]> Git Repo - linux.git/blame - fs/f2fs/segment.c
f2fs: wrap discard policy
[linux.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK
2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
1d7be270 19#include <linux/freezer.h>
1eb1ef4a 20#include <linux/sched/signal.h>
351df4b2
JK
21
22#include "f2fs.h"
23#include "segment.h"
24#include "node.h"
5f656541 25#include "gc.h"
9e4ded3f 26#include "trace.h"
6ec178da 27#include <trace/events/f2fs.h>
351df4b2 28
9a7f143a
CL
29#define __reverse_ffz(x) __reverse_ffs(~(x))
30
7fd9e544 31static struct kmem_cache *discard_entry_slab;
b01a9201 32static struct kmem_cache *discard_cmd_slab;
184a5cd2 33static struct kmem_cache *sit_entry_set_slab;
88b88a66 34static struct kmem_cache *inmem_entry_slab;
7fd9e544 35
f96999c3
JK
36static unsigned long __reverse_ulong(unsigned char *str)
37{
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41#if BITS_PER_LONG == 64
42 shift = 56;
43#endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49}
50
9a7f143a
CL
51/*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55static inline unsigned long __reverse_ffs(unsigned long word)
56{
57 int num = 0;
58
59#if BITS_PER_LONG == 64
f96999c3 60 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 61 num += 32;
f96999c3 62 else
9a7f143a 63 word >>= 32;
9a7f143a 64#endif
f96999c3 65 if ((word & 0xffff0000) == 0)
9a7f143a 66 num += 16;
f96999c3 67 else
9a7f143a 68 word >>= 16;
f96999c3
JK
69
70 if ((word & 0xff00) == 0)
9a7f143a 71 num += 8;
f96999c3 72 else
9a7f143a 73 word >>= 8;
f96999c3 74
9a7f143a
CL
75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
f96999c3 79
9a7f143a
CL
80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
f96999c3 84
9a7f143a
CL
85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88}
89
90/*
e1c42045 91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 93 * @size must be integral times of unsigned long.
9a7f143a 94 * Example:
f96999c3
JK
95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL
98 */
99static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101{
102 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 103 unsigned long result = size;
9a7f143a 104 unsigned long tmp;
9a7f143a
CL
105
106 if (offset >= size)
107 return size;
108
692223d1 109 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 110 offset %= BITS_PER_LONG;
f96999c3 111
692223d1
FL
112 while (1) {
113 if (*p == 0)
114 goto pass;
9a7f143a 115
f96999c3 116 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL
117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 121 if (tmp)
692223d1
FL
122 goto found;
123pass:
124 if (size <= BITS_PER_LONG)
125 break;
9a7f143a 126 size -= BITS_PER_LONG;
692223d1 127 offset = 0;
f96999c3 128 p++;
9a7f143a 129 }
692223d1
FL
130 return result;
131found:
132 return result - size + __reverse_ffs(tmp);
9a7f143a
CL
133}
134
135static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137{
138 const unsigned long *p = addr + BIT_WORD(offset);
80609448 139 unsigned long result = size;
9a7f143a 140 unsigned long tmp;
9a7f143a
CL
141
142 if (offset >= size)
143 return size;
144
80609448 145 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 146 offset %= BITS_PER_LONG;
80609448
JK
147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
f96999c3 152 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK
153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
f96999c3 158 if (tmp != ~0UL)
80609448
JK
159 goto found;
160pass:
161 if (size <= BITS_PER_LONG)
162 break;
9a7f143a 163 size -= BITS_PER_LONG;
80609448 164 offset = 0;
f96999c3 165 p++;
9a7f143a 166 }
80609448
JK
167 return result;
168found:
169 return result - size + __reverse_ffz(tmp);
9a7f143a
CL
170}
171
b3a97a2a
JK
172bool need_SSR(struct f2fs_sb_info *sbi)
173{
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177
178 if (test_opt(sbi, LFS))
179 return false;
180 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
184 2 * reserved_sections(sbi));
185}
186
88b88a66
JK
187void register_inmem_page(struct inode *inode, struct page *page)
188{
189 struct f2fs_inode_info *fi = F2FS_I(inode);
190 struct inmem_pages *new;
9be32d72 191
9e4ded3f 192 f2fs_trace_pid(page);
0722b101 193
decd36b6
CY
194 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
195 SetPagePrivate(page);
196
88b88a66
JK
197 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
198
199 /* add atomic page indices to the list */
200 new->page = page;
201 INIT_LIST_HEAD(&new->list);
decd36b6 202
88b88a66
JK
203 /* increase reference count with clean state */
204 mutex_lock(&fi->inmem_lock);
205 get_page(page);
206 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 207 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 208 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK
209
210 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK
211}
212
28bc106b
CY
213static int __revoke_inmem_pages(struct inode *inode,
214 struct list_head *head, bool drop, bool recover)
29b96b54 215{
28bc106b 216 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 217 struct inmem_pages *cur, *tmp;
28bc106b 218 int err = 0;
29b96b54
CY
219
220 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY
221 struct page *page = cur->page;
222
223 if (drop)
224 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
225
226 lock_page(page);
29b96b54 227
28bc106b
CY
228 if (recover) {
229 struct dnode_of_data dn;
230 struct node_info ni;
231
232 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 233retry:
28bc106b 234 set_new_dnode(&dn, inode, NULL, NULL, 0);
7f2b4e8e
CY
235 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
236 if (err) {
237 if (err == -ENOMEM) {
238 congestion_wait(BLK_RW_ASYNC, HZ/50);
239 cond_resched();
240 goto retry;
241 }
28bc106b
CY
242 err = -EAGAIN;
243 goto next;
244 }
245 get_node_info(sbi, dn.nid, &ni);
246 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
247 cur->old_addr, ni.version, true, true);
248 f2fs_put_dnode(&dn);
249 }
250next:
63c52d78
JK
251 /* we don't need to invalidate this in the sccessful status */
252 if (drop || recover)
253 ClearPageUptodate(page);
28bc106b 254 set_page_private(page, 0);
c81ced05 255 ClearPagePrivate(page);
28bc106b 256 f2fs_put_page(page, 1);
29b96b54
CY
257
258 list_del(&cur->list);
259 kmem_cache_free(inmem_entry_slab, cur);
260 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
261 }
28bc106b 262 return err;
29b96b54
CY
263}
264
265void drop_inmem_pages(struct inode *inode)
266{
267 struct f2fs_inode_info *fi = F2FS_I(inode);
268
269 mutex_lock(&fi->inmem_lock);
28bc106b 270 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54 271 mutex_unlock(&fi->inmem_lock);
5fe45743
CY
272
273 clear_inode_flag(inode, FI_ATOMIC_FILE);
84a23fbe 274 clear_inode_flag(inode, FI_HOT_DATA);
5fe45743 275 stat_dec_atomic_write(inode);
29b96b54
CY
276}
277
8c242db9
JK
278void drop_inmem_page(struct inode *inode, struct page *page)
279{
280 struct f2fs_inode_info *fi = F2FS_I(inode);
281 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
282 struct list_head *head = &fi->inmem_pages;
283 struct inmem_pages *cur = NULL;
284
285 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
286
287 mutex_lock(&fi->inmem_lock);
288 list_for_each_entry(cur, head, list) {
289 if (cur->page == page)
290 break;
291 }
292
293 f2fs_bug_on(sbi, !cur || cur->page != page);
294 list_del(&cur->list);
295 mutex_unlock(&fi->inmem_lock);
296
297 dec_page_count(sbi, F2FS_INMEM_PAGES);
298 kmem_cache_free(inmem_entry_slab, cur);
299
300 ClearPageUptodate(page);
301 set_page_private(page, 0);
302 ClearPagePrivate(page);
303 f2fs_put_page(page, 0);
304
305 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
306}
307
28bc106b
CY
308static int __commit_inmem_pages(struct inode *inode,
309 struct list_head *revoke_list)
88b88a66
JK
310{
311 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
312 struct f2fs_inode_info *fi = F2FS_I(inode);
313 struct inmem_pages *cur, *tmp;
88b88a66 314 struct f2fs_io_info fio = {
05ca3632 315 .sbi = sbi,
39d787be 316 .ino = inode->i_ino,
88b88a66 317 .type = DATA,
04d328de 318 .op = REQ_OP_WRITE,
70fd7614 319 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 320 .io_type = FS_DATA_IO,
88b88a66 321 };
942fd319 322 pgoff_t last_idx = ULONG_MAX;
edb27dee 323 int err = 0;
88b88a66 324
88b88a66 325 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY
326 struct page *page = cur->page;
327
328 lock_page(page);
329 if (page->mapping == inode->i_mapping) {
330 trace_f2fs_commit_inmem_page(page, INMEM);
331
332 set_page_dirty(page);
333 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 334 if (clear_page_dirty_for_io(page)) {
29b96b54 335 inode_dec_dirty_pages(inode);
933439c8
CY
336 remove_dirty_inode(inode);
337 }
640cc189 338retry:
28bc106b 339 fio.page = page;
e959c8f5 340 fio.old_blkaddr = NULL_ADDR;
4d978078 341 fio.encrypted_page = NULL;
cc15620b 342 fio.need_lock = LOCK_DONE;
29b96b54
CY
343 err = do_write_data_page(&fio);
344 if (err) {
640cc189
JK
345 if (err == -ENOMEM) {
346 congestion_wait(BLK_RW_ASYNC, HZ/50);
347 cond_resched();
348 goto retry;
349 }
28bc106b 350 unlock_page(page);
29b96b54 351 break;
70c640b1 352 }
28bc106b
CY
353 /* record old blkaddr for revoking */
354 cur->old_addr = fio.old_blkaddr;
942fd319 355 last_idx = page->index;
28bc106b
CY
356 }
357 unlock_page(page);
358 list_move_tail(&cur->list, revoke_list);
88b88a66 359 }
29b96b54 360
942fd319 361 if (last_idx != ULONG_MAX)
b9109b0e 362 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b
CY
363
364 if (!err)
365 __revoke_inmem_pages(inode, revoke_list, false, false);
366
29b96b54
CY
367 return err;
368}
369
370int commit_inmem_pages(struct inode *inode)
371{
372 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
373 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY
374 struct list_head revoke_list;
375 int err;
29b96b54 376
28bc106b 377 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY
378 f2fs_balance_fs(sbi, true);
379 f2fs_lock_op(sbi);
380
5fe45743
CY
381 set_inode_flag(inode, FI_ATOMIC_COMMIT);
382
29b96b54 383 mutex_lock(&fi->inmem_lock);
28bc106b
CY
384 err = __commit_inmem_pages(inode, &revoke_list);
385 if (err) {
386 int ret;
387 /*
388 * try to revoke all committed pages, but still we could fail
389 * due to no memory or other reason, if that happened, EAGAIN
390 * will be returned, which means in such case, transaction is
391 * already not integrity, caller should use journal to do the
392 * recovery or rewrite & commit last transaction. For other
393 * error number, revoking was done by filesystem itself.
394 */
395 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
396 if (ret)
397 err = ret;
398
399 /* drop all uncommitted pages */
400 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
401 }
88b88a66
JK
402 mutex_unlock(&fi->inmem_lock);
403
5fe45743
CY
404 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
405
29b96b54 406 f2fs_unlock_op(sbi);
edb27dee 407 return err;
88b88a66
JK
408}
409
0a8165d7 410/*
351df4b2
JK
411 * This function balances dirty node and dentry pages.
412 * In addition, it controls garbage collection.
413 */
2c4db1a6 414void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 415{
0f348028 416#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY
417 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
418 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 419 f2fs_stop_checkpoint(sbi, false);
55523519 420 }
0f348028
CY
421#endif
422
e589c2c4 423 /* balance_fs_bg is able to be pending */
a7881893 424 if (need && excess_cached_nats(sbi))
e589c2c4
JK
425 f2fs_balance_fs_bg(sbi);
426
351df4b2 427 /*
029cd28c
JK
428 * We should do GC or end up with checkpoint, if there are so many dirty
429 * dir/node pages without enough free segments.
351df4b2 430 */
7f3037a5 431 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 432 mutex_lock(&sbi->gc_mutex);
e066b83c 433 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK
434 }
435}
436
4660f9c0
JK
437void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
438{
1dcc336b 439 /* try to shrink extent cache when there is no enough memory */
554df79e
JK
440 if (!available_free_memory(sbi, EXTENT_CACHE))
441 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 442
1b38dc8e
JK
443 /* check the # of cached NAT entries */
444 if (!available_free_memory(sbi, NAT_ENTRIES))
445 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
446
31696580 447 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK
448 try_to_free_nids(sbi, MAX_FREE_NIDS);
449 else
22ad0b6a 450 build_free_nids(sbi, false, false);
31696580 451
1c0f4bf5 452 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 453 return;
31696580 454
1b38dc8e
JK
455 /* checkpoint is the only way to shrink partial cached entries */
456 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 457 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY
458 excess_prefree_segs(sbi) ||
459 excess_dirty_nats(sbi) ||
f455c8a5 460 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY
461 if (test_opt(sbi, DATA_FLUSH)) {
462 struct blk_plug plug;
463
464 blk_start_plug(&plug);
36b35a0d 465 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY
466 blk_finish_plug(&plug);
467 }
4660f9c0 468 f2fs_sync_fs(sbi->sb, true);
42190d2a 469 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 470 }
4660f9c0
JK
471}
472
20fda56b
KM
473static int __submit_flush_wait(struct f2fs_sb_info *sbi,
474 struct block_device *bdev)
3c62be17
JK
475{
476 struct bio *bio = f2fs_bio_alloc(0);
477 int ret;
478
3adc5fcb 479 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 480 bio_set_dev(bio, bdev);
3c62be17
JK
481 ret = submit_bio_wait(bio);
482 bio_put(bio);
20fda56b
KM
483
484 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
485 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK
486 return ret;
487}
488
39d787be 489static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 490{
39d787be 491 int ret = 0;
3c62be17
JK
492 int i;
493
39d787be
CY
494 if (!sbi->s_ndevs)
495 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 496
39d787be
CY
497 for (i = 0; i < sbi->s_ndevs; i++) {
498 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
499 continue;
20fda56b
KM
500 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
501 if (ret)
502 break;
3c62be17
JK
503 }
504 return ret;
505}
506
2163d198 507static int issue_flush_thread(void *data)
6b4afdd7
JK
508{
509 struct f2fs_sb_info *sbi = data;
b01a9201 510 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 511 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK
512repeat:
513 if (kthread_should_stop())
514 return 0;
515
dc6febb6
CY
516 sb_start_intwrite(sbi->sb);
517
721bd4d5 518 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK
519 struct flush_cmd *cmd, *next;
520 int ret;
521
721bd4d5
GZ
522 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
523 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
524
39d787be
CY
525 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
526
527 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY
528 atomic_inc(&fcc->issued_flush);
529
721bd4d5
GZ
530 llist_for_each_entry_safe(cmd, next,
531 fcc->dispatch_list, llnode) {
6b4afdd7 532 cmd->ret = ret;
6b4afdd7
JK
533 complete(&cmd->wait);
534 }
a688b9d9 535 fcc->dispatch_list = NULL;
6b4afdd7
JK
536 }
537
dc6febb6
CY
538 sb_end_intwrite(sbi->sb);
539
a688b9d9 540 wait_event_interruptible(*q,
721bd4d5 541 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK
542 goto repeat;
543}
544
39d787be 545int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 546{
b01a9201 547 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 548 struct flush_cmd cmd;
8b8dd65f 549 int ret;
6b4afdd7 550
0f7b2abd
JK
551 if (test_opt(sbi, NOBARRIER))
552 return 0;
553
8b8dd65f 554 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 555 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY
556 atomic_inc(&fcc->issued_flush);
557 return ret;
558 }
740432f8 559
39d787be
CY
560 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
561 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY
562 atomic_dec(&fcc->issing_flush);
563
564 atomic_inc(&fcc->issued_flush);
740432f8
JK
565 return ret;
566 }
6b4afdd7 567
39d787be 568 cmd.ino = ino;
adf8d90b 569 init_completion(&cmd.wait);
6b4afdd7 570
721bd4d5 571 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 572
6f890df0
CY
573 /* update issue_list before we wake up issue_flush thread */
574 smp_mb();
575
576 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 577 wake_up(&fcc->flush_wait_queue);
6b4afdd7 578
5eba8c5d
JK
579 if (fcc->f2fs_issue_flush) {
580 wait_for_completion(&cmd.wait);
8b8dd65f 581 atomic_dec(&fcc->issing_flush);
5eba8c5d 582 } else {
d3238691
CY
583 struct llist_node *list;
584
585 list = llist_del_all(&fcc->issue_list);
586 if (!list) {
587 wait_for_completion(&cmd.wait);
588 atomic_dec(&fcc->issing_flush);
589 } else {
590 struct flush_cmd *tmp, *next;
591
39d787be 592 ret = submit_flush_wait(sbi, ino);
d3238691
CY
593
594 llist_for_each_entry_safe(tmp, next, list, llnode) {
595 if (tmp == &cmd) {
596 cmd.ret = ret;
597 atomic_dec(&fcc->issing_flush);
598 continue;
599 }
600 tmp->ret = ret;
601 complete(&tmp->wait);
602 }
603 }
5eba8c5d 604 }
adf8d90b
CY
605
606 return cmd.ret;
6b4afdd7
JK
607}
608
2163d198
GZ
609int create_flush_cmd_control(struct f2fs_sb_info *sbi)
610{
611 dev_t dev = sbi->sb->s_bdev->bd_dev;
612 struct flush_cmd_control *fcc;
613 int err = 0;
614
b01a9201
JK
615 if (SM_I(sbi)->fcc_info) {
616 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS
617 if (fcc->f2fs_issue_flush)
618 return err;
5eba8c5d
JK
619 goto init_thread;
620 }
621
2163d198
GZ
622 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
623 if (!fcc)
624 return -ENOMEM;
8b8dd65f
CY
625 atomic_set(&fcc->issued_flush, 0);
626 atomic_set(&fcc->issing_flush, 0);
2163d198 627 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 628 init_llist_head(&fcc->issue_list);
b01a9201 629 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH
630 if (!test_opt(sbi, FLUSH_MERGE))
631 return err;
632
5eba8c5d 633init_thread:
2163d198
GZ
634 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
635 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
636 if (IS_ERR(fcc->f2fs_issue_flush)) {
637 err = PTR_ERR(fcc->f2fs_issue_flush);
638 kfree(fcc);
b01a9201 639 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ
640 return err;
641 }
2163d198
GZ
642
643 return err;
644}
645
5eba8c5d 646void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 647{
b01a9201 648 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 649
5eba8c5d
JK
650 if (fcc && fcc->f2fs_issue_flush) {
651 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
652
653 fcc->f2fs_issue_flush = NULL;
654 kthread_stop(flush_thread);
655 }
656 if (free) {
657 kfree(fcc);
b01a9201 658 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 659 }
2163d198
GZ
660}
661
1228b482
CY
662int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
663{
664 int ret = 0, i;
665
666 if (!sbi->s_ndevs)
667 return 0;
668
669 for (i = 1; i < sbi->s_ndevs; i++) {
670 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
671 continue;
672 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
673 if (ret)
674 break;
675
676 spin_lock(&sbi->dev_lock);
677 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
678 spin_unlock(&sbi->dev_lock);
679 }
680
681 return ret;
682}
683
351df4b2
JK
684static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
685 enum dirty_type dirty_type)
686{
687 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
688
689 /* need not be added */
690 if (IS_CURSEG(sbi, segno))
691 return;
692
693 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
694 dirty_i->nr_dirty[dirty_type]++;
695
696 if (dirty_type == DIRTY) {
697 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 698 enum dirty_type t = sentry->type;
b2f2c390 699
ec325b52
JK
700 if (unlikely(t >= DIRTY)) {
701 f2fs_bug_on(sbi, 1);
702 return;
703 }
4625d6aa
CL
704 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
705 dirty_i->nr_dirty[t]++;
351df4b2
JK
706 }
707}
708
709static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
710 enum dirty_type dirty_type)
711{
712 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
713
714 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
715 dirty_i->nr_dirty[dirty_type]--;
716
717 if (dirty_type == DIRTY) {
4625d6aa
CL
718 struct seg_entry *sentry = get_seg_entry(sbi, segno);
719 enum dirty_type t = sentry->type;
720
721 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
722 dirty_i->nr_dirty[t]--;
b2f2c390 723
302bd348 724 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 725 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 726 dirty_i->victim_secmap);
351df4b2
JK
727 }
728}
729
0a8165d7 730/*
351df4b2
JK
731 * Should not occur error such as -ENOMEM.
732 * Adding dirty entry into seglist is not critical operation.
733 * If a given segment is one of current working segments, it won't be added.
734 */
8d8451af 735static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK
736{
737 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
738 unsigned short valid_blocks;
739
740 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
741 return;
742
743 mutex_lock(&dirty_i->seglist_lock);
744
302bd348 745 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK
746
747 if (valid_blocks == 0) {
748 __locate_dirty_segment(sbi, segno, PRE);
749 __remove_dirty_segment(sbi, segno, DIRTY);
750 } else if (valid_blocks < sbi->blocks_per_seg) {
751 __locate_dirty_segment(sbi, segno, DIRTY);
752 } else {
753 /* Recovery routine with SSR needs this */
754 __remove_dirty_segment(sbi, segno, DIRTY);
755 }
756
757 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK
758}
759
004b6862 760static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK
761 struct block_device *bdev, block_t lstart,
762 block_t start, block_t len)
275b66b0 763{
0b54fb84 764 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 765 struct list_head *pend_list;
b01a9201 766 struct discard_cmd *dc;
275b66b0 767
ba48a33e
CY
768 f2fs_bug_on(sbi, !len);
769
770 pend_list = &dcc->pend_list[plist_idx(len)];
771
b01a9201
JK
772 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
773 INIT_LIST_HEAD(&dc->list);
c81abe34 774 dc->bdev = bdev;
b01a9201 775 dc->lstart = lstart;
c81abe34 776 dc->start = start;
b01a9201 777 dc->len = len;
ec9895ad 778 dc->ref = 0;
15469963 779 dc->state = D_PREP;
c81abe34 780 dc->error = 0;
b01a9201 781 init_completion(&dc->wait);
22d375dd 782 list_add_tail(&dc->list, pend_list);
5f32366a 783 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 784 dcc->undiscard_blks += len;
004b6862
CY
785
786 return dc;
787}
788
789static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
790 struct block_device *bdev, block_t lstart,
791 block_t start, block_t len,
792 struct rb_node *parent, struct rb_node **p)
793{
794 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
795 struct discard_cmd *dc;
796
797 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
798
799 rb_link_node(&dc->rb_node, parent, p);
800 rb_insert_color(&dc->rb_node, &dcc->root);
801
802 return dc;
15469963
JK
803}
804
004b6862
CY
805static void __detach_discard_cmd(struct discard_cmd_control *dcc,
806 struct discard_cmd *dc)
15469963 807{
dcc9165d 808 if (dc->state == D_DONE)
004b6862
CY
809 atomic_dec(&dcc->issing_discard);
810
811 list_del(&dc->list);
812 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 813 dcc->undiscard_blks -= dc->len;
004b6862
CY
814
815 kmem_cache_free(discard_cmd_slab, dc);
816
817 atomic_dec(&dcc->discard_cmd_cnt);
818}
819
820static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
821 struct discard_cmd *dc)
822{
823 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 824
d9703d90
CY
825 f2fs_bug_on(sbi, dc->ref);
826
c81abe34
JK
827 if (dc->error == -EOPNOTSUPP)
828 dc->error = 0;
15469963 829
c81abe34 830 if (dc->error)
15469963 831 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY
832 "Issue discard(%u, %u, %u) failed, ret: %d",
833 dc->lstart, dc->start, dc->len, dc->error);
004b6862 834 __detach_discard_cmd(dcc, dc);
275b66b0
CY
835}
836
c81abe34
JK
837static void f2fs_submit_discard_endio(struct bio *bio)
838{
839 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
840
4e4cbee9 841 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 842 dc->state = D_DONE;
e31b9821 843 complete_all(&dc->wait);
c81abe34
JK
844 bio_put(bio);
845}
846
6915ea9d
CY
847void __check_sit_bitmap(struct f2fs_sb_info *sbi,
848 block_t start, block_t end)
849{
850#ifdef CONFIG_F2FS_CHECK_FS
851 struct seg_entry *sentry;
852 unsigned int segno;
853 block_t blk = start;
854 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
855 unsigned long *map;
856
857 while (blk < end) {
858 segno = GET_SEGNO(sbi, blk);
859 sentry = get_seg_entry(sbi, segno);
860 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
861
008396e1
YS
862 if (end < START_BLOCK(sbi, segno + 1))
863 size = GET_BLKOFF_FROM_SEG0(sbi, end);
864 else
865 size = max_blocks;
6915ea9d
CY
866 map = (unsigned long *)(sentry->cur_valid_map);
867 offset = __find_rev_next_bit(map, size, offset);
868 f2fs_bug_on(sbi, offset != size);
008396e1 869 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY
870 }
871#endif
872}
873
c81abe34
JK
874/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
875static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
8412663d 876 struct discard_cmd *dc, bool fstrim)
c81abe34
JK
877{
878 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
8412663d
CY
879 struct list_head *wait_list = fstrim ? &(dcc->fstrim_list) :
880 &(dcc->wait_list);
c81abe34 881 struct bio *bio = NULL;
ecc9aa00 882 int flag = dcc->dpolicy.sync ? REQ_SYNC : 0;
c81abe34
JK
883
884 if (dc->state != D_PREP)
885 return;
886
0243a5f9
CY
887 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
888
c81abe34
JK
889 dc->error = __blkdev_issue_discard(dc->bdev,
890 SECTOR_FROM_BLOCK(dc->start),
891 SECTOR_FROM_BLOCK(dc->len),
892 GFP_NOFS, 0, &bio);
893 if (!dc->error) {
894 /* should keep before submission to avoid D_DONE right away */
895 dc->state = D_SUBMIT;
8b8dd65f
CY
896 atomic_inc(&dcc->issued_discard);
897 atomic_inc(&dcc->issing_discard);
c81abe34
JK
898 if (bio) {
899 bio->bi_private = dc;
900 bio->bi_end_io = f2fs_submit_discard_endio;
ecc9aa00 901 bio->bi_opf |= flag;
c81abe34 902 submit_bio(bio);
8412663d 903 list_move_tail(&dc->list, wait_list);
6915ea9d 904 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY
905
906 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK
907 }
908 } else {
909 __remove_discard_cmd(sbi, dc);
910 }
911}
912
004b6862
CY
913static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
914 struct block_device *bdev, block_t lstart,
915 block_t start, block_t len,
916 struct rb_node **insert_p,
917 struct rb_node *insert_parent)
c81abe34 918{
004b6862
CY
919 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
920 struct rb_node **p = &dcc->root.rb_node;
921 struct rb_node *parent = NULL;
922 struct discard_cmd *dc = NULL;
923
924 if (insert_p && insert_parent) {
925 parent = insert_parent;
926 p = insert_p;
927 goto do_insert;
928 }
c81abe34 929
004b6862
CY
930 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
931do_insert:
932 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
933 if (!dc)
934 return NULL;
c81abe34 935
004b6862 936 return dc;
c81abe34
JK
937}
938
ba48a33e
CY
939static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
940 struct discard_cmd *dc)
941{
942 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
943}
944
3d6a650f
YH
945static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
946 struct discard_cmd *dc, block_t blkaddr)
947{
ba48a33e 948 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY
949 struct discard_info di = dc->di;
950 bool modified = false;
3d6a650f 951
004b6862 952 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH
953 __remove_discard_cmd(sbi, dc);
954 return;
955 }
956
d84d1cbd
CY
957 dcc->undiscard_blks -= di.len;
958
004b6862 959 if (blkaddr > di.lstart) {
3d6a650f 960 dc->len = blkaddr - dc->lstart;
d84d1cbd 961 dcc->undiscard_blks += dc->len;
ba48a33e 962 __relocate_discard_cmd(dcc, dc);
004b6862
CY
963 modified = true;
964 }
965
966 if (blkaddr < di.lstart + di.len - 1) {
967 if (modified) {
968 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
969 di.start + blkaddr + 1 - di.lstart,
970 di.lstart + di.len - 1 - blkaddr,
971 NULL, NULL);
972 } else {
973 dc->lstart++;
974 dc->len--;
975 dc->start++;
d84d1cbd 976 dcc->undiscard_blks += dc->len;
ba48a33e 977 __relocate_discard_cmd(dcc, dc);
004b6862 978 }
3d6a650f
YH
979 }
980}
981
004b6862
CY
982static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
983 struct block_device *bdev, block_t lstart,
984 block_t start, block_t len)
275b66b0 985{
0b54fb84 986 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY
987 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
988 struct discard_cmd *dc;
989 struct discard_info di = {0};
990 struct rb_node **insert_p = NULL, *insert_parent = NULL;
991 block_t end = lstart + len;
275b66b0 992
15469963 993 mutex_lock(&dcc->cmd_lock);
40465257 994
004b6862
CY
995 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
996 NULL, lstart,
997 (struct rb_entry **)&prev_dc,
998 (struct rb_entry **)&next_dc,
999 &insert_p, &insert_parent, true);
1000 if (dc)
1001 prev_dc = dc;
1002
1003 if (!prev_dc) {
1004 di.lstart = lstart;
1005 di.len = next_dc ? next_dc->lstart - lstart : len;
1006 di.len = min(di.len, len);
1007 di.start = start;
22d375dd 1008 }
15469963 1009
004b6862
CY
1010 while (1) {
1011 struct rb_node *node;
1012 bool merged = false;
1013 struct discard_cmd *tdc = NULL;
1014
1015 if (prev_dc) {
1016 di.lstart = prev_dc->lstart + prev_dc->len;
1017 if (di.lstart < lstart)
1018 di.lstart = lstart;
1019 if (di.lstart >= end)
1020 break;
1021
1022 if (!next_dc || next_dc->lstart > end)
1023 di.len = end - di.lstart;
1024 else
1025 di.len = next_dc->lstart - di.lstart;
1026 di.start = start + di.lstart - lstart;
1027 }
1028
1029 if (!di.len)
1030 goto next;
1031
1032 if (prev_dc && prev_dc->state == D_PREP &&
1033 prev_dc->bdev == bdev &&
1034 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1035 prev_dc->di.len += di.len;
d84d1cbd 1036 dcc->undiscard_blks += di.len;
ba48a33e 1037 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY
1038 di = prev_dc->di;
1039 tdc = prev_dc;
1040 merged = true;
1041 }
1042
1043 if (next_dc && next_dc->state == D_PREP &&
1044 next_dc->bdev == bdev &&
1045 __is_discard_front_mergeable(&di, &next_dc->di)) {
1046 next_dc->di.lstart = di.lstart;
1047 next_dc->di.len += di.len;
1048 next_dc->di.start = di.start;
d84d1cbd 1049 dcc->undiscard_blks += di.len;
ba48a33e 1050 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY
1051 if (tdc)
1052 __remove_discard_cmd(sbi, tdc);
004b6862 1053 merged = true;
4e6a8d9b 1054 }
004b6862 1055
df0f6b44 1056 if (!merged) {
004b6862
CY
1057 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1058 di.len, NULL, NULL);
df0f6b44 1059 }
004b6862
CY
1060 next:
1061 prev_dc = next_dc;
1062 if (!prev_dc)
1063 break;
1064
1065 node = rb_next(&prev_dc->rb_node);
1066 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1067 }
1068
1069 mutex_unlock(&dcc->cmd_lock);
1070}
1071
1072static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1073 struct block_device *bdev, block_t blkstart, block_t blklen)
1074{
1075 block_t lblkstart = blkstart;
1076
0243a5f9 1077 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY
1078
1079 if (sbi->s_ndevs) {
1080 int devi = f2fs_target_device_index(sbi, blkstart);
1081
1082 blkstart -= FDEV(devi).start_blk;
1083 }
1084 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY
1085 return 0;
1086}
1087
8412663d
CY
1088static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
1089 unsigned int start, unsigned int end,
1090 unsigned int granularity)
1091{
1092 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1093 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1094 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1095 struct discard_cmd *dc;
ecc9aa00 1096 struct discard_policy *dpolicy = &dcc->dpolicy;
8412663d
CY
1097 struct blk_plug plug;
1098 int issued;
1099
1100next:
1101 issued = 0;
1102
1103 mutex_lock(&dcc->cmd_lock);
1104 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1105
1106 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1107 NULL, start,
1108 (struct rb_entry **)&prev_dc,
1109 (struct rb_entry **)&next_dc,
1110 &insert_p, &insert_parent, true);
1111 if (!dc)
1112 dc = next_dc;
1113
1114 blk_start_plug(&plug);
1115
1116 while (dc && dc->lstart <= end) {
1117 struct rb_node *node;
1118
1119 if (dc->len < granularity)
1120 goto skip;
1121
1122 if (dc->state != D_PREP) {
1123 list_move_tail(&dc->list, &dcc->fstrim_list);
1124 goto skip;
1125 }
1126
1127 __submit_discard_cmd(sbi, dc, true);
1128
ecc9aa00 1129 if (++issued >= dpolicy->max_requests) {
8412663d
CY
1130 start = dc->lstart + dc->len;
1131
1132 blk_finish_plug(&plug);
1133 mutex_unlock(&dcc->cmd_lock);
1134
1135 schedule();
1136
1137 goto next;
1138 }
1139skip:
1140 node = rb_next(&dc->rb_node);
1141 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1142
1143 if (fatal_signal_pending(current))
1144 break;
1145 }
1146
1147 blk_finish_plug(&plug);
1148 mutex_unlock(&dcc->cmd_lock);
1149}
1150
969d1b18 1151static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
bd5b0738
CY
1152{
1153 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1154 struct list_head *pend_list;
1155 struct discard_cmd *dc, *tmp;
1156 struct blk_plug plug;
ecc9aa00 1157 struct discard_policy *dpolicy = &dcc->dpolicy;
969d1b18
CY
1158 int iter = 0, issued = 0;
1159 int i;
e6c6de18 1160 bool io_interrupted = false;
bd5b0738
CY
1161
1162 mutex_lock(&dcc->cmd_lock);
963932a9
YH
1163 f2fs_bug_on(sbi,
1164 !__check_rb_tree_consistence(sbi, &dcc->root));
bd5b0738 1165 blk_start_plug(&plug);
969d1b18
CY
1166 for (i = MAX_PLIST_NUM - 1;
1167 i >= 0 && plist_issue(dcc->pend_list_tag[i]); i--) {
bd5b0738
CY
1168 pend_list = &dcc->pend_list[i];
1169 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1170 f2fs_bug_on(sbi, dc->state != D_PREP);
1171
969d1b18
CY
1172 /* Hurry up to finish fstrim */
1173 if (dcc->pend_list_tag[i] & P_TRIM) {
8412663d 1174 __submit_discard_cmd(sbi, dc, false);
969d1b18
CY
1175 issued++;
1176 continue;
1177 }
1178
e6c6de18 1179 if (!issue_cond) {
8412663d 1180 __submit_discard_cmd(sbi, dc, false);
969d1b18 1181 issued++;
e6c6de18
CY
1182 continue;
1183 }
1184
ecc9aa00
CY
1185 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1186 !is_idle(sbi)) {
e6c6de18 1187 io_interrupted = true;
ecc9aa00 1188 goto skip;
969d1b18 1189 }
e6c6de18 1190
ecc9aa00
CY
1191 __submit_discard_cmd(sbi, dc, false);
1192 issued++;
1193skip:
1194 if (++iter >= dpolicy->max_requests)
bd5b0738
CY
1195 goto out;
1196 }
969d1b18
CY
1197 if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM)
1198 dcc->pend_list_tag[i] &= (~P_TRIM);
bd5b0738
CY
1199 }
1200out:
1201 blk_finish_plug(&plug);
1202 mutex_unlock(&dcc->cmd_lock);
969d1b18 1203
e6c6de18
CY
1204 if (!issued && io_interrupted)
1205 issued = -1;
1206
969d1b18
CY
1207 return issued;
1208}
1209
1210static void __drop_discard_cmd(struct f2fs_sb_info *sbi)
1211{
1212 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1213 struct list_head *pend_list;
1214 struct discard_cmd *dc, *tmp;
1215 int i;
1216
1217 mutex_lock(&dcc->cmd_lock);
1218 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1219 pend_list = &dcc->pend_list[i];
1220 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1221 f2fs_bug_on(sbi, dc->state != D_PREP);
1222 __remove_discard_cmd(sbi, dc);
1223 }
1224 }
1225 mutex_unlock(&dcc->cmd_lock);
bd5b0738
CY
1226}
1227
2a510c00
CY
1228static void __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1229 struct discard_cmd *dc)
1230{
1231 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1232
1233 wait_for_completion_io(&dc->wait);
1234 mutex_lock(&dcc->cmd_lock);
1235 f2fs_bug_on(sbi, dc->state != D_DONE);
1236 dc->ref--;
1237 if (!dc->ref)
1238 __remove_discard_cmd(sbi, dc);
1239 mutex_unlock(&dcc->cmd_lock);
1240}
1241
8412663d
CY
1242static void __wait_discard_cmd_range(struct f2fs_sb_info *sbi, bool wait_cond,
1243 block_t start, block_t end,
1244 unsigned int granularity,
1245 bool fstrim)
63a94fa1
CY
1246{
1247 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
8412663d
CY
1248 struct list_head *wait_list = fstrim ? &(dcc->fstrim_list) :
1249 &(dcc->wait_list);
63a94fa1 1250 struct discard_cmd *dc, *tmp;
6afae633
CY
1251 bool need_wait;
1252
1253next:
1254 need_wait = false;
63a94fa1
CY
1255
1256 mutex_lock(&dcc->cmd_lock);
1257 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY
1258 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1259 continue;
1260 if (dc->len < granularity)
1261 continue;
6afae633 1262 if (!wait_cond || (dc->state == D_DONE && !dc->ref)) {
63a94fa1
CY
1263 wait_for_completion_io(&dc->wait);
1264 __remove_discard_cmd(sbi, dc);
6afae633
CY
1265 } else {
1266 dc->ref++;
1267 need_wait = true;
1268 break;
63a94fa1
CY
1269 }
1270 }
1271 mutex_unlock(&dcc->cmd_lock);
6afae633
CY
1272
1273 if (need_wait) {
2a510c00 1274 __wait_one_discard_bio(sbi, dc);
6afae633
CY
1275 goto next;
1276 }
63a94fa1
CY
1277}
1278
8412663d
CY
1279static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1280{
1281 __wait_discard_cmd_range(sbi, wait_cond, 0, UINT_MAX, 1, false);
1282}
1283
004b6862
CY
1284/* This should be covered by global mutex, &sit_i->sentry_lock */
1285void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1286{
1287 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1288 struct discard_cmd *dc;
ec9895ad 1289 bool need_wait = false;
004b6862
CY
1290
1291 mutex_lock(&dcc->cmd_lock);
004b6862
CY
1292 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1293 if (dc) {
ec9895ad
CY
1294 if (dc->state == D_PREP) {
1295 __punch_discard_cmd(sbi, dc, blkaddr);
1296 } else {
1297 dc->ref++;
1298 need_wait = true;
1299 }
275b66b0 1300 }
d431413f 1301 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1302
2a510c00
CY
1303 if (need_wait)
1304 __wait_one_discard_bio(sbi, dc);
d431413f
CY
1305}
1306
cce13252
CY
1307void stop_discard_thread(struct f2fs_sb_info *sbi)
1308{
1309 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1310
1311 if (dcc && dcc->f2fs_issue_discard) {
1312 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1313
1314 dcc->f2fs_issue_discard = NULL;
1315 kthread_stop(discard_thread);
ec9895ad 1316 }
d431413f
CY
1317}
1318
8412663d
CY
1319/* This comes from f2fs_put_super */
1320void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
d431413f 1321{
bd5b0738 1322 __issue_discard_cmd(sbi, false);
969d1b18 1323 __drop_discard_cmd(sbi);
8412663d 1324 __wait_all_discard_cmd(sbi, false);
275b66b0
CY
1325}
1326
969d1b18
CY
1327static void mark_discard_range_all(struct f2fs_sb_info *sbi)
1328{
1329 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1330 int i;
1331
1332 mutex_lock(&dcc->cmd_lock);
1333 for (i = 0; i < MAX_PLIST_NUM; i++)
1334 dcc->pend_list_tag[i] |= P_TRIM;
1335 mutex_unlock(&dcc->cmd_lock);
1336}
1337
15469963
JK
1338static int issue_discard_thread(void *data)
1339{
1340 struct f2fs_sb_info *sbi = data;
1341 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1342 wait_queue_head_t *q = &dcc->discard_wait_queue;
ecc9aa00 1343 struct discard_policy *dpolicy = &dcc->dpolicy;
969d1b18
CY
1344 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1345 int issued;
15469963 1346
1d7be270 1347 set_freezable();
15469963 1348
1d7be270 1349 do {
969d1b18
CY
1350 wait_event_interruptible_timeout(*q,
1351 kthread_should_stop() || freezing(current) ||
1352 dcc->discard_wake,
1353 msecs_to_jiffies(wait_ms));
1d7be270
JK
1354 if (try_to_freeze())
1355 continue;
1356 if (kthread_should_stop())
1357 return 0;
15469963 1358
5f656541 1359 if (dcc->discard_wake) {
969d1b18 1360 dcc->discard_wake = 0;
5f656541
JK
1361 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1362 mark_discard_range_all(sbi);
1363 }
969d1b18 1364
dc6febb6
CY
1365 sb_start_intwrite(sbi->sb);
1366
969d1b18
CY
1367 issued = __issue_discard_cmd(sbi, true);
1368 if (issued) {
8412663d 1369 __wait_all_discard_cmd(sbi, true);
ecc9aa00 1370 wait_ms = dpolicy->min_interval;
969d1b18 1371 } else {
ecc9aa00 1372 wait_ms = dpolicy->max_interval;
969d1b18 1373 }
1d7be270 1374
dc6febb6 1375 sb_end_intwrite(sbi->sb);
1d7be270 1376
1d7be270
JK
1377 } while (!kthread_should_stop());
1378 return 0;
15469963
JK
1379}
1380
f46e8809 1381#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK
1382static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1383 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1384{
92592285 1385 sector_t sector, nr_sects;
10a875f8 1386 block_t lblkstart = blkstart;
3c62be17
JK
1387 int devi = 0;
1388
1389 if (sbi->s_ndevs) {
1390 devi = f2fs_target_device_index(sbi, blkstart);
1391 blkstart -= FDEV(devi).start_blk;
1392 }
f46e8809
DLM
1393
1394 /*
1395 * We need to know the type of the zone: for conventional zones,
1396 * use regular discard if the drive supports it. For sequential
1397 * zones, reset the zone write pointer.
1398 */
3c62be17 1399 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM
1400
1401 case BLK_ZONE_TYPE_CONVENTIONAL:
1402 if (!blk_queue_discard(bdev_get_queue(bdev)))
1403 return 0;
c81abe34 1404 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM
1405 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1406 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK
1407 sector = SECTOR_FROM_BLOCK(blkstart);
1408 nr_sects = SECTOR_FROM_BLOCK(blklen);
1409
1410 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1411 nr_sects != bdev_zone_sectors(bdev)) {
1412 f2fs_msg(sbi->sb, KERN_INFO,
1413 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1414 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1415 blkstart, blklen);
1416 return -EIO;
1417 }
d50aaeec 1418 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM
1419 return blkdev_reset_zones(bdev, sector,
1420 nr_sects, GFP_NOFS);
1421 default:
1422 /* Unknown zone type: broken device ? */
1423 return -EIO;
1424 }
1425}
1426#endif
1427
3c62be17
JK
1428static int __issue_discard_async(struct f2fs_sb_info *sbi,
1429 struct block_device *bdev, block_t blkstart, block_t blklen)
1430{
1431#ifdef CONFIG_BLK_DEV_ZONED
1432 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1433 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1434 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1435#endif
c81abe34 1436 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK
1437}
1438
1e87a78d 1439static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK
1440 block_t blkstart, block_t blklen)
1441{
3c62be17
JK
1442 sector_t start = blkstart, len = 0;
1443 struct block_device *bdev;
a66cdd98
JK
1444 struct seg_entry *se;
1445 unsigned int offset;
1446 block_t i;
3c62be17
JK
1447 int err = 0;
1448
1449 bdev = f2fs_target_device(sbi, blkstart, NULL);
1450
1451 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1452 if (i != start) {
1453 struct block_device *bdev2 =
1454 f2fs_target_device(sbi, i, NULL);
1455
1456 if (bdev2 != bdev) {
1457 err = __issue_discard_async(sbi, bdev,
1458 start, len);
1459 if (err)
1460 return err;
1461 bdev = bdev2;
1462 start = i;
1463 len = 0;
1464 }
1465 }
a66cdd98 1466
a66cdd98
JK
1467 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1468 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1469
1470 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1471 sbi->discard_blks--;
1472 }
f46e8809 1473
3c62be17
JK
1474 if (len)
1475 err = __issue_discard_async(sbi, bdev, start, len);
1476 return err;
1e87a78d
JK
1477}
1478
25290fa5
JK
1479static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1480 bool check_only)
adf4983b 1481{
b2955550
JK
1482 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1483 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1484 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK
1485 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1486 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1487 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1488 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1489 unsigned int start = 0, end = -1;
c473f1a9 1490 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1491 struct discard_entry *de = NULL;
46f84c2c 1492 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK
1493 int i;
1494
3e025740 1495 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1496 return false;
b2955550 1497
a66cdd98
JK
1498 if (!force) {
1499 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK
1500 SM_I(sbi)->dcc_info->nr_discards >=
1501 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1502 return false;
4b2fecc8
JK
1503 }
1504
b2955550
JK
1505 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1506 for (i = 0; i < entries; i++)
a66cdd98 1507 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1508 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1509
0b54fb84
JK
1510 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1511 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK
1512 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1513 if (start >= max_blocks)
1514 break;
1515
1516 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH
1517 if (force && start && end != max_blocks
1518 && (end - start) < cpc->trim_minlen)
1519 continue;
1520
25290fa5
JK
1521 if (check_only)
1522 return true;
1523
a7eeb823
CY
1524 if (!de) {
1525 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1526 GFP_F2FS_ZERO);
1527 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1528 list_add_tail(&de->list, head);
1529 }
1530
1531 for (i = start; i < end; i++)
1532 __set_bit_le(i, (void *)de->discard_map);
1533
1534 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1535 }
25290fa5 1536 return false;
b2955550
JK
1537}
1538
4b2fecc8
JK
1539void release_discard_addrs(struct f2fs_sb_info *sbi)
1540{
46f84c2c 1541 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK
1542 struct discard_entry *entry, *this;
1543
1544 /* drop caches */
1545 list_for_each_entry_safe(entry, this, head, list) {
1546 list_del(&entry->list);
1547 kmem_cache_free(discard_entry_slab, entry);
1548 }
1549}
1550
0a8165d7 1551/*
351df4b2
JK
1552 * Should call clear_prefree_segments after checkpoint is done.
1553 */
1554static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1555{
1556 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1557 unsigned int segno;
351df4b2
JK
1558
1559 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1560 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1561 __set_test_and_free(sbi, segno);
351df4b2
JK
1562 mutex_unlock(&dirty_i->seglist_lock);
1563}
1564
836b5a63 1565void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1566{
969d1b18
CY
1567 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1568 struct list_head *head = &dcc->entry_list;
2d7b822a 1569 struct discard_entry *entry, *this;
351df4b2 1570 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1571 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1572 unsigned int start = 0, end = -1;
36abef4e 1573 unsigned int secno, start_segno;
c473f1a9 1574 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK
1575
1576 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1577
351df4b2 1578 while (1) {
29e59c14 1579 int i;
7cd8558b
JK
1580 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1581 if (start >= MAIN_SEGS(sbi))
351df4b2 1582 break;
7cd8558b
JK
1583 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1584 start + 1);
29e59c14
CL
1585
1586 for (i = start; i < end; i++)
1587 clear_bit(i, prefree_map);
1588
1589 dirty_i->nr_dirty[PRE] -= end - start;
1590
650d3c4e 1591 if (!test_opt(sbi, DISCARD))
29e59c14 1592 continue;
351df4b2 1593
650d3c4e
YH
1594 if (force && start >= cpc->trim_start &&
1595 (end - 1) <= cpc->trim_end)
1596 continue;
1597
36abef4e
JK
1598 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1599 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1600 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK
1601 continue;
1602 }
1603next:
4ddb1a4d
JK
1604 secno = GET_SEC_FROM_SEG(sbi, start);
1605 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1606 if (!IS_CURSEC(sbi, secno) &&
302bd348 1607 !get_valid_blocks(sbi, start, true))
36abef4e
JK
1608 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1609 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1610
1611 start = start_segno + sbi->segs_per_sec;
1612 if (start < end)
1613 goto next;
8b107f5b
JK
1614 else
1615 end = start - 1;
351df4b2
JK
1616 }
1617 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK
1618
1619 /* send small discards */
2d7b822a 1620 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY
1621 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1622 bool is_valid = test_bit_le(0, entry->discard_map);
1623
1624find_next:
1625 if (is_valid) {
1626 next_pos = find_next_zero_bit_le(entry->discard_map,
1627 sbi->blocks_per_seg, cur_pos);
1628 len = next_pos - cur_pos;
1629
acfd2810
DLM
1630 if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1631 (force && len < cpc->trim_minlen))
a7eeb823
CY
1632 goto skip;
1633
1634 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1635 len);
1636 cpc->trimmed += len;
1637 total_len += len;
1638 } else {
1639 next_pos = find_next_bit_le(entry->discard_map,
1640 sbi->blocks_per_seg, cur_pos);
1641 }
836b5a63 1642skip:
a7eeb823
CY
1643 cur_pos = next_pos;
1644 is_valid = !is_valid;
1645
1646 if (cur_pos < sbi->blocks_per_seg)
1647 goto find_next;
1648
b2955550 1649 list_del(&entry->list);
969d1b18 1650 dcc->nr_discards -= total_len;
b2955550
JK
1651 kmem_cache_free(discard_entry_slab, entry);
1652 }
34e159da 1653
01983c71 1654 wake_up_discard_thread(sbi, false);
351df4b2
JK
1655}
1656
ecc9aa00
CY
1657static void inline init_discard_policy(struct discard_cmd_control *dcc)
1658{
1659 struct discard_policy *dpolicy = &dcc->dpolicy;
1660
1661 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1662 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1663 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1664 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1665 dpolicy->io_aware = true;
1666 dpolicy->sync = true;
1667}
1668
8ed59745 1669static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1670{
15469963 1671 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1672 struct discard_cmd_control *dcc;
ba48a33e 1673 int err = 0, i;
0b54fb84
JK
1674
1675 if (SM_I(sbi)->dcc_info) {
1676 dcc = SM_I(sbi)->dcc_info;
1677 goto init_thread;
1678 }
1679
1680 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1681 if (!dcc)
1682 return -ENOMEM;
1683
969d1b18 1684 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1685 INIT_LIST_HEAD(&dcc->entry_list);
969d1b18 1686 for (i = 0; i < MAX_PLIST_NUM; i++) {
ba48a33e 1687 INIT_LIST_HEAD(&dcc->pend_list[i]);
969d1b18
CY
1688 if (i >= dcc->discard_granularity - 1)
1689 dcc->pend_list_tag[i] |= P_ACTIVE;
1690 }
46f84c2c 1691 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 1692 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 1693 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY
1694 atomic_set(&dcc->issued_discard, 0);
1695 atomic_set(&dcc->issing_discard, 0);
5f32366a 1696 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1697 dcc->nr_discards = 0;
d618ebaf 1698 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1699 dcc->undiscard_blks = 0;
004b6862 1700 dcc->root = RB_ROOT;
0b54fb84 1701
ecc9aa00
CY
1702 init_discard_policy(dcc);
1703
15469963 1704 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK
1705 SM_I(sbi)->dcc_info = dcc;
1706init_thread:
15469963
JK
1707 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1708 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1709 if (IS_ERR(dcc->f2fs_issue_discard)) {
1710 err = PTR_ERR(dcc->f2fs_issue_discard);
1711 kfree(dcc);
1712 SM_I(sbi)->dcc_info = NULL;
1713 return err;
1714 }
1715
0b54fb84
JK
1716 return err;
1717}
1718
f099405f 1719static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK
1720{
1721 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1722
f099405f
CY
1723 if (!dcc)
1724 return;
1725
cce13252 1726 stop_discard_thread(sbi);
f099405f
CY
1727
1728 kfree(dcc);
1729 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK
1730}
1731
184a5cd2 1732static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK
1733{
1734 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY
1735
1736 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1737 sit_i->dirty_sentries++;
184a5cd2
CY
1738 return false;
1739 }
1740
1741 return true;
351df4b2
JK
1742}
1743
1744static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1745 unsigned int segno, int modified)
1746{
1747 struct seg_entry *se = get_seg_entry(sbi, segno);
1748 se->type = type;
1749 if (modified)
1750 __mark_sit_entry_dirty(sbi, segno);
1751}
1752
1753static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1754{
1755 struct seg_entry *se;
1756 unsigned int segno, offset;
1757 long int new_vblocks;
6415fedc
YS
1758 bool exist;
1759#ifdef CONFIG_F2FS_CHECK_FS
1760 bool mir_exist;
1761#endif
351df4b2
JK
1762
1763 segno = GET_SEGNO(sbi, blkaddr);
1764
1765 se = get_seg_entry(sbi, segno);
1766 new_vblocks = se->valid_blocks + del;
491c0854 1767 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1768
9850cf4a 1769 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK
1770 (new_vblocks > sbi->blocks_per_seg)));
1771
1772 se->valid_blocks = new_vblocks;
1773 se->mtime = get_mtime(sbi);
1774 SIT_I(sbi)->max_mtime = se->mtime;
1775
1776 /* Update valid block bitmap */
1777 if (del > 0) {
6415fedc 1778 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1779#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS
1780 mir_exist = f2fs_test_and_set_bit(offset,
1781 se->cur_valid_map_mir);
1782 if (unlikely(exist != mir_exist)) {
1783 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1784 "when setting bitmap, blk:%u, old bit:%d",
1785 blkaddr, exist);
05796763 1786 f2fs_bug_on(sbi, 1);
6415fedc 1787 }
355e7891 1788#endif
6415fedc
YS
1789 if (unlikely(exist)) {
1790 f2fs_msg(sbi->sb, KERN_ERR,
1791 "Bitmap was wrongly set, blk:%u", blkaddr);
1792 f2fs_bug_on(sbi, 1);
35ee82ca
YS
1793 se->valid_blocks--;
1794 del = 0;
355e7891 1795 }
6415fedc 1796
3e025740
JK
1797 if (f2fs_discard_en(sbi) &&
1798 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1799 sbi->discard_blks--;
720037f9
JK
1800
1801 /* don't overwrite by SSR to keep node chain */
1802 if (se->type == CURSEG_WARM_NODE) {
1803 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1804 se->ckpt_valid_blocks++;
1805 }
351df4b2 1806 } else {
6415fedc 1807 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1808#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS
1809 mir_exist = f2fs_test_and_clear_bit(offset,
1810 se->cur_valid_map_mir);
1811 if (unlikely(exist != mir_exist)) {
1812 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1813 "when clearing bitmap, blk:%u, old bit:%d",
1814 blkaddr, exist);
05796763 1815 f2fs_bug_on(sbi, 1);
6415fedc 1816 }
355e7891 1817#endif
6415fedc
YS
1818 if (unlikely(!exist)) {
1819 f2fs_msg(sbi->sb, KERN_ERR,
1820 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1821 f2fs_bug_on(sbi, 1);
35ee82ca
YS
1822 se->valid_blocks++;
1823 del = 0;
355e7891 1824 }
6415fedc 1825
3e025740
JK
1826 if (f2fs_discard_en(sbi) &&
1827 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1828 sbi->discard_blks++;
351df4b2
JK
1829 }
1830 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1831 se->ckpt_valid_blocks += del;
1832
1833 __mark_sit_entry_dirty(sbi, segno);
1834
1835 /* update total number of valid blocks to be written in ckpt area */
1836 SIT_I(sbi)->written_valid_blocks += del;
1837
1838 if (sbi->segs_per_sec > 1)
1839 get_sec_entry(sbi, segno)->valid_blocks += del;
1840}
1841
5e443818 1842void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1843{
5e443818
JK
1844 update_sit_entry(sbi, new, 1);
1845 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1846 update_sit_entry(sbi, old, -1);
1847
1848 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1849 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK
1850}
1851
1852void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1853{
1854 unsigned int segno = GET_SEGNO(sbi, addr);
1855 struct sit_info *sit_i = SIT_I(sbi);
1856
9850cf4a 1857 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK
1858 if (addr == NEW_ADDR)
1859 return;
1860
1861 /* add it into sit main buffer */
1862 mutex_lock(&sit_i->sentry_lock);
1863
1864 update_sit_entry(sbi, addr, -1);
1865
1866 /* add it into dirty seglist */
1867 locate_dirty_segment(sbi, segno);
1868
1869 mutex_unlock(&sit_i->sentry_lock);
1870}
1871
6e2c64ad
JK
1872bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1873{
1874 struct sit_info *sit_i = SIT_I(sbi);
1875 unsigned int segno, offset;
1876 struct seg_entry *se;
1877 bool is_cp = false;
1878
1879 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1880 return true;
1881
1882 mutex_lock(&sit_i->sentry_lock);
1883
1884 segno = GET_SEGNO(sbi, blkaddr);
1885 se = get_seg_entry(sbi, segno);
1886 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1887
1888 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1889 is_cp = true;
1890
1891 mutex_unlock(&sit_i->sentry_lock);
1892
1893 return is_cp;
1894}
1895
0a8165d7 1896/*
351df4b2
JK
1897 * This function should be resided under the curseg_mutex lock
1898 */
1899static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1900 struct f2fs_summary *sum)
351df4b2
JK
1901{
1902 struct curseg_info *curseg = CURSEG_I(sbi, type);
1903 void *addr = curseg->sum_blk;
e79efe3b 1904 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1905 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK
1906}
1907
0a8165d7 1908/*
351df4b2
JK
1909 * Calculate the number of current summary pages for writing
1910 */
3fa06d7b 1911int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1912{
351df4b2 1913 int valid_sum_count = 0;
9a47938b 1914 int i, sum_in_page;
351df4b2
JK
1915
1916 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1917 if (sbi->ckpt->alloc_type[i] == SSR)
1918 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY
1919 else {
1920 if (for_ra)
1921 valid_sum_count += le16_to_cpu(
1922 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1923 else
1924 valid_sum_count += curseg_blkoff(sbi, i);
1925 }
351df4b2
JK
1926 }
1927
09cbfeaf 1928 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL
1929 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1930 if (valid_sum_count <= sum_in_page)
351df4b2 1931 return 1;
9a47938b 1932 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1933 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK
1934 return 2;
1935 return 3;
1936}
1937
0a8165d7 1938/*
351df4b2
JK
1939 * Caller should put this summary page
1940 */
1941struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1942{
1943 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1944}
1945
381722d2 1946void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK
1947{
1948 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY
1949 void *dst = page_address(page);
1950
1951 if (src)
09cbfeaf 1952 memcpy(dst, src, PAGE_SIZE);
381722d2 1953 else
09cbfeaf 1954 memset(dst, 0, PAGE_SIZE);
351df4b2
JK
1955 set_page_dirty(page);
1956 f2fs_put_page(page, 1);
1957}
1958
381722d2
CY
1959static void write_sum_page(struct f2fs_sb_info *sbi,
1960 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1961{
1962 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1963}
1964
b7ad7512
CY
1965static void write_current_sum_page(struct f2fs_sb_info *sbi,
1966 int type, block_t blk_addr)
1967{
1968 struct curseg_info *curseg = CURSEG_I(sbi, type);
1969 struct page *page = grab_meta_page(sbi, blk_addr);
1970 struct f2fs_summary_block *src = curseg->sum_blk;
1971 struct f2fs_summary_block *dst;
1972
1973 dst = (struct f2fs_summary_block *)page_address(page);
1974
1975 mutex_lock(&curseg->curseg_mutex);
1976
1977 down_read(&curseg->journal_rwsem);
1978 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1979 up_read(&curseg->journal_rwsem);
1980
1981 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1982 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1983
1984 mutex_unlock(&curseg->curseg_mutex);
1985
1986 set_page_dirty(page);
1987 f2fs_put_page(page, 1);
1988}
1989
a7881893
JK
1990static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1991{
1992 struct curseg_info *curseg = CURSEG_I(sbi, type);
1993 unsigned int segno = curseg->segno + 1;
1994 struct free_segmap_info *free_i = FREE_I(sbi);
1995
1996 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1997 return !test_bit(segno, free_i->free_segmap);
1998 return 0;
1999}
2000
0a8165d7 2001/*
351df4b2
JK
2002 * Find a new segment from the free segments bitmap to right order
2003 * This function should be returned with success, otherwise BUG
2004 */
2005static void get_new_segment(struct f2fs_sb_info *sbi,
2006 unsigned int *newseg, bool new_sec, int dir)
2007{
2008 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2009 unsigned int segno, secno, zoneno;
7cd8558b 2010 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK
2011 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2012 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK
2013 unsigned int left_start = hint;
2014 bool init = true;
2015 int go_left = 0;
2016 int i;
2017
1a118ccf 2018 spin_lock(&free_i->segmap_lock);
351df4b2
JK
2019
2020 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2021 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK
2022 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2023 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK
2024 goto got_it;
2025 }
2026find_other_zone:
7cd8558b
JK
2027 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2028 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK
2029 if (dir == ALLOC_RIGHT) {
2030 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK
2031 MAIN_SECS(sbi), 0);
2032 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK
2033 } else {
2034 go_left = 1;
2035 left_start = hint - 1;
2036 }
2037 }
2038 if (go_left == 0)
2039 goto skip_left;
2040
2041 while (test_bit(left_start, free_i->free_secmap)) {
2042 if (left_start > 0) {
2043 left_start--;
2044 continue;
2045 }
2046 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK
2047 MAIN_SECS(sbi), 0);
2048 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK
2049 break;
2050 }
2051 secno = left_start;
2052skip_left:
2053 hint = secno;
4ddb1a4d
JK
2054 segno = GET_SEG_FROM_SEC(sbi, secno);
2055 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK
2056
2057 /* give up on finding another zone */
2058 if (!init)
2059 goto got_it;
2060 if (sbi->secs_per_zone == 1)
2061 goto got_it;
2062 if (zoneno == old_zoneno)
2063 goto got_it;
2064 if (dir == ALLOC_LEFT) {
2065 if (!go_left && zoneno + 1 >= total_zones)
2066 goto got_it;
2067 if (go_left && zoneno == 0)
2068 goto got_it;
2069 }
2070 for (i = 0; i < NR_CURSEG_TYPE; i++)
2071 if (CURSEG_I(sbi, i)->zone == zoneno)
2072 break;
2073
2074 if (i < NR_CURSEG_TYPE) {
2075 /* zone is in user, try another */
2076 if (go_left)
2077 hint = zoneno * sbi->secs_per_zone - 1;
2078 else if (zoneno + 1 >= total_zones)
2079 hint = 0;
2080 else
2081 hint = (zoneno + 1) * sbi->secs_per_zone;
2082 init = false;
2083 goto find_other_zone;
2084 }
2085got_it:
2086 /* set it as dirty segment in free segmap */
9850cf4a 2087 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK
2088 __set_inuse(sbi, segno);
2089 *newseg = segno;
1a118ccf 2090 spin_unlock(&free_i->segmap_lock);
351df4b2
JK
2091}
2092
2093static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2094{
2095 struct curseg_info *curseg = CURSEG_I(sbi, type);
2096 struct summary_footer *sum_footer;
2097
2098 curseg->segno = curseg->next_segno;
4ddb1a4d 2099 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK
2100 curseg->next_blkoff = 0;
2101 curseg->next_segno = NULL_SEGNO;
2102
2103 sum_footer = &(curseg->sum_blk->footer);
2104 memset(sum_footer, 0, sizeof(struct summary_footer));
2105 if (IS_DATASEG(type))
2106 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2107 if (IS_NODESEG(type))
2108 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2109 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2110}
2111
7a20b8a6
JK
2112static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2113{
a7881893
JK
2114 /* if segs_per_sec is large than 1, we need to keep original policy. */
2115 if (sbi->segs_per_sec != 1)
2116 return CURSEG_I(sbi, type)->segno;
2117
7a20b8a6
JK
2118 if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
2119 return 0;
2120
e066b83c
JK
2121 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2122 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
7a20b8a6
JK
2123 return CURSEG_I(sbi, type)->segno;
2124}
2125
0a8165d7 2126/*
351df4b2
JK
2127 * Allocate a current working segment.
2128 * This function always allocates a free segment in LFS manner.
2129 */
2130static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2131{
2132 struct curseg_info *curseg = CURSEG_I(sbi, type);
2133 unsigned int segno = curseg->segno;
2134 int dir = ALLOC_LEFT;
2135
2136 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2137 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK
2138 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2139 dir = ALLOC_RIGHT;
2140
2141 if (test_opt(sbi, NOHEAP))
2142 dir = ALLOC_RIGHT;
2143
7a20b8a6 2144 segno = __get_next_segno(sbi, type);
351df4b2
JK
2145 get_new_segment(sbi, &segno, new_sec, dir);
2146 curseg->next_segno = segno;
2147 reset_curseg(sbi, type, 1);
2148 curseg->alloc_type = LFS;
2149}
2150
2151static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2152 struct curseg_info *seg, block_t start)
2153{
2154 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2155 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2156 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL
2157 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2158 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2159 int i, pos;
2160
2161 for (i = 0; i < entries; i++)
2162 target_map[i] = ckpt_map[i] | cur_map[i];
2163
2164 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2165
2166 seg->next_blkoff = pos;
351df4b2
JK
2167}
2168
0a8165d7 2169/*
351df4b2
JK
2170 * If a segment is written by LFS manner, next block offset is just obtained
2171 * by increasing the current block offset. However, if a segment is written by
2172 * SSR manner, next block offset obtained by calling __next_free_blkoff
2173 */
2174static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2175 struct curseg_info *seg)
2176{
2177 if (seg->alloc_type == SSR)
2178 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2179 else
2180 seg->next_blkoff++;
2181}
2182
0a8165d7 2183/*
e1c42045 2184 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK
2185 * manner, so it should recover the existing segment information of valid blocks
2186 */
025d63a4 2187static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK
2188{
2189 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2190 struct curseg_info *curseg = CURSEG_I(sbi, type);
2191 unsigned int new_segno = curseg->next_segno;
2192 struct f2fs_summary_block *sum_node;
2193 struct page *sum_page;
2194
2195 write_sum_page(sbi, curseg->sum_blk,
2196 GET_SUM_BLOCK(sbi, curseg->segno));
2197 __set_test_and_inuse(sbi, new_segno);
2198
2199 mutex_lock(&dirty_i->seglist_lock);
2200 __remove_dirty_segment(sbi, new_segno, PRE);
2201 __remove_dirty_segment(sbi, new_segno, DIRTY);
2202 mutex_unlock(&dirty_i->seglist_lock);
2203
2204 reset_curseg(sbi, type, 1);
2205 curseg->alloc_type = SSR;
2206 __next_free_blkoff(sbi, curseg, 0);
2207
025d63a4
CY
2208 sum_page = get_sum_page(sbi, new_segno);
2209 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2210 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2211 f2fs_put_page(sum_page, 1);
351df4b2
JK
2212}
2213
43727527
JK
2214static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2215{
2216 struct curseg_info *curseg = CURSEG_I(sbi, type);
2217 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2218 unsigned segno = NULL_SEGNO;
d27c3d89
CY
2219 int i, cnt;
2220 bool reversed = false;
c192f7a4
JK
2221
2222 /* need_SSR() already forces to do this */
e066b83c
JK
2223 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2224 curseg->next_segno = segno;
c192f7a4 2225 return 1;
e066b83c 2226 }
43727527 2227
70d625cb
JK
2228 /* For node segments, let's do SSR more intensively */
2229 if (IS_NODESEG(type)) {
d27c3d89
CY
2230 if (type >= CURSEG_WARM_NODE) {
2231 reversed = true;
2232 i = CURSEG_COLD_NODE;
2233 } else {
2234 i = CURSEG_HOT_NODE;
2235 }
2236 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2237 } else {
d27c3d89
CY
2238 if (type >= CURSEG_WARM_DATA) {
2239 reversed = true;
2240 i = CURSEG_COLD_DATA;
2241 } else {
2242 i = CURSEG_HOT_DATA;
2243 }
2244 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2245 }
43727527 2246
d27c3d89 2247 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK
2248 if (i == type)
2249 continue;
e066b83c
JK
2250 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2251 curseg->next_segno = segno;
43727527 2252 return 1;
e066b83c 2253 }
c192f7a4 2254 }
43727527
JK
2255 return 0;
2256}
2257
351df4b2
JK
2258/*
2259 * flush out current segment and replace it with new segment
2260 * This function should be returned with success, otherwise BUG
2261 */
2262static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2263 int type, bool force)
2264{
a7881893
JK
2265 struct curseg_info *curseg = CURSEG_I(sbi, type);
2266
7b405275 2267 if (force)
351df4b2 2268 new_curseg(sbi, type, true);
5b6c6be2
JK
2269 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2270 type == CURSEG_WARM_NODE)
351df4b2 2271 new_curseg(sbi, type, false);
a7881893
JK
2272 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2273 new_curseg(sbi, type, false);
351df4b2 2274 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2275 change_curseg(sbi, type);
351df4b2
JK
2276 else
2277 new_curseg(sbi, type, false);
dcdfff65 2278
a7881893 2279 stat_inc_seg_type(sbi, curseg);
351df4b2
JK
2280}
2281
2282void allocate_new_segments(struct f2fs_sb_info *sbi)
2283{
6ae1be13
JK
2284 struct curseg_info *curseg;
2285 unsigned int old_segno;
351df4b2
JK
2286 int i;
2287
6ae1be13
JK
2288 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2289 curseg = CURSEG_I(sbi, i);
2290 old_segno = curseg->segno;
2291 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2292 locate_dirty_segment(sbi, old_segno);
2293 }
351df4b2
JK
2294}
2295
2296static const struct segment_allocation default_salloc_ops = {
2297 .allocate_segment = allocate_segment_by_default,
2298};
2299
25290fa5
JK
2300bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2301{
2302 __u64 trim_start = cpc->trim_start;
2303 bool has_candidate = false;
2304
2305 mutex_lock(&SIT_I(sbi)->sentry_lock);
2306 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2307 if (add_discard_addrs(sbi, cpc, true)) {
2308 has_candidate = true;
2309 break;
2310 }
2311 }
2312 mutex_unlock(&SIT_I(sbi)->sentry_lock);
2313
2314 cpc->trim_start = trim_start;
2315 return has_candidate;
2316}
2317
4b2fecc8
JK
2318int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2319{
f7ef9b83
JK
2320 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2321 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
8412663d
CY
2322 unsigned int start_segno, end_segno, cur_segno;
2323 block_t start_block, end_block;
4b2fecc8 2324 struct cp_control cpc;
c34f42e2 2325 int err = 0;
4b2fecc8 2326
836b5a63 2327 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK
2328 return -EINVAL;
2329
9bd27ae4 2330 cpc.trimmed = 0;
7cd8558b 2331 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK
2332 goto out;
2333
ed214a11
YH
2334 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2335 f2fs_msg(sbi->sb, KERN_WARNING,
2336 "Found FS corruption, run fsck to fix.");
2337 goto out;
2338 }
2339
4b2fecc8 2340 /* start/end segment number in main_area */
7cd8558b
JK
2341 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2342 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2343 GET_SEGNO(sbi, end);
8412663d
CY
2344
2345 start_block = START_BLOCK(sbi, start_segno);
2346 end_block = START_BLOCK(sbi, end_segno + 1);
2347
4b2fecc8 2348 cpc.reason = CP_DISCARD;
836b5a63 2349 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK
2350
2351 /* do checkpoint to issue discard commands safely */
8412663d
CY
2352 for (cur_segno = start_segno; cur_segno <= end_segno;
2353 cur_segno = cpc.trim_end + 1) {
2354 cpc.trim_start = cur_segno;
a66cdd98
JK
2355
2356 if (sbi->discard_blks == 0)
2357 break;
2358 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2359 cpc.trim_end = end_segno;
2360 else
2361 cpc.trim_end = min_t(unsigned int,
8412663d 2362 rounddown(cur_segno +
bba681cb
JK
2363 BATCHED_TRIM_SEGMENTS(sbi),
2364 sbi->segs_per_sec) - 1, end_segno);
2365
2366 mutex_lock(&sbi->gc_mutex);
c34f42e2 2367 err = write_checkpoint(sbi, &cpc);
bba681cb 2368 mutex_unlock(&sbi->gc_mutex);
e9328353
CY
2369 if (err)
2370 break;
74fa5f3d
CY
2371
2372 schedule();
bba681cb 2373 }
8412663d
CY
2374
2375 start_block = START_BLOCK(sbi, start_segno);
2376 end_block = START_BLOCK(sbi, min(cur_segno, end_segno) + 1);
2377
2378 __issue_discard_cmd_range(sbi, start_block, end_block, cpc.trim_minlen);
2379 __wait_discard_cmd_range(sbi, true, start_block, end_block,
2380 cpc.trim_minlen, true);
4b2fecc8 2381out:
f7ef9b83 2382 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 2383 return err;
4b2fecc8
JK
2384}
2385
351df4b2
JK
2386static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2387{
2388 struct curseg_info *curseg = CURSEG_I(sbi, type);
2389 if (curseg->next_blkoff < sbi->blocks_per_seg)
2390 return true;
2391 return false;
2392}
2393
81377bd6 2394static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2395{
81377bd6 2396 if (fio->type == DATA)
351df4b2
JK
2397 return CURSEG_HOT_DATA;
2398 else
2399 return CURSEG_HOT_NODE;
2400}
2401
81377bd6 2402static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2403{
81377bd6
JK
2404 if (fio->type == DATA) {
2405 struct inode *inode = fio->page->mapping->host;
351df4b2
JK
2406
2407 if (S_ISDIR(inode->i_mode))
2408 return CURSEG_HOT_DATA;
2409 else
2410 return CURSEG_COLD_DATA;
2411 } else {
81377bd6 2412 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2413 return CURSEG_WARM_NODE;
351df4b2
JK
2414 else
2415 return CURSEG_COLD_NODE;
2416 }
2417}
2418
81377bd6 2419static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2420{
81377bd6
JK
2421 if (fio->type == DATA) {
2422 struct inode *inode = fio->page->mapping->host;
351df4b2 2423
81377bd6 2424 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2425 return CURSEG_COLD_DATA;
ef095d19
JK
2426 if (is_inode_flag_set(inode, FI_HOT_DATA))
2427 return CURSEG_HOT_DATA;
2428 return CURSEG_WARM_DATA;
351df4b2 2429 } else {
81377bd6
JK
2430 if (IS_DNODE(fio->page))
2431 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2432 CURSEG_HOT_NODE;
ef095d19 2433 return CURSEG_COLD_NODE;
351df4b2
JK
2434 }
2435}
2436
81377bd6 2437static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2438{
a912b54d
JK
2439 int type = 0;
2440
81377bd6 2441 switch (fio->sbi->active_logs) {
351df4b2 2442 case 2:
a912b54d
JK
2443 type = __get_segment_type_2(fio);
2444 break;
351df4b2 2445 case 4:
a912b54d
JK
2446 type = __get_segment_type_4(fio);
2447 break;
2448 case 6:
2449 type = __get_segment_type_6(fio);
2450 break;
2451 default:
2452 f2fs_bug_on(fio->sbi, true);
351df4b2 2453 }
81377bd6 2454
a912b54d
JK
2455 if (IS_HOT(type))
2456 fio->temp = HOT;
2457 else if (IS_WARM(type))
2458 fio->temp = WARM;
2459 else
2460 fio->temp = COLD;
2461 return type;
351df4b2
JK
2462}
2463
bfad7c2d
JK
2464void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2465 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY
2466 struct f2fs_summary *sum, int type,
2467 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK
2468{
2469 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2470 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK
2471
2472 mutex_lock(&curseg->curseg_mutex);
21cb1d99 2473 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK
2474
2475 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2476
4e6a8d9b
JK
2477 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2478
351df4b2
JK
2479 /*
2480 * __add_sum_entry should be resided under the curseg_mutex
2481 * because, this function updates a summary entry in the
2482 * current summary block.
2483 */
e79efe3b 2484 __add_sum_entry(sbi, type, sum);
351df4b2 2485
351df4b2 2486 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK
2487
2488 stat_inc_block_count(sbi, curseg);
351df4b2 2489
c6f82fe9
JK
2490 if (!__has_curseg_space(sbi, type))
2491 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2 2492 /*
c6f82fe9
JK
2493 * SIT information should be updated after segment allocation,
2494 * since we need to keep dirty segments precisely under SSR.
351df4b2
JK
2495 */
2496 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 2497
351df4b2
JK
2498 mutex_unlock(&sit_i->sentry_lock);
2499
704956ec 2500 if (page && IS_NODESEG(type)) {
351df4b2
JK
2501 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2502
704956ec
CY
2503 f2fs_inode_chksum_set(sbi, page);
2504 }
2505
fb830fc5
CY
2506 if (add_list) {
2507 struct f2fs_bio_info *io;
2508
2509 INIT_LIST_HEAD(&fio->list);
2510 fio->in_list = true;
2511 io = sbi->write_io[fio->type] + fio->temp;
2512 spin_lock(&io->io_lock);
2513 list_add_tail(&fio->list, &io->io_list);
2514 spin_unlock(&io->io_lock);
2515 }
2516
bfad7c2d
JK
2517 mutex_unlock(&curseg->curseg_mutex);
2518}
2519
39d787be
CY
2520static void update_device_state(struct f2fs_io_info *fio)
2521{
2522 struct f2fs_sb_info *sbi = fio->sbi;
2523 unsigned int devidx;
2524
2525 if (!sbi->s_ndevs)
2526 return;
2527
2528 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2529
2530 /* update device state for fsync */
2531 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY
2532
2533 /* update device state for checkpoint */
2534 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2535 spin_lock(&sbi->dev_lock);
2536 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2537 spin_unlock(&sbi->dev_lock);
2538 }
39d787be
CY
2539}
2540
05ca3632 2541static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2542{
81377bd6 2543 int type = __get_segment_type(fio);
0a595eba 2544 int err;
bfad7c2d 2545
0a595eba 2546reallocate:
7a9d7548 2547 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2548 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2549
351df4b2 2550 /* writeout dirty page into bdev */
b9109b0e 2551 err = f2fs_submit_page_write(fio);
0a595eba
JK
2552 if (err == -EAGAIN) {
2553 fio->old_blkaddr = fio->new_blkaddr;
2554 goto reallocate;
39d787be
CY
2555 } else if (!err) {
2556 update_device_state(fio);
0a595eba 2557 }
351df4b2
JK
2558}
2559
b0af6d49
CY
2560void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2561 enum iostat_type io_type)
351df4b2 2562{
458e6197 2563 struct f2fs_io_info fio = {
05ca3632 2564 .sbi = sbi,
458e6197 2565 .type = META,
04d328de 2566 .op = REQ_OP_WRITE,
70fd7614 2567 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY
2568 .old_blkaddr = page->index,
2569 .new_blkaddr = page->index,
05ca3632 2570 .page = page,
4375a336 2571 .encrypted_page = NULL,
fb830fc5 2572 .in_list = false,
458e6197
JK
2573 };
2574
2b947003 2575 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2576 fio.op_flags &= ~REQ_META;
2b947003 2577
351df4b2 2578 set_page_writeback(page);
b9109b0e 2579 f2fs_submit_page_write(&fio);
b0af6d49
CY
2580
2581 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK
2582}
2583
05ca3632 2584void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK
2585{
2586 struct f2fs_summary sum;
05ca3632 2587
351df4b2 2588 set_summary(&sum, nid, 0, 0);
05ca3632 2589 do_write_page(&sum, fio);
b0af6d49
CY
2590
2591 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK
2592}
2593
05ca3632 2594void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2595{
05ca3632 2596 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK
2597 struct f2fs_summary sum;
2598 struct node_info ni;
2599
9850cf4a 2600 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK
2601 get_node_info(sbi, dn->nid, &ni);
2602 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2603 do_write_page(&sum, fio);
f28b3434 2604 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY
2605
2606 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK
2607}
2608
d1b3e72d 2609int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2610{
b0af6d49
CY
2611 int err;
2612
7a9d7548 2613 fio->new_blkaddr = fio->old_blkaddr;
05ca3632 2614 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY
2615
2616 err = f2fs_submit_page_bio(fio);
39d787be
CY
2617 if (!err)
2618 update_device_state(fio);
b0af6d49
CY
2619
2620 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2621
2622 return err;
351df4b2
JK
2623}
2624
4356e48e 2625void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2626 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2627 bool recover_curseg, bool recover_newaddr)
351df4b2
JK
2628{
2629 struct sit_info *sit_i = SIT_I(sbi);
2630 struct curseg_info *curseg;
2631 unsigned int segno, old_cursegno;
2632 struct seg_entry *se;
2633 int type;
19f106bc 2634 unsigned short old_blkoff;
351df4b2
JK
2635
2636 segno = GET_SEGNO(sbi, new_blkaddr);
2637 se = get_seg_entry(sbi, segno);
2638 type = se->type;
2639
19f106bc
CY
2640 if (!recover_curseg) {
2641 /* for recovery flow */
2642 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2643 if (old_blkaddr == NULL_ADDR)
2644 type = CURSEG_COLD_DATA;
2645 else
2646 type = CURSEG_WARM_DATA;
2647 }
2648 } else {
2649 if (!IS_CURSEG(sbi, segno))
351df4b2
JK
2650 type = CURSEG_WARM_DATA;
2651 }
19f106bc 2652
351df4b2
JK
2653 curseg = CURSEG_I(sbi, type);
2654
2655 mutex_lock(&curseg->curseg_mutex);
2656 mutex_lock(&sit_i->sentry_lock);
2657
2658 old_cursegno = curseg->segno;
19f106bc 2659 old_blkoff = curseg->next_blkoff;
351df4b2
JK
2660
2661 /* change the current segment */
2662 if (segno != curseg->segno) {
2663 curseg->next_segno = segno;
025d63a4 2664 change_curseg(sbi, type);
351df4b2
JK
2665 }
2666
491c0854 2667 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2668 __add_sum_entry(sbi, type, sum);
351df4b2 2669
28bc106b 2670 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK
2671 update_sit_entry(sbi, new_blkaddr, 1);
2672 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2673 update_sit_entry(sbi, old_blkaddr, -1);
2674
2675 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2676 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2677
351df4b2 2678 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2679
19f106bc
CY
2680 if (recover_curseg) {
2681 if (old_cursegno != curseg->segno) {
2682 curseg->next_segno = old_cursegno;
025d63a4 2683 change_curseg(sbi, type);
19f106bc
CY
2684 }
2685 curseg->next_blkoff = old_blkoff;
2686 }
2687
351df4b2
JK
2688 mutex_unlock(&sit_i->sentry_lock);
2689 mutex_unlock(&curseg->curseg_mutex);
2690}
2691
528e3459
CY
2692void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2693 block_t old_addr, block_t new_addr,
28bc106b
CY
2694 unsigned char version, bool recover_curseg,
2695 bool recover_newaddr)
528e3459
CY
2696{
2697 struct f2fs_summary sum;
2698
2699 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2700
28bc106b
CY
2701 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2702 recover_curseg, recover_newaddr);
528e3459 2703
f28b3434 2704 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY
2705}
2706
93dfe2ac 2707void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2708 enum page_type type, bool ordered)
93dfe2ac 2709{
93dfe2ac 2710 if (PageWriteback(page)) {
4081363f
JK
2711 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2712
b9109b0e
JK
2713 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2714 0, page->index, type);
fec1d657
JK
2715 if (ordered)
2716 wait_on_page_writeback(page);
2717 else
2718 wait_for_stable_page(page);
93dfe2ac
JK
2719 }
2720}
2721
d4c759ee 2722void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY
2723{
2724 struct page *cpage;
2725
5d4c0af4 2726 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY
2727 return;
2728
08b39fbd
CY
2729 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2730 if (cpage) {
fec1d657 2731 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY
2732 f2fs_put_page(cpage, 1);
2733 }
2734}
2735
351df4b2
JK
2736static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2737{
2738 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2739 struct curseg_info *seg_i;
2740 unsigned char *kaddr;
2741 struct page *page;
2742 block_t start;
2743 int i, j, offset;
2744
2745 start = start_sum_block(sbi);
2746
2747 page = get_meta_page(sbi, start++);
2748 kaddr = (unsigned char *)page_address(page);
2749
2750 /* Step 1: restore nat cache */
2751 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2752 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK
2753
2754 /* Step 2: restore sit cache */
2755 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2756 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK
2757 offset = 2 * SUM_JOURNAL_SIZE;
2758
2759 /* Step 3: restore summary entries */
2760 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2761 unsigned short blk_off;
2762 unsigned int segno;
2763
2764 seg_i = CURSEG_I(sbi, i);
2765 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2766 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2767 seg_i->next_segno = segno;
2768 reset_curseg(sbi, i, 0);
2769 seg_i->alloc_type = ckpt->alloc_type[i];
2770 seg_i->next_blkoff = blk_off;
2771
2772 if (seg_i->alloc_type == SSR)
2773 blk_off = sbi->blocks_per_seg;
2774
2775 for (j = 0; j < blk_off; j++) {
2776 struct f2fs_summary *s;
2777 s = (struct f2fs_summary *)(kaddr + offset);
2778 seg_i->sum_blk->entries[j] = *s;
2779 offset += SUMMARY_SIZE;
09cbfeaf 2780 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK
2781 SUM_FOOTER_SIZE)
2782 continue;
2783
2784 f2fs_put_page(page, 1);
2785 page = NULL;
2786
2787 page = get_meta_page(sbi, start++);
2788 kaddr = (unsigned char *)page_address(page);
2789 offset = 0;
2790 }
2791 }
2792 f2fs_put_page(page, 1);
2793 return 0;
2794}
2795
2796static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2797{
2798 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2799 struct f2fs_summary_block *sum;
2800 struct curseg_info *curseg;
2801 struct page *new;
2802 unsigned short blk_off;
2803 unsigned int segno = 0;
2804 block_t blk_addr = 0;
2805
2806 /* get segment number and block addr */
2807 if (IS_DATASEG(type)) {
2808 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2809 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2810 CURSEG_HOT_DATA]);
119ee914 2811 if (__exist_node_summaries(sbi))
351df4b2
JK
2812 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2813 else
2814 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2815 } else {
2816 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2817 CURSEG_HOT_NODE]);
2818 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2819 CURSEG_HOT_NODE]);
119ee914 2820 if (__exist_node_summaries(sbi))
351df4b2
JK
2821 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2822 type - CURSEG_HOT_NODE);
2823 else
2824 blk_addr = GET_SUM_BLOCK(sbi, segno);
2825 }
2826
2827 new = get_meta_page(sbi, blk_addr);
2828 sum = (struct f2fs_summary_block *)page_address(new);
2829
2830 if (IS_NODESEG(type)) {
119ee914 2831 if (__exist_node_summaries(sbi)) {
351df4b2
JK
2832 struct f2fs_summary *ns = &sum->entries[0];
2833 int i;
2834 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2835 ns->version = 0;
2836 ns->ofs_in_node = 0;
2837 }
2838 } else {
d653788a
GZ
2839 int err;
2840
2841 err = restore_node_summary(sbi, segno, sum);
2842 if (err) {
351df4b2 2843 f2fs_put_page(new, 1);
d653788a 2844 return err;
351df4b2
JK
2845 }
2846 }
2847 }
2848
2849 /* set uncompleted segment to curseg */
2850 curseg = CURSEG_I(sbi, type);
2851 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY
2852
2853 /* update journal info */
2854 down_write(&curseg->journal_rwsem);
2855 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2856 up_write(&curseg->journal_rwsem);
2857
2858 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2859 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK
2860 curseg->next_segno = segno;
2861 reset_curseg(sbi, type, 0);
2862 curseg->alloc_type = ckpt->alloc_type[type];
2863 curseg->next_blkoff = blk_off;
2864 mutex_unlock(&curseg->curseg_mutex);
2865 f2fs_put_page(new, 1);
2866 return 0;
2867}
2868
2869static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2870{
21d3f8e1
JQ
2871 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2872 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 2873 int type = CURSEG_HOT_DATA;
e4fc5fbf 2874 int err;
351df4b2 2875
aaec2b1d 2876 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY
2877 int npages = npages_for_summary_flush(sbi, true);
2878
2879 if (npages >= 2)
2880 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2881 META_CP, true);
3fa06d7b 2882
351df4b2
JK
2883 /* restore for compacted data summary */
2884 if (read_compacted_summaries(sbi))
2885 return -EINVAL;
2886 type = CURSEG_HOT_NODE;
2887 }
2888
119ee914 2889 if (__exist_node_summaries(sbi))
3fa06d7b 2890 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2891 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2892
e4fc5fbf
CY
2893 for (; type <= CURSEG_COLD_NODE; type++) {
2894 err = read_normal_summaries(sbi, type);
2895 if (err)
2896 return err;
2897 }
2898
21d3f8e1
JQ
2899 /* sanity check for summary blocks */
2900 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2901 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2902 return -EINVAL;
2903
351df4b2
JK
2904 return 0;
2905}
2906
2907static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2908{
2909 struct page *page;
2910 unsigned char *kaddr;
2911 struct f2fs_summary *summary;
2912 struct curseg_info *seg_i;
2913 int written_size = 0;
2914 int i, j;
2915
2916 page = grab_meta_page(sbi, blkaddr++);
2917 kaddr = (unsigned char *)page_address(page);
2918
2919 /* Step 1: write nat cache */
2920 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2921 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK
2922 written_size += SUM_JOURNAL_SIZE;
2923
2924 /* Step 2: write sit cache */
2925 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2926 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK
2927 written_size += SUM_JOURNAL_SIZE;
2928
351df4b2
JK
2929 /* Step 3: write summary entries */
2930 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2931 unsigned short blkoff;
2932 seg_i = CURSEG_I(sbi, i);
2933 if (sbi->ckpt->alloc_type[i] == SSR)
2934 blkoff = sbi->blocks_per_seg;
2935 else
2936 blkoff = curseg_blkoff(sbi, i);
2937
2938 for (j = 0; j < blkoff; j++) {
2939 if (!page) {
2940 page = grab_meta_page(sbi, blkaddr++);
2941 kaddr = (unsigned char *)page_address(page);
2942 written_size = 0;
2943 }
2944 summary = (struct f2fs_summary *)(kaddr + written_size);
2945 *summary = seg_i->sum_blk->entries[j];
2946 written_size += SUMMARY_SIZE;
351df4b2 2947
09cbfeaf 2948 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK
2949 SUM_FOOTER_SIZE)
2950 continue;
2951
e8d61a74 2952 set_page_dirty(page);
351df4b2
JK
2953 f2fs_put_page(page, 1);
2954 page = NULL;
2955 }
2956 }
e8d61a74
CY
2957 if (page) {
2958 set_page_dirty(page);
351df4b2 2959 f2fs_put_page(page, 1);
e8d61a74 2960 }
351df4b2
JK
2961}
2962
2963static void write_normal_summaries(struct f2fs_sb_info *sbi,
2964 block_t blkaddr, int type)
2965{
2966 int i, end;
2967 if (IS_DATASEG(type))
2968 end = type + NR_CURSEG_DATA_TYPE;
2969 else
2970 end = type + NR_CURSEG_NODE_TYPE;
2971
b7ad7512
CY
2972 for (i = type; i < end; i++)
2973 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK
2974}
2975
2976void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2977{
aaec2b1d 2978 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK
2979 write_compacted_summaries(sbi, start_blk);
2980 else
2981 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2982}
2983
2984void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2985{
119ee914 2986 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK
2987}
2988
dfc08a12 2989int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK
2990 unsigned int val, int alloc)
2991{
2992 int i;
2993
2994 if (type == NAT_JOURNAL) {
dfc08a12
CY
2995 for (i = 0; i < nats_in_cursum(journal); i++) {
2996 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK
2997 return i;
2998 }
dfc08a12
CY
2999 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3000 return update_nats_in_cursum(journal, 1);
351df4b2 3001 } else if (type == SIT_JOURNAL) {
dfc08a12
CY
3002 for (i = 0; i < sits_in_cursum(journal); i++)
3003 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3004 return i;
dfc08a12
CY
3005 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3006 return update_sits_in_cursum(journal, 1);
351df4b2
JK
3007 }
3008 return -1;
3009}
3010
3011static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3012 unsigned int segno)
3013{
2cc22186 3014 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK
3015}
3016
3017static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3018 unsigned int start)
3019{
3020 struct sit_info *sit_i = SIT_I(sbi);
3021 struct page *src_page, *dst_page;
3022 pgoff_t src_off, dst_off;
3023 void *src_addr, *dst_addr;
3024
3025 src_off = current_sit_addr(sbi, start);
3026 dst_off = next_sit_addr(sbi, src_off);
3027
3028 /* get current sit block page without lock */
3029 src_page = get_meta_page(sbi, src_off);
3030 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 3031 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK
3032
3033 src_addr = page_address(src_page);
3034 dst_addr = page_address(dst_page);
09cbfeaf 3035 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK
3036
3037 set_page_dirty(dst_page);
3038 f2fs_put_page(src_page, 1);
3039
3040 set_to_next_sit(sit_i, start);
3041
3042 return dst_page;
3043}
3044
184a5cd2
CY
3045static struct sit_entry_set *grab_sit_entry_set(void)
3046{
3047 struct sit_entry_set *ses =
80c54505 3048 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY
3049
3050 ses->entry_cnt = 0;
3051 INIT_LIST_HEAD(&ses->set_list);
3052 return ses;
3053}
3054
3055static void release_sit_entry_set(struct sit_entry_set *ses)
3056{
3057 list_del(&ses->set_list);
3058 kmem_cache_free(sit_entry_set_slab, ses);
3059}
3060
3061static void adjust_sit_entry_set(struct sit_entry_set *ses,
3062 struct list_head *head)
3063{
3064 struct sit_entry_set *next = ses;
3065
3066 if (list_is_last(&ses->set_list, head))
3067 return;
3068
3069 list_for_each_entry_continue(next, head, set_list)
3070 if (ses->entry_cnt <= next->entry_cnt)
3071 break;
3072
3073 list_move_tail(&ses->set_list, &next->set_list);
3074}
3075
3076static void add_sit_entry(unsigned int segno, struct list_head *head)
3077{
3078 struct sit_entry_set *ses;
3079 unsigned int start_segno = START_SEGNO(segno);
3080
3081 list_for_each_entry(ses, head, set_list) {
3082 if (ses->start_segno == start_segno) {
3083 ses->entry_cnt++;
3084 adjust_sit_entry_set(ses, head);
3085 return;
3086 }
3087 }
3088
3089 ses = grab_sit_entry_set();
3090
3091 ses->start_segno = start_segno;
3092 ses->entry_cnt++;
3093 list_add(&ses->set_list, head);
3094}
3095
3096static void add_sits_in_set(struct f2fs_sb_info *sbi)
3097{
3098 struct f2fs_sm_info *sm_info = SM_I(sbi);
3099 struct list_head *set_list = &sm_info->sit_entry_set;
3100 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY
3101 unsigned int segno;
3102
7cd8558b 3103 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY
3104 add_sit_entry(segno, set_list);
3105}
3106
3107static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK
3108{
3109 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3110 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK
3111 int i;
3112
b7ad7512 3113 down_write(&curseg->journal_rwsem);
dfc08a12 3114 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY
3115 unsigned int segno;
3116 bool dirtied;
3117
dfc08a12 3118 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY
3119 dirtied = __mark_sit_entry_dirty(sbi, segno);
3120
3121 if (!dirtied)
3122 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3123 }
dfc08a12 3124 update_sits_in_cursum(journal, -i);
b7ad7512 3125 up_write(&curseg->journal_rwsem);
351df4b2
JK
3126}
3127
0a8165d7 3128/*
351df4b2
JK
3129 * CP calls this function, which flushes SIT entries including sit_journal,
3130 * and moves prefree segs to free segs.
3131 */
4b2fecc8 3132void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK
3133{
3134 struct sit_info *sit_i = SIT_I(sbi);
3135 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3136 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3137 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY
3138 struct sit_entry_set *ses, *tmp;
3139 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3140 bool to_journal = true;
4b2fecc8 3141 struct seg_entry *se;
351df4b2 3142
351df4b2
JK
3143 mutex_lock(&sit_i->sentry_lock);
3144
2b11a74b
WL
3145 if (!sit_i->dirty_sentries)
3146 goto out;
3147
351df4b2 3148 /*
184a5cd2
CY
3149 * add and account sit entries of dirty bitmap in sit entry
3150 * set temporarily
351df4b2 3151 */
184a5cd2 3152 add_sits_in_set(sbi);
351df4b2 3153
184a5cd2
CY
3154 /*
3155 * if there are no enough space in journal to store dirty sit
3156 * entries, remove all entries from journal and add and account
3157 * them in sit entry set.
3158 */
dfc08a12 3159 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3160 remove_sits_in_journal(sbi);
b2955550 3161
184a5cd2
CY
3162 /*
3163 * there are two steps to flush sit entries:
3164 * #1, flush sit entries to journal in current cold data summary block.
3165 * #2, flush sit entries to sit page.
3166 */
3167 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3168 struct page *page = NULL;
184a5cd2
CY
3169 struct f2fs_sit_block *raw_sit = NULL;
3170 unsigned int start_segno = ses->start_segno;
3171 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3172 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY
3173 unsigned int segno = start_segno;
3174
3175 if (to_journal &&
dfc08a12 3176 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY
3177 to_journal = false;
3178
b7ad7512
CY
3179 if (to_journal) {
3180 down_write(&curseg->journal_rwsem);
3181 } else {
184a5cd2
CY
3182 page = get_next_sit_page(sbi, start_segno);
3183 raw_sit = page_address(page);
351df4b2 3184 }
351df4b2 3185
184a5cd2
CY
3186 /* flush dirty sit entries in region of current sit set */
3187 for_each_set_bit_from(segno, bitmap, end) {
3188 int offset, sit_offset;
4b2fecc8
JK
3189
3190 se = get_seg_entry(sbi, segno);
184a5cd2
CY
3191
3192 /* add discard candidates */
c473f1a9 3193 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3194 cpc->trim_start = segno;
25290fa5 3195 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3196 }
184a5cd2
CY
3197
3198 if (to_journal) {
dfc08a12 3199 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY
3200 SIT_JOURNAL, segno, 1);
3201 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3202 segno_in_journal(journal, offset) =
184a5cd2
CY
3203 cpu_to_le32(segno);
3204 seg_info_to_raw_sit(se,
dfc08a12 3205 &sit_in_journal(journal, offset));
184a5cd2
CY
3206 } else {
3207 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3208 seg_info_to_raw_sit(se,
3209 &raw_sit->entries[sit_offset]);
3210 }
351df4b2 3211
184a5cd2
CY
3212 __clear_bit(segno, bitmap);
3213 sit_i->dirty_sentries--;
3214 ses->entry_cnt--;
351df4b2
JK
3215 }
3216
b7ad7512
CY
3217 if (to_journal)
3218 up_write(&curseg->journal_rwsem);
3219 else
184a5cd2
CY
3220 f2fs_put_page(page, 1);
3221
3222 f2fs_bug_on(sbi, ses->entry_cnt);
3223 release_sit_entry_set(ses);
351df4b2 3224 }
184a5cd2
CY
3225
3226 f2fs_bug_on(sbi, !list_empty(head));
3227 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3228out:
c473f1a9 3229 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH
3230 __u64 trim_start = cpc->trim_start;
3231
4b2fecc8 3232 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3233 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH
3234
3235 cpc->trim_start = trim_start;
4b2fecc8 3236 }
351df4b2 3237 mutex_unlock(&sit_i->sentry_lock);
351df4b2 3238
351df4b2
JK
3239 set_prefree_as_free_segments(sbi);
3240}
3241
3242static int build_sit_info(struct f2fs_sb_info *sbi)
3243{
3244 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK
3245 struct sit_info *sit_i;
3246 unsigned int sit_segs, start;
ae27d62e 3247 char *src_bitmap;
351df4b2
JK
3248 unsigned int bitmap_size;
3249
3250 /* allocate memory for SIT information */
3251 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3252 if (!sit_i)
3253 return -ENOMEM;
3254
3255 SM_I(sbi)->sit_info = sit_i;
3256
a7c3e901 3257 sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
39307a8e 3258 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK
3259 if (!sit_i->sentries)
3260 return -ENOMEM;
3261
7cd8558b 3262 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3263 sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK
3264 if (!sit_i->dirty_sentries_bitmap)
3265 return -ENOMEM;
3266
7cd8558b 3267 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK
3268 sit_i->sentries[start].cur_valid_map
3269 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3270 sit_i->sentries[start].ckpt_valid_map
3271 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3272 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3273 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3274 return -ENOMEM;
3e025740 3275
355e7891
CY
3276#ifdef CONFIG_F2FS_CHECK_FS
3277 sit_i->sentries[start].cur_valid_map_mir
3278 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3279 if (!sit_i->sentries[start].cur_valid_map_mir)
3280 return -ENOMEM;
3281#endif
3282
3e025740
JK
3283 if (f2fs_discard_en(sbi)) {
3284 sit_i->sentries[start].discard_map
3285 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3286 if (!sit_i->sentries[start].discard_map)
3287 return -ENOMEM;
3288 }
351df4b2
JK
3289 }
3290
60a3b782
JK
3291 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3292 if (!sit_i->tmp_map)
3293 return -ENOMEM;
3294
351df4b2 3295 if (sbi->segs_per_sec > 1) {
a7c3e901 3296 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
39307a8e 3297 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK
3298 if (!sit_i->sec_entries)
3299 return -ENOMEM;
3300 }
3301
3302 /* get information related with SIT */
3303 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3304
3305 /* setup SIT bitmap from ckeckpoint pack */
3306 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3307 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3308
ae27d62e
CY
3309 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3310 if (!sit_i->sit_bitmap)
351df4b2 3311 return -ENOMEM;
351df4b2 3312
ae27d62e
CY
3313#ifdef CONFIG_F2FS_CHECK_FS
3314 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3315 if (!sit_i->sit_bitmap_mir)
3316 return -ENOMEM;
3317#endif
3318
351df4b2
JK
3319 /* init SIT information */
3320 sit_i->s_ops = &default_salloc_ops;
3321
3322 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3323 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3324 sit_i->written_valid_blocks = 0;
351df4b2
JK
3325 sit_i->bitmap_size = bitmap_size;
3326 sit_i->dirty_sentries = 0;
3327 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3328 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3329 sit_i->mounted_time = ktime_get_real_seconds();
351df4b2
JK
3330 mutex_init(&sit_i->sentry_lock);
3331 return 0;
3332}
3333
3334static int build_free_segmap(struct f2fs_sb_info *sbi)
3335{
351df4b2
JK
3336 struct free_segmap_info *free_i;
3337 unsigned int bitmap_size, sec_bitmap_size;
3338
3339 /* allocate memory for free segmap information */
3340 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3341 if (!free_i)
3342 return -ENOMEM;
3343
3344 SM_I(sbi)->free_info = free_i;
3345
7cd8558b 3346 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3347 free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK
3348 if (!free_i->free_segmap)
3349 return -ENOMEM;
3350
7cd8558b 3351 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
a7c3e901 3352 free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK
3353 if (!free_i->free_secmap)
3354 return -ENOMEM;
3355
3356 /* set all segments as dirty temporarily */
3357 memset(free_i->free_segmap, 0xff, bitmap_size);
3358 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3359
3360 /* init free segmap information */
7cd8558b 3361 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK
3362 free_i->free_segments = 0;
3363 free_i->free_sections = 0;
1a118ccf 3364 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK
3365 return 0;
3366}
3367
3368static int build_curseg(struct f2fs_sb_info *sbi)
3369{
1042d60f 3370 struct curseg_info *array;
351df4b2
JK
3371 int i;
3372
b434babf 3373 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK
3374 if (!array)
3375 return -ENOMEM;
3376
3377 SM_I(sbi)->curseg_array = array;
3378
3379 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3380 mutex_init(&array[i].curseg_mutex);
09cbfeaf 3381 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK
3382 if (!array[i].sum_blk)
3383 return -ENOMEM;
b7ad7512
CY
3384 init_rwsem(&array[i].journal_rwsem);
3385 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3386 GFP_KERNEL);
3387 if (!array[i].journal)
3388 return -ENOMEM;
351df4b2
JK
3389 array[i].segno = NULL_SEGNO;
3390 array[i].next_blkoff = 0;
3391 }
3392 return restore_curseg_summaries(sbi);
3393}
3394
3395static void build_sit_entries(struct f2fs_sb_info *sbi)
3396{
3397 struct sit_info *sit_i = SIT_I(sbi);
3398 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3399 struct f2fs_journal *journal = curseg->journal;
9c094040
YH
3400 struct seg_entry *se;
3401 struct f2fs_sit_entry sit;
74de593a
CY
3402 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3403 unsigned int i, start, end;
3404 unsigned int readed, start_blk = 0;
351df4b2 3405
74de593a 3406 do {
664ba972
JK
3407 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3408 META_SIT, true);
74de593a
CY
3409
3410 start = start_blk * sit_i->sents_per_block;
3411 end = (start_blk + readed) * sit_i->sents_per_block;
3412
7cd8558b 3413 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3414 struct f2fs_sit_block *sit_blk;
74de593a
CY
3415 struct page *page;
3416
9c094040 3417 se = &sit_i->sentries[start];
74de593a
CY
3418 page = get_current_sit_page(sbi, start);
3419 sit_blk = (struct f2fs_sit_block *)page_address(page);
3420 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3421 f2fs_put_page(page, 1);
d600af23 3422
74de593a
CY
3423 check_block_count(sbi, start, &sit);
3424 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK
3425
3426 /* build discard map only one time */
3e025740 3427 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY
3428 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3429 memset(se->discard_map, 0xff,
3430 SIT_VBLOCK_MAP_SIZE);
3431 } else {
3432 memcpy(se->discard_map,
3433 se->cur_valid_map,
3434 SIT_VBLOCK_MAP_SIZE);
3435 sbi->discard_blks +=
3436 sbi->blocks_per_seg -
3437 se->valid_blocks;
3438 }
3e025740 3439 }
a66cdd98 3440
d600af23
CY
3441 if (sbi->segs_per_sec > 1)
3442 get_sec_entry(sbi, start)->valid_blocks +=
3443 se->valid_blocks;
351df4b2 3444 }
74de593a
CY
3445 start_blk += readed;
3446 } while (start_blk < sit_blk_cnt);
d600af23
CY
3447
3448 down_read(&curseg->journal_rwsem);
3449 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY
3450 unsigned int old_valid_blocks;
3451
3452 start = le32_to_cpu(segno_in_journal(journal, i));
3453 se = &sit_i->sentries[start];
3454 sit = sit_in_journal(journal, i);
3455
3456 old_valid_blocks = se->valid_blocks;
3457
3458 check_block_count(sbi, start, &sit);
3459 seg_info_from_raw_sit(se, &sit);
3460
3461 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY
3462 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3463 memset(se->discard_map, 0xff,
3464 SIT_VBLOCK_MAP_SIZE);
3465 } else {
3466 memcpy(se->discard_map, se->cur_valid_map,
3467 SIT_VBLOCK_MAP_SIZE);
3468 sbi->discard_blks += old_valid_blocks -
3469 se->valid_blocks;
3470 }
d600af23
CY
3471 }
3472
3473 if (sbi->segs_per_sec > 1)
3474 get_sec_entry(sbi, start)->valid_blocks +=
3475 se->valid_blocks - old_valid_blocks;
3476 }
3477 up_read(&curseg->journal_rwsem);
351df4b2
JK
3478}
3479
3480static void init_free_segmap(struct f2fs_sb_info *sbi)
3481{
3482 unsigned int start;
3483 int type;
3484
7cd8558b 3485 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK
3486 struct seg_entry *sentry = get_seg_entry(sbi, start);
3487 if (!sentry->valid_blocks)
3488 __set_free(sbi, start);
c79b7ff1
JK
3489 else
3490 SIT_I(sbi)->written_valid_blocks +=
3491 sentry->valid_blocks;
351df4b2
JK
3492 }
3493
3494 /* set use the current segments */
3495 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3496 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3497 __set_test_and_inuse(sbi, curseg_t->segno);
3498 }
3499}
3500
3501static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3502{
3503 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3504 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3505 unsigned int segno = 0, offset = 0;
351df4b2
JK
3506 unsigned short valid_blocks;
3507
8736fbf0 3508 while (1) {
351df4b2 3509 /* find dirty segment based on free segmap */
7cd8558b
JK
3510 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3511 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK
3512 break;
3513 offset = segno + 1;
302bd348 3514 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3515 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3516 continue;
ec325b52
JK
3517 if (valid_blocks > sbi->blocks_per_seg) {
3518 f2fs_bug_on(sbi, 1);
3519 continue;
3520 }
351df4b2
JK
3521 mutex_lock(&dirty_i->seglist_lock);
3522 __locate_dirty_segment(sbi, segno, DIRTY);
3523 mutex_unlock(&dirty_i->seglist_lock);
3524 }
3525}
3526
5ec4e49f 3527static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK
3528{
3529 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3530 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3531
a7c3e901 3532 dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 3533 if (!dirty_i->victim_secmap)
351df4b2
JK
3534 return -ENOMEM;
3535 return 0;
3536}
3537
3538static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3539{
3540 struct dirty_seglist_info *dirty_i;
3541 unsigned int bitmap_size, i;
3542
3543 /* allocate memory for dirty segments list information */
3544 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3545 if (!dirty_i)
3546 return -ENOMEM;
3547
3548 SM_I(sbi)->dirty_info = dirty_i;
3549 mutex_init(&dirty_i->seglist_lock);
3550
7cd8558b 3551 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK
3552
3553 for (i = 0; i < NR_DIRTY_TYPE; i++) {
a7c3e901 3554 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK
3555 if (!dirty_i->dirty_segmap[i])
3556 return -ENOMEM;
3557 }
3558
3559 init_dirty_segmap(sbi);
5ec4e49f 3560 return init_victim_secmap(sbi);
351df4b2
JK
3561}
3562
0a8165d7 3563/*
351df4b2
JK
3564 * Update min, max modified time for cost-benefit GC algorithm
3565 */
3566static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3567{
3568 struct sit_info *sit_i = SIT_I(sbi);
3569 unsigned int segno;
3570
3571 mutex_lock(&sit_i->sentry_lock);
3572
3573 sit_i->min_mtime = LLONG_MAX;
3574
7cd8558b 3575 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK
3576 unsigned int i;
3577 unsigned long long mtime = 0;
3578
3579 for (i = 0; i < sbi->segs_per_sec; i++)
3580 mtime += get_seg_entry(sbi, segno + i)->mtime;
3581
3582 mtime = div_u64(mtime, sbi->segs_per_sec);
3583
3584 if (sit_i->min_mtime > mtime)
3585 sit_i->min_mtime = mtime;
3586 }
3587 sit_i->max_mtime = get_mtime(sbi);
3588 mutex_unlock(&sit_i->sentry_lock);
3589}
3590
3591int build_segment_manager(struct f2fs_sb_info *sbi)
3592{
3593 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3594 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3595 struct f2fs_sm_info *sm_info;
351df4b2
JK
3596 int err;
3597
3598 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3599 if (!sm_info)
3600 return -ENOMEM;
3601
3602 /* init sm info */
3603 sbi->sm_info = sm_info;
351df4b2
JK
3604 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3605 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3606 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3607 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3608 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3609 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3610 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK
3611 sm_info->rec_prefree_segments = sm_info->main_segments *
3612 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK
3613 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3614 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3615
52763a4b
JK
3616 if (!test_opt(sbi, LFS))
3617 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3618 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3619 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3620 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
351df4b2 3621
bba681cb
JK
3622 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3623
184a5cd2
CY
3624 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3625
d4fdf8ba 3626 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ
3627 err = create_flush_cmd_control(sbi);
3628 if (err)
a688b9d9 3629 return err;
6b4afdd7
JK
3630 }
3631
0b54fb84
JK
3632 err = create_discard_cmd_control(sbi);
3633 if (err)
3634 return err;
3635
351df4b2
JK
3636 err = build_sit_info(sbi);
3637 if (err)
3638 return err;
3639 err = build_free_segmap(sbi);
3640 if (err)
3641 return err;
3642 err = build_curseg(sbi);
3643 if (err)
3644 return err;
3645
3646 /* reinit free segmap based on SIT */
3647 build_sit_entries(sbi);
3648
3649 init_free_segmap(sbi);
3650 err = build_dirty_segmap(sbi);
3651 if (err)
3652 return err;
3653
3654 init_min_max_mtime(sbi);
3655 return 0;
3656}
3657
3658static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3659 enum dirty_type dirty_type)
3660{
3661 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3662
3663 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3664 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK
3665 dirty_i->nr_dirty[dirty_type] = 0;
3666 mutex_unlock(&dirty_i->seglist_lock);
3667}
3668
5ec4e49f 3669static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK
3670{
3671 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3672 kvfree(dirty_i->victim_secmap);
351df4b2
JK
3673}
3674
3675static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3676{
3677 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3678 int i;
3679
3680 if (!dirty_i)
3681 return;
3682
3683 /* discard pre-free/dirty segments list */
3684 for (i = 0; i < NR_DIRTY_TYPE; i++)
3685 discard_dirty_segmap(sbi, i);
3686
5ec4e49f 3687 destroy_victim_secmap(sbi);
351df4b2
JK
3688 SM_I(sbi)->dirty_info = NULL;
3689 kfree(dirty_i);
3690}
3691
3692static void destroy_curseg(struct f2fs_sb_info *sbi)
3693{
3694 struct curseg_info *array = SM_I(sbi)->curseg_array;
3695 int i;
3696
3697 if (!array)
3698 return;
3699 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3700 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3701 kfree(array[i].sum_blk);
b7ad7512
CY
3702 kfree(array[i].journal);
3703 }
351df4b2
JK
3704 kfree(array);
3705}
3706
3707static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3708{
3709 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3710 if (!free_i)
3711 return;
3712 SM_I(sbi)->free_info = NULL;
39307a8e
JK
3713 kvfree(free_i->free_segmap);
3714 kvfree(free_i->free_secmap);
351df4b2
JK
3715 kfree(free_i);
3716}
3717
3718static void destroy_sit_info(struct f2fs_sb_info *sbi)
3719{
3720 struct sit_info *sit_i = SIT_I(sbi);
3721 unsigned int start;
3722
3723 if (!sit_i)
3724 return;
3725
3726 if (sit_i->sentries) {
7cd8558b 3727 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3728 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY
3729#ifdef CONFIG_F2FS_CHECK_FS
3730 kfree(sit_i->sentries[start].cur_valid_map_mir);
3731#endif
351df4b2 3732 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3733 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK
3734 }
3735 }
60a3b782
JK
3736 kfree(sit_i->tmp_map);
3737
39307a8e
JK
3738 kvfree(sit_i->sentries);
3739 kvfree(sit_i->sec_entries);
3740 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK
3741
3742 SM_I(sbi)->sit_info = NULL;
3743 kfree(sit_i->sit_bitmap);
ae27d62e
CY
3744#ifdef CONFIG_F2FS_CHECK_FS
3745 kfree(sit_i->sit_bitmap_mir);
3746#endif
351df4b2
JK
3747 kfree(sit_i);
3748}
3749
3750void destroy_segment_manager(struct f2fs_sb_info *sbi)
3751{
3752 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3753
3b03f724
CY
3754 if (!sm_info)
3755 return;
5eba8c5d 3756 destroy_flush_cmd_control(sbi, true);
f099405f 3757 destroy_discard_cmd_control(sbi);
351df4b2
JK
3758 destroy_dirty_segmap(sbi);
3759 destroy_curseg(sbi);
3760 destroy_free_segmap(sbi);
3761 destroy_sit_info(sbi);
3762 sbi->sm_info = NULL;
3763 kfree(sm_info);
3764}
7fd9e544
JK
3765
3766int __init create_segment_manager_caches(void)
3767{
3768 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3769 sizeof(struct discard_entry));
7fd9e544 3770 if (!discard_entry_slab)
184a5cd2
CY
3771 goto fail;
3772
b01a9201
JK
3773 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3774 sizeof(struct discard_cmd));
3775 if (!discard_cmd_slab)
6ab2a308 3776 goto destroy_discard_entry;
275b66b0 3777
184a5cd2 3778 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3779 sizeof(struct sit_entry_set));
184a5cd2 3780 if (!sit_entry_set_slab)
b01a9201 3781 goto destroy_discard_cmd;
88b88a66
JK
3782
3783 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3784 sizeof(struct inmem_pages));
3785 if (!inmem_entry_slab)
3786 goto destroy_sit_entry_set;
7fd9e544 3787 return 0;
184a5cd2 3788
88b88a66
JK
3789destroy_sit_entry_set:
3790 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK
3791destroy_discard_cmd:
3792 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3793destroy_discard_entry:
184a5cd2
CY
3794 kmem_cache_destroy(discard_entry_slab);
3795fail:
3796 return -ENOMEM;
7fd9e544
JK
3797}
3798
3799void destroy_segment_manager_caches(void)
3800{
184a5cd2 3801 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3802 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3803 kmem_cache_destroy(discard_entry_slab);
88b88a66 3804 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3805}
This page took 0.909115 seconds and 4 git commands to generate.