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