1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * fs/f2fs/segment.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/sched/mm.h>
13 #include <linux/prefetch.h>
14 #include <linux/kthread.h>
15 #include <linux/swap.h>
16 #include <linux/timer.h>
17 #include <linux/freezer.h>
18 #include <linux/sched/signal.h>
19 #include <linux/random.h>
20
21 #include "f2fs.h"
22 #include "segment.h"
23 #include "node.h"
24 #include "gc.h"
25 #include "iostat.h"
26 #include <trace/events/f2fs.h>
27
28 #define __reverse_ffz(x) __reverse_ffs(~(x))
29
30 static struct kmem_cache *discard_entry_slab;
31 static struct kmem_cache *discard_cmd_slab;
32 static struct kmem_cache *sit_entry_set_slab;
33 static struct kmem_cache *revoke_entry_slab;
34
__reverse_ulong(unsigned char * str)35 static unsigned long __reverse_ulong(unsigned char *str)
36 {
37 unsigned long tmp = 0;
38 int shift = 24, idx = 0;
39
40 #if BITS_PER_LONG == 64
41 shift = 56;
42 #endif
43 while (shift >= 0) {
44 tmp |= (unsigned long)str[idx++] << shift;
45 shift -= BITS_PER_BYTE;
46 }
47 return tmp;
48 }
49
50 /*
51 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
52 * MSB and LSB are reversed in a byte by f2fs_set_bit.
53 */
__reverse_ffs(unsigned long word)54 static inline unsigned long __reverse_ffs(unsigned long word)
55 {
56 int num = 0;
57
58 #if BITS_PER_LONG == 64
59 if ((word & 0xffffffff00000000UL) == 0)
60 num += 32;
61 else
62 word >>= 32;
63 #endif
64 if ((word & 0xffff0000) == 0)
65 num += 16;
66 else
67 word >>= 16;
68
69 if ((word & 0xff00) == 0)
70 num += 8;
71 else
72 word >>= 8;
73
74 if ((word & 0xf0) == 0)
75 num += 4;
76 else
77 word >>= 4;
78
79 if ((word & 0xc) == 0)
80 num += 2;
81 else
82 word >>= 2;
83
84 if ((word & 0x2) == 0)
85 num += 1;
86 return num;
87 }
88
89 /*
90 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
91 * f2fs_set_bit makes MSB and LSB reversed in a byte.
92 * @size must be integral times of unsigned long.
93 * Example:
94 * MSB <--> LSB
95 * f2fs_set_bit(0, bitmap) => 1000 0000
96 * f2fs_set_bit(7, bitmap) => 0000 0001
97 */
__find_rev_next_bit(const unsigned long * addr,unsigned long size,unsigned long offset)98 static unsigned long __find_rev_next_bit(const unsigned long *addr,
99 unsigned long size, unsigned long offset)
100 {
101 const unsigned long *p = addr + BIT_WORD(offset);
102 unsigned long result = size;
103 unsigned long tmp;
104
105 if (offset >= size)
106 return size;
107
108 size -= (offset & ~(BITS_PER_LONG - 1));
109 offset %= BITS_PER_LONG;
110
111 while (1) {
112 if (*p == 0)
113 goto pass;
114
115 tmp = __reverse_ulong((unsigned char *)p);
116
117 tmp &= ~0UL >> offset;
118 if (size < BITS_PER_LONG)
119 tmp &= (~0UL << (BITS_PER_LONG - size));
120 if (tmp)
121 goto found;
122 pass:
123 if (size <= BITS_PER_LONG)
124 break;
125 size -= BITS_PER_LONG;
126 offset = 0;
127 p++;
128 }
129 return result;
130 found:
131 return result - size + __reverse_ffs(tmp);
132 }
133
__find_rev_next_zero_bit(const unsigned long * addr,unsigned long size,unsigned long offset)134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
135 unsigned long size, unsigned long offset)
136 {
137 const unsigned long *p = addr + BIT_WORD(offset);
138 unsigned long result = size;
139 unsigned long tmp;
140
141 if (offset >= size)
142 return size;
143
144 size -= (offset & ~(BITS_PER_LONG - 1));
145 offset %= BITS_PER_LONG;
146
147 while (1) {
148 if (*p == ~0UL)
149 goto pass;
150
151 tmp = __reverse_ulong((unsigned char *)p);
152
153 if (offset)
154 tmp |= ~0UL << (BITS_PER_LONG - offset);
155 if (size < BITS_PER_LONG)
156 tmp |= ~0UL >> size;
157 if (tmp != ~0UL)
158 goto found;
159 pass:
160 if (size <= BITS_PER_LONG)
161 break;
162 size -= BITS_PER_LONG;
163 offset = 0;
164 p++;
165 }
166 return result;
167 found:
168 return result - size + __reverse_ffz(tmp);
169 }
170
f2fs_need_SSR(struct f2fs_sb_info * sbi)171 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
172 {
173 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
174 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
175 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
176
177 if (f2fs_lfs_mode(sbi))
178 return false;
179 if (sbi->gc_mode == GC_URGENT_HIGH)
180 return true;
181 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
182 return true;
183
184 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
185 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
186 }
187
f2fs_abort_atomic_write(struct inode * inode,bool clean)188 void f2fs_abort_atomic_write(struct inode *inode, bool clean)
189 {
190 struct f2fs_inode_info *fi = F2FS_I(inode);
191
192 if (!f2fs_is_atomic_file(inode))
193 return;
194
195 if (clean)
196 truncate_inode_pages_final(inode->i_mapping);
197
198 release_atomic_write_cnt(inode);
199 clear_inode_flag(inode, FI_ATOMIC_COMMITTED);
200 clear_inode_flag(inode, FI_ATOMIC_REPLACE);
201 clear_inode_flag(inode, FI_ATOMIC_FILE);
202 if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) {
203 clear_inode_flag(inode, FI_ATOMIC_DIRTIED);
204 f2fs_mark_inode_dirty_sync(inode, true);
205 }
206 stat_dec_atomic_inode(inode);
207
208 F2FS_I(inode)->atomic_write_task = NULL;
209
210 if (clean) {
211 f2fs_i_size_write(inode, fi->original_i_size);
212 fi->original_i_size = 0;
213 }
214 /* avoid stale dirty inode during eviction */
215 sync_inode_metadata(inode, 0);
216 }
217
__replace_atomic_write_block(struct inode * inode,pgoff_t index,block_t new_addr,block_t * old_addr,bool recover)218 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
219 block_t new_addr, block_t *old_addr, bool recover)
220 {
221 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
222 struct dnode_of_data dn;
223 struct node_info ni;
224 int err;
225
226 retry:
227 set_new_dnode(&dn, inode, NULL, NULL, 0);
228 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE);
229 if (err) {
230 if (err == -ENOMEM) {
231 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
232 goto retry;
233 }
234 return err;
235 }
236
237 err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
238 if (err) {
239 f2fs_put_dnode(&dn);
240 return err;
241 }
242
243 if (recover) {
244 /* dn.data_blkaddr is always valid */
245 if (!__is_valid_data_blkaddr(new_addr)) {
246 if (new_addr == NULL_ADDR)
247 dec_valid_block_count(sbi, inode, 1);
248 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
249 f2fs_update_data_blkaddr(&dn, new_addr);
250 } else {
251 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
252 new_addr, ni.version, true, true);
253 }
254 } else {
255 blkcnt_t count = 1;
256
257 err = inc_valid_block_count(sbi, inode, &count, true);
258 if (err) {
259 f2fs_put_dnode(&dn);
260 return err;
261 }
262
263 *old_addr = dn.data_blkaddr;
264 f2fs_truncate_data_blocks_range(&dn, 1);
265 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
266
267 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
268 ni.version, true, false);
269 }
270
271 f2fs_put_dnode(&dn);
272
273 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode,
274 index, old_addr ? *old_addr : 0, new_addr, recover);
275 return 0;
276 }
277
__complete_revoke_list(struct inode * inode,struct list_head * head,bool revoke)278 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
279 bool revoke)
280 {
281 struct revoke_entry *cur, *tmp;
282 pgoff_t start_index = 0;
283 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
284
285 list_for_each_entry_safe(cur, tmp, head, list) {
286 if (revoke) {
287 __replace_atomic_write_block(inode, cur->index,
288 cur->old_addr, NULL, true);
289 } else if (truncate) {
290 f2fs_truncate_hole(inode, start_index, cur->index);
291 start_index = cur->index + 1;
292 }
293
294 list_del(&cur->list);
295 kmem_cache_free(revoke_entry_slab, cur);
296 }
297
298 if (!revoke && truncate)
299 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
300 }
301
__f2fs_commit_atomic_write(struct inode * inode)302 static int __f2fs_commit_atomic_write(struct inode *inode)
303 {
304 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
305 struct f2fs_inode_info *fi = F2FS_I(inode);
306 struct inode *cow_inode = fi->cow_inode;
307 struct revoke_entry *new;
308 struct list_head revoke_list;
309 block_t blkaddr;
310 struct dnode_of_data dn;
311 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
312 pgoff_t off = 0, blen, index;
313 int ret = 0, i;
314
315 INIT_LIST_HEAD(&revoke_list);
316
317 while (len) {
318 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
319
320 set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
321 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
322 if (ret && ret != -ENOENT) {
323 goto out;
324 } else if (ret == -ENOENT) {
325 ret = 0;
326 if (dn.max_level == 0)
327 goto out;
328 goto next;
329 }
330
331 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
332 len);
333 index = off;
334 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
335 blkaddr = f2fs_data_blkaddr(&dn);
336
337 if (!__is_valid_data_blkaddr(blkaddr)) {
338 continue;
339 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
340 DATA_GENERIC_ENHANCE)) {
341 f2fs_put_dnode(&dn);
342 ret = -EFSCORRUPTED;
343 goto out;
344 }
345
346 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
347 true, NULL);
348
349 ret = __replace_atomic_write_block(inode, index, blkaddr,
350 &new->old_addr, false);
351 if (ret) {
352 f2fs_put_dnode(&dn);
353 kmem_cache_free(revoke_entry_slab, new);
354 goto out;
355 }
356
357 f2fs_update_data_blkaddr(&dn, NULL_ADDR);
358 new->index = index;
359 list_add_tail(&new->list, &revoke_list);
360 }
361 f2fs_put_dnode(&dn);
362 next:
363 off += blen;
364 len -= blen;
365 }
366
367 out:
368 if (ret) {
369 sbi->revoked_atomic_block += fi->atomic_write_cnt;
370 } else {
371 sbi->committed_atomic_block += fi->atomic_write_cnt;
372 set_inode_flag(inode, FI_ATOMIC_COMMITTED);
373 if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) {
374 clear_inode_flag(inode, FI_ATOMIC_DIRTIED);
375 f2fs_mark_inode_dirty_sync(inode, true);
376 }
377 }
378
379 __complete_revoke_list(inode, &revoke_list, ret ? true : false);
380
381 return ret;
382 }
383
f2fs_commit_atomic_write(struct inode * inode)384 int f2fs_commit_atomic_write(struct inode *inode)
385 {
386 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
387 struct f2fs_inode_info *fi = F2FS_I(inode);
388 int err;
389
390 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
391 if (err)
392 return err;
393
394 f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
395 f2fs_lock_op(sbi);
396
397 err = __f2fs_commit_atomic_write(inode);
398
399 f2fs_unlock_op(sbi);
400 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
401
402 return err;
403 }
404
405 /*
406 * This function balances dirty node and dentry pages.
407 * In addition, it controls garbage collection.
408 */
f2fs_balance_fs(struct f2fs_sb_info * sbi,bool need)409 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
410 {
411 if (f2fs_cp_error(sbi))
412 return;
413
414 if (time_to_inject(sbi, FAULT_CHECKPOINT))
415 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
416
417 /* balance_fs_bg is able to be pending */
418 if (need && excess_cached_nats(sbi))
419 f2fs_balance_fs_bg(sbi, false);
420
421 if (!f2fs_is_checkpoint_ready(sbi))
422 return;
423
424 /*
425 * We should do GC or end up with checkpoint, if there are so many dirty
426 * dir/node pages without enough free segments.
427 */
428 if (has_enough_free_secs(sbi, 0, 0))
429 return;
430
431 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
432 sbi->gc_thread->f2fs_gc_task) {
433 DEFINE_WAIT(wait);
434
435 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
436 TASK_UNINTERRUPTIBLE);
437 wake_up(&sbi->gc_thread->gc_wait_queue_head);
438 io_schedule();
439 finish_wait(&sbi->gc_thread->fggc_wq, &wait);
440 } else {
441 struct f2fs_gc_control gc_control = {
442 .victim_segno = NULL_SEGNO,
443 .init_gc_type = BG_GC,
444 .no_bg_gc = true,
445 .should_migrate_blocks = false,
446 .err_gc_skipped = false,
447 .nr_free_secs = 1 };
448 f2fs_down_write(&sbi->gc_lock);
449 stat_inc_gc_call_count(sbi, FOREGROUND);
450 f2fs_gc(sbi, &gc_control);
451 }
452 }
453
excess_dirty_threshold(struct f2fs_sb_info * sbi)454 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
455 {
456 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
457 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
458 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
459 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
460 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
461 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
462 unsigned int threshold =
463 SEGS_TO_BLKS(sbi, (factor * DEFAULT_DIRTY_THRESHOLD));
464 unsigned int global_threshold = threshold * 3 / 2;
465
466 if (dents >= threshold || qdata >= threshold ||
467 nodes >= threshold || meta >= threshold ||
468 imeta >= threshold)
469 return true;
470 return dents + qdata + nodes + meta + imeta > global_threshold;
471 }
472
f2fs_balance_fs_bg(struct f2fs_sb_info * sbi,bool from_bg)473 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
474 {
475 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
476 return;
477
478 /* try to shrink extent cache when there is no enough memory */
479 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
480 f2fs_shrink_read_extent_tree(sbi,
481 READ_EXTENT_CACHE_SHRINK_NUMBER);
482
483 /* try to shrink age extent cache when there is no enough memory */
484 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
485 f2fs_shrink_age_extent_tree(sbi,
486 AGE_EXTENT_CACHE_SHRINK_NUMBER);
487
488 /* check the # of cached NAT entries */
489 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
490 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
491
492 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
493 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
494 else
495 f2fs_build_free_nids(sbi, false, false);
496
497 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
498 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
499 goto do_sync;
500
501 /* there is background inflight IO or foreground operation recently */
502 if (is_inflight_io(sbi, REQ_TIME) ||
503 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
504 return;
505
506 /* exceed periodical checkpoint timeout threshold */
507 if (f2fs_time_over(sbi, CP_TIME))
508 goto do_sync;
509
510 /* checkpoint is the only way to shrink partial cached entries */
511 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
512 f2fs_available_free_memory(sbi, INO_ENTRIES))
513 return;
514
515 do_sync:
516 if (test_opt(sbi, DATA_FLUSH) && from_bg) {
517 struct blk_plug plug;
518
519 mutex_lock(&sbi->flush_lock);
520
521 blk_start_plug(&plug);
522 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
523 blk_finish_plug(&plug);
524
525 mutex_unlock(&sbi->flush_lock);
526 }
527 stat_inc_cp_call_count(sbi, BACKGROUND);
528 f2fs_sync_fs(sbi->sb, 1);
529 }
530
__submit_flush_wait(struct f2fs_sb_info * sbi,struct block_device * bdev)531 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
532 struct block_device *bdev)
533 {
534 int ret = blkdev_issue_flush(bdev);
535
536 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
537 test_opt(sbi, FLUSH_MERGE), ret);
538 if (!ret)
539 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
540 return ret;
541 }
542
submit_flush_wait(struct f2fs_sb_info * sbi,nid_t ino)543 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
544 {
545 int ret = 0;
546 int i;
547
548 if (!f2fs_is_multi_device(sbi))
549 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
550
551 for (i = 0; i < sbi->s_ndevs; i++) {
552 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
553 continue;
554 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
555 if (ret)
556 break;
557 }
558 return ret;
559 }
560
issue_flush_thread(void * data)561 static int issue_flush_thread(void *data)
562 {
563 struct f2fs_sb_info *sbi = data;
564 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
565 wait_queue_head_t *q = &fcc->flush_wait_queue;
566 repeat:
567 if (kthread_should_stop())
568 return 0;
569
570 if (!llist_empty(&fcc->issue_list)) {
571 struct flush_cmd *cmd, *next;
572 int ret;
573
574 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
575 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
576
577 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
578
579 ret = submit_flush_wait(sbi, cmd->ino);
580 atomic_inc(&fcc->issued_flush);
581
582 llist_for_each_entry_safe(cmd, next,
583 fcc->dispatch_list, llnode) {
584 cmd->ret = ret;
585 complete(&cmd->wait);
586 }
587 fcc->dispatch_list = NULL;
588 }
589
590 wait_event_interruptible(*q,
591 kthread_should_stop() || !llist_empty(&fcc->issue_list));
592 goto repeat;
593 }
594
f2fs_issue_flush(struct f2fs_sb_info * sbi,nid_t ino)595 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
596 {
597 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
598 struct flush_cmd cmd;
599 int ret;
600
601 if (test_opt(sbi, NOBARRIER))
602 return 0;
603
604 if (!test_opt(sbi, FLUSH_MERGE)) {
605 atomic_inc(&fcc->queued_flush);
606 ret = submit_flush_wait(sbi, ino);
607 atomic_dec(&fcc->queued_flush);
608 atomic_inc(&fcc->issued_flush);
609 return ret;
610 }
611
612 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
613 f2fs_is_multi_device(sbi)) {
614 ret = submit_flush_wait(sbi, ino);
615 atomic_dec(&fcc->queued_flush);
616
617 atomic_inc(&fcc->issued_flush);
618 return ret;
619 }
620
621 cmd.ino = ino;
622 init_completion(&cmd.wait);
623
624 llist_add(&cmd.llnode, &fcc->issue_list);
625
626 /*
627 * update issue_list before we wake up issue_flush thread, this
628 * smp_mb() pairs with another barrier in ___wait_event(), see
629 * more details in comments of waitqueue_active().
630 */
631 smp_mb();
632
633 if (waitqueue_active(&fcc->flush_wait_queue))
634 wake_up(&fcc->flush_wait_queue);
635
636 if (fcc->f2fs_issue_flush) {
637 wait_for_completion(&cmd.wait);
638 atomic_dec(&fcc->queued_flush);
639 } else {
640 struct llist_node *list;
641
642 list = llist_del_all(&fcc->issue_list);
643 if (!list) {
644 wait_for_completion(&cmd.wait);
645 atomic_dec(&fcc->queued_flush);
646 } else {
647 struct flush_cmd *tmp, *next;
648
649 ret = submit_flush_wait(sbi, ino);
650
651 llist_for_each_entry_safe(tmp, next, list, llnode) {
652 if (tmp == &cmd) {
653 cmd.ret = ret;
654 atomic_dec(&fcc->queued_flush);
655 continue;
656 }
657 tmp->ret = ret;
658 complete(&tmp->wait);
659 }
660 }
661 }
662
663 return cmd.ret;
664 }
665
f2fs_create_flush_cmd_control(struct f2fs_sb_info * sbi)666 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
667 {
668 dev_t dev = sbi->sb->s_bdev->bd_dev;
669 struct flush_cmd_control *fcc;
670
671 if (SM_I(sbi)->fcc_info) {
672 fcc = SM_I(sbi)->fcc_info;
673 if (fcc->f2fs_issue_flush)
674 return 0;
675 goto init_thread;
676 }
677
678 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
679 if (!fcc)
680 return -ENOMEM;
681 atomic_set(&fcc->issued_flush, 0);
682 atomic_set(&fcc->queued_flush, 0);
683 init_waitqueue_head(&fcc->flush_wait_queue);
684 init_llist_head(&fcc->issue_list);
685 SM_I(sbi)->fcc_info = fcc;
686 if (!test_opt(sbi, FLUSH_MERGE))
687 return 0;
688
689 init_thread:
690 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
691 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
692 if (IS_ERR(fcc->f2fs_issue_flush)) {
693 int err = PTR_ERR(fcc->f2fs_issue_flush);
694
695 fcc->f2fs_issue_flush = NULL;
696 return err;
697 }
698
699 return 0;
700 }
701
f2fs_destroy_flush_cmd_control(struct f2fs_sb_info * sbi,bool free)702 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
703 {
704 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
705
706 if (fcc && fcc->f2fs_issue_flush) {
707 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
708
709 fcc->f2fs_issue_flush = NULL;
710 kthread_stop(flush_thread);
711 }
712 if (free) {
713 kfree(fcc);
714 SM_I(sbi)->fcc_info = NULL;
715 }
716 }
717
f2fs_flush_device_cache(struct f2fs_sb_info * sbi)718 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
719 {
720 int ret = 0, i;
721
722 if (!f2fs_is_multi_device(sbi))
723 return 0;
724
725 if (test_opt(sbi, NOBARRIER))
726 return 0;
727
728 for (i = 1; i < sbi->s_ndevs; i++) {
729 int count = DEFAULT_RETRY_IO_COUNT;
730
731 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
732 continue;
733
734 do {
735 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
736 if (ret)
737 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
738 } while (ret && --count);
739
740 if (ret) {
741 f2fs_stop_checkpoint(sbi, false,
742 STOP_CP_REASON_FLUSH_FAIL);
743 break;
744 }
745
746 spin_lock(&sbi->dev_lock);
747 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
748 spin_unlock(&sbi->dev_lock);
749 }
750
751 return ret;
752 }
753
__locate_dirty_segment(struct f2fs_sb_info * sbi,unsigned int segno,enum dirty_type dirty_type)754 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
755 enum dirty_type dirty_type)
756 {
757 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
758
759 /* need not be added */
760 if (IS_CURSEG(sbi, segno))
761 return;
762
763 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
764 dirty_i->nr_dirty[dirty_type]++;
765
766 if (dirty_type == DIRTY) {
767 struct seg_entry *sentry = get_seg_entry(sbi, segno);
768 enum dirty_type t = sentry->type;
769
770 if (unlikely(t >= DIRTY)) {
771 f2fs_bug_on(sbi, 1);
772 return;
773 }
774 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
775 dirty_i->nr_dirty[t]++;
776
777 if (__is_large_section(sbi)) {
778 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
779 block_t valid_blocks =
780 get_valid_blocks(sbi, segno, true);
781
782 f2fs_bug_on(sbi,
783 (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
784 !valid_blocks) ||
785 valid_blocks == CAP_BLKS_PER_SEC(sbi));
786
787 if (!IS_CURSEC(sbi, secno))
788 set_bit(secno, dirty_i->dirty_secmap);
789 }
790 }
791 }
792
__remove_dirty_segment(struct f2fs_sb_info * sbi,unsigned int segno,enum dirty_type dirty_type)793 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
794 enum dirty_type dirty_type)
795 {
796 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
797 block_t valid_blocks;
798
799 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
800 dirty_i->nr_dirty[dirty_type]--;
801
802 if (dirty_type == DIRTY) {
803 struct seg_entry *sentry = get_seg_entry(sbi, segno);
804 enum dirty_type t = sentry->type;
805
806 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
807 dirty_i->nr_dirty[t]--;
808
809 valid_blocks = get_valid_blocks(sbi, segno, true);
810 if (valid_blocks == 0) {
811 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
812 dirty_i->victim_secmap);
813 #ifdef CONFIG_F2FS_CHECK_FS
814 clear_bit(segno, SIT_I(sbi)->invalid_segmap);
815 #endif
816 }
817 if (__is_large_section(sbi)) {
818 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
819
820 if (!valid_blocks ||
821 valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
822 clear_bit(secno, dirty_i->dirty_secmap);
823 return;
824 }
825
826 if (!IS_CURSEC(sbi, secno))
827 set_bit(secno, dirty_i->dirty_secmap);
828 }
829 }
830 }
831
832 /*
833 * Should not occur error such as -ENOMEM.
834 * Adding dirty entry into seglist is not critical operation.
835 * If a given segment is one of current working segments, it won't be added.
836 */
locate_dirty_segment(struct f2fs_sb_info * sbi,unsigned int segno)837 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
838 {
839 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
840 unsigned short valid_blocks, ckpt_valid_blocks;
841 unsigned int usable_blocks;
842
843 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
844 return;
845
846 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
847 mutex_lock(&dirty_i->seglist_lock);
848
849 valid_blocks = get_valid_blocks(sbi, segno, false);
850 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
851
852 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
853 ckpt_valid_blocks == usable_blocks)) {
854 __locate_dirty_segment(sbi, segno, PRE);
855 __remove_dirty_segment(sbi, segno, DIRTY);
856 } else if (valid_blocks < usable_blocks) {
857 __locate_dirty_segment(sbi, segno, DIRTY);
858 } else {
859 /* Recovery routine with SSR needs this */
860 __remove_dirty_segment(sbi, segno, DIRTY);
861 }
862
863 mutex_unlock(&dirty_i->seglist_lock);
864 }
865
866 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
f2fs_dirty_to_prefree(struct f2fs_sb_info * sbi)867 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
868 {
869 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
870 unsigned int segno;
871
872 mutex_lock(&dirty_i->seglist_lock);
873 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
874 if (get_valid_blocks(sbi, segno, false))
875 continue;
876 if (IS_CURSEG(sbi, segno))
877 continue;
878 __locate_dirty_segment(sbi, segno, PRE);
879 __remove_dirty_segment(sbi, segno, DIRTY);
880 }
881 mutex_unlock(&dirty_i->seglist_lock);
882 }
883
f2fs_get_unusable_blocks(struct f2fs_sb_info * sbi)884 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
885 {
886 int ovp_hole_segs =
887 (overprovision_segments(sbi) - reserved_segments(sbi));
888 block_t ovp_holes = SEGS_TO_BLKS(sbi, ovp_hole_segs);
889 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
890 block_t holes[2] = {0, 0}; /* DATA and NODE */
891 block_t unusable;
892 struct seg_entry *se;
893 unsigned int segno;
894
895 mutex_lock(&dirty_i->seglist_lock);
896 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
897 se = get_seg_entry(sbi, segno);
898 if (IS_NODESEG(se->type))
899 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
900 se->valid_blocks;
901 else
902 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
903 se->valid_blocks;
904 }
905 mutex_unlock(&dirty_i->seglist_lock);
906
907 unusable = max(holes[DATA], holes[NODE]);
908 if (unusable > ovp_holes)
909 return unusable - ovp_holes;
910 return 0;
911 }
912
f2fs_disable_cp_again(struct f2fs_sb_info * sbi,block_t unusable)913 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
914 {
915 int ovp_hole_segs =
916 (overprovision_segments(sbi) - reserved_segments(sbi));
917
918 if (F2FS_OPTION(sbi).unusable_cap_perc == 100)
919 return 0;
920 if (unusable > F2FS_OPTION(sbi).unusable_cap)
921 return -EAGAIN;
922 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
923 dirty_segments(sbi) > ovp_hole_segs)
924 return -EAGAIN;
925 if (has_not_enough_free_secs(sbi, 0, 0))
926 return -EAGAIN;
927 return 0;
928 }
929
930 /* This is only used by SBI_CP_DISABLED */
get_free_segment(struct f2fs_sb_info * sbi)931 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
932 {
933 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
934 unsigned int segno = 0;
935
936 mutex_lock(&dirty_i->seglist_lock);
937 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
938 if (get_valid_blocks(sbi, segno, false))
939 continue;
940 if (get_ckpt_valid_blocks(sbi, segno, false))
941 continue;
942 mutex_unlock(&dirty_i->seglist_lock);
943 return segno;
944 }
945 mutex_unlock(&dirty_i->seglist_lock);
946 return NULL_SEGNO;
947 }
948
__create_discard_cmd(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t lstart,block_t start,block_t len)949 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
950 struct block_device *bdev, block_t lstart,
951 block_t start, block_t len)
952 {
953 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
954 struct list_head *pend_list;
955 struct discard_cmd *dc;
956
957 f2fs_bug_on(sbi, !len);
958
959 pend_list = &dcc->pend_list[plist_idx(len)];
960
961 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
962 INIT_LIST_HEAD(&dc->list);
963 dc->bdev = bdev;
964 dc->di.lstart = lstart;
965 dc->di.start = start;
966 dc->di.len = len;
967 dc->ref = 0;
968 dc->state = D_PREP;
969 dc->queued = 0;
970 dc->error = 0;
971 init_completion(&dc->wait);
972 list_add_tail(&dc->list, pend_list);
973 spin_lock_init(&dc->lock);
974 dc->bio_ref = 0;
975 atomic_inc(&dcc->discard_cmd_cnt);
976 dcc->undiscard_blks += len;
977
978 return dc;
979 }
980
f2fs_check_discard_tree(struct f2fs_sb_info * sbi)981 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
982 {
983 #ifdef CONFIG_F2FS_CHECK_FS
984 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
985 struct rb_node *cur = rb_first_cached(&dcc->root), *next;
986 struct discard_cmd *cur_dc, *next_dc;
987
988 while (cur) {
989 next = rb_next(cur);
990 if (!next)
991 return true;
992
993 cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
994 next_dc = rb_entry(next, struct discard_cmd, rb_node);
995
996 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
997 f2fs_info(sbi, "broken discard_rbtree, "
998 "cur(%u, %u) next(%u, %u)",
999 cur_dc->di.lstart, cur_dc->di.len,
1000 next_dc->di.lstart, next_dc->di.len);
1001 return false;
1002 }
1003 cur = next;
1004 }
1005 #endif
1006 return true;
1007 }
1008
__lookup_discard_cmd(struct f2fs_sb_info * sbi,block_t blkaddr)1009 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
1010 block_t blkaddr)
1011 {
1012 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1013 struct rb_node *node = dcc->root.rb_root.rb_node;
1014 struct discard_cmd *dc;
1015
1016 while (node) {
1017 dc = rb_entry(node, struct discard_cmd, rb_node);
1018
1019 if (blkaddr < dc->di.lstart)
1020 node = node->rb_left;
1021 else if (blkaddr >= dc->di.lstart + dc->di.len)
1022 node = node->rb_right;
1023 else
1024 return dc;
1025 }
1026 return NULL;
1027 }
1028
__lookup_discard_cmd_ret(struct rb_root_cached * root,block_t blkaddr,struct discard_cmd ** prev_entry,struct discard_cmd ** next_entry,struct rb_node *** insert_p,struct rb_node ** insert_parent)1029 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1030 block_t blkaddr,
1031 struct discard_cmd **prev_entry,
1032 struct discard_cmd **next_entry,
1033 struct rb_node ***insert_p,
1034 struct rb_node **insert_parent)
1035 {
1036 struct rb_node **pnode = &root->rb_root.rb_node;
1037 struct rb_node *parent = NULL, *tmp_node;
1038 struct discard_cmd *dc;
1039
1040 *insert_p = NULL;
1041 *insert_parent = NULL;
1042 *prev_entry = NULL;
1043 *next_entry = NULL;
1044
1045 if (RB_EMPTY_ROOT(&root->rb_root))
1046 return NULL;
1047
1048 while (*pnode) {
1049 parent = *pnode;
1050 dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1051
1052 if (blkaddr < dc->di.lstart)
1053 pnode = &(*pnode)->rb_left;
1054 else if (blkaddr >= dc->di.lstart + dc->di.len)
1055 pnode = &(*pnode)->rb_right;
1056 else
1057 goto lookup_neighbors;
1058 }
1059
1060 *insert_p = pnode;
1061 *insert_parent = parent;
1062
1063 dc = rb_entry(parent, struct discard_cmd, rb_node);
1064 tmp_node = parent;
1065 if (parent && blkaddr > dc->di.lstart)
1066 tmp_node = rb_next(parent);
1067 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1068
1069 tmp_node = parent;
1070 if (parent && blkaddr < dc->di.lstart)
1071 tmp_node = rb_prev(parent);
1072 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1073 return NULL;
1074
1075 lookup_neighbors:
1076 /* lookup prev node for merging backward later */
1077 tmp_node = rb_prev(&dc->rb_node);
1078 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1079
1080 /* lookup next node for merging frontward later */
1081 tmp_node = rb_next(&dc->rb_node);
1082 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1083 return dc;
1084 }
1085
__detach_discard_cmd(struct discard_cmd_control * dcc,struct discard_cmd * dc)1086 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1087 struct discard_cmd *dc)
1088 {
1089 if (dc->state == D_DONE)
1090 atomic_sub(dc->queued, &dcc->queued_discard);
1091
1092 list_del(&dc->list);
1093 rb_erase_cached(&dc->rb_node, &dcc->root);
1094 dcc->undiscard_blks -= dc->di.len;
1095
1096 kmem_cache_free(discard_cmd_slab, dc);
1097
1098 atomic_dec(&dcc->discard_cmd_cnt);
1099 }
1100
__remove_discard_cmd(struct f2fs_sb_info * sbi,struct discard_cmd * dc)1101 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1102 struct discard_cmd *dc)
1103 {
1104 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1105 unsigned long flags;
1106
1107 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1108
1109 spin_lock_irqsave(&dc->lock, flags);
1110 if (dc->bio_ref) {
1111 spin_unlock_irqrestore(&dc->lock, flags);
1112 return;
1113 }
1114 spin_unlock_irqrestore(&dc->lock, flags);
1115
1116 f2fs_bug_on(sbi, dc->ref);
1117
1118 if (dc->error == -EOPNOTSUPP)
1119 dc->error = 0;
1120
1121 if (dc->error)
1122 f2fs_info_ratelimited(sbi,
1123 "Issue discard(%u, %u, %u) failed, ret: %d",
1124 dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1125 __detach_discard_cmd(dcc, dc);
1126 }
1127
f2fs_submit_discard_endio(struct bio * bio)1128 static void f2fs_submit_discard_endio(struct bio *bio)
1129 {
1130 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1131 unsigned long flags;
1132
1133 spin_lock_irqsave(&dc->lock, flags);
1134 if (!dc->error)
1135 dc->error = blk_status_to_errno(bio->bi_status);
1136 dc->bio_ref--;
1137 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1138 dc->state = D_DONE;
1139 complete_all(&dc->wait);
1140 }
1141 spin_unlock_irqrestore(&dc->lock, flags);
1142 bio_put(bio);
1143 }
1144
__check_sit_bitmap(struct f2fs_sb_info * sbi,block_t start,block_t end)1145 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1146 block_t start, block_t end)
1147 {
1148 #ifdef CONFIG_F2FS_CHECK_FS
1149 struct seg_entry *sentry;
1150 unsigned int segno;
1151 block_t blk = start;
1152 unsigned long offset, size, *map;
1153
1154 while (blk < end) {
1155 segno = GET_SEGNO(sbi, blk);
1156 sentry = get_seg_entry(sbi, segno);
1157 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1158
1159 if (end < START_BLOCK(sbi, segno + 1))
1160 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1161 else
1162 size = BLKS_PER_SEG(sbi);
1163 map = (unsigned long *)(sentry->cur_valid_map);
1164 offset = __find_rev_next_bit(map, size, offset);
1165 f2fs_bug_on(sbi, offset != size);
1166 blk = START_BLOCK(sbi, segno + 1);
1167 }
1168 #endif
1169 }
1170
__init_discard_policy(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy,int discard_type,unsigned int granularity)1171 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1172 struct discard_policy *dpolicy,
1173 int discard_type, unsigned int granularity)
1174 {
1175 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1176
1177 /* common policy */
1178 dpolicy->type = discard_type;
1179 dpolicy->sync = true;
1180 dpolicy->ordered = false;
1181 dpolicy->granularity = granularity;
1182
1183 dpolicy->max_requests = dcc->max_discard_request;
1184 dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1185 dpolicy->timeout = false;
1186
1187 if (discard_type == DPOLICY_BG) {
1188 dpolicy->min_interval = dcc->min_discard_issue_time;
1189 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1190 dpolicy->max_interval = dcc->max_discard_issue_time;
1191 if (dcc->discard_io_aware == DPOLICY_IO_AWARE_ENABLE)
1192 dpolicy->io_aware = true;
1193 else if (dcc->discard_io_aware == DPOLICY_IO_AWARE_DISABLE)
1194 dpolicy->io_aware = false;
1195 dpolicy->sync = false;
1196 dpolicy->ordered = true;
1197 if (utilization(sbi) > dcc->discard_urgent_util) {
1198 dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1199 if (atomic_read(&dcc->discard_cmd_cnt))
1200 dpolicy->max_interval =
1201 dcc->min_discard_issue_time;
1202 }
1203 } else if (discard_type == DPOLICY_FORCE) {
1204 dpolicy->min_interval = dcc->min_discard_issue_time;
1205 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1206 dpolicy->max_interval = dcc->max_discard_issue_time;
1207 dpolicy->io_aware = false;
1208 } else if (discard_type == DPOLICY_FSTRIM) {
1209 dpolicy->io_aware = false;
1210 } else if (discard_type == DPOLICY_UMOUNT) {
1211 dpolicy->io_aware = false;
1212 /* we need to issue all to keep CP_TRIMMED_FLAG */
1213 dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1214 dpolicy->timeout = true;
1215 }
1216 }
1217
1218 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1219 struct block_device *bdev, block_t lstart,
1220 block_t start, block_t len);
1221
1222 #ifdef CONFIG_BLK_DEV_ZONED
__submit_zone_reset_cmd(struct f2fs_sb_info * sbi,struct discard_cmd * dc,blk_opf_t flag,struct list_head * wait_list,unsigned int * issued)1223 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
1224 struct discard_cmd *dc, blk_opf_t flag,
1225 struct list_head *wait_list,
1226 unsigned int *issued)
1227 {
1228 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1229 struct block_device *bdev = dc->bdev;
1230 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
1231 unsigned long flags;
1232
1233 trace_f2fs_issue_reset_zone(bdev, dc->di.start);
1234
1235 spin_lock_irqsave(&dc->lock, flags);
1236 dc->state = D_SUBMIT;
1237 dc->bio_ref++;
1238 spin_unlock_irqrestore(&dc->lock, flags);
1239
1240 if (issued)
1241 (*issued)++;
1242
1243 atomic_inc(&dcc->queued_discard);
1244 dc->queued++;
1245 list_move_tail(&dc->list, wait_list);
1246
1247 /* sanity check on discard range */
1248 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
1249
1250 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
1251 bio->bi_private = dc;
1252 bio->bi_end_io = f2fs_submit_discard_endio;
1253 submit_bio(bio);
1254
1255 atomic_inc(&dcc->issued_discard);
1256 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE);
1257 }
1258 #endif
1259
1260 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
__submit_discard_cmd(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy,struct discard_cmd * dc,int * issued)1261 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1262 struct discard_policy *dpolicy,
1263 struct discard_cmd *dc, int *issued)
1264 {
1265 struct block_device *bdev = dc->bdev;
1266 unsigned int max_discard_blocks =
1267 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1268 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1269 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1270 &(dcc->fstrim_list) : &(dcc->wait_list);
1271 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1272 block_t lstart, start, len, total_len;
1273 int err = 0;
1274
1275 if (dc->state != D_PREP)
1276 return 0;
1277
1278 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1279 return 0;
1280
1281 #ifdef CONFIG_BLK_DEV_ZONED
1282 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) {
1283 int devi = f2fs_bdev_index(sbi, bdev);
1284
1285 if (devi < 0)
1286 return -EINVAL;
1287
1288 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1289 __submit_zone_reset_cmd(sbi, dc, flag,
1290 wait_list, issued);
1291 return 0;
1292 }
1293
1294 /*
1295 * Issue discard for conventional zones only if the device
1296 * supports discard.
1297 */
1298 if (!bdev_max_discard_sectors(bdev))
1299 return -EOPNOTSUPP;
1300 }
1301 #endif
1302
1303 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1304
1305 lstart = dc->di.lstart;
1306 start = dc->di.start;
1307 len = dc->di.len;
1308 total_len = len;
1309
1310 dc->di.len = 0;
1311
1312 while (total_len && *issued < dpolicy->max_requests && !err) {
1313 struct bio *bio = NULL;
1314 unsigned long flags;
1315 bool last = true;
1316
1317 if (len > max_discard_blocks) {
1318 len = max_discard_blocks;
1319 last = false;
1320 }
1321
1322 (*issued)++;
1323 if (*issued == dpolicy->max_requests)
1324 last = true;
1325
1326 dc->di.len += len;
1327
1328 if (time_to_inject(sbi, FAULT_DISCARD)) {
1329 err = -EIO;
1330 } else {
1331 err = __blkdev_issue_discard(bdev,
1332 SECTOR_FROM_BLOCK(start),
1333 SECTOR_FROM_BLOCK(len),
1334 GFP_NOFS, &bio);
1335 }
1336 if (err) {
1337 spin_lock_irqsave(&dc->lock, flags);
1338 if (dc->state == D_PARTIAL)
1339 dc->state = D_SUBMIT;
1340 spin_unlock_irqrestore(&dc->lock, flags);
1341
1342 break;
1343 }
1344
1345 f2fs_bug_on(sbi, !bio);
1346
1347 /*
1348 * should keep before submission to avoid D_DONE
1349 * right away
1350 */
1351 spin_lock_irqsave(&dc->lock, flags);
1352 if (last)
1353 dc->state = D_SUBMIT;
1354 else
1355 dc->state = D_PARTIAL;
1356 dc->bio_ref++;
1357 spin_unlock_irqrestore(&dc->lock, flags);
1358
1359 atomic_inc(&dcc->queued_discard);
1360 dc->queued++;
1361 list_move_tail(&dc->list, wait_list);
1362
1363 /* sanity check on discard range */
1364 __check_sit_bitmap(sbi, lstart, lstart + len);
1365
1366 bio->bi_private = dc;
1367 bio->bi_end_io = f2fs_submit_discard_endio;
1368 bio->bi_opf |= flag;
1369 submit_bio(bio);
1370
1371 atomic_inc(&dcc->issued_discard);
1372
1373 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1374
1375 lstart += len;
1376 start += len;
1377 total_len -= len;
1378 len = total_len;
1379 }
1380
1381 if (!err && len) {
1382 dcc->undiscard_blks -= len;
1383 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1384 }
1385 return err;
1386 }
1387
__insert_discard_cmd(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t lstart,block_t start,block_t len)1388 static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1389 struct block_device *bdev, block_t lstart,
1390 block_t start, block_t len)
1391 {
1392 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1393 struct rb_node **p = &dcc->root.rb_root.rb_node;
1394 struct rb_node *parent = NULL;
1395 struct discard_cmd *dc;
1396 bool leftmost = true;
1397
1398 /* look up rb tree to find parent node */
1399 while (*p) {
1400 parent = *p;
1401 dc = rb_entry(parent, struct discard_cmd, rb_node);
1402
1403 if (lstart < dc->di.lstart) {
1404 p = &(*p)->rb_left;
1405 } else if (lstart >= dc->di.lstart + dc->di.len) {
1406 p = &(*p)->rb_right;
1407 leftmost = false;
1408 } else {
1409 /* Let's skip to add, if exists */
1410 return;
1411 }
1412 }
1413
1414 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1415
1416 rb_link_node(&dc->rb_node, parent, p);
1417 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1418 }
1419
__relocate_discard_cmd(struct discard_cmd_control * dcc,struct discard_cmd * dc)1420 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1421 struct discard_cmd *dc)
1422 {
1423 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1424 }
1425
__punch_discard_cmd(struct f2fs_sb_info * sbi,struct discard_cmd * dc,block_t blkaddr)1426 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1427 struct discard_cmd *dc, block_t blkaddr)
1428 {
1429 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1430 struct discard_info di = dc->di;
1431 bool modified = false;
1432
1433 if (dc->state == D_DONE || dc->di.len == 1) {
1434 __remove_discard_cmd(sbi, dc);
1435 return;
1436 }
1437
1438 dcc->undiscard_blks -= di.len;
1439
1440 if (blkaddr > di.lstart) {
1441 dc->di.len = blkaddr - dc->di.lstart;
1442 dcc->undiscard_blks += dc->di.len;
1443 __relocate_discard_cmd(dcc, dc);
1444 modified = true;
1445 }
1446
1447 if (blkaddr < di.lstart + di.len - 1) {
1448 if (modified) {
1449 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1450 di.start + blkaddr + 1 - di.lstart,
1451 di.lstart + di.len - 1 - blkaddr);
1452 } else {
1453 dc->di.lstart++;
1454 dc->di.len--;
1455 dc->di.start++;
1456 dcc->undiscard_blks += dc->di.len;
1457 __relocate_discard_cmd(dcc, dc);
1458 }
1459 }
1460 }
1461
__update_discard_tree_range(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t lstart,block_t start,block_t len)1462 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1463 struct block_device *bdev, block_t lstart,
1464 block_t start, block_t len)
1465 {
1466 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1467 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1468 struct discard_cmd *dc;
1469 struct discard_info di = {0};
1470 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1471 unsigned int max_discard_blocks =
1472 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1473 block_t end = lstart + len;
1474
1475 dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1476 &prev_dc, &next_dc, &insert_p, &insert_parent);
1477 if (dc)
1478 prev_dc = dc;
1479
1480 if (!prev_dc) {
1481 di.lstart = lstart;
1482 di.len = next_dc ? next_dc->di.lstart - lstart : len;
1483 di.len = min(di.len, len);
1484 di.start = start;
1485 }
1486
1487 while (1) {
1488 struct rb_node *node;
1489 bool merged = false;
1490 struct discard_cmd *tdc = NULL;
1491
1492 if (prev_dc) {
1493 di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1494 if (di.lstart < lstart)
1495 di.lstart = lstart;
1496 if (di.lstart >= end)
1497 break;
1498
1499 if (!next_dc || next_dc->di.lstart > end)
1500 di.len = end - di.lstart;
1501 else
1502 di.len = next_dc->di.lstart - di.lstart;
1503 di.start = start + di.lstart - lstart;
1504 }
1505
1506 if (!di.len)
1507 goto next;
1508
1509 if (prev_dc && prev_dc->state == D_PREP &&
1510 prev_dc->bdev == bdev &&
1511 __is_discard_back_mergeable(&di, &prev_dc->di,
1512 max_discard_blocks)) {
1513 prev_dc->di.len += di.len;
1514 dcc->undiscard_blks += di.len;
1515 __relocate_discard_cmd(dcc, prev_dc);
1516 di = prev_dc->di;
1517 tdc = prev_dc;
1518 merged = true;
1519 }
1520
1521 if (next_dc && next_dc->state == D_PREP &&
1522 next_dc->bdev == bdev &&
1523 __is_discard_front_mergeable(&di, &next_dc->di,
1524 max_discard_blocks)) {
1525 next_dc->di.lstart = di.lstart;
1526 next_dc->di.len += di.len;
1527 next_dc->di.start = di.start;
1528 dcc->undiscard_blks += di.len;
1529 __relocate_discard_cmd(dcc, next_dc);
1530 if (tdc)
1531 __remove_discard_cmd(sbi, tdc);
1532 merged = true;
1533 }
1534
1535 if (!merged)
1536 __insert_discard_cmd(sbi, bdev,
1537 di.lstart, di.start, di.len);
1538 next:
1539 prev_dc = next_dc;
1540 if (!prev_dc)
1541 break;
1542
1543 node = rb_next(&prev_dc->rb_node);
1544 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1545 }
1546 }
1547
1548 #ifdef CONFIG_BLK_DEV_ZONED
__queue_zone_reset_cmd(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t blkstart,block_t lblkstart,block_t blklen)1549 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
1550 struct block_device *bdev, block_t blkstart, block_t lblkstart,
1551 block_t blklen)
1552 {
1553 trace_f2fs_queue_reset_zone(bdev, blkstart);
1554
1555 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1556 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
1557 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1558 }
1559 #endif
1560
__queue_discard_cmd(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t blkstart,block_t blklen)1561 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1562 struct block_device *bdev, block_t blkstart, block_t blklen)
1563 {
1564 block_t lblkstart = blkstart;
1565
1566 if (!f2fs_bdev_support_discard(bdev))
1567 return;
1568
1569 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1570
1571 if (f2fs_is_multi_device(sbi)) {
1572 int devi = f2fs_target_device_index(sbi, blkstart);
1573
1574 blkstart -= FDEV(devi).start_blk;
1575 }
1576 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1577 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1578 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1579 }
1580
__issue_discard_cmd_orderly(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy,int * issued)1581 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1582 struct discard_policy *dpolicy, int *issued)
1583 {
1584 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1585 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1586 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1587 struct discard_cmd *dc;
1588 struct blk_plug plug;
1589 bool io_interrupted = false;
1590
1591 mutex_lock(&dcc->cmd_lock);
1592 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1593 &prev_dc, &next_dc, &insert_p, &insert_parent);
1594 if (!dc)
1595 dc = next_dc;
1596
1597 blk_start_plug(&plug);
1598
1599 while (dc) {
1600 struct rb_node *node;
1601 int err = 0;
1602
1603 if (dc->state != D_PREP)
1604 goto next;
1605
1606 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1607 io_interrupted = true;
1608 break;
1609 }
1610
1611 dcc->next_pos = dc->di.lstart + dc->di.len;
1612 err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1613
1614 if (*issued >= dpolicy->max_requests)
1615 break;
1616 next:
1617 node = rb_next(&dc->rb_node);
1618 if (err)
1619 __remove_discard_cmd(sbi, dc);
1620 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1621 }
1622
1623 blk_finish_plug(&plug);
1624
1625 if (!dc)
1626 dcc->next_pos = 0;
1627
1628 mutex_unlock(&dcc->cmd_lock);
1629
1630 if (!(*issued) && io_interrupted)
1631 *issued = -1;
1632 }
1633 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1634 struct discard_policy *dpolicy);
1635
__issue_discard_cmd(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy)1636 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1637 struct discard_policy *dpolicy)
1638 {
1639 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1640 struct list_head *pend_list;
1641 struct discard_cmd *dc, *tmp;
1642 struct blk_plug plug;
1643 int i, issued;
1644 bool io_interrupted = false;
1645
1646 if (dpolicy->timeout)
1647 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1648
1649 retry:
1650 issued = 0;
1651 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1652 if (dpolicy->timeout &&
1653 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1654 break;
1655
1656 if (i + 1 < dpolicy->granularity)
1657 break;
1658
1659 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1660 __issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1661 return issued;
1662 }
1663
1664 pend_list = &dcc->pend_list[i];
1665
1666 mutex_lock(&dcc->cmd_lock);
1667 if (list_empty(pend_list))
1668 goto next;
1669 if (unlikely(dcc->rbtree_check))
1670 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1671 blk_start_plug(&plug);
1672 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1673 f2fs_bug_on(sbi, dc->state != D_PREP);
1674
1675 if (dpolicy->timeout &&
1676 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1677 break;
1678
1679 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1680 !is_idle(sbi, DISCARD_TIME)) {
1681 io_interrupted = true;
1682 break;
1683 }
1684
1685 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1686
1687 if (issued >= dpolicy->max_requests)
1688 break;
1689 }
1690 blk_finish_plug(&plug);
1691 next:
1692 mutex_unlock(&dcc->cmd_lock);
1693
1694 if (issued >= dpolicy->max_requests || io_interrupted)
1695 break;
1696 }
1697
1698 if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1699 __wait_all_discard_cmd(sbi, dpolicy);
1700 goto retry;
1701 }
1702
1703 if (!issued && io_interrupted)
1704 issued = -1;
1705
1706 return issued;
1707 }
1708
__drop_discard_cmd(struct f2fs_sb_info * sbi)1709 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1710 {
1711 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1712 struct list_head *pend_list;
1713 struct discard_cmd *dc, *tmp;
1714 int i;
1715 bool dropped = false;
1716
1717 mutex_lock(&dcc->cmd_lock);
1718 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1719 pend_list = &dcc->pend_list[i];
1720 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1721 f2fs_bug_on(sbi, dc->state != D_PREP);
1722 __remove_discard_cmd(sbi, dc);
1723 dropped = true;
1724 }
1725 }
1726 mutex_unlock(&dcc->cmd_lock);
1727
1728 return dropped;
1729 }
1730
f2fs_drop_discard_cmd(struct f2fs_sb_info * sbi)1731 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1732 {
1733 __drop_discard_cmd(sbi);
1734 }
1735
__wait_one_discard_bio(struct f2fs_sb_info * sbi,struct discard_cmd * dc)1736 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1737 struct discard_cmd *dc)
1738 {
1739 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1740 unsigned int len = 0;
1741
1742 wait_for_completion_io(&dc->wait);
1743 mutex_lock(&dcc->cmd_lock);
1744 f2fs_bug_on(sbi, dc->state != D_DONE);
1745 dc->ref--;
1746 if (!dc->ref) {
1747 if (!dc->error)
1748 len = dc->di.len;
1749 __remove_discard_cmd(sbi, dc);
1750 }
1751 mutex_unlock(&dcc->cmd_lock);
1752
1753 return len;
1754 }
1755
__wait_discard_cmd_range(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy,block_t start,block_t end)1756 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1757 struct discard_policy *dpolicy,
1758 block_t start, block_t end)
1759 {
1760 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1761 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1762 &(dcc->fstrim_list) : &(dcc->wait_list);
1763 struct discard_cmd *dc = NULL, *iter, *tmp;
1764 unsigned int trimmed = 0;
1765
1766 next:
1767 dc = NULL;
1768
1769 mutex_lock(&dcc->cmd_lock);
1770 list_for_each_entry_safe(iter, tmp, wait_list, list) {
1771 if (iter->di.lstart + iter->di.len <= start ||
1772 end <= iter->di.lstart)
1773 continue;
1774 if (iter->di.len < dpolicy->granularity)
1775 continue;
1776 if (iter->state == D_DONE && !iter->ref) {
1777 wait_for_completion_io(&iter->wait);
1778 if (!iter->error)
1779 trimmed += iter->di.len;
1780 __remove_discard_cmd(sbi, iter);
1781 } else {
1782 iter->ref++;
1783 dc = iter;
1784 break;
1785 }
1786 }
1787 mutex_unlock(&dcc->cmd_lock);
1788
1789 if (dc) {
1790 trimmed += __wait_one_discard_bio(sbi, dc);
1791 goto next;
1792 }
1793
1794 return trimmed;
1795 }
1796
__wait_all_discard_cmd(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy)1797 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1798 struct discard_policy *dpolicy)
1799 {
1800 struct discard_policy dp;
1801 unsigned int discard_blks;
1802
1803 if (dpolicy)
1804 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1805
1806 /* wait all */
1807 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1808 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1809 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1810 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1811
1812 return discard_blks;
1813 }
1814
1815 /* This should be covered by global mutex, &sit_i->sentry_lock */
f2fs_wait_discard_bio(struct f2fs_sb_info * sbi,block_t blkaddr)1816 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1817 {
1818 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1819 struct discard_cmd *dc;
1820 bool need_wait = false;
1821
1822 mutex_lock(&dcc->cmd_lock);
1823 dc = __lookup_discard_cmd(sbi, blkaddr);
1824 #ifdef CONFIG_BLK_DEV_ZONED
1825 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
1826 int devi = f2fs_bdev_index(sbi, dc->bdev);
1827
1828 if (devi < 0) {
1829 mutex_unlock(&dcc->cmd_lock);
1830 return;
1831 }
1832
1833 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1834 /* force submit zone reset */
1835 if (dc->state == D_PREP)
1836 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
1837 &dcc->wait_list, NULL);
1838 dc->ref++;
1839 mutex_unlock(&dcc->cmd_lock);
1840 /* wait zone reset */
1841 __wait_one_discard_bio(sbi, dc);
1842 return;
1843 }
1844 }
1845 #endif
1846 if (dc) {
1847 if (dc->state == D_PREP) {
1848 __punch_discard_cmd(sbi, dc, blkaddr);
1849 } else {
1850 dc->ref++;
1851 need_wait = true;
1852 }
1853 }
1854 mutex_unlock(&dcc->cmd_lock);
1855
1856 if (need_wait)
1857 __wait_one_discard_bio(sbi, dc);
1858 }
1859
f2fs_stop_discard_thread(struct f2fs_sb_info * sbi)1860 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1861 {
1862 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1863
1864 if (dcc && dcc->f2fs_issue_discard) {
1865 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1866
1867 dcc->f2fs_issue_discard = NULL;
1868 kthread_stop(discard_thread);
1869 }
1870 }
1871
1872 /**
1873 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1874 * @sbi: the f2fs_sb_info data for discard cmd to issue
1875 *
1876 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1877 *
1878 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1879 */
f2fs_issue_discard_timeout(struct f2fs_sb_info * sbi)1880 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1881 {
1882 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1883 struct discard_policy dpolicy;
1884 bool dropped;
1885
1886 if (!atomic_read(&dcc->discard_cmd_cnt))
1887 return true;
1888
1889 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1890 dcc->discard_granularity);
1891 __issue_discard_cmd(sbi, &dpolicy);
1892 dropped = __drop_discard_cmd(sbi);
1893
1894 /* just to make sure there is no pending discard commands */
1895 __wait_all_discard_cmd(sbi, NULL);
1896
1897 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1898 return !dropped;
1899 }
1900
issue_discard_thread(void * data)1901 static int issue_discard_thread(void *data)
1902 {
1903 struct f2fs_sb_info *sbi = data;
1904 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1905 wait_queue_head_t *q = &dcc->discard_wait_queue;
1906 struct discard_policy dpolicy;
1907 unsigned int wait_ms = dcc->min_discard_issue_time;
1908 int issued;
1909
1910 set_freezable();
1911
1912 do {
1913 wait_event_freezable_timeout(*q,
1914 kthread_should_stop() || dcc->discard_wake,
1915 msecs_to_jiffies(wait_ms));
1916
1917 if (sbi->gc_mode == GC_URGENT_HIGH ||
1918 !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1919 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1920 MIN_DISCARD_GRANULARITY);
1921 else
1922 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1923 dcc->discard_granularity);
1924
1925 if (dcc->discard_wake)
1926 dcc->discard_wake = false;
1927
1928 /* clean up pending candidates before going to sleep */
1929 if (atomic_read(&dcc->queued_discard))
1930 __wait_all_discard_cmd(sbi, NULL);
1931
1932 if (f2fs_readonly(sbi->sb))
1933 continue;
1934 if (kthread_should_stop())
1935 return 0;
1936 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1937 !atomic_read(&dcc->discard_cmd_cnt)) {
1938 wait_ms = dpolicy.max_interval;
1939 continue;
1940 }
1941
1942 sb_start_intwrite(sbi->sb);
1943
1944 issued = __issue_discard_cmd(sbi, &dpolicy);
1945 if (issued > 0) {
1946 __wait_all_discard_cmd(sbi, &dpolicy);
1947 wait_ms = dpolicy.min_interval;
1948 } else if (issued == -1) {
1949 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1950 if (!wait_ms)
1951 wait_ms = dpolicy.mid_interval;
1952 } else {
1953 wait_ms = dpolicy.max_interval;
1954 }
1955 if (!atomic_read(&dcc->discard_cmd_cnt))
1956 wait_ms = dpolicy.max_interval;
1957
1958 sb_end_intwrite(sbi->sb);
1959
1960 } while (!kthread_should_stop());
1961 return 0;
1962 }
1963
1964 #ifdef CONFIG_BLK_DEV_ZONED
__f2fs_issue_discard_zone(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t blkstart,block_t blklen)1965 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1966 struct block_device *bdev, block_t blkstart, block_t blklen)
1967 {
1968 sector_t sector, nr_sects;
1969 block_t lblkstart = blkstart;
1970 int devi = 0;
1971 u64 remainder = 0;
1972
1973 if (f2fs_is_multi_device(sbi)) {
1974 devi = f2fs_target_device_index(sbi, blkstart);
1975 if (blkstart < FDEV(devi).start_blk ||
1976 blkstart > FDEV(devi).end_blk) {
1977 f2fs_err(sbi, "Invalid block %x", blkstart);
1978 return -EIO;
1979 }
1980 blkstart -= FDEV(devi).start_blk;
1981 }
1982
1983 /* For sequential zones, reset the zone write pointer */
1984 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1985 sector = SECTOR_FROM_BLOCK(blkstart);
1986 nr_sects = SECTOR_FROM_BLOCK(blklen);
1987 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
1988
1989 if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
1990 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1991 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1992 blkstart, blklen);
1993 return -EIO;
1994 }
1995
1996 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) {
1997 unsigned int nofs_flags;
1998 int ret;
1999
2000 trace_f2fs_issue_reset_zone(bdev, blkstart);
2001 nofs_flags = memalloc_nofs_save();
2002 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
2003 sector, nr_sects);
2004 memalloc_nofs_restore(nofs_flags);
2005 return ret;
2006 }
2007
2008 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen);
2009 return 0;
2010 }
2011
2012 /* For conventional zones, use regular discard if supported */
2013 __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
2014 return 0;
2015 }
2016 #endif
2017
__issue_discard_async(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t blkstart,block_t blklen)2018 static int __issue_discard_async(struct f2fs_sb_info *sbi,
2019 struct block_device *bdev, block_t blkstart, block_t blklen)
2020 {
2021 #ifdef CONFIG_BLK_DEV_ZONED
2022 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
2023 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
2024 #endif
2025 __queue_discard_cmd(sbi, bdev, blkstart, blklen);
2026 return 0;
2027 }
2028
f2fs_issue_discard(struct f2fs_sb_info * sbi,block_t blkstart,block_t blklen)2029 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
2030 block_t blkstart, block_t blklen)
2031 {
2032 sector_t start = blkstart, len = 0;
2033 struct block_device *bdev;
2034 struct seg_entry *se;
2035 unsigned int offset;
2036 block_t i;
2037 int err = 0;
2038
2039 bdev = f2fs_target_device(sbi, blkstart, NULL);
2040
2041 for (i = blkstart; i < blkstart + blklen; i++, len++) {
2042 if (i != start) {
2043 struct block_device *bdev2 =
2044 f2fs_target_device(sbi, i, NULL);
2045
2046 if (bdev2 != bdev) {
2047 err = __issue_discard_async(sbi, bdev,
2048 start, len);
2049 if (err)
2050 return err;
2051 bdev = bdev2;
2052 start = i;
2053 len = 0;
2054 }
2055 }
2056
2057 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
2058 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
2059
2060 if (f2fs_block_unit_discard(sbi) &&
2061 !f2fs_test_and_set_bit(offset, se->discard_map))
2062 sbi->discard_blks--;
2063 }
2064
2065 if (len)
2066 err = __issue_discard_async(sbi, bdev, start, len);
2067 return err;
2068 }
2069
add_discard_addrs(struct f2fs_sb_info * sbi,struct cp_control * cpc,bool check_only)2070 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
2071 bool check_only)
2072 {
2073 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2074 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
2075 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2076 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2077 unsigned long *discard_map = (unsigned long *)se->discard_map;
2078 unsigned long *dmap = SIT_I(sbi)->tmp_map;
2079 unsigned int start = 0, end = -1;
2080 bool force = (cpc->reason & CP_DISCARD);
2081 struct discard_entry *de = NULL;
2082 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
2083 int i;
2084
2085 if (se->valid_blocks == BLKS_PER_SEG(sbi) ||
2086 !f2fs_hw_support_discard(sbi) ||
2087 !f2fs_block_unit_discard(sbi))
2088 return false;
2089
2090 if (!force) {
2091 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
2092 SM_I(sbi)->dcc_info->nr_discards >=
2093 SM_I(sbi)->dcc_info->max_discards)
2094 return false;
2095 }
2096
2097 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
2098 for (i = 0; i < entries; i++)
2099 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
2100 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
2101
2102 while (force || SM_I(sbi)->dcc_info->nr_discards <=
2103 SM_I(sbi)->dcc_info->max_discards) {
2104 start = __find_rev_next_bit(dmap, BLKS_PER_SEG(sbi), end + 1);
2105 if (start >= BLKS_PER_SEG(sbi))
2106 break;
2107
2108 end = __find_rev_next_zero_bit(dmap,
2109 BLKS_PER_SEG(sbi), start + 1);
2110 if (force && start && end != BLKS_PER_SEG(sbi) &&
2111 (end - start) < cpc->trim_minlen)
2112 continue;
2113
2114 if (check_only)
2115 return true;
2116
2117 if (!de) {
2118 de = f2fs_kmem_cache_alloc(discard_entry_slab,
2119 GFP_F2FS_ZERO, true, NULL);
2120 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
2121 list_add_tail(&de->list, head);
2122 }
2123
2124 for (i = start; i < end; i++)
2125 __set_bit_le(i, (void *)de->discard_map);
2126
2127 SM_I(sbi)->dcc_info->nr_discards += end - start;
2128 }
2129 return false;
2130 }
2131
release_discard_addr(struct discard_entry * entry)2132 static void release_discard_addr(struct discard_entry *entry)
2133 {
2134 list_del(&entry->list);
2135 kmem_cache_free(discard_entry_slab, entry);
2136 }
2137
f2fs_release_discard_addrs(struct f2fs_sb_info * sbi)2138 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2139 {
2140 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2141 struct discard_entry *entry, *this;
2142
2143 /* drop caches */
2144 list_for_each_entry_safe(entry, this, head, list)
2145 release_discard_addr(entry);
2146 }
2147
2148 /*
2149 * Should call f2fs_clear_prefree_segments after checkpoint is done.
2150 */
set_prefree_as_free_segments(struct f2fs_sb_info * sbi)2151 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2152 {
2153 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2154 unsigned int segno;
2155
2156 mutex_lock(&dirty_i->seglist_lock);
2157 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2158 __set_test_and_free(sbi, segno, false);
2159 mutex_unlock(&dirty_i->seglist_lock);
2160 }
2161
f2fs_clear_prefree_segments(struct f2fs_sb_info * sbi,struct cp_control * cpc)2162 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2163 struct cp_control *cpc)
2164 {
2165 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2166 struct list_head *head = &dcc->entry_list;
2167 struct discard_entry *entry, *this;
2168 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2169 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2170 unsigned int start = 0, end = -1;
2171 unsigned int secno, start_segno;
2172 bool force = (cpc->reason & CP_DISCARD);
2173 bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2174 DISCARD_UNIT_SECTION;
2175
2176 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2177 section_alignment = true;
2178
2179 mutex_lock(&dirty_i->seglist_lock);
2180
2181 while (1) {
2182 int i;
2183
2184 if (section_alignment && end != -1)
2185 end--;
2186 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2187 if (start >= MAIN_SEGS(sbi))
2188 break;
2189 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2190 start + 1);
2191
2192 if (section_alignment) {
2193 start = rounddown(start, SEGS_PER_SEC(sbi));
2194 end = roundup(end, SEGS_PER_SEC(sbi));
2195 }
2196
2197 for (i = start; i < end; i++) {
2198 if (test_and_clear_bit(i, prefree_map))
2199 dirty_i->nr_dirty[PRE]--;
2200 }
2201
2202 if (!f2fs_realtime_discard_enable(sbi))
2203 continue;
2204
2205 if (force && start >= cpc->trim_start &&
2206 (end - 1) <= cpc->trim_end)
2207 continue;
2208
2209 /* Should cover 2MB zoned device for zone-based reset */
2210 if (!f2fs_sb_has_blkzoned(sbi) &&
2211 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2212 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2213 SEGS_TO_BLKS(sbi, end - start));
2214 continue;
2215 }
2216 next:
2217 secno = GET_SEC_FROM_SEG(sbi, start);
2218 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2219 if (!IS_CURSEC(sbi, secno) &&
2220 !get_valid_blocks(sbi, start, true))
2221 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2222 BLKS_PER_SEC(sbi));
2223
2224 start = start_segno + SEGS_PER_SEC(sbi);
2225 if (start < end)
2226 goto next;
2227 else
2228 end = start - 1;
2229 }
2230 mutex_unlock(&dirty_i->seglist_lock);
2231
2232 if (!f2fs_block_unit_discard(sbi))
2233 goto wakeup;
2234
2235 /* send small discards */
2236 list_for_each_entry_safe(entry, this, head, list) {
2237 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2238 bool is_valid = test_bit_le(0, entry->discard_map);
2239
2240 find_next:
2241 if (is_valid) {
2242 next_pos = find_next_zero_bit_le(entry->discard_map,
2243 BLKS_PER_SEG(sbi), cur_pos);
2244 len = next_pos - cur_pos;
2245
2246 if (f2fs_sb_has_blkzoned(sbi) ||
2247 (force && len < cpc->trim_minlen))
2248 goto skip;
2249
2250 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2251 len);
2252 total_len += len;
2253 } else {
2254 next_pos = find_next_bit_le(entry->discard_map,
2255 BLKS_PER_SEG(sbi), cur_pos);
2256 }
2257 skip:
2258 cur_pos = next_pos;
2259 is_valid = !is_valid;
2260
2261 if (cur_pos < BLKS_PER_SEG(sbi))
2262 goto find_next;
2263
2264 release_discard_addr(entry);
2265 dcc->nr_discards -= total_len;
2266 }
2267
2268 wakeup:
2269 wake_up_discard_thread(sbi, false);
2270 }
2271
f2fs_start_discard_thread(struct f2fs_sb_info * sbi)2272 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2273 {
2274 dev_t dev = sbi->sb->s_bdev->bd_dev;
2275 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2276 int err = 0;
2277
2278 if (f2fs_sb_has_readonly(sbi)) {
2279 f2fs_info(sbi,
2280 "Skip to start discard thread for readonly image");
2281 return 0;
2282 }
2283
2284 if (!f2fs_realtime_discard_enable(sbi))
2285 return 0;
2286
2287 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2288 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2289 if (IS_ERR(dcc->f2fs_issue_discard)) {
2290 err = PTR_ERR(dcc->f2fs_issue_discard);
2291 dcc->f2fs_issue_discard = NULL;
2292 }
2293
2294 return err;
2295 }
2296
create_discard_cmd_control(struct f2fs_sb_info * sbi)2297 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2298 {
2299 struct discard_cmd_control *dcc;
2300 int err = 0, i;
2301
2302 if (SM_I(sbi)->dcc_info) {
2303 dcc = SM_I(sbi)->dcc_info;
2304 goto init_thread;
2305 }
2306
2307 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2308 if (!dcc)
2309 return -ENOMEM;
2310
2311 dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2312 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2313 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2314 dcc->discard_io_aware = DPOLICY_IO_AWARE_ENABLE;
2315 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2316 dcc->discard_granularity = BLKS_PER_SEG(sbi);
2317 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2318 dcc->discard_granularity = BLKS_PER_SEC(sbi);
2319
2320 INIT_LIST_HEAD(&dcc->entry_list);
2321 for (i = 0; i < MAX_PLIST_NUM; i++)
2322 INIT_LIST_HEAD(&dcc->pend_list[i]);
2323 INIT_LIST_HEAD(&dcc->wait_list);
2324 INIT_LIST_HEAD(&dcc->fstrim_list);
2325 mutex_init(&dcc->cmd_lock);
2326 atomic_set(&dcc->issued_discard, 0);
2327 atomic_set(&dcc->queued_discard, 0);
2328 atomic_set(&dcc->discard_cmd_cnt, 0);
2329 dcc->nr_discards = 0;
2330 dcc->max_discards = SEGS_TO_BLKS(sbi, MAIN_SEGS(sbi));
2331 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2332 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2333 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2334 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2335 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2336 dcc->undiscard_blks = 0;
2337 dcc->next_pos = 0;
2338 dcc->root = RB_ROOT_CACHED;
2339 dcc->rbtree_check = false;
2340
2341 init_waitqueue_head(&dcc->discard_wait_queue);
2342 SM_I(sbi)->dcc_info = dcc;
2343 init_thread:
2344 err = f2fs_start_discard_thread(sbi);
2345 if (err) {
2346 kfree(dcc);
2347 SM_I(sbi)->dcc_info = NULL;
2348 }
2349
2350 return err;
2351 }
2352
destroy_discard_cmd_control(struct f2fs_sb_info * sbi)2353 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2354 {
2355 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2356
2357 if (!dcc)
2358 return;
2359
2360 f2fs_stop_discard_thread(sbi);
2361
2362 /*
2363 * Recovery can cache discard commands, so in error path of
2364 * fill_super(), it needs to give a chance to handle them.
2365 */
2366 f2fs_issue_discard_timeout(sbi);
2367
2368 kfree(dcc);
2369 SM_I(sbi)->dcc_info = NULL;
2370 }
2371
__mark_sit_entry_dirty(struct f2fs_sb_info * sbi,unsigned int segno)2372 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2373 {
2374 struct sit_info *sit_i = SIT_I(sbi);
2375
2376 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2377 sit_i->dirty_sentries++;
2378 return false;
2379 }
2380
2381 return true;
2382 }
2383
__set_sit_entry_type(struct f2fs_sb_info * sbi,int type,unsigned int segno,int modified)2384 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2385 unsigned int segno, int modified)
2386 {
2387 struct seg_entry *se = get_seg_entry(sbi, segno);
2388
2389 se->type = type;
2390 if (modified)
2391 __mark_sit_entry_dirty(sbi, segno);
2392 }
2393
get_segment_mtime(struct f2fs_sb_info * sbi,block_t blkaddr)2394 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2395 block_t blkaddr)
2396 {
2397 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2398
2399 if (segno == NULL_SEGNO)
2400 return 0;
2401 return get_seg_entry(sbi, segno)->mtime;
2402 }
2403
update_segment_mtime(struct f2fs_sb_info * sbi,block_t blkaddr,unsigned long long old_mtime)2404 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2405 unsigned long long old_mtime)
2406 {
2407 struct seg_entry *se;
2408 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2409 unsigned long long ctime = get_mtime(sbi, false);
2410 unsigned long long mtime = old_mtime ? old_mtime : ctime;
2411
2412 if (segno == NULL_SEGNO)
2413 return;
2414
2415 se = get_seg_entry(sbi, segno);
2416
2417 if (!se->mtime)
2418 se->mtime = mtime;
2419 else
2420 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2421 se->valid_blocks + 1);
2422
2423 if (ctime > SIT_I(sbi)->max_mtime)
2424 SIT_I(sbi)->max_mtime = ctime;
2425 }
2426
update_sit_entry(struct f2fs_sb_info * sbi,block_t blkaddr,int del)2427 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2428 {
2429 struct seg_entry *se;
2430 unsigned int segno, offset;
2431 long int new_vblocks;
2432 bool exist;
2433 #ifdef CONFIG_F2FS_CHECK_FS
2434 bool mir_exist;
2435 #endif
2436
2437 segno = GET_SEGNO(sbi, blkaddr);
2438 if (segno == NULL_SEGNO)
2439 return;
2440
2441 se = get_seg_entry(sbi, segno);
2442 new_vblocks = se->valid_blocks + del;
2443 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2444
2445 f2fs_bug_on(sbi, (new_vblocks < 0 ||
2446 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2447
2448 se->valid_blocks = new_vblocks;
2449
2450 /* Update valid block bitmap */
2451 if (del > 0) {
2452 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2453 #ifdef CONFIG_F2FS_CHECK_FS
2454 mir_exist = f2fs_test_and_set_bit(offset,
2455 se->cur_valid_map_mir);
2456 if (unlikely(exist != mir_exist)) {
2457 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2458 blkaddr, exist);
2459 f2fs_bug_on(sbi, 1);
2460 }
2461 #endif
2462 if (unlikely(exist)) {
2463 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2464 blkaddr);
2465 f2fs_bug_on(sbi, 1);
2466 se->valid_blocks--;
2467 del = 0;
2468 }
2469
2470 if (f2fs_block_unit_discard(sbi) &&
2471 !f2fs_test_and_set_bit(offset, se->discard_map))
2472 sbi->discard_blks--;
2473
2474 /*
2475 * SSR should never reuse block which is checkpointed
2476 * or newly invalidated.
2477 */
2478 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2479 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2480 se->ckpt_valid_blocks++;
2481 }
2482 } else {
2483 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2484 #ifdef CONFIG_F2FS_CHECK_FS
2485 mir_exist = f2fs_test_and_clear_bit(offset,
2486 se->cur_valid_map_mir);
2487 if (unlikely(exist != mir_exist)) {
2488 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2489 blkaddr, exist);
2490 f2fs_bug_on(sbi, 1);
2491 }
2492 #endif
2493 if (unlikely(!exist)) {
2494 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2495 blkaddr);
2496 f2fs_bug_on(sbi, 1);
2497 se->valid_blocks++;
2498 del = 0;
2499 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2500 /*
2501 * If checkpoints are off, we must not reuse data that
2502 * was used in the previous checkpoint. If it was used
2503 * before, we must track that to know how much space we
2504 * really have.
2505 */
2506 if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2507 spin_lock(&sbi->stat_lock);
2508 sbi->unusable_block_count++;
2509 spin_unlock(&sbi->stat_lock);
2510 }
2511 }
2512
2513 if (f2fs_block_unit_discard(sbi) &&
2514 f2fs_test_and_clear_bit(offset, se->discard_map))
2515 sbi->discard_blks++;
2516 }
2517 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2518 se->ckpt_valid_blocks += del;
2519
2520 __mark_sit_entry_dirty(sbi, segno);
2521
2522 /* update total number of valid blocks to be written in ckpt area */
2523 SIT_I(sbi)->written_valid_blocks += del;
2524
2525 if (__is_large_section(sbi))
2526 get_sec_entry(sbi, segno)->valid_blocks += del;
2527 }
2528
f2fs_invalidate_blocks(struct f2fs_sb_info * sbi,block_t addr)2529 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2530 {
2531 unsigned int segno = GET_SEGNO(sbi, addr);
2532 struct sit_info *sit_i = SIT_I(sbi);
2533
2534 f2fs_bug_on(sbi, addr == NULL_ADDR);
2535 if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2536 return;
2537
2538 f2fs_invalidate_internal_cache(sbi, addr);
2539
2540 /* add it into sit main buffer */
2541 down_write(&sit_i->sentry_lock);
2542
2543 update_segment_mtime(sbi, addr, 0);
2544 update_sit_entry(sbi, addr, -1);
2545
2546 /* add it into dirty seglist */
2547 locate_dirty_segment(sbi, segno);
2548
2549 up_write(&sit_i->sentry_lock);
2550 }
2551
f2fs_is_checkpointed_data(struct f2fs_sb_info * sbi,block_t blkaddr)2552 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2553 {
2554 struct sit_info *sit_i = SIT_I(sbi);
2555 unsigned int segno, offset;
2556 struct seg_entry *se;
2557 bool is_cp = false;
2558
2559 if (!__is_valid_data_blkaddr(blkaddr))
2560 return true;
2561
2562 down_read(&sit_i->sentry_lock);
2563
2564 segno = GET_SEGNO(sbi, blkaddr);
2565 se = get_seg_entry(sbi, segno);
2566 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2567
2568 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2569 is_cp = true;
2570
2571 up_read(&sit_i->sentry_lock);
2572
2573 return is_cp;
2574 }
2575
f2fs_curseg_valid_blocks(struct f2fs_sb_info * sbi,int type)2576 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2577 {
2578 struct curseg_info *curseg = CURSEG_I(sbi, type);
2579
2580 if (sbi->ckpt->alloc_type[type] == SSR)
2581 return BLKS_PER_SEG(sbi);
2582 return curseg->next_blkoff;
2583 }
2584
2585 /*
2586 * Calculate the number of current summary pages for writing
2587 */
f2fs_npages_for_summary_flush(struct f2fs_sb_info * sbi,bool for_ra)2588 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2589 {
2590 int valid_sum_count = 0;
2591 int i, sum_in_page;
2592
2593 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2594 if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2595 valid_sum_count +=
2596 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2597 else
2598 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2599 }
2600
2601 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2602 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2603 if (valid_sum_count <= sum_in_page)
2604 return 1;
2605 else if ((valid_sum_count - sum_in_page) <=
2606 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2607 return 2;
2608 return 3;
2609 }
2610
2611 /*
2612 * Caller should put this summary page
2613 */
f2fs_get_sum_page(struct f2fs_sb_info * sbi,unsigned int segno)2614 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2615 {
2616 if (unlikely(f2fs_cp_error(sbi)))
2617 return ERR_PTR(-EIO);
2618 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2619 }
2620
f2fs_update_meta_page(struct f2fs_sb_info * sbi,void * src,block_t blk_addr)2621 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2622 void *src, block_t blk_addr)
2623 {
2624 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2625
2626 memcpy(page_address(page), src, PAGE_SIZE);
2627 set_page_dirty(page);
2628 f2fs_put_page(page, 1);
2629 }
2630
write_sum_page(struct f2fs_sb_info * sbi,struct f2fs_summary_block * sum_blk,block_t blk_addr)2631 static void write_sum_page(struct f2fs_sb_info *sbi,
2632 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2633 {
2634 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2635 }
2636
write_current_sum_page(struct f2fs_sb_info * sbi,int type,block_t blk_addr)2637 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2638 int type, block_t blk_addr)
2639 {
2640 struct curseg_info *curseg = CURSEG_I(sbi, type);
2641 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2642 struct f2fs_summary_block *src = curseg->sum_blk;
2643 struct f2fs_summary_block *dst;
2644
2645 dst = (struct f2fs_summary_block *)page_address(page);
2646 memset(dst, 0, PAGE_SIZE);
2647
2648 mutex_lock(&curseg->curseg_mutex);
2649
2650 down_read(&curseg->journal_rwsem);
2651 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2652 up_read(&curseg->journal_rwsem);
2653
2654 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2655 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2656
2657 mutex_unlock(&curseg->curseg_mutex);
2658
2659 set_page_dirty(page);
2660 f2fs_put_page(page, 1);
2661 }
2662
is_next_segment_free(struct f2fs_sb_info * sbi,struct curseg_info * curseg)2663 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2664 struct curseg_info *curseg)
2665 {
2666 unsigned int segno = curseg->segno + 1;
2667 struct free_segmap_info *free_i = FREE_I(sbi);
2668
2669 if (segno < MAIN_SEGS(sbi) && segno % SEGS_PER_SEC(sbi))
2670 return !test_bit(segno, free_i->free_segmap);
2671 return 0;
2672 }
2673
2674 /*
2675 * Find a new segment from the free segments bitmap to right order
2676 * This function should be returned with success, otherwise BUG
2677 */
get_new_segment(struct f2fs_sb_info * sbi,unsigned int * newseg,bool new_sec,bool pinning)2678 static int get_new_segment(struct f2fs_sb_info *sbi,
2679 unsigned int *newseg, bool new_sec, bool pinning)
2680 {
2681 struct free_segmap_info *free_i = FREE_I(sbi);
2682 unsigned int segno, secno, zoneno;
2683 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2684 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2685 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2686 bool init = true;
2687 int i;
2688 int ret = 0;
2689
2690 spin_lock(&free_i->segmap_lock);
2691
2692 if (time_to_inject(sbi, FAULT_NO_SEGMENT)) {
2693 ret = -ENOSPC;
2694 goto out_unlock;
2695 }
2696
2697 if (!new_sec && ((*newseg + 1) % SEGS_PER_SEC(sbi))) {
2698 segno = find_next_zero_bit(free_i->free_segmap,
2699 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2700 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2701 goto got_it;
2702 }
2703
2704 #ifdef CONFIG_BLK_DEV_ZONED
2705 /*
2706 * If we format f2fs on zoned storage, let's try to get pinned sections
2707 * from beginning of the storage, which should be a conventional one.
2708 */
2709 if (f2fs_sb_has_blkzoned(sbi)) {
2710 /* Prioritize writing to conventional zones */
2711 if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_PRIOR_CONV || pinning)
2712 segno = 0;
2713 else
2714 segno = max(first_zoned_segno(sbi), *newseg);
2715 hint = GET_SEC_FROM_SEG(sbi, segno);
2716 }
2717 #endif
2718
2719 find_other_zone:
2720 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2721
2722 #ifdef CONFIG_BLK_DEV_ZONED
2723 if (secno >= MAIN_SECS(sbi) && f2fs_sb_has_blkzoned(sbi)) {
2724 /* Write only to sequential zones */
2725 if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_ONLY_SEQ) {
2726 hint = GET_SEC_FROM_SEG(sbi, first_zoned_segno(sbi));
2727 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2728 } else
2729 secno = find_first_zero_bit(free_i->free_secmap,
2730 MAIN_SECS(sbi));
2731 if (secno >= MAIN_SECS(sbi)) {
2732 ret = -ENOSPC;
2733 f2fs_bug_on(sbi, 1);
2734 goto out_unlock;
2735 }
2736 }
2737 #endif
2738
2739 if (secno >= MAIN_SECS(sbi)) {
2740 secno = find_first_zero_bit(free_i->free_secmap,
2741 MAIN_SECS(sbi));
2742 if (secno >= MAIN_SECS(sbi)) {
2743 ret = -ENOSPC;
2744 f2fs_bug_on(sbi, 1);
2745 goto out_unlock;
2746 }
2747 }
2748 segno = GET_SEG_FROM_SEC(sbi, secno);
2749 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2750
2751 /* give up on finding another zone */
2752 if (!init)
2753 goto got_it;
2754 if (sbi->secs_per_zone == 1)
2755 goto got_it;
2756 if (zoneno == old_zoneno)
2757 goto got_it;
2758 for (i = 0; i < NR_CURSEG_TYPE; i++)
2759 if (CURSEG_I(sbi, i)->zone == zoneno)
2760 break;
2761
2762 if (i < NR_CURSEG_TYPE) {
2763 /* zone is in user, try another */
2764 if (zoneno + 1 >= total_zones)
2765 hint = 0;
2766 else
2767 hint = (zoneno + 1) * sbi->secs_per_zone;
2768 init = false;
2769 goto find_other_zone;
2770 }
2771 got_it:
2772 /* set it as dirty segment in free segmap */
2773 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2774
2775 /* no free section in conventional zone */
2776 if (new_sec && pinning &&
2777 !f2fs_valid_pinned_area(sbi, START_BLOCK(sbi, segno))) {
2778 ret = -EAGAIN;
2779 goto out_unlock;
2780 }
2781 __set_inuse(sbi, segno);
2782 *newseg = segno;
2783 out_unlock:
2784 spin_unlock(&free_i->segmap_lock);
2785
2786 if (ret == -ENOSPC)
2787 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_NO_SEGMENT);
2788 return ret;
2789 }
2790
reset_curseg(struct f2fs_sb_info * sbi,int type,int modified)2791 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2792 {
2793 struct curseg_info *curseg = CURSEG_I(sbi, type);
2794 struct summary_footer *sum_footer;
2795 unsigned short seg_type = curseg->seg_type;
2796
2797 /* only happen when get_new_segment() fails */
2798 if (curseg->next_segno == NULL_SEGNO)
2799 return;
2800
2801 curseg->inited = true;
2802 curseg->segno = curseg->next_segno;
2803 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2804 curseg->next_blkoff = 0;
2805 curseg->next_segno = NULL_SEGNO;
2806
2807 sum_footer = &(curseg->sum_blk->footer);
2808 memset(sum_footer, 0, sizeof(struct summary_footer));
2809
2810 sanity_check_seg_type(sbi, seg_type);
2811
2812 if (IS_DATASEG(seg_type))
2813 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2814 if (IS_NODESEG(seg_type))
2815 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2816 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2817 }
2818
__get_next_segno(struct f2fs_sb_info * sbi,int type)2819 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2820 {
2821 struct curseg_info *curseg = CURSEG_I(sbi, type);
2822 unsigned short seg_type = curseg->seg_type;
2823
2824 sanity_check_seg_type(sbi, seg_type);
2825 if (__is_large_section(sbi)) {
2826 if (f2fs_need_rand_seg(sbi)) {
2827 unsigned int hint = GET_SEC_FROM_SEG(sbi, curseg->segno);
2828
2829 if (GET_SEC_FROM_SEG(sbi, curseg->segno + 1) != hint)
2830 return curseg->segno;
2831 return get_random_u32_inclusive(curseg->segno + 1,
2832 GET_SEG_FROM_SEC(sbi, hint + 1) - 1);
2833 }
2834 return curseg->segno;
2835 } else if (f2fs_need_rand_seg(sbi)) {
2836 return get_random_u32_below(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi));
2837 }
2838
2839 /* inmem log may not locate on any segment after mount */
2840 if (!curseg->inited)
2841 return 0;
2842
2843 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2844 return 0;
2845
2846 if (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))
2847 return 0;
2848
2849 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2850 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2851
2852 /* find segments from 0 to reuse freed segments */
2853 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2854 return 0;
2855
2856 return curseg->segno;
2857 }
2858
2859 /*
2860 * Allocate a current working segment.
2861 * This function always allocates a free segment in LFS manner.
2862 */
new_curseg(struct f2fs_sb_info * sbi,int type,bool new_sec)2863 static int new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2864 {
2865 struct curseg_info *curseg = CURSEG_I(sbi, type);
2866 unsigned int segno = curseg->segno;
2867 bool pinning = type == CURSEG_COLD_DATA_PINNED;
2868 int ret;
2869
2870 if (curseg->inited)
2871 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, segno));
2872
2873 segno = __get_next_segno(sbi, type);
2874 ret = get_new_segment(sbi, &segno, new_sec, pinning);
2875 if (ret) {
2876 if (ret == -ENOSPC)
2877 curseg->segno = NULL_SEGNO;
2878 return ret;
2879 }
2880
2881 curseg->next_segno = segno;
2882 reset_curseg(sbi, type, 1);
2883 curseg->alloc_type = LFS;
2884 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2885 curseg->fragment_remained_chunk =
2886 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2887 return 0;
2888 }
2889
__next_free_blkoff(struct f2fs_sb_info * sbi,int segno,block_t start)2890 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2891 int segno, block_t start)
2892 {
2893 struct seg_entry *se = get_seg_entry(sbi, segno);
2894 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2895 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2896 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2897 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2898 int i;
2899
2900 for (i = 0; i < entries; i++)
2901 target_map[i] = ckpt_map[i] | cur_map[i];
2902
2903 return __find_rev_next_zero_bit(target_map, BLKS_PER_SEG(sbi), start);
2904 }
2905
f2fs_find_next_ssr_block(struct f2fs_sb_info * sbi,struct curseg_info * seg)2906 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
2907 struct curseg_info *seg)
2908 {
2909 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
2910 }
2911
f2fs_segment_has_free_slot(struct f2fs_sb_info * sbi,int segno)2912 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2913 {
2914 return __next_free_blkoff(sbi, segno, 0) < BLKS_PER_SEG(sbi);
2915 }
2916
2917 /*
2918 * This function always allocates a used segment(from dirty seglist) by SSR
2919 * manner, so it should recover the existing segment information of valid blocks
2920 */
change_curseg(struct f2fs_sb_info * sbi,int type)2921 static int change_curseg(struct f2fs_sb_info *sbi, int type)
2922 {
2923 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2924 struct curseg_info *curseg = CURSEG_I(sbi, type);
2925 unsigned int new_segno = curseg->next_segno;
2926 struct f2fs_summary_block *sum_node;
2927 struct page *sum_page;
2928
2929 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2930
2931 __set_test_and_inuse(sbi, new_segno);
2932
2933 mutex_lock(&dirty_i->seglist_lock);
2934 __remove_dirty_segment(sbi, new_segno, PRE);
2935 __remove_dirty_segment(sbi, new_segno, DIRTY);
2936 mutex_unlock(&dirty_i->seglist_lock);
2937
2938 reset_curseg(sbi, type, 1);
2939 curseg->alloc_type = SSR;
2940 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2941
2942 sum_page = f2fs_get_sum_page(sbi, new_segno);
2943 if (IS_ERR(sum_page)) {
2944 /* GC won't be able to use stale summary pages by cp_error */
2945 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2946 return PTR_ERR(sum_page);
2947 }
2948 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2949 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2950 f2fs_put_page(sum_page, 1);
2951 return 0;
2952 }
2953
2954 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2955 int alloc_mode, unsigned long long age);
2956
get_atssr_segment(struct f2fs_sb_info * sbi,int type,int target_type,int alloc_mode,unsigned long long age)2957 static int get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2958 int target_type, int alloc_mode,
2959 unsigned long long age)
2960 {
2961 struct curseg_info *curseg = CURSEG_I(sbi, type);
2962 int ret = 0;
2963
2964 curseg->seg_type = target_type;
2965
2966 if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2967 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2968
2969 curseg->seg_type = se->type;
2970 ret = change_curseg(sbi, type);
2971 } else {
2972 /* allocate cold segment by default */
2973 curseg->seg_type = CURSEG_COLD_DATA;
2974 ret = new_curseg(sbi, type, true);
2975 }
2976 stat_inc_seg_type(sbi, curseg);
2977 return ret;
2978 }
2979
__f2fs_init_atgc_curseg(struct f2fs_sb_info * sbi,bool force)2980 static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi, bool force)
2981 {
2982 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2983 int ret = 0;
2984
2985 if (!sbi->am.atgc_enabled && !force)
2986 return 0;
2987
2988 f2fs_down_read(&SM_I(sbi)->curseg_lock);
2989
2990 mutex_lock(&curseg->curseg_mutex);
2991 down_write(&SIT_I(sbi)->sentry_lock);
2992
2993 ret = get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC,
2994 CURSEG_COLD_DATA, SSR, 0);
2995
2996 up_write(&SIT_I(sbi)->sentry_lock);
2997 mutex_unlock(&curseg->curseg_mutex);
2998
2999 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3000 return ret;
3001 }
3002
f2fs_init_inmem_curseg(struct f2fs_sb_info * sbi)3003 int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
3004 {
3005 return __f2fs_init_atgc_curseg(sbi, false);
3006 }
3007
f2fs_reinit_atgc_curseg(struct f2fs_sb_info * sbi)3008 int f2fs_reinit_atgc_curseg(struct f2fs_sb_info *sbi)
3009 {
3010 int ret;
3011
3012 if (!test_opt(sbi, ATGC))
3013 return 0;
3014 if (sbi->am.atgc_enabled)
3015 return 0;
3016 if (le64_to_cpu(F2FS_CKPT(sbi)->elapsed_time) <
3017 sbi->am.age_threshold)
3018 return 0;
3019
3020 ret = __f2fs_init_atgc_curseg(sbi, true);
3021 if (!ret) {
3022 sbi->am.atgc_enabled = true;
3023 f2fs_info(sbi, "reenabled age threshold GC");
3024 }
3025 return ret;
3026 }
3027
__f2fs_save_inmem_curseg(struct f2fs_sb_info * sbi,int type)3028 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
3029 {
3030 struct curseg_info *curseg = CURSEG_I(sbi, type);
3031
3032 mutex_lock(&curseg->curseg_mutex);
3033 if (!curseg->inited)
3034 goto out;
3035
3036 if (get_valid_blocks(sbi, curseg->segno, false)) {
3037 write_sum_page(sbi, curseg->sum_blk,
3038 GET_SUM_BLOCK(sbi, curseg->segno));
3039 } else {
3040 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
3041 __set_test_and_free(sbi, curseg->segno, true);
3042 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
3043 }
3044 out:
3045 mutex_unlock(&curseg->curseg_mutex);
3046 }
3047
f2fs_save_inmem_curseg(struct f2fs_sb_info * sbi)3048 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
3049 {
3050 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
3051
3052 if (sbi->am.atgc_enabled)
3053 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
3054 }
3055
__f2fs_restore_inmem_curseg(struct f2fs_sb_info * sbi,int type)3056 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
3057 {
3058 struct curseg_info *curseg = CURSEG_I(sbi, type);
3059
3060 mutex_lock(&curseg->curseg_mutex);
3061 if (!curseg->inited)
3062 goto out;
3063 if (get_valid_blocks(sbi, curseg->segno, false))
3064 goto out;
3065
3066 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
3067 __set_test_and_inuse(sbi, curseg->segno);
3068 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
3069 out:
3070 mutex_unlock(&curseg->curseg_mutex);
3071 }
3072
f2fs_restore_inmem_curseg(struct f2fs_sb_info * sbi)3073 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
3074 {
3075 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
3076
3077 if (sbi->am.atgc_enabled)
3078 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
3079 }
3080
get_ssr_segment(struct f2fs_sb_info * sbi,int type,int alloc_mode,unsigned long long age)3081 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
3082 int alloc_mode, unsigned long long age)
3083 {
3084 struct curseg_info *curseg = CURSEG_I(sbi, type);
3085 unsigned segno = NULL_SEGNO;
3086 unsigned short seg_type = curseg->seg_type;
3087 int i, cnt;
3088 bool reversed = false;
3089
3090 sanity_check_seg_type(sbi, seg_type);
3091
3092 /* f2fs_need_SSR() already forces to do this */
3093 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type,
3094 alloc_mode, age, false)) {
3095 curseg->next_segno = segno;
3096 return 1;
3097 }
3098
3099 /* For node segments, let's do SSR more intensively */
3100 if (IS_NODESEG(seg_type)) {
3101 if (seg_type >= CURSEG_WARM_NODE) {
3102 reversed = true;
3103 i = CURSEG_COLD_NODE;
3104 } else {
3105 i = CURSEG_HOT_NODE;
3106 }
3107 cnt = NR_CURSEG_NODE_TYPE;
3108 } else {
3109 if (seg_type >= CURSEG_WARM_DATA) {
3110 reversed = true;
3111 i = CURSEG_COLD_DATA;
3112 } else {
3113 i = CURSEG_HOT_DATA;
3114 }
3115 cnt = NR_CURSEG_DATA_TYPE;
3116 }
3117
3118 for (; cnt-- > 0; reversed ? i-- : i++) {
3119 if (i == seg_type)
3120 continue;
3121 if (!f2fs_get_victim(sbi, &segno, BG_GC, i,
3122 alloc_mode, age, false)) {
3123 curseg->next_segno = segno;
3124 return 1;
3125 }
3126 }
3127
3128 /* find valid_blocks=0 in dirty list */
3129 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
3130 segno = get_free_segment(sbi);
3131 if (segno != NULL_SEGNO) {
3132 curseg->next_segno = segno;
3133 return 1;
3134 }
3135 }
3136 return 0;
3137 }
3138
need_new_seg(struct f2fs_sb_info * sbi,int type)3139 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
3140 {
3141 struct curseg_info *curseg = CURSEG_I(sbi, type);
3142
3143 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
3144 curseg->seg_type == CURSEG_WARM_NODE)
3145 return true;
3146 if (curseg->alloc_type == LFS && is_next_segment_free(sbi, curseg) &&
3147 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
3148 return true;
3149 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
3150 return true;
3151 return false;
3152 }
3153
f2fs_allocate_segment_for_resize(struct f2fs_sb_info * sbi,int type,unsigned int start,unsigned int end)3154 int f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
3155 unsigned int start, unsigned int end)
3156 {
3157 struct curseg_info *curseg = CURSEG_I(sbi, type);
3158 unsigned int segno;
3159 int ret = 0;
3160
3161 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3162 mutex_lock(&curseg->curseg_mutex);
3163 down_write(&SIT_I(sbi)->sentry_lock);
3164
3165 segno = CURSEG_I(sbi, type)->segno;
3166 if (segno < start || segno > end)
3167 goto unlock;
3168
3169 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
3170 ret = change_curseg(sbi, type);
3171 else
3172 ret = new_curseg(sbi, type, true);
3173
3174 stat_inc_seg_type(sbi, curseg);
3175
3176 locate_dirty_segment(sbi, segno);
3177 unlock:
3178 up_write(&SIT_I(sbi)->sentry_lock);
3179
3180 if (segno != curseg->segno)
3181 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3182 type, segno, curseg->segno);
3183
3184 mutex_unlock(&curseg->curseg_mutex);
3185 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3186 return ret;
3187 }
3188
__allocate_new_segment(struct f2fs_sb_info * sbi,int type,bool new_sec,bool force)3189 static int __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3190 bool new_sec, bool force)
3191 {
3192 struct curseg_info *curseg = CURSEG_I(sbi, type);
3193 unsigned int old_segno;
3194 int err = 0;
3195
3196 if (type == CURSEG_COLD_DATA_PINNED && !curseg->inited)
3197 goto allocate;
3198
3199 if (!force && curseg->inited &&
3200 !curseg->next_blkoff &&
3201 !get_valid_blocks(sbi, curseg->segno, new_sec) &&
3202 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3203 return 0;
3204
3205 allocate:
3206 old_segno = curseg->segno;
3207 err = new_curseg(sbi, type, true);
3208 if (err)
3209 return err;
3210 stat_inc_seg_type(sbi, curseg);
3211 locate_dirty_segment(sbi, old_segno);
3212 return 0;
3213 }
3214
f2fs_allocate_new_section(struct f2fs_sb_info * sbi,int type,bool force)3215 int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3216 {
3217 int ret;
3218
3219 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3220 down_write(&SIT_I(sbi)->sentry_lock);
3221 ret = __allocate_new_segment(sbi, type, true, force);
3222 up_write(&SIT_I(sbi)->sentry_lock);
3223 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3224
3225 return ret;
3226 }
3227
f2fs_allocate_pinning_section(struct f2fs_sb_info * sbi)3228 int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi)
3229 {
3230 int err;
3231 bool gc_required = true;
3232
3233 retry:
3234 f2fs_lock_op(sbi);
3235 err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3236 f2fs_unlock_op(sbi);
3237
3238 if (f2fs_sb_has_blkzoned(sbi) && err == -EAGAIN && gc_required) {
3239 f2fs_down_write(&sbi->gc_lock);
3240 err = f2fs_gc_range(sbi, 0, GET_SEGNO(sbi, FDEV(0).end_blk), true, 1);
3241 f2fs_up_write(&sbi->gc_lock);
3242
3243 gc_required = false;
3244 if (!err)
3245 goto retry;
3246 }
3247
3248 return err;
3249 }
3250
f2fs_allocate_new_segments(struct f2fs_sb_info * sbi)3251 int f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3252 {
3253 int i;
3254 int err = 0;
3255
3256 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3257 down_write(&SIT_I(sbi)->sentry_lock);
3258 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3259 err += __allocate_new_segment(sbi, i, false, false);
3260 up_write(&SIT_I(sbi)->sentry_lock);
3261 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3262
3263 return err;
3264 }
3265
f2fs_exist_trim_candidates(struct f2fs_sb_info * sbi,struct cp_control * cpc)3266 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3267 struct cp_control *cpc)
3268 {
3269 __u64 trim_start = cpc->trim_start;
3270 bool has_candidate = false;
3271
3272 down_write(&SIT_I(sbi)->sentry_lock);
3273 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3274 if (add_discard_addrs(sbi, cpc, true)) {
3275 has_candidate = true;
3276 break;
3277 }
3278 }
3279 up_write(&SIT_I(sbi)->sentry_lock);
3280
3281 cpc->trim_start = trim_start;
3282 return has_candidate;
3283 }
3284
__issue_discard_cmd_range(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy,unsigned int start,unsigned int end)3285 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3286 struct discard_policy *dpolicy,
3287 unsigned int start, unsigned int end)
3288 {
3289 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3290 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3291 struct rb_node **insert_p = NULL, *insert_parent = NULL;
3292 struct discard_cmd *dc;
3293 struct blk_plug plug;
3294 int issued;
3295 unsigned int trimmed = 0;
3296
3297 next:
3298 issued = 0;
3299
3300 mutex_lock(&dcc->cmd_lock);
3301 if (unlikely(dcc->rbtree_check))
3302 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3303
3304 dc = __lookup_discard_cmd_ret(&dcc->root, start,
3305 &prev_dc, &next_dc, &insert_p, &insert_parent);
3306 if (!dc)
3307 dc = next_dc;
3308
3309 blk_start_plug(&plug);
3310
3311 while (dc && dc->di.lstart <= end) {
3312 struct rb_node *node;
3313 int err = 0;
3314
3315 if (dc->di.len < dpolicy->granularity)
3316 goto skip;
3317
3318 if (dc->state != D_PREP) {
3319 list_move_tail(&dc->list, &dcc->fstrim_list);
3320 goto skip;
3321 }
3322
3323 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3324
3325 if (issued >= dpolicy->max_requests) {
3326 start = dc->di.lstart + dc->di.len;
3327
3328 if (err)
3329 __remove_discard_cmd(sbi, dc);
3330
3331 blk_finish_plug(&plug);
3332 mutex_unlock(&dcc->cmd_lock);
3333 trimmed += __wait_all_discard_cmd(sbi, NULL);
3334 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3335 goto next;
3336 }
3337 skip:
3338 node = rb_next(&dc->rb_node);
3339 if (err)
3340 __remove_discard_cmd(sbi, dc);
3341 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3342
3343 if (fatal_signal_pending(current))
3344 break;
3345 }
3346
3347 blk_finish_plug(&plug);
3348 mutex_unlock(&dcc->cmd_lock);
3349
3350 return trimmed;
3351 }
3352
f2fs_trim_fs(struct f2fs_sb_info * sbi,struct fstrim_range * range)3353 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3354 {
3355 __u64 start = F2FS_BYTES_TO_BLK(range->start);
3356 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3357 unsigned int start_segno, end_segno;
3358 block_t start_block, end_block;
3359 struct cp_control cpc;
3360 struct discard_policy dpolicy;
3361 unsigned long long trimmed = 0;
3362 int err = 0;
3363 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3364
3365 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3366 return -EINVAL;
3367
3368 if (end < MAIN_BLKADDR(sbi))
3369 goto out;
3370
3371 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3372 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3373 return -EFSCORRUPTED;
3374 }
3375
3376 /* start/end segment number in main_area */
3377 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3378 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3379 GET_SEGNO(sbi, end);
3380 if (need_align) {
3381 start_segno = rounddown(start_segno, SEGS_PER_SEC(sbi));
3382 end_segno = roundup(end_segno + 1, SEGS_PER_SEC(sbi)) - 1;
3383 }
3384
3385 cpc.reason = CP_DISCARD;
3386 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3387 cpc.trim_start = start_segno;
3388 cpc.trim_end = end_segno;
3389
3390 if (sbi->discard_blks == 0)
3391 goto out;
3392
3393 f2fs_down_write(&sbi->gc_lock);
3394 stat_inc_cp_call_count(sbi, TOTAL_CALL);
3395 err = f2fs_write_checkpoint(sbi, &cpc);
3396 f2fs_up_write(&sbi->gc_lock);
3397 if (err)
3398 goto out;
3399
3400 /*
3401 * We filed discard candidates, but actually we don't need to wait for
3402 * all of them, since they'll be issued in idle time along with runtime
3403 * discard option. User configuration looks like using runtime discard
3404 * or periodic fstrim instead of it.
3405 */
3406 if (f2fs_realtime_discard_enable(sbi))
3407 goto out;
3408
3409 start_block = START_BLOCK(sbi, start_segno);
3410 end_block = START_BLOCK(sbi, end_segno + 1);
3411
3412 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3413 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3414 start_block, end_block);
3415
3416 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3417 start_block, end_block);
3418 out:
3419 if (!err)
3420 range->len = F2FS_BLK_TO_BYTES(trimmed);
3421 return err;
3422 }
3423
f2fs_rw_hint_to_seg_type(struct f2fs_sb_info * sbi,enum rw_hint hint)3424 int f2fs_rw_hint_to_seg_type(struct f2fs_sb_info *sbi, enum rw_hint hint)
3425 {
3426 if (F2FS_OPTION(sbi).active_logs == 2)
3427 return CURSEG_HOT_DATA;
3428 else if (F2FS_OPTION(sbi).active_logs == 4)
3429 return CURSEG_COLD_DATA;
3430
3431 /* active_log == 6 */
3432 switch (hint) {
3433 case WRITE_LIFE_SHORT:
3434 return CURSEG_HOT_DATA;
3435 case WRITE_LIFE_EXTREME:
3436 return CURSEG_COLD_DATA;
3437 default:
3438 return CURSEG_WARM_DATA;
3439 }
3440 }
3441
3442 /*
3443 * This returns write hints for each segment type. This hints will be
3444 * passed down to block layer as below by default.
3445 *
3446 * User F2FS Block
3447 * ---- ---- -----
3448 * META WRITE_LIFE_NONE|REQ_META
3449 * HOT_NODE WRITE_LIFE_NONE
3450 * WARM_NODE WRITE_LIFE_MEDIUM
3451 * COLD_NODE WRITE_LIFE_LONG
3452 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
3453 * extension list " "
3454 *
3455 * -- buffered io
3456 * COLD_DATA WRITE_LIFE_EXTREME
3457 * HOT_DATA WRITE_LIFE_SHORT
3458 * WARM_DATA WRITE_LIFE_NOT_SET
3459 *
3460 * -- direct io
3461 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3462 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3463 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
3464 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
3465 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
3466 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
3467 */
f2fs_io_type_to_rw_hint(struct f2fs_sb_info * sbi,enum page_type type,enum temp_type temp)3468 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3469 enum page_type type, enum temp_type temp)
3470 {
3471 switch (type) {
3472 case DATA:
3473 switch (temp) {
3474 case WARM:
3475 return WRITE_LIFE_NOT_SET;
3476 case HOT:
3477 return WRITE_LIFE_SHORT;
3478 case COLD:
3479 return WRITE_LIFE_EXTREME;
3480 default:
3481 return WRITE_LIFE_NONE;
3482 }
3483 case NODE:
3484 switch (temp) {
3485 case WARM:
3486 return WRITE_LIFE_MEDIUM;
3487 case HOT:
3488 return WRITE_LIFE_NONE;
3489 case COLD:
3490 return WRITE_LIFE_LONG;
3491 default:
3492 return WRITE_LIFE_NONE;
3493 }
3494 case META:
3495 return WRITE_LIFE_NONE;
3496 default:
3497 return WRITE_LIFE_NONE;
3498 }
3499 }
3500
__get_segment_type_2(struct f2fs_io_info * fio)3501 static int __get_segment_type_2(struct f2fs_io_info *fio)
3502 {
3503 if (fio->type == DATA)
3504 return CURSEG_HOT_DATA;
3505 else
3506 return CURSEG_HOT_NODE;
3507 }
3508
__get_segment_type_4(struct f2fs_io_info * fio)3509 static int __get_segment_type_4(struct f2fs_io_info *fio)
3510 {
3511 if (fio->type == DATA) {
3512 struct inode *inode = fio->page->mapping->host;
3513
3514 if (S_ISDIR(inode->i_mode))
3515 return CURSEG_HOT_DATA;
3516 else
3517 return CURSEG_COLD_DATA;
3518 } else {
3519 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3520 return CURSEG_WARM_NODE;
3521 else
3522 return CURSEG_COLD_NODE;
3523 }
3524 }
3525
__get_age_segment_type(struct inode * inode,pgoff_t pgofs)3526 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3527 {
3528 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3529 struct extent_info ei = {};
3530
3531 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3532 if (!ei.age)
3533 return NO_CHECK_TYPE;
3534 if (ei.age <= sbi->hot_data_age_threshold)
3535 return CURSEG_HOT_DATA;
3536 if (ei.age <= sbi->warm_data_age_threshold)
3537 return CURSEG_WARM_DATA;
3538 return CURSEG_COLD_DATA;
3539 }
3540 return NO_CHECK_TYPE;
3541 }
3542
__get_segment_type_6(struct f2fs_io_info * fio)3543 static int __get_segment_type_6(struct f2fs_io_info *fio)
3544 {
3545 if (fio->type == DATA) {
3546 struct inode *inode = fio->page->mapping->host;
3547 int type;
3548
3549 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3550 return CURSEG_COLD_DATA_PINNED;
3551
3552 if (page_private_gcing(fio->page)) {
3553 if (fio->sbi->am.atgc_enabled &&
3554 (fio->io_type == FS_DATA_IO) &&
3555 (fio->sbi->gc_mode != GC_URGENT_HIGH) &&
3556 __is_valid_data_blkaddr(fio->old_blkaddr) &&
3557 !is_inode_flag_set(inode, FI_OPU_WRITE))
3558 return CURSEG_ALL_DATA_ATGC;
3559 else
3560 return CURSEG_COLD_DATA;
3561 }
3562 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3563 return CURSEG_COLD_DATA;
3564
3565 type = __get_age_segment_type(inode,
3566 page_folio(fio->page)->index);
3567 if (type != NO_CHECK_TYPE)
3568 return type;
3569
3570 if (file_is_hot(inode) ||
3571 is_inode_flag_set(inode, FI_HOT_DATA) ||
3572 f2fs_is_cow_file(inode))
3573 return CURSEG_HOT_DATA;
3574 return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
3575 inode->i_write_hint);
3576 } else {
3577 if (IS_DNODE(fio->page))
3578 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3579 CURSEG_HOT_NODE;
3580 return CURSEG_COLD_NODE;
3581 }
3582 }
3583
f2fs_get_segment_temp(int seg_type)3584 int f2fs_get_segment_temp(int seg_type)
3585 {
3586 if (IS_HOT(seg_type))
3587 return HOT;
3588 else if (IS_WARM(seg_type))
3589 return WARM;
3590 return COLD;
3591 }
3592
__get_segment_type(struct f2fs_io_info * fio)3593 static int __get_segment_type(struct f2fs_io_info *fio)
3594 {
3595 int type = 0;
3596
3597 switch (F2FS_OPTION(fio->sbi).active_logs) {
3598 case 2:
3599 type = __get_segment_type_2(fio);
3600 break;
3601 case 4:
3602 type = __get_segment_type_4(fio);
3603 break;
3604 case 6:
3605 type = __get_segment_type_6(fio);
3606 break;
3607 default:
3608 f2fs_bug_on(fio->sbi, true);
3609 }
3610
3611 fio->temp = f2fs_get_segment_temp(type);
3612
3613 return type;
3614 }
3615
f2fs_randomize_chunk(struct f2fs_sb_info * sbi,struct curseg_info * seg)3616 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3617 struct curseg_info *seg)
3618 {
3619 /* To allocate block chunks in different sizes, use random number */
3620 if (--seg->fragment_remained_chunk > 0)
3621 return;
3622
3623 seg->fragment_remained_chunk =
3624 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3625 seg->next_blkoff +=
3626 get_random_u32_inclusive(1, sbi->max_fragment_hole);
3627 }
3628
reset_curseg_fields(struct curseg_info * curseg)3629 static void reset_curseg_fields(struct curseg_info *curseg)
3630 {
3631 curseg->inited = false;
3632 curseg->segno = NULL_SEGNO;
3633 curseg->next_segno = 0;
3634 }
3635
f2fs_allocate_data_block(struct f2fs_sb_info * sbi,struct page * page,block_t old_blkaddr,block_t * new_blkaddr,struct f2fs_summary * sum,int type,struct f2fs_io_info * fio)3636 int f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3637 block_t old_blkaddr, block_t *new_blkaddr,
3638 struct f2fs_summary *sum, int type,
3639 struct f2fs_io_info *fio)
3640 {
3641 struct sit_info *sit_i = SIT_I(sbi);
3642 struct curseg_info *curseg = CURSEG_I(sbi, type);
3643 unsigned long long old_mtime;
3644 bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3645 struct seg_entry *se = NULL;
3646 bool segment_full = false;
3647 int ret = 0;
3648
3649 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3650
3651 mutex_lock(&curseg->curseg_mutex);
3652 down_write(&sit_i->sentry_lock);
3653
3654 if (curseg->segno == NULL_SEGNO) {
3655 ret = -ENOSPC;
3656 goto out_err;
3657 }
3658
3659 if (from_gc) {
3660 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3661 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3662 sanity_check_seg_type(sbi, se->type);
3663 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3664 }
3665 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3666
3667 f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi));
3668
3669 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3670
3671 curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3672 if (curseg->alloc_type == SSR) {
3673 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3674 } else {
3675 curseg->next_blkoff++;
3676 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3677 f2fs_randomize_chunk(sbi, curseg);
3678 }
3679 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3680 segment_full = true;
3681 stat_inc_block_count(sbi, curseg);
3682
3683 if (from_gc) {
3684 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3685 } else {
3686 update_segment_mtime(sbi, old_blkaddr, 0);
3687 old_mtime = 0;
3688 }
3689 update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3690
3691 /*
3692 * SIT information should be updated before segment allocation,
3693 * since SSR needs latest valid block information.
3694 */
3695 update_sit_entry(sbi, *new_blkaddr, 1);
3696 update_sit_entry(sbi, old_blkaddr, -1);
3697
3698 /*
3699 * If the current segment is full, flush it out and replace it with a
3700 * new segment.
3701 */
3702 if (segment_full) {
3703 if (type == CURSEG_COLD_DATA_PINNED &&
3704 !((curseg->segno + 1) % sbi->segs_per_sec)) {
3705 write_sum_page(sbi, curseg->sum_blk,
3706 GET_SUM_BLOCK(sbi, curseg->segno));
3707 reset_curseg_fields(curseg);
3708 goto skip_new_segment;
3709 }
3710
3711 if (from_gc) {
3712 ret = get_atssr_segment(sbi, type, se->type,
3713 AT_SSR, se->mtime);
3714 } else {
3715 if (need_new_seg(sbi, type))
3716 ret = new_curseg(sbi, type, false);
3717 else
3718 ret = change_curseg(sbi, type);
3719 stat_inc_seg_type(sbi, curseg);
3720 }
3721
3722 if (ret)
3723 goto out_err;
3724 }
3725
3726 skip_new_segment:
3727 /*
3728 * segment dirty status should be updated after segment allocation,
3729 * so we just need to update status only one time after previous
3730 * segment being closed.
3731 */
3732 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3733 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3734
3735 if (IS_DATASEG(curseg->seg_type))
3736 atomic64_inc(&sbi->allocated_data_blocks);
3737
3738 up_write(&sit_i->sentry_lock);
3739
3740 if (page && IS_NODESEG(curseg->seg_type)) {
3741 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3742
3743 f2fs_inode_chksum_set(sbi, page);
3744 }
3745
3746 if (fio) {
3747 struct f2fs_bio_info *io;
3748
3749 INIT_LIST_HEAD(&fio->list);
3750 fio->in_list = 1;
3751 io = sbi->write_io[fio->type] + fio->temp;
3752 spin_lock(&io->io_lock);
3753 list_add_tail(&fio->list, &io->io_list);
3754 spin_unlock(&io->io_lock);
3755 }
3756
3757 mutex_unlock(&curseg->curseg_mutex);
3758 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3759 return 0;
3760
3761 out_err:
3762 *new_blkaddr = NULL_ADDR;
3763 up_write(&sit_i->sentry_lock);
3764 mutex_unlock(&curseg->curseg_mutex);
3765 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3766 return ret;
3767 }
3768
f2fs_update_device_state(struct f2fs_sb_info * sbi,nid_t ino,block_t blkaddr,unsigned int blkcnt)3769 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3770 block_t blkaddr, unsigned int blkcnt)
3771 {
3772 if (!f2fs_is_multi_device(sbi))
3773 return;
3774
3775 while (1) {
3776 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3777 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3778
3779 /* update device state for fsync */
3780 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3781
3782 /* update device state for checkpoint */
3783 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3784 spin_lock(&sbi->dev_lock);
3785 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3786 spin_unlock(&sbi->dev_lock);
3787 }
3788
3789 if (blkcnt <= blks)
3790 break;
3791 blkcnt -= blks;
3792 blkaddr += blks;
3793 }
3794 }
3795
do_write_page(struct f2fs_summary * sum,struct f2fs_io_info * fio)3796 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3797 {
3798 int type = __get_segment_type(fio);
3799 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3800
3801 if (keep_order)
3802 f2fs_down_read(&fio->sbi->io_order_lock);
3803
3804 if (f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3805 &fio->new_blkaddr, sum, type, fio)) {
3806 if (fscrypt_inode_uses_fs_layer_crypto(fio->page->mapping->host))
3807 fscrypt_finalize_bounce_page(&fio->encrypted_page);
3808 end_page_writeback(fio->page);
3809 if (f2fs_in_warm_node_list(fio->sbi, fio->page))
3810 f2fs_del_fsync_node_entry(fio->sbi, fio->page);
3811 goto out;
3812 }
3813 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3814 f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr);
3815
3816 /* writeout dirty page into bdev */
3817 f2fs_submit_page_write(fio);
3818
3819 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3820 out:
3821 if (keep_order)
3822 f2fs_up_read(&fio->sbi->io_order_lock);
3823 }
3824
f2fs_do_write_meta_page(struct f2fs_sb_info * sbi,struct folio * folio,enum iostat_type io_type)3825 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct folio *folio,
3826 enum iostat_type io_type)
3827 {
3828 struct f2fs_io_info fio = {
3829 .sbi = sbi,
3830 .type = META,
3831 .temp = HOT,
3832 .op = REQ_OP_WRITE,
3833 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3834 .old_blkaddr = folio->index,
3835 .new_blkaddr = folio->index,
3836 .page = folio_page(folio, 0),
3837 .encrypted_page = NULL,
3838 .in_list = 0,
3839 };
3840
3841 if (unlikely(folio->index >= MAIN_BLKADDR(sbi)))
3842 fio.op_flags &= ~REQ_META;
3843
3844 folio_start_writeback(folio);
3845 f2fs_submit_page_write(&fio);
3846
3847 stat_inc_meta_count(sbi, folio->index);
3848 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3849 }
3850
f2fs_do_write_node_page(unsigned int nid,struct f2fs_io_info * fio)3851 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3852 {
3853 struct f2fs_summary sum;
3854
3855 set_summary(&sum, nid, 0, 0);
3856 do_write_page(&sum, fio);
3857
3858 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3859 }
3860
f2fs_outplace_write_data(struct dnode_of_data * dn,struct f2fs_io_info * fio)3861 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3862 struct f2fs_io_info *fio)
3863 {
3864 struct f2fs_sb_info *sbi = fio->sbi;
3865 struct f2fs_summary sum;
3866
3867 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3868 if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
3869 f2fs_update_age_extent_cache(dn);
3870 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3871 do_write_page(&sum, fio);
3872 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3873
3874 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3875 }
3876
f2fs_inplace_write_data(struct f2fs_io_info * fio)3877 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3878 {
3879 int err;
3880 struct f2fs_sb_info *sbi = fio->sbi;
3881 unsigned int segno;
3882
3883 fio->new_blkaddr = fio->old_blkaddr;
3884 /* i/o temperature is needed for passing down write hints */
3885 __get_segment_type(fio);
3886
3887 segno = GET_SEGNO(sbi, fio->new_blkaddr);
3888
3889 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3890 set_sbi_flag(sbi, SBI_NEED_FSCK);
3891 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3892 __func__, segno);
3893 err = -EFSCORRUPTED;
3894 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3895 goto drop_bio;
3896 }
3897
3898 if (f2fs_cp_error(sbi)) {
3899 err = -EIO;
3900 goto drop_bio;
3901 }
3902
3903 if (fio->meta_gc)
3904 f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1);
3905
3906 stat_inc_inplace_blocks(fio->sbi);
3907
3908 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi))
3909 err = f2fs_merge_page_bio(fio);
3910 else
3911 err = f2fs_submit_page_bio(fio);
3912 if (!err) {
3913 f2fs_update_device_state(fio->sbi, fio->ino,
3914 fio->new_blkaddr, 1);
3915 f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3916 fio->io_type, F2FS_BLKSIZE);
3917 }
3918
3919 return err;
3920 drop_bio:
3921 if (fio->bio && *(fio->bio)) {
3922 struct bio *bio = *(fio->bio);
3923
3924 bio->bi_status = BLK_STS_IOERR;
3925 bio_endio(bio);
3926 *(fio->bio) = NULL;
3927 }
3928 return err;
3929 }
3930
__f2fs_get_curseg(struct f2fs_sb_info * sbi,unsigned int segno)3931 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3932 unsigned int segno)
3933 {
3934 int i;
3935
3936 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3937 if (CURSEG_I(sbi, i)->segno == segno)
3938 break;
3939 }
3940 return i;
3941 }
3942
f2fs_do_replace_block(struct f2fs_sb_info * sbi,struct f2fs_summary * sum,block_t old_blkaddr,block_t new_blkaddr,bool recover_curseg,bool recover_newaddr,bool from_gc)3943 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3944 block_t old_blkaddr, block_t new_blkaddr,
3945 bool recover_curseg, bool recover_newaddr,
3946 bool from_gc)
3947 {
3948 struct sit_info *sit_i = SIT_I(sbi);
3949 struct curseg_info *curseg;
3950 unsigned int segno, old_cursegno;
3951 struct seg_entry *se;
3952 int type;
3953 unsigned short old_blkoff;
3954 unsigned char old_alloc_type;
3955
3956 segno = GET_SEGNO(sbi, new_blkaddr);
3957 se = get_seg_entry(sbi, segno);
3958 type = se->type;
3959
3960 f2fs_down_write(&SM_I(sbi)->curseg_lock);
3961
3962 if (!recover_curseg) {
3963 /* for recovery flow */
3964 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3965 if (old_blkaddr == NULL_ADDR)
3966 type = CURSEG_COLD_DATA;
3967 else
3968 type = CURSEG_WARM_DATA;
3969 }
3970 } else {
3971 if (IS_CURSEG(sbi, segno)) {
3972 /* se->type is volatile as SSR allocation */
3973 type = __f2fs_get_curseg(sbi, segno);
3974 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3975 } else {
3976 type = CURSEG_WARM_DATA;
3977 }
3978 }
3979
3980 f2fs_bug_on(sbi, !IS_DATASEG(type));
3981 curseg = CURSEG_I(sbi, type);
3982
3983 mutex_lock(&curseg->curseg_mutex);
3984 down_write(&sit_i->sentry_lock);
3985
3986 old_cursegno = curseg->segno;
3987 old_blkoff = curseg->next_blkoff;
3988 old_alloc_type = curseg->alloc_type;
3989
3990 /* change the current segment */
3991 if (segno != curseg->segno) {
3992 curseg->next_segno = segno;
3993 if (change_curseg(sbi, type))
3994 goto out_unlock;
3995 }
3996
3997 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3998 curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3999
4000 if (!recover_curseg || recover_newaddr) {
4001 if (!from_gc)
4002 update_segment_mtime(sbi, new_blkaddr, 0);
4003 update_sit_entry(sbi, new_blkaddr, 1);
4004 }
4005 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
4006 f2fs_invalidate_internal_cache(sbi, old_blkaddr);
4007 if (!from_gc)
4008 update_segment_mtime(sbi, old_blkaddr, 0);
4009 update_sit_entry(sbi, old_blkaddr, -1);
4010 }
4011
4012 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
4013 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
4014
4015 locate_dirty_segment(sbi, old_cursegno);
4016
4017 if (recover_curseg) {
4018 if (old_cursegno != curseg->segno) {
4019 curseg->next_segno = old_cursegno;
4020 if (change_curseg(sbi, type))
4021 goto out_unlock;
4022 }
4023 curseg->next_blkoff = old_blkoff;
4024 curseg->alloc_type = old_alloc_type;
4025 }
4026
4027 out_unlock:
4028 up_write(&sit_i->sentry_lock);
4029 mutex_unlock(&curseg->curseg_mutex);
4030 f2fs_up_write(&SM_I(sbi)->curseg_lock);
4031 }
4032
f2fs_replace_block(struct f2fs_sb_info * sbi,struct dnode_of_data * dn,block_t old_addr,block_t new_addr,unsigned char version,bool recover_curseg,bool recover_newaddr)4033 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
4034 block_t old_addr, block_t new_addr,
4035 unsigned char version, bool recover_curseg,
4036 bool recover_newaddr)
4037 {
4038 struct f2fs_summary sum;
4039
4040 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
4041
4042 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
4043 recover_curseg, recover_newaddr, false);
4044
4045 f2fs_update_data_blkaddr(dn, new_addr);
4046 }
4047
f2fs_wait_on_page_writeback(struct page * page,enum page_type type,bool ordered,bool locked)4048 void f2fs_wait_on_page_writeback(struct page *page,
4049 enum page_type type, bool ordered, bool locked)
4050 {
4051 if (folio_test_writeback(page_folio(page))) {
4052 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
4053
4054 /* submit cached LFS IO */
4055 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
4056 /* submit cached IPU IO */
4057 f2fs_submit_merged_ipu_write(sbi, NULL, page);
4058 if (ordered) {
4059 wait_on_page_writeback(page);
4060 f2fs_bug_on(sbi, locked &&
4061 folio_test_writeback(page_folio(page)));
4062 } else {
4063 wait_for_stable_page(page);
4064 }
4065 }
4066 }
4067
f2fs_wait_on_block_writeback(struct inode * inode,block_t blkaddr)4068 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
4069 {
4070 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4071 struct page *cpage;
4072
4073 if (!f2fs_meta_inode_gc_required(inode))
4074 return;
4075
4076 if (!__is_valid_data_blkaddr(blkaddr))
4077 return;
4078
4079 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
4080 if (cpage) {
4081 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
4082 f2fs_put_page(cpage, 1);
4083 }
4084 }
4085
f2fs_wait_on_block_writeback_range(struct inode * inode,block_t blkaddr,block_t len)4086 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
4087 block_t len)
4088 {
4089 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4090 block_t i;
4091
4092 if (!f2fs_meta_inode_gc_required(inode))
4093 return;
4094
4095 for (i = 0; i < len; i++)
4096 f2fs_wait_on_block_writeback(inode, blkaddr + i);
4097
4098 f2fs_truncate_meta_inode_pages(sbi, blkaddr, len);
4099 }
4100
read_compacted_summaries(struct f2fs_sb_info * sbi)4101 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
4102 {
4103 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4104 struct curseg_info *seg_i;
4105 unsigned char *kaddr;
4106 struct page *page;
4107 block_t start;
4108 int i, j, offset;
4109
4110 start = start_sum_block(sbi);
4111
4112 page = f2fs_get_meta_page(sbi, start++);
4113 if (IS_ERR(page))
4114 return PTR_ERR(page);
4115 kaddr = (unsigned char *)page_address(page);
4116
4117 /* Step 1: restore nat cache */
4118 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4119 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
4120
4121 /* Step 2: restore sit cache */
4122 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4123 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
4124 offset = 2 * SUM_JOURNAL_SIZE;
4125
4126 /* Step 3: restore summary entries */
4127 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4128 unsigned short blk_off;
4129 unsigned int segno;
4130
4131 seg_i = CURSEG_I(sbi, i);
4132 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
4133 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
4134 seg_i->next_segno = segno;
4135 reset_curseg(sbi, i, 0);
4136 seg_i->alloc_type = ckpt->alloc_type[i];
4137 seg_i->next_blkoff = blk_off;
4138
4139 if (seg_i->alloc_type == SSR)
4140 blk_off = BLKS_PER_SEG(sbi);
4141
4142 for (j = 0; j < blk_off; j++) {
4143 struct f2fs_summary *s;
4144
4145 s = (struct f2fs_summary *)(kaddr + offset);
4146 seg_i->sum_blk->entries[j] = *s;
4147 offset += SUMMARY_SIZE;
4148 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
4149 SUM_FOOTER_SIZE)
4150 continue;
4151
4152 f2fs_put_page(page, 1);
4153 page = NULL;
4154
4155 page = f2fs_get_meta_page(sbi, start++);
4156 if (IS_ERR(page))
4157 return PTR_ERR(page);
4158 kaddr = (unsigned char *)page_address(page);
4159 offset = 0;
4160 }
4161 }
4162 f2fs_put_page(page, 1);
4163 return 0;
4164 }
4165
read_normal_summaries(struct f2fs_sb_info * sbi,int type)4166 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
4167 {
4168 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4169 struct f2fs_summary_block *sum;
4170 struct curseg_info *curseg;
4171 struct page *new;
4172 unsigned short blk_off;
4173 unsigned int segno = 0;
4174 block_t blk_addr = 0;
4175 int err = 0;
4176
4177 /* get segment number and block addr */
4178 if (IS_DATASEG(type)) {
4179 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
4180 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
4181 CURSEG_HOT_DATA]);
4182 if (__exist_node_summaries(sbi))
4183 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
4184 else
4185 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
4186 } else {
4187 segno = le32_to_cpu(ckpt->cur_node_segno[type -
4188 CURSEG_HOT_NODE]);
4189 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
4190 CURSEG_HOT_NODE]);
4191 if (__exist_node_summaries(sbi))
4192 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
4193 type - CURSEG_HOT_NODE);
4194 else
4195 blk_addr = GET_SUM_BLOCK(sbi, segno);
4196 }
4197
4198 new = f2fs_get_meta_page(sbi, blk_addr);
4199 if (IS_ERR(new))
4200 return PTR_ERR(new);
4201 sum = (struct f2fs_summary_block *)page_address(new);
4202
4203 if (IS_NODESEG(type)) {
4204 if (__exist_node_summaries(sbi)) {
4205 struct f2fs_summary *ns = &sum->entries[0];
4206 int i;
4207
4208 for (i = 0; i < BLKS_PER_SEG(sbi); i++, ns++) {
4209 ns->version = 0;
4210 ns->ofs_in_node = 0;
4211 }
4212 } else {
4213 err = f2fs_restore_node_summary(sbi, segno, sum);
4214 if (err)
4215 goto out;
4216 }
4217 }
4218
4219 /* set uncompleted segment to curseg */
4220 curseg = CURSEG_I(sbi, type);
4221 mutex_lock(&curseg->curseg_mutex);
4222
4223 /* update journal info */
4224 down_write(&curseg->journal_rwsem);
4225 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
4226 up_write(&curseg->journal_rwsem);
4227
4228 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
4229 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
4230 curseg->next_segno = segno;
4231 reset_curseg(sbi, type, 0);
4232 curseg->alloc_type = ckpt->alloc_type[type];
4233 curseg->next_blkoff = blk_off;
4234 mutex_unlock(&curseg->curseg_mutex);
4235 out:
4236 f2fs_put_page(new, 1);
4237 return err;
4238 }
4239
restore_curseg_summaries(struct f2fs_sb_info * sbi)4240 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
4241 {
4242 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
4243 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
4244 int type = CURSEG_HOT_DATA;
4245 int err;
4246
4247 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4248 int npages = f2fs_npages_for_summary_flush(sbi, true);
4249
4250 if (npages >= 2)
4251 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
4252 META_CP, true);
4253
4254 /* restore for compacted data summary */
4255 err = read_compacted_summaries(sbi);
4256 if (err)
4257 return err;
4258 type = CURSEG_HOT_NODE;
4259 }
4260
4261 if (__exist_node_summaries(sbi))
4262 f2fs_ra_meta_pages(sbi,
4263 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4264 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4265
4266 for (; type <= CURSEG_COLD_NODE; type++) {
4267 err = read_normal_summaries(sbi, type);
4268 if (err)
4269 return err;
4270 }
4271
4272 /* sanity check for summary blocks */
4273 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4274 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4275 f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4276 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4277 return -EINVAL;
4278 }
4279
4280 return 0;
4281 }
4282
write_compacted_summaries(struct f2fs_sb_info * sbi,block_t blkaddr)4283 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4284 {
4285 struct page *page;
4286 unsigned char *kaddr;
4287 struct f2fs_summary *summary;
4288 struct curseg_info *seg_i;
4289 int written_size = 0;
4290 int i, j;
4291
4292 page = f2fs_grab_meta_page(sbi, blkaddr++);
4293 kaddr = (unsigned char *)page_address(page);
4294 memset(kaddr, 0, PAGE_SIZE);
4295
4296 /* Step 1: write nat cache */
4297 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4298 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4299 written_size += SUM_JOURNAL_SIZE;
4300
4301 /* Step 2: write sit cache */
4302 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4303 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4304 written_size += SUM_JOURNAL_SIZE;
4305
4306 /* Step 3: write summary entries */
4307 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4308 seg_i = CURSEG_I(sbi, i);
4309 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) {
4310 if (!page) {
4311 page = f2fs_grab_meta_page(sbi, blkaddr++);
4312 kaddr = (unsigned char *)page_address(page);
4313 memset(kaddr, 0, PAGE_SIZE);
4314 written_size = 0;
4315 }
4316 summary = (struct f2fs_summary *)(kaddr + written_size);
4317 *summary = seg_i->sum_blk->entries[j];
4318 written_size += SUMMARY_SIZE;
4319
4320 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4321 SUM_FOOTER_SIZE)
4322 continue;
4323
4324 set_page_dirty(page);
4325 f2fs_put_page(page, 1);
4326 page = NULL;
4327 }
4328 }
4329 if (page) {
4330 set_page_dirty(page);
4331 f2fs_put_page(page, 1);
4332 }
4333 }
4334
write_normal_summaries(struct f2fs_sb_info * sbi,block_t blkaddr,int type)4335 static void write_normal_summaries(struct f2fs_sb_info *sbi,
4336 block_t blkaddr, int type)
4337 {
4338 int i, end;
4339
4340 if (IS_DATASEG(type))
4341 end = type + NR_CURSEG_DATA_TYPE;
4342 else
4343 end = type + NR_CURSEG_NODE_TYPE;
4344
4345 for (i = type; i < end; i++)
4346 write_current_sum_page(sbi, i, blkaddr + (i - type));
4347 }
4348
f2fs_write_data_summaries(struct f2fs_sb_info * sbi,block_t start_blk)4349 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4350 {
4351 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4352 write_compacted_summaries(sbi, start_blk);
4353 else
4354 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4355 }
4356
f2fs_write_node_summaries(struct f2fs_sb_info * sbi,block_t start_blk)4357 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4358 {
4359 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4360 }
4361
f2fs_lookup_journal_in_cursum(struct f2fs_journal * journal,int type,unsigned int val,int alloc)4362 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4363 unsigned int val, int alloc)
4364 {
4365 int i;
4366
4367 if (type == NAT_JOURNAL) {
4368 for (i = 0; i < nats_in_cursum(journal); i++) {
4369 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4370 return i;
4371 }
4372 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4373 return update_nats_in_cursum(journal, 1);
4374 } else if (type == SIT_JOURNAL) {
4375 for (i = 0; i < sits_in_cursum(journal); i++)
4376 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4377 return i;
4378 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4379 return update_sits_in_cursum(journal, 1);
4380 }
4381 return -1;
4382 }
4383
get_current_sit_page(struct f2fs_sb_info * sbi,unsigned int segno)4384 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4385 unsigned int segno)
4386 {
4387 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4388 }
4389
get_next_sit_page(struct f2fs_sb_info * sbi,unsigned int start)4390 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4391 unsigned int start)
4392 {
4393 struct sit_info *sit_i = SIT_I(sbi);
4394 struct page *page;
4395 pgoff_t src_off, dst_off;
4396
4397 src_off = current_sit_addr(sbi, start);
4398 dst_off = next_sit_addr(sbi, src_off);
4399
4400 page = f2fs_grab_meta_page(sbi, dst_off);
4401 seg_info_to_sit_page(sbi, page, start);
4402
4403 set_page_dirty(page);
4404 set_to_next_sit(sit_i, start);
4405
4406 return page;
4407 }
4408
grab_sit_entry_set(void)4409 static struct sit_entry_set *grab_sit_entry_set(void)
4410 {
4411 struct sit_entry_set *ses =
4412 f2fs_kmem_cache_alloc(sit_entry_set_slab,
4413 GFP_NOFS, true, NULL);
4414
4415 ses->entry_cnt = 0;
4416 INIT_LIST_HEAD(&ses->set_list);
4417 return ses;
4418 }
4419
release_sit_entry_set(struct sit_entry_set * ses)4420 static void release_sit_entry_set(struct sit_entry_set *ses)
4421 {
4422 list_del(&ses->set_list);
4423 kmem_cache_free(sit_entry_set_slab, ses);
4424 }
4425
adjust_sit_entry_set(struct sit_entry_set * ses,struct list_head * head)4426 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4427 struct list_head *head)
4428 {
4429 struct sit_entry_set *next = ses;
4430
4431 if (list_is_last(&ses->set_list, head))
4432 return;
4433
4434 list_for_each_entry_continue(next, head, set_list)
4435 if (ses->entry_cnt <= next->entry_cnt) {
4436 list_move_tail(&ses->set_list, &next->set_list);
4437 return;
4438 }
4439
4440 list_move_tail(&ses->set_list, head);
4441 }
4442
add_sit_entry(unsigned int segno,struct list_head * head)4443 static void add_sit_entry(unsigned int segno, struct list_head *head)
4444 {
4445 struct sit_entry_set *ses;
4446 unsigned int start_segno = START_SEGNO(segno);
4447
4448 list_for_each_entry(ses, head, set_list) {
4449 if (ses->start_segno == start_segno) {
4450 ses->entry_cnt++;
4451 adjust_sit_entry_set(ses, head);
4452 return;
4453 }
4454 }
4455
4456 ses = grab_sit_entry_set();
4457
4458 ses->start_segno = start_segno;
4459 ses->entry_cnt++;
4460 list_add(&ses->set_list, head);
4461 }
4462
add_sits_in_set(struct f2fs_sb_info * sbi)4463 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4464 {
4465 struct f2fs_sm_info *sm_info = SM_I(sbi);
4466 struct list_head *set_list = &sm_info->sit_entry_set;
4467 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4468 unsigned int segno;
4469
4470 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4471 add_sit_entry(segno, set_list);
4472 }
4473
remove_sits_in_journal(struct f2fs_sb_info * sbi)4474 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4475 {
4476 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4477 struct f2fs_journal *journal = curseg->journal;
4478 int i;
4479
4480 down_write(&curseg->journal_rwsem);
4481 for (i = 0; i < sits_in_cursum(journal); i++) {
4482 unsigned int segno;
4483 bool dirtied;
4484
4485 segno = le32_to_cpu(segno_in_journal(journal, i));
4486 dirtied = __mark_sit_entry_dirty(sbi, segno);
4487
4488 if (!dirtied)
4489 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4490 }
4491 update_sits_in_cursum(journal, -i);
4492 up_write(&curseg->journal_rwsem);
4493 }
4494
4495 /*
4496 * CP calls this function, which flushes SIT entries including sit_journal,
4497 * and moves prefree segs to free segs.
4498 */
f2fs_flush_sit_entries(struct f2fs_sb_info * sbi,struct cp_control * cpc)4499 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4500 {
4501 struct sit_info *sit_i = SIT_I(sbi);
4502 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4503 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4504 struct f2fs_journal *journal = curseg->journal;
4505 struct sit_entry_set *ses, *tmp;
4506 struct list_head *head = &SM_I(sbi)->sit_entry_set;
4507 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4508 struct seg_entry *se;
4509
4510 down_write(&sit_i->sentry_lock);
4511
4512 if (!sit_i->dirty_sentries)
4513 goto out;
4514
4515 /*
4516 * add and account sit entries of dirty bitmap in sit entry
4517 * set temporarily
4518 */
4519 add_sits_in_set(sbi);
4520
4521 /*
4522 * if there are no enough space in journal to store dirty sit
4523 * entries, remove all entries from journal and add and account
4524 * them in sit entry set.
4525 */
4526 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4527 !to_journal)
4528 remove_sits_in_journal(sbi);
4529
4530 /*
4531 * there are two steps to flush sit entries:
4532 * #1, flush sit entries to journal in current cold data summary block.
4533 * #2, flush sit entries to sit page.
4534 */
4535 list_for_each_entry_safe(ses, tmp, head, set_list) {
4536 struct page *page = NULL;
4537 struct f2fs_sit_block *raw_sit = NULL;
4538 unsigned int start_segno = ses->start_segno;
4539 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4540 (unsigned long)MAIN_SEGS(sbi));
4541 unsigned int segno = start_segno;
4542
4543 if (to_journal &&
4544 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4545 to_journal = false;
4546
4547 if (to_journal) {
4548 down_write(&curseg->journal_rwsem);
4549 } else {
4550 page = get_next_sit_page(sbi, start_segno);
4551 raw_sit = page_address(page);
4552 }
4553
4554 /* flush dirty sit entries in region of current sit set */
4555 for_each_set_bit_from(segno, bitmap, end) {
4556 int offset, sit_offset;
4557
4558 se = get_seg_entry(sbi, segno);
4559 #ifdef CONFIG_F2FS_CHECK_FS
4560 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4561 SIT_VBLOCK_MAP_SIZE))
4562 f2fs_bug_on(sbi, 1);
4563 #endif
4564
4565 /* add discard candidates */
4566 if (!(cpc->reason & CP_DISCARD)) {
4567 cpc->trim_start = segno;
4568 add_discard_addrs(sbi, cpc, false);
4569 }
4570
4571 if (to_journal) {
4572 offset = f2fs_lookup_journal_in_cursum(journal,
4573 SIT_JOURNAL, segno, 1);
4574 f2fs_bug_on(sbi, offset < 0);
4575 segno_in_journal(journal, offset) =
4576 cpu_to_le32(segno);
4577 seg_info_to_raw_sit(se,
4578 &sit_in_journal(journal, offset));
4579 check_block_count(sbi, segno,
4580 &sit_in_journal(journal, offset));
4581 } else {
4582 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4583 seg_info_to_raw_sit(se,
4584 &raw_sit->entries[sit_offset]);
4585 check_block_count(sbi, segno,
4586 &raw_sit->entries[sit_offset]);
4587 }
4588
4589 __clear_bit(segno, bitmap);
4590 sit_i->dirty_sentries--;
4591 ses->entry_cnt--;
4592 }
4593
4594 if (to_journal)
4595 up_write(&curseg->journal_rwsem);
4596 else
4597 f2fs_put_page(page, 1);
4598
4599 f2fs_bug_on(sbi, ses->entry_cnt);
4600 release_sit_entry_set(ses);
4601 }
4602
4603 f2fs_bug_on(sbi, !list_empty(head));
4604 f2fs_bug_on(sbi, sit_i->dirty_sentries);
4605 out:
4606 if (cpc->reason & CP_DISCARD) {
4607 __u64 trim_start = cpc->trim_start;
4608
4609 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4610 add_discard_addrs(sbi, cpc, false);
4611
4612 cpc->trim_start = trim_start;
4613 }
4614 up_write(&sit_i->sentry_lock);
4615
4616 set_prefree_as_free_segments(sbi);
4617 }
4618
build_sit_info(struct f2fs_sb_info * sbi)4619 static int build_sit_info(struct f2fs_sb_info *sbi)
4620 {
4621 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4622 struct sit_info *sit_i;
4623 unsigned int sit_segs, start;
4624 char *src_bitmap, *bitmap;
4625 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4626 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4627
4628 /* allocate memory for SIT information */
4629 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4630 if (!sit_i)
4631 return -ENOMEM;
4632
4633 SM_I(sbi)->sit_info = sit_i;
4634
4635 sit_i->sentries =
4636 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4637 MAIN_SEGS(sbi)),
4638 GFP_KERNEL);
4639 if (!sit_i->sentries)
4640 return -ENOMEM;
4641
4642 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4643 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4644 GFP_KERNEL);
4645 if (!sit_i->dirty_sentries_bitmap)
4646 return -ENOMEM;
4647
4648 #ifdef CONFIG_F2FS_CHECK_FS
4649 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4650 #else
4651 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4652 #endif
4653 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4654 if (!sit_i->bitmap)
4655 return -ENOMEM;
4656
4657 bitmap = sit_i->bitmap;
4658
4659 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4660 sit_i->sentries[start].cur_valid_map = bitmap;
4661 bitmap += SIT_VBLOCK_MAP_SIZE;
4662
4663 sit_i->sentries[start].ckpt_valid_map = bitmap;
4664 bitmap += SIT_VBLOCK_MAP_SIZE;
4665
4666 #ifdef CONFIG_F2FS_CHECK_FS
4667 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4668 bitmap += SIT_VBLOCK_MAP_SIZE;
4669 #endif
4670
4671 if (discard_map) {
4672 sit_i->sentries[start].discard_map = bitmap;
4673 bitmap += SIT_VBLOCK_MAP_SIZE;
4674 }
4675 }
4676
4677 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4678 if (!sit_i->tmp_map)
4679 return -ENOMEM;
4680
4681 if (__is_large_section(sbi)) {
4682 sit_i->sec_entries =
4683 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4684 MAIN_SECS(sbi)),
4685 GFP_KERNEL);
4686 if (!sit_i->sec_entries)
4687 return -ENOMEM;
4688 }
4689
4690 /* get information related with SIT */
4691 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4692
4693 /* setup SIT bitmap from ckeckpoint pack */
4694 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4695 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4696
4697 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4698 if (!sit_i->sit_bitmap)
4699 return -ENOMEM;
4700
4701 #ifdef CONFIG_F2FS_CHECK_FS
4702 sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4703 sit_bitmap_size, GFP_KERNEL);
4704 if (!sit_i->sit_bitmap_mir)
4705 return -ENOMEM;
4706
4707 sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4708 main_bitmap_size, GFP_KERNEL);
4709 if (!sit_i->invalid_segmap)
4710 return -ENOMEM;
4711 #endif
4712
4713 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4714 sit_i->sit_blocks = SEGS_TO_BLKS(sbi, sit_segs);
4715 sit_i->written_valid_blocks = 0;
4716 sit_i->bitmap_size = sit_bitmap_size;
4717 sit_i->dirty_sentries = 0;
4718 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4719 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4720 sit_i->mounted_time = ktime_get_boottime_seconds();
4721 init_rwsem(&sit_i->sentry_lock);
4722 return 0;
4723 }
4724
build_free_segmap(struct f2fs_sb_info * sbi)4725 static int build_free_segmap(struct f2fs_sb_info *sbi)
4726 {
4727 struct free_segmap_info *free_i;
4728 unsigned int bitmap_size, sec_bitmap_size;
4729
4730 /* allocate memory for free segmap information */
4731 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4732 if (!free_i)
4733 return -ENOMEM;
4734
4735 SM_I(sbi)->free_info = free_i;
4736
4737 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4738 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4739 if (!free_i->free_segmap)
4740 return -ENOMEM;
4741
4742 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4743 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4744 if (!free_i->free_secmap)
4745 return -ENOMEM;
4746
4747 /* set all segments as dirty temporarily */
4748 memset(free_i->free_segmap, 0xff, bitmap_size);
4749 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4750
4751 /* init free segmap information */
4752 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4753 free_i->free_segments = 0;
4754 free_i->free_sections = 0;
4755 spin_lock_init(&free_i->segmap_lock);
4756 return 0;
4757 }
4758
build_curseg(struct f2fs_sb_info * sbi)4759 static int build_curseg(struct f2fs_sb_info *sbi)
4760 {
4761 struct curseg_info *array;
4762 int i;
4763
4764 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4765 sizeof(*array)), GFP_KERNEL);
4766 if (!array)
4767 return -ENOMEM;
4768
4769 SM_I(sbi)->curseg_array = array;
4770
4771 for (i = 0; i < NO_CHECK_TYPE; i++) {
4772 mutex_init(&array[i].curseg_mutex);
4773 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4774 if (!array[i].sum_blk)
4775 return -ENOMEM;
4776 init_rwsem(&array[i].journal_rwsem);
4777 array[i].journal = f2fs_kzalloc(sbi,
4778 sizeof(struct f2fs_journal), GFP_KERNEL);
4779 if (!array[i].journal)
4780 return -ENOMEM;
4781 if (i < NR_PERSISTENT_LOG)
4782 array[i].seg_type = CURSEG_HOT_DATA + i;
4783 else if (i == CURSEG_COLD_DATA_PINNED)
4784 array[i].seg_type = CURSEG_COLD_DATA;
4785 else if (i == CURSEG_ALL_DATA_ATGC)
4786 array[i].seg_type = CURSEG_COLD_DATA;
4787 reset_curseg_fields(&array[i]);
4788 }
4789 return restore_curseg_summaries(sbi);
4790 }
4791
build_sit_entries(struct f2fs_sb_info * sbi)4792 static int build_sit_entries(struct f2fs_sb_info *sbi)
4793 {
4794 struct sit_info *sit_i = SIT_I(sbi);
4795 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4796 struct f2fs_journal *journal = curseg->journal;
4797 struct seg_entry *se;
4798 struct f2fs_sit_entry sit;
4799 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4800 unsigned int i, start, end;
4801 unsigned int readed, start_blk = 0;
4802 int err = 0;
4803 block_t sit_valid_blocks[2] = {0, 0};
4804
4805 do {
4806 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4807 META_SIT, true);
4808
4809 start = start_blk * sit_i->sents_per_block;
4810 end = (start_blk + readed) * sit_i->sents_per_block;
4811
4812 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4813 struct f2fs_sit_block *sit_blk;
4814 struct page *page;
4815
4816 se = &sit_i->sentries[start];
4817 page = get_current_sit_page(sbi, start);
4818 if (IS_ERR(page))
4819 return PTR_ERR(page);
4820 sit_blk = (struct f2fs_sit_block *)page_address(page);
4821 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4822 f2fs_put_page(page, 1);
4823
4824 err = check_block_count(sbi, start, &sit);
4825 if (err)
4826 return err;
4827 seg_info_from_raw_sit(se, &sit);
4828
4829 if (se->type >= NR_PERSISTENT_LOG) {
4830 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4831 se->type, start);
4832 f2fs_handle_error(sbi,
4833 ERROR_INCONSISTENT_SUM_TYPE);
4834 return -EFSCORRUPTED;
4835 }
4836
4837 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4838
4839 if (!f2fs_block_unit_discard(sbi))
4840 goto init_discard_map_done;
4841
4842 /* build discard map only one time */
4843 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4844 memset(se->discard_map, 0xff,
4845 SIT_VBLOCK_MAP_SIZE);
4846 goto init_discard_map_done;
4847 }
4848 memcpy(se->discard_map, se->cur_valid_map,
4849 SIT_VBLOCK_MAP_SIZE);
4850 sbi->discard_blks += BLKS_PER_SEG(sbi) -
4851 se->valid_blocks;
4852 init_discard_map_done:
4853 if (__is_large_section(sbi))
4854 get_sec_entry(sbi, start)->valid_blocks +=
4855 se->valid_blocks;
4856 }
4857 start_blk += readed;
4858 } while (start_blk < sit_blk_cnt);
4859
4860 down_read(&curseg->journal_rwsem);
4861 for (i = 0; i < sits_in_cursum(journal); i++) {
4862 unsigned int old_valid_blocks;
4863
4864 start = le32_to_cpu(segno_in_journal(journal, i));
4865 if (start >= MAIN_SEGS(sbi)) {
4866 f2fs_err(sbi, "Wrong journal entry on segno %u",
4867 start);
4868 err = -EFSCORRUPTED;
4869 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4870 break;
4871 }
4872
4873 se = &sit_i->sentries[start];
4874 sit = sit_in_journal(journal, i);
4875
4876 old_valid_blocks = se->valid_blocks;
4877
4878 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4879
4880 err = check_block_count(sbi, start, &sit);
4881 if (err)
4882 break;
4883 seg_info_from_raw_sit(se, &sit);
4884
4885 if (se->type >= NR_PERSISTENT_LOG) {
4886 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4887 se->type, start);
4888 err = -EFSCORRUPTED;
4889 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4890 break;
4891 }
4892
4893 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4894
4895 if (f2fs_block_unit_discard(sbi)) {
4896 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4897 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4898 } else {
4899 memcpy(se->discard_map, se->cur_valid_map,
4900 SIT_VBLOCK_MAP_SIZE);
4901 sbi->discard_blks += old_valid_blocks;
4902 sbi->discard_blks -= se->valid_blocks;
4903 }
4904 }
4905
4906 if (__is_large_section(sbi)) {
4907 get_sec_entry(sbi, start)->valid_blocks +=
4908 se->valid_blocks;
4909 get_sec_entry(sbi, start)->valid_blocks -=
4910 old_valid_blocks;
4911 }
4912 }
4913 up_read(&curseg->journal_rwsem);
4914
4915 if (err)
4916 return err;
4917
4918 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4919 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4920 sit_valid_blocks[NODE], valid_node_count(sbi));
4921 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4922 return -EFSCORRUPTED;
4923 }
4924
4925 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4926 valid_user_blocks(sbi)) {
4927 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4928 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4929 valid_user_blocks(sbi));
4930 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4931 return -EFSCORRUPTED;
4932 }
4933
4934 return 0;
4935 }
4936
init_free_segmap(struct f2fs_sb_info * sbi)4937 static void init_free_segmap(struct f2fs_sb_info *sbi)
4938 {
4939 unsigned int start;
4940 int type;
4941 struct seg_entry *sentry;
4942
4943 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4944 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4945 continue;
4946 sentry = get_seg_entry(sbi, start);
4947 if (!sentry->valid_blocks)
4948 __set_free(sbi, start);
4949 else
4950 SIT_I(sbi)->written_valid_blocks +=
4951 sentry->valid_blocks;
4952 }
4953
4954 /* set use the current segments */
4955 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4956 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4957
4958 __set_test_and_inuse(sbi, curseg_t->segno);
4959 }
4960 }
4961
init_dirty_segmap(struct f2fs_sb_info * sbi)4962 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4963 {
4964 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4965 struct free_segmap_info *free_i = FREE_I(sbi);
4966 unsigned int segno = 0, offset = 0, secno;
4967 block_t valid_blocks, usable_blks_in_seg;
4968
4969 while (1) {
4970 /* find dirty segment based on free segmap */
4971 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4972 if (segno >= MAIN_SEGS(sbi))
4973 break;
4974 offset = segno + 1;
4975 valid_blocks = get_valid_blocks(sbi, segno, false);
4976 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4977 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4978 continue;
4979 if (valid_blocks > usable_blks_in_seg) {
4980 f2fs_bug_on(sbi, 1);
4981 continue;
4982 }
4983 mutex_lock(&dirty_i->seglist_lock);
4984 __locate_dirty_segment(sbi, segno, DIRTY);
4985 mutex_unlock(&dirty_i->seglist_lock);
4986 }
4987
4988 if (!__is_large_section(sbi))
4989 return;
4990
4991 mutex_lock(&dirty_i->seglist_lock);
4992 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
4993 valid_blocks = get_valid_blocks(sbi, segno, true);
4994 secno = GET_SEC_FROM_SEG(sbi, segno);
4995
4996 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4997 continue;
4998 if (IS_CURSEC(sbi, secno))
4999 continue;
5000 set_bit(secno, dirty_i->dirty_secmap);
5001 }
5002 mutex_unlock(&dirty_i->seglist_lock);
5003 }
5004
init_victim_secmap(struct f2fs_sb_info * sbi)5005 static int init_victim_secmap(struct f2fs_sb_info *sbi)
5006 {
5007 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5008 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
5009
5010 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5011 if (!dirty_i->victim_secmap)
5012 return -ENOMEM;
5013
5014 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5015 if (!dirty_i->pinned_secmap)
5016 return -ENOMEM;
5017
5018 dirty_i->pinned_secmap_cnt = 0;
5019 dirty_i->enable_pin_section = true;
5020 return 0;
5021 }
5022
build_dirty_segmap(struct f2fs_sb_info * sbi)5023 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
5024 {
5025 struct dirty_seglist_info *dirty_i;
5026 unsigned int bitmap_size, i;
5027
5028 /* allocate memory for dirty segments list information */
5029 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
5030 GFP_KERNEL);
5031 if (!dirty_i)
5032 return -ENOMEM;
5033
5034 SM_I(sbi)->dirty_info = dirty_i;
5035 mutex_init(&dirty_i->seglist_lock);
5036
5037 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
5038
5039 for (i = 0; i < NR_DIRTY_TYPE; i++) {
5040 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
5041 GFP_KERNEL);
5042 if (!dirty_i->dirty_segmap[i])
5043 return -ENOMEM;
5044 }
5045
5046 if (__is_large_section(sbi)) {
5047 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
5048 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
5049 bitmap_size, GFP_KERNEL);
5050 if (!dirty_i->dirty_secmap)
5051 return -ENOMEM;
5052 }
5053
5054 init_dirty_segmap(sbi);
5055 return init_victim_secmap(sbi);
5056 }
5057
sanity_check_curseg(struct f2fs_sb_info * sbi)5058 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
5059 {
5060 int i;
5061
5062 /*
5063 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
5064 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
5065 */
5066 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5067 struct curseg_info *curseg = CURSEG_I(sbi, i);
5068 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
5069 unsigned int blkofs = curseg->next_blkoff;
5070
5071 if (f2fs_sb_has_readonly(sbi) &&
5072 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
5073 continue;
5074
5075 sanity_check_seg_type(sbi, curseg->seg_type);
5076
5077 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
5078 f2fs_err(sbi,
5079 "Current segment has invalid alloc_type:%d",
5080 curseg->alloc_type);
5081 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
5082 return -EFSCORRUPTED;
5083 }
5084
5085 if (f2fs_test_bit(blkofs, se->cur_valid_map))
5086 goto out;
5087
5088 if (curseg->alloc_type == SSR)
5089 continue;
5090
5091 for (blkofs += 1; blkofs < BLKS_PER_SEG(sbi); blkofs++) {
5092 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
5093 continue;
5094 out:
5095 f2fs_err(sbi,
5096 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
5097 i, curseg->segno, curseg->alloc_type,
5098 curseg->next_blkoff, blkofs);
5099 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
5100 return -EFSCORRUPTED;
5101 }
5102 }
5103 return 0;
5104 }
5105
5106 #ifdef CONFIG_BLK_DEV_ZONED
check_zone_write_pointer(struct f2fs_sb_info * sbi,struct f2fs_dev_info * fdev,struct blk_zone * zone)5107 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
5108 struct f2fs_dev_info *fdev,
5109 struct blk_zone *zone)
5110 {
5111 unsigned int zone_segno;
5112 block_t zone_block, valid_block_cnt;
5113 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
5114 int ret;
5115 unsigned int nofs_flags;
5116
5117 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5118 return 0;
5119
5120 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
5121 zone_segno = GET_SEGNO(sbi, zone_block);
5122
5123 /*
5124 * Skip check of zones cursegs point to, since
5125 * fix_curseg_write_pointer() checks them.
5126 */
5127 if (zone_segno >= MAIN_SEGS(sbi))
5128 return 0;
5129
5130 /*
5131 * Get # of valid block of the zone.
5132 */
5133 valid_block_cnt = get_valid_blocks(sbi, zone_segno, true);
5134 if (IS_CURSEC(sbi, GET_SEC_FROM_SEG(sbi, zone_segno))) {
5135 f2fs_notice(sbi, "Open zones: valid block[0x%x,0x%x] cond[%s]",
5136 zone_segno, valid_block_cnt,
5137 blk_zone_cond_str(zone->cond));
5138 return 0;
5139 }
5140
5141 if ((!valid_block_cnt && zone->cond == BLK_ZONE_COND_EMPTY) ||
5142 (valid_block_cnt && zone->cond == BLK_ZONE_COND_FULL))
5143 return 0;
5144
5145 if (!valid_block_cnt) {
5146 f2fs_notice(sbi, "Zone without valid block has non-zero write "
5147 "pointer. Reset the write pointer: cond[%s]",
5148 blk_zone_cond_str(zone->cond));
5149 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
5150 zone->len >> log_sectors_per_block);
5151 if (ret)
5152 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5153 fdev->path, ret);
5154 return ret;
5155 }
5156
5157 /*
5158 * If there are valid blocks and the write pointer doesn't match
5159 * with them, we need to report the inconsistency and fill
5160 * the zone till the end to close the zone. This inconsistency
5161 * does not cause write error because the zone will not be
5162 * selected for write operation until it get discarded.
5163 */
5164 f2fs_notice(sbi, "Valid blocks are not aligned with write "
5165 "pointer: valid block[0x%x,0x%x] cond[%s]",
5166 zone_segno, valid_block_cnt, blk_zone_cond_str(zone->cond));
5167
5168 nofs_flags = memalloc_nofs_save();
5169 ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH,
5170 zone->start, zone->len);
5171 memalloc_nofs_restore(nofs_flags);
5172 if (ret == -EOPNOTSUPP) {
5173 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp,
5174 zone->len - (zone->wp - zone->start),
5175 GFP_NOFS, 0);
5176 if (ret)
5177 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)",
5178 fdev->path, ret);
5179 } else if (ret) {
5180 f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)",
5181 fdev->path, ret);
5182 }
5183
5184 return ret;
5185 }
5186
get_target_zoned_dev(struct f2fs_sb_info * sbi,block_t zone_blkaddr)5187 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
5188 block_t zone_blkaddr)
5189 {
5190 int i;
5191
5192 for (i = 0; i < sbi->s_ndevs; i++) {
5193 if (!bdev_is_zoned(FDEV(i).bdev))
5194 continue;
5195 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
5196 zone_blkaddr <= FDEV(i).end_blk))
5197 return &FDEV(i);
5198 }
5199
5200 return NULL;
5201 }
5202
report_one_zone_cb(struct blk_zone * zone,unsigned int idx,void * data)5203 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
5204 void *data)
5205 {
5206 memcpy(data, zone, sizeof(struct blk_zone));
5207 return 0;
5208 }
5209
fix_curseg_write_pointer(struct f2fs_sb_info * sbi,int type)5210 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
5211 {
5212 struct curseg_info *cs = CURSEG_I(sbi, type);
5213 struct f2fs_dev_info *zbd;
5214 struct blk_zone zone;
5215 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
5216 block_t cs_zone_block, wp_block;
5217 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
5218 sector_t zone_sector;
5219 int err;
5220
5221 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5222 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5223
5224 zbd = get_target_zoned_dev(sbi, cs_zone_block);
5225 if (!zbd)
5226 return 0;
5227
5228 /* report zone for the sector the curseg points to */
5229 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5230 << log_sectors_per_block;
5231 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5232 report_one_zone_cb, &zone);
5233 if (err != 1) {
5234 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5235 zbd->path, err);
5236 return err;
5237 }
5238
5239 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5240 return 0;
5241
5242 /*
5243 * When safely unmounted in the previous mount, we could use current
5244 * segments. Otherwise, allocate new sections.
5245 */
5246 if (is_set_ckpt_flags(sbi, CP_UMOUNT_FLAG)) {
5247 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
5248 wp_segno = GET_SEGNO(sbi, wp_block);
5249 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
5250 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
5251
5252 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
5253 wp_sector_off == 0)
5254 return 0;
5255
5256 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
5257 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", type, cs->segno,
5258 cs->next_blkoff, wp_segno, wp_blkoff);
5259 }
5260
5261 /* Allocate a new section if it's not new. */
5262 if (cs->next_blkoff ||
5263 cs->segno != GET_SEG_FROM_SEC(sbi, GET_ZONE_FROM_SEC(sbi, cs_section))) {
5264 unsigned int old_segno = cs->segno, old_blkoff = cs->next_blkoff;
5265
5266 f2fs_allocate_new_section(sbi, type, true);
5267 f2fs_notice(sbi, "Assign new section to curseg[%d]: "
5268 "[0x%x,0x%x] -> [0x%x,0x%x]",
5269 type, old_segno, old_blkoff,
5270 cs->segno, cs->next_blkoff);
5271 }
5272
5273 /* check consistency of the zone curseg pointed to */
5274 if (check_zone_write_pointer(sbi, zbd, &zone))
5275 return -EIO;
5276
5277 /* check newly assigned zone */
5278 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5279 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5280
5281 zbd = get_target_zoned_dev(sbi, cs_zone_block);
5282 if (!zbd)
5283 return 0;
5284
5285 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5286 << log_sectors_per_block;
5287 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5288 report_one_zone_cb, &zone);
5289 if (err != 1) {
5290 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5291 zbd->path, err);
5292 return err;
5293 }
5294
5295 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5296 return 0;
5297
5298 if (zone.wp != zone.start) {
5299 f2fs_notice(sbi,
5300 "New zone for curseg[%d] is not yet discarded. "
5301 "Reset the zone: curseg[0x%x,0x%x]",
5302 type, cs->segno, cs->next_blkoff);
5303 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block,
5304 zone.len >> log_sectors_per_block);
5305 if (err) {
5306 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5307 zbd->path, err);
5308 return err;
5309 }
5310 }
5311
5312 return 0;
5313 }
5314
f2fs_fix_curseg_write_pointer(struct f2fs_sb_info * sbi)5315 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5316 {
5317 int i, ret;
5318
5319 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5320 ret = fix_curseg_write_pointer(sbi, i);
5321 if (ret)
5322 return ret;
5323 }
5324
5325 return 0;
5326 }
5327
5328 struct check_zone_write_pointer_args {
5329 struct f2fs_sb_info *sbi;
5330 struct f2fs_dev_info *fdev;
5331 };
5332
check_zone_write_pointer_cb(struct blk_zone * zone,unsigned int idx,void * data)5333 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5334 void *data)
5335 {
5336 struct check_zone_write_pointer_args *args;
5337
5338 args = (struct check_zone_write_pointer_args *)data;
5339
5340 return check_zone_write_pointer(args->sbi, args->fdev, zone);
5341 }
5342
f2fs_check_write_pointer(struct f2fs_sb_info * sbi)5343 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5344 {
5345 int i, ret;
5346 struct check_zone_write_pointer_args args;
5347
5348 for (i = 0; i < sbi->s_ndevs; i++) {
5349 if (!bdev_is_zoned(FDEV(i).bdev))
5350 continue;
5351
5352 args.sbi = sbi;
5353 args.fdev = &FDEV(i);
5354 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5355 check_zone_write_pointer_cb, &args);
5356 if (ret < 0)
5357 return ret;
5358 }
5359
5360 return 0;
5361 }
5362
5363 /*
5364 * Return the number of usable blocks in a segment. The number of blocks
5365 * returned is always equal to the number of blocks in a segment for
5366 * segments fully contained within a sequential zone capacity or a
5367 * conventional zone. For segments partially contained in a sequential
5368 * zone capacity, the number of usable blocks up to the zone capacity
5369 * is returned. 0 is returned in all other cases.
5370 */
f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info * sbi,unsigned int segno)5371 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5372 struct f2fs_sb_info *sbi, unsigned int segno)
5373 {
5374 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5375 unsigned int secno;
5376
5377 if (!sbi->unusable_blocks_per_sec)
5378 return BLKS_PER_SEG(sbi);
5379
5380 secno = GET_SEC_FROM_SEG(sbi, segno);
5381 seg_start = START_BLOCK(sbi, segno);
5382 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5383 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5384
5385 /*
5386 * If segment starts before zone capacity and spans beyond
5387 * zone capacity, then usable blocks are from seg start to
5388 * zone capacity. If the segment starts after the zone capacity,
5389 * then there are no usable blocks.
5390 */
5391 if (seg_start >= sec_cap_blkaddr)
5392 return 0;
5393 if (seg_start + BLKS_PER_SEG(sbi) > sec_cap_blkaddr)
5394 return sec_cap_blkaddr - seg_start;
5395
5396 return BLKS_PER_SEG(sbi);
5397 }
5398 #else
f2fs_fix_curseg_write_pointer(struct f2fs_sb_info * sbi)5399 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5400 {
5401 return 0;
5402 }
5403
f2fs_check_write_pointer(struct f2fs_sb_info * sbi)5404 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5405 {
5406 return 0;
5407 }
5408
f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info * sbi,unsigned int segno)5409 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5410 unsigned int segno)
5411 {
5412 return 0;
5413 }
5414
5415 #endif
f2fs_usable_blks_in_seg(struct f2fs_sb_info * sbi,unsigned int segno)5416 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5417 unsigned int segno)
5418 {
5419 if (f2fs_sb_has_blkzoned(sbi))
5420 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5421
5422 return BLKS_PER_SEG(sbi);
5423 }
5424
f2fs_usable_segs_in_sec(struct f2fs_sb_info * sbi)5425 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi)
5426 {
5427 if (f2fs_sb_has_blkzoned(sbi))
5428 return CAP_SEGS_PER_SEC(sbi);
5429
5430 return SEGS_PER_SEC(sbi);
5431 }
5432
5433 /*
5434 * Update min, max modified time for cost-benefit GC algorithm
5435 */
init_min_max_mtime(struct f2fs_sb_info * sbi)5436 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5437 {
5438 struct sit_info *sit_i = SIT_I(sbi);
5439 unsigned int segno;
5440
5441 down_write(&sit_i->sentry_lock);
5442
5443 sit_i->min_mtime = ULLONG_MAX;
5444
5445 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
5446 unsigned int i;
5447 unsigned long long mtime = 0;
5448
5449 for (i = 0; i < SEGS_PER_SEC(sbi); i++)
5450 mtime += get_seg_entry(sbi, segno + i)->mtime;
5451
5452 mtime = div_u64(mtime, SEGS_PER_SEC(sbi));
5453
5454 if (sit_i->min_mtime > mtime)
5455 sit_i->min_mtime = mtime;
5456 }
5457 sit_i->max_mtime = get_mtime(sbi, false);
5458 sit_i->dirty_max_mtime = 0;
5459 up_write(&sit_i->sentry_lock);
5460 }
5461
f2fs_build_segment_manager(struct f2fs_sb_info * sbi)5462 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5463 {
5464 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5465 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5466 struct f2fs_sm_info *sm_info;
5467 int err;
5468
5469 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5470 if (!sm_info)
5471 return -ENOMEM;
5472
5473 /* init sm info */
5474 sbi->sm_info = sm_info;
5475 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5476 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5477 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5478 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5479 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5480 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5481 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5482 sm_info->rec_prefree_segments = sm_info->main_segments *
5483 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5484 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5485 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5486
5487 if (!f2fs_lfs_mode(sbi))
5488 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5489 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5490 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5491 sm_info->min_seq_blocks = BLKS_PER_SEG(sbi);
5492 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5493 sm_info->min_ssr_sections = reserved_sections(sbi);
5494
5495 INIT_LIST_HEAD(&sm_info->sit_entry_set);
5496
5497 init_f2fs_rwsem(&sm_info->curseg_lock);
5498
5499 err = f2fs_create_flush_cmd_control(sbi);
5500 if (err)
5501 return err;
5502
5503 err = create_discard_cmd_control(sbi);
5504 if (err)
5505 return err;
5506
5507 err = build_sit_info(sbi);
5508 if (err)
5509 return err;
5510 err = build_free_segmap(sbi);
5511 if (err)
5512 return err;
5513 err = build_curseg(sbi);
5514 if (err)
5515 return err;
5516
5517 /* reinit free segmap based on SIT */
5518 err = build_sit_entries(sbi);
5519 if (err)
5520 return err;
5521
5522 init_free_segmap(sbi);
5523 err = build_dirty_segmap(sbi);
5524 if (err)
5525 return err;
5526
5527 err = sanity_check_curseg(sbi);
5528 if (err)
5529 return err;
5530
5531 init_min_max_mtime(sbi);
5532 return 0;
5533 }
5534
discard_dirty_segmap(struct f2fs_sb_info * sbi,enum dirty_type dirty_type)5535 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5536 enum dirty_type dirty_type)
5537 {
5538 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5539
5540 mutex_lock(&dirty_i->seglist_lock);
5541 kvfree(dirty_i->dirty_segmap[dirty_type]);
5542 dirty_i->nr_dirty[dirty_type] = 0;
5543 mutex_unlock(&dirty_i->seglist_lock);
5544 }
5545
destroy_victim_secmap(struct f2fs_sb_info * sbi)5546 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5547 {
5548 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5549
5550 kvfree(dirty_i->pinned_secmap);
5551 kvfree(dirty_i->victim_secmap);
5552 }
5553
destroy_dirty_segmap(struct f2fs_sb_info * sbi)5554 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5555 {
5556 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5557 int i;
5558
5559 if (!dirty_i)
5560 return;
5561
5562 /* discard pre-free/dirty segments list */
5563 for (i = 0; i < NR_DIRTY_TYPE; i++)
5564 discard_dirty_segmap(sbi, i);
5565
5566 if (__is_large_section(sbi)) {
5567 mutex_lock(&dirty_i->seglist_lock);
5568 kvfree(dirty_i->dirty_secmap);
5569 mutex_unlock(&dirty_i->seglist_lock);
5570 }
5571
5572 destroy_victim_secmap(sbi);
5573 SM_I(sbi)->dirty_info = NULL;
5574 kfree(dirty_i);
5575 }
5576
destroy_curseg(struct f2fs_sb_info * sbi)5577 static void destroy_curseg(struct f2fs_sb_info *sbi)
5578 {
5579 struct curseg_info *array = SM_I(sbi)->curseg_array;
5580 int i;
5581
5582 if (!array)
5583 return;
5584 SM_I(sbi)->curseg_array = NULL;
5585 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5586 kfree(array[i].sum_blk);
5587 kfree(array[i].journal);
5588 }
5589 kfree(array);
5590 }
5591
destroy_free_segmap(struct f2fs_sb_info * sbi)5592 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5593 {
5594 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5595
5596 if (!free_i)
5597 return;
5598 SM_I(sbi)->free_info = NULL;
5599 kvfree(free_i->free_segmap);
5600 kvfree(free_i->free_secmap);
5601 kfree(free_i);
5602 }
5603
destroy_sit_info(struct f2fs_sb_info * sbi)5604 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5605 {
5606 struct sit_info *sit_i = SIT_I(sbi);
5607
5608 if (!sit_i)
5609 return;
5610
5611 if (sit_i->sentries)
5612 kvfree(sit_i->bitmap);
5613 kfree(sit_i->tmp_map);
5614
5615 kvfree(sit_i->sentries);
5616 kvfree(sit_i->sec_entries);
5617 kvfree(sit_i->dirty_sentries_bitmap);
5618
5619 SM_I(sbi)->sit_info = NULL;
5620 kvfree(sit_i->sit_bitmap);
5621 #ifdef CONFIG_F2FS_CHECK_FS
5622 kvfree(sit_i->sit_bitmap_mir);
5623 kvfree(sit_i->invalid_segmap);
5624 #endif
5625 kfree(sit_i);
5626 }
5627
f2fs_destroy_segment_manager(struct f2fs_sb_info * sbi)5628 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5629 {
5630 struct f2fs_sm_info *sm_info = SM_I(sbi);
5631
5632 if (!sm_info)
5633 return;
5634 f2fs_destroy_flush_cmd_control(sbi, true);
5635 destroy_discard_cmd_control(sbi);
5636 destroy_dirty_segmap(sbi);
5637 destroy_curseg(sbi);
5638 destroy_free_segmap(sbi);
5639 destroy_sit_info(sbi);
5640 sbi->sm_info = NULL;
5641 kfree(sm_info);
5642 }
5643
f2fs_create_segment_manager_caches(void)5644 int __init f2fs_create_segment_manager_caches(void)
5645 {
5646 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5647 sizeof(struct discard_entry));
5648 if (!discard_entry_slab)
5649 goto fail;
5650
5651 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5652 sizeof(struct discard_cmd));
5653 if (!discard_cmd_slab)
5654 goto destroy_discard_entry;
5655
5656 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5657 sizeof(struct sit_entry_set));
5658 if (!sit_entry_set_slab)
5659 goto destroy_discard_cmd;
5660
5661 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5662 sizeof(struct revoke_entry));
5663 if (!revoke_entry_slab)
5664 goto destroy_sit_entry_set;
5665 return 0;
5666
5667 destroy_sit_entry_set:
5668 kmem_cache_destroy(sit_entry_set_slab);
5669 destroy_discard_cmd:
5670 kmem_cache_destroy(discard_cmd_slab);
5671 destroy_discard_entry:
5672 kmem_cache_destroy(discard_entry_slab);
5673 fail:
5674 return -ENOMEM;
5675 }
5676
f2fs_destroy_segment_manager_caches(void)5677 void f2fs_destroy_segment_manager_caches(void)
5678 {
5679 kmem_cache_destroy(sit_entry_set_slab);
5680 kmem_cache_destroy(discard_cmd_slab);
5681 kmem_cache_destroy(discard_entry_slab);
5682 kmem_cache_destroy(revoke_entry_slab);
5683 }
5684