Loading...
Note: File does not exist in v3.1.
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 stat_dec_atomic_inode(inode);
203
204 F2FS_I(inode)->atomic_write_task = NULL;
205
206 if (clean) {
207 f2fs_i_size_write(inode, fi->original_i_size);
208 fi->original_i_size = 0;
209 }
210 /* avoid stale dirty inode during eviction */
211 sync_inode_metadata(inode, 0);
212}
213
214static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
215 block_t new_addr, block_t *old_addr, bool recover)
216{
217 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
218 struct dnode_of_data dn;
219 struct node_info ni;
220 int err;
221
222retry:
223 set_new_dnode(&dn, inode, NULL, NULL, 0);
224 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE);
225 if (err) {
226 if (err == -ENOMEM) {
227 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
228 goto retry;
229 }
230 return err;
231 }
232
233 err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
234 if (err) {
235 f2fs_put_dnode(&dn);
236 return err;
237 }
238
239 if (recover) {
240 /* dn.data_blkaddr is always valid */
241 if (!__is_valid_data_blkaddr(new_addr)) {
242 if (new_addr == NULL_ADDR)
243 dec_valid_block_count(sbi, inode, 1);
244 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
245 f2fs_update_data_blkaddr(&dn, new_addr);
246 } else {
247 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
248 new_addr, ni.version, true, true);
249 }
250 } else {
251 blkcnt_t count = 1;
252
253 err = inc_valid_block_count(sbi, inode, &count, true);
254 if (err) {
255 f2fs_put_dnode(&dn);
256 return err;
257 }
258
259 *old_addr = dn.data_blkaddr;
260 f2fs_truncate_data_blocks_range(&dn, 1);
261 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
262
263 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
264 ni.version, true, false);
265 }
266
267 f2fs_put_dnode(&dn);
268
269 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode,
270 index, old_addr ? *old_addr : 0, new_addr, recover);
271 return 0;
272}
273
274static void __complete_revoke_list(struct inode *inode, struct list_head *head,
275 bool revoke)
276{
277 struct revoke_entry *cur, *tmp;
278 pgoff_t start_index = 0;
279 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
280
281 list_for_each_entry_safe(cur, tmp, head, list) {
282 if (revoke) {
283 __replace_atomic_write_block(inode, cur->index,
284 cur->old_addr, NULL, true);
285 } else if (truncate) {
286 f2fs_truncate_hole(inode, start_index, cur->index);
287 start_index = cur->index + 1;
288 }
289
290 list_del(&cur->list);
291 kmem_cache_free(revoke_entry_slab, cur);
292 }
293
294 if (!revoke && truncate)
295 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
296}
297
298static int __f2fs_commit_atomic_write(struct inode *inode)
299{
300 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
301 struct f2fs_inode_info *fi = F2FS_I(inode);
302 struct inode *cow_inode = fi->cow_inode;
303 struct revoke_entry *new;
304 struct list_head revoke_list;
305 block_t blkaddr;
306 struct dnode_of_data dn;
307 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
308 pgoff_t off = 0, blen, index;
309 int ret = 0, i;
310
311 INIT_LIST_HEAD(&revoke_list);
312
313 while (len) {
314 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
315
316 set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
317 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
318 if (ret && ret != -ENOENT) {
319 goto out;
320 } else if (ret == -ENOENT) {
321 ret = 0;
322 if (dn.max_level == 0)
323 goto out;
324 goto next;
325 }
326
327 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
328 len);
329 index = off;
330 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
331 blkaddr = f2fs_data_blkaddr(&dn);
332
333 if (!__is_valid_data_blkaddr(blkaddr)) {
334 continue;
335 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
336 DATA_GENERIC_ENHANCE)) {
337 f2fs_put_dnode(&dn);
338 ret = -EFSCORRUPTED;
339 goto out;
340 }
341
342 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
343 true, NULL);
344
345 ret = __replace_atomic_write_block(inode, index, blkaddr,
346 &new->old_addr, false);
347 if (ret) {
348 f2fs_put_dnode(&dn);
349 kmem_cache_free(revoke_entry_slab, new);
350 goto out;
351 }
352
353 f2fs_update_data_blkaddr(&dn, NULL_ADDR);
354 new->index = index;
355 list_add_tail(&new->list, &revoke_list);
356 }
357 f2fs_put_dnode(&dn);
358next:
359 off += blen;
360 len -= blen;
361 }
362
363out:
364 if (ret) {
365 sbi->revoked_atomic_block += fi->atomic_write_cnt;
366 } else {
367 sbi->committed_atomic_block += fi->atomic_write_cnt;
368 set_inode_flag(inode, FI_ATOMIC_COMMITTED);
369 }
370
371 __complete_revoke_list(inode, &revoke_list, ret ? true : false);
372
373 return ret;
374}
375
376int f2fs_commit_atomic_write(struct inode *inode)
377{
378 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
379 struct f2fs_inode_info *fi = F2FS_I(inode);
380 int err;
381
382 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
383 if (err)
384 return err;
385
386 f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
387 f2fs_lock_op(sbi);
388
389 err = __f2fs_commit_atomic_write(inode);
390
391 f2fs_unlock_op(sbi);
392 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
393
394 return err;
395}
396
397/*
398 * This function balances dirty node and dentry pages.
399 * In addition, it controls garbage collection.
400 */
401void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
402{
403 if (f2fs_cp_error(sbi))
404 return;
405
406 if (time_to_inject(sbi, FAULT_CHECKPOINT))
407 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
408
409 /* balance_fs_bg is able to be pending */
410 if (need && excess_cached_nats(sbi))
411 f2fs_balance_fs_bg(sbi, false);
412
413 if (!f2fs_is_checkpoint_ready(sbi))
414 return;
415
416 /*
417 * We should do GC or end up with checkpoint, if there are so many dirty
418 * dir/node pages without enough free segments.
419 */
420 if (has_enough_free_secs(sbi, 0, 0))
421 return;
422
423 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
424 sbi->gc_thread->f2fs_gc_task) {
425 DEFINE_WAIT(wait);
426
427 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
428 TASK_UNINTERRUPTIBLE);
429 wake_up(&sbi->gc_thread->gc_wait_queue_head);
430 io_schedule();
431 finish_wait(&sbi->gc_thread->fggc_wq, &wait);
432 } else {
433 struct f2fs_gc_control gc_control = {
434 .victim_segno = NULL_SEGNO,
435 .init_gc_type = BG_GC,
436 .no_bg_gc = true,
437 .should_migrate_blocks = false,
438 .err_gc_skipped = false,
439 .nr_free_secs = 1 };
440 f2fs_down_write(&sbi->gc_lock);
441 stat_inc_gc_call_count(sbi, FOREGROUND);
442 f2fs_gc(sbi, &gc_control);
443 }
444}
445
446static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
447{
448 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
449 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
450 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
451 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
452 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
453 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
454 unsigned int threshold =
455 SEGS_TO_BLKS(sbi, (factor * DEFAULT_DIRTY_THRESHOLD));
456 unsigned int global_threshold = threshold * 3 / 2;
457
458 if (dents >= threshold || qdata >= threshold ||
459 nodes >= threshold || meta >= threshold ||
460 imeta >= threshold)
461 return true;
462 return dents + qdata + nodes + meta + imeta > global_threshold;
463}
464
465void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
466{
467 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
468 return;
469
470 /* try to shrink extent cache when there is no enough memory */
471 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
472 f2fs_shrink_read_extent_tree(sbi,
473 READ_EXTENT_CACHE_SHRINK_NUMBER);
474
475 /* try to shrink age extent cache when there is no enough memory */
476 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
477 f2fs_shrink_age_extent_tree(sbi,
478 AGE_EXTENT_CACHE_SHRINK_NUMBER);
479
480 /* check the # of cached NAT entries */
481 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
482 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
483
484 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
485 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
486 else
487 f2fs_build_free_nids(sbi, false, false);
488
489 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
490 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
491 goto do_sync;
492
493 /* there is background inflight IO or foreground operation recently */
494 if (is_inflight_io(sbi, REQ_TIME) ||
495 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
496 return;
497
498 /* exceed periodical checkpoint timeout threshold */
499 if (f2fs_time_over(sbi, CP_TIME))
500 goto do_sync;
501
502 /* checkpoint is the only way to shrink partial cached entries */
503 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
504 f2fs_available_free_memory(sbi, INO_ENTRIES))
505 return;
506
507do_sync:
508 if (test_opt(sbi, DATA_FLUSH) && from_bg) {
509 struct blk_plug plug;
510
511 mutex_lock(&sbi->flush_lock);
512
513 blk_start_plug(&plug);
514 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
515 blk_finish_plug(&plug);
516
517 mutex_unlock(&sbi->flush_lock);
518 }
519 stat_inc_cp_call_count(sbi, BACKGROUND);
520 f2fs_sync_fs(sbi->sb, 1);
521}
522
523static int __submit_flush_wait(struct f2fs_sb_info *sbi,
524 struct block_device *bdev)
525{
526 int ret = blkdev_issue_flush(bdev);
527
528 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
529 test_opt(sbi, FLUSH_MERGE), ret);
530 if (!ret)
531 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
532 return ret;
533}
534
535static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
536{
537 int ret = 0;
538 int i;
539
540 if (!f2fs_is_multi_device(sbi))
541 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
542
543 for (i = 0; i < sbi->s_ndevs; i++) {
544 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
545 continue;
546 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
547 if (ret)
548 break;
549 }
550 return ret;
551}
552
553static int issue_flush_thread(void *data)
554{
555 struct f2fs_sb_info *sbi = data;
556 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
557 wait_queue_head_t *q = &fcc->flush_wait_queue;
558repeat:
559 if (kthread_should_stop())
560 return 0;
561
562 if (!llist_empty(&fcc->issue_list)) {
563 struct flush_cmd *cmd, *next;
564 int ret;
565
566 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
567 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
568
569 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
570
571 ret = submit_flush_wait(sbi, cmd->ino);
572 atomic_inc(&fcc->issued_flush);
573
574 llist_for_each_entry_safe(cmd, next,
575 fcc->dispatch_list, llnode) {
576 cmd->ret = ret;
577 complete(&cmd->wait);
578 }
579 fcc->dispatch_list = NULL;
580 }
581
582 wait_event_interruptible(*q,
583 kthread_should_stop() || !llist_empty(&fcc->issue_list));
584 goto repeat;
585}
586
587int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
588{
589 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
590 struct flush_cmd cmd;
591 int ret;
592
593 if (test_opt(sbi, NOBARRIER))
594 return 0;
595
596 if (!test_opt(sbi, FLUSH_MERGE)) {
597 atomic_inc(&fcc->queued_flush);
598 ret = submit_flush_wait(sbi, ino);
599 atomic_dec(&fcc->queued_flush);
600 atomic_inc(&fcc->issued_flush);
601 return ret;
602 }
603
604 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
605 f2fs_is_multi_device(sbi)) {
606 ret = submit_flush_wait(sbi, ino);
607 atomic_dec(&fcc->queued_flush);
608
609 atomic_inc(&fcc->issued_flush);
610 return ret;
611 }
612
613 cmd.ino = ino;
614 init_completion(&cmd.wait);
615
616 llist_add(&cmd.llnode, &fcc->issue_list);
617
618 /*
619 * update issue_list before we wake up issue_flush thread, this
620 * smp_mb() pairs with another barrier in ___wait_event(), see
621 * more details in comments of waitqueue_active().
622 */
623 smp_mb();
624
625 if (waitqueue_active(&fcc->flush_wait_queue))
626 wake_up(&fcc->flush_wait_queue);
627
628 if (fcc->f2fs_issue_flush) {
629 wait_for_completion(&cmd.wait);
630 atomic_dec(&fcc->queued_flush);
631 } else {
632 struct llist_node *list;
633
634 list = llist_del_all(&fcc->issue_list);
635 if (!list) {
636 wait_for_completion(&cmd.wait);
637 atomic_dec(&fcc->queued_flush);
638 } else {
639 struct flush_cmd *tmp, *next;
640
641 ret = submit_flush_wait(sbi, ino);
642
643 llist_for_each_entry_safe(tmp, next, list, llnode) {
644 if (tmp == &cmd) {
645 cmd.ret = ret;
646 atomic_dec(&fcc->queued_flush);
647 continue;
648 }
649 tmp->ret = ret;
650 complete(&tmp->wait);
651 }
652 }
653 }
654
655 return cmd.ret;
656}
657
658int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
659{
660 dev_t dev = sbi->sb->s_bdev->bd_dev;
661 struct flush_cmd_control *fcc;
662
663 if (SM_I(sbi)->fcc_info) {
664 fcc = SM_I(sbi)->fcc_info;
665 if (fcc->f2fs_issue_flush)
666 return 0;
667 goto init_thread;
668 }
669
670 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
671 if (!fcc)
672 return -ENOMEM;
673 atomic_set(&fcc->issued_flush, 0);
674 atomic_set(&fcc->queued_flush, 0);
675 init_waitqueue_head(&fcc->flush_wait_queue);
676 init_llist_head(&fcc->issue_list);
677 SM_I(sbi)->fcc_info = fcc;
678 if (!test_opt(sbi, FLUSH_MERGE))
679 return 0;
680
681init_thread:
682 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
683 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
684 if (IS_ERR(fcc->f2fs_issue_flush)) {
685 int err = PTR_ERR(fcc->f2fs_issue_flush);
686
687 fcc->f2fs_issue_flush = NULL;
688 return err;
689 }
690
691 return 0;
692}
693
694void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
695{
696 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
697
698 if (fcc && fcc->f2fs_issue_flush) {
699 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
700
701 fcc->f2fs_issue_flush = NULL;
702 kthread_stop(flush_thread);
703 }
704 if (free) {
705 kfree(fcc);
706 SM_I(sbi)->fcc_info = NULL;
707 }
708}
709
710int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
711{
712 int ret = 0, i;
713
714 if (!f2fs_is_multi_device(sbi))
715 return 0;
716
717 if (test_opt(sbi, NOBARRIER))
718 return 0;
719
720 for (i = 1; i < sbi->s_ndevs; i++) {
721 int count = DEFAULT_RETRY_IO_COUNT;
722
723 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
724 continue;
725
726 do {
727 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
728 if (ret)
729 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
730 } while (ret && --count);
731
732 if (ret) {
733 f2fs_stop_checkpoint(sbi, false,
734 STOP_CP_REASON_FLUSH_FAIL);
735 break;
736 }
737
738 spin_lock(&sbi->dev_lock);
739 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
740 spin_unlock(&sbi->dev_lock);
741 }
742
743 return ret;
744}
745
746static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
747 enum dirty_type dirty_type)
748{
749 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
750
751 /* need not be added */
752 if (IS_CURSEG(sbi, segno))
753 return;
754
755 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
756 dirty_i->nr_dirty[dirty_type]++;
757
758 if (dirty_type == DIRTY) {
759 struct seg_entry *sentry = get_seg_entry(sbi, segno);
760 enum dirty_type t = sentry->type;
761
762 if (unlikely(t >= DIRTY)) {
763 f2fs_bug_on(sbi, 1);
764 return;
765 }
766 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
767 dirty_i->nr_dirty[t]++;
768
769 if (__is_large_section(sbi)) {
770 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
771 block_t valid_blocks =
772 get_valid_blocks(sbi, segno, true);
773
774 f2fs_bug_on(sbi, unlikely(!valid_blocks ||
775 valid_blocks == CAP_BLKS_PER_SEC(sbi)));
776
777 if (!IS_CURSEC(sbi, secno))
778 set_bit(secno, dirty_i->dirty_secmap);
779 }
780 }
781}
782
783static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
784 enum dirty_type dirty_type)
785{
786 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
787 block_t valid_blocks;
788
789 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
790 dirty_i->nr_dirty[dirty_type]--;
791
792 if (dirty_type == DIRTY) {
793 struct seg_entry *sentry = get_seg_entry(sbi, segno);
794 enum dirty_type t = sentry->type;
795
796 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
797 dirty_i->nr_dirty[t]--;
798
799 valid_blocks = get_valid_blocks(sbi, segno, true);
800 if (valid_blocks == 0) {
801 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
802 dirty_i->victim_secmap);
803#ifdef CONFIG_F2FS_CHECK_FS
804 clear_bit(segno, SIT_I(sbi)->invalid_segmap);
805#endif
806 }
807 if (__is_large_section(sbi)) {
808 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
809
810 if (!valid_blocks ||
811 valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
812 clear_bit(secno, dirty_i->dirty_secmap);
813 return;
814 }
815
816 if (!IS_CURSEC(sbi, secno))
817 set_bit(secno, dirty_i->dirty_secmap);
818 }
819 }
820}
821
822/*
823 * Should not occur error such as -ENOMEM.
824 * Adding dirty entry into seglist is not critical operation.
825 * If a given segment is one of current working segments, it won't be added.
826 */
827static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
828{
829 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
830 unsigned short valid_blocks, ckpt_valid_blocks;
831 unsigned int usable_blocks;
832
833 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
834 return;
835
836 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
837 mutex_lock(&dirty_i->seglist_lock);
838
839 valid_blocks = get_valid_blocks(sbi, segno, false);
840 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
841
842 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
843 ckpt_valid_blocks == usable_blocks)) {
844 __locate_dirty_segment(sbi, segno, PRE);
845 __remove_dirty_segment(sbi, segno, DIRTY);
846 } else if (valid_blocks < usable_blocks) {
847 __locate_dirty_segment(sbi, segno, DIRTY);
848 } else {
849 /* Recovery routine with SSR needs this */
850 __remove_dirty_segment(sbi, segno, DIRTY);
851 }
852
853 mutex_unlock(&dirty_i->seglist_lock);
854}
855
856/* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
857void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
858{
859 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
860 unsigned int segno;
861
862 mutex_lock(&dirty_i->seglist_lock);
863 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
864 if (get_valid_blocks(sbi, segno, false))
865 continue;
866 if (IS_CURSEG(sbi, segno))
867 continue;
868 __locate_dirty_segment(sbi, segno, PRE);
869 __remove_dirty_segment(sbi, segno, DIRTY);
870 }
871 mutex_unlock(&dirty_i->seglist_lock);
872}
873
874block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
875{
876 int ovp_hole_segs =
877 (overprovision_segments(sbi) - reserved_segments(sbi));
878 block_t ovp_holes = SEGS_TO_BLKS(sbi, ovp_hole_segs);
879 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
880 block_t holes[2] = {0, 0}; /* DATA and NODE */
881 block_t unusable;
882 struct seg_entry *se;
883 unsigned int segno;
884
885 mutex_lock(&dirty_i->seglist_lock);
886 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
887 se = get_seg_entry(sbi, segno);
888 if (IS_NODESEG(se->type))
889 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
890 se->valid_blocks;
891 else
892 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
893 se->valid_blocks;
894 }
895 mutex_unlock(&dirty_i->seglist_lock);
896
897 unusable = max(holes[DATA], holes[NODE]);
898 if (unusable > ovp_holes)
899 return unusable - ovp_holes;
900 return 0;
901}
902
903int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
904{
905 int ovp_hole_segs =
906 (overprovision_segments(sbi) - reserved_segments(sbi));
907
908 if (F2FS_OPTION(sbi).unusable_cap_perc == 100)
909 return 0;
910 if (unusable > F2FS_OPTION(sbi).unusable_cap)
911 return -EAGAIN;
912 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
913 dirty_segments(sbi) > ovp_hole_segs)
914 return -EAGAIN;
915 if (has_not_enough_free_secs(sbi, 0, 0))
916 return -EAGAIN;
917 return 0;
918}
919
920/* This is only used by SBI_CP_DISABLED */
921static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
922{
923 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
924 unsigned int segno = 0;
925
926 mutex_lock(&dirty_i->seglist_lock);
927 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
928 if (get_valid_blocks(sbi, segno, false))
929 continue;
930 if (get_ckpt_valid_blocks(sbi, segno, false))
931 continue;
932 mutex_unlock(&dirty_i->seglist_lock);
933 return segno;
934 }
935 mutex_unlock(&dirty_i->seglist_lock);
936 return NULL_SEGNO;
937}
938
939static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
940 struct block_device *bdev, block_t lstart,
941 block_t start, block_t len)
942{
943 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
944 struct list_head *pend_list;
945 struct discard_cmd *dc;
946
947 f2fs_bug_on(sbi, !len);
948
949 pend_list = &dcc->pend_list[plist_idx(len)];
950
951 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
952 INIT_LIST_HEAD(&dc->list);
953 dc->bdev = bdev;
954 dc->di.lstart = lstart;
955 dc->di.start = start;
956 dc->di.len = len;
957 dc->ref = 0;
958 dc->state = D_PREP;
959 dc->queued = 0;
960 dc->error = 0;
961 init_completion(&dc->wait);
962 list_add_tail(&dc->list, pend_list);
963 spin_lock_init(&dc->lock);
964 dc->bio_ref = 0;
965 atomic_inc(&dcc->discard_cmd_cnt);
966 dcc->undiscard_blks += len;
967
968 return dc;
969}
970
971static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
972{
973#ifdef CONFIG_F2FS_CHECK_FS
974 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
975 struct rb_node *cur = rb_first_cached(&dcc->root), *next;
976 struct discard_cmd *cur_dc, *next_dc;
977
978 while (cur) {
979 next = rb_next(cur);
980 if (!next)
981 return true;
982
983 cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
984 next_dc = rb_entry(next, struct discard_cmd, rb_node);
985
986 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
987 f2fs_info(sbi, "broken discard_rbtree, "
988 "cur(%u, %u) next(%u, %u)",
989 cur_dc->di.lstart, cur_dc->di.len,
990 next_dc->di.lstart, next_dc->di.len);
991 return false;
992 }
993 cur = next;
994 }
995#endif
996 return true;
997}
998
999static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
1000 block_t blkaddr)
1001{
1002 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1003 struct rb_node *node = dcc->root.rb_root.rb_node;
1004 struct discard_cmd *dc;
1005
1006 while (node) {
1007 dc = rb_entry(node, struct discard_cmd, rb_node);
1008
1009 if (blkaddr < dc->di.lstart)
1010 node = node->rb_left;
1011 else if (blkaddr >= dc->di.lstart + dc->di.len)
1012 node = node->rb_right;
1013 else
1014 return dc;
1015 }
1016 return NULL;
1017}
1018
1019static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1020 block_t blkaddr,
1021 struct discard_cmd **prev_entry,
1022 struct discard_cmd **next_entry,
1023 struct rb_node ***insert_p,
1024 struct rb_node **insert_parent)
1025{
1026 struct rb_node **pnode = &root->rb_root.rb_node;
1027 struct rb_node *parent = NULL, *tmp_node;
1028 struct discard_cmd *dc;
1029
1030 *insert_p = NULL;
1031 *insert_parent = NULL;
1032 *prev_entry = NULL;
1033 *next_entry = NULL;
1034
1035 if (RB_EMPTY_ROOT(&root->rb_root))
1036 return NULL;
1037
1038 while (*pnode) {
1039 parent = *pnode;
1040 dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1041
1042 if (blkaddr < dc->di.lstart)
1043 pnode = &(*pnode)->rb_left;
1044 else if (blkaddr >= dc->di.lstart + dc->di.len)
1045 pnode = &(*pnode)->rb_right;
1046 else
1047 goto lookup_neighbors;
1048 }
1049
1050 *insert_p = pnode;
1051 *insert_parent = parent;
1052
1053 dc = rb_entry(parent, struct discard_cmd, rb_node);
1054 tmp_node = parent;
1055 if (parent && blkaddr > dc->di.lstart)
1056 tmp_node = rb_next(parent);
1057 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1058
1059 tmp_node = parent;
1060 if (parent && blkaddr < dc->di.lstart)
1061 tmp_node = rb_prev(parent);
1062 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1063 return NULL;
1064
1065lookup_neighbors:
1066 /* lookup prev node for merging backward later */
1067 tmp_node = rb_prev(&dc->rb_node);
1068 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1069
1070 /* lookup next node for merging frontward later */
1071 tmp_node = rb_next(&dc->rb_node);
1072 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1073 return dc;
1074}
1075
1076static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1077 struct discard_cmd *dc)
1078{
1079 if (dc->state == D_DONE)
1080 atomic_sub(dc->queued, &dcc->queued_discard);
1081
1082 list_del(&dc->list);
1083 rb_erase_cached(&dc->rb_node, &dcc->root);
1084 dcc->undiscard_blks -= dc->di.len;
1085
1086 kmem_cache_free(discard_cmd_slab, dc);
1087
1088 atomic_dec(&dcc->discard_cmd_cnt);
1089}
1090
1091static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1092 struct discard_cmd *dc)
1093{
1094 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1095 unsigned long flags;
1096
1097 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1098
1099 spin_lock_irqsave(&dc->lock, flags);
1100 if (dc->bio_ref) {
1101 spin_unlock_irqrestore(&dc->lock, flags);
1102 return;
1103 }
1104 spin_unlock_irqrestore(&dc->lock, flags);
1105
1106 f2fs_bug_on(sbi, dc->ref);
1107
1108 if (dc->error == -EOPNOTSUPP)
1109 dc->error = 0;
1110
1111 if (dc->error)
1112 printk_ratelimited(
1113 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1114 KERN_INFO, sbi->sb->s_id,
1115 dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1116 __detach_discard_cmd(dcc, dc);
1117}
1118
1119static void f2fs_submit_discard_endio(struct bio *bio)
1120{
1121 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1122 unsigned long flags;
1123
1124 spin_lock_irqsave(&dc->lock, flags);
1125 if (!dc->error)
1126 dc->error = blk_status_to_errno(bio->bi_status);
1127 dc->bio_ref--;
1128 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1129 dc->state = D_DONE;
1130 complete_all(&dc->wait);
1131 }
1132 spin_unlock_irqrestore(&dc->lock, flags);
1133 bio_put(bio);
1134}
1135
1136static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1137 block_t start, block_t end)
1138{
1139#ifdef CONFIG_F2FS_CHECK_FS
1140 struct seg_entry *sentry;
1141 unsigned int segno;
1142 block_t blk = start;
1143 unsigned long offset, size, *map;
1144
1145 while (blk < end) {
1146 segno = GET_SEGNO(sbi, blk);
1147 sentry = get_seg_entry(sbi, segno);
1148 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1149
1150 if (end < START_BLOCK(sbi, segno + 1))
1151 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1152 else
1153 size = BLKS_PER_SEG(sbi);
1154 map = (unsigned long *)(sentry->cur_valid_map);
1155 offset = __find_rev_next_bit(map, size, offset);
1156 f2fs_bug_on(sbi, offset != size);
1157 blk = START_BLOCK(sbi, segno + 1);
1158 }
1159#endif
1160}
1161
1162static void __init_discard_policy(struct f2fs_sb_info *sbi,
1163 struct discard_policy *dpolicy,
1164 int discard_type, unsigned int granularity)
1165{
1166 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1167
1168 /* common policy */
1169 dpolicy->type = discard_type;
1170 dpolicy->sync = true;
1171 dpolicy->ordered = false;
1172 dpolicy->granularity = granularity;
1173
1174 dpolicy->max_requests = dcc->max_discard_request;
1175 dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1176 dpolicy->timeout = false;
1177
1178 if (discard_type == DPOLICY_BG) {
1179 dpolicy->min_interval = dcc->min_discard_issue_time;
1180 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1181 dpolicy->max_interval = dcc->max_discard_issue_time;
1182 if (dcc->discard_io_aware == DPOLICY_IO_AWARE_ENABLE)
1183 dpolicy->io_aware = true;
1184 else if (dcc->discard_io_aware == DPOLICY_IO_AWARE_DISABLE)
1185 dpolicy->io_aware = false;
1186 dpolicy->sync = false;
1187 dpolicy->ordered = true;
1188 if (utilization(sbi) > dcc->discard_urgent_util) {
1189 dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1190 if (atomic_read(&dcc->discard_cmd_cnt))
1191 dpolicy->max_interval =
1192 dcc->min_discard_issue_time;
1193 }
1194 } else if (discard_type == DPOLICY_FORCE) {
1195 dpolicy->min_interval = dcc->min_discard_issue_time;
1196 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1197 dpolicy->max_interval = dcc->max_discard_issue_time;
1198 dpolicy->io_aware = false;
1199 } else if (discard_type == DPOLICY_FSTRIM) {
1200 dpolicy->io_aware = false;
1201 } else if (discard_type == DPOLICY_UMOUNT) {
1202 dpolicy->io_aware = false;
1203 /* we need to issue all to keep CP_TRIMMED_FLAG */
1204 dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1205 dpolicy->timeout = true;
1206 }
1207}
1208
1209static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1210 struct block_device *bdev, block_t lstart,
1211 block_t start, block_t len);
1212
1213#ifdef CONFIG_BLK_DEV_ZONED
1214static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
1215 struct discard_cmd *dc, blk_opf_t flag,
1216 struct list_head *wait_list,
1217 unsigned int *issued)
1218{
1219 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1220 struct block_device *bdev = dc->bdev;
1221 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
1222 unsigned long flags;
1223
1224 trace_f2fs_issue_reset_zone(bdev, dc->di.start);
1225
1226 spin_lock_irqsave(&dc->lock, flags);
1227 dc->state = D_SUBMIT;
1228 dc->bio_ref++;
1229 spin_unlock_irqrestore(&dc->lock, flags);
1230
1231 if (issued)
1232 (*issued)++;
1233
1234 atomic_inc(&dcc->queued_discard);
1235 dc->queued++;
1236 list_move_tail(&dc->list, wait_list);
1237
1238 /* sanity check on discard range */
1239 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
1240
1241 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
1242 bio->bi_private = dc;
1243 bio->bi_end_io = f2fs_submit_discard_endio;
1244 submit_bio(bio);
1245
1246 atomic_inc(&dcc->issued_discard);
1247 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE);
1248}
1249#endif
1250
1251/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1252static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1253 struct discard_policy *dpolicy,
1254 struct discard_cmd *dc, int *issued)
1255{
1256 struct block_device *bdev = dc->bdev;
1257 unsigned int max_discard_blocks =
1258 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1259 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1260 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1261 &(dcc->fstrim_list) : &(dcc->wait_list);
1262 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1263 block_t lstart, start, len, total_len;
1264 int err = 0;
1265
1266 if (dc->state != D_PREP)
1267 return 0;
1268
1269 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1270 return 0;
1271
1272#ifdef CONFIG_BLK_DEV_ZONED
1273 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) {
1274 int devi = f2fs_bdev_index(sbi, bdev);
1275
1276 if (devi < 0)
1277 return -EINVAL;
1278
1279 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1280 __submit_zone_reset_cmd(sbi, dc, flag,
1281 wait_list, issued);
1282 return 0;
1283 }
1284 }
1285#endif
1286
1287 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1288
1289 lstart = dc->di.lstart;
1290 start = dc->di.start;
1291 len = dc->di.len;
1292 total_len = len;
1293
1294 dc->di.len = 0;
1295
1296 while (total_len && *issued < dpolicy->max_requests && !err) {
1297 struct bio *bio = NULL;
1298 unsigned long flags;
1299 bool last = true;
1300
1301 if (len > max_discard_blocks) {
1302 len = max_discard_blocks;
1303 last = false;
1304 }
1305
1306 (*issued)++;
1307 if (*issued == dpolicy->max_requests)
1308 last = true;
1309
1310 dc->di.len += len;
1311
1312 if (time_to_inject(sbi, FAULT_DISCARD)) {
1313 err = -EIO;
1314 } else {
1315 err = __blkdev_issue_discard(bdev,
1316 SECTOR_FROM_BLOCK(start),
1317 SECTOR_FROM_BLOCK(len),
1318 GFP_NOFS, &bio);
1319 }
1320 if (err) {
1321 spin_lock_irqsave(&dc->lock, flags);
1322 if (dc->state == D_PARTIAL)
1323 dc->state = D_SUBMIT;
1324 spin_unlock_irqrestore(&dc->lock, flags);
1325
1326 break;
1327 }
1328
1329 f2fs_bug_on(sbi, !bio);
1330
1331 /*
1332 * should keep before submission to avoid D_DONE
1333 * right away
1334 */
1335 spin_lock_irqsave(&dc->lock, flags);
1336 if (last)
1337 dc->state = D_SUBMIT;
1338 else
1339 dc->state = D_PARTIAL;
1340 dc->bio_ref++;
1341 spin_unlock_irqrestore(&dc->lock, flags);
1342
1343 atomic_inc(&dcc->queued_discard);
1344 dc->queued++;
1345 list_move_tail(&dc->list, wait_list);
1346
1347 /* sanity check on discard range */
1348 __check_sit_bitmap(sbi, lstart, lstart + len);
1349
1350 bio->bi_private = dc;
1351 bio->bi_end_io = f2fs_submit_discard_endio;
1352 bio->bi_opf |= flag;
1353 submit_bio(bio);
1354
1355 atomic_inc(&dcc->issued_discard);
1356
1357 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1358
1359 lstart += len;
1360 start += len;
1361 total_len -= len;
1362 len = total_len;
1363 }
1364
1365 if (!err && len) {
1366 dcc->undiscard_blks -= len;
1367 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1368 }
1369 return err;
1370}
1371
1372static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1373 struct block_device *bdev, block_t lstart,
1374 block_t start, block_t len)
1375{
1376 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1377 struct rb_node **p = &dcc->root.rb_root.rb_node;
1378 struct rb_node *parent = NULL;
1379 struct discard_cmd *dc;
1380 bool leftmost = true;
1381
1382 /* look up rb tree to find parent node */
1383 while (*p) {
1384 parent = *p;
1385 dc = rb_entry(parent, struct discard_cmd, rb_node);
1386
1387 if (lstart < dc->di.lstart) {
1388 p = &(*p)->rb_left;
1389 } else if (lstart >= dc->di.lstart + dc->di.len) {
1390 p = &(*p)->rb_right;
1391 leftmost = false;
1392 } else {
1393 /* Let's skip to add, if exists */
1394 return;
1395 }
1396 }
1397
1398 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1399
1400 rb_link_node(&dc->rb_node, parent, p);
1401 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1402}
1403
1404static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1405 struct discard_cmd *dc)
1406{
1407 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1408}
1409
1410static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1411 struct discard_cmd *dc, block_t blkaddr)
1412{
1413 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1414 struct discard_info di = dc->di;
1415 bool modified = false;
1416
1417 if (dc->state == D_DONE || dc->di.len == 1) {
1418 __remove_discard_cmd(sbi, dc);
1419 return;
1420 }
1421
1422 dcc->undiscard_blks -= di.len;
1423
1424 if (blkaddr > di.lstart) {
1425 dc->di.len = blkaddr - dc->di.lstart;
1426 dcc->undiscard_blks += dc->di.len;
1427 __relocate_discard_cmd(dcc, dc);
1428 modified = true;
1429 }
1430
1431 if (blkaddr < di.lstart + di.len - 1) {
1432 if (modified) {
1433 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1434 di.start + blkaddr + 1 - di.lstart,
1435 di.lstart + di.len - 1 - blkaddr);
1436 } else {
1437 dc->di.lstart++;
1438 dc->di.len--;
1439 dc->di.start++;
1440 dcc->undiscard_blks += dc->di.len;
1441 __relocate_discard_cmd(dcc, dc);
1442 }
1443 }
1444}
1445
1446static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1447 struct block_device *bdev, block_t lstart,
1448 block_t start, block_t len)
1449{
1450 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1451 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1452 struct discard_cmd *dc;
1453 struct discard_info di = {0};
1454 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1455 unsigned int max_discard_blocks =
1456 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1457 block_t end = lstart + len;
1458
1459 dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1460 &prev_dc, &next_dc, &insert_p, &insert_parent);
1461 if (dc)
1462 prev_dc = dc;
1463
1464 if (!prev_dc) {
1465 di.lstart = lstart;
1466 di.len = next_dc ? next_dc->di.lstart - lstart : len;
1467 di.len = min(di.len, len);
1468 di.start = start;
1469 }
1470
1471 while (1) {
1472 struct rb_node *node;
1473 bool merged = false;
1474 struct discard_cmd *tdc = NULL;
1475
1476 if (prev_dc) {
1477 di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1478 if (di.lstart < lstart)
1479 di.lstart = lstart;
1480 if (di.lstart >= end)
1481 break;
1482
1483 if (!next_dc || next_dc->di.lstart > end)
1484 di.len = end - di.lstart;
1485 else
1486 di.len = next_dc->di.lstart - di.lstart;
1487 di.start = start + di.lstart - lstart;
1488 }
1489
1490 if (!di.len)
1491 goto next;
1492
1493 if (prev_dc && prev_dc->state == D_PREP &&
1494 prev_dc->bdev == bdev &&
1495 __is_discard_back_mergeable(&di, &prev_dc->di,
1496 max_discard_blocks)) {
1497 prev_dc->di.len += di.len;
1498 dcc->undiscard_blks += di.len;
1499 __relocate_discard_cmd(dcc, prev_dc);
1500 di = prev_dc->di;
1501 tdc = prev_dc;
1502 merged = true;
1503 }
1504
1505 if (next_dc && next_dc->state == D_PREP &&
1506 next_dc->bdev == bdev &&
1507 __is_discard_front_mergeable(&di, &next_dc->di,
1508 max_discard_blocks)) {
1509 next_dc->di.lstart = di.lstart;
1510 next_dc->di.len += di.len;
1511 next_dc->di.start = di.start;
1512 dcc->undiscard_blks += di.len;
1513 __relocate_discard_cmd(dcc, next_dc);
1514 if (tdc)
1515 __remove_discard_cmd(sbi, tdc);
1516 merged = true;
1517 }
1518
1519 if (!merged)
1520 __insert_discard_cmd(sbi, bdev,
1521 di.lstart, di.start, di.len);
1522 next:
1523 prev_dc = next_dc;
1524 if (!prev_dc)
1525 break;
1526
1527 node = rb_next(&prev_dc->rb_node);
1528 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1529 }
1530}
1531
1532#ifdef CONFIG_BLK_DEV_ZONED
1533static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
1534 struct block_device *bdev, block_t blkstart, block_t lblkstart,
1535 block_t blklen)
1536{
1537 trace_f2fs_queue_reset_zone(bdev, blkstart);
1538
1539 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1540 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
1541 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1542}
1543#endif
1544
1545static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1546 struct block_device *bdev, block_t blkstart, block_t blklen)
1547{
1548 block_t lblkstart = blkstart;
1549
1550 if (!f2fs_bdev_support_discard(bdev))
1551 return;
1552
1553 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1554
1555 if (f2fs_is_multi_device(sbi)) {
1556 int devi = f2fs_target_device_index(sbi, blkstart);
1557
1558 blkstart -= FDEV(devi).start_blk;
1559 }
1560 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1561 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1562 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1563}
1564
1565static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1566 struct discard_policy *dpolicy, int *issued)
1567{
1568 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1569 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1570 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1571 struct discard_cmd *dc;
1572 struct blk_plug plug;
1573 bool io_interrupted = false;
1574
1575 mutex_lock(&dcc->cmd_lock);
1576 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1577 &prev_dc, &next_dc, &insert_p, &insert_parent);
1578 if (!dc)
1579 dc = next_dc;
1580
1581 blk_start_plug(&plug);
1582
1583 while (dc) {
1584 struct rb_node *node;
1585 int err = 0;
1586
1587 if (dc->state != D_PREP)
1588 goto next;
1589
1590 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1591 io_interrupted = true;
1592 break;
1593 }
1594
1595 dcc->next_pos = dc->di.lstart + dc->di.len;
1596 err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1597
1598 if (*issued >= dpolicy->max_requests)
1599 break;
1600next:
1601 node = rb_next(&dc->rb_node);
1602 if (err)
1603 __remove_discard_cmd(sbi, dc);
1604 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1605 }
1606
1607 blk_finish_plug(&plug);
1608
1609 if (!dc)
1610 dcc->next_pos = 0;
1611
1612 mutex_unlock(&dcc->cmd_lock);
1613
1614 if (!(*issued) && io_interrupted)
1615 *issued = -1;
1616}
1617static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1618 struct discard_policy *dpolicy);
1619
1620static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1621 struct discard_policy *dpolicy)
1622{
1623 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1624 struct list_head *pend_list;
1625 struct discard_cmd *dc, *tmp;
1626 struct blk_plug plug;
1627 int i, issued;
1628 bool io_interrupted = false;
1629
1630 if (dpolicy->timeout)
1631 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1632
1633retry:
1634 issued = 0;
1635 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1636 if (dpolicy->timeout &&
1637 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1638 break;
1639
1640 if (i + 1 < dpolicy->granularity)
1641 break;
1642
1643 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1644 __issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1645 return issued;
1646 }
1647
1648 pend_list = &dcc->pend_list[i];
1649
1650 mutex_lock(&dcc->cmd_lock);
1651 if (list_empty(pend_list))
1652 goto next;
1653 if (unlikely(dcc->rbtree_check))
1654 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1655 blk_start_plug(&plug);
1656 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1657 f2fs_bug_on(sbi, dc->state != D_PREP);
1658
1659 if (dpolicy->timeout &&
1660 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1661 break;
1662
1663 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1664 !is_idle(sbi, DISCARD_TIME)) {
1665 io_interrupted = true;
1666 break;
1667 }
1668
1669 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1670
1671 if (issued >= dpolicy->max_requests)
1672 break;
1673 }
1674 blk_finish_plug(&plug);
1675next:
1676 mutex_unlock(&dcc->cmd_lock);
1677
1678 if (issued >= dpolicy->max_requests || io_interrupted)
1679 break;
1680 }
1681
1682 if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1683 __wait_all_discard_cmd(sbi, dpolicy);
1684 goto retry;
1685 }
1686
1687 if (!issued && io_interrupted)
1688 issued = -1;
1689
1690 return issued;
1691}
1692
1693static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1694{
1695 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1696 struct list_head *pend_list;
1697 struct discard_cmd *dc, *tmp;
1698 int i;
1699 bool dropped = false;
1700
1701 mutex_lock(&dcc->cmd_lock);
1702 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1703 pend_list = &dcc->pend_list[i];
1704 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1705 f2fs_bug_on(sbi, dc->state != D_PREP);
1706 __remove_discard_cmd(sbi, dc);
1707 dropped = true;
1708 }
1709 }
1710 mutex_unlock(&dcc->cmd_lock);
1711
1712 return dropped;
1713}
1714
1715void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1716{
1717 __drop_discard_cmd(sbi);
1718}
1719
1720static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1721 struct discard_cmd *dc)
1722{
1723 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1724 unsigned int len = 0;
1725
1726 wait_for_completion_io(&dc->wait);
1727 mutex_lock(&dcc->cmd_lock);
1728 f2fs_bug_on(sbi, dc->state != D_DONE);
1729 dc->ref--;
1730 if (!dc->ref) {
1731 if (!dc->error)
1732 len = dc->di.len;
1733 __remove_discard_cmd(sbi, dc);
1734 }
1735 mutex_unlock(&dcc->cmd_lock);
1736
1737 return len;
1738}
1739
1740static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1741 struct discard_policy *dpolicy,
1742 block_t start, block_t end)
1743{
1744 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1745 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1746 &(dcc->fstrim_list) : &(dcc->wait_list);
1747 struct discard_cmd *dc = NULL, *iter, *tmp;
1748 unsigned int trimmed = 0;
1749
1750next:
1751 dc = NULL;
1752
1753 mutex_lock(&dcc->cmd_lock);
1754 list_for_each_entry_safe(iter, tmp, wait_list, list) {
1755 if (iter->di.lstart + iter->di.len <= start ||
1756 end <= iter->di.lstart)
1757 continue;
1758 if (iter->di.len < dpolicy->granularity)
1759 continue;
1760 if (iter->state == D_DONE && !iter->ref) {
1761 wait_for_completion_io(&iter->wait);
1762 if (!iter->error)
1763 trimmed += iter->di.len;
1764 __remove_discard_cmd(sbi, iter);
1765 } else {
1766 iter->ref++;
1767 dc = iter;
1768 break;
1769 }
1770 }
1771 mutex_unlock(&dcc->cmd_lock);
1772
1773 if (dc) {
1774 trimmed += __wait_one_discard_bio(sbi, dc);
1775 goto next;
1776 }
1777
1778 return trimmed;
1779}
1780
1781static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1782 struct discard_policy *dpolicy)
1783{
1784 struct discard_policy dp;
1785 unsigned int discard_blks;
1786
1787 if (dpolicy)
1788 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1789
1790 /* wait all */
1791 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1792 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1793 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1794 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1795
1796 return discard_blks;
1797}
1798
1799/* This should be covered by global mutex, &sit_i->sentry_lock */
1800static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1801{
1802 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1803 struct discard_cmd *dc;
1804 bool need_wait = false;
1805
1806 mutex_lock(&dcc->cmd_lock);
1807 dc = __lookup_discard_cmd(sbi, blkaddr);
1808#ifdef CONFIG_BLK_DEV_ZONED
1809 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
1810 int devi = f2fs_bdev_index(sbi, dc->bdev);
1811
1812 if (devi < 0) {
1813 mutex_unlock(&dcc->cmd_lock);
1814 return;
1815 }
1816
1817 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1818 /* force submit zone reset */
1819 if (dc->state == D_PREP)
1820 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
1821 &dcc->wait_list, NULL);
1822 dc->ref++;
1823 mutex_unlock(&dcc->cmd_lock);
1824 /* wait zone reset */
1825 __wait_one_discard_bio(sbi, dc);
1826 return;
1827 }
1828 }
1829#endif
1830 if (dc) {
1831 if (dc->state == D_PREP) {
1832 __punch_discard_cmd(sbi, dc, blkaddr);
1833 } else {
1834 dc->ref++;
1835 need_wait = true;
1836 }
1837 }
1838 mutex_unlock(&dcc->cmd_lock);
1839
1840 if (need_wait)
1841 __wait_one_discard_bio(sbi, dc);
1842}
1843
1844void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1845{
1846 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1847
1848 if (dcc && dcc->f2fs_issue_discard) {
1849 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1850
1851 dcc->f2fs_issue_discard = NULL;
1852 kthread_stop(discard_thread);
1853 }
1854}
1855
1856/**
1857 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1858 * @sbi: the f2fs_sb_info data for discard cmd to issue
1859 *
1860 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1861 *
1862 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1863 */
1864bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1865{
1866 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1867 struct discard_policy dpolicy;
1868 bool dropped;
1869
1870 if (!atomic_read(&dcc->discard_cmd_cnt))
1871 return true;
1872
1873 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1874 dcc->discard_granularity);
1875 __issue_discard_cmd(sbi, &dpolicy);
1876 dropped = __drop_discard_cmd(sbi);
1877
1878 /* just to make sure there is no pending discard commands */
1879 __wait_all_discard_cmd(sbi, NULL);
1880
1881 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1882 return !dropped;
1883}
1884
1885static int issue_discard_thread(void *data)
1886{
1887 struct f2fs_sb_info *sbi = data;
1888 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1889 wait_queue_head_t *q = &dcc->discard_wait_queue;
1890 struct discard_policy dpolicy;
1891 unsigned int wait_ms = dcc->min_discard_issue_time;
1892 int issued;
1893
1894 set_freezable();
1895
1896 do {
1897 wait_event_freezable_timeout(*q,
1898 kthread_should_stop() || dcc->discard_wake,
1899 msecs_to_jiffies(wait_ms));
1900
1901 if (sbi->gc_mode == GC_URGENT_HIGH ||
1902 !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1903 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1904 MIN_DISCARD_GRANULARITY);
1905 else
1906 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1907 dcc->discard_granularity);
1908
1909 if (dcc->discard_wake)
1910 dcc->discard_wake = false;
1911
1912 /* clean up pending candidates before going to sleep */
1913 if (atomic_read(&dcc->queued_discard))
1914 __wait_all_discard_cmd(sbi, NULL);
1915
1916 if (f2fs_readonly(sbi->sb))
1917 continue;
1918 if (kthread_should_stop())
1919 return 0;
1920 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1921 !atomic_read(&dcc->discard_cmd_cnt)) {
1922 wait_ms = dpolicy.max_interval;
1923 continue;
1924 }
1925
1926 sb_start_intwrite(sbi->sb);
1927
1928 issued = __issue_discard_cmd(sbi, &dpolicy);
1929 if (issued > 0) {
1930 __wait_all_discard_cmd(sbi, &dpolicy);
1931 wait_ms = dpolicy.min_interval;
1932 } else if (issued == -1) {
1933 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1934 if (!wait_ms)
1935 wait_ms = dpolicy.mid_interval;
1936 } else {
1937 wait_ms = dpolicy.max_interval;
1938 }
1939 if (!atomic_read(&dcc->discard_cmd_cnt))
1940 wait_ms = dpolicy.max_interval;
1941
1942 sb_end_intwrite(sbi->sb);
1943
1944 } while (!kthread_should_stop());
1945 return 0;
1946}
1947
1948#ifdef CONFIG_BLK_DEV_ZONED
1949static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1950 struct block_device *bdev, block_t blkstart, block_t blklen)
1951{
1952 sector_t sector, nr_sects;
1953 block_t lblkstart = blkstart;
1954 int devi = 0;
1955 u64 remainder = 0;
1956
1957 if (f2fs_is_multi_device(sbi)) {
1958 devi = f2fs_target_device_index(sbi, blkstart);
1959 if (blkstart < FDEV(devi).start_blk ||
1960 blkstart > FDEV(devi).end_blk) {
1961 f2fs_err(sbi, "Invalid block %x", blkstart);
1962 return -EIO;
1963 }
1964 blkstart -= FDEV(devi).start_blk;
1965 }
1966
1967 /* For sequential zones, reset the zone write pointer */
1968 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1969 sector = SECTOR_FROM_BLOCK(blkstart);
1970 nr_sects = SECTOR_FROM_BLOCK(blklen);
1971 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
1972
1973 if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
1974 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1975 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1976 blkstart, blklen);
1977 return -EIO;
1978 }
1979
1980 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) {
1981 unsigned int nofs_flags;
1982 int ret;
1983
1984 trace_f2fs_issue_reset_zone(bdev, blkstart);
1985 nofs_flags = memalloc_nofs_save();
1986 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1987 sector, nr_sects);
1988 memalloc_nofs_restore(nofs_flags);
1989 return ret;
1990 }
1991
1992 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen);
1993 return 0;
1994 }
1995
1996 /* For conventional zones, use regular discard if supported */
1997 __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1998 return 0;
1999}
2000#endif
2001
2002static int __issue_discard_async(struct f2fs_sb_info *sbi,
2003 struct block_device *bdev, block_t blkstart, block_t blklen)
2004{
2005#ifdef CONFIG_BLK_DEV_ZONED
2006 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
2007 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
2008#endif
2009 __queue_discard_cmd(sbi, bdev, blkstart, blklen);
2010 return 0;
2011}
2012
2013static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
2014 block_t blkstart, block_t blklen)
2015{
2016 sector_t start = blkstart, len = 0;
2017 struct block_device *bdev;
2018 struct seg_entry *se;
2019 unsigned int offset;
2020 block_t i;
2021 int err = 0;
2022
2023 bdev = f2fs_target_device(sbi, blkstart, NULL);
2024
2025 for (i = blkstart; i < blkstart + blklen; i++, len++) {
2026 if (i != start) {
2027 struct block_device *bdev2 =
2028 f2fs_target_device(sbi, i, NULL);
2029
2030 if (bdev2 != bdev) {
2031 err = __issue_discard_async(sbi, bdev,
2032 start, len);
2033 if (err)
2034 return err;
2035 bdev = bdev2;
2036 start = i;
2037 len = 0;
2038 }
2039 }
2040
2041 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
2042 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
2043
2044 if (f2fs_block_unit_discard(sbi) &&
2045 !f2fs_test_and_set_bit(offset, se->discard_map))
2046 sbi->discard_blks--;
2047 }
2048
2049 if (len)
2050 err = __issue_discard_async(sbi, bdev, start, len);
2051 return err;
2052}
2053
2054static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
2055 bool check_only)
2056{
2057 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2058 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
2059 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2060 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2061 unsigned long *discard_map = (unsigned long *)se->discard_map;
2062 unsigned long *dmap = SIT_I(sbi)->tmp_map;
2063 unsigned int start = 0, end = -1;
2064 bool force = (cpc->reason & CP_DISCARD);
2065 struct discard_entry *de = NULL;
2066 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
2067 int i;
2068
2069 if (se->valid_blocks == BLKS_PER_SEG(sbi) ||
2070 !f2fs_hw_support_discard(sbi) ||
2071 !f2fs_block_unit_discard(sbi))
2072 return false;
2073
2074 if (!force) {
2075 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
2076 SM_I(sbi)->dcc_info->nr_discards >=
2077 SM_I(sbi)->dcc_info->max_discards)
2078 return false;
2079 }
2080
2081 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
2082 for (i = 0; i < entries; i++)
2083 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
2084 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
2085
2086 while (force || SM_I(sbi)->dcc_info->nr_discards <=
2087 SM_I(sbi)->dcc_info->max_discards) {
2088 start = __find_rev_next_bit(dmap, BLKS_PER_SEG(sbi), end + 1);
2089 if (start >= BLKS_PER_SEG(sbi))
2090 break;
2091
2092 end = __find_rev_next_zero_bit(dmap,
2093 BLKS_PER_SEG(sbi), start + 1);
2094 if (force && start && end != BLKS_PER_SEG(sbi) &&
2095 (end - start) < cpc->trim_minlen)
2096 continue;
2097
2098 if (check_only)
2099 return true;
2100
2101 if (!de) {
2102 de = f2fs_kmem_cache_alloc(discard_entry_slab,
2103 GFP_F2FS_ZERO, true, NULL);
2104 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
2105 list_add_tail(&de->list, head);
2106 }
2107
2108 for (i = start; i < end; i++)
2109 __set_bit_le(i, (void *)de->discard_map);
2110
2111 SM_I(sbi)->dcc_info->nr_discards += end - start;
2112 }
2113 return false;
2114}
2115
2116static void release_discard_addr(struct discard_entry *entry)
2117{
2118 list_del(&entry->list);
2119 kmem_cache_free(discard_entry_slab, entry);
2120}
2121
2122void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2123{
2124 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2125 struct discard_entry *entry, *this;
2126
2127 /* drop caches */
2128 list_for_each_entry_safe(entry, this, head, list)
2129 release_discard_addr(entry);
2130}
2131
2132/*
2133 * Should call f2fs_clear_prefree_segments after checkpoint is done.
2134 */
2135static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2136{
2137 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2138 unsigned int segno;
2139
2140 mutex_lock(&dirty_i->seglist_lock);
2141 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2142 __set_test_and_free(sbi, segno, false);
2143 mutex_unlock(&dirty_i->seglist_lock);
2144}
2145
2146void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2147 struct cp_control *cpc)
2148{
2149 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2150 struct list_head *head = &dcc->entry_list;
2151 struct discard_entry *entry, *this;
2152 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2153 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2154 unsigned int start = 0, end = -1;
2155 unsigned int secno, start_segno;
2156 bool force = (cpc->reason & CP_DISCARD);
2157 bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2158 DISCARD_UNIT_SECTION;
2159
2160 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2161 section_alignment = true;
2162
2163 mutex_lock(&dirty_i->seglist_lock);
2164
2165 while (1) {
2166 int i;
2167
2168 if (section_alignment && end != -1)
2169 end--;
2170 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2171 if (start >= MAIN_SEGS(sbi))
2172 break;
2173 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2174 start + 1);
2175
2176 if (section_alignment) {
2177 start = rounddown(start, SEGS_PER_SEC(sbi));
2178 end = roundup(end, SEGS_PER_SEC(sbi));
2179 }
2180
2181 for (i = start; i < end; i++) {
2182 if (test_and_clear_bit(i, prefree_map))
2183 dirty_i->nr_dirty[PRE]--;
2184 }
2185
2186 if (!f2fs_realtime_discard_enable(sbi))
2187 continue;
2188
2189 if (force && start >= cpc->trim_start &&
2190 (end - 1) <= cpc->trim_end)
2191 continue;
2192
2193 /* Should cover 2MB zoned device for zone-based reset */
2194 if (!f2fs_sb_has_blkzoned(sbi) &&
2195 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2196 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2197 SEGS_TO_BLKS(sbi, end - start));
2198 continue;
2199 }
2200next:
2201 secno = GET_SEC_FROM_SEG(sbi, start);
2202 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2203 if (!IS_CURSEC(sbi, secno) &&
2204 !get_valid_blocks(sbi, start, true))
2205 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2206 BLKS_PER_SEC(sbi));
2207
2208 start = start_segno + SEGS_PER_SEC(sbi);
2209 if (start < end)
2210 goto next;
2211 else
2212 end = start - 1;
2213 }
2214 mutex_unlock(&dirty_i->seglist_lock);
2215
2216 if (!f2fs_block_unit_discard(sbi))
2217 goto wakeup;
2218
2219 /* send small discards */
2220 list_for_each_entry_safe(entry, this, head, list) {
2221 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2222 bool is_valid = test_bit_le(0, entry->discard_map);
2223
2224find_next:
2225 if (is_valid) {
2226 next_pos = find_next_zero_bit_le(entry->discard_map,
2227 BLKS_PER_SEG(sbi), cur_pos);
2228 len = next_pos - cur_pos;
2229
2230 if (f2fs_sb_has_blkzoned(sbi) ||
2231 (force && len < cpc->trim_minlen))
2232 goto skip;
2233
2234 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2235 len);
2236 total_len += len;
2237 } else {
2238 next_pos = find_next_bit_le(entry->discard_map,
2239 BLKS_PER_SEG(sbi), cur_pos);
2240 }
2241skip:
2242 cur_pos = next_pos;
2243 is_valid = !is_valid;
2244
2245 if (cur_pos < BLKS_PER_SEG(sbi))
2246 goto find_next;
2247
2248 release_discard_addr(entry);
2249 dcc->nr_discards -= total_len;
2250 }
2251
2252wakeup:
2253 wake_up_discard_thread(sbi, false);
2254}
2255
2256int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2257{
2258 dev_t dev = sbi->sb->s_bdev->bd_dev;
2259 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2260 int err = 0;
2261
2262 if (f2fs_sb_has_readonly(sbi)) {
2263 f2fs_info(sbi,
2264 "Skip to start discard thread for readonly image");
2265 return 0;
2266 }
2267
2268 if (!f2fs_realtime_discard_enable(sbi))
2269 return 0;
2270
2271 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2272 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2273 if (IS_ERR(dcc->f2fs_issue_discard)) {
2274 err = PTR_ERR(dcc->f2fs_issue_discard);
2275 dcc->f2fs_issue_discard = NULL;
2276 }
2277
2278 return err;
2279}
2280
2281static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2282{
2283 struct discard_cmd_control *dcc;
2284 int err = 0, i;
2285
2286 if (SM_I(sbi)->dcc_info) {
2287 dcc = SM_I(sbi)->dcc_info;
2288 goto init_thread;
2289 }
2290
2291 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2292 if (!dcc)
2293 return -ENOMEM;
2294
2295 dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2296 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2297 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2298 dcc->discard_io_aware = DPOLICY_IO_AWARE_ENABLE;
2299 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2300 dcc->discard_granularity = BLKS_PER_SEG(sbi);
2301 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2302 dcc->discard_granularity = BLKS_PER_SEC(sbi);
2303
2304 INIT_LIST_HEAD(&dcc->entry_list);
2305 for (i = 0; i < MAX_PLIST_NUM; i++)
2306 INIT_LIST_HEAD(&dcc->pend_list[i]);
2307 INIT_LIST_HEAD(&dcc->wait_list);
2308 INIT_LIST_HEAD(&dcc->fstrim_list);
2309 mutex_init(&dcc->cmd_lock);
2310 atomic_set(&dcc->issued_discard, 0);
2311 atomic_set(&dcc->queued_discard, 0);
2312 atomic_set(&dcc->discard_cmd_cnt, 0);
2313 dcc->nr_discards = 0;
2314 dcc->max_discards = SEGS_TO_BLKS(sbi, MAIN_SEGS(sbi));
2315 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2316 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2317 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2318 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2319 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2320 dcc->undiscard_blks = 0;
2321 dcc->next_pos = 0;
2322 dcc->root = RB_ROOT_CACHED;
2323 dcc->rbtree_check = false;
2324
2325 init_waitqueue_head(&dcc->discard_wait_queue);
2326 SM_I(sbi)->dcc_info = dcc;
2327init_thread:
2328 err = f2fs_start_discard_thread(sbi);
2329 if (err) {
2330 kfree(dcc);
2331 SM_I(sbi)->dcc_info = NULL;
2332 }
2333
2334 return err;
2335}
2336
2337static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2338{
2339 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2340
2341 if (!dcc)
2342 return;
2343
2344 f2fs_stop_discard_thread(sbi);
2345
2346 /*
2347 * Recovery can cache discard commands, so in error path of
2348 * fill_super(), it needs to give a chance to handle them.
2349 */
2350 f2fs_issue_discard_timeout(sbi);
2351
2352 kfree(dcc);
2353 SM_I(sbi)->dcc_info = NULL;
2354}
2355
2356static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2357{
2358 struct sit_info *sit_i = SIT_I(sbi);
2359
2360 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2361 sit_i->dirty_sentries++;
2362 return false;
2363 }
2364
2365 return true;
2366}
2367
2368static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2369 unsigned int segno, int modified)
2370{
2371 struct seg_entry *se = get_seg_entry(sbi, segno);
2372
2373 se->type = type;
2374 if (modified)
2375 __mark_sit_entry_dirty(sbi, segno);
2376}
2377
2378static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2379 block_t blkaddr)
2380{
2381 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2382
2383 if (segno == NULL_SEGNO)
2384 return 0;
2385 return get_seg_entry(sbi, segno)->mtime;
2386}
2387
2388static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2389 unsigned long long old_mtime)
2390{
2391 struct seg_entry *se;
2392 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2393 unsigned long long ctime = get_mtime(sbi, false);
2394 unsigned long long mtime = old_mtime ? old_mtime : ctime;
2395
2396 if (segno == NULL_SEGNO)
2397 return;
2398
2399 se = get_seg_entry(sbi, segno);
2400
2401 if (!se->mtime)
2402 se->mtime = mtime;
2403 else
2404 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2405 se->valid_blocks + 1);
2406
2407 if (ctime > SIT_I(sbi)->max_mtime)
2408 SIT_I(sbi)->max_mtime = ctime;
2409}
2410
2411static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2412{
2413 struct seg_entry *se;
2414 unsigned int segno, offset;
2415 long int new_vblocks;
2416 bool exist;
2417#ifdef CONFIG_F2FS_CHECK_FS
2418 bool mir_exist;
2419#endif
2420
2421 segno = GET_SEGNO(sbi, blkaddr);
2422 if (segno == NULL_SEGNO)
2423 return;
2424
2425 se = get_seg_entry(sbi, segno);
2426 new_vblocks = se->valid_blocks + del;
2427 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2428
2429 f2fs_bug_on(sbi, (new_vblocks < 0 ||
2430 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2431
2432 se->valid_blocks = new_vblocks;
2433
2434 /* Update valid block bitmap */
2435 if (del > 0) {
2436 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2437#ifdef CONFIG_F2FS_CHECK_FS
2438 mir_exist = f2fs_test_and_set_bit(offset,
2439 se->cur_valid_map_mir);
2440 if (unlikely(exist != mir_exist)) {
2441 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2442 blkaddr, exist);
2443 f2fs_bug_on(sbi, 1);
2444 }
2445#endif
2446 if (unlikely(exist)) {
2447 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2448 blkaddr);
2449 f2fs_bug_on(sbi, 1);
2450 se->valid_blocks--;
2451 del = 0;
2452 }
2453
2454 if (f2fs_block_unit_discard(sbi) &&
2455 !f2fs_test_and_set_bit(offset, se->discard_map))
2456 sbi->discard_blks--;
2457
2458 /*
2459 * SSR should never reuse block which is checkpointed
2460 * or newly invalidated.
2461 */
2462 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2463 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2464 se->ckpt_valid_blocks++;
2465 }
2466 } else {
2467 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2468#ifdef CONFIG_F2FS_CHECK_FS
2469 mir_exist = f2fs_test_and_clear_bit(offset,
2470 se->cur_valid_map_mir);
2471 if (unlikely(exist != mir_exist)) {
2472 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2473 blkaddr, exist);
2474 f2fs_bug_on(sbi, 1);
2475 }
2476#endif
2477 if (unlikely(!exist)) {
2478 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2479 blkaddr);
2480 f2fs_bug_on(sbi, 1);
2481 se->valid_blocks++;
2482 del = 0;
2483 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2484 /*
2485 * If checkpoints are off, we must not reuse data that
2486 * was used in the previous checkpoint. If it was used
2487 * before, we must track that to know how much space we
2488 * really have.
2489 */
2490 if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2491 spin_lock(&sbi->stat_lock);
2492 sbi->unusable_block_count++;
2493 spin_unlock(&sbi->stat_lock);
2494 }
2495 }
2496
2497 if (f2fs_block_unit_discard(sbi) &&
2498 f2fs_test_and_clear_bit(offset, se->discard_map))
2499 sbi->discard_blks++;
2500 }
2501 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2502 se->ckpt_valid_blocks += del;
2503
2504 __mark_sit_entry_dirty(sbi, segno);
2505
2506 /* update total number of valid blocks to be written in ckpt area */
2507 SIT_I(sbi)->written_valid_blocks += del;
2508
2509 if (__is_large_section(sbi))
2510 get_sec_entry(sbi, segno)->valid_blocks += del;
2511}
2512
2513void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2514{
2515 unsigned int segno = GET_SEGNO(sbi, addr);
2516 struct sit_info *sit_i = SIT_I(sbi);
2517
2518 f2fs_bug_on(sbi, addr == NULL_ADDR);
2519 if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2520 return;
2521
2522 f2fs_invalidate_internal_cache(sbi, addr);
2523
2524 /* add it into sit main buffer */
2525 down_write(&sit_i->sentry_lock);
2526
2527 update_segment_mtime(sbi, addr, 0);
2528 update_sit_entry(sbi, addr, -1);
2529
2530 /* add it into dirty seglist */
2531 locate_dirty_segment(sbi, segno);
2532
2533 up_write(&sit_i->sentry_lock);
2534}
2535
2536bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2537{
2538 struct sit_info *sit_i = SIT_I(sbi);
2539 unsigned int segno, offset;
2540 struct seg_entry *se;
2541 bool is_cp = false;
2542
2543 if (!__is_valid_data_blkaddr(blkaddr))
2544 return true;
2545
2546 down_read(&sit_i->sentry_lock);
2547
2548 segno = GET_SEGNO(sbi, blkaddr);
2549 se = get_seg_entry(sbi, segno);
2550 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2551
2552 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2553 is_cp = true;
2554
2555 up_read(&sit_i->sentry_lock);
2556
2557 return is_cp;
2558}
2559
2560static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2561{
2562 struct curseg_info *curseg = CURSEG_I(sbi, type);
2563
2564 if (sbi->ckpt->alloc_type[type] == SSR)
2565 return BLKS_PER_SEG(sbi);
2566 return curseg->next_blkoff;
2567}
2568
2569/*
2570 * Calculate the number of current summary pages for writing
2571 */
2572int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2573{
2574 int valid_sum_count = 0;
2575 int i, sum_in_page;
2576
2577 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2578 if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2579 valid_sum_count +=
2580 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2581 else
2582 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2583 }
2584
2585 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2586 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2587 if (valid_sum_count <= sum_in_page)
2588 return 1;
2589 else if ((valid_sum_count - sum_in_page) <=
2590 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2591 return 2;
2592 return 3;
2593}
2594
2595/*
2596 * Caller should put this summary page
2597 */
2598struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2599{
2600 if (unlikely(f2fs_cp_error(sbi)))
2601 return ERR_PTR(-EIO);
2602 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2603}
2604
2605void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2606 void *src, block_t blk_addr)
2607{
2608 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2609
2610 memcpy(page_address(page), src, PAGE_SIZE);
2611 set_page_dirty(page);
2612 f2fs_put_page(page, 1);
2613}
2614
2615static void write_sum_page(struct f2fs_sb_info *sbi,
2616 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2617{
2618 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2619}
2620
2621static void write_current_sum_page(struct f2fs_sb_info *sbi,
2622 int type, block_t blk_addr)
2623{
2624 struct curseg_info *curseg = CURSEG_I(sbi, type);
2625 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2626 struct f2fs_summary_block *src = curseg->sum_blk;
2627 struct f2fs_summary_block *dst;
2628
2629 dst = (struct f2fs_summary_block *)page_address(page);
2630 memset(dst, 0, PAGE_SIZE);
2631
2632 mutex_lock(&curseg->curseg_mutex);
2633
2634 down_read(&curseg->journal_rwsem);
2635 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2636 up_read(&curseg->journal_rwsem);
2637
2638 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2639 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2640
2641 mutex_unlock(&curseg->curseg_mutex);
2642
2643 set_page_dirty(page);
2644 f2fs_put_page(page, 1);
2645}
2646
2647static int is_next_segment_free(struct f2fs_sb_info *sbi,
2648 struct curseg_info *curseg, int type)
2649{
2650 unsigned int segno = curseg->segno + 1;
2651 struct free_segmap_info *free_i = FREE_I(sbi);
2652
2653 if (segno < MAIN_SEGS(sbi) && segno % SEGS_PER_SEC(sbi))
2654 return !test_bit(segno, free_i->free_segmap);
2655 return 0;
2656}
2657
2658/*
2659 * Find a new segment from the free segments bitmap to right order
2660 * This function should be returned with success, otherwise BUG
2661 */
2662static int get_new_segment(struct f2fs_sb_info *sbi,
2663 unsigned int *newseg, bool new_sec, bool pinning)
2664{
2665 struct free_segmap_info *free_i = FREE_I(sbi);
2666 unsigned int segno, secno, zoneno;
2667 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2668 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2669 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2670 bool init = true;
2671 int i;
2672 int ret = 0;
2673
2674 spin_lock(&free_i->segmap_lock);
2675
2676 if (time_to_inject(sbi, FAULT_NO_SEGMENT)) {
2677 ret = -ENOSPC;
2678 goto out_unlock;
2679 }
2680
2681 if (!new_sec && ((*newseg + 1) % SEGS_PER_SEC(sbi))) {
2682 segno = find_next_zero_bit(free_i->free_segmap,
2683 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2684 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2685 goto got_it;
2686 }
2687
2688 /*
2689 * If we format f2fs on zoned storage, let's try to get pinned sections
2690 * from beginning of the storage, which should be a conventional one.
2691 */
2692 if (f2fs_sb_has_blkzoned(sbi)) {
2693 segno = pinning ? 0 : max(first_zoned_segno(sbi), *newseg);
2694 hint = GET_SEC_FROM_SEG(sbi, segno);
2695 }
2696
2697find_other_zone:
2698 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2699 if (secno >= MAIN_SECS(sbi)) {
2700 secno = find_first_zero_bit(free_i->free_secmap,
2701 MAIN_SECS(sbi));
2702 if (secno >= MAIN_SECS(sbi)) {
2703 ret = -ENOSPC;
2704 goto out_unlock;
2705 }
2706 }
2707 segno = GET_SEG_FROM_SEC(sbi, secno);
2708 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2709
2710 /* give up on finding another zone */
2711 if (!init)
2712 goto got_it;
2713 if (sbi->secs_per_zone == 1)
2714 goto got_it;
2715 if (zoneno == old_zoneno)
2716 goto got_it;
2717 for (i = 0; i < NR_CURSEG_TYPE; i++)
2718 if (CURSEG_I(sbi, i)->zone == zoneno)
2719 break;
2720
2721 if (i < NR_CURSEG_TYPE) {
2722 /* zone is in user, try another */
2723 if (zoneno + 1 >= total_zones)
2724 hint = 0;
2725 else
2726 hint = (zoneno + 1) * sbi->secs_per_zone;
2727 init = false;
2728 goto find_other_zone;
2729 }
2730got_it:
2731 /* set it as dirty segment in free segmap */
2732 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2733
2734 /* no free section in conventional zone */
2735 if (new_sec && pinning &&
2736 !f2fs_valid_pinned_area(sbi, START_BLOCK(sbi, segno))) {
2737 ret = -EAGAIN;
2738 goto out_unlock;
2739 }
2740 __set_inuse(sbi, segno);
2741 *newseg = segno;
2742out_unlock:
2743 spin_unlock(&free_i->segmap_lock);
2744
2745 if (ret == -ENOSPC) {
2746 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_NO_SEGMENT);
2747 f2fs_bug_on(sbi, 1);
2748 }
2749 return ret;
2750}
2751
2752static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2753{
2754 struct curseg_info *curseg = CURSEG_I(sbi, type);
2755 struct summary_footer *sum_footer;
2756 unsigned short seg_type = curseg->seg_type;
2757
2758 /* only happen when get_new_segment() fails */
2759 if (curseg->next_segno == NULL_SEGNO)
2760 return;
2761
2762 curseg->inited = true;
2763 curseg->segno = curseg->next_segno;
2764 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2765 curseg->next_blkoff = 0;
2766 curseg->next_segno = NULL_SEGNO;
2767
2768 sum_footer = &(curseg->sum_blk->footer);
2769 memset(sum_footer, 0, sizeof(struct summary_footer));
2770
2771 sanity_check_seg_type(sbi, seg_type);
2772
2773 if (IS_DATASEG(seg_type))
2774 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2775 if (IS_NODESEG(seg_type))
2776 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2777 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2778}
2779
2780static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2781{
2782 struct curseg_info *curseg = CURSEG_I(sbi, type);
2783 unsigned short seg_type = curseg->seg_type;
2784
2785 sanity_check_seg_type(sbi, seg_type);
2786 if (f2fs_need_rand_seg(sbi))
2787 return get_random_u32_below(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi));
2788
2789 if (__is_large_section(sbi))
2790 return curseg->segno;
2791
2792 /* inmem log may not locate on any segment after mount */
2793 if (!curseg->inited)
2794 return 0;
2795
2796 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2797 return 0;
2798
2799 if (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))
2800 return 0;
2801
2802 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2803 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2804
2805 /* find segments from 0 to reuse freed segments */
2806 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2807 return 0;
2808
2809 return curseg->segno;
2810}
2811
2812/*
2813 * Allocate a current working segment.
2814 * This function always allocates a free segment in LFS manner.
2815 */
2816static int new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2817{
2818 struct curseg_info *curseg = CURSEG_I(sbi, type);
2819 unsigned int segno = curseg->segno;
2820 bool pinning = type == CURSEG_COLD_DATA_PINNED;
2821 int ret;
2822
2823 if (curseg->inited)
2824 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, segno));
2825
2826 segno = __get_next_segno(sbi, type);
2827 ret = get_new_segment(sbi, &segno, new_sec, pinning);
2828 if (ret) {
2829 if (ret == -ENOSPC)
2830 curseg->segno = NULL_SEGNO;
2831 return ret;
2832 }
2833
2834 curseg->next_segno = segno;
2835 reset_curseg(sbi, type, 1);
2836 curseg->alloc_type = LFS;
2837 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2838 curseg->fragment_remained_chunk =
2839 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2840 return 0;
2841}
2842
2843static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2844 int segno, block_t start)
2845{
2846 struct seg_entry *se = get_seg_entry(sbi, segno);
2847 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2848 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2849 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2850 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2851 int i;
2852
2853 for (i = 0; i < entries; i++)
2854 target_map[i] = ckpt_map[i] | cur_map[i];
2855
2856 return __find_rev_next_zero_bit(target_map, BLKS_PER_SEG(sbi), start);
2857}
2858
2859static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
2860 struct curseg_info *seg)
2861{
2862 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
2863}
2864
2865bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2866{
2867 return __next_free_blkoff(sbi, segno, 0) < BLKS_PER_SEG(sbi);
2868}
2869
2870/*
2871 * This function always allocates a used segment(from dirty seglist) by SSR
2872 * manner, so it should recover the existing segment information of valid blocks
2873 */
2874static int change_curseg(struct f2fs_sb_info *sbi, int type)
2875{
2876 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2877 struct curseg_info *curseg = CURSEG_I(sbi, type);
2878 unsigned int new_segno = curseg->next_segno;
2879 struct f2fs_summary_block *sum_node;
2880 struct page *sum_page;
2881
2882 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2883
2884 __set_test_and_inuse(sbi, new_segno);
2885
2886 mutex_lock(&dirty_i->seglist_lock);
2887 __remove_dirty_segment(sbi, new_segno, PRE);
2888 __remove_dirty_segment(sbi, new_segno, DIRTY);
2889 mutex_unlock(&dirty_i->seglist_lock);
2890
2891 reset_curseg(sbi, type, 1);
2892 curseg->alloc_type = SSR;
2893 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2894
2895 sum_page = f2fs_get_sum_page(sbi, new_segno);
2896 if (IS_ERR(sum_page)) {
2897 /* GC won't be able to use stale summary pages by cp_error */
2898 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2899 return PTR_ERR(sum_page);
2900 }
2901 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2902 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2903 f2fs_put_page(sum_page, 1);
2904 return 0;
2905}
2906
2907static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2908 int alloc_mode, unsigned long long age);
2909
2910static int get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2911 int target_type, int alloc_mode,
2912 unsigned long long age)
2913{
2914 struct curseg_info *curseg = CURSEG_I(sbi, type);
2915 int ret = 0;
2916
2917 curseg->seg_type = target_type;
2918
2919 if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2920 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2921
2922 curseg->seg_type = se->type;
2923 ret = change_curseg(sbi, type);
2924 } else {
2925 /* allocate cold segment by default */
2926 curseg->seg_type = CURSEG_COLD_DATA;
2927 ret = new_curseg(sbi, type, true);
2928 }
2929 stat_inc_seg_type(sbi, curseg);
2930 return ret;
2931}
2932
2933static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2934{
2935 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2936 int ret = 0;
2937
2938 if (!sbi->am.atgc_enabled)
2939 return 0;
2940
2941 f2fs_down_read(&SM_I(sbi)->curseg_lock);
2942
2943 mutex_lock(&curseg->curseg_mutex);
2944 down_write(&SIT_I(sbi)->sentry_lock);
2945
2946 ret = get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC,
2947 CURSEG_COLD_DATA, SSR, 0);
2948
2949 up_write(&SIT_I(sbi)->sentry_lock);
2950 mutex_unlock(&curseg->curseg_mutex);
2951
2952 f2fs_up_read(&SM_I(sbi)->curseg_lock);
2953 return ret;
2954}
2955int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2956{
2957 return __f2fs_init_atgc_curseg(sbi);
2958}
2959
2960static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2961{
2962 struct curseg_info *curseg = CURSEG_I(sbi, type);
2963
2964 mutex_lock(&curseg->curseg_mutex);
2965 if (!curseg->inited)
2966 goto out;
2967
2968 if (get_valid_blocks(sbi, curseg->segno, false)) {
2969 write_sum_page(sbi, curseg->sum_blk,
2970 GET_SUM_BLOCK(sbi, curseg->segno));
2971 } else {
2972 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2973 __set_test_and_free(sbi, curseg->segno, true);
2974 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2975 }
2976out:
2977 mutex_unlock(&curseg->curseg_mutex);
2978}
2979
2980void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2981{
2982 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2983
2984 if (sbi->am.atgc_enabled)
2985 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2986}
2987
2988static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2989{
2990 struct curseg_info *curseg = CURSEG_I(sbi, type);
2991
2992 mutex_lock(&curseg->curseg_mutex);
2993 if (!curseg->inited)
2994 goto out;
2995 if (get_valid_blocks(sbi, curseg->segno, false))
2996 goto out;
2997
2998 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2999 __set_test_and_inuse(sbi, curseg->segno);
3000 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
3001out:
3002 mutex_unlock(&curseg->curseg_mutex);
3003}
3004
3005void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
3006{
3007 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
3008
3009 if (sbi->am.atgc_enabled)
3010 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
3011}
3012
3013static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
3014 int alloc_mode, unsigned long long age)
3015{
3016 struct curseg_info *curseg = CURSEG_I(sbi, type);
3017 unsigned segno = NULL_SEGNO;
3018 unsigned short seg_type = curseg->seg_type;
3019 int i, cnt;
3020 bool reversed = false;
3021
3022 sanity_check_seg_type(sbi, seg_type);
3023
3024 /* f2fs_need_SSR() already forces to do this */
3025 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
3026 curseg->next_segno = segno;
3027 return 1;
3028 }
3029
3030 /* For node segments, let's do SSR more intensively */
3031 if (IS_NODESEG(seg_type)) {
3032 if (seg_type >= CURSEG_WARM_NODE) {
3033 reversed = true;
3034 i = CURSEG_COLD_NODE;
3035 } else {
3036 i = CURSEG_HOT_NODE;
3037 }
3038 cnt = NR_CURSEG_NODE_TYPE;
3039 } else {
3040 if (seg_type >= CURSEG_WARM_DATA) {
3041 reversed = true;
3042 i = CURSEG_COLD_DATA;
3043 } else {
3044 i = CURSEG_HOT_DATA;
3045 }
3046 cnt = NR_CURSEG_DATA_TYPE;
3047 }
3048
3049 for (; cnt-- > 0; reversed ? i-- : i++) {
3050 if (i == seg_type)
3051 continue;
3052 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
3053 curseg->next_segno = segno;
3054 return 1;
3055 }
3056 }
3057
3058 /* find valid_blocks=0 in dirty list */
3059 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
3060 segno = get_free_segment(sbi);
3061 if (segno != NULL_SEGNO) {
3062 curseg->next_segno = segno;
3063 return 1;
3064 }
3065 }
3066 return 0;
3067}
3068
3069static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
3070{
3071 struct curseg_info *curseg = CURSEG_I(sbi, type);
3072
3073 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
3074 curseg->seg_type == CURSEG_WARM_NODE)
3075 return true;
3076 if (curseg->alloc_type == LFS &&
3077 is_next_segment_free(sbi, curseg, type) &&
3078 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
3079 return true;
3080 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
3081 return true;
3082 return false;
3083}
3084
3085int f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
3086 unsigned int start, unsigned int end)
3087{
3088 struct curseg_info *curseg = CURSEG_I(sbi, type);
3089 unsigned int segno;
3090 int ret = 0;
3091
3092 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3093 mutex_lock(&curseg->curseg_mutex);
3094 down_write(&SIT_I(sbi)->sentry_lock);
3095
3096 segno = CURSEG_I(sbi, type)->segno;
3097 if (segno < start || segno > end)
3098 goto unlock;
3099
3100 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
3101 ret = change_curseg(sbi, type);
3102 else
3103 ret = new_curseg(sbi, type, true);
3104
3105 stat_inc_seg_type(sbi, curseg);
3106
3107 locate_dirty_segment(sbi, segno);
3108unlock:
3109 up_write(&SIT_I(sbi)->sentry_lock);
3110
3111 if (segno != curseg->segno)
3112 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3113 type, segno, curseg->segno);
3114
3115 mutex_unlock(&curseg->curseg_mutex);
3116 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3117 return ret;
3118}
3119
3120static int __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3121 bool new_sec, bool force)
3122{
3123 struct curseg_info *curseg = CURSEG_I(sbi, type);
3124 unsigned int old_segno;
3125 int err = 0;
3126
3127 if (type == CURSEG_COLD_DATA_PINNED && !curseg->inited)
3128 goto allocate;
3129
3130 if (!force && curseg->inited &&
3131 !curseg->next_blkoff &&
3132 !get_valid_blocks(sbi, curseg->segno, new_sec) &&
3133 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3134 return 0;
3135
3136allocate:
3137 old_segno = curseg->segno;
3138 err = new_curseg(sbi, type, true);
3139 if (err)
3140 return err;
3141 stat_inc_seg_type(sbi, curseg);
3142 locate_dirty_segment(sbi, old_segno);
3143 return 0;
3144}
3145
3146int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3147{
3148 int ret;
3149
3150 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3151 down_write(&SIT_I(sbi)->sentry_lock);
3152 ret = __allocate_new_segment(sbi, type, true, force);
3153 up_write(&SIT_I(sbi)->sentry_lock);
3154 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3155
3156 return ret;
3157}
3158
3159int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi)
3160{
3161 int err;
3162 bool gc_required = true;
3163
3164retry:
3165 f2fs_lock_op(sbi);
3166 err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3167 f2fs_unlock_op(sbi);
3168
3169 if (f2fs_sb_has_blkzoned(sbi) && err == -EAGAIN && gc_required) {
3170 f2fs_down_write(&sbi->gc_lock);
3171 err = f2fs_gc_range(sbi, 0, GET_SEGNO(sbi, FDEV(0).end_blk), true, 1);
3172 f2fs_up_write(&sbi->gc_lock);
3173
3174 gc_required = false;
3175 if (!err)
3176 goto retry;
3177 }
3178
3179 return err;
3180}
3181
3182int f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3183{
3184 int i;
3185 int err = 0;
3186
3187 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3188 down_write(&SIT_I(sbi)->sentry_lock);
3189 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3190 err += __allocate_new_segment(sbi, i, false, false);
3191 up_write(&SIT_I(sbi)->sentry_lock);
3192 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3193
3194 return err;
3195}
3196
3197bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3198 struct cp_control *cpc)
3199{
3200 __u64 trim_start = cpc->trim_start;
3201 bool has_candidate = false;
3202
3203 down_write(&SIT_I(sbi)->sentry_lock);
3204 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3205 if (add_discard_addrs(sbi, cpc, true)) {
3206 has_candidate = true;
3207 break;
3208 }
3209 }
3210 up_write(&SIT_I(sbi)->sentry_lock);
3211
3212 cpc->trim_start = trim_start;
3213 return has_candidate;
3214}
3215
3216static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3217 struct discard_policy *dpolicy,
3218 unsigned int start, unsigned int end)
3219{
3220 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3221 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3222 struct rb_node **insert_p = NULL, *insert_parent = NULL;
3223 struct discard_cmd *dc;
3224 struct blk_plug plug;
3225 int issued;
3226 unsigned int trimmed = 0;
3227
3228next:
3229 issued = 0;
3230
3231 mutex_lock(&dcc->cmd_lock);
3232 if (unlikely(dcc->rbtree_check))
3233 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3234
3235 dc = __lookup_discard_cmd_ret(&dcc->root, start,
3236 &prev_dc, &next_dc, &insert_p, &insert_parent);
3237 if (!dc)
3238 dc = next_dc;
3239
3240 blk_start_plug(&plug);
3241
3242 while (dc && dc->di.lstart <= end) {
3243 struct rb_node *node;
3244 int err = 0;
3245
3246 if (dc->di.len < dpolicy->granularity)
3247 goto skip;
3248
3249 if (dc->state != D_PREP) {
3250 list_move_tail(&dc->list, &dcc->fstrim_list);
3251 goto skip;
3252 }
3253
3254 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3255
3256 if (issued >= dpolicy->max_requests) {
3257 start = dc->di.lstart + dc->di.len;
3258
3259 if (err)
3260 __remove_discard_cmd(sbi, dc);
3261
3262 blk_finish_plug(&plug);
3263 mutex_unlock(&dcc->cmd_lock);
3264 trimmed += __wait_all_discard_cmd(sbi, NULL);
3265 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3266 goto next;
3267 }
3268skip:
3269 node = rb_next(&dc->rb_node);
3270 if (err)
3271 __remove_discard_cmd(sbi, dc);
3272 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3273
3274 if (fatal_signal_pending(current))
3275 break;
3276 }
3277
3278 blk_finish_plug(&plug);
3279 mutex_unlock(&dcc->cmd_lock);
3280
3281 return trimmed;
3282}
3283
3284int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3285{
3286 __u64 start = F2FS_BYTES_TO_BLK(range->start);
3287 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3288 unsigned int start_segno, end_segno;
3289 block_t start_block, end_block;
3290 struct cp_control cpc;
3291 struct discard_policy dpolicy;
3292 unsigned long long trimmed = 0;
3293 int err = 0;
3294 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3295
3296 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3297 return -EINVAL;
3298
3299 if (end < MAIN_BLKADDR(sbi))
3300 goto out;
3301
3302 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3303 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3304 return -EFSCORRUPTED;
3305 }
3306
3307 /* start/end segment number in main_area */
3308 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3309 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3310 GET_SEGNO(sbi, end);
3311 if (need_align) {
3312 start_segno = rounddown(start_segno, SEGS_PER_SEC(sbi));
3313 end_segno = roundup(end_segno + 1, SEGS_PER_SEC(sbi)) - 1;
3314 }
3315
3316 cpc.reason = CP_DISCARD;
3317 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3318 cpc.trim_start = start_segno;
3319 cpc.trim_end = end_segno;
3320
3321 if (sbi->discard_blks == 0)
3322 goto out;
3323
3324 f2fs_down_write(&sbi->gc_lock);
3325 stat_inc_cp_call_count(sbi, TOTAL_CALL);
3326 err = f2fs_write_checkpoint(sbi, &cpc);
3327 f2fs_up_write(&sbi->gc_lock);
3328 if (err)
3329 goto out;
3330
3331 /*
3332 * We filed discard candidates, but actually we don't need to wait for
3333 * all of them, since they'll be issued in idle time along with runtime
3334 * discard option. User configuration looks like using runtime discard
3335 * or periodic fstrim instead of it.
3336 */
3337 if (f2fs_realtime_discard_enable(sbi))
3338 goto out;
3339
3340 start_block = START_BLOCK(sbi, start_segno);
3341 end_block = START_BLOCK(sbi, end_segno + 1);
3342
3343 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3344 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3345 start_block, end_block);
3346
3347 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3348 start_block, end_block);
3349out:
3350 if (!err)
3351 range->len = F2FS_BLK_TO_BYTES(trimmed);
3352 return err;
3353}
3354
3355int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3356{
3357 switch (hint) {
3358 case WRITE_LIFE_SHORT:
3359 return CURSEG_HOT_DATA;
3360 case WRITE_LIFE_EXTREME:
3361 return CURSEG_COLD_DATA;
3362 default:
3363 return CURSEG_WARM_DATA;
3364 }
3365}
3366
3367static int __get_segment_type_2(struct f2fs_io_info *fio)
3368{
3369 if (fio->type == DATA)
3370 return CURSEG_HOT_DATA;
3371 else
3372 return CURSEG_HOT_NODE;
3373}
3374
3375static int __get_segment_type_4(struct f2fs_io_info *fio)
3376{
3377 if (fio->type == DATA) {
3378 struct inode *inode = fio->page->mapping->host;
3379
3380 if (S_ISDIR(inode->i_mode))
3381 return CURSEG_HOT_DATA;
3382 else
3383 return CURSEG_COLD_DATA;
3384 } else {
3385 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3386 return CURSEG_WARM_NODE;
3387 else
3388 return CURSEG_COLD_NODE;
3389 }
3390}
3391
3392static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3393{
3394 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3395 struct extent_info ei = {};
3396
3397 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3398 if (!ei.age)
3399 return NO_CHECK_TYPE;
3400 if (ei.age <= sbi->hot_data_age_threshold)
3401 return CURSEG_HOT_DATA;
3402 if (ei.age <= sbi->warm_data_age_threshold)
3403 return CURSEG_WARM_DATA;
3404 return CURSEG_COLD_DATA;
3405 }
3406 return NO_CHECK_TYPE;
3407}
3408
3409static int __get_segment_type_6(struct f2fs_io_info *fio)
3410{
3411 if (fio->type == DATA) {
3412 struct inode *inode = fio->page->mapping->host;
3413 int type;
3414
3415 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3416 return CURSEG_COLD_DATA_PINNED;
3417
3418 if (page_private_gcing(fio->page)) {
3419 if (fio->sbi->am.atgc_enabled &&
3420 (fio->io_type == FS_DATA_IO) &&
3421 (fio->sbi->gc_mode != GC_URGENT_HIGH))
3422 return CURSEG_ALL_DATA_ATGC;
3423 else
3424 return CURSEG_COLD_DATA;
3425 }
3426 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3427 return CURSEG_COLD_DATA;
3428
3429 type = __get_age_segment_type(inode, fio->page->index);
3430 if (type != NO_CHECK_TYPE)
3431 return type;
3432
3433 if (file_is_hot(inode) ||
3434 is_inode_flag_set(inode, FI_HOT_DATA) ||
3435 f2fs_is_cow_file(inode))
3436 return CURSEG_HOT_DATA;
3437 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3438 } else {
3439 if (IS_DNODE(fio->page))
3440 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3441 CURSEG_HOT_NODE;
3442 return CURSEG_COLD_NODE;
3443 }
3444}
3445
3446static int __get_segment_type(struct f2fs_io_info *fio)
3447{
3448 int type = 0;
3449
3450 switch (F2FS_OPTION(fio->sbi).active_logs) {
3451 case 2:
3452 type = __get_segment_type_2(fio);
3453 break;
3454 case 4:
3455 type = __get_segment_type_4(fio);
3456 break;
3457 case 6:
3458 type = __get_segment_type_6(fio);
3459 break;
3460 default:
3461 f2fs_bug_on(fio->sbi, true);
3462 }
3463
3464 if (IS_HOT(type))
3465 fio->temp = HOT;
3466 else if (IS_WARM(type))
3467 fio->temp = WARM;
3468 else
3469 fio->temp = COLD;
3470 return type;
3471}
3472
3473static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3474 struct curseg_info *seg)
3475{
3476 /* To allocate block chunks in different sizes, use random number */
3477 if (--seg->fragment_remained_chunk > 0)
3478 return;
3479
3480 seg->fragment_remained_chunk =
3481 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3482 seg->next_blkoff +=
3483 get_random_u32_inclusive(1, sbi->max_fragment_hole);
3484}
3485
3486static void reset_curseg_fields(struct curseg_info *curseg)
3487{
3488 curseg->inited = false;
3489 curseg->segno = NULL_SEGNO;
3490 curseg->next_segno = 0;
3491}
3492
3493int f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3494 block_t old_blkaddr, block_t *new_blkaddr,
3495 struct f2fs_summary *sum, int type,
3496 struct f2fs_io_info *fio)
3497{
3498 struct sit_info *sit_i = SIT_I(sbi);
3499 struct curseg_info *curseg = CURSEG_I(sbi, type);
3500 unsigned long long old_mtime;
3501 bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3502 struct seg_entry *se = NULL;
3503 bool segment_full = false;
3504 int ret = 0;
3505
3506 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3507
3508 mutex_lock(&curseg->curseg_mutex);
3509 down_write(&sit_i->sentry_lock);
3510
3511 if (curseg->segno == NULL_SEGNO) {
3512 ret = -ENOSPC;
3513 goto out_err;
3514 }
3515
3516 if (from_gc) {
3517 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3518 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3519 sanity_check_seg_type(sbi, se->type);
3520 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3521 }
3522 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3523
3524 f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi));
3525
3526 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3527
3528 curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3529 if (curseg->alloc_type == SSR) {
3530 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3531 } else {
3532 curseg->next_blkoff++;
3533 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3534 f2fs_randomize_chunk(sbi, curseg);
3535 }
3536 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3537 segment_full = true;
3538 stat_inc_block_count(sbi, curseg);
3539
3540 if (from_gc) {
3541 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3542 } else {
3543 update_segment_mtime(sbi, old_blkaddr, 0);
3544 old_mtime = 0;
3545 }
3546 update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3547
3548 /*
3549 * SIT information should be updated before segment allocation,
3550 * since SSR needs latest valid block information.
3551 */
3552 update_sit_entry(sbi, *new_blkaddr, 1);
3553 update_sit_entry(sbi, old_blkaddr, -1);
3554
3555 /*
3556 * If the current segment is full, flush it out and replace it with a
3557 * new segment.
3558 */
3559 if (segment_full) {
3560 if (type == CURSEG_COLD_DATA_PINNED &&
3561 !((curseg->segno + 1) % sbi->segs_per_sec)) {
3562 write_sum_page(sbi, curseg->sum_blk,
3563 GET_SUM_BLOCK(sbi, curseg->segno));
3564 reset_curseg_fields(curseg);
3565 goto skip_new_segment;
3566 }
3567
3568 if (from_gc) {
3569 ret = get_atssr_segment(sbi, type, se->type,
3570 AT_SSR, se->mtime);
3571 } else {
3572 if (need_new_seg(sbi, type))
3573 ret = new_curseg(sbi, type, false);
3574 else
3575 ret = change_curseg(sbi, type);
3576 stat_inc_seg_type(sbi, curseg);
3577 }
3578
3579 if (ret)
3580 goto out_err;
3581 }
3582
3583skip_new_segment:
3584 /*
3585 * segment dirty status should be updated after segment allocation,
3586 * so we just need to update status only one time after previous
3587 * segment being closed.
3588 */
3589 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3590 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3591
3592 if (IS_DATASEG(curseg->seg_type))
3593 atomic64_inc(&sbi->allocated_data_blocks);
3594
3595 up_write(&sit_i->sentry_lock);
3596
3597 if (page && IS_NODESEG(curseg->seg_type)) {
3598 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3599
3600 f2fs_inode_chksum_set(sbi, page);
3601 }
3602
3603 if (fio) {
3604 struct f2fs_bio_info *io;
3605
3606 INIT_LIST_HEAD(&fio->list);
3607 fio->in_list = 1;
3608 io = sbi->write_io[fio->type] + fio->temp;
3609 spin_lock(&io->io_lock);
3610 list_add_tail(&fio->list, &io->io_list);
3611 spin_unlock(&io->io_lock);
3612 }
3613
3614 mutex_unlock(&curseg->curseg_mutex);
3615 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3616 return 0;
3617out_err:
3618 *new_blkaddr = NULL_ADDR;
3619 up_write(&sit_i->sentry_lock);
3620 mutex_unlock(&curseg->curseg_mutex);
3621 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3622 return ret;
3623
3624}
3625
3626void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3627 block_t blkaddr, unsigned int blkcnt)
3628{
3629 if (!f2fs_is_multi_device(sbi))
3630 return;
3631
3632 while (1) {
3633 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3634 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3635
3636 /* update device state for fsync */
3637 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3638
3639 /* update device state for checkpoint */
3640 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3641 spin_lock(&sbi->dev_lock);
3642 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3643 spin_unlock(&sbi->dev_lock);
3644 }
3645
3646 if (blkcnt <= blks)
3647 break;
3648 blkcnt -= blks;
3649 blkaddr += blks;
3650 }
3651}
3652
3653static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3654{
3655 int type = __get_segment_type(fio);
3656 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3657
3658 if (keep_order)
3659 f2fs_down_read(&fio->sbi->io_order_lock);
3660
3661 if (f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3662 &fio->new_blkaddr, sum, type, fio)) {
3663 if (fscrypt_inode_uses_fs_layer_crypto(fio->page->mapping->host))
3664 fscrypt_finalize_bounce_page(&fio->encrypted_page);
3665 if (PageWriteback(fio->page))
3666 end_page_writeback(fio->page);
3667 if (f2fs_in_warm_node_list(fio->sbi, fio->page))
3668 f2fs_del_fsync_node_entry(fio->sbi, fio->page);
3669 goto out;
3670 }
3671 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3672 f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr);
3673
3674 /* writeout dirty page into bdev */
3675 f2fs_submit_page_write(fio);
3676
3677 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3678out:
3679 if (keep_order)
3680 f2fs_up_read(&fio->sbi->io_order_lock);
3681}
3682
3683void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3684 enum iostat_type io_type)
3685{
3686 struct f2fs_io_info fio = {
3687 .sbi = sbi,
3688 .type = META,
3689 .temp = HOT,
3690 .op = REQ_OP_WRITE,
3691 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3692 .old_blkaddr = page->index,
3693 .new_blkaddr = page->index,
3694 .page = page,
3695 .encrypted_page = NULL,
3696 .in_list = 0,
3697 };
3698
3699 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3700 fio.op_flags &= ~REQ_META;
3701
3702 set_page_writeback(page);
3703 f2fs_submit_page_write(&fio);
3704
3705 stat_inc_meta_count(sbi, page->index);
3706 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3707}
3708
3709void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3710{
3711 struct f2fs_summary sum;
3712
3713 set_summary(&sum, nid, 0, 0);
3714 do_write_page(&sum, fio);
3715
3716 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3717}
3718
3719void f2fs_outplace_write_data(struct dnode_of_data *dn,
3720 struct f2fs_io_info *fio)
3721{
3722 struct f2fs_sb_info *sbi = fio->sbi;
3723 struct f2fs_summary sum;
3724
3725 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3726 if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
3727 f2fs_update_age_extent_cache(dn);
3728 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3729 do_write_page(&sum, fio);
3730 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3731
3732 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3733}
3734
3735int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3736{
3737 int err;
3738 struct f2fs_sb_info *sbi = fio->sbi;
3739 unsigned int segno;
3740
3741 fio->new_blkaddr = fio->old_blkaddr;
3742 /* i/o temperature is needed for passing down write hints */
3743 __get_segment_type(fio);
3744
3745 segno = GET_SEGNO(sbi, fio->new_blkaddr);
3746
3747 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3748 set_sbi_flag(sbi, SBI_NEED_FSCK);
3749 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3750 __func__, segno);
3751 err = -EFSCORRUPTED;
3752 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3753 goto drop_bio;
3754 }
3755
3756 if (f2fs_cp_error(sbi)) {
3757 err = -EIO;
3758 goto drop_bio;
3759 }
3760
3761 if (fio->post_read)
3762 f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1);
3763
3764 stat_inc_inplace_blocks(fio->sbi);
3765
3766 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi))
3767 err = f2fs_merge_page_bio(fio);
3768 else
3769 err = f2fs_submit_page_bio(fio);
3770 if (!err) {
3771 f2fs_update_device_state(fio->sbi, fio->ino,
3772 fio->new_blkaddr, 1);
3773 f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3774 fio->io_type, F2FS_BLKSIZE);
3775 }
3776
3777 return err;
3778drop_bio:
3779 if (fio->bio && *(fio->bio)) {
3780 struct bio *bio = *(fio->bio);
3781
3782 bio->bi_status = BLK_STS_IOERR;
3783 bio_endio(bio);
3784 *(fio->bio) = NULL;
3785 }
3786 return err;
3787}
3788
3789static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3790 unsigned int segno)
3791{
3792 int i;
3793
3794 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3795 if (CURSEG_I(sbi, i)->segno == segno)
3796 break;
3797 }
3798 return i;
3799}
3800
3801void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3802 block_t old_blkaddr, block_t new_blkaddr,
3803 bool recover_curseg, bool recover_newaddr,
3804 bool from_gc)
3805{
3806 struct sit_info *sit_i = SIT_I(sbi);
3807 struct curseg_info *curseg;
3808 unsigned int segno, old_cursegno;
3809 struct seg_entry *se;
3810 int type;
3811 unsigned short old_blkoff;
3812 unsigned char old_alloc_type;
3813
3814 segno = GET_SEGNO(sbi, new_blkaddr);
3815 se = get_seg_entry(sbi, segno);
3816 type = se->type;
3817
3818 f2fs_down_write(&SM_I(sbi)->curseg_lock);
3819
3820 if (!recover_curseg) {
3821 /* for recovery flow */
3822 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3823 if (old_blkaddr == NULL_ADDR)
3824 type = CURSEG_COLD_DATA;
3825 else
3826 type = CURSEG_WARM_DATA;
3827 }
3828 } else {
3829 if (IS_CURSEG(sbi, segno)) {
3830 /* se->type is volatile as SSR allocation */
3831 type = __f2fs_get_curseg(sbi, segno);
3832 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3833 } else {
3834 type = CURSEG_WARM_DATA;
3835 }
3836 }
3837
3838 f2fs_bug_on(sbi, !IS_DATASEG(type));
3839 curseg = CURSEG_I(sbi, type);
3840
3841 mutex_lock(&curseg->curseg_mutex);
3842 down_write(&sit_i->sentry_lock);
3843
3844 old_cursegno = curseg->segno;
3845 old_blkoff = curseg->next_blkoff;
3846 old_alloc_type = curseg->alloc_type;
3847
3848 /* change the current segment */
3849 if (segno != curseg->segno) {
3850 curseg->next_segno = segno;
3851 if (change_curseg(sbi, type))
3852 goto out_unlock;
3853 }
3854
3855 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3856 curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3857
3858 if (!recover_curseg || recover_newaddr) {
3859 if (!from_gc)
3860 update_segment_mtime(sbi, new_blkaddr, 0);
3861 update_sit_entry(sbi, new_blkaddr, 1);
3862 }
3863 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3864 f2fs_invalidate_internal_cache(sbi, old_blkaddr);
3865 if (!from_gc)
3866 update_segment_mtime(sbi, old_blkaddr, 0);
3867 update_sit_entry(sbi, old_blkaddr, -1);
3868 }
3869
3870 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3871 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3872
3873 locate_dirty_segment(sbi, old_cursegno);
3874
3875 if (recover_curseg) {
3876 if (old_cursegno != curseg->segno) {
3877 curseg->next_segno = old_cursegno;
3878 if (change_curseg(sbi, type))
3879 goto out_unlock;
3880 }
3881 curseg->next_blkoff = old_blkoff;
3882 curseg->alloc_type = old_alloc_type;
3883 }
3884
3885out_unlock:
3886 up_write(&sit_i->sentry_lock);
3887 mutex_unlock(&curseg->curseg_mutex);
3888 f2fs_up_write(&SM_I(sbi)->curseg_lock);
3889}
3890
3891void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3892 block_t old_addr, block_t new_addr,
3893 unsigned char version, bool recover_curseg,
3894 bool recover_newaddr)
3895{
3896 struct f2fs_summary sum;
3897
3898 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3899
3900 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3901 recover_curseg, recover_newaddr, false);
3902
3903 f2fs_update_data_blkaddr(dn, new_addr);
3904}
3905
3906void f2fs_wait_on_page_writeback(struct page *page,
3907 enum page_type type, bool ordered, bool locked)
3908{
3909 if (PageWriteback(page)) {
3910 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3911
3912 /* submit cached LFS IO */
3913 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3914 /* submit cached IPU IO */
3915 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3916 if (ordered) {
3917 wait_on_page_writeback(page);
3918 f2fs_bug_on(sbi, locked && PageWriteback(page));
3919 } else {
3920 wait_for_stable_page(page);
3921 }
3922 }
3923}
3924
3925void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3926{
3927 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3928 struct page *cpage;
3929
3930 if (!f2fs_post_read_required(inode))
3931 return;
3932
3933 if (!__is_valid_data_blkaddr(blkaddr))
3934 return;
3935
3936 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3937 if (cpage) {
3938 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3939 f2fs_put_page(cpage, 1);
3940 }
3941}
3942
3943void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3944 block_t len)
3945{
3946 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3947 block_t i;
3948
3949 if (!f2fs_post_read_required(inode))
3950 return;
3951
3952 for (i = 0; i < len; i++)
3953 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3954
3955 f2fs_truncate_meta_inode_pages(sbi, blkaddr, len);
3956}
3957
3958static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3959{
3960 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3961 struct curseg_info *seg_i;
3962 unsigned char *kaddr;
3963 struct page *page;
3964 block_t start;
3965 int i, j, offset;
3966
3967 start = start_sum_block(sbi);
3968
3969 page = f2fs_get_meta_page(sbi, start++);
3970 if (IS_ERR(page))
3971 return PTR_ERR(page);
3972 kaddr = (unsigned char *)page_address(page);
3973
3974 /* Step 1: restore nat cache */
3975 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3976 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3977
3978 /* Step 2: restore sit cache */
3979 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3980 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3981 offset = 2 * SUM_JOURNAL_SIZE;
3982
3983 /* Step 3: restore summary entries */
3984 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3985 unsigned short blk_off;
3986 unsigned int segno;
3987
3988 seg_i = CURSEG_I(sbi, i);
3989 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3990 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3991 seg_i->next_segno = segno;
3992 reset_curseg(sbi, i, 0);
3993 seg_i->alloc_type = ckpt->alloc_type[i];
3994 seg_i->next_blkoff = blk_off;
3995
3996 if (seg_i->alloc_type == SSR)
3997 blk_off = BLKS_PER_SEG(sbi);
3998
3999 for (j = 0; j < blk_off; j++) {
4000 struct f2fs_summary *s;
4001
4002 s = (struct f2fs_summary *)(kaddr + offset);
4003 seg_i->sum_blk->entries[j] = *s;
4004 offset += SUMMARY_SIZE;
4005 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
4006 SUM_FOOTER_SIZE)
4007 continue;
4008
4009 f2fs_put_page(page, 1);
4010 page = NULL;
4011
4012 page = f2fs_get_meta_page(sbi, start++);
4013 if (IS_ERR(page))
4014 return PTR_ERR(page);
4015 kaddr = (unsigned char *)page_address(page);
4016 offset = 0;
4017 }
4018 }
4019 f2fs_put_page(page, 1);
4020 return 0;
4021}
4022
4023static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
4024{
4025 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4026 struct f2fs_summary_block *sum;
4027 struct curseg_info *curseg;
4028 struct page *new;
4029 unsigned short blk_off;
4030 unsigned int segno = 0;
4031 block_t blk_addr = 0;
4032 int err = 0;
4033
4034 /* get segment number and block addr */
4035 if (IS_DATASEG(type)) {
4036 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
4037 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
4038 CURSEG_HOT_DATA]);
4039 if (__exist_node_summaries(sbi))
4040 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
4041 else
4042 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
4043 } else {
4044 segno = le32_to_cpu(ckpt->cur_node_segno[type -
4045 CURSEG_HOT_NODE]);
4046 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
4047 CURSEG_HOT_NODE]);
4048 if (__exist_node_summaries(sbi))
4049 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
4050 type - CURSEG_HOT_NODE);
4051 else
4052 blk_addr = GET_SUM_BLOCK(sbi, segno);
4053 }
4054
4055 new = f2fs_get_meta_page(sbi, blk_addr);
4056 if (IS_ERR(new))
4057 return PTR_ERR(new);
4058 sum = (struct f2fs_summary_block *)page_address(new);
4059
4060 if (IS_NODESEG(type)) {
4061 if (__exist_node_summaries(sbi)) {
4062 struct f2fs_summary *ns = &sum->entries[0];
4063 int i;
4064
4065 for (i = 0; i < BLKS_PER_SEG(sbi); i++, ns++) {
4066 ns->version = 0;
4067 ns->ofs_in_node = 0;
4068 }
4069 } else {
4070 err = f2fs_restore_node_summary(sbi, segno, sum);
4071 if (err)
4072 goto out;
4073 }
4074 }
4075
4076 /* set uncompleted segment to curseg */
4077 curseg = CURSEG_I(sbi, type);
4078 mutex_lock(&curseg->curseg_mutex);
4079
4080 /* update journal info */
4081 down_write(&curseg->journal_rwsem);
4082 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
4083 up_write(&curseg->journal_rwsem);
4084
4085 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
4086 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
4087 curseg->next_segno = segno;
4088 reset_curseg(sbi, type, 0);
4089 curseg->alloc_type = ckpt->alloc_type[type];
4090 curseg->next_blkoff = blk_off;
4091 mutex_unlock(&curseg->curseg_mutex);
4092out:
4093 f2fs_put_page(new, 1);
4094 return err;
4095}
4096
4097static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
4098{
4099 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
4100 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
4101 int type = CURSEG_HOT_DATA;
4102 int err;
4103
4104 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4105 int npages = f2fs_npages_for_summary_flush(sbi, true);
4106
4107 if (npages >= 2)
4108 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
4109 META_CP, true);
4110
4111 /* restore for compacted data summary */
4112 err = read_compacted_summaries(sbi);
4113 if (err)
4114 return err;
4115 type = CURSEG_HOT_NODE;
4116 }
4117
4118 if (__exist_node_summaries(sbi))
4119 f2fs_ra_meta_pages(sbi,
4120 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4121 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4122
4123 for (; type <= CURSEG_COLD_NODE; type++) {
4124 err = read_normal_summaries(sbi, type);
4125 if (err)
4126 return err;
4127 }
4128
4129 /* sanity check for summary blocks */
4130 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4131 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4132 f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4133 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4134 return -EINVAL;
4135 }
4136
4137 return 0;
4138}
4139
4140static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4141{
4142 struct page *page;
4143 unsigned char *kaddr;
4144 struct f2fs_summary *summary;
4145 struct curseg_info *seg_i;
4146 int written_size = 0;
4147 int i, j;
4148
4149 page = f2fs_grab_meta_page(sbi, blkaddr++);
4150 kaddr = (unsigned char *)page_address(page);
4151 memset(kaddr, 0, PAGE_SIZE);
4152
4153 /* Step 1: write nat cache */
4154 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4155 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4156 written_size += SUM_JOURNAL_SIZE;
4157
4158 /* Step 2: write sit cache */
4159 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4160 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4161 written_size += SUM_JOURNAL_SIZE;
4162
4163 /* Step 3: write summary entries */
4164 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4165 seg_i = CURSEG_I(sbi, i);
4166 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) {
4167 if (!page) {
4168 page = f2fs_grab_meta_page(sbi, blkaddr++);
4169 kaddr = (unsigned char *)page_address(page);
4170 memset(kaddr, 0, PAGE_SIZE);
4171 written_size = 0;
4172 }
4173 summary = (struct f2fs_summary *)(kaddr + written_size);
4174 *summary = seg_i->sum_blk->entries[j];
4175 written_size += SUMMARY_SIZE;
4176
4177 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4178 SUM_FOOTER_SIZE)
4179 continue;
4180
4181 set_page_dirty(page);
4182 f2fs_put_page(page, 1);
4183 page = NULL;
4184 }
4185 }
4186 if (page) {
4187 set_page_dirty(page);
4188 f2fs_put_page(page, 1);
4189 }
4190}
4191
4192static void write_normal_summaries(struct f2fs_sb_info *sbi,
4193 block_t blkaddr, int type)
4194{
4195 int i, end;
4196
4197 if (IS_DATASEG(type))
4198 end = type + NR_CURSEG_DATA_TYPE;
4199 else
4200 end = type + NR_CURSEG_NODE_TYPE;
4201
4202 for (i = type; i < end; i++)
4203 write_current_sum_page(sbi, i, blkaddr + (i - type));
4204}
4205
4206void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4207{
4208 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4209 write_compacted_summaries(sbi, start_blk);
4210 else
4211 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4212}
4213
4214void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4215{
4216 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4217}
4218
4219int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4220 unsigned int val, int alloc)
4221{
4222 int i;
4223
4224 if (type == NAT_JOURNAL) {
4225 for (i = 0; i < nats_in_cursum(journal); i++) {
4226 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4227 return i;
4228 }
4229 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4230 return update_nats_in_cursum(journal, 1);
4231 } else if (type == SIT_JOURNAL) {
4232 for (i = 0; i < sits_in_cursum(journal); i++)
4233 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4234 return i;
4235 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4236 return update_sits_in_cursum(journal, 1);
4237 }
4238 return -1;
4239}
4240
4241static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4242 unsigned int segno)
4243{
4244 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4245}
4246
4247static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4248 unsigned int start)
4249{
4250 struct sit_info *sit_i = SIT_I(sbi);
4251 struct page *page;
4252 pgoff_t src_off, dst_off;
4253
4254 src_off = current_sit_addr(sbi, start);
4255 dst_off = next_sit_addr(sbi, src_off);
4256
4257 page = f2fs_grab_meta_page(sbi, dst_off);
4258 seg_info_to_sit_page(sbi, page, start);
4259
4260 set_page_dirty(page);
4261 set_to_next_sit(sit_i, start);
4262
4263 return page;
4264}
4265
4266static struct sit_entry_set *grab_sit_entry_set(void)
4267{
4268 struct sit_entry_set *ses =
4269 f2fs_kmem_cache_alloc(sit_entry_set_slab,
4270 GFP_NOFS, true, NULL);
4271
4272 ses->entry_cnt = 0;
4273 INIT_LIST_HEAD(&ses->set_list);
4274 return ses;
4275}
4276
4277static void release_sit_entry_set(struct sit_entry_set *ses)
4278{
4279 list_del(&ses->set_list);
4280 kmem_cache_free(sit_entry_set_slab, ses);
4281}
4282
4283static void adjust_sit_entry_set(struct sit_entry_set *ses,
4284 struct list_head *head)
4285{
4286 struct sit_entry_set *next = ses;
4287
4288 if (list_is_last(&ses->set_list, head))
4289 return;
4290
4291 list_for_each_entry_continue(next, head, set_list)
4292 if (ses->entry_cnt <= next->entry_cnt) {
4293 list_move_tail(&ses->set_list, &next->set_list);
4294 return;
4295 }
4296
4297 list_move_tail(&ses->set_list, head);
4298}
4299
4300static void add_sit_entry(unsigned int segno, struct list_head *head)
4301{
4302 struct sit_entry_set *ses;
4303 unsigned int start_segno = START_SEGNO(segno);
4304
4305 list_for_each_entry(ses, head, set_list) {
4306 if (ses->start_segno == start_segno) {
4307 ses->entry_cnt++;
4308 adjust_sit_entry_set(ses, head);
4309 return;
4310 }
4311 }
4312
4313 ses = grab_sit_entry_set();
4314
4315 ses->start_segno = start_segno;
4316 ses->entry_cnt++;
4317 list_add(&ses->set_list, head);
4318}
4319
4320static void add_sits_in_set(struct f2fs_sb_info *sbi)
4321{
4322 struct f2fs_sm_info *sm_info = SM_I(sbi);
4323 struct list_head *set_list = &sm_info->sit_entry_set;
4324 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4325 unsigned int segno;
4326
4327 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4328 add_sit_entry(segno, set_list);
4329}
4330
4331static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4332{
4333 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4334 struct f2fs_journal *journal = curseg->journal;
4335 int i;
4336
4337 down_write(&curseg->journal_rwsem);
4338 for (i = 0; i < sits_in_cursum(journal); i++) {
4339 unsigned int segno;
4340 bool dirtied;
4341
4342 segno = le32_to_cpu(segno_in_journal(journal, i));
4343 dirtied = __mark_sit_entry_dirty(sbi, segno);
4344
4345 if (!dirtied)
4346 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4347 }
4348 update_sits_in_cursum(journal, -i);
4349 up_write(&curseg->journal_rwsem);
4350}
4351
4352/*
4353 * CP calls this function, which flushes SIT entries including sit_journal,
4354 * and moves prefree segs to free segs.
4355 */
4356void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4357{
4358 struct sit_info *sit_i = SIT_I(sbi);
4359 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4360 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4361 struct f2fs_journal *journal = curseg->journal;
4362 struct sit_entry_set *ses, *tmp;
4363 struct list_head *head = &SM_I(sbi)->sit_entry_set;
4364 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4365 struct seg_entry *se;
4366
4367 down_write(&sit_i->sentry_lock);
4368
4369 if (!sit_i->dirty_sentries)
4370 goto out;
4371
4372 /*
4373 * add and account sit entries of dirty bitmap in sit entry
4374 * set temporarily
4375 */
4376 add_sits_in_set(sbi);
4377
4378 /*
4379 * if there are no enough space in journal to store dirty sit
4380 * entries, remove all entries from journal and add and account
4381 * them in sit entry set.
4382 */
4383 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4384 !to_journal)
4385 remove_sits_in_journal(sbi);
4386
4387 /*
4388 * there are two steps to flush sit entries:
4389 * #1, flush sit entries to journal in current cold data summary block.
4390 * #2, flush sit entries to sit page.
4391 */
4392 list_for_each_entry_safe(ses, tmp, head, set_list) {
4393 struct page *page = NULL;
4394 struct f2fs_sit_block *raw_sit = NULL;
4395 unsigned int start_segno = ses->start_segno;
4396 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4397 (unsigned long)MAIN_SEGS(sbi));
4398 unsigned int segno = start_segno;
4399
4400 if (to_journal &&
4401 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4402 to_journal = false;
4403
4404 if (to_journal) {
4405 down_write(&curseg->journal_rwsem);
4406 } else {
4407 page = get_next_sit_page(sbi, start_segno);
4408 raw_sit = page_address(page);
4409 }
4410
4411 /* flush dirty sit entries in region of current sit set */
4412 for_each_set_bit_from(segno, bitmap, end) {
4413 int offset, sit_offset;
4414
4415 se = get_seg_entry(sbi, segno);
4416#ifdef CONFIG_F2FS_CHECK_FS
4417 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4418 SIT_VBLOCK_MAP_SIZE))
4419 f2fs_bug_on(sbi, 1);
4420#endif
4421
4422 /* add discard candidates */
4423 if (!(cpc->reason & CP_DISCARD)) {
4424 cpc->trim_start = segno;
4425 add_discard_addrs(sbi, cpc, false);
4426 }
4427
4428 if (to_journal) {
4429 offset = f2fs_lookup_journal_in_cursum(journal,
4430 SIT_JOURNAL, segno, 1);
4431 f2fs_bug_on(sbi, offset < 0);
4432 segno_in_journal(journal, offset) =
4433 cpu_to_le32(segno);
4434 seg_info_to_raw_sit(se,
4435 &sit_in_journal(journal, offset));
4436 check_block_count(sbi, segno,
4437 &sit_in_journal(journal, offset));
4438 } else {
4439 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4440 seg_info_to_raw_sit(se,
4441 &raw_sit->entries[sit_offset]);
4442 check_block_count(sbi, segno,
4443 &raw_sit->entries[sit_offset]);
4444 }
4445
4446 __clear_bit(segno, bitmap);
4447 sit_i->dirty_sentries--;
4448 ses->entry_cnt--;
4449 }
4450
4451 if (to_journal)
4452 up_write(&curseg->journal_rwsem);
4453 else
4454 f2fs_put_page(page, 1);
4455
4456 f2fs_bug_on(sbi, ses->entry_cnt);
4457 release_sit_entry_set(ses);
4458 }
4459
4460 f2fs_bug_on(sbi, !list_empty(head));
4461 f2fs_bug_on(sbi, sit_i->dirty_sentries);
4462out:
4463 if (cpc->reason & CP_DISCARD) {
4464 __u64 trim_start = cpc->trim_start;
4465
4466 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4467 add_discard_addrs(sbi, cpc, false);
4468
4469 cpc->trim_start = trim_start;
4470 }
4471 up_write(&sit_i->sentry_lock);
4472
4473 set_prefree_as_free_segments(sbi);
4474}
4475
4476static int build_sit_info(struct f2fs_sb_info *sbi)
4477{
4478 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4479 struct sit_info *sit_i;
4480 unsigned int sit_segs, start;
4481 char *src_bitmap, *bitmap;
4482 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4483 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4484
4485 /* allocate memory for SIT information */
4486 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4487 if (!sit_i)
4488 return -ENOMEM;
4489
4490 SM_I(sbi)->sit_info = sit_i;
4491
4492 sit_i->sentries =
4493 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4494 MAIN_SEGS(sbi)),
4495 GFP_KERNEL);
4496 if (!sit_i->sentries)
4497 return -ENOMEM;
4498
4499 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4500 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4501 GFP_KERNEL);
4502 if (!sit_i->dirty_sentries_bitmap)
4503 return -ENOMEM;
4504
4505#ifdef CONFIG_F2FS_CHECK_FS
4506 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4507#else
4508 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4509#endif
4510 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4511 if (!sit_i->bitmap)
4512 return -ENOMEM;
4513
4514 bitmap = sit_i->bitmap;
4515
4516 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4517 sit_i->sentries[start].cur_valid_map = bitmap;
4518 bitmap += SIT_VBLOCK_MAP_SIZE;
4519
4520 sit_i->sentries[start].ckpt_valid_map = bitmap;
4521 bitmap += SIT_VBLOCK_MAP_SIZE;
4522
4523#ifdef CONFIG_F2FS_CHECK_FS
4524 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4525 bitmap += SIT_VBLOCK_MAP_SIZE;
4526#endif
4527
4528 if (discard_map) {
4529 sit_i->sentries[start].discard_map = bitmap;
4530 bitmap += SIT_VBLOCK_MAP_SIZE;
4531 }
4532 }
4533
4534 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4535 if (!sit_i->tmp_map)
4536 return -ENOMEM;
4537
4538 if (__is_large_section(sbi)) {
4539 sit_i->sec_entries =
4540 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4541 MAIN_SECS(sbi)),
4542 GFP_KERNEL);
4543 if (!sit_i->sec_entries)
4544 return -ENOMEM;
4545 }
4546
4547 /* get information related with SIT */
4548 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4549
4550 /* setup SIT bitmap from ckeckpoint pack */
4551 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4552 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4553
4554 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4555 if (!sit_i->sit_bitmap)
4556 return -ENOMEM;
4557
4558#ifdef CONFIG_F2FS_CHECK_FS
4559 sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4560 sit_bitmap_size, GFP_KERNEL);
4561 if (!sit_i->sit_bitmap_mir)
4562 return -ENOMEM;
4563
4564 sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4565 main_bitmap_size, GFP_KERNEL);
4566 if (!sit_i->invalid_segmap)
4567 return -ENOMEM;
4568#endif
4569
4570 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4571 sit_i->sit_blocks = SEGS_TO_BLKS(sbi, sit_segs);
4572 sit_i->written_valid_blocks = 0;
4573 sit_i->bitmap_size = sit_bitmap_size;
4574 sit_i->dirty_sentries = 0;
4575 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4576 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4577 sit_i->mounted_time = ktime_get_boottime_seconds();
4578 init_rwsem(&sit_i->sentry_lock);
4579 return 0;
4580}
4581
4582static int build_free_segmap(struct f2fs_sb_info *sbi)
4583{
4584 struct free_segmap_info *free_i;
4585 unsigned int bitmap_size, sec_bitmap_size;
4586
4587 /* allocate memory for free segmap information */
4588 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4589 if (!free_i)
4590 return -ENOMEM;
4591
4592 SM_I(sbi)->free_info = free_i;
4593
4594 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4595 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4596 if (!free_i->free_segmap)
4597 return -ENOMEM;
4598
4599 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4600 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4601 if (!free_i->free_secmap)
4602 return -ENOMEM;
4603
4604 /* set all segments as dirty temporarily */
4605 memset(free_i->free_segmap, 0xff, bitmap_size);
4606 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4607
4608 /* init free segmap information */
4609 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4610 free_i->free_segments = 0;
4611 free_i->free_sections = 0;
4612 spin_lock_init(&free_i->segmap_lock);
4613 return 0;
4614}
4615
4616static int build_curseg(struct f2fs_sb_info *sbi)
4617{
4618 struct curseg_info *array;
4619 int i;
4620
4621 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4622 sizeof(*array)), GFP_KERNEL);
4623 if (!array)
4624 return -ENOMEM;
4625
4626 SM_I(sbi)->curseg_array = array;
4627
4628 for (i = 0; i < NO_CHECK_TYPE; i++) {
4629 mutex_init(&array[i].curseg_mutex);
4630 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4631 if (!array[i].sum_blk)
4632 return -ENOMEM;
4633 init_rwsem(&array[i].journal_rwsem);
4634 array[i].journal = f2fs_kzalloc(sbi,
4635 sizeof(struct f2fs_journal), GFP_KERNEL);
4636 if (!array[i].journal)
4637 return -ENOMEM;
4638 if (i < NR_PERSISTENT_LOG)
4639 array[i].seg_type = CURSEG_HOT_DATA + i;
4640 else if (i == CURSEG_COLD_DATA_PINNED)
4641 array[i].seg_type = CURSEG_COLD_DATA;
4642 else if (i == CURSEG_ALL_DATA_ATGC)
4643 array[i].seg_type = CURSEG_COLD_DATA;
4644 reset_curseg_fields(&array[i]);
4645 }
4646 return restore_curseg_summaries(sbi);
4647}
4648
4649static int build_sit_entries(struct f2fs_sb_info *sbi)
4650{
4651 struct sit_info *sit_i = SIT_I(sbi);
4652 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4653 struct f2fs_journal *journal = curseg->journal;
4654 struct seg_entry *se;
4655 struct f2fs_sit_entry sit;
4656 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4657 unsigned int i, start, end;
4658 unsigned int readed, start_blk = 0;
4659 int err = 0;
4660 block_t sit_valid_blocks[2] = {0, 0};
4661
4662 do {
4663 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4664 META_SIT, true);
4665
4666 start = start_blk * sit_i->sents_per_block;
4667 end = (start_blk + readed) * sit_i->sents_per_block;
4668
4669 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4670 struct f2fs_sit_block *sit_blk;
4671 struct page *page;
4672
4673 se = &sit_i->sentries[start];
4674 page = get_current_sit_page(sbi, start);
4675 if (IS_ERR(page))
4676 return PTR_ERR(page);
4677 sit_blk = (struct f2fs_sit_block *)page_address(page);
4678 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4679 f2fs_put_page(page, 1);
4680
4681 err = check_block_count(sbi, start, &sit);
4682 if (err)
4683 return err;
4684 seg_info_from_raw_sit(se, &sit);
4685
4686 if (se->type >= NR_PERSISTENT_LOG) {
4687 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4688 se->type, start);
4689 f2fs_handle_error(sbi,
4690 ERROR_INCONSISTENT_SUM_TYPE);
4691 return -EFSCORRUPTED;
4692 }
4693
4694 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4695
4696 if (!f2fs_block_unit_discard(sbi))
4697 goto init_discard_map_done;
4698
4699 /* build discard map only one time */
4700 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4701 memset(se->discard_map, 0xff,
4702 SIT_VBLOCK_MAP_SIZE);
4703 goto init_discard_map_done;
4704 }
4705 memcpy(se->discard_map, se->cur_valid_map,
4706 SIT_VBLOCK_MAP_SIZE);
4707 sbi->discard_blks += BLKS_PER_SEG(sbi) -
4708 se->valid_blocks;
4709init_discard_map_done:
4710 if (__is_large_section(sbi))
4711 get_sec_entry(sbi, start)->valid_blocks +=
4712 se->valid_blocks;
4713 }
4714 start_blk += readed;
4715 } while (start_blk < sit_blk_cnt);
4716
4717 down_read(&curseg->journal_rwsem);
4718 for (i = 0; i < sits_in_cursum(journal); i++) {
4719 unsigned int old_valid_blocks;
4720
4721 start = le32_to_cpu(segno_in_journal(journal, i));
4722 if (start >= MAIN_SEGS(sbi)) {
4723 f2fs_err(sbi, "Wrong journal entry on segno %u",
4724 start);
4725 err = -EFSCORRUPTED;
4726 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4727 break;
4728 }
4729
4730 se = &sit_i->sentries[start];
4731 sit = sit_in_journal(journal, i);
4732
4733 old_valid_blocks = se->valid_blocks;
4734
4735 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4736
4737 err = check_block_count(sbi, start, &sit);
4738 if (err)
4739 break;
4740 seg_info_from_raw_sit(se, &sit);
4741
4742 if (se->type >= NR_PERSISTENT_LOG) {
4743 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4744 se->type, start);
4745 err = -EFSCORRUPTED;
4746 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4747 break;
4748 }
4749
4750 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4751
4752 if (f2fs_block_unit_discard(sbi)) {
4753 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4754 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4755 } else {
4756 memcpy(se->discard_map, se->cur_valid_map,
4757 SIT_VBLOCK_MAP_SIZE);
4758 sbi->discard_blks += old_valid_blocks;
4759 sbi->discard_blks -= se->valid_blocks;
4760 }
4761 }
4762
4763 if (__is_large_section(sbi)) {
4764 get_sec_entry(sbi, start)->valid_blocks +=
4765 se->valid_blocks;
4766 get_sec_entry(sbi, start)->valid_blocks -=
4767 old_valid_blocks;
4768 }
4769 }
4770 up_read(&curseg->journal_rwsem);
4771
4772 if (err)
4773 return err;
4774
4775 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4776 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4777 sit_valid_blocks[NODE], valid_node_count(sbi));
4778 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4779 return -EFSCORRUPTED;
4780 }
4781
4782 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4783 valid_user_blocks(sbi)) {
4784 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4785 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4786 valid_user_blocks(sbi));
4787 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4788 return -EFSCORRUPTED;
4789 }
4790
4791 return 0;
4792}
4793
4794static void init_free_segmap(struct f2fs_sb_info *sbi)
4795{
4796 unsigned int start;
4797 int type;
4798 struct seg_entry *sentry;
4799
4800 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4801 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4802 continue;
4803 sentry = get_seg_entry(sbi, start);
4804 if (!sentry->valid_blocks)
4805 __set_free(sbi, start);
4806 else
4807 SIT_I(sbi)->written_valid_blocks +=
4808 sentry->valid_blocks;
4809 }
4810
4811 /* set use the current segments */
4812 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4813 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4814
4815 __set_test_and_inuse(sbi, curseg_t->segno);
4816 }
4817}
4818
4819static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4820{
4821 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4822 struct free_segmap_info *free_i = FREE_I(sbi);
4823 unsigned int segno = 0, offset = 0, secno;
4824 block_t valid_blocks, usable_blks_in_seg;
4825
4826 while (1) {
4827 /* find dirty segment based on free segmap */
4828 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4829 if (segno >= MAIN_SEGS(sbi))
4830 break;
4831 offset = segno + 1;
4832 valid_blocks = get_valid_blocks(sbi, segno, false);
4833 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4834 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4835 continue;
4836 if (valid_blocks > usable_blks_in_seg) {
4837 f2fs_bug_on(sbi, 1);
4838 continue;
4839 }
4840 mutex_lock(&dirty_i->seglist_lock);
4841 __locate_dirty_segment(sbi, segno, DIRTY);
4842 mutex_unlock(&dirty_i->seglist_lock);
4843 }
4844
4845 if (!__is_large_section(sbi))
4846 return;
4847
4848 mutex_lock(&dirty_i->seglist_lock);
4849 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
4850 valid_blocks = get_valid_blocks(sbi, segno, true);
4851 secno = GET_SEC_FROM_SEG(sbi, segno);
4852
4853 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4854 continue;
4855 if (IS_CURSEC(sbi, secno))
4856 continue;
4857 set_bit(secno, dirty_i->dirty_secmap);
4858 }
4859 mutex_unlock(&dirty_i->seglist_lock);
4860}
4861
4862static int init_victim_secmap(struct f2fs_sb_info *sbi)
4863{
4864 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4865 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4866
4867 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4868 if (!dirty_i->victim_secmap)
4869 return -ENOMEM;
4870
4871 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4872 if (!dirty_i->pinned_secmap)
4873 return -ENOMEM;
4874
4875 dirty_i->pinned_secmap_cnt = 0;
4876 dirty_i->enable_pin_section = true;
4877 return 0;
4878}
4879
4880static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4881{
4882 struct dirty_seglist_info *dirty_i;
4883 unsigned int bitmap_size, i;
4884
4885 /* allocate memory for dirty segments list information */
4886 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4887 GFP_KERNEL);
4888 if (!dirty_i)
4889 return -ENOMEM;
4890
4891 SM_I(sbi)->dirty_info = dirty_i;
4892 mutex_init(&dirty_i->seglist_lock);
4893
4894 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4895
4896 for (i = 0; i < NR_DIRTY_TYPE; i++) {
4897 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4898 GFP_KERNEL);
4899 if (!dirty_i->dirty_segmap[i])
4900 return -ENOMEM;
4901 }
4902
4903 if (__is_large_section(sbi)) {
4904 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4905 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4906 bitmap_size, GFP_KERNEL);
4907 if (!dirty_i->dirty_secmap)
4908 return -ENOMEM;
4909 }
4910
4911 init_dirty_segmap(sbi);
4912 return init_victim_secmap(sbi);
4913}
4914
4915static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4916{
4917 int i;
4918
4919 /*
4920 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4921 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4922 */
4923 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4924 struct curseg_info *curseg = CURSEG_I(sbi, i);
4925 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4926 unsigned int blkofs = curseg->next_blkoff;
4927
4928 if (f2fs_sb_has_readonly(sbi) &&
4929 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4930 continue;
4931
4932 sanity_check_seg_type(sbi, curseg->seg_type);
4933
4934 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4935 f2fs_err(sbi,
4936 "Current segment has invalid alloc_type:%d",
4937 curseg->alloc_type);
4938 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4939 return -EFSCORRUPTED;
4940 }
4941
4942 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4943 goto out;
4944
4945 if (curseg->alloc_type == SSR)
4946 continue;
4947
4948 for (blkofs += 1; blkofs < BLKS_PER_SEG(sbi); blkofs++) {
4949 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4950 continue;
4951out:
4952 f2fs_err(sbi,
4953 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4954 i, curseg->segno, curseg->alloc_type,
4955 curseg->next_blkoff, blkofs);
4956 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4957 return -EFSCORRUPTED;
4958 }
4959 }
4960 return 0;
4961}
4962
4963#ifdef CONFIG_BLK_DEV_ZONED
4964static const char *f2fs_zone_status[BLK_ZONE_COND_OFFLINE + 1] = {
4965 [BLK_ZONE_COND_NOT_WP] = "NOT_WP",
4966 [BLK_ZONE_COND_EMPTY] = "EMPTY",
4967 [BLK_ZONE_COND_IMP_OPEN] = "IMPLICIT_OPEN",
4968 [BLK_ZONE_COND_EXP_OPEN] = "EXPLICIT_OPEN",
4969 [BLK_ZONE_COND_CLOSED] = "CLOSED",
4970 [BLK_ZONE_COND_READONLY] = "READONLY",
4971 [BLK_ZONE_COND_FULL] = "FULL",
4972 [BLK_ZONE_COND_OFFLINE] = "OFFLINE",
4973};
4974
4975static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4976 struct f2fs_dev_info *fdev,
4977 struct blk_zone *zone)
4978{
4979 unsigned int zone_segno;
4980 block_t zone_block, valid_block_cnt;
4981 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4982 int ret;
4983 unsigned int nofs_flags;
4984
4985 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4986 return 0;
4987
4988 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4989 zone_segno = GET_SEGNO(sbi, zone_block);
4990
4991 /*
4992 * Skip check of zones cursegs point to, since
4993 * fix_curseg_write_pointer() checks them.
4994 */
4995 if (zone_segno >= MAIN_SEGS(sbi))
4996 return 0;
4997
4998 /*
4999 * Get # of valid block of the zone.
5000 */
5001 valid_block_cnt = get_valid_blocks(sbi, zone_segno, true);
5002 if (IS_CURSEC(sbi, GET_SEC_FROM_SEG(sbi, zone_segno))) {
5003 f2fs_notice(sbi, "Open zones: valid block[0x%x,0x%x] cond[%s]",
5004 zone_segno, valid_block_cnt,
5005 f2fs_zone_status[zone->cond]);
5006 return 0;
5007 }
5008
5009 if ((!valid_block_cnt && zone->cond == BLK_ZONE_COND_EMPTY) ||
5010 (valid_block_cnt && zone->cond == BLK_ZONE_COND_FULL))
5011 return 0;
5012
5013 if (!valid_block_cnt) {
5014 f2fs_notice(sbi, "Zone without valid block has non-zero write "
5015 "pointer. Reset the write pointer: cond[%s]",
5016 f2fs_zone_status[zone->cond]);
5017 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
5018 zone->len >> log_sectors_per_block);
5019 if (ret)
5020 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5021 fdev->path, ret);
5022 return ret;
5023 }
5024
5025 /*
5026 * If there are valid blocks and the write pointer doesn't match
5027 * with them, we need to report the inconsistency and fill
5028 * the zone till the end to close the zone. This inconsistency
5029 * does not cause write error because the zone will not be
5030 * selected for write operation until it get discarded.
5031 */
5032 f2fs_notice(sbi, "Valid blocks are not aligned with write "
5033 "pointer: valid block[0x%x,0x%x] cond[%s]",
5034 zone_segno, valid_block_cnt, f2fs_zone_status[zone->cond]);
5035
5036 nofs_flags = memalloc_nofs_save();
5037 ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH,
5038 zone->start, zone->len);
5039 memalloc_nofs_restore(nofs_flags);
5040 if (ret == -EOPNOTSUPP) {
5041 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp,
5042 zone->len - (zone->wp - zone->start),
5043 GFP_NOFS, 0);
5044 if (ret)
5045 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)",
5046 fdev->path, ret);
5047 } else if (ret) {
5048 f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)",
5049 fdev->path, ret);
5050 }
5051
5052 return ret;
5053}
5054
5055static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
5056 block_t zone_blkaddr)
5057{
5058 int i;
5059
5060 for (i = 0; i < sbi->s_ndevs; i++) {
5061 if (!bdev_is_zoned(FDEV(i).bdev))
5062 continue;
5063 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
5064 zone_blkaddr <= FDEV(i).end_blk))
5065 return &FDEV(i);
5066 }
5067
5068 return NULL;
5069}
5070
5071static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
5072 void *data)
5073{
5074 memcpy(data, zone, sizeof(struct blk_zone));
5075 return 0;
5076}
5077
5078static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
5079{
5080 struct curseg_info *cs = CURSEG_I(sbi, type);
5081 struct f2fs_dev_info *zbd;
5082 struct blk_zone zone;
5083 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
5084 block_t cs_zone_block, wp_block;
5085 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
5086 sector_t zone_sector;
5087 int err;
5088
5089 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5090 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5091
5092 zbd = get_target_zoned_dev(sbi, cs_zone_block);
5093 if (!zbd)
5094 return 0;
5095
5096 /* report zone for the sector the curseg points to */
5097 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5098 << log_sectors_per_block;
5099 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5100 report_one_zone_cb, &zone);
5101 if (err != 1) {
5102 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5103 zbd->path, err);
5104 return err;
5105 }
5106
5107 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5108 return 0;
5109
5110 /*
5111 * When safely unmounted in the previous mount, we could use current
5112 * segments. Otherwise, allocate new sections.
5113 */
5114 if (is_set_ckpt_flags(sbi, CP_UMOUNT_FLAG)) {
5115 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
5116 wp_segno = GET_SEGNO(sbi, wp_block);
5117 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
5118 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
5119
5120 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
5121 wp_sector_off == 0)
5122 return 0;
5123
5124 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
5125 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", type, cs->segno,
5126 cs->next_blkoff, wp_segno, wp_blkoff);
5127 }
5128
5129 /* Allocate a new section if it's not new. */
5130 if (cs->next_blkoff) {
5131 unsigned int old_segno = cs->segno, old_blkoff = cs->next_blkoff;
5132
5133 f2fs_allocate_new_section(sbi, type, true);
5134 f2fs_notice(sbi, "Assign new section to curseg[%d]: "
5135 "[0x%x,0x%x] -> [0x%x,0x%x]",
5136 type, old_segno, old_blkoff,
5137 cs->segno, cs->next_blkoff);
5138 }
5139
5140 /* check consistency of the zone curseg pointed to */
5141 if (check_zone_write_pointer(sbi, zbd, &zone))
5142 return -EIO;
5143
5144 /* check newly assigned zone */
5145 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5146 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5147
5148 zbd = get_target_zoned_dev(sbi, cs_zone_block);
5149 if (!zbd)
5150 return 0;
5151
5152 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5153 << log_sectors_per_block;
5154 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5155 report_one_zone_cb, &zone);
5156 if (err != 1) {
5157 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5158 zbd->path, err);
5159 return err;
5160 }
5161
5162 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5163 return 0;
5164
5165 if (zone.wp != zone.start) {
5166 f2fs_notice(sbi,
5167 "New zone for curseg[%d] is not yet discarded. "
5168 "Reset the zone: curseg[0x%x,0x%x]",
5169 type, cs->segno, cs->next_blkoff);
5170 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block,
5171 zone.len >> log_sectors_per_block);
5172 if (err) {
5173 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5174 zbd->path, err);
5175 return err;
5176 }
5177 }
5178
5179 return 0;
5180}
5181
5182int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5183{
5184 int i, ret;
5185
5186 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5187 ret = fix_curseg_write_pointer(sbi, i);
5188 if (ret)
5189 return ret;
5190 }
5191
5192 return 0;
5193}
5194
5195struct check_zone_write_pointer_args {
5196 struct f2fs_sb_info *sbi;
5197 struct f2fs_dev_info *fdev;
5198};
5199
5200static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5201 void *data)
5202{
5203 struct check_zone_write_pointer_args *args;
5204
5205 args = (struct check_zone_write_pointer_args *)data;
5206
5207 return check_zone_write_pointer(args->sbi, args->fdev, zone);
5208}
5209
5210int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5211{
5212 int i, ret;
5213 struct check_zone_write_pointer_args args;
5214
5215 for (i = 0; i < sbi->s_ndevs; i++) {
5216 if (!bdev_is_zoned(FDEV(i).bdev))
5217 continue;
5218
5219 args.sbi = sbi;
5220 args.fdev = &FDEV(i);
5221 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5222 check_zone_write_pointer_cb, &args);
5223 if (ret < 0)
5224 return ret;
5225 }
5226
5227 return 0;
5228}
5229
5230/*
5231 * Return the number of usable blocks in a segment. The number of blocks
5232 * returned is always equal to the number of blocks in a segment for
5233 * segments fully contained within a sequential zone capacity or a
5234 * conventional zone. For segments partially contained in a sequential
5235 * zone capacity, the number of usable blocks up to the zone capacity
5236 * is returned. 0 is returned in all other cases.
5237 */
5238static inline unsigned int f2fs_usable_zone_blks_in_seg(
5239 struct f2fs_sb_info *sbi, unsigned int segno)
5240{
5241 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5242 unsigned int secno;
5243
5244 if (!sbi->unusable_blocks_per_sec)
5245 return BLKS_PER_SEG(sbi);
5246
5247 secno = GET_SEC_FROM_SEG(sbi, segno);
5248 seg_start = START_BLOCK(sbi, segno);
5249 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5250 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5251
5252 /*
5253 * If segment starts before zone capacity and spans beyond
5254 * zone capacity, then usable blocks are from seg start to
5255 * zone capacity. If the segment starts after the zone capacity,
5256 * then there are no usable blocks.
5257 */
5258 if (seg_start >= sec_cap_blkaddr)
5259 return 0;
5260 if (seg_start + BLKS_PER_SEG(sbi) > sec_cap_blkaddr)
5261 return sec_cap_blkaddr - seg_start;
5262
5263 return BLKS_PER_SEG(sbi);
5264}
5265#else
5266int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5267{
5268 return 0;
5269}
5270
5271int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5272{
5273 return 0;
5274}
5275
5276static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5277 unsigned int segno)
5278{
5279 return 0;
5280}
5281
5282#endif
5283unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5284 unsigned int segno)
5285{
5286 if (f2fs_sb_has_blkzoned(sbi))
5287 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5288
5289 return BLKS_PER_SEG(sbi);
5290}
5291
5292unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5293 unsigned int segno)
5294{
5295 if (f2fs_sb_has_blkzoned(sbi))
5296 return CAP_SEGS_PER_SEC(sbi);
5297
5298 return SEGS_PER_SEC(sbi);
5299}
5300
5301/*
5302 * Update min, max modified time for cost-benefit GC algorithm
5303 */
5304static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5305{
5306 struct sit_info *sit_i = SIT_I(sbi);
5307 unsigned int segno;
5308
5309 down_write(&sit_i->sentry_lock);
5310
5311 sit_i->min_mtime = ULLONG_MAX;
5312
5313 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
5314 unsigned int i;
5315 unsigned long long mtime = 0;
5316
5317 for (i = 0; i < SEGS_PER_SEC(sbi); i++)
5318 mtime += get_seg_entry(sbi, segno + i)->mtime;
5319
5320 mtime = div_u64(mtime, SEGS_PER_SEC(sbi));
5321
5322 if (sit_i->min_mtime > mtime)
5323 sit_i->min_mtime = mtime;
5324 }
5325 sit_i->max_mtime = get_mtime(sbi, false);
5326 sit_i->dirty_max_mtime = 0;
5327 up_write(&sit_i->sentry_lock);
5328}
5329
5330int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5331{
5332 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5333 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5334 struct f2fs_sm_info *sm_info;
5335 int err;
5336
5337 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5338 if (!sm_info)
5339 return -ENOMEM;
5340
5341 /* init sm info */
5342 sbi->sm_info = sm_info;
5343 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5344 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5345 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5346 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5347 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5348 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5349 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5350 sm_info->rec_prefree_segments = sm_info->main_segments *
5351 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5352 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5353 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5354
5355 if (!f2fs_lfs_mode(sbi))
5356 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5357 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5358 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5359 sm_info->min_seq_blocks = BLKS_PER_SEG(sbi);
5360 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5361 sm_info->min_ssr_sections = reserved_sections(sbi);
5362
5363 INIT_LIST_HEAD(&sm_info->sit_entry_set);
5364
5365 init_f2fs_rwsem(&sm_info->curseg_lock);
5366
5367 err = f2fs_create_flush_cmd_control(sbi);
5368 if (err)
5369 return err;
5370
5371 err = create_discard_cmd_control(sbi);
5372 if (err)
5373 return err;
5374
5375 err = build_sit_info(sbi);
5376 if (err)
5377 return err;
5378 err = build_free_segmap(sbi);
5379 if (err)
5380 return err;
5381 err = build_curseg(sbi);
5382 if (err)
5383 return err;
5384
5385 /* reinit free segmap based on SIT */
5386 err = build_sit_entries(sbi);
5387 if (err)
5388 return err;
5389
5390 init_free_segmap(sbi);
5391 err = build_dirty_segmap(sbi);
5392 if (err)
5393 return err;
5394
5395 err = sanity_check_curseg(sbi);
5396 if (err)
5397 return err;
5398
5399 init_min_max_mtime(sbi);
5400 return 0;
5401}
5402
5403static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5404 enum dirty_type dirty_type)
5405{
5406 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5407
5408 mutex_lock(&dirty_i->seglist_lock);
5409 kvfree(dirty_i->dirty_segmap[dirty_type]);
5410 dirty_i->nr_dirty[dirty_type] = 0;
5411 mutex_unlock(&dirty_i->seglist_lock);
5412}
5413
5414static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5415{
5416 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5417
5418 kvfree(dirty_i->pinned_secmap);
5419 kvfree(dirty_i->victim_secmap);
5420}
5421
5422static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5423{
5424 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5425 int i;
5426
5427 if (!dirty_i)
5428 return;
5429
5430 /* discard pre-free/dirty segments list */
5431 for (i = 0; i < NR_DIRTY_TYPE; i++)
5432 discard_dirty_segmap(sbi, i);
5433
5434 if (__is_large_section(sbi)) {
5435 mutex_lock(&dirty_i->seglist_lock);
5436 kvfree(dirty_i->dirty_secmap);
5437 mutex_unlock(&dirty_i->seglist_lock);
5438 }
5439
5440 destroy_victim_secmap(sbi);
5441 SM_I(sbi)->dirty_info = NULL;
5442 kfree(dirty_i);
5443}
5444
5445static void destroy_curseg(struct f2fs_sb_info *sbi)
5446{
5447 struct curseg_info *array = SM_I(sbi)->curseg_array;
5448 int i;
5449
5450 if (!array)
5451 return;
5452 SM_I(sbi)->curseg_array = NULL;
5453 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5454 kfree(array[i].sum_blk);
5455 kfree(array[i].journal);
5456 }
5457 kfree(array);
5458}
5459
5460static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5461{
5462 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5463
5464 if (!free_i)
5465 return;
5466 SM_I(sbi)->free_info = NULL;
5467 kvfree(free_i->free_segmap);
5468 kvfree(free_i->free_secmap);
5469 kfree(free_i);
5470}
5471
5472static void destroy_sit_info(struct f2fs_sb_info *sbi)
5473{
5474 struct sit_info *sit_i = SIT_I(sbi);
5475
5476 if (!sit_i)
5477 return;
5478
5479 if (sit_i->sentries)
5480 kvfree(sit_i->bitmap);
5481 kfree(sit_i->tmp_map);
5482
5483 kvfree(sit_i->sentries);
5484 kvfree(sit_i->sec_entries);
5485 kvfree(sit_i->dirty_sentries_bitmap);
5486
5487 SM_I(sbi)->sit_info = NULL;
5488 kvfree(sit_i->sit_bitmap);
5489#ifdef CONFIG_F2FS_CHECK_FS
5490 kvfree(sit_i->sit_bitmap_mir);
5491 kvfree(sit_i->invalid_segmap);
5492#endif
5493 kfree(sit_i);
5494}
5495
5496void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5497{
5498 struct f2fs_sm_info *sm_info = SM_I(sbi);
5499
5500 if (!sm_info)
5501 return;
5502 f2fs_destroy_flush_cmd_control(sbi, true);
5503 destroy_discard_cmd_control(sbi);
5504 destroy_dirty_segmap(sbi);
5505 destroy_curseg(sbi);
5506 destroy_free_segmap(sbi);
5507 destroy_sit_info(sbi);
5508 sbi->sm_info = NULL;
5509 kfree(sm_info);
5510}
5511
5512int __init f2fs_create_segment_manager_caches(void)
5513{
5514 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5515 sizeof(struct discard_entry));
5516 if (!discard_entry_slab)
5517 goto fail;
5518
5519 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5520 sizeof(struct discard_cmd));
5521 if (!discard_cmd_slab)
5522 goto destroy_discard_entry;
5523
5524 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5525 sizeof(struct sit_entry_set));
5526 if (!sit_entry_set_slab)
5527 goto destroy_discard_cmd;
5528
5529 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5530 sizeof(struct revoke_entry));
5531 if (!revoke_entry_slab)
5532 goto destroy_sit_entry_set;
5533 return 0;
5534
5535destroy_sit_entry_set:
5536 kmem_cache_destroy(sit_entry_set_slab);
5537destroy_discard_cmd:
5538 kmem_cache_destroy(discard_cmd_slab);
5539destroy_discard_entry:
5540 kmem_cache_destroy(discard_entry_slab);
5541fail:
5542 return -ENOMEM;
5543}
5544
5545void f2fs_destroy_segment_manager_caches(void)
5546{
5547 kmem_cache_destroy(sit_entry_set_slab);
5548 kmem_cache_destroy(discard_cmd_slab);
5549 kmem_cache_destroy(discard_entry_slab);
5550 kmem_cache_destroy(revoke_entry_slab);
5551}