSEG_HOT_DATA; 3642 else 3643 return CURSEG_HOT_NODE; 3644} 3645 3646static int __get_segment_type_4(struct f2fs_io_info *fio) 3647{ 3648 if (fio->type == DATA) { 3649 struct inode *inode = fio_inode(fio); 3650 3651 if (S_ISDIR(inode->i_mode)) 3652 return CURSEG_HOT_DATA; 3653 else 3654 return CURSEG_COLD_DATA; 3655 } else { 3656 if (IS_DNODE(fio->folio) && is_cold_node(fio->folio)) 3657 return CURSEG_WARM_NODE; 3658 else 3659 return CURSEG_COLD_NODE; 3660 } 3661} 3662 3663static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs) 3664{ 3665 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3666 struct extent_info ei = {}; 3667 3668 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) { 3669 if (!ei.age) 3670 return NO_CHECK_TYPE; 3671 if (ei.age <= sbi->hot_data_age_threshold) 3672 return CURSEG_HOT_DATA; 3673 if (ei.age <= sbi->warm_data_age_threshold) 3674 return CURSEG_WARM_DATA; 3675 return CURSEG_COLD_DATA; 3676 } 3677 return NO_CHECK_TYPE; 3678} 3679 3680static int __get_segment_type_6(struct f2fs_io_info *fio) 3681{ 3682 if (fio->type == DATA) { 3683 struct inode *inode = fio_inode(fio); 3684 int type; 3685 3686 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 3687 return CURSEG_COLD_DATA_PINNED; 3688 3689 if (page_private_gcing(fio->page)) { 3690 if (fio->sbi->am.atgc_enabled && 3691 (fio->io_type == FS_DATA_IO) && 3692 (fio->sbi->gc_mode != GC_URGENT_HIGH) && 3693 __is_valid_data_blkaddr(fio->old_blkaddr) && 3694 !is_inode_flag_set(inode, FI_OPU_WRITE)) 3695 return CURSEG_ALL_DATA_ATGC; 3696 else 3697 return CURSEG_COLD_DATA; 3698 } 3699 if (file_is_cold(inode) || f2fs_need_compress_data(inode)) 3700 return CURSEG_COLD_DATA; 3701 3702 type = __get_age_segment_type(inode, fio->folio->index); 3703 if (type != NO_CHECK_TYPE) 3704 return type; 3705 3706 if (file_is_hot(inode) || 3707 is_inode_flag_set(inode, FI_HOT_DATA) || 3708 f2fs_is_cow_file(inode) || 3709 is_inode_flag_set(inode, FI_NEED_IPU)) 3710 return CURSEG_HOT_DATA; 3711 return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode), 3712 inode->i_write_hint); 3713 } else { 3714 if (IS_DNODE(fio->folio)) 3715 return is_cold_node(fio->folio) ? CURSEG_WARM_NODE : 3716 CURSEG_HOT_NODE; 3717 return CURSEG_COLD_NODE; 3718 } 3719} 3720 3721enum temp_type f2fs_get_segment_temp(struct f2fs_sb_info *sbi, 3722 enum log_type type) 3723{ 3724 struct curseg_info *curseg = CURSEG_I(sbi, type); 3725 enum temp_type temp = COLD; 3726 3727 switch (curseg->seg_type) { 3728 case CURSEG_HOT_NODE: 3729 case CURSEG_HOT_DATA: 3730 temp = HOT; 3731 break; 3732 case CURSEG_WARM_NODE: 3733 case CURSEG_WARM_DATA: 3734 temp = WARM; 3735 break; 3736 case CURSEG_COLD_NODE: 3737 case CURSEG_COLD_DATA: 3738 temp = COLD; 3739 break; 3740 default: 3741 f2fs_bug_on(sbi, 1); 3742 } 3743 3744 return temp; 3745} 3746 3747static int __get_segment_type(struct f2fs_io_info *fio) 3748{ 3749 enum log_type type = CURSEG_HOT_DATA; 3750 3751 switch (F2FS_OPTION(fio->sbi).active_logs) { 3752 case 2: 3753 type = __get_segment_type_2(fio); 3754 break; 3755 case 4: 3756 type = __get_segment_type_4(fio); 3757 break; 3758 case 6: 3759 type = __get_segment_type_6(fio); 3760 break; 3761 default: 3762 f2fs_bug_on(fio->sbi, true); 3763 } 3764 3765 fio->temp = f2fs_get_segment_temp(fio->sbi, type); 3766 3767 return type; 3768} 3769 3770static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi, 3771 struct curseg_info *seg) 3772{ 3773 /* To allocate block chunks in different sizes, use random number */ 3774 if (--seg->fragment_remained_chunk > 0) 3775 return; 3776 3777 seg->fragment_remained_chunk = 3778 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 3779 seg->next_blkoff += 3780 get_random_u32_inclusive(1, sbi->max_fragment_hole); 3781} 3782 3783int f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct folio *folio, 3784 block_t old_blkaddr, block_t *new_blkaddr, 3785 struct f2fs_summary *sum, int type, 3786 struct f2fs_io_info *fio) 3787{ 3788 struct sit_info *sit_i = SIT_I(sbi); 3789 struct curseg_info *curseg = CURSEG_I(sbi, type); 3790 unsigned long long old_mtime; 3791 bool from_gc = (type == CURSEG_ALL_DATA_ATGC); 3792 struct seg_entry *se = NULL; 3793 bool segment_full = false; 3794 int ret = 0; 3795 3796 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3797 3798 mutex_lock(&curseg->curseg_mutex); 3799 down_write(&sit_i->sentry_lock); 3800 3801 if (curseg->segno == NULL_SEGNO) { 3802 ret = -ENOSPC; 3803 goto out_err; 3804 } 3805 3806 if (from_gc) { 3807 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO); 3808 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr)); 3809 sanity_check_seg_type(sbi, se->type); 3810 f2fs_bug_on(sbi, IS_NODESEG(se->type)); 3811 } 3812 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3813 3814 f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi)); 3815 3816 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3817 3818 sum_entries(curseg->sum_blk)[curseg->next_blkoff] = *sum; 3819 if (curseg->alloc_type == SSR) { 3820 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg); 3821 } else { 3822 curseg->next_blkoff++; 3823 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 3824 f2fs_randomize_chunk(sbi, curseg); 3825 } 3826 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno)) 3827 segment_full = true; 3828 stat_inc_block_count(sbi, curseg); 3829 3830 if (from_gc) { 3831 old_mtime = get_segment_mtime(sbi, old_blkaddr); 3832 } else { 3833 update_segment_mtime(sbi, old_blkaddr, 0); 3834 old_mtime = 0; 3835 } 3836 update_segment_mtime(sbi, *new_blkaddr, old_mtime); 3837 3838 /* 3839 * SIT information should be updated before segment allocation, 3840 * since SSR needs latest valid block information. 3841 */ 3842 update_sit_entry(sbi, *new_blkaddr, 1); 3843 update_sit_entry(sbi, old_blkaddr, -1); 3844 3845 /* 3846 * If the current segment is full, flush it out and replace it with a 3847 * new segment. 3848 */ 3849 if (segment_full) { 3850 if (type == CURSEG_COLD_DATA_PINNED && 3851 !((curseg->segno + 1) % sbi->segs_per_sec)) { 3852 write_sum_page(sbi, curseg->sum_blk, curseg->segno); 3853 reset_curseg_fields(curseg); 3854 goto skip_new_segment; 3855 } 3856 3857 if (from_gc) { 3858 ret = get_atssr_segment(sbi, type, se->type, 3859 AT_SSR, se->mtime); 3860 } else { 3861 if (need_new_seg(sbi, type)) 3862 ret = new_curseg(sbi, type, false); 3863 else 3864 ret = change_curseg(sbi, type); 3865 stat_inc_seg_type(sbi, curseg); 3866 } 3867 3868 if (ret) 3869 goto out_err; 3870 } 3871 3872skip_new_segment: 3873 /* 3874 * segment dirty status should be updated after segment allocation, 3875 * so we just need to update status only one time after previous 3876 * segment being closed. 3877 */ 3878 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3879 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3880 3881 if (IS_DATASEG(curseg->seg_type)) { 3882 unsigned long long new_val; 3883 3884 new_val = atomic64_inc_return(&sbi->allocated_data_blocks); 3885 if (unlikely(new_val == ULLONG_MAX)) 3886 atomic64_set(&sbi->allocated_data_blocks, 0); 3887 } 3888 3889 up_write(&sit_i->sentry_lock); 3890 3891 if (folio && IS_NODESEG(curseg->seg_type)) { 3892 fill_node_footer_blkaddr(folio, NEXT_FREE_BLKADDR(sbi, curseg)); 3893 3894 f2fs_inode_chksum_set(sbi, folio); 3895 } 3896 3897 if (fio) { 3898 struct f2fs_bio_info *io; 3899 3900 INIT_LIST_HEAD(&fio->list); 3901 fio->in_list = 1; 3902 io = sbi->write_io[fio->type] + fio->temp; 3903 spin_lock(&io->io_lock); 3904 list_add_tail(&fio->list, &io->io_list); 3905 spin_unlock(&io->io_lock); 3906 } 3907 3908 mutex_unlock(&curseg->curseg_mutex); 3909 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3910 return 0; 3911 3912out_err: 3913 *new_blkaddr = NULL_ADDR; 3914 up_write(&sit_i->sentry_lock); 3915 mutex_unlock(&curseg->curseg_mutex); 3916 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3917 return ret; 3918} 3919 3920void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino, 3921 block_t blkaddr, unsigned int blkcnt) 3922{ 3923 if (!f2fs_is_multi_device(sbi)) 3924 return; 3925 3926 while (1) { 3927 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr); 3928 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1; 3929 3930 /* update device state for fsync */ 3931 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO); 3932 3933 /* update device state for checkpoint */ 3934 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3935 spin_lock(&sbi->dev_lock); 3936 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3937 spin_unlock(&sbi->dev_lock); 3938 } 3939 3940 if (blkcnt <= blks) 3941 break; 3942 blkcnt -= blks; 3943 blkaddr += blks; 3944 } 3945} 3946 3947static int log_type_to_seg_type(enum log_type type) 3948{ 3949 int seg_type = CURSEG_COLD_DATA; 3950 3951 switch (type) { 3952 case CURSEG_HOT_DATA: 3953 case CURSEG_WARM_DATA: 3954 case CURSEG_COLD_DATA: 3955 case CURSEG_HOT_NODE: 3956 case CURSEG_WARM_NODE: 3957 case CURSEG_COLD_NODE: 3958 seg_type = (int)type; 3959 break; 3960 case CURSEG_COLD_DATA_PINNED: 3961 case CURSEG_ALL_DATA_ATGC: 3962 seg_type = CURSEG_COLD_DATA; 3963 break; 3964 default: 3965 break; 3966 } 3967 return seg_type; 3968} 3969 3970static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3971{ 3972 struct folio *folio = fio->folio; 3973 enum log_type type = __get_segment_type(fio); 3974 int seg_type = log_type_to_seg_type(type); 3975 bool keep_order = (f2fs_lfs_mode(fio->sbi) && 3976 seg_type == CURSEG_COLD_DATA); 3977 int err; 3978 3979 if (keep_order) 3980 f2fs_down_read(&fio->sbi->io_order_lock); 3981 3982 err = f2fs_allocate_data_block(fio->sbi, folio, fio->old_blkaddr, 3983 &fio->new_blkaddr, sum, type, fio); 3984 if (unlikely(err)) { 3985 f2fs_err_ratelimited(fio->sbi, 3986 "%s Failed to allocate data block, ino:%u, index:%lu, type:%d, old_blkaddr:0x%x, new_blkaddr:0x%x, err:%d", 3987 __func__, fio->ino, folio->index, type, 3988 fio->old_blkaddr, fio->new_blkaddr, err); 3989 if (fscrypt_inode_uses_fs_layer_crypto(folio->mapping->host)) 3990 fscrypt_finalize_bounce_page(&fio->encrypted_page); 3991 folio_end_writeback(folio); 3992 if (f2fs_in_warm_node_list(fio->sbi, folio)) 3993 f2fs_del_fsync_node_entry(fio->sbi, folio); 3994 f2fs_bug_on(fio->sbi, !is_set_ckpt_flags(fio->sbi, 3995 CP_ERROR_FLAG)); 3996 goto out; 3997 } 3998 3999 f2fs_bug_on(fio->sbi, !f2fs_is_valid_blkaddr_raw(fio->sbi, 4000 fio->new_blkaddr, DATA_GENERIC_ENHANCE)); 4001 4002 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) 4003 f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr, 1); 4004 4005 /* writeout dirty page into bdev */ 4006 f2fs_submit_page_write(fio); 4007 4008 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1); 4009out: 4010 if (keep_order) 4011 f2fs_up_read(&fio->sbi->io_order_lock); 4012} 4013 4014void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct folio *folio, 4015 enum iostat_type io_type) 4016{ 4017 struct f2fs_io_info fio = { 4018 .sbi = sbi, 4019 .type = META, 4020 .temp = HOT, 4021 .op = REQ_OP_WRITE, 4022 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 4023 .old_blkaddr = folio->index, 4024 .new_blkaddr = folio->index, 4025 .folio = folio, 4026 .encrypted_page = NULL, 4027 .in_list = 0, 4028 }; 4029 4030 if (unlikely(folio->index >= MAIN_BLKADDR(sbi))) 4031 fio.op_flags &= ~REQ_META; 4032 4033 folio_start_writeback(folio); 4034 f2fs_submit_page_write(&fio); 4035 4036 stat_inc_meta_count(sbi, folio->index); 4037 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE); 4038} 4039 4040void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 4041{ 4042 struct f2fs_summary sum; 4043 4044 set_summary(&sum, nid, 0, 0); 4045 do_write_page(&sum, fio); 4046 4047 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE); 4048} 4049 4050void f2fs_outplace_write_data(struct dnode_of_data *dn, 4051 struct f2fs_io_info *fio) 4052{ 4053 struct f2fs_sb_info *sbi = fio->sbi; 4054 struct f2fs_summary sum; 4055 4056 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 4057 if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO) 4058 f2fs_update_age_extent_cache(dn); 4059 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 4060 do_write_page(&sum, fio); 4061 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 4062 4063 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE); 4064} 4065 4066int f2fs_inplace_write_data(struct f2fs_io_info *fio) 4067{ 4068 int err; 4069 struct f2fs_sb_info *sbi = fio->sbi; 4070 unsigned int segno; 4071 4072 fio->new_blkaddr = fio->old_blkaddr; 4073 /* i/o temperature is needed for passing down write hints */ 4074 __get_segment_type(fio); 4075 4076 segno = GET_SEGNO(sbi, fio->new_blkaddr); 4077 4078 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 4079 set_sbi_flag(sbi, SBI_NEED_FSCK); 4080 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 4081 __func__, segno); 4082 err = -EFSCORRUPTED; 4083 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 4084 goto drop_bio; 4085 } 4086 4087 if (f2fs_cp_error(sbi)) { 4088 err = -EIO; 4089 goto drop_bio; 4090 } 4091 4092 if (fio->meta_gc) 4093 f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1); 4094 4095 stat_inc_inplace_blocks(fio->sbi); 4096 4097 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi)) 4098 err = f2fs_merge_page_bio(fio); 4099 else 4100 err = f2fs_submit_page_bio(fio); 4101 if (!err) { 4102 f2fs_update_device_state(fio->sbi, fio->ino, 4103 fio->new_blkaddr, 1); 4104 f2fs_update_iostat(fio->sbi, fio_inode(fio), 4105 fio->io_type, F2FS_BLKSIZE); 4106 } 4107 4108 return err; 4109drop_bio: 4110 if (fio->bio && *(fio->bio)) { 4111 struct bio *bio = *(fio->bio); 4112 4113 bio->bi_status = BLK_STS_IOERR; 4114 bio_endio(bio); 4115 *(fio->bio) = NULL; 4116 } 4117 return err; 4118} 4119 4120static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 4121 unsigned int segno) 4122{ 4123 int i; 4124 4125 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 4126 if (CURSEG_I(sbi, i)->segno == segno) 4127 break; 4128 } 4129 return i; 4130} 4131 4132void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 4133 block_t old_blkaddr, block_t new_blkaddr, 4134 bool recover_curseg, bool recover_newaddr, 4135 bool from_gc) 4136{ 4137 struct sit_info *sit_i = SIT_I(sbi); 4138 struct curseg_info *curseg; 4139 unsigned int segno, old_cursegno; 4140 struct seg_entry *se; 4141 int type; 4142 unsigned short old_blkoff; 4143 unsigned char old_alloc_type; 4144 4145 segno = GET_SEGNO(sbi, new_blkaddr); 4146 se = get_seg_entry(sbi, segno); 4147 type = se->type; 4148 4149 f2fs_down_write(&SM_I(sbi)->curseg_lock); 4150 4151 if (!recover_curseg) { 4152 /* for recovery flow */ 4153 if (se->valid_blocks == 0 && !is_curseg(sbi, segno)) { 4154 if (old_blkaddr == NULL_ADDR) 4155 type = CURSEG_COLD_DATA; 4156 else 4157 type = CURSEG_WARM_DATA; 4158 } 4159 } else { 4160 if (is_curseg(sbi, segno)) { 4161 /* se->type is volatile as SSR allocation */ 4162 type = __f2fs_get_curseg(sbi, segno); 4163 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 4164 } else { 4165 type = CURSEG_WARM_DATA; 4166 } 4167 } 4168 4169 curseg = CURSEG_I(sbi, type); 4170 f2fs_bug_on(sbi, !IS_DATASEG(curseg->seg_type)); 4171 4172 mutex_lock(&curseg->curseg_mutex); 4173 down_write(&sit_i->sentry_lock); 4174 4175 old_cursegno = curseg->segno; 4176 old_blkoff = curseg->next_blkoff; 4177 old_alloc_type = curseg->alloc_type; 4178 4179 /* change the current segment */ 4180 if (segno != curseg->segno) { 4181 curseg->next_segno = segno; 4182 if (change_curseg(sbi, type)) 4183 goto out_unlock; 4184 } 4185 4186 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 4187 sum_entries(curseg->sum_blk)[curseg->next_blkoff] = *sum; 4188 4189 if (!recover_curseg || recover_newaddr) { 4190 if (!from_gc) 4191 update_segment_mtime(sbi, new_blkaddr, 0); 4192 update_sit_entry(sbi, new_blkaddr, 1); 4193 } 4194 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 4195 f2fs_invalidate_internal_cache(sbi, old_blkaddr, 1); 4196 if (!from_gc) 4197 update_segment_mtime(sbi, old_blkaddr, 0); 4198 update_sit_entry(sbi, old_blkaddr, -1); 4199 } 4200 4201 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 4202 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 4203 4204 locate_dirty_segment(sbi, old_cursegno); 4205 4206 if (recover_curseg) { 4207 if (old_cursegno != curseg->segno) { 4208 curseg->next_segno = old_cursegno; 4209 if (change_curseg(sbi, type)) 4210 goto out_unlock; 4211 } 4212 curseg->next_blkoff = old_blkoff; 4213 curseg->alloc_type = old_alloc_type; 4214 } 4215 4216out_unlock: 4217 up_write(&sit_i->sentry_lock); 4218 mutex_unlock(&curseg->curseg_mutex); 4219 f2fs_up_write(&SM_I(sbi)->curseg_lock); 4220} 4221 4222void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 4223 block_t old_addr, block_t new_addr, 4224 unsigned char version, bool recover_curseg, 4225 bool recover_newaddr) 4226{ 4227 struct f2fs_summary sum; 4228 4229 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 4230 4231 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 4232 recover_curseg, recover_newaddr, false); 4233 4234 f2fs_update_data_blkaddr(dn, new_addr); 4235} 4236 4237void f2fs_folio_wait_writeback(struct folio *folio, enum page_type type, 4238 bool ordered, bool locked) 4239{ 4240 if (folio_test_writeback(folio)) { 4241 struct f2fs_sb_info *sbi = F2FS_F_SB(folio); 4242 4243 /* submit cached LFS IO */ 4244 f2fs_submit_merged_write_cond(sbi, NULL, folio, 0, type); 4245 /* submit cached IPU IO */ 4246 f2fs_submit_merged_ipu_write(sbi, NULL, folio); 4247 if (ordered) { 4248 folio_wait_writeback(folio); 4249 f2fs_bug_on(sbi, locked && folio_test_writeback(folio)); 4250 } else { 4251 folio_wait_stable(folio); 4252 } 4253 } 4254} 4255 4256void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 4257{ 4258 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4259 struct folio *cfolio; 4260 4261 if (!f2fs_meta_inode_gc_required(inode)) 4262 return; 4263 4264 if (!__is_valid_data_blkaddr(blkaddr)) 4265 return; 4266 4267 cfolio = filemap_lock_folio(META_MAPPING(sbi), blkaddr); 4268 if (!IS_ERR(cfolio)) { 4269 f2fs_folio_wait_writeback(cfolio, DATA, true, true); 4270 f2fs_folio_put(cfolio, true); 4271 } 4272} 4273 4274void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 4275 block_t len) 4276{ 4277 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4278 block_t i; 4279 4280 if (!f2fs_meta_inode_gc_required(inode)) 4281 return; 4282 4283 for (i = 0; i < len; i++) 4284 f2fs_wait_on_block_writeback(inode, blkaddr + i); 4285 4286 f2fs_truncate_meta_inode_pages(sbi, blkaddr, len); 4287} 4288 4289static int read_compacted_summaries(struct f2fs_sb_info *sbi) 4290{ 4291 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 4292 struct curseg_info *seg_i; 4293 unsigned char *kaddr; 4294 struct folio *folio; 4295 block_t start; 4296 int i, j, offset; 4297 4298 start = start_sum_block(sbi); 4299 4300 folio = f2fs_get_meta_folio(sbi, start++); 4301 if (IS_ERR(folio)) 4302 return PTR_ERR(folio); 4303 kaddr = folio_address(folio); 4304 4305 /* Step 1: restore nat cache */ 4306 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4307 memcpy(seg_i->journal, kaddr, sbi->sum_journal_size); 4308 4309 /* Step 2: restore sit cache */ 4310 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4311 memcpy(seg_i->journal, kaddr + sbi->sum_journal_size, sbi->sum_journal_size); 4312 offset = 2 * sbi->sum_journal_size; 4313 4314 /* Step 3: restore summary entries */ 4315 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4316 unsigned short blk_off; 4317 unsigned int segno; 4318 4319 seg_i = CURSEG_I(sbi, i); 4320 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 4321 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 4322 seg_i->next_segno = segno; 4323 reset_curseg(sbi, i, 0); 4324 seg_i->alloc_type = ckpt->alloc_type[i]; 4325 seg_i->next_blkoff = blk_off; 4326 4327 if (seg_i->alloc_type == SSR) 4328 blk_off = BLKS_PER_SEG(sbi); 4329 4330 for (j = 0; j < blk_off; j++) { 4331 struct f2fs_summary *s; 4332 4333 s = (struct f2fs_summary *)(kaddr + offset); 4334 sum_entries(seg_i->sum_blk)[j] = *s; 4335 offset += SUMMARY_SIZE; 4336 if (offset + SUMMARY_SIZE <= sbi->blocksize - 4337 SUM_FOOTER_SIZE) 4338 continue; 4339 4340 f2fs_folio_put(folio, true); 4341 4342 folio = f2fs_get_meta_folio(sbi, start++); 4343 if (IS_ERR(folio)) 4344 return PTR_ERR(folio); 4345 kaddr = folio_address(folio); 4346 offset = 0; 4347 } 4348 } 4349 f2fs_folio_put(folio, true); 4350 return 0; 4351} 4352 4353static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 4354{ 4355 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 4356 struct f2fs_summary_block *sum; 4357 struct curseg_info *curseg; 4358 struct folio *new; 4359 unsigned short blk_off; 4360 unsigned int segno = 0; 4361 block_t blk_addr = 0; 4362 int err = 0; 4363 4364 /* get segment number and block addr */ 4365 if (IS_DATASEG(type)) { 4366 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 4367 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 4368 CURSEG_HOT_DATA]); 4369 if (__exist_node_summaries(sbi)) 4370 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type); 4371 else 4372 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 4373 } else { 4374 segno = le32_to_cpu(ckpt->cur_node_segno[type - 4375 CURSEG_HOT_NODE]); 4376 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 4377 CURSEG_HOT_NODE]); 4378 if (__exist_node_summaries(sbi)) 4379 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 4380 type - CURSEG_HOT_NODE); 4381 else 4382 blk_addr = GET_SUM_BLOCK(sbi, segno); 4383 } 4384 4385 new = f2fs_get_meta_folio(sbi, blk_addr); 4386 if (IS_ERR(new)) 4387 return PTR_ERR(new); 4388 sum = folio_address(new); 4389 4390 if (IS_NODESEG(type)) { 4391 if (__exist_node_summaries(sbi)) { 4392 struct f2fs_summary *ns = sum_entries(sum); 4393 int i; 4394 4395 for (i = 0; i < BLKS_PER_SEG(sbi); i++, ns++) { 4396 ns->version = 0; 4397 ns->ofs_in_node = 0; 4398 } 4399 } else { 4400 err = f2fs_restore_node_summary(sbi, segno, sum); 4401 if (err) 4402 goto out; 4403 } 4404 } 4405 4406 /* set uncompleted segment to curseg */ 4407 curseg = CURSEG_I(sbi, type); 4408 mutex_lock(&curseg->curseg_mutex); 4409 4410 /* update journal info */ 4411 down_write(&curseg->journal_rwsem); 4412 memcpy(curseg->journal, sum_journal(sbi, sum), sbi->sum_journal_size); 4413 up_write(&curseg->journal_rwsem); 4414 4415 memcpy(sum_entries(curseg->sum_blk), sum_entries(sum), 4416 sbi->sum_entry_size); 4417 memcpy(sum_footer(sbi, curseg->sum_blk), sum_footer(sbi, sum), 4418 SUM_FOOTER_SIZE); 4419 curseg->next_segno = segno; 4420 reset_curseg(sbi, type, 0); 4421 curseg->alloc_type = ckpt->alloc_type[type]; 4422 curseg->next_blkoff = blk_off; 4423 mutex_unlock(&curseg->curseg_mutex); 4424out: 4425 f2fs_folio_put(new, true); 4426 return err; 4427} 4428 4429static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 4430{ 4431 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 4432 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 4433 int type = CURSEG_HOT_DATA; 4434 int err; 4435 4436 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 4437 int npages = f2fs_npages_for_summary_flush(sbi, true); 4438 4439 if (npages >= 2) 4440 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 4441 META_CP, true); 4442 4443 /* restore for compacted data summary */ 4444 err = read_compacted_summaries(sbi); 4445 if (err) 4446 return err; 4447 type = CURSEG_HOT_NODE; 4448 } 4449 4450 if (__exist_node_summaries(sbi)) 4451 f2fs_ra_meta_pages(sbi, 4452 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type), 4453 NR_CURSEG_PERSIST_TYPE - type, META_CP, true); 4454 4455 for (; type <= CURSEG_COLD_NODE; type++) { 4456 err = read_normal_summaries(sbi, type); 4457 if (err) 4458 return err; 4459 } 4460 4461 /* sanity check for summary blocks */ 4462 if (nats_in_cursum(nat_j) > sbi->nat_journal_entries || 4463 sits_in_cursum(sit_j) > sbi->sit_journal_entries) { 4464 f2fs_err(sbi, "invalid journal entries nats %u sits %u", 4465 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 4466 return -EINVAL; 4467 } 4468 4469 return 0; 4470} 4471 4472static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 4473{ 4474 struct folio *folio; 4475 unsigned char *kaddr; 4476 struct f2fs_summary *summary; 4477 struct curseg_info *seg_i; 4478 int written_size = 0; 4479 int i, j; 4480 4481 folio = f2fs_grab_meta_folio(sbi, blkaddr++); 4482 kaddr = folio_address(folio); 4483 memset(kaddr, 0, PAGE_SIZE); 4484 4485 /* Step 1: write nat cache */ 4486 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4487 memcpy(kaddr, seg_i->journal, sbi->sum_journal_size); 4488 written_size += sbi->sum_journal_size; 4489 4490 /* Step 2: write sit cache */ 4491 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4492 memcpy(kaddr + written_size, seg_i->journal, sbi->sum_journal_size); 4493 written_size += sbi->sum_journal_size; 4494 4495 /* Step 3: write summary entries */ 4496 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4497 seg_i = CURSEG_I(sbi, i); 4498 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) { 4499 if (!folio) { 4500 folio = f2fs_grab_meta_folio(sbi, blkaddr++); 4501 kaddr = folio_address(folio); 4502 memset(kaddr, 0, PAGE_SIZE); 4503 written_size = 0; 4504 } 4505 summary = (struct f2fs_summary *)(kaddr + written_size); 4506 *summary = sum_entries(seg_i->sum_blk)[j]; 4507 written_size += SUMMARY_SIZE; 4508 4509 if (written_size + SUMMARY_SIZE <= sbi->blocksize - 4510 SUM_FOOTER_SIZE) 4511 continue; 4512 4513 folio_mark_dirty(folio); 4514 f2fs_folio_put(folio, true); 4515 folio = NULL; 4516 } 4517 } 4518 if (folio) { 4519 folio_mark_dirty(folio); 4520 f2fs_folio_put(folio, true); 4521 } 4522} 4523 4524static void write_normal_summaries(struct f2fs_sb_info *sbi, 4525 block_t blkaddr, int type) 4526{ 4527 int i, end; 4528 4529 if (IS_DATASEG(type)) 4530 end = type + NR_CURSEG_DATA_TYPE; 4531 else 4532 end = type + NR_CURSEG_NODE_TYPE; 4533 4534 for (i = type; i < end; i++) 4535 write_current_sum_page(sbi, i, blkaddr + (i - type)); 4536} 4537 4538void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4539{ 4540 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 4541 write_compacted_summaries(sbi, start_blk); 4542 else 4543 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 4544} 4545 4546void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4547{ 4548 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 4549} 4550 4551int f2fs_lookup_journal_in_cursum(struct f2fs_sb_info *sbi, 4552 struct f2fs_journal *journal, int type, 4553 unsigned int val, int alloc) 4554{ 4555 int i; 4556 4557 if (type == NAT_JOURNAL) { 4558 for (i = 0; i < nats_in_cursum(journal); i++) { 4559 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 4560 return i; 4561 } 4562 if (alloc && __has_cursum_space(sbi, journal, 1, NAT_JOURNAL)) 4563 return update_nats_in_cursum(journal, 1); 4564 } else if (type == SIT_JOURNAL) { 4565 for (i = 0; i < sits_in_cursum(journal); i++) 4566 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 4567 return i; 4568 if (alloc && __has_cursum_space(sbi, journal, 1, SIT_JOURNAL)) 4569 return update_sits_in_cursum(journal, 1); 4570 } 4571 return -1; 4572} 4573 4574static struct folio *get_current_sit_folio(struct f2fs_sb_info *sbi, 4575 unsigned int segno) 4576{ 4577 return f2fs_get_meta_folio(sbi, current_sit_addr(sbi, segno)); 4578} 4579 4580static struct folio *get_next_sit_folio(struct f2fs_sb_info *sbi, 4581 unsigned int start) 4582{ 4583 struct sit_info *sit_i = SIT_I(sbi); 4584 struct folio *folio; 4585 pgoff_t src_off, dst_off; 4586 4587 src_off = current_sit_addr(sbi, start); 4588 dst_off = next_sit_addr(sbi, src_off); 4589 4590 folio = f2fs_grab_meta_folio(sbi, dst_off); 4591 seg_info_to_sit_folio(sbi, folio, start); 4592 4593 folio_mark_dirty(folio); 4594 set_to_next_sit(sit_i, start); 4595 4596 return folio; 4597} 4598 4599static struct sit_entry_set *grab_sit_entry_set(void) 4600{ 4601 struct sit_entry_set *ses = 4602 f2fs_kmem_cache_alloc(sit_entry_set_slab, 4603 GFP_NOFS, true, NULL); 4604 4605 ses->entry_cnt = 0; 4606 INIT_LIST_HEAD(&ses->set_list); 4607 return ses; 4608} 4609 4610static void release_sit_entry_set(struct sit_entry_set *ses) 4611{ 4612 list_del(&ses->set_list); 4613 kmem_cache_free(sit_entry_set_slab, ses); 4614} 4615 4616static void adjust_sit_entry_set(struct sit_entry_set *ses, 4617 struct list_head *head) 4618{ 4619 struct sit_entry_set *next = ses; 4620 4621 if (list_is_last(&ses->set_list, head)) 4622 return; 4623 4624 list_for_each_entry_continue(next, head, set_list) 4625 if (ses->entry_cnt <= next->entry_cnt) { 4626 list_move_tail(&ses->set_list, &next->set_list); 4627 return; 4628 } 4629 4630 list_move_tail(&ses->set_list, head); 4631} 4632 4633static void add_sit_entry(unsigned int segno, struct list_head *head) 4634{ 4635 struct sit_entry_set *ses; 4636 unsigned int start_segno = START_SEGNO(segno); 4637 4638 list_for_each_entry(ses, head, set_list) { 4639 if (ses->start_segno == start_segno) { 4640 ses->entry_cnt++; 4641 adjust_sit_entry_set(ses, head); 4642 return; 4643 } 4644 } 4645 4646 ses = grab_sit_entry_set(); 4647 4648 ses->start_segno = start_segno; 4649 ses->entry_cnt++; 4650 list_add(&ses->set_list, head); 4651} 4652 4653static void add_sits_in_set(struct f2fs_sb_info *sbi) 4654{ 4655 struct f2fs_sm_info *sm_info = SM_I(sbi); 4656 struct list_head *set_list = &sm_info->sit_entry_set; 4657 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 4658 unsigned int segno; 4659 4660 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 4661 add_sit_entry(segno, set_list); 4662} 4663 4664static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 4665{ 4666 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4667 struct f2fs_journal *journal = curseg->journal; 4668 int i; 4669 4670 down_write(&curseg->journal_rwsem); 4671 for (i = 0; i < sits_in_cursum(journal); i++) { 4672 unsigned int segno; 4673 bool dirtied; 4674 4675 segno = le32_to_cpu(segno_in_journal(journal, i)); 4676 dirtied = __mark_sit_entry_dirty(sbi, segno); 4677 4678 if (!dirtied) 4679 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 4680 } 4681 update_sits_in_cursum(journal, -i); 4682 up_write(&curseg->journal_rwsem); 4683} 4684 4685/* 4686 * CP calls this function, which flushes SIT entries including sit_journal, 4687 * and moves prefree segs to free segs. 4688 */ 4689void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 4690{ 4691 struct sit_info *sit_i = SIT_I(sbi); 4692 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 4693 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4694 struct f2fs_journal *journal = curseg->journal; 4695 struct sit_entry_set *ses, *tmp; 4696 struct list_head *head = &SM_I(sbi)->sit_entry_set; 4697 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 4698 struct seg_entry *se; 4699 4700 down_write(&sit_i->sentry_lock); 4701 4702 if (!sit_i->dirty_sentries) 4703 goto out; 4704 4705 /* 4706 * add and account sit entries of dirty bitmap in sit entry 4707 * set temporarily 4708 */ 4709 add_sits_in_set(sbi); 4710 4711 /* 4712 * if there are no enough space in journal to store dirty sit 4713 * entries, remove all entries from journal and add and account 4714 * them in sit entry set. 4715 */ 4716 if (!__has_cursum_space(sbi, journal, 4717 sit_i->dirty_sentries, SIT_JOURNAL) || !to_journal) 4718 remove_sits_in_journal(sbi); 4719 4720 /* 4721 * there are two steps to flush sit entries: 4722 * #1, flush sit entries to journal in current cold data summary block. 4723 * #2, flush sit entries to sit page. 4724 */ 4725 list_for_each_entry_safe(ses, tmp, head, set_list) { 4726 struct folio *folio = NULL; 4727 struct f2fs_sit_block *raw_sit = NULL; 4728 unsigned int start_segno = ses->start_segno; 4729 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 4730 (unsigned long)MAIN_SEGS(sbi)); 4731 unsigned int segno = start_segno; 4732 4733 if (to_journal && 4734 !__has_cursum_space(sbi, journal, ses->entry_cnt, 4735 SIT_JOURNAL)) 4736 to_journal = false; 4737 4738 if (to_journal) { 4739 down_write(&curseg->journal_rwsem); 4740 } else { 4741 folio = get_next_sit_folio(sbi, start_segno); 4742 raw_sit = folio_address(folio); 4743 } 4744 4745 /* flush dirty sit entries in region of current sit set */ 4746 for_each_set_bit_from(segno, bitmap, end) { 4747 int offset, sit_offset; 4748 4749 se = get_seg_entry(sbi, segno); 4750#ifdef CONFIG_F2FS_CHECK_FS 4751 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 4752 SIT_VBLOCK_MAP_SIZE)) 4753 f2fs_bug_on(sbi, 1); 4754#endif 4755 4756 /* add discard candidates */ 4757 if (!(cpc->reason & CP_DISCARD)) { 4758 cpc->trim_start = segno; 4759 add_discard_addrs(sbi, cpc, false); 4760 } 4761 4762 if (to_journal) { 4763 offset = f2fs_lookup_journal_in_cursum(sbi, journal, 4764 SIT_JOURNAL, segno, 1); 4765 f2fs_bug_on(sbi, offset < 0); 4766 segno_in_journal(journal, offset) = 4767 cpu_to_le32(segno); 4768 seg_info_to_raw_sit(se, 4769 &sit_in_journal(journal, offset)); 4770 check_block_count(sbi, segno, 4771 &sit_in_journal(journal, offset)); 4772 } else { 4773 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 4774 seg_info_to_raw_sit(se, 4775 &raw_sit->entries[sit_offset]); 4776 check_block_count(sbi, segno, 4777 &raw_sit->entries[sit_offset]); 4778 } 4779 4780 /* update ckpt_valid_block */ 4781 if (__is_large_section(sbi)) { 4782 set_ckpt_valid_blocks(sbi, segno); 4783 sanity_check_valid_blocks(sbi, segno); 4784 } 4785 4786 __clear_bit(segno, bitmap); 4787 sit_i->dirty_sentries--; 4788 ses->entry_cnt--; 4789 } 4790 4791 if (to_journal) 4792 up_write(&curseg->journal_rwsem); 4793 else 4794 f2fs_folio_put(folio, true); 4795 4796 f2fs_bug_on(sbi, ses->entry_cnt); 4797 release_sit_entry_set(ses); 4798 } 4799 4800 f2fs_bug_on(sbi, !list_empty(head)); 4801 f2fs_bug_on(sbi, sit_i->dirty_sentries); 4802out: 4803 if (cpc->reason & CP_DISCARD) { 4804 __u64 trim_start = cpc->trim_start; 4805 4806 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 4807 add_discard_addrs(sbi, cpc, false); 4808 4809 cpc->trim_start = trim_start; 4810 } 4811 up_write(&sit_i->sentry_lock); 4812 4813 set_prefree_as_free_segments(sbi); 4814} 4815 4816static int build_sit_info(struct f2fs_sb_info *sbi) 4817{ 4818 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4819 struct sit_info *sit_i; 4820 unsigned int sit_segs, start; 4821 char *src_bitmap, *bitmap; 4822 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 4823 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0; 4824 4825 /* allocate memory for SIT information */ 4826 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4827 if (!sit_i) 4828 return -ENOMEM; 4829 4830 SM_I(sbi)->sit_info = sit_i; 4831 4832 sit_i->sentries = 4833 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4834 MAIN_SEGS(sbi)), 4835 GFP_KERNEL); 4836 if (!sit_i->sentries) 4837 return -ENOMEM; 4838 4839 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4840 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4841 GFP_KERNEL); 4842 if (!sit_i->dirty_sentries_bitmap) 4843 return -ENOMEM; 4844 4845#ifdef CONFIG_F2FS_CHECK_FS 4846 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map); 4847#else 4848 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map); 4849#endif 4850 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4851 if (!sit_i->bitmap) 4852 return -ENOMEM; 4853 4854 bitmap = sit_i->bitmap; 4855 4856 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4857 sit_i->sentries[start].cur_valid_map = bitmap; 4858 bitmap += SIT_VBLOCK_MAP_SIZE; 4859 4860 sit_i->sentries[start].ckpt_valid_map = bitmap; 4861 bitmap += SIT_VBLOCK_MAP_SIZE; 4862 4863#ifdef CONFIG_F2FS_CHECK_FS 4864 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4865 bitmap += SIT_VBLOCK_MAP_SIZE; 4866#endif 4867 4868 if (discard_map) { 4869 sit_i->sentries[start].discard_map = bitmap; 4870 bitmap += SIT_VBLOCK_MAP_SIZE; 4871 } 4872 } 4873 4874 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4875 if (!sit_i->tmp_map) 4876 return -ENOMEM; 4877 4878 if (__is_large_section(sbi)) { 4879 sit_i->sec_entries = 4880 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4881 MAIN_SECS(sbi)), 4882 GFP_KERNEL); 4883 if (!sit_i->sec_entries) 4884 return -ENOMEM; 4885 } 4886 4887 /* get information related with SIT */ 4888 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4889 4890 /* setup SIT bitmap from ckeckpoint pack */ 4891 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4892 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4893 4894 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4895 if (!sit_i->sit_bitmap) 4896 return -ENOMEM; 4897 4898#ifdef CONFIG_F2FS_CHECK_FS 4899 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4900 sit_bitmap_size, GFP_KERNEL); 4901 if (!sit_i->sit_bitmap_mir) 4902 return -ENOMEM; 4903 4904 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4905 main_bitmap_size, GFP_KERNEL); 4906 if (!sit_i->invalid_segmap) 4907 return -ENOMEM; 4908#endif 4909 4910 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4911 sit_i->sit_blocks = SEGS_TO_BLKS(sbi, sit_segs); 4912 sit_i->written_valid_blocks = 0; 4913 sit_i->bitmap_size = sit_bitmap_size; 4914 sit_i->dirty_sentries = 0; 4915 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4916 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4917 sit_i->mounted_time = ktime_get_boottime_seconds(); 4918 init_rwsem(&sit_i->sentry_lock); 4919 return 0; 4920} 4921 4922static int build_free_segmap(struct f2fs_sb_info *sbi) 4923{ 4924 struct free_segmap_info *free_i; 4925 unsigned int bitmap_size, sec_bitmap_size; 4926 4927 /* allocate memory for free segmap information */ 4928 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4929 if (!free_i) 4930 return -ENOMEM; 4931 4932 SM_I(sbi)->free_info = free_i; 4933 4934 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4935 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4936 if (!free_i->free_segmap) 4937 return -ENOMEM; 4938 4939 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4940 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4941 if (!free_i->free_secmap) 4942 return -ENOMEM; 4943 4944 /* set all segments as dirty temporarily */ 4945 memset(free_i->free_segmap, 0xff, bitmap_size); 4946 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4947 4948 /* init free segmap information */ 4949 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4950 free_i->free_segments = 0; 4951 free_i->free_sections = 0; 4952 spin_lock_init(&free_i->segmap_lock); 4953 return 0; 4954} 4955 4956static int build_curseg(struct f2fs_sb_info *sbi) 4957{ 4958 struct curseg_info *array; 4959 int i; 4960 4961 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, 4962 sizeof(*array)), GFP_KERNEL); 4963 if (!array) 4964 return -ENOMEM; 4965 4966 SM_I(sbi)->curseg_array = array; 4967 4968 for (i = 0; i < NO_CHECK_TYPE; i++) { 4969 mutex_init(&array[i].curseg_mutex); 4970 array[i].sum_blk = f2fs_kzalloc(sbi, sbi->sum_blocksize, 4971 GFP_KERNEL); 4972 if (!array[i].sum_blk) 4973 return -ENOMEM; 4974 init_rwsem(&array[i].journal_rwsem); 4975 array[i].journal = f2fs_kzalloc(sbi, 4976 sbi->sum_journal_size, GFP_KERNEL); 4977 if (!array[i].journal) 4978 return -ENOMEM; 4979 array[i].seg_type = log_type_to_seg_type(i); 4980 reset_curseg_fields(&array[i]); 4981 } 4982 return restore_curseg_summaries(sbi); 4983} 4984 4985static int build_sit_entries(struct f2fs_sb_info *sbi) 4986{ 4987 struct sit_info *sit_i = SIT_I(sbi); 4988 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4989 struct f2fs_journal *journal = curseg->journal; 4990 struct seg_entry *se; 4991 struct f2fs_sit_entry sit; 4992 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4993 unsigned int i, start, end; 4994 unsigned int readed, start_blk = 0; 4995 int err = 0; 4996 block_t sit_valid_blocks[2] = {0, 0}; 4997 4998 do { 4999 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS, 5000 META_SIT, true); 5001 5002 start = start_blk * sit_i->sents_per_block; 5003 end = (start_blk + readed) * sit_i->sents_per_block; 5004 5005 for (; start < end && start < MAIN_SEGS(sbi); start++) { 5006 struct f2fs_sit_block *sit_blk; 5007 struct folio *folio; 5008 5009 se = &sit_i->sentries[start]; 5010 folio = get_current_sit_folio(sbi, start); 5011 if (IS_ERR(folio)) 5012 return PTR_ERR(folio); 5013 sit_blk = folio_address(folio); 5014 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 5015 f2fs_folio_put(folio, true); 5016 5017 err = check_block_count(sbi, start, &sit); 5018 if (err) 5019 return err; 5020 seg_info_from_raw_sit(se, &sit); 5021 5022 if (se->type >= NR_PERSISTENT_LOG) { 5023 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 5024 se->type, start); 5025 f2fs_handle_error(sbi, 5026 ERROR_INCONSISTENT_SUM_TYPE); 5027 return -EFSCORRUPTED; 5028 } 5029 5030 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 5031 5032 if (!f2fs_block_unit_discard(sbi)) 5033 goto init_discard_map_done; 5034 5035 /* build discard map only one time */ 5036 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 5037 memset(se->discard_map, 0xff, 5038 SIT_VBLOCK_MAP_SIZE); 5039 goto init_discard_map_done; 5040 } 5041 memcpy(se->discard_map, se->cur_valid_map, 5042 SIT_VBLOCK_MAP_SIZE); 5043 sbi->discard_blks += BLKS_PER_SEG(sbi) - 5044 se->valid_blocks; 5045init_discard_map_done: 5046 if (__is_large_section(sbi)) 5047 get_sec_entry(sbi, start)->valid_blocks += 5048 se->valid_blocks; 5049 } 5050 start_blk += readed; 5051 } while (start_blk < sit_blk_cnt); 5052 5053 down_read(&curseg->journal_rwsem); 5054 for (i = 0; i < sits_in_cursum(journal); i++) { 5055 unsigned int old_valid_blocks; 5056 5057 start = le32_to_cpu(segno_in_journal(journal, i)); 5058 if (start >= MAIN_SEGS(sbi)) { 5059 f2fs_err(sbi, "Wrong journal entry on segno %u", 5060 start); 5061 err = -EFSCORRUPTED; 5062 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL); 5063 break; 5064 } 5065 5066 se = &sit_i->sentries[start]; 5067 sit = sit_in_journal(journal, i); 5068 5069 old_valid_blocks = se->valid_blocks; 5070 5071 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks; 5072 5073 err = check_block_count(sbi, start, &sit); 5074 if (err) 5075 break; 5076 seg_info_from_raw_sit(se, &sit); 5077 5078 if (se->type >= NR_PERSISTENT_LOG) { 5079 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 5080 se->type, start); 5081 err = -EFSCORRUPTED; 5082 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 5083 break; 5084 } 5085 5086 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 5087 5088 if (f2fs_block_unit_discard(sbi)) { 5089 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 5090 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 5091 } else { 5092 memcpy(se->discard_map, se->cur_valid_map, 5093 SIT_VBLOCK_MAP_SIZE); 5094 sbi->discard_blks += old_valid_blocks; 5095 sbi->discard_blks -= se->valid_blocks; 5096 } 5097 } 5098 5099 if (__is_large_section(sbi)) { 5100 get_sec_entry(sbi, start)->valid_blocks += 5101 se->valid_blocks; 5102 get_sec_entry(sbi, start)->valid_blocks -= 5103 old_valid_blocks; 5104 } 5105 } 5106 up_read(&curseg->journal_rwsem); 5107 5108 /* update ckpt_valid_block */ 5109 if (__is_large_section(sbi)) { 5110 unsigned int segno; 5111 5112 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) { 5113 set_ckpt_valid_blocks(sbi, segno); 5114 sanity_check_valid_blocks(sbi, segno); 5115 } 5116 } 5117 5118 if (err) 5119 return err; 5120 5121 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) { 5122 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 5123 sit_valid_blocks[NODE], valid_node_count(sbi)); 5124 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT); 5125 return -EFSCORRUPTED; 5126 } 5127 5128 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] > 5129 valid_user_blocks(sbi)) { 5130 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u", 5131 sit_valid_blocks[DATA], sit_valid_blocks[NODE], 5132 valid_user_blocks(sbi)); 5133 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT); 5134 return -EFSCORRUPTED; 5135 } 5136 5137 return 0; 5138} 5139 5140static void init_free_segmap(struct f2fs_sb_info *sbi) 5141{ 5142 unsigned int start; 5143 int type; 5144 struct seg_entry *sentry; 5145 5146 for (start = 0; start < MAIN_SEGS(sbi); start++) { 5147 if (f2fs_usable_blks_in_seg(sbi, start) == 0) 5148 continue; 5149 sentry = get_seg_entry(sbi, start); 5150 if (!sentry->valid_blocks) 5151 __set_free(sbi, start); 5152 else 5153 SIT_I(sbi)->written_valid_blocks += 5154 sentry->valid_blocks; 5155 } 5156 5157 /* set use the current segments */ 5158 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 5159 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 5160 5161 __set_test_and_inuse(sbi, curseg_t->segno); 5162 } 5163} 5164 5165static void init_dirty_segmap(struct f2fs_sb_info *sbi) 5166{ 5167 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5168 struct free_segmap_info *free_i = FREE_I(sbi); 5169 unsigned int segno = 0, offset = 0, secno; 5170 block_t valid_blocks, usable_blks_in_seg; 5171 5172 while (1) { 5173 /* find dirty segment based on free segmap */ 5174 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 5175 if (segno >= MAIN_SEGS(sbi)) 5176 break; 5177 offset = segno + 1; 5178 valid_blocks = get_valid_blocks(sbi, segno, false); 5179 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 5180 if (valid_blocks == usable_blks_in_seg || !valid_blocks) 5181 continue; 5182 if (valid_blocks > usable_blks_in_seg) { 5183 f2fs_bug_on(sbi, 1); 5184 continue; 5185 } 5186 mutex_lock(&dirty_i->seglist_lock); 5187 __locate_dirty_segment(sbi, segno, DIRTY); 5188 mutex_unlock(&dirty_i->seglist_lock); 5189 } 5190 5191 if (!__is_large_section(sbi)) 5192 return; 5193 5194 mutex_lock(&dirty_i->seglist_lock); 5195 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) { 5196 valid_blocks = get_valid_blocks(sbi, segno, true); 5197 secno = GET_SEC_FROM_SEG(sbi, segno); 5198 5199 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi)) 5200 continue; 5201 if (is_cursec(sbi, secno)) 5202 continue; 5203 set_bit(secno, dirty_i->dirty_secmap); 5204 } 5205 mutex_unlock(&dirty_i->seglist_lock); 5206} 5207 5208static int init_victim_secmap(struct f2fs_sb_info *sbi) 5209{ 5210 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5211 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 5212 5213 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 5214 if (!dirty_i->victim_secmap) 5215 return -ENOMEM; 5216 5217 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 5218 if (!dirty_i->pinned_secmap) 5219 return -ENOMEM; 5220 5221 dirty_i->pinned_secmap_cnt = 0; 5222 dirty_i->enable_pin_section = true; 5223 return 0; 5224} 5225 5226static int build_dirty_segmap(struct f2fs_sb_info *sbi) 5227{ 5228 struct dirty_seglist_info *dirty_i; 5229 unsigned int bitmap_size, i; 5230 5231 /* allocate memory for dirty segments list information */ 5232 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 5233 GFP_KERNEL); 5234 if (!dirty_i) 5235 return -ENOMEM; 5236 5237 SM_I(sbi)->dirty_info = dirty_i; 5238 mutex_init(&dirty_i->seglist_lock); 5239 5240 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 5241 5242 for (i = 0; i < NR_DIRTY_TYPE; i++) { 5243 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 5244 GFP_KERNEL); 5245 if (!dirty_i->dirty_segmap[i]) 5246 return -ENOMEM; 5247 } 5248 5249 if (__is_large_section(sbi)) { 5250 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 5251 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi, 5252 bitmap_size, GFP_KERNEL); 5253 if (!dirty_i->dirty_secmap) 5254 return -ENOMEM; 5255 } 5256 5257 init_dirty_segmap(sbi); 5258 return init_victim_secmap(sbi); 5259} 5260 5261static int sanity_check_curseg(struct f2fs_sb_info *sbi) 5262{ 5263 int i; 5264 5265 /* 5266 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 5267 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 5268 */ 5269 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5270 struct curseg_info *curseg = CURSEG_I(sbi, i); 5271 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 5272 unsigned int blkofs = curseg->next_blkoff; 5273 5274 if (f2fs_sb_has_readonly(sbi) && 5275 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE) 5276 continue; 5277 5278 sanity_check_seg_type(sbi, curseg->seg_type); 5279 5280 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) { 5281 f2fs_err(sbi, 5282 "Current segment has invalid alloc_type:%d", 5283 curseg->alloc_type); 5284 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 5285 return -EFSCORRUPTED; 5286 } 5287 5288 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 5289 goto out; 5290 5291 if (curseg->alloc_type == SSR) 5292 continue; 5293 5294 for (blkofs += 1; blkofs < BLKS_PER_SEG(sbi); blkofs++) { 5295 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 5296 continue; 5297out: 5298 f2fs_err(sbi, 5299 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 5300 i, curseg->segno, curseg->alloc_type, 5301 curseg->next_blkoff, blkofs); 5302 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 5303 return -EFSCORRUPTED; 5304 } 5305 } 5306 return 0; 5307} 5308 5309#ifdef CONFIG_BLK_DEV_ZONED 5310static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 5311 struct f2fs_dev_info *fdev, 5312 struct blk_zone *zone) 5313{ 5314 unsigned int zone_segno; 5315 block_t zone_block, valid_block_cnt; 5316 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 5317 int ret; 5318 unsigned int nofs_flags; 5319 5320 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5321 return 0; 5322 5323 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 5324 zone_segno = GET_SEGNO(sbi, zone_block); 5325 5326 /* 5327 * Skip check of zones cursegs point to, since 5328 * fix_curseg_write_pointer() checks them. 5329 */ 5330 if (zone_segno >= MAIN_SEGS(sbi)) 5331 return 0; 5332 5333 /* 5334 * Get # of valid block of the zone. 5335 */ 5336 valid_block_cnt = get_valid_blocks(sbi, zone_segno, true); 5337 if (is_cursec(sbi, GET_SEC_FROM_SEG(sbi, zone_segno))) { 5338 f2fs_notice(sbi, "Open zones: valid block[0x%x,0x%x] cond[%s]", 5339 zone_segno, valid_block_cnt, 5340 blk_zone_cond_str(zone->cond)); 5341 return 0; 5342 } 5343 5344 if ((!valid_block_cnt && zone->cond == BLK_ZONE_COND_EMPTY) || 5345 (valid_block_cnt && zone->cond == BLK_ZONE_COND_FULL)) 5346 return 0; 5347 5348 if (!valid_block_cnt) { 5349 f2fs_notice(sbi, "Zone without valid block has non-zero write " 5350 "pointer. Reset the write pointer: cond[%s]", 5351 blk_zone_cond_str(zone->cond)); 5352 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 5353 zone->len >> log_sectors_per_block); 5354 if (ret) 5355 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5356 fdev->path, ret); 5357 return ret; 5358 } 5359 5360 /* 5361 * If there are valid blocks and the write pointer doesn't match 5362 * with them, we need to report the inconsistency and fill 5363 * the zone till the end to close the zone. This inconsistency 5364 * does not cause write error because the zone will not be 5365 * selected for write operation until it get discarded. 5366 */ 5367 f2fs_notice(sbi, "Valid blocks are not aligned with write " 5368 "pointer: valid block[0x%x,0x%x] cond[%s]", 5369 zone_segno, valid_block_cnt, blk_zone_cond_str(zone->cond)); 5370 5371 nofs_flags = memalloc_nofs_save(); 5372 ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH, 5373 zone->start, zone->len); 5374 memalloc_nofs_restore(nofs_flags); 5375 if (ret == -EOPNOTSUPP) { 5376 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp, 5377 zone->len - (zone->wp - zone->start), 5378 GFP_NOFS, 0); 5379 if (ret) 5380 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)", 5381 fdev->path, ret); 5382 } else if (ret) { 5383 f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)", 5384 fdev->path, ret); 5385 } 5386 5387 return ret; 5388} 5389 5390static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 5391 block_t zone_blkaddr) 5392{ 5393 int i; 5394 5395 for (i = 0; i < sbi->s_ndevs; i++) { 5396 if (!bdev_is_zoned(FDEV(i).bdev)) 5397 continue; 5398 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 5399 zone_blkaddr <= FDEV(i).end_blk)) 5400 return &FDEV(i); 5401 } 5402 5403 return NULL; 5404} 5405 5406static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 5407 void *data) 5408{ 5409 memcpy(data, zone, sizeof(struct blk_zone)); 5410 return 0; 5411} 5412 5413static int do_fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 5414{ 5415 struct curseg_info *cs = CURSEG_I(sbi, type); 5416 struct f2fs_dev_info *zbd; 5417 struct blk_zone zone; 5418 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 5419 block_t cs_zone_block, wp_block; 5420 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 5421 sector_t zone_sector; 5422 int err; 5423 5424 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5425 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5426 5427 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5428 if (!zbd) 5429 return 0; 5430 5431 /* report zone for the sector the curseg points to */ 5432 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5433 << log_sectors_per_block; 5434 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5435 report_one_zone_cb, &zone); 5436 if (err != 1) { 5437 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5438 zbd->path, err); 5439 return err; 5440 } 5441 5442 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5443 return 0; 5444 5445 /* 5446 * When safely unmounted in the previous mount, we could use current 5447 * segments. Otherwise, allocate new sections. 5448 */ 5449 if (is_set_ckpt_flags(sbi, CP_UMOUNT_FLAG)) { 5450 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 5451 wp_segno = GET_SEGNO(sbi, wp_block); 5452 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 5453 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 5454 5455 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 5456 wp_sector_off == 0) 5457 return 0; 5458 5459 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 5460 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", type, cs->segno, 5461 cs->next_blkoff, wp_segno, wp_blkoff); 5462 } 5463 5464 /* Allocate a new section if it's not new. */ 5465 if (cs->next_blkoff || 5466 cs->segno != GET_SEG_FROM_SEC(sbi, GET_ZONE_FROM_SEC(sbi, cs_section))) { 5467 unsigned int old_segno = cs->segno, old_blkoff = cs->next_blkoff; 5468 5469 f2fs_allocate_new_section(sbi, type, true); 5470 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 5471 "[0x%x,0x%x] -> [0x%x,0x%x]", 5472 type, old_segno, old_blkoff, 5473 cs->segno, cs->next_blkoff); 5474 } 5475 5476 /* check consistency of the zone curseg pointed to */ 5477 if (check_zone_write_pointer(sbi, zbd, &zone)) 5478 return -EIO; 5479 5480 /* check newly assigned zone */ 5481 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5482 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5483 5484 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5485 if (!zbd) 5486 return 0; 5487 5488 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5489 << log_sectors_per_block; 5490 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5491 report_one_zone_cb, &zone); 5492 if (err != 1) { 5493 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5494 zbd->path, err); 5495 return err; 5496 } 5497 5498 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5499 return 0; 5500 5501 if (zone.wp != zone.start) { 5502 f2fs_notice(sbi, 5503 "New zone for curseg[%d] is not yet discarded. " 5504 "Reset the zone: curseg[0x%x,0x%x]", 5505 type, cs->segno, cs->next_blkoff); 5506 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block, 5507 zone.len >> log_sectors_per_block); 5508 if (err) { 5509 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5510 zbd->path, err); 5511 return err; 5512 } 5513 } 5514 5515 return 0; 5516} 5517 5518static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5519{ 5520 int i, ret; 5521 5522 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5523 ret = do_fix_curseg_write_pointer(sbi, i); 5524 if (ret) 5525 return ret; 5526 } 5527 5528 return 0; 5529} 5530 5531struct check_zone_write_pointer_args { 5532 struct f2fs_sb_info *sbi; 5533 struct f2fs_dev_info *fdev; 5534}; 5535 5536static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 5537 void *data) 5538{ 5539 struct check_zone_write_pointer_args *args; 5540 5541 args = (struct check_zone_write_pointer_args *)data; 5542 5543 return check_zone_write_pointer(args->sbi, args->fdev, zone); 5544} 5545 5546static int check_write_pointer(struct f2fs_sb_info *sbi) 5547{ 5548 int i, ret; 5549 struct check_zone_write_pointer_args args; 5550 5551 for (i = 0; i < sbi->s_ndevs; i++) { 5552 if (!bdev_is_zoned(FDEV(i).bdev)) 5553 continue; 5554 5555 args.sbi = sbi; 5556 args.fdev = &FDEV(i); 5557 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 5558 check_zone_write_pointer_cb, &args); 5559 if (ret < 0) 5560 return ret; 5561 } 5562 5563 return 0; 5564} 5565 5566int f2fs_check_and_fix_write_pointer(struct f2fs_sb_info *sbi) 5567{ 5568 int ret; 5569 5570 if (!f2fs_sb_has_blkzoned(sbi) || f2fs_readonly(sbi->sb) || 5571 f2fs_hw_is_readonly(sbi)) 5572 return 0; 5573 5574 f2fs_notice(sbi, "Checking entire write pointers"); 5575 ret = fix_curseg_write_pointer(sbi); 5576 if (!ret) 5577 ret = check_write_pointer(sbi); 5578 return ret; 5579} 5580 5581/* 5582 * Return the number of usable blocks in a segment. The number of blocks 5583 * returned is always equal to the number of blocks in a segment for 5584 * segments fully contained within a sequential zone capacity or a 5585 * conventional zone. For segments partially contained in a sequential 5586 * zone capacity, the number of usable blocks up to the zone capacity 5587 * is returned. 0 is returned in all other cases. 5588 */ 5589static inline unsigned int f2fs_usable_zone_blks_in_seg( 5590 struct f2fs_sb_info *sbi, unsigned int segno) 5591{ 5592 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr; 5593 unsigned int secno; 5594 5595 if (!sbi->unusable_blocks_per_sec) 5596 return BLKS_PER_SEG(sbi); 5597 5598 secno = GET_SEC_FROM_SEG(sbi, segno); 5599 seg_start = START_BLOCK(sbi, segno); 5600 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5601 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi); 5602 5603 /* 5604 * If segment starts before zone capacity and spans beyond 5605 * zone capacity, then usable blocks are from seg start to 5606 * zone capacity. If the segment starts after the zone capacity, 5607 * then there are no usable blocks. 5608 */ 5609 if (seg_start >= sec_cap_blkaddr) 5610 return 0; 5611 if (seg_start + BLKS_PER_SEG(sbi) > sec_cap_blkaddr) 5612 return sec_cap_blkaddr - seg_start; 5613 5614 return BLKS_PER_SEG(sbi); 5615} 5616#else 5617int f2fs_check_and_fix_write_pointer(struct f2fs_sb_info *sbi) 5618{ 5619 return 0; 5620} 5621 5622static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi, 5623 unsigned int segno) 5624{ 5625 return 0; 5626} 5627 5628#endif 5629unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi, 5630 unsigned int segno) 5631{ 5632 if (f2fs_sb_has_blkzoned(sbi)) 5633 return f2fs_usable_zone_blks_in_seg(sbi, segno); 5634 5635 return BLKS_PER_SEG(sbi); 5636} 5637 5638unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi) 5639{ 5640 if (f2fs_sb_has_blkzoned(sbi)) 5641 return CAP_SEGS_PER_SEC(sbi); 5642 5643 return SEGS_PER_SEC(sbi); 5644} 5645 5646unsigned long long f2fs_get_section_mtime(struct f2fs_sb_info *sbi, 5647 unsigned int segno) 5648{ 5649 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi); 5650 unsigned int secno = 0, start = 0; 5651 unsigned int total_valid_blocks = 0; 5652 unsigned long long mtime = 0; 5653 unsigned int i = 0; 5654 5655 secno = GET_SEC_FROM_SEG(sbi, segno); 5656 start = GET_SEG_FROM_SEC(sbi, secno); 5657 5658 if (!__is_large_section(sbi)) { 5659 mtime = get_seg_entry(sbi, start + i)->mtime; 5660 goto out; 5661 } 5662 5663 for (i = 0; i < usable_segs_per_sec; i++) { 5664 /* for large section, only check the mtime of valid segments */ 5665 struct seg_entry *se = get_seg_entry(sbi, start+i); 5666 5667 mtime += se->mtime * se->valid_blocks; 5668 total_valid_blocks += se->valid_blocks; 5669 } 5670 5671 if (total_valid_blocks == 0) 5672 return INVALID_MTIME; 5673 5674 mtime = div_u64(mtime, total_valid_blocks); 5675out: 5676 if (unlikely(mtime == INVALID_MTIME)) 5677 mtime -= 1; 5678 return mtime; 5679} 5680 5681/* 5682 * Update min, max modified time for cost-benefit GC algorithm 5683 */ 5684static void init_min_max_mtime(struct f2fs_sb_info *sbi) 5685{ 5686 struct sit_info *sit_i = SIT_I(sbi); 5687 unsigned int segno; 5688 5689 down_write(&sit_i->sentry_lock); 5690 5691 sit_i->min_mtime = ULLONG_MAX; 5692 5693 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) { 5694 unsigned long long mtime = 0; 5695 5696 mtime = f2fs_get_section_mtime(sbi, segno); 5697 5698 if (sit_i->min_mtime > mtime) 5699 sit_i->min_mtime = mtime; 5700 } 5701 sit_i->max_mtime = get_mtime(sbi, false); 5702 sit_i->dirty_max_mtime = 0; 5703 up_write(&sit_i->sentry_lock); 5704} 5705 5706int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 5707{ 5708 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 5709 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 5710 struct f2fs_sm_info *sm_info; 5711 int err; 5712 5713 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 5714 if (!sm_info) 5715 return -ENOMEM; 5716 5717 /* init sm info */ 5718 sbi->sm_info = sm_info; 5719 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 5720 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 5721 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 5722 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 5723 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 5724 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 5725 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 5726 sm_info->rec_prefree_segments = sm_info->main_segments * 5727 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 5728 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 5729 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 5730 5731 if (!f2fs_lfs_mode(sbi)) 5732 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC); 5733 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 5734 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 5735 sm_info->min_seq_blocks = BLKS_PER_SEG(sbi); 5736 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 5737 sm_info->min_ssr_sections = reserved_sections(sbi); 5738 5739 INIT_LIST_HEAD(&sm_info->sit_entry_set); 5740 5741 init_f2fs_rwsem(&sm_info->curseg_lock); 5742 5743 err = f2fs_create_flush_cmd_control(sbi); 5744 if (err) 5745 return err; 5746 5747 err = create_discard_cmd_control(sbi); 5748 if (err) 5749 return err; 5750 5751 err = build_sit_info(sbi); 5752 if (err) 5753 return err; 5754 err = build_free_segmap(sbi); 5755 if (err) 5756 return err; 5757 err = build_curseg(sbi); 5758 if (err) 5759 return err; 5760 5761 /* reinit free segmap based on SIT */ 5762 err = build_sit_entries(sbi); 5763 if (err) 5764 return err; 5765 5766 init_free_segmap(sbi); 5767 err = build_dirty_segmap(sbi); 5768 if (err) 5769 return err; 5770 5771 err = sanity_check_curseg(sbi); 5772 if (err) 5773 return err; 5774 5775 init_min_max_mtime(sbi); 5776 return 0; 5777} 5778 5779static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 5780 enum dirty_type dirty_type) 5781{ 5782 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5783 5784 mutex_lock(&dirty_i->seglist_lock); 5785 kvfree(dirty_i->dirty_segmap[dirty_type]); 5786 dirty_i->nr_dirty[dirty_type] = 0; 5787 mutex_unlock(&dirty_i->seglist_lock); 5788} 5789 5790static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 5791{ 5792 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5793 5794 kvfree(dirty_i->pinned_secmap); 5795 kvfree(dirty_i->victim_secmap); 5796} 5797 5798static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 5799{ 5800 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5801 int i; 5802 5803 if (!dirty_i) 5804 return; 5805 5806 /* discard pre-free/dirty segments list */ 5807 for (i = 0; i < NR_DIRTY_TYPE; i++) 5808 discard_dirty_segmap(sbi, i); 5809 5810 if (__is_large_section(sbi)) { 5811 mutex_lock(&dirty_i->seglist_lock); 5812 kvfree(dirty_i->dirty_secmap); 5813 mutex_unlock(&dirty_i->seglist_lock); 5814 } 5815 5816 destroy_victim_secmap(sbi); 5817 SM_I(sbi)->dirty_info = NULL; 5818 kfree(dirty_i); 5819} 5820 5821static void destroy_curseg(struct f2fs_sb_info *sbi) 5822{ 5823 struct curseg_info *array = SM_I(sbi)->curseg_array; 5824 int i; 5825 5826 if (!array) 5827 return; 5828 SM_I(sbi)->curseg_array = NULL; 5829 for (i = 0; i < NR_CURSEG_TYPE; i++) { 5830 kfree(array[i].sum_blk); 5831 kfree(array[i].journal); 5832 } 5833 kfree(array); 5834} 5835 5836static void destroy_free_segmap(struct f2fs_sb_info *sbi) 5837{ 5838 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 5839 5840 if (!free_i) 5841 return; 5842 SM_I(sbi)->free_info = NULL; 5843 kvfree(free_i->free_segmap); 5844 kvfree(free_i->free_secmap); 5845 kfree(free_i); 5846} 5847 5848static void destroy_sit_info(struct f2fs_sb_info *sbi) 5849{ 5850 struct sit_info *sit_i = SIT_I(sbi); 5851 5852 if (!sit_i) 5853 return; 5854 5855 if (sit_i->sentries) 5856 kvfree(sit_i->bitmap); 5857 kfree(sit_i->tmp_map); 5858 5859 kvfree(sit_i->sentries); 5860 kvfree(sit_i->sec_entries); 5861 kvfree(sit_i->dirty_sentries_bitmap); 5862 5863 SM_I(sbi)->sit_info = NULL; 5864 kfree(sit_i->sit_bitmap); 5865#ifdef CONFIG_F2FS_CHECK_FS 5866 kfree(sit_i->sit_bitmap_mir); 5867 kvfree(sit_i->invalid_segmap); 5868#endif 5869 kfree(sit_i); 5870} 5871 5872void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 5873{ 5874 struct f2fs_sm_info *sm_info = SM_I(sbi); 5875 5876 if (!sm_info) 5877 return; 5878 f2fs_destroy_flush_cmd_control(sbi, true); 5879 destroy_discard_cmd_control(sbi); 5880 destroy_dirty_segmap(sbi); 5881 destroy_curseg(sbi); 5882 destroy_free_segmap(sbi); 5883 destroy_sit_info(sbi); 5884 sbi->sm_info = NULL; 5885 kfree(sm_info); 5886} 5887 5888int __init f2fs_create_segment_manager_caches(void) 5889{ 5890 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 5891 sizeof(struct discard_entry)); 5892 if (!discard_entry_slab) 5893 goto fail; 5894 5895 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 5896 sizeof(struct discard_cmd)); 5897 if (!discard_cmd_slab) 5898 goto destroy_discard_entry; 5899 5900 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 5901 sizeof(struct sit_entry_set)); 5902 if (!sit_entry_set_slab) 5903 goto destroy_discard_cmd; 5904 5905 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry", 5906 sizeof(struct revoke_entry)); 5907 if (!revoke_entry_slab) 5908 goto destroy_sit_entry_set; 5909 return 0; 5910 5911destroy_sit_entry_set: 5912 kmem_cache_destroy(sit_entry_set_slab); 5913destroy_discard_cmd: 5914 kmem_cache_destroy(discard_cmd_slab); 5915destroy_discard_entry: 5916 kmem_cache_destroy(discard_entry_slab); 5917fail: 5918 return -ENOMEM; 5919} 5920 5921void f2fs_destroy_segment_manager_caches(void) 5922{ 5923 kmem_cache_destroy(sit_entry_set_slab); 5924 kmem_cache_destroy(discard_cmd_slab); 5925 kmem_cache_destroy(discard_entry_slab); 5926 kmem_cache_destroy(revoke_entry_slab); 5927}
~/
Read-only file preview.