projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: fix out-of-order execution in f2fs_issue_flush
[linux-2.6-block.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>
351df4b2
JK		 20
21#include "f2fs.h"
22#include "segment.h"
23#include "node.h"
5f656541 24#include "gc.h"
9e4ded3f 25#include "trace.h"
6ec178da 26#include <trace/events/f2fs.h>
351df4b2 27
9a7f143a
CL		 28#define __reverse_ffz(x) __reverse_ffs(~(x))
29
7fd9e544 30static struct kmem_cache *discard_entry_slab;
b01a9201 31static struct kmem_cache *discard_cmd_slab;
184a5cd2 32static struct kmem_cache *sit_entry_set_slab;
88b88a66 33static struct kmem_cache *inmem_entry_slab;
7fd9e544 34
f96999c3
JK		 35static unsigned long __reverse_ulong(unsigned char *str)
36{
37 unsigned long tmp = 0;
38 int shift = 24, idx = 0;
39
40#if BITS_PER_LONG == 64
41 shift = 56;
42#endif
43 while (shift >= 0) {
44 tmp |= (unsigned long)str[idx++] << shift;
45 shift -= BITS_PER_BYTE;
46 }
47 return tmp;
48}
49
9a7f143a
CL		 50/*
51 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
52 * MSB and LSB are reversed in a byte by f2fs_set_bit.
53 */
54static inline unsigned long __reverse_ffs(unsigned long word)
55{
56 int num = 0;
57
58#if BITS_PER_LONG == 64
f96999c3 59 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 60 num += 32;
f96999c3 61 else
9a7f143a 62 word >>= 32;
9a7f143a 63#endif
f96999c3 64 if ((word & 0xffff0000) == 0)
9a7f143a 65 num += 16;
f96999c3 66 else
9a7f143a 67 word >>= 16;
f96999c3
JK		 68
69 if ((word & 0xff00) == 0)
9a7f143a 70 num += 8;
f96999c3 71 else
9a7f143a 72 word >>= 8;
f96999c3 73
9a7f143a
CL		 74 if ((word & 0xf0) == 0)
75 num += 4;
76 else
77 word >>= 4;
f96999c3 78
9a7f143a
CL		 79 if ((word & 0xc) == 0)
80 num += 2;
81 else
82 word >>= 2;
f96999c3 83
9a7f143a
CL		 84 if ((word & 0x2) == 0)
85 num += 1;
86 return num;
87}
88
89/*
e1c42045 90 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 91 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 92 * @size must be integral times of unsigned long.
9a7f143a 93 * Example:
f96999c3
JK		 94 * MSB <--> LSB
95 * f2fs_set_bit(0, bitmap) => 1000 0000
96 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 97 */
98static unsigned long __find_rev_next_bit(const unsigned long *addr,
99 unsigned long size, unsigned long offset)
100{
101 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 102 unsigned long result = size;
9a7f143a 103 unsigned long tmp;
9a7f143a
CL		 104
105 if (offset >= size)
106 return size;
107
692223d1 108 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 109 offset %= BITS_PER_LONG;
f96999c3 110
692223d1
FL		 111 while (1) {
112 if (*p == 0)
113 goto pass;
9a7f143a 114
f96999c3 115 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 116
117 tmp &= ~0UL >> offset;
118 if (size < BITS_PER_LONG)
119 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 120 if (tmp)
692223d1
FL		 121 goto found;
122pass:
123 if (size <= BITS_PER_LONG)
124 break;
9a7f143a 125 size -= BITS_PER_LONG;
692223d1 126 offset = 0;
f96999c3 127 p++;
9a7f143a 128 }
692223d1
FL		 129 return result;
130found:
131 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 132}
133
134static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
135 unsigned long size, unsigned long offset)
136{
137 const unsigned long *p = addr + BIT_WORD(offset);
80609448 138 unsigned long result = size;
9a7f143a 139 unsigned long tmp;
9a7f143a
CL		 140
141 if (offset >= size)
142 return size;
143
80609448 144 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 145 offset %= BITS_PER_LONG;
80609448
JK		 146
147 while (1) {
148 if (*p == ~0UL)
149 goto pass;
150
f96999c3 151 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 152
153 if (offset)
154 tmp |= ~0UL << (BITS_PER_LONG - offset);
155 if (size < BITS_PER_LONG)
156 tmp |= ~0UL >> size;
f96999c3 157 if (tmp != ~0UL)
80609448
JK		 158 goto found;
159pass:
160 if (size <= BITS_PER_LONG)
161 break;
9a7f143a 162 size -= BITS_PER_LONG;
80609448 163 offset = 0;
f96999c3 164 p++;
9a7f143a 165 }
80609448
JK		 166 return result;
167found:
168 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 169}
170
88b88a66
JK		 171void register_inmem_page(struct inode *inode, struct page *page)
172{
173 struct f2fs_inode_info *fi = F2FS_I(inode);
174 struct inmem_pages *new;
9be32d72 175
9e4ded3f 176 f2fs_trace_pid(page);
0722b101 177
decd36b6
CY		 178 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
179 SetPagePrivate(page);
180
88b88a66
JK		 181 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
182
183 /* add atomic page indices to the list */
184 new->page = page;
185 INIT_LIST_HEAD(&new->list);
decd36b6 186
88b88a66
JK		 187 /* increase reference count with clean state */
188 mutex_lock(&fi->inmem_lock);
189 get_page(page);
190 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 191 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 192 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 193
194 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 195}
196
28bc106b
CY		 197static int __revoke_inmem_pages(struct inode *inode,
198 struct list_head *head, bool drop, bool recover)
29b96b54 199{
28bc106b 200 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 201 struct inmem_pages *cur, *tmp;
28bc106b 202 int err = 0;
29b96b54
CY		 203
204 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 205 struct page *page = cur->page;
206
207 if (drop)
208 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
209
210 lock_page(page);
29b96b54 211
28bc106b
CY		 212 if (recover) {
213 struct dnode_of_data dn;
214 struct node_info ni;
215
216 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 217retry:
28bc106b 218 set_new_dnode(&dn, inode, NULL, NULL, 0);
7f2b4e8e
CY		 219 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
220 if (err) {
221 if (err == -ENOMEM) {
222 congestion_wait(BLK_RW_ASYNC, HZ/50);
223 cond_resched();
224 goto retry;
225 }
28bc106b
CY		 226 err = -EAGAIN;
227 goto next;
228 }
229 get_node_info(sbi, dn.nid, &ni);
230 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
231 cur->old_addr, ni.version, true, true);
232 f2fs_put_dnode(&dn);
233 }
234next:
63c52d78
JK		 235 /* we don't need to invalidate this in the sccessful status */
236 if (drop || recover)
237 ClearPageUptodate(page);
28bc106b 238 set_page_private(page, 0);
c81ced05 239 ClearPagePrivate(page);
28bc106b 240 f2fs_put_page(page, 1);
29b96b54
CY		 241
242 list_del(&cur->list);
243 kmem_cache_free(inmem_entry_slab, cur);
244 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
245 }
28bc106b 246 return err;
29b96b54
CY		 247}
248
249void drop_inmem_pages(struct inode *inode)
250{
251 struct f2fs_inode_info *fi = F2FS_I(inode);
252
253 mutex_lock(&fi->inmem_lock);
28bc106b 254 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54 255 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 256
257 clear_inode_flag(inode, FI_ATOMIC_FILE);
258 stat_dec_atomic_write(inode);
29b96b54
CY		 259}
260
8c242db9
JK		 261void drop_inmem_page(struct inode *inode, struct page *page)
262{
263 struct f2fs_inode_info *fi = F2FS_I(inode);
264 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
265 struct list_head *head = &fi->inmem_pages;
266 struct inmem_pages *cur = NULL;
267
268 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
269
270 mutex_lock(&fi->inmem_lock);
271 list_for_each_entry(cur, head, list) {
272 if (cur->page == page)
273 break;
274 }
275
276 f2fs_bug_on(sbi, !cur || cur->page != page);
277 list_del(&cur->list);
278 mutex_unlock(&fi->inmem_lock);
279
280 dec_page_count(sbi, F2FS_INMEM_PAGES);
281 kmem_cache_free(inmem_entry_slab, cur);
282
283 ClearPageUptodate(page);
284 set_page_private(page, 0);
285 ClearPagePrivate(page);
286 f2fs_put_page(page, 0);
287
288 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
289}
290
28bc106b
CY		 291static int __commit_inmem_pages(struct inode *inode,
292 struct list_head *revoke_list)
88b88a66
JK		 293{
294 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
295 struct f2fs_inode_info *fi = F2FS_I(inode);
296 struct inmem_pages *cur, *tmp;
88b88a66 297 struct f2fs_io_info fio = {
05ca3632 298 .sbi = sbi,
88b88a66 299 .type = DATA,
04d328de 300 .op = REQ_OP_WRITE,
70fd7614 301 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 302 .io_type = FS_DATA_IO,
88b88a66 303 };
942fd319 304 pgoff_t last_idx = ULONG_MAX;
edb27dee 305 int err = 0;
88b88a66 306
88b88a66 307 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 308 struct page *page = cur->page;
309
310 lock_page(page);
311 if (page->mapping == inode->i_mapping) {
312 trace_f2fs_commit_inmem_page(page, INMEM);
313
314 set_page_dirty(page);
315 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 316 if (clear_page_dirty_for_io(page)) {
29b96b54 317 inode_dec_dirty_pages(inode);
933439c8
CY		 318 remove_dirty_inode(inode);
319 }
640cc189 320retry:
28bc106b 321 fio.page = page;
e959c8f5 322 fio.old_blkaddr = NULL_ADDR;
4d978078 323 fio.encrypted_page = NULL;
cc15620b 324 fio.need_lock = LOCK_DONE;
29b96b54
CY		 325 err = do_write_data_page(&fio);
326 if (err) {
640cc189
JK		 327 if (err == -ENOMEM) {
328 congestion_wait(BLK_RW_ASYNC, HZ/50);
329 cond_resched();
330 goto retry;
331 }
28bc106b 332 unlock_page(page);
29b96b54 333 break;
70c640b1 334 }
28bc106b
CY		 335 /* record old blkaddr for revoking */
336 cur->old_addr = fio.old_blkaddr;
942fd319 337 last_idx = page->index;
28bc106b
CY		 338 }
339 unlock_page(page);
340 list_move_tail(&cur->list, revoke_list);
88b88a66 341 }
29b96b54 342
942fd319 343 if (last_idx != ULONG_MAX)
b9109b0e 344 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b
CY		 345
346 if (!err)
347 __revoke_inmem_pages(inode, revoke_list, false, false);
348
29b96b54
CY		 349 return err;
350}
351
352int commit_inmem_pages(struct inode *inode)
353{
354 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
355 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 356 struct list_head revoke_list;
357 int err;
29b96b54 358
28bc106b 359 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 360 f2fs_balance_fs(sbi, true);
361 f2fs_lock_op(sbi);
362
5fe45743
CY		 363 set_inode_flag(inode, FI_ATOMIC_COMMIT);
364
29b96b54 365 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 366 err = __commit_inmem_pages(inode, &revoke_list);
367 if (err) {
368 int ret;
369 /*
370 * try to revoke all committed pages, but still we could fail
371 * due to no memory or other reason, if that happened, EAGAIN
372 * will be returned, which means in such case, transaction is
373 * already not integrity, caller should use journal to do the
374 * recovery or rewrite & commit last transaction. For other
375 * error number, revoking was done by filesystem itself.
376 */
377 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
378 if (ret)
379 err = ret;
380
381 /* drop all uncommitted pages */
382 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
383 }
88b88a66
JK		 384 mutex_unlock(&fi->inmem_lock);
385
5fe45743
CY		 386 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
387
29b96b54 388 f2fs_unlock_op(sbi);
edb27dee 389 return err;
88b88a66
JK		 390}
391
0a8165d7 392/*
351df4b2
JK		 393 * This function balances dirty node and dentry pages.
394 * In addition, it controls garbage collection.
395 */
2c4db1a6 396void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 397{
0f348028 398#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 399 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
400 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 401 f2fs_stop_checkpoint(sbi, false);
55523519 402 }
0f348028
CY		 403#endif
404
e589c2c4 405 /* balance_fs_bg is able to be pending */
a7881893 406 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 407 f2fs_balance_fs_bg(sbi);
408
351df4b2 409 /*
029cd28c
JK		 410 * We should do GC or end up with checkpoint, if there are so many dirty
411 * dir/node pages without enough free segments.
351df4b2 412 */
7f3037a5 413 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 414 mutex_lock(&sbi->gc_mutex);
e066b83c 415 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 416 }
417}
418
4660f9c0
JK		 419void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
420{
1dcc336b 421 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 422 if (!available_free_memory(sbi, EXTENT_CACHE))
423 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 424
1b38dc8e
JK		 425 /* check the # of cached NAT entries */
426 if (!available_free_memory(sbi, NAT_ENTRIES))
427 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
428
31696580 429 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 430 try_to_free_nids(sbi, MAX_FREE_NIDS);
431 else
22ad0b6a 432 build_free_nids(sbi, false, false);
31696580 433
1c0f4bf5 434 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 435 return;
31696580 436
1b38dc8e
JK		 437 /* checkpoint is the only way to shrink partial cached entries */
438 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 439 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 440 excess_prefree_segs(sbi) ||
441 excess_dirty_nats(sbi) ||
f455c8a5 442 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 443 if (test_opt(sbi, DATA_FLUSH)) {
444 struct blk_plug plug;
445
446 blk_start_plug(&plug);
36b35a0d 447 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 448 blk_finish_plug(&plug);
449 }
4660f9c0 450 f2fs_sync_fs(sbi->sb, true);
42190d2a 451 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 452 }
4660f9c0
JK		 453}
454
20fda56b
KM		 455static int __submit_flush_wait(struct f2fs_sb_info *sbi,
456 struct block_device *bdev)
3c62be17
JK		 457{
458 struct bio *bio = f2fs_bio_alloc(0);
459 int ret;
460
3adc5fcb 461 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3c62be17
JK		 462 bio->bi_bdev = bdev;
463 ret = submit_bio_wait(bio);
464 bio_put(bio);
20fda56b
KM		 465
466 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
467 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 468 return ret;
469}
470
471static int submit_flush_wait(struct f2fs_sb_info *sbi)
472{
20fda56b 473 int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev);
3c62be17
JK		 474 int i;
475
20fda56b
KM		 476 if (!sbi->s_ndevs || ret)
477 return ret;
478
479 for (i = 1; i < sbi->s_ndevs; i++) {
480 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
481 if (ret)
482 break;
3c62be17
JK		 483 }
484 return ret;
485}
486
2163d198 487static int issue_flush_thread(void *data)
6b4afdd7
JK		 488{
489 struct f2fs_sb_info *sbi = data;
b01a9201 490 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 491 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 492repeat:
493 if (kthread_should_stop())
494 return 0;
495
dc6febb6
CY		 496 sb_start_intwrite(sbi->sb);
497
721bd4d5 498 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 499 struct flush_cmd *cmd, *next;
500 int ret;
501
721bd4d5
GZ		 502 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
503 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
504
3c62be17 505 ret = submit_flush_wait(sbi);
8b8dd65f
CY		 506 atomic_inc(&fcc->issued_flush);
507
721bd4d5
GZ		 508 llist_for_each_entry_safe(cmd, next,
509 fcc->dispatch_list, llnode) {
6b4afdd7 510 cmd->ret = ret;
6b4afdd7
JK		 511 complete(&cmd->wait);
512 }
a688b9d9 513 fcc->dispatch_list = NULL;
6b4afdd7
JK		 514 }
515
dc6febb6
CY		 516 sb_end_intwrite(sbi->sb);
517
a688b9d9 518 wait_event_interruptible(*q,
721bd4d5 519 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 520 goto repeat;
521}
522
523int f2fs_issue_flush(struct f2fs_sb_info *sbi)
524{
b01a9201 525 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 526 struct flush_cmd cmd;
8b8dd65f 527 int ret;
6b4afdd7 528
0f7b2abd
JK		 529 if (test_opt(sbi, NOBARRIER))
530 return 0;
531
8b8dd65f
CY		 532 if (!test_opt(sbi, FLUSH_MERGE)) {
533 ret = submit_flush_wait(sbi);
534 atomic_inc(&fcc->issued_flush);
535 return ret;
536 }
740432f8 537
8b8dd65f
CY		 538 if (!atomic_read(&fcc->issing_flush)) {
539 atomic_inc(&fcc->issing_flush);
3c62be17 540 ret = submit_flush_wait(sbi);
8b8dd65f
CY		 541 atomic_dec(&fcc->issing_flush);
542
543 atomic_inc(&fcc->issued_flush);
740432f8
JK		 544 return ret;
545 }
6b4afdd7 546
adf8d90b 547 init_completion(&cmd.wait);
6b4afdd7 548
8b8dd65f 549 atomic_inc(&fcc->issing_flush);
721bd4d5 550 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 551
6f890df0
CY		 552 /* update issue_list before we wake up issue_flush thread */
553 smp_mb();
554
555 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 556 wake_up(&fcc->flush_wait_queue);
6b4afdd7 557
5eba8c5d
JK		 558 if (fcc->f2fs_issue_flush) {
559 wait_for_completion(&cmd.wait);
8b8dd65f 560 atomic_dec(&fcc->issing_flush);
5eba8c5d
JK		 561 } else {
562 llist_del_all(&fcc->issue_list);
8b8dd65f 563 atomic_set(&fcc->issing_flush, 0);
5eba8c5d 564 }
adf8d90b
CY		 565
566 return cmd.ret;
6b4afdd7
JK		 567}
568
2163d198
GZ		 569int create_flush_cmd_control(struct f2fs_sb_info *sbi)
570{
571 dev_t dev = sbi->sb->s_bdev->bd_dev;
572 struct flush_cmd_control *fcc;
573 int err = 0;
574
b01a9201
JK		 575 if (SM_I(sbi)->fcc_info) {
576 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 577 if (fcc->f2fs_issue_flush)
578 return err;
5eba8c5d
JK		 579 goto init_thread;
580 }
581
2163d198
GZ		 582 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
583 if (!fcc)
584 return -ENOMEM;
8b8dd65f
CY		 585 atomic_set(&fcc->issued_flush, 0);
586 atomic_set(&fcc->issing_flush, 0);
2163d198 587 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 588 init_llist_head(&fcc->issue_list);
b01a9201 589 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 590 if (!test_opt(sbi, FLUSH_MERGE))
591 return err;
592
5eba8c5d 593init_thread:
2163d198
GZ		 594 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
595 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
596 if (IS_ERR(fcc->f2fs_issue_flush)) {
597 err = PTR_ERR(fcc->f2fs_issue_flush);
598 kfree(fcc);
b01a9201 599 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 600 return err;
601 }
2163d198
GZ		 602
603 return err;
604}
605
5eba8c5d 606void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 607{
b01a9201 608 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 609
5eba8c5d
JK		 610 if (fcc && fcc->f2fs_issue_flush) {
611 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
612
613 fcc->f2fs_issue_flush = NULL;
614 kthread_stop(flush_thread);
615 }
616 if (free) {
617 kfree(fcc);
b01a9201 618 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 619 }
2163d198
GZ		 620}
621
351df4b2
JK		 622static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
623 enum dirty_type dirty_type)
624{
625 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
626
627 /* need not be added */
628 if (IS_CURSEG(sbi, segno))
629 return;
630
631 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
632 dirty_i->nr_dirty[dirty_type]++;
633
634 if (dirty_type == DIRTY) {
635 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 636 enum dirty_type t = sentry->type;
b2f2c390 637
ec325b52
JK		 638 if (unlikely(t >= DIRTY)) {
639 f2fs_bug_on(sbi, 1);
640 return;
641 }
4625d6aa
CL		 642 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
643 dirty_i->nr_dirty[t]++;
351df4b2
JK		 644 }
645}
646
647static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
648 enum dirty_type dirty_type)
649{
650 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
651
652 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
653 dirty_i->nr_dirty[dirty_type]--;
654
655 if (dirty_type == DIRTY) {
4625d6aa
CL		 656 struct seg_entry *sentry = get_seg_entry(sbi, segno);
657 enum dirty_type t = sentry->type;
658
659 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
660 dirty_i->nr_dirty[t]--;
b2f2c390 661
302bd348 662 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 663 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 664 dirty_i->victim_secmap);
351df4b2
JK		 665 }
666}
667
0a8165d7 668/*
351df4b2
JK		 669 * Should not occur error such as -ENOMEM.
670 * Adding dirty entry into seglist is not critical operation.
671 * If a given segment is one of current working segments, it won't be added.
672 */
8d8451af 673static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 674{
675 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
676 unsigned short valid_blocks;
677
678 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
679 return;
680
681 mutex_lock(&dirty_i->seglist_lock);
682
302bd348 683 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 684
685 if (valid_blocks == 0) {
686 __locate_dirty_segment(sbi, segno, PRE);
687 __remove_dirty_segment(sbi, segno, DIRTY);
688 } else if (valid_blocks < sbi->blocks_per_seg) {
689 __locate_dirty_segment(sbi, segno, DIRTY);
690 } else {
691 /* Recovery routine with SSR needs this */
692 __remove_dirty_segment(sbi, segno, DIRTY);
693 }
694
695 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 696}
697
004b6862 698static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 699 struct block_device *bdev, block_t lstart,
700 block_t start, block_t len)
275b66b0 701{
0b54fb84 702 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 703 struct list_head *pend_list;
b01a9201 704 struct discard_cmd *dc;
275b66b0 705
ba48a33e
CY		 706 f2fs_bug_on(sbi, !len);
707
708 pend_list = &dcc->pend_list[plist_idx(len)];
709
b01a9201
JK		 710 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
711 INIT_LIST_HEAD(&dc->list);
c81abe34 712 dc->bdev = bdev;
b01a9201 713 dc->lstart = lstart;
c81abe34 714 dc->start = start;
b01a9201 715 dc->len = len;
ec9895ad 716 dc->ref = 0;
15469963 717 dc->state = D_PREP;
c81abe34 718 dc->error = 0;
b01a9201 719 init_completion(&dc->wait);
22d375dd 720 list_add_tail(&dc->list, pend_list);
5f32366a 721 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 722 dcc->undiscard_blks += len;
004b6862
CY		 723
724 return dc;
725}
726
727static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
728 struct block_device *bdev, block_t lstart,
729 block_t start, block_t len,
730 struct rb_node *parent, struct rb_node **p)
731{
732 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
733 struct discard_cmd *dc;
734
735 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
736
737 rb_link_node(&dc->rb_node, parent, p);
738 rb_insert_color(&dc->rb_node, &dcc->root);
739
740 return dc;
15469963
JK		 741}
742
004b6862
CY		 743static void __detach_discard_cmd(struct discard_cmd_control *dcc,
744 struct discard_cmd *dc)
15469963 745{
dcc9165d 746 if (dc->state == D_DONE)
004b6862
CY		 747 atomic_dec(&dcc->issing_discard);
748
749 list_del(&dc->list);
750 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 751 dcc->undiscard_blks -= dc->len;
004b6862
CY		 752
753 kmem_cache_free(discard_cmd_slab, dc);
754
755 atomic_dec(&dcc->discard_cmd_cnt);
756}
757
758static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
759 struct discard_cmd *dc)
760{
761 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 762
d9703d90
CY		 763 f2fs_bug_on(sbi, dc->ref);
764
c81abe34
JK		 765 if (dc->error == -EOPNOTSUPP)
766 dc->error = 0;
15469963 767
c81abe34 768 if (dc->error)
15469963 769 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 770 "Issue discard(%u, %u, %u) failed, ret: %d",
771 dc->lstart, dc->start, dc->len, dc->error);
004b6862 772 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 773}
774
c81abe34
JK		 775static void f2fs_submit_discard_endio(struct bio *bio)
776{
777 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
778
4e4cbee9 779 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 780 dc->state = D_DONE;
e31b9821 781 complete_all(&dc->wait);
c81abe34
JK		 782 bio_put(bio);
783}
784
6915ea9d
CY		 785void __check_sit_bitmap(struct f2fs_sb_info *sbi,
786 block_t start, block_t end)
787{
788#ifdef CONFIG_F2FS_CHECK_FS
789 struct seg_entry *sentry;
790 unsigned int segno;
791 block_t blk = start;
792 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
793 unsigned long *map;
794
795 while (blk < end) {
796 segno = GET_SEGNO(sbi, blk);
797 sentry = get_seg_entry(sbi, segno);
798 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
799
008396e1
YS		 800 if (end < START_BLOCK(sbi, segno + 1))
801 size = GET_BLKOFF_FROM_SEG0(sbi, end);
802 else
803 size = max_blocks;
6915ea9d
CY		 804 map = (unsigned long *)(sentry->cur_valid_map);
805 offset = __find_rev_next_bit(map, size, offset);
806 f2fs_bug_on(sbi, offset != size);
008396e1 807 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 808 }
809#endif
810}
811
c81abe34
JK		 812/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
813static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
814 struct discard_cmd *dc)
815{
816 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
817 struct bio *bio = NULL;
818
819 if (dc->state != D_PREP)
820 return;
821
0243a5f9
CY		 822 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
823
c81abe34
JK		 824 dc->error = __blkdev_issue_discard(dc->bdev,
825 SECTOR_FROM_BLOCK(dc->start),
826 SECTOR_FROM_BLOCK(dc->len),
827 GFP_NOFS, 0, &bio);
828 if (!dc->error) {
829 /* should keep before submission to avoid D_DONE right away */
830 dc->state = D_SUBMIT;
8b8dd65f
CY		 831 atomic_inc(&dcc->issued_discard);
832 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 833 if (bio) {
834 bio->bi_private = dc;
835 bio->bi_end_io = f2fs_submit_discard_endio;
836 bio->bi_opf |= REQ_SYNC;
837 submit_bio(bio);
46f84c2c 838 list_move_tail(&dc->list, &dcc->wait_list);
6915ea9d 839 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 840
841 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 842 }
843 } else {
844 __remove_discard_cmd(sbi, dc);
845 }
846}
847
004b6862
CY		 848static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
849 struct block_device *bdev, block_t lstart,
850 block_t start, block_t len,
851 struct rb_node **insert_p,
852 struct rb_node *insert_parent)
c81abe34 853{
004b6862
CY		 854 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
855 struct rb_node **p = &dcc->root.rb_node;
856 struct rb_node *parent = NULL;
857 struct discard_cmd *dc = NULL;
858
859 if (insert_p && insert_parent) {
860 parent = insert_parent;
861 p = insert_p;
862 goto do_insert;
863 }
c81abe34 864
004b6862
CY		 865 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
866do_insert:
867 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
868 if (!dc)
869 return NULL;
c81abe34 870
004b6862 871 return dc;
c81abe34
JK		 872}
873
ba48a33e
CY		 874static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
875 struct discard_cmd *dc)
876{
877 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
878}
879
3d6a650f
YH		 880static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
881 struct discard_cmd *dc, block_t blkaddr)
882{
ba48a33e 883 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 884 struct discard_info di = dc->di;
885 bool modified = false;
3d6a650f 886
004b6862 887 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 888 __remove_discard_cmd(sbi, dc);
889 return;
890 }
891
d84d1cbd
CY		 892 dcc->undiscard_blks -= di.len;
893
004b6862 894 if (blkaddr > di.lstart) {
3d6a650f 895 dc->len = blkaddr - dc->lstart;
d84d1cbd 896 dcc->undiscard_blks += dc->len;
ba48a33e 897 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 898 modified = true;
899 }
900
901 if (blkaddr < di.lstart + di.len - 1) {
902 if (modified) {
903 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
904 di.start + blkaddr + 1 - di.lstart,
905 di.lstart + di.len - 1 - blkaddr,
906 NULL, NULL);
907 } else {
908 dc->lstart++;
909 dc->len--;
910 dc->start++;
d84d1cbd 911 dcc->undiscard_blks += dc->len;
ba48a33e 912 __relocate_discard_cmd(dcc, dc);
004b6862 913 }
3d6a650f
YH		 914 }
915}
916
004b6862
CY		 917static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
918 struct block_device *bdev, block_t lstart,
919 block_t start, block_t len)
275b66b0 920{
0b54fb84 921 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 922 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
923 struct discard_cmd *dc;
924 struct discard_info di = {0};
925 struct rb_node **insert_p = NULL, *insert_parent = NULL;
926 block_t end = lstart + len;
275b66b0 927
15469963 928 mutex_lock(&dcc->cmd_lock);
40465257 929
004b6862
CY		 930 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
931 NULL, lstart,
932 (struct rb_entry **)&prev_dc,
933 (struct rb_entry **)&next_dc,
934 &insert_p, &insert_parent, true);
935 if (dc)
936 prev_dc = dc;
937
938 if (!prev_dc) {
939 di.lstart = lstart;
940 di.len = next_dc ? next_dc->lstart - lstart : len;
941 di.len = min(di.len, len);
942 di.start = start;
22d375dd 943 }
15469963 944
004b6862
CY		 945 while (1) {
946 struct rb_node *node;
947 bool merged = false;
948 struct discard_cmd *tdc = NULL;
949
950 if (prev_dc) {
951 di.lstart = prev_dc->lstart + prev_dc->len;
952 if (di.lstart < lstart)
953 di.lstart = lstart;
954 if (di.lstart >= end)
955 break;
956
957 if (!next_dc || next_dc->lstart > end)
958 di.len = end - di.lstart;
959 else
960 di.len = next_dc->lstart - di.lstart;
961 di.start = start + di.lstart - lstart;
962 }
963
964 if (!di.len)
965 goto next;
966
967 if (prev_dc && prev_dc->state == D_PREP &&
968 prev_dc->bdev == bdev &&
969 __is_discard_back_mergeable(&di, &prev_dc->di)) {
970 prev_dc->di.len += di.len;
d84d1cbd 971 dcc->undiscard_blks += di.len;
ba48a33e 972 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 973 di = prev_dc->di;
974 tdc = prev_dc;
975 merged = true;
976 }
977
978 if (next_dc && next_dc->state == D_PREP &&
979 next_dc->bdev == bdev &&
980 __is_discard_front_mergeable(&di, &next_dc->di)) {
981 next_dc->di.lstart = di.lstart;
982 next_dc->di.len += di.len;
983 next_dc->di.start = di.start;
d84d1cbd 984 dcc->undiscard_blks += di.len;
ba48a33e 985 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 986 if (tdc)
987 __remove_discard_cmd(sbi, tdc);
004b6862 988 merged = true;
4e6a8d9b 989 }
004b6862 990
df0f6b44 991 if (!merged) {
004b6862
CY		 992 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
993 di.len, NULL, NULL);
df0f6b44 994 }
004b6862
CY		 995 next:
996 prev_dc = next_dc;
997 if (!prev_dc)
998 break;
999
1000 node = rb_next(&prev_dc->rb_node);
1001 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1002 }
1003
1004 mutex_unlock(&dcc->cmd_lock);
1005}
1006
1007static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1008 struct block_device *bdev, block_t blkstart, block_t blklen)
1009{
1010 block_t lblkstart = blkstart;
1011
0243a5f9 1012 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1013
1014 if (sbi->s_ndevs) {
1015 int devi = f2fs_target_device_index(sbi, blkstart);
1016
1017 blkstart -= FDEV(devi).start_blk;
1018 }
1019 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1020 return 0;
1021}
1022
969d1b18 1023static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
bd5b0738
CY		 1024{
1025 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1026 struct list_head *pend_list;
1027 struct discard_cmd *dc, *tmp;
1028 struct blk_plug plug;
969d1b18
CY		 1029 int iter = 0, issued = 0;
1030 int i;
bd5b0738
CY		 1031
1032 mutex_lock(&dcc->cmd_lock);
963932a9
YH		 1033 f2fs_bug_on(sbi,
1034 !__check_rb_tree_consistence(sbi, &dcc->root));
bd5b0738 1035 blk_start_plug(&plug);
969d1b18
CY		 1036 for (i = MAX_PLIST_NUM - 1;
1037 i >= 0 && plist_issue(dcc->pend_list_tag[i]); i--) {
bd5b0738
CY		 1038 pend_list = &dcc->pend_list[i];
1039 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1040 f2fs_bug_on(sbi, dc->state != D_PREP);
1041
969d1b18
CY		 1042 /* Hurry up to finish fstrim */
1043 if (dcc->pend_list_tag[i] & P_TRIM) {
1044 __submit_discard_cmd(sbi, dc);
1045 issued++;
1046 continue;
1047 }
1048
1049 if (!issue_cond || is_idle(sbi)) {
1050 issued++;
bd5b0738 1051 __submit_discard_cmd(sbi, dc);
969d1b18 1052 }
bd5b0738
CY		 1053 if (issue_cond && iter++ > DISCARD_ISSUE_RATE)
1054 goto out;
1055 }
969d1b18
CY		 1056 if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM)
1057 dcc->pend_list_tag[i] &= (~P_TRIM);
bd5b0738
CY		 1058 }
1059out:
1060 blk_finish_plug(&plug);
1061 mutex_unlock(&dcc->cmd_lock);
969d1b18
CY		 1062
1063 return issued;
1064}
1065
1066static void __drop_discard_cmd(struct f2fs_sb_info *sbi)
1067{
1068 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1069 struct list_head *pend_list;
1070 struct discard_cmd *dc, *tmp;
1071 int i;
1072
1073 mutex_lock(&dcc->cmd_lock);
1074 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1075 pend_list = &dcc->pend_list[i];
1076 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1077 f2fs_bug_on(sbi, dc->state != D_PREP);
1078 __remove_discard_cmd(sbi, dc);
1079 }
1080 }
1081 mutex_unlock(&dcc->cmd_lock);
bd5b0738
CY		 1082}
1083
2a510c00
CY		 1084static void __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1085 struct discard_cmd *dc)
1086{
1087 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1088
1089 wait_for_completion_io(&dc->wait);
1090 mutex_lock(&dcc->cmd_lock);
1091 f2fs_bug_on(sbi, dc->state != D_DONE);
1092 dc->ref--;
1093 if (!dc->ref)
1094 __remove_discard_cmd(sbi, dc);
1095 mutex_unlock(&dcc->cmd_lock);
1096}
1097
63a94fa1
CY		 1098static void __wait_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1099{
1100 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1101 struct list_head *wait_list = &(dcc->wait_list);
1102 struct discard_cmd *dc, *tmp;
6afae633
CY		 1103 bool need_wait;
1104
1105next:
1106 need_wait = false;
63a94fa1
CY		 1107
1108 mutex_lock(&dcc->cmd_lock);
1109 list_for_each_entry_safe(dc, tmp, wait_list, list) {
6afae633 1110 if (!wait_cond || (dc->state == D_DONE && !dc->ref)) {
63a94fa1
CY		 1111 wait_for_completion_io(&dc->wait);
1112 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1113 } else {
1114 dc->ref++;
1115 need_wait = true;
1116 break;
63a94fa1
CY		 1117 }
1118 }
1119 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1120
1121 if (need_wait) {
2a510c00 1122 __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1123 goto next;
1124 }
63a94fa1
CY		 1125}
1126
004b6862
CY		 1127/* This should be covered by global mutex, &sit_i->sentry_lock */
1128void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1129{
1130 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1131 struct discard_cmd *dc;
ec9895ad 1132 bool need_wait = false;
004b6862
CY		 1133
1134 mutex_lock(&dcc->cmd_lock);
004b6862
CY		 1135 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1136 if (dc) {
ec9895ad
CY		 1137 if (dc->state == D_PREP) {
1138 __punch_discard_cmd(sbi, dc, blkaddr);
1139 } else {
1140 dc->ref++;
1141 need_wait = true;
1142 }
275b66b0 1143 }
d431413f 1144 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1145
2a510c00
CY		 1146 if (need_wait)
1147 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1148}
1149
cce13252
CY		 1150void stop_discard_thread(struct f2fs_sb_info *sbi)
1151{
1152 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1153
1154 if (dcc && dcc->f2fs_issue_discard) {
1155 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1156
1157 dcc->f2fs_issue_discard = NULL;
1158 kthread_stop(discard_thread);
ec9895ad 1159 }
d431413f
CY		 1160}
1161
1162/* This comes from f2fs_put_super */
1163void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1164{
bd5b0738 1165 __issue_discard_cmd(sbi, false);
969d1b18 1166 __drop_discard_cmd(sbi);
63a94fa1 1167 __wait_discard_cmd(sbi, false);
275b66b0
CY		 1168}
1169
969d1b18
CY		 1170static void mark_discard_range_all(struct f2fs_sb_info *sbi)
1171{
1172 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1173 int i;
1174
1175 mutex_lock(&dcc->cmd_lock);
1176 for (i = 0; i < MAX_PLIST_NUM; i++)
1177 dcc->pend_list_tag[i] |= P_TRIM;
1178 mutex_unlock(&dcc->cmd_lock);
1179}
1180
15469963
JK		 1181static int issue_discard_thread(void *data)
1182{
1183 struct f2fs_sb_info *sbi = data;
1184 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1185 wait_queue_head_t *q = &dcc->discard_wait_queue;
969d1b18
CY		 1186 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1187 int issued;
15469963 1188
1d7be270 1189 set_freezable();
15469963 1190
1d7be270 1191 do {
969d1b18
CY		 1192 wait_event_interruptible_timeout(*q,
1193 kthread_should_stop() || freezing(current) ||
1194 dcc->discard_wake,
1195 msecs_to_jiffies(wait_ms));
1d7be270
JK		 1196 if (try_to_freeze())
1197 continue;
1198 if (kthread_should_stop())
1199 return 0;
15469963 1200
5f656541 1201 if (dcc->discard_wake) {
969d1b18 1202 dcc->discard_wake = 0;
5f656541
JK		 1203 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1204 mark_discard_range_all(sbi);
1205 }
969d1b18 1206
dc6febb6
CY		 1207 sb_start_intwrite(sbi->sb);
1208
969d1b18
CY		 1209 issued = __issue_discard_cmd(sbi, true);
1210 if (issued) {
1211 __wait_discard_cmd(sbi, true);
1212 wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1213 } else {
1214 wait_ms = DEF_MAX_DISCARD_ISSUE_TIME;
1215 }
1d7be270 1216
dc6febb6
CY		 1217 sb_end_intwrite(sbi->sb);
1218
1d7be270
JK		 1219 } while (!kthread_should_stop());
1220 return 0;
15469963
JK		 1221}
1222
f46e8809 1223#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1224static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1225 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1226{
92592285 1227 sector_t sector, nr_sects;
10a875f8 1228 block_t lblkstart = blkstart;
3c62be17
JK		 1229 int devi = 0;
1230
1231 if (sbi->s_ndevs) {
1232 devi = f2fs_target_device_index(sbi, blkstart);
1233 blkstart -= FDEV(devi).start_blk;
1234 }
f46e8809
DLM		 1235
1236 /*
1237 * We need to know the type of the zone: for conventional zones,
1238 * use regular discard if the drive supports it. For sequential
1239 * zones, reset the zone write pointer.
1240 */
3c62be17 1241 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1242
1243 case BLK_ZONE_TYPE_CONVENTIONAL:
1244 if (!blk_queue_discard(bdev_get_queue(bdev)))
1245 return 0;
c81abe34 1246 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1247 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1248 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1249 sector = SECTOR_FROM_BLOCK(blkstart);
1250 nr_sects = SECTOR_FROM_BLOCK(blklen);
1251
1252 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1253 nr_sects != bdev_zone_sectors(bdev)) {
1254 f2fs_msg(sbi->sb, KERN_INFO,
1255 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1256 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1257 blkstart, blklen);
1258 return -EIO;
1259 }
d50aaeec 1260 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1261 return blkdev_reset_zones(bdev, sector,
1262 nr_sects, GFP_NOFS);
1263 default:
1264 /* Unknown zone type: broken device ? */
1265 return -EIO;
1266 }
1267}
1268#endif
1269
3c62be17
JK		 1270static int __issue_discard_async(struct f2fs_sb_info *sbi,
1271 struct block_device *bdev, block_t blkstart, block_t blklen)
1272{
1273#ifdef CONFIG_BLK_DEV_ZONED
1274 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1275 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1276 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1277#endif
c81abe34 1278 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1279}
1280
1e87a78d 1281static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1282 block_t blkstart, block_t blklen)
1283{
3c62be17
JK		 1284 sector_t start = blkstart, len = 0;
1285 struct block_device *bdev;
a66cdd98
JK		 1286 struct seg_entry *se;
1287 unsigned int offset;
1288 block_t i;
3c62be17
JK		 1289 int err = 0;
1290
1291 bdev = f2fs_target_device(sbi, blkstart, NULL);
1292
1293 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1294 if (i != start) {
1295 struct block_device *bdev2 =
1296 f2fs_target_device(sbi, i, NULL);
1297
1298 if (bdev2 != bdev) {
1299 err = __issue_discard_async(sbi, bdev,
1300 start, len);
1301 if (err)
1302 return err;
1303 bdev = bdev2;
1304 start = i;
1305 len = 0;
1306 }
1307 }
a66cdd98 1308
a66cdd98
JK		 1309 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1310 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1311
1312 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1313 sbi->discard_blks--;
1314 }
f46e8809 1315
3c62be17
JK		 1316 if (len)
1317 err = __issue_discard_async(sbi, bdev, start, len);
1318 return err;
1e87a78d
JK		 1319}
1320
25290fa5
JK		 1321static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1322 bool check_only)
adf4983b 1323{
b2955550
JK		 1324 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1325 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1326 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1327 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1328 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1329 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1330 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1331 unsigned int start = 0, end = -1;
c473f1a9 1332 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1333 struct discard_entry *de = NULL;
46f84c2c 1334 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1335 int i;
1336
3e025740 1337 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1338 return false;
b2955550 1339
a66cdd98
JK		 1340 if (!force) {
1341 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1342 SM_I(sbi)->dcc_info->nr_discards >=
1343 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1344 return false;
4b2fecc8
JK		 1345 }
1346
b2955550
JK		 1347 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1348 for (i = 0; i < entries; i++)
a66cdd98 1349 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1350 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1351
0b54fb84
JK		 1352 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1353 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1354 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1355 if (start >= max_blocks)
1356 break;
1357
1358 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1359 if (force && start && end != max_blocks
1360 && (end - start) < cpc->trim_minlen)
1361 continue;
1362
25290fa5
JK		 1363 if (check_only)
1364 return true;
1365
a7eeb823
CY		 1366 if (!de) {
1367 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1368 GFP_F2FS_ZERO);
1369 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1370 list_add_tail(&de->list, head);
1371 }
1372
1373 for (i = start; i < end; i++)
1374 __set_bit_le(i, (void *)de->discard_map);
1375
1376 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1377 }
25290fa5 1378 return false;
b2955550
JK		 1379}
1380
4b2fecc8
JK		 1381void release_discard_addrs(struct f2fs_sb_info *sbi)
1382{
46f84c2c 1383 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1384 struct discard_entry *entry, *this;
1385
1386 /* drop caches */
1387 list_for_each_entry_safe(entry, this, head, list) {
1388 list_del(&entry->list);
1389 kmem_cache_free(discard_entry_slab, entry);
1390 }
1391}
1392
0a8165d7 1393/*
351df4b2
JK		 1394 * Should call clear_prefree_segments after checkpoint is done.
1395 */
1396static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1397{
1398 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1399 unsigned int segno;
351df4b2
JK		 1400
1401 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1402 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1403 __set_test_and_free(sbi, segno);
351df4b2
JK		 1404 mutex_unlock(&dirty_i->seglist_lock);
1405}
1406
836b5a63 1407void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1408{
969d1b18
CY		 1409 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1410 struct list_head *head = &dcc->entry_list;
2d7b822a 1411 struct discard_entry *entry, *this;
351df4b2 1412 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1413 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1414 unsigned int start = 0, end = -1;
36abef4e 1415 unsigned int secno, start_segno;
c473f1a9 1416 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK		 1417
1418 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1419
351df4b2 1420 while (1) {
29e59c14 1421 int i;
7cd8558b
JK		 1422 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1423 if (start >= MAIN_SEGS(sbi))
351df4b2 1424 break;
7cd8558b
JK		 1425 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1426 start + 1);
29e59c14
CL		 1427
1428 for (i = start; i < end; i++)
1429 clear_bit(i, prefree_map);
1430
1431 dirty_i->nr_dirty[PRE] -= end - start;
1432
650d3c4e 1433 if (!test_opt(sbi, DISCARD))
29e59c14 1434 continue;
351df4b2 1435
650d3c4e
YH		 1436 if (force && start >= cpc->trim_start &&
1437 (end - 1) <= cpc->trim_end)
1438 continue;
1439
36abef4e
JK		 1440 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1441 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1442 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1443 continue;
1444 }
1445next:
4ddb1a4d
JK		 1446 secno = GET_SEC_FROM_SEG(sbi, start);
1447 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1448 if (!IS_CURSEC(sbi, secno) &&
302bd348 1449 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1450 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1451 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1452
1453 start = start_segno + sbi->segs_per_sec;
1454 if (start < end)
1455 goto next;
8b107f5b
JK		 1456 else
1457 end = start - 1;
351df4b2
JK		 1458 }
1459 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1460
1461 /* send small discards */
2d7b822a 1462 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1463 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1464 bool is_valid = test_bit_le(0, entry->discard_map);
1465
1466find_next:
1467 if (is_valid) {
1468 next_pos = find_next_zero_bit_le(entry->discard_map,
1469 sbi->blocks_per_seg, cur_pos);
1470 len = next_pos - cur_pos;
1471
acfd2810
DLM		 1472 if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1473 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1474 goto skip;
1475
1476 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1477 len);
1478 cpc->trimmed += len;
1479 total_len += len;
1480 } else {
1481 next_pos = find_next_bit_le(entry->discard_map,
1482 sbi->blocks_per_seg, cur_pos);
1483 }
836b5a63 1484skip:
a7eeb823
CY		 1485 cur_pos = next_pos;
1486 is_valid = !is_valid;
1487
1488 if (cur_pos < sbi->blocks_per_seg)
1489 goto find_next;
1490
b2955550 1491 list_del(&entry->list);
969d1b18 1492 dcc->nr_discards -= total_len;
b2955550
JK		 1493 kmem_cache_free(discard_entry_slab, entry);
1494 }
34e159da 1495
969d1b18
CY		 1496 dcc->discard_wake = 1;
1497 wake_up_interruptible_all(&dcc->discard_wait_queue);
351df4b2
JK		 1498}
1499
8ed59745 1500static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1501{
15469963 1502 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1503 struct discard_cmd_control *dcc;
ba48a33e 1504 int err = 0, i;
0b54fb84
JK		 1505
1506 if (SM_I(sbi)->dcc_info) {
1507 dcc = SM_I(sbi)->dcc_info;
1508 goto init_thread;
1509 }
1510
1511 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1512 if (!dcc)
1513 return -ENOMEM;
1514
969d1b18 1515 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1516 INIT_LIST_HEAD(&dcc->entry_list);
969d1b18 1517 for (i = 0; i < MAX_PLIST_NUM; i++) {
ba48a33e 1518 INIT_LIST_HEAD(&dcc->pend_list[i]);
969d1b18
CY		 1519 if (i >= dcc->discard_granularity - 1)
1520 dcc->pend_list_tag[i] |= P_ACTIVE;
1521 }
46f84c2c 1522 INIT_LIST_HEAD(&dcc->wait_list);
15469963 1523 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1524 atomic_set(&dcc->issued_discard, 0);
1525 atomic_set(&dcc->issing_discard, 0);
5f32366a 1526 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1527 dcc->nr_discards = 0;
d618ebaf 1528 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1529 dcc->undiscard_blks = 0;
004b6862 1530 dcc->root = RB_ROOT;
0b54fb84 1531
15469963 1532 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1533 SM_I(sbi)->dcc_info = dcc;
1534init_thread:
15469963
JK		 1535 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1536 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1537 if (IS_ERR(dcc->f2fs_issue_discard)) {
1538 err = PTR_ERR(dcc->f2fs_issue_discard);
1539 kfree(dcc);
1540 SM_I(sbi)->dcc_info = NULL;
1541 return err;
1542 }
1543
0b54fb84
JK		 1544 return err;
1545}
1546
f099405f 1547static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1548{
1549 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1550
f099405f
CY		 1551 if (!dcc)
1552 return;
1553
cce13252 1554 stop_discard_thread(sbi);
f099405f
CY		 1555
1556 kfree(dcc);
1557 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1558}
1559
184a5cd2 1560static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1561{
1562 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1563
1564 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1565 sit_i->dirty_sentries++;
184a5cd2
CY		 1566 return false;
1567 }
1568
1569 return true;
351df4b2
JK		 1570}
1571
1572static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1573 unsigned int segno, int modified)
1574{
1575 struct seg_entry *se = get_seg_entry(sbi, segno);
1576 se->type = type;
1577 if (modified)
1578 __mark_sit_entry_dirty(sbi, segno);
1579}
1580
1581static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1582{
1583 struct seg_entry *se;
1584 unsigned int segno, offset;
1585 long int new_vblocks;
6415fedc
YS		 1586 bool exist;
1587#ifdef CONFIG_F2FS_CHECK_FS
1588 bool mir_exist;
1589#endif
351df4b2
JK		 1590
1591 segno = GET_SEGNO(sbi, blkaddr);
1592
1593 se = get_seg_entry(sbi, segno);
1594 new_vblocks = se->valid_blocks + del;
491c0854 1595 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1596
9850cf4a 1597 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1598 (new_vblocks > sbi->blocks_per_seg)));
1599
1600 se->valid_blocks = new_vblocks;
1601 se->mtime = get_mtime(sbi);
1602 SIT_I(sbi)->max_mtime = se->mtime;
1603
1604 /* Update valid block bitmap */
1605 if (del > 0) {
6415fedc 1606 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1607#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1608 mir_exist = f2fs_test_and_set_bit(offset,
1609 se->cur_valid_map_mir);
1610 if (unlikely(exist != mir_exist)) {
1611 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1612 "when setting bitmap, blk:%u, old bit:%d",
1613 blkaddr, exist);
05796763 1614 f2fs_bug_on(sbi, 1);
6415fedc 1615 }
355e7891 1616#endif
6415fedc
YS		 1617 if (unlikely(exist)) {
1618 f2fs_msg(sbi->sb, KERN_ERR,
1619 "Bitmap was wrongly set, blk:%u", blkaddr);
1620 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1621 se->valid_blocks--;
1622 del = 0;
355e7891 1623 }
6415fedc 1624
3e025740
JK		 1625 if (f2fs_discard_en(sbi) &&
1626 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1627 sbi->discard_blks--;
720037f9
JK		 1628
1629 /* don't overwrite by SSR to keep node chain */
1630 if (se->type == CURSEG_WARM_NODE) {
1631 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1632 se->ckpt_valid_blocks++;
1633 }
351df4b2 1634 } else {
6415fedc 1635 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1636#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1637 mir_exist = f2fs_test_and_clear_bit(offset,
1638 se->cur_valid_map_mir);
1639 if (unlikely(exist != mir_exist)) {
1640 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1641 "when clearing bitmap, blk:%u, old bit:%d",
1642 blkaddr, exist);
05796763 1643 f2fs_bug_on(sbi, 1);
6415fedc 1644 }
355e7891 1645#endif
6415fedc
YS		 1646 if (unlikely(!exist)) {
1647 f2fs_msg(sbi->sb, KERN_ERR,
1648 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1649 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1650 se->valid_blocks++;
1651 del = 0;
355e7891 1652 }
6415fedc 1653
3e025740
JK		 1654 if (f2fs_discard_en(sbi) &&
1655 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1656 sbi->discard_blks++;
351df4b2
JK		 1657 }
1658 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1659 se->ckpt_valid_blocks += del;
1660
1661 __mark_sit_entry_dirty(sbi, segno);
1662
1663 /* update total number of valid blocks to be written in ckpt area */
1664 SIT_I(sbi)->written_valid_blocks += del;
1665
1666 if (sbi->segs_per_sec > 1)
1667 get_sec_entry(sbi, segno)->valid_blocks += del;
1668}
1669
5e443818 1670void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1671{
5e443818
JK		 1672 update_sit_entry(sbi, new, 1);
1673 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1674 update_sit_entry(sbi, old, -1);
1675
1676 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1677 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 1678}
1679
1680void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1681{
1682 unsigned int segno = GET_SEGNO(sbi, addr);
1683 struct sit_info *sit_i = SIT_I(sbi);
1684
9850cf4a 1685 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1686 if (addr == NEW_ADDR)
1687 return;
1688
1689 /* add it into sit main buffer */
1690 mutex_lock(&sit_i->sentry_lock);
1691
1692 update_sit_entry(sbi, addr, -1);
1693
1694 /* add it into dirty seglist */
1695 locate_dirty_segment(sbi, segno);
1696
1697 mutex_unlock(&sit_i->sentry_lock);
1698}
1699
6e2c64ad
JK		 1700bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1701{
1702 struct sit_info *sit_i = SIT_I(sbi);
1703 unsigned int segno, offset;
1704 struct seg_entry *se;
1705 bool is_cp = false;
1706
1707 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1708 return true;
1709
1710 mutex_lock(&sit_i->sentry_lock);
1711
1712 segno = GET_SEGNO(sbi, blkaddr);
1713 se = get_seg_entry(sbi, segno);
1714 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1715
1716 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1717 is_cp = true;
1718
1719 mutex_unlock(&sit_i->sentry_lock);
1720
1721 return is_cp;
1722}
1723
0a8165d7 1724/*
351df4b2
JK		 1725 * This function should be resided under the curseg_mutex lock
1726 */
1727static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1728 struct f2fs_summary *sum)
351df4b2
JK		 1729{
1730 struct curseg_info *curseg = CURSEG_I(sbi, type);
1731 void *addr = curseg->sum_blk;
e79efe3b 1732 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1733 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1734}
1735
0a8165d7 1736/*
351df4b2
JK		 1737 * Calculate the number of current summary pages for writing
1738 */
3fa06d7b 1739int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1740{
351df4b2 1741 int valid_sum_count = 0;
9a47938b 1742 int i, sum_in_page;
351df4b2
JK		 1743
1744 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1745 if (sbi->ckpt->alloc_type[i] == SSR)
1746 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1747 else {
1748 if (for_ra)
1749 valid_sum_count += le16_to_cpu(
1750 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1751 else
1752 valid_sum_count += curseg_blkoff(sbi, i);
1753 }
351df4b2
JK		 1754 }
1755
09cbfeaf 1756 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1757 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1758 if (valid_sum_count <= sum_in_page)
351df4b2 1759 return 1;
9a47938b 1760 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1761 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1762 return 2;
1763 return 3;
1764}
1765
0a8165d7 1766/*
351df4b2
JK		 1767 * Caller should put this summary page
1768 */
1769struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1770{
1771 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1772}
1773
381722d2 1774void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1775{
1776 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1777 void *dst = page_address(page);
1778
1779 if (src)
09cbfeaf 1780 memcpy(dst, src, PAGE_SIZE);
381722d2 1781 else
09cbfeaf 1782 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1783 set_page_dirty(page);
1784 f2fs_put_page(page, 1);
1785}
1786
381722d2
CY		 1787static void write_sum_page(struct f2fs_sb_info *sbi,
1788 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1789{
1790 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1791}
1792
b7ad7512
CY		 1793static void write_current_sum_page(struct f2fs_sb_info *sbi,
1794 int type, block_t blk_addr)
1795{
1796 struct curseg_info *curseg = CURSEG_I(sbi, type);
1797 struct page *page = grab_meta_page(sbi, blk_addr);
1798 struct f2fs_summary_block *src = curseg->sum_blk;
1799 struct f2fs_summary_block *dst;
1800
1801 dst = (struct f2fs_summary_block *)page_address(page);
1802
1803 mutex_lock(&curseg->curseg_mutex);
1804
1805 down_read(&curseg->journal_rwsem);
1806 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1807 up_read(&curseg->journal_rwsem);
1808
1809 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1810 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1811
1812 mutex_unlock(&curseg->curseg_mutex);
1813
1814 set_page_dirty(page);
1815 f2fs_put_page(page, 1);
1816}
1817
a7881893
JK		 1818static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1819{
1820 struct curseg_info *curseg = CURSEG_I(sbi, type);
1821 unsigned int segno = curseg->segno + 1;
1822 struct free_segmap_info *free_i = FREE_I(sbi);
1823
1824 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1825 return !test_bit(segno, free_i->free_segmap);
1826 return 0;
1827}
1828
0a8165d7 1829/*
351df4b2
JK		 1830 * Find a new segment from the free segments bitmap to right order
1831 * This function should be returned with success, otherwise BUG
1832 */
1833static void get_new_segment(struct f2fs_sb_info *sbi,
1834 unsigned int *newseg, bool new_sec, int dir)
1835{
1836 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1837 unsigned int segno, secno, zoneno;
7cd8558b 1838 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 1839 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1840 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 1841 unsigned int left_start = hint;
1842 bool init = true;
1843 int go_left = 0;
1844 int i;
1845
1a118ccf 1846 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1847
1848 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1849 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 1850 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
1851 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 1852 goto got_it;
1853 }
1854find_other_zone:
7cd8558b
JK		 1855 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1856 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1857 if (dir == ALLOC_RIGHT) {
1858 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1859 MAIN_SECS(sbi), 0);
1860 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1861 } else {
1862 go_left = 1;
1863 left_start = hint - 1;
1864 }
1865 }
1866 if (go_left == 0)
1867 goto skip_left;
1868
1869 while (test_bit(left_start, free_i->free_secmap)) {
1870 if (left_start > 0) {
1871 left_start--;
1872 continue;
1873 }
1874 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1875 MAIN_SECS(sbi), 0);
1876 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1877 break;
1878 }
1879 secno = left_start;
1880skip_left:
1881 hint = secno;
4ddb1a4d
JK		 1882 segno = GET_SEG_FROM_SEC(sbi, secno);
1883 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 1884
1885 /* give up on finding another zone */
1886 if (!init)
1887 goto got_it;
1888 if (sbi->secs_per_zone == 1)
1889 goto got_it;
1890 if (zoneno == old_zoneno)
1891 goto got_it;
1892 if (dir == ALLOC_LEFT) {
1893 if (!go_left && zoneno + 1 >= total_zones)
1894 goto got_it;
1895 if (go_left && zoneno == 0)
1896 goto got_it;
1897 }
1898 for (i = 0; i < NR_CURSEG_TYPE; i++)
1899 if (CURSEG_I(sbi, i)->zone == zoneno)
1900 break;
1901
1902 if (i < NR_CURSEG_TYPE) {
1903 /* zone is in user, try another */
1904 if (go_left)
1905 hint = zoneno * sbi->secs_per_zone - 1;
1906 else if (zoneno + 1 >= total_zones)
1907 hint = 0;
1908 else
1909 hint = (zoneno + 1) * sbi->secs_per_zone;
1910 init = false;
1911 goto find_other_zone;
1912 }
1913got_it:
1914 /* set it as dirty segment in free segmap */
9850cf4a 1915 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1916 __set_inuse(sbi, segno);
1917 *newseg = segno;
1a118ccf 1918 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1919}
1920
1921static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1922{
1923 struct curseg_info *curseg = CURSEG_I(sbi, type);
1924 struct summary_footer *sum_footer;
1925
1926 curseg->segno = curseg->next_segno;
4ddb1a4d 1927 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 1928 curseg->next_blkoff = 0;
1929 curseg->next_segno = NULL_SEGNO;
1930
1931 sum_footer = &(curseg->sum_blk->footer);
1932 memset(sum_footer, 0, sizeof(struct summary_footer));
1933 if (IS_DATASEG(type))
1934 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1935 if (IS_NODESEG(type))
1936 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1937 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1938}
1939
7a20b8a6
JK		 1940static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
1941{
a7881893
JK		 1942 /* if segs_per_sec is large than 1, we need to keep original policy. */
1943 if (sbi->segs_per_sec != 1)
1944 return CURSEG_I(sbi, type)->segno;
1945
7a20b8a6
JK		 1946 if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
1947 return 0;
1948
e066b83c
JK		 1949 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
1950 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
7a20b8a6
JK		 1951 return CURSEG_I(sbi, type)->segno;
1952}
1953
0a8165d7 1954/*
351df4b2
JK		 1955 * Allocate a current working segment.
1956 * This function always allocates a free segment in LFS manner.
1957 */
1958static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1959{
1960 struct curseg_info *curseg = CURSEG_I(sbi, type);
1961 unsigned int segno = curseg->segno;
1962 int dir = ALLOC_LEFT;
1963
1964 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1965 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1966 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1967 dir = ALLOC_RIGHT;
1968
1969 if (test_opt(sbi, NOHEAP))
1970 dir = ALLOC_RIGHT;
1971
7a20b8a6 1972 segno = __get_next_segno(sbi, type);
351df4b2
JK		 1973 get_new_segment(sbi, &segno, new_sec, dir);
1974 curseg->next_segno = segno;
1975 reset_curseg(sbi, type, 1);
1976 curseg->alloc_type = LFS;
1977}
1978
1979static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1980 struct curseg_info *seg, block_t start)
1981{
1982 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1983 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1984 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1985 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1986 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1987 int i, pos;
1988
1989 for (i = 0; i < entries; i++)
1990 target_map[i] = ckpt_map[i] | cur_map[i];
1991
1992 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1993
1994 seg->next_blkoff = pos;
351df4b2
JK		 1995}
1996
0a8165d7 1997/*
351df4b2
JK		 1998 * If a segment is written by LFS manner, next block offset is just obtained
1999 * by increasing the current block offset. However, if a segment is written by
2000 * SSR manner, next block offset obtained by calling __next_free_blkoff
2001 */
2002static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2003 struct curseg_info *seg)
2004{
2005 if (seg->alloc_type == SSR)
2006 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2007 else
2008 seg->next_blkoff++;
2009}
2010
0a8165d7 2011/*
e1c42045 2012 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2013 * manner, so it should recover the existing segment information of valid blocks
2014 */
2015static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
2016{
2017 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2018 struct curseg_info *curseg = CURSEG_I(sbi, type);
2019 unsigned int new_segno = curseg->next_segno;
2020 struct f2fs_summary_block *sum_node;
2021 struct page *sum_page;
2022
2023 write_sum_page(sbi, curseg->sum_blk,
2024 GET_SUM_BLOCK(sbi, curseg->segno));
2025 __set_test_and_inuse(sbi, new_segno);
2026
2027 mutex_lock(&dirty_i->seglist_lock);
2028 __remove_dirty_segment(sbi, new_segno, PRE);
2029 __remove_dirty_segment(sbi, new_segno, DIRTY);
2030 mutex_unlock(&dirty_i->seglist_lock);
2031
2032 reset_curseg(sbi, type, 1);
2033 curseg->alloc_type = SSR;
2034 __next_free_blkoff(sbi, curseg, 0);
2035
2036 if (reuse) {
2037 sum_page = get_sum_page(sbi, new_segno);
2038 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2039 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2040 f2fs_put_page(sum_page, 1);
2041 }
2042}
2043
43727527
JK		 2044static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2045{
2046 struct curseg_info *curseg = CURSEG_I(sbi, type);
2047 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2048 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2049 int i, cnt;
2050 bool reversed = false;
c192f7a4
JK		 2051
2052 /* need_SSR() already forces to do this */
e066b83c
JK		 2053 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2054 curseg->next_segno = segno;
c192f7a4 2055 return 1;
e066b83c 2056 }
43727527 2057
70d625cb
JK		 2058 /* For node segments, let's do SSR more intensively */
2059 if (IS_NODESEG(type)) {
d27c3d89
CY		 2060 if (type >= CURSEG_WARM_NODE) {
2061 reversed = true;
2062 i = CURSEG_COLD_NODE;
2063 } else {
2064 i = CURSEG_HOT_NODE;
2065 }
2066 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2067 } else {
d27c3d89
CY		 2068 if (type >= CURSEG_WARM_DATA) {
2069 reversed = true;
2070 i = CURSEG_COLD_DATA;
2071 } else {
2072 i = CURSEG_HOT_DATA;
2073 }
2074 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2075 }
43727527 2076
d27c3d89 2077 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2078 if (i == type)
2079 continue;
e066b83c
JK		 2080 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2081 curseg->next_segno = segno;
43727527 2082 return 1;
e066b83c 2083 }
c192f7a4 2084 }
43727527
JK		 2085 return 0;
2086}
2087
351df4b2
JK		 2088/*
2089 * flush out current segment and replace it with new segment
2090 * This function should be returned with success, otherwise BUG
2091 */
2092static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2093 int type, bool force)
2094{
a7881893
JK		 2095 struct curseg_info *curseg = CURSEG_I(sbi, type);
2096
7b405275 2097 if (force)
351df4b2 2098 new_curseg(sbi, type, true);
5b6c6be2
JK		 2099 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2100 type == CURSEG_WARM_NODE)
351df4b2 2101 new_curseg(sbi, type, false);
a7881893
JK		 2102 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2103 new_curseg(sbi, type, false);
351df4b2
JK		 2104 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2105 change_curseg(sbi, type, true);
2106 else
2107 new_curseg(sbi, type, false);
dcdfff65 2108
a7881893 2109 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2110}
2111
2112void allocate_new_segments(struct f2fs_sb_info *sbi)
2113{
6ae1be13
JK		 2114 struct curseg_info *curseg;
2115 unsigned int old_segno;
351df4b2
JK		 2116 int i;
2117
6ae1be13
JK		 2118 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2119 curseg = CURSEG_I(sbi, i);
2120 old_segno = curseg->segno;
2121 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2122 locate_dirty_segment(sbi, old_segno);
2123 }
351df4b2
JK		 2124}
2125
2126static const struct segment_allocation default_salloc_ops = {
2127 .allocate_segment = allocate_segment_by_default,
2128};
2129
25290fa5
JK		 2130bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2131{
2132 __u64 trim_start = cpc->trim_start;
2133 bool has_candidate = false;
2134
2135 mutex_lock(&SIT_I(sbi)->sentry_lock);
2136 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2137 if (add_discard_addrs(sbi, cpc, true)) {
2138 has_candidate = true;
2139 break;
2140 }
2141 }
2142 mutex_unlock(&SIT_I(sbi)->sentry_lock);
2143
2144 cpc->trim_start = trim_start;
2145 return has_candidate;
2146}
2147
4b2fecc8
JK		 2148int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2149{
f7ef9b83
JK		 2150 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2151 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 2152 unsigned int start_segno, end_segno;
2153 struct cp_control cpc;
c34f42e2 2154 int err = 0;
4b2fecc8 2155
836b5a63 2156 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2157 return -EINVAL;
2158
9bd27ae4 2159 cpc.trimmed = 0;
7cd8558b 2160 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 2161 goto out;
2162
ed214a11
YH		 2163 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2164 f2fs_msg(sbi->sb, KERN_WARNING,
2165 "Found FS corruption, run fsck to fix.");
2166 goto out;
2167 }
2168
4b2fecc8 2169 /* start/end segment number in main_area */
7cd8558b
JK		 2170 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2171 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2172 GET_SEGNO(sbi, end);
4b2fecc8 2173 cpc.reason = CP_DISCARD;
836b5a63 2174 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 2175
2176 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 2177 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
2178 cpc.trim_start = start_segno;
a66cdd98
JK		 2179
2180 if (sbi->discard_blks == 0)
2181 break;
2182 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2183 cpc.trim_end = end_segno;
2184 else
2185 cpc.trim_end = min_t(unsigned int,
2186 rounddown(start_segno +
bba681cb
JK		 2187 BATCHED_TRIM_SEGMENTS(sbi),
2188 sbi->segs_per_sec) - 1, end_segno);
2189
2190 mutex_lock(&sbi->gc_mutex);
c34f42e2 2191 err = write_checkpoint(sbi, &cpc);
bba681cb 2192 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 2193 if (err)
2194 break;
74fa5f3d
CY		 2195
2196 schedule();
bba681cb 2197 }
969d1b18
CY		 2198 /* It's time to issue all the filed discards */
2199 mark_discard_range_all(sbi);
4b2fecc8 2200out:
f7ef9b83 2201 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 2202 return err;
4b2fecc8
JK		 2203}
2204
351df4b2
JK		 2205static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2206{
2207 struct curseg_info *curseg = CURSEG_I(sbi, type);
2208 if (curseg->next_blkoff < sbi->blocks_per_seg)
2209 return true;
2210 return false;
2211}
2212
81377bd6 2213static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2214{
81377bd6 2215 if (fio->type == DATA)
351df4b2
JK		 2216 return CURSEG_HOT_DATA;
2217 else
2218 return CURSEG_HOT_NODE;
2219}
2220
81377bd6 2221static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2222{
81377bd6
JK		 2223 if (fio->type == DATA) {
2224 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2225
2226 if (S_ISDIR(inode->i_mode))
2227 return CURSEG_HOT_DATA;
2228 else
2229 return CURSEG_COLD_DATA;
2230 } else {
81377bd6 2231 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2232 return CURSEG_WARM_NODE;
351df4b2
JK		 2233 else
2234 return CURSEG_COLD_NODE;
2235 }
2236}
2237
81377bd6 2238static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2239{
81377bd6
JK		 2240 if (fio->type == DATA) {
2241 struct inode *inode = fio->page->mapping->host;
351df4b2 2242
81377bd6 2243 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2244 return CURSEG_COLD_DATA;
ef095d19
JK		 2245 if (is_inode_flag_set(inode, FI_HOT_DATA))
2246 return CURSEG_HOT_DATA;
2247 return CURSEG_WARM_DATA;
351df4b2 2248 } else {
81377bd6
JK		 2249 if (IS_DNODE(fio->page))
2250 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2251 CURSEG_HOT_NODE;
ef095d19 2252 return CURSEG_COLD_NODE;
351df4b2
JK		 2253 }
2254}
2255
81377bd6 2256static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2257{
a912b54d
JK		 2258 int type = 0;
2259
81377bd6 2260 switch (fio->sbi->active_logs) {
351df4b2 2261 case 2:
a912b54d
JK		 2262 type = __get_segment_type_2(fio);
2263 break;
351df4b2 2264 case 4:
a912b54d
JK		 2265 type = __get_segment_type_4(fio);
2266 break;
2267 case 6:
2268 type = __get_segment_type_6(fio);
2269 break;
2270 default:
2271 f2fs_bug_on(fio->sbi, true);
351df4b2 2272 }
81377bd6 2273
a912b54d
JK		 2274 if (IS_HOT(type))
2275 fio->temp = HOT;
2276 else if (IS_WARM(type))
2277 fio->temp = WARM;
2278 else
2279 fio->temp = COLD;
2280 return type;
351df4b2
JK		 2281}
2282
bfad7c2d
JK		 2283void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2284 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2285 struct f2fs_summary *sum, int type,
2286 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2287{
2288 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2289 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 2290
2291 mutex_lock(&curseg->curseg_mutex);
21cb1d99 2292 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK		 2293
2294 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2295
4e6a8d9b
JK		 2296 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2297
351df4b2
JK		 2298 /*
2299 * __add_sum_entry should be resided under the curseg_mutex
2300 * because, this function updates a summary entry in the
2301 * current summary block.
2302 */
e79efe3b 2303 __add_sum_entry(sbi, type, sum);
351df4b2 2304
351df4b2 2305 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2306
2307 stat_inc_block_count(sbi, curseg);
351df4b2 2308
c6f82fe9
JK		 2309 if (!__has_curseg_space(sbi, type))
2310 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2 2311 /*
c6f82fe9
JK		 2312 * SIT information should be updated after segment allocation,
2313 * since we need to keep dirty segments precisely under SSR.
351df4b2
JK		 2314 */
2315 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 2316
351df4b2
JK		 2317 mutex_unlock(&sit_i->sentry_lock);
2318
704956ec 2319 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2320 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2321
704956ec
CY		 2322 f2fs_inode_chksum_set(sbi, page);
2323 }
2324
fb830fc5
CY		 2325 if (add_list) {
2326 struct f2fs_bio_info *io;
2327
2328 INIT_LIST_HEAD(&fio->list);
2329 fio->in_list = true;
2330 io = sbi->write_io[fio->type] + fio->temp;
2331 spin_lock(&io->io_lock);
2332 list_add_tail(&fio->list, &io->io_list);
2333 spin_unlock(&io->io_lock);
2334 }
2335
bfad7c2d
JK		 2336 mutex_unlock(&curseg->curseg_mutex);
2337}
2338
05ca3632 2339static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2340{
81377bd6 2341 int type = __get_segment_type(fio);
0a595eba 2342 int err;
bfad7c2d 2343
0a595eba 2344reallocate:
7a9d7548 2345 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2346 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2347
351df4b2 2348 /* writeout dirty page into bdev */
b9109b0e 2349 err = f2fs_submit_page_write(fio);
0a595eba
JK		 2350 if (err == -EAGAIN) {
2351 fio->old_blkaddr = fio->new_blkaddr;
2352 goto reallocate;
2353 }
351df4b2
JK		 2354}
2355
b0af6d49
CY		 2356void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2357 enum iostat_type io_type)
351df4b2 2358{
458e6197 2359 struct f2fs_io_info fio = {
05ca3632 2360 .sbi = sbi,
458e6197 2361 .type = META,
04d328de 2362 .op = REQ_OP_WRITE,
70fd7614 2363 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2364 .old_blkaddr = page->index,
2365 .new_blkaddr = page->index,
05ca3632 2366 .page = page,
4375a336 2367 .encrypted_page = NULL,
fb830fc5 2368 .in_list = false,
458e6197
JK		 2369 };
2370
2b947003 2371 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2372 fio.op_flags &= ~REQ_META;
2b947003 2373
351df4b2 2374 set_page_writeback(page);
b9109b0e 2375 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2376
2377 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2378}
2379
05ca3632 2380void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2381{
2382 struct f2fs_summary sum;
05ca3632 2383
351df4b2 2384 set_summary(&sum, nid, 0, 0);
05ca3632 2385 do_write_page(&sum, fio);
b0af6d49
CY		 2386
2387 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2388}
2389
05ca3632 2390void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2391{
05ca3632 2392 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2393 struct f2fs_summary sum;
2394 struct node_info ni;
2395
9850cf4a 2396 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2397 get_node_info(sbi, dn->nid, &ni);
2398 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2399 do_write_page(&sum, fio);
f28b3434 2400 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2401
2402 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2403}
2404
d1b3e72d 2405int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2406{
b0af6d49
CY		 2407 int err;
2408
7a9d7548 2409 fio->new_blkaddr = fio->old_blkaddr;
05ca3632 2410 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2411
2412 err = f2fs_submit_page_bio(fio);
2413
2414 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2415
2416 return err;
351df4b2
JK		 2417}
2418
4356e48e 2419void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2420 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2421 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2422{
2423 struct sit_info *sit_i = SIT_I(sbi);
2424 struct curseg_info *curseg;
2425 unsigned int segno, old_cursegno;
2426 struct seg_entry *se;
2427 int type;
19f106bc 2428 unsigned short old_blkoff;
351df4b2
JK		 2429
2430 segno = GET_SEGNO(sbi, new_blkaddr);
2431 se = get_seg_entry(sbi, segno);
2432 type = se->type;
2433
19f106bc
CY		 2434 if (!recover_curseg) {
2435 /* for recovery flow */
2436 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2437 if (old_blkaddr == NULL_ADDR)
2438 type = CURSEG_COLD_DATA;
2439 else
2440 type = CURSEG_WARM_DATA;
2441 }
2442 } else {
2443 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 2444 type = CURSEG_WARM_DATA;
2445 }
19f106bc 2446
351df4b2
JK		 2447 curseg = CURSEG_I(sbi, type);
2448
2449 mutex_lock(&curseg->curseg_mutex);
2450 mutex_lock(&sit_i->sentry_lock);
2451
2452 old_cursegno = curseg->segno;
19f106bc 2453 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2454
2455 /* change the current segment */
2456 if (segno != curseg->segno) {
2457 curseg->next_segno = segno;
2458 change_curseg(sbi, type, true);
2459 }
2460
491c0854 2461 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2462 __add_sum_entry(sbi, type, sum);
351df4b2 2463
28bc106b 2464 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2465 update_sit_entry(sbi, new_blkaddr, 1);
2466 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2467 update_sit_entry(sbi, old_blkaddr, -1);
2468
2469 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2470 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2471
351df4b2 2472 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2473
19f106bc
CY		 2474 if (recover_curseg) {
2475 if (old_cursegno != curseg->segno) {
2476 curseg->next_segno = old_cursegno;
2477 change_curseg(sbi, type, true);
2478 }
2479 curseg->next_blkoff = old_blkoff;
2480 }
2481
351df4b2
JK		 2482 mutex_unlock(&sit_i->sentry_lock);
2483 mutex_unlock(&curseg->curseg_mutex);
2484}
2485
528e3459
CY		 2486void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2487 block_t old_addr, block_t new_addr,
28bc106b
CY		 2488 unsigned char version, bool recover_curseg,
2489 bool recover_newaddr)
528e3459
CY		 2490{
2491 struct f2fs_summary sum;
2492
2493 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2494
28bc106b
CY		 2495 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2496 recover_curseg, recover_newaddr);
528e3459 2497
f28b3434 2498 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2499}
2500
93dfe2ac 2501void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2502 enum page_type type, bool ordered)
93dfe2ac 2503{
93dfe2ac 2504 if (PageWriteback(page)) {
4081363f
JK		 2505 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2506
b9109b0e
JK		 2507 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2508 0, page->index, type);
fec1d657
JK		 2509 if (ordered)
2510 wait_on_page_writeback(page);
2511 else
2512 wait_for_stable_page(page);
93dfe2ac
JK		 2513 }
2514}
2515
08b39fbd
CY		 2516void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
2517 block_t blkaddr)
2518{
2519 struct page *cpage;
2520
5d4c0af4 2521 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2522 return;
2523
08b39fbd
CY		 2524 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2525 if (cpage) {
fec1d657 2526 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2527 f2fs_put_page(cpage, 1);
2528 }
2529}
2530
351df4b2
JK		 2531static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2532{
2533 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2534 struct curseg_info *seg_i;
2535 unsigned char *kaddr;
2536 struct page *page;
2537 block_t start;
2538 int i, j, offset;
2539
2540 start = start_sum_block(sbi);
2541
2542 page = get_meta_page(sbi, start++);
2543 kaddr = (unsigned char *)page_address(page);
2544
2545 /* Step 1: restore nat cache */
2546 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2547 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2548
2549 /* Step 2: restore sit cache */
2550 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2551 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2552 offset = 2 * SUM_JOURNAL_SIZE;
2553
2554 /* Step 3: restore summary entries */
2555 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2556 unsigned short blk_off;
2557 unsigned int segno;
2558
2559 seg_i = CURSEG_I(sbi, i);
2560 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2561 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2562 seg_i->next_segno = segno;
2563 reset_curseg(sbi, i, 0);
2564 seg_i->alloc_type = ckpt->alloc_type[i];
2565 seg_i->next_blkoff = blk_off;
2566
2567 if (seg_i->alloc_type == SSR)
2568 blk_off = sbi->blocks_per_seg;
2569
2570 for (j = 0; j < blk_off; j++) {
2571 struct f2fs_summary *s;
2572 s = (struct f2fs_summary *)(kaddr + offset);
2573 seg_i->sum_blk->entries[j] = *s;
2574 offset += SUMMARY_SIZE;
09cbfeaf 2575 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2576 SUM_FOOTER_SIZE)
2577 continue;
2578
2579 f2fs_put_page(page, 1);
2580 page = NULL;
2581
2582 page = get_meta_page(sbi, start++);
2583 kaddr = (unsigned char *)page_address(page);
2584 offset = 0;
2585 }
2586 }
2587 f2fs_put_page(page, 1);
2588 return 0;
2589}
2590
2591static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2592{
2593 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2594 struct f2fs_summary_block *sum;
2595 struct curseg_info *curseg;
2596 struct page *new;
2597 unsigned short blk_off;
2598 unsigned int segno = 0;
2599 block_t blk_addr = 0;
2600
2601 /* get segment number and block addr */
2602 if (IS_DATASEG(type)) {
2603 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2604 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2605 CURSEG_HOT_DATA]);
119ee914 2606 if (__exist_node_summaries(sbi))
351df4b2
JK		 2607 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2608 else
2609 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2610 } else {
2611 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2612 CURSEG_HOT_NODE]);
2613 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2614 CURSEG_HOT_NODE]);
119ee914 2615 if (__exist_node_summaries(sbi))
351df4b2
JK		 2616 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2617 type - CURSEG_HOT_NODE);
2618 else
2619 blk_addr = GET_SUM_BLOCK(sbi, segno);
2620 }
2621
2622 new = get_meta_page(sbi, blk_addr);
2623 sum = (struct f2fs_summary_block *)page_address(new);
2624
2625 if (IS_NODESEG(type)) {
119ee914 2626 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2627 struct f2fs_summary *ns = &sum->entries[0];
2628 int i;
2629 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2630 ns->version = 0;
2631 ns->ofs_in_node = 0;
2632 }
2633 } else {
d653788a
GZ		 2634 int err;
2635
2636 err = restore_node_summary(sbi, segno, sum);
2637 if (err) {
351df4b2 2638 f2fs_put_page(new, 1);
d653788a 2639 return err;
351df4b2
JK		 2640 }
2641 }
2642 }
2643
2644 /* set uncompleted segment to curseg */
2645 curseg = CURSEG_I(sbi, type);
2646 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2647
2648 /* update journal info */
2649 down_write(&curseg->journal_rwsem);
2650 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2651 up_write(&curseg->journal_rwsem);
2652
2653 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2654 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2655 curseg->next_segno = segno;
2656 reset_curseg(sbi, type, 0);
2657 curseg->alloc_type = ckpt->alloc_type[type];
2658 curseg->next_blkoff = blk_off;
2659 mutex_unlock(&curseg->curseg_mutex);
2660 f2fs_put_page(new, 1);
2661 return 0;
2662}
2663
2664static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2665{
21d3f8e1
JQ		 2666 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2667 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 2668 int type = CURSEG_HOT_DATA;
e4fc5fbf 2669 int err;
351df4b2 2670
aaec2b1d 2671 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2672 int npages = npages_for_summary_flush(sbi, true);
2673
2674 if (npages >= 2)
2675 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2676 META_CP, true);
3fa06d7b 2677
351df4b2
JK		 2678 /* restore for compacted data summary */
2679 if (read_compacted_summaries(sbi))
2680 return -EINVAL;
2681 type = CURSEG_HOT_NODE;
2682 }
2683
119ee914 2684 if (__exist_node_summaries(sbi))
3fa06d7b 2685 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2686 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2687
e4fc5fbf
CY		 2688 for (; type <= CURSEG_COLD_NODE; type++) {
2689 err = read_normal_summaries(sbi, type);
2690 if (err)
2691 return err;
2692 }
2693
21d3f8e1
JQ		 2694 /* sanity check for summary blocks */
2695 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2696 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2697 return -EINVAL;
2698
351df4b2
JK		 2699 return 0;
2700}
2701
2702static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2703{
2704 struct page *page;
2705 unsigned char *kaddr;
2706 struct f2fs_summary *summary;
2707 struct curseg_info *seg_i;
2708 int written_size = 0;
2709 int i, j;
2710
2711 page = grab_meta_page(sbi, blkaddr++);
2712 kaddr = (unsigned char *)page_address(page);
2713
2714 /* Step 1: write nat cache */
2715 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2716 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2717 written_size += SUM_JOURNAL_SIZE;
2718
2719 /* Step 2: write sit cache */
2720 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2721 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2722 written_size += SUM_JOURNAL_SIZE;
2723
351df4b2
JK		 2724 /* Step 3: write summary entries */
2725 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2726 unsigned short blkoff;
2727 seg_i = CURSEG_I(sbi, i);
2728 if (sbi->ckpt->alloc_type[i] == SSR)
2729 blkoff = sbi->blocks_per_seg;
2730 else
2731 blkoff = curseg_blkoff(sbi, i);
2732
2733 for (j = 0; j < blkoff; j++) {
2734 if (!page) {
2735 page = grab_meta_page(sbi, blkaddr++);
2736 kaddr = (unsigned char *)page_address(page);
2737 written_size = 0;
2738 }
2739 summary = (struct f2fs_summary *)(kaddr + written_size);
2740 *summary = seg_i->sum_blk->entries[j];
2741 written_size += SUMMARY_SIZE;
351df4b2 2742
09cbfeaf 2743 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2744 SUM_FOOTER_SIZE)
2745 continue;
2746
e8d61a74 2747 set_page_dirty(page);
351df4b2
JK		 2748 f2fs_put_page(page, 1);
2749 page = NULL;
2750 }
2751 }
e8d61a74
CY		 2752 if (page) {
2753 set_page_dirty(page);
351df4b2 2754 f2fs_put_page(page, 1);
e8d61a74 2755 }
351df4b2
JK		 2756}
2757
2758static void write_normal_summaries(struct f2fs_sb_info *sbi,
2759 block_t blkaddr, int type)
2760{
2761 int i, end;
2762 if (IS_DATASEG(type))
2763 end = type + NR_CURSEG_DATA_TYPE;
2764 else
2765 end = type + NR_CURSEG_NODE_TYPE;
2766
b7ad7512
CY		 2767 for (i = type; i < end; i++)
2768 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 2769}
2770
2771void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2772{
aaec2b1d 2773 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 2774 write_compacted_summaries(sbi, start_blk);
2775 else
2776 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2777}
2778
2779void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2780{
119ee914 2781 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 2782}
2783
dfc08a12 2784int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 2785 unsigned int val, int alloc)
2786{
2787 int i;
2788
2789 if (type == NAT_JOURNAL) {
dfc08a12
CY		 2790 for (i = 0; i < nats_in_cursum(journal); i++) {
2791 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 2792 return i;
2793 }
dfc08a12
CY		 2794 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2795 return update_nats_in_cursum(journal, 1);
351df4b2 2796 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 2797 for (i = 0; i < sits_in_cursum(journal); i++)
2798 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 2799 return i;
dfc08a12
CY		 2800 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2801 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 2802 }
2803 return -1;
2804}
2805
2806static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2807 unsigned int segno)
2808{
2cc22186 2809 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 2810}
2811
2812static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2813 unsigned int start)
2814{
2815 struct sit_info *sit_i = SIT_I(sbi);
2816 struct page *src_page, *dst_page;
2817 pgoff_t src_off, dst_off;
2818 void *src_addr, *dst_addr;
2819
2820 src_off = current_sit_addr(sbi, start);
2821 dst_off = next_sit_addr(sbi, src_off);
2822
2823 /* get current sit block page without lock */
2824 src_page = get_meta_page(sbi, src_off);
2825 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 2826 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 2827
2828 src_addr = page_address(src_page);
2829 dst_addr = page_address(dst_page);
09cbfeaf 2830 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 2831
2832 set_page_dirty(dst_page);
2833 f2fs_put_page(src_page, 1);
2834
2835 set_to_next_sit(sit_i, start);
2836
2837 return dst_page;
2838}
2839
184a5cd2
CY		 2840static struct sit_entry_set *grab_sit_entry_set(void)
2841{
2842 struct sit_entry_set *ses =
80c54505 2843 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 2844
2845 ses->entry_cnt = 0;
2846 INIT_LIST_HEAD(&ses->set_list);
2847 return ses;
2848}
2849
2850static void release_sit_entry_set(struct sit_entry_set *ses)
2851{
2852 list_del(&ses->set_list);
2853 kmem_cache_free(sit_entry_set_slab, ses);
2854}
2855
2856static void adjust_sit_entry_set(struct sit_entry_set *ses,
2857 struct list_head *head)
2858{
2859 struct sit_entry_set *next = ses;
2860
2861 if (list_is_last(&ses->set_list, head))
2862 return;
2863
2864 list_for_each_entry_continue(next, head, set_list)
2865 if (ses->entry_cnt <= next->entry_cnt)
2866 break;
2867
2868 list_move_tail(&ses->set_list, &next->set_list);
2869}
2870
2871static void add_sit_entry(unsigned int segno, struct list_head *head)
2872{
2873 struct sit_entry_set *ses;
2874 unsigned int start_segno = START_SEGNO(segno);
2875
2876 list_for_each_entry(ses, head, set_list) {
2877 if (ses->start_segno == start_segno) {
2878 ses->entry_cnt++;
2879 adjust_sit_entry_set(ses, head);
2880 return;
2881 }
2882 }
2883
2884 ses = grab_sit_entry_set();
2885
2886 ses->start_segno = start_segno;
2887 ses->entry_cnt++;
2888 list_add(&ses->set_list, head);
2889}
2890
2891static void add_sits_in_set(struct f2fs_sb_info *sbi)
2892{
2893 struct f2fs_sm_info *sm_info = SM_I(sbi);
2894 struct list_head *set_list = &sm_info->sit_entry_set;
2895 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 2896 unsigned int segno;
2897
7cd8558b 2898 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 2899 add_sit_entry(segno, set_list);
2900}
2901
2902static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 2903{
2904 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2905 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 2906 int i;
2907
b7ad7512 2908 down_write(&curseg->journal_rwsem);
dfc08a12 2909 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 2910 unsigned int segno;
2911 bool dirtied;
2912
dfc08a12 2913 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 2914 dirtied = __mark_sit_entry_dirty(sbi, segno);
2915
2916 if (!dirtied)
2917 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 2918 }
dfc08a12 2919 update_sits_in_cursum(journal, -i);
b7ad7512 2920 up_write(&curseg->journal_rwsem);
351df4b2
JK		 2921}
2922
0a8165d7 2923/*
351df4b2
JK		 2924 * CP calls this function, which flushes SIT entries including sit_journal,
2925 * and moves prefree segs to free segs.
2926 */
4b2fecc8 2927void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 2928{
2929 struct sit_info *sit_i = SIT_I(sbi);
2930 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2931 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2932 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 2933 struct sit_entry_set *ses, *tmp;
2934 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 2935 bool to_journal = true;
4b2fecc8 2936 struct seg_entry *se;
351df4b2 2937
351df4b2
JK		 2938 mutex_lock(&sit_i->sentry_lock);
2939
2b11a74b
WL		 2940 if (!sit_i->dirty_sentries)
2941 goto out;
2942
351df4b2 2943 /*
184a5cd2
CY		 2944 * add and account sit entries of dirty bitmap in sit entry
2945 * set temporarily
351df4b2 2946 */
184a5cd2 2947 add_sits_in_set(sbi);
351df4b2 2948
184a5cd2
CY		 2949 /*
2950 * if there are no enough space in journal to store dirty sit
2951 * entries, remove all entries from journal and add and account
2952 * them in sit entry set.
2953 */
dfc08a12 2954 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2955 remove_sits_in_journal(sbi);
b2955550 2956
184a5cd2
CY		 2957 /*
2958 * there are two steps to flush sit entries:
2959 * #1, flush sit entries to journal in current cold data summary block.
2960 * #2, flush sit entries to sit page.
2961 */
2962 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2963 struct page *page = NULL;
184a5cd2
CY		 2964 struct f2fs_sit_block *raw_sit = NULL;
2965 unsigned int start_segno = ses->start_segno;
2966 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2967 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2968 unsigned int segno = start_segno;
2969
2970 if (to_journal &&
dfc08a12 2971 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2972 to_journal = false;
2973
b7ad7512
CY		 2974 if (to_journal) {
2975 down_write(&curseg->journal_rwsem);
2976 } else {
184a5cd2
CY		 2977 page = get_next_sit_page(sbi, start_segno);
2978 raw_sit = page_address(page);
351df4b2 2979 }
351df4b2 2980
184a5cd2
CY		 2981 /* flush dirty sit entries in region of current sit set */
2982 for_each_set_bit_from(segno, bitmap, end) {
2983 int offset, sit_offset;
4b2fecc8
JK		 2984
2985 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2986
2987 /* add discard candidates */
c473f1a9 2988 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 2989 cpc->trim_start = segno;
25290fa5 2990 add_discard_addrs(sbi, cpc, false);
4b2fecc8 2991 }
184a5cd2
CY		 2992
2993 if (to_journal) {
dfc08a12 2994 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2995 SIT_JOURNAL, segno, 1);
2996 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2997 segno_in_journal(journal, offset) =
184a5cd2
CY		 2998 cpu_to_le32(segno);
2999 seg_info_to_raw_sit(se,
dfc08a12 3000 &sit_in_journal(journal, offset));
184a5cd2
CY		 3001 } else {
3002 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3003 seg_info_to_raw_sit(se,
3004 &raw_sit->entries[sit_offset]);
3005 }
351df4b2 3006
184a5cd2
CY		 3007 __clear_bit(segno, bitmap);
3008 sit_i->dirty_sentries--;
3009 ses->entry_cnt--;
351df4b2
JK		 3010 }
3011
b7ad7512
CY		 3012 if (to_journal)
3013 up_write(&curseg->journal_rwsem);
3014 else
184a5cd2
CY		 3015 f2fs_put_page(page, 1);
3016
3017 f2fs_bug_on(sbi, ses->entry_cnt);
3018 release_sit_entry_set(ses);
351df4b2 3019 }
184a5cd2
CY		 3020
3021 f2fs_bug_on(sbi, !list_empty(head));
3022 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3023out:
c473f1a9 3024 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3025 __u64 trim_start = cpc->trim_start;
3026
4b2fecc8 3027 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3028 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3029
3030 cpc->trim_start = trim_start;
4b2fecc8 3031 }
351df4b2 3032 mutex_unlock(&sit_i->sentry_lock);
351df4b2 3033
351df4b2
JK		 3034 set_prefree_as_free_segments(sbi);
3035}
3036
3037static int build_sit_info(struct f2fs_sb_info *sbi)
3038{
3039 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3040 struct sit_info *sit_i;
3041 unsigned int sit_segs, start;
ae27d62e 3042 char *src_bitmap;
351df4b2
JK		 3043 unsigned int bitmap_size;
3044
3045 /* allocate memory for SIT information */
3046 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3047 if (!sit_i)
3048 return -ENOMEM;
3049
3050 SM_I(sbi)->sit_info = sit_i;
3051
a7c3e901 3052 sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
39307a8e 3053 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3054 if (!sit_i->sentries)
3055 return -ENOMEM;
3056
7cd8558b 3057 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3058 sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3059 if (!sit_i->dirty_sentries_bitmap)
3060 return -ENOMEM;
3061
7cd8558b 3062 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3063 sit_i->sentries[start].cur_valid_map
3064 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3065 sit_i->sentries[start].ckpt_valid_map
3066 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3067 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3068 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3069 return -ENOMEM;
3e025740 3070
355e7891
CY		 3071#ifdef CONFIG_F2FS_CHECK_FS
3072 sit_i->sentries[start].cur_valid_map_mir
3073 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3074 if (!sit_i->sentries[start].cur_valid_map_mir)
3075 return -ENOMEM;
3076#endif
3077
3e025740
JK		 3078 if (f2fs_discard_en(sbi)) {
3079 sit_i->sentries[start].discard_map
3080 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3081 if (!sit_i->sentries[start].discard_map)
3082 return -ENOMEM;
3083 }
351df4b2
JK		 3084 }
3085
60a3b782
JK		 3086 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3087 if (!sit_i->tmp_map)
3088 return -ENOMEM;
3089
351df4b2 3090 if (sbi->segs_per_sec > 1) {
a7c3e901 3091 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
39307a8e 3092 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3093 if (!sit_i->sec_entries)
3094 return -ENOMEM;
3095 }
3096
3097 /* get information related with SIT */
3098 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3099
3100 /* setup SIT bitmap from ckeckpoint pack */
3101 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3102 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3103
ae27d62e
CY		 3104 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3105 if (!sit_i->sit_bitmap)
351df4b2 3106 return -ENOMEM;
351df4b2 3107
ae27d62e
CY		 3108#ifdef CONFIG_F2FS_CHECK_FS
3109 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3110 if (!sit_i->sit_bitmap_mir)
3111 return -ENOMEM;
3112#endif
3113
351df4b2
JK		 3114 /* init SIT information */
3115 sit_i->s_ops = &default_salloc_ops;
3116
3117 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3118 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3119 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3120 sit_i->bitmap_size = bitmap_size;
3121 sit_i->dirty_sentries = 0;
3122 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3123 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3124 sit_i->mounted_time = ktime_get_real_seconds();
351df4b2
JK		 3125 mutex_init(&sit_i->sentry_lock);
3126 return 0;
3127}
3128
3129static int build_free_segmap(struct f2fs_sb_info *sbi)
3130{
351df4b2
JK		 3131 struct free_segmap_info *free_i;
3132 unsigned int bitmap_size, sec_bitmap_size;
3133
3134 /* allocate memory for free segmap information */
3135 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3136 if (!free_i)
3137 return -ENOMEM;
3138
3139 SM_I(sbi)->free_info = free_i;
3140
7cd8558b 3141 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3142 free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3143 if (!free_i->free_segmap)
3144 return -ENOMEM;
3145
7cd8558b 3146 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
a7c3e901 3147 free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3148 if (!free_i->free_secmap)
3149 return -ENOMEM;
3150
3151 /* set all segments as dirty temporarily */
3152 memset(free_i->free_segmap, 0xff, bitmap_size);
3153 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3154
3155 /* init free segmap information */
7cd8558b 3156 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3157 free_i->free_segments = 0;
3158 free_i->free_sections = 0;
1a118ccf 3159 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3160 return 0;
3161}
3162
3163static int build_curseg(struct f2fs_sb_info *sbi)
3164{
1042d60f 3165 struct curseg_info *array;
351df4b2
JK		 3166 int i;
3167
b434babf 3168 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 3169 if (!array)
3170 return -ENOMEM;
3171
3172 SM_I(sbi)->curseg_array = array;
3173
3174 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3175 mutex_init(&array[i].curseg_mutex);
09cbfeaf 3176 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3177 if (!array[i].sum_blk)
3178 return -ENOMEM;
b7ad7512
CY		 3179 init_rwsem(&array[i].journal_rwsem);
3180 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3181 GFP_KERNEL);
3182 if (!array[i].journal)
3183 return -ENOMEM;
351df4b2
JK		 3184 array[i].segno = NULL_SEGNO;
3185 array[i].next_blkoff = 0;
3186 }
3187 return restore_curseg_summaries(sbi);
3188}
3189
3190static void build_sit_entries(struct f2fs_sb_info *sbi)
3191{
3192 struct sit_info *sit_i = SIT_I(sbi);
3193 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3194 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3195 struct seg_entry *se;
3196 struct f2fs_sit_entry sit;
74de593a
CY		 3197 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3198 unsigned int i, start, end;
3199 unsigned int readed, start_blk = 0;
351df4b2 3200
74de593a 3201 do {
664ba972
JK		 3202 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3203 META_SIT, true);
74de593a
CY		 3204
3205 start = start_blk * sit_i->sents_per_block;
3206 end = (start_blk + readed) * sit_i->sents_per_block;
3207
7cd8558b 3208 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3209 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3210 struct page *page;
3211
9c094040 3212 se = &sit_i->sentries[start];
74de593a
CY		 3213 page = get_current_sit_page(sbi, start);
3214 sit_blk = (struct f2fs_sit_block *)page_address(page);
3215 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3216 f2fs_put_page(page, 1);
d600af23 3217
74de593a
CY		 3218 check_block_count(sbi, start, &sit);
3219 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 3220
3221 /* build discard map only one time */
3e025740 3222 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3223 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3224 memset(se->discard_map, 0xff,
3225 SIT_VBLOCK_MAP_SIZE);
3226 } else {
3227 memcpy(se->discard_map,
3228 se->cur_valid_map,
3229 SIT_VBLOCK_MAP_SIZE);
3230 sbi->discard_blks +=
3231 sbi->blocks_per_seg -
3232 se->valid_blocks;
3233 }
3e025740 3234 }
a66cdd98 3235
d600af23
CY		 3236 if (sbi->segs_per_sec > 1)
3237 get_sec_entry(sbi, start)->valid_blocks +=
3238 se->valid_blocks;
351df4b2 3239 }
74de593a
CY		 3240 start_blk += readed;
3241 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3242
3243 down_read(&curseg->journal_rwsem);
3244 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3245 unsigned int old_valid_blocks;
3246
3247 start = le32_to_cpu(segno_in_journal(journal, i));
3248 se = &sit_i->sentries[start];
3249 sit = sit_in_journal(journal, i);
3250
3251 old_valid_blocks = se->valid_blocks;
3252
3253 check_block_count(sbi, start, &sit);
3254 seg_info_from_raw_sit(se, &sit);
3255
3256 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3257 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3258 memset(se->discard_map, 0xff,
3259 SIT_VBLOCK_MAP_SIZE);
3260 } else {
3261 memcpy(se->discard_map, se->cur_valid_map,
3262 SIT_VBLOCK_MAP_SIZE);
3263 sbi->discard_blks += old_valid_blocks -
3264 se->valid_blocks;
3265 }
d600af23
CY		 3266 }
3267
3268 if (sbi->segs_per_sec > 1)
3269 get_sec_entry(sbi, start)->valid_blocks +=
3270 se->valid_blocks - old_valid_blocks;
3271 }
3272 up_read(&curseg->journal_rwsem);
351df4b2
JK		 3273}
3274
3275static void init_free_segmap(struct f2fs_sb_info *sbi)
3276{
3277 unsigned int start;
3278 int type;
3279
7cd8558b 3280 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3281 struct seg_entry *sentry = get_seg_entry(sbi, start);
3282 if (!sentry->valid_blocks)
3283 __set_free(sbi, start);
c79b7ff1
JK		 3284 else
3285 SIT_I(sbi)->written_valid_blocks +=
3286 sentry->valid_blocks;
351df4b2
JK		 3287 }
3288
3289 /* set use the current segments */
3290 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3291 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3292 __set_test_and_inuse(sbi, curseg_t->segno);
3293 }
3294}
3295
3296static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3297{
3298 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3299 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3300 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3301 unsigned short valid_blocks;
3302
8736fbf0 3303 while (1) {
351df4b2 3304 /* find dirty segment based on free segmap */
7cd8558b
JK		 3305 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3306 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3307 break;
3308 offset = segno + 1;
302bd348 3309 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3310 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3311 continue;
ec325b52
JK		 3312 if (valid_blocks > sbi->blocks_per_seg) {
3313 f2fs_bug_on(sbi, 1);
3314 continue;
3315 }
351df4b2
JK		 3316 mutex_lock(&dirty_i->seglist_lock);
3317 __locate_dirty_segment(sbi, segno, DIRTY);
3318 mutex_unlock(&dirty_i->seglist_lock);
3319 }
3320}
3321
5ec4e49f 3322static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3323{
3324 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3325 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3326
a7c3e901 3327 dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 3328 if (!dirty_i->victim_secmap)
351df4b2
JK		 3329 return -ENOMEM;
3330 return 0;
3331}
3332
3333static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3334{
3335 struct dirty_seglist_info *dirty_i;
3336 unsigned int bitmap_size, i;
3337
3338 /* allocate memory for dirty segments list information */
3339 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3340 if (!dirty_i)
3341 return -ENOMEM;
3342
3343 SM_I(sbi)->dirty_info = dirty_i;
3344 mutex_init(&dirty_i->seglist_lock);
3345
7cd8558b 3346 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3347
3348 for (i = 0; i < NR_DIRTY_TYPE; i++) {
a7c3e901 3349 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3350 if (!dirty_i->dirty_segmap[i])
3351 return -ENOMEM;
3352 }
3353
3354 init_dirty_segmap(sbi);
5ec4e49f 3355 return init_victim_secmap(sbi);
351df4b2
JK		 3356}
3357
0a8165d7 3358/*
351df4b2
JK		 3359 * Update min, max modified time for cost-benefit GC algorithm
3360 */
3361static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3362{
3363 struct sit_info *sit_i = SIT_I(sbi);
3364 unsigned int segno;
3365
3366 mutex_lock(&sit_i->sentry_lock);
3367
3368 sit_i->min_mtime = LLONG_MAX;
3369
7cd8558b 3370 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3371 unsigned int i;
3372 unsigned long long mtime = 0;
3373
3374 for (i = 0; i < sbi->segs_per_sec; i++)
3375 mtime += get_seg_entry(sbi, segno + i)->mtime;
3376
3377 mtime = div_u64(mtime, sbi->segs_per_sec);
3378
3379 if (sit_i->min_mtime > mtime)
3380 sit_i->min_mtime = mtime;
3381 }
3382 sit_i->max_mtime = get_mtime(sbi);
3383 mutex_unlock(&sit_i->sentry_lock);
3384}
3385
3386int build_segment_manager(struct f2fs_sb_info *sbi)
3387{
3388 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3389 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3390 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3391 int err;
3392
3393 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3394 if (!sm_info)
3395 return -ENOMEM;
3396
3397 /* init sm info */
3398 sbi->sm_info = sm_info;
351df4b2
JK		 3399 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3400 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3401 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3402 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3403 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3404 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3405 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3406 sm_info->rec_prefree_segments = sm_info->main_segments *
3407 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3408 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3409 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3410
52763a4b
JK		 3411 if (!test_opt(sbi, LFS))
3412 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3413 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3414 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3415 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
351df4b2 3416
bba681cb
JK		 3417 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3418
184a5cd2
CY		 3419 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3420
d4fdf8ba 3421 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3422 err = create_flush_cmd_control(sbi);
3423 if (err)
a688b9d9 3424 return err;
6b4afdd7
JK		 3425 }
3426
0b54fb84
JK		 3427 err = create_discard_cmd_control(sbi);
3428 if (err)
3429 return err;
3430
351df4b2
JK		 3431 err = build_sit_info(sbi);
3432 if (err)
3433 return err;
3434 err = build_free_segmap(sbi);
3435 if (err)
3436 return err;
3437 err = build_curseg(sbi);
3438 if (err)
3439 return err;
3440
3441 /* reinit free segmap based on SIT */
3442 build_sit_entries(sbi);
3443
3444 init_free_segmap(sbi);
3445 err = build_dirty_segmap(sbi);
3446 if (err)
3447 return err;
3448
3449 init_min_max_mtime(sbi);
3450 return 0;
3451}
3452
3453static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3454 enum dirty_type dirty_type)
3455{
3456 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3457
3458 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3459 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3460 dirty_i->nr_dirty[dirty_type] = 0;
3461 mutex_unlock(&dirty_i->seglist_lock);
3462}
3463
5ec4e49f 3464static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3465{
3466 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3467 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3468}
3469
3470static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3471{
3472 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3473 int i;
3474
3475 if (!dirty_i)
3476 return;
3477
3478 /* discard pre-free/dirty segments list */
3479 for (i = 0; i < NR_DIRTY_TYPE; i++)
3480 discard_dirty_segmap(sbi, i);
3481
5ec4e49f 3482 destroy_victim_secmap(sbi);
351df4b2
JK		 3483 SM_I(sbi)->dirty_info = NULL;
3484 kfree(dirty_i);
3485}
3486
3487static void destroy_curseg(struct f2fs_sb_info *sbi)
3488{
3489 struct curseg_info *array = SM_I(sbi)->curseg_array;
3490 int i;
3491
3492 if (!array)
3493 return;
3494 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3495 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3496 kfree(array[i].sum_blk);
b7ad7512
CY		 3497 kfree(array[i].journal);
3498 }
351df4b2
JK		 3499 kfree(array);
3500}
3501
3502static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3503{
3504 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3505 if (!free_i)
3506 return;
3507 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3508 kvfree(free_i->free_segmap);
3509 kvfree(free_i->free_secmap);
351df4b2
JK		 3510 kfree(free_i);
3511}
3512
3513static void destroy_sit_info(struct f2fs_sb_info *sbi)
3514{
3515 struct sit_info *sit_i = SIT_I(sbi);
3516 unsigned int start;
3517
3518 if (!sit_i)
3519 return;
3520
3521 if (sit_i->sentries) {
7cd8558b 3522 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3523 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 3524#ifdef CONFIG_F2FS_CHECK_FS
3525 kfree(sit_i->sentries[start].cur_valid_map_mir);
3526#endif
351df4b2 3527 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3528 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3529 }
3530 }
60a3b782
JK		 3531 kfree(sit_i->tmp_map);
3532
39307a8e
JK		 3533 kvfree(sit_i->sentries);
3534 kvfree(sit_i->sec_entries);
3535 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3536
3537 SM_I(sbi)->sit_info = NULL;
3538 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 3539#ifdef CONFIG_F2FS_CHECK_FS
3540 kfree(sit_i->sit_bitmap_mir);
3541#endif
351df4b2
JK		 3542 kfree(sit_i);
3543}
3544
3545void destroy_segment_manager(struct f2fs_sb_info *sbi)
3546{
3547 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3548
3b03f724
CY		 3549 if (!sm_info)
3550 return;
5eba8c5d 3551 destroy_flush_cmd_control(sbi, true);
f099405f 3552 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3553 destroy_dirty_segmap(sbi);
3554 destroy_curseg(sbi);
3555 destroy_free_segmap(sbi);
3556 destroy_sit_info(sbi);
3557 sbi->sm_info = NULL;
3558 kfree(sm_info);
3559}
7fd9e544
JK		 3560
3561int __init create_segment_manager_caches(void)
3562{
3563 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3564 sizeof(struct discard_entry));
7fd9e544 3565 if (!discard_entry_slab)
184a5cd2
CY		 3566 goto fail;
3567
b01a9201
JK		 3568 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3569 sizeof(struct discard_cmd));
3570 if (!discard_cmd_slab)
6ab2a308 3571 goto destroy_discard_entry;
275b66b0 3572
184a5cd2 3573 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3574 sizeof(struct sit_entry_set));
184a5cd2 3575 if (!sit_entry_set_slab)
b01a9201 3576 goto destroy_discard_cmd;
88b88a66
JK		 3577
3578 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3579 sizeof(struct inmem_pages));
3580 if (!inmem_entry_slab)
3581 goto destroy_sit_entry_set;
7fd9e544 3582 return 0;
184a5cd2 3583
88b88a66
JK		 3584destroy_sit_entry_set:
3585 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 3586destroy_discard_cmd:
3587 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3588destroy_discard_entry:
184a5cd2
CY		 3589 kmem_cache_destroy(discard_entry_slab);
3590fail:
3591 return -ENOMEM;
7fd9e544
JK		 3592}
3593
3594void destroy_segment_manager_caches(void)
3595{
184a5cd2 3596 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3597 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3598 kmem_cache_destroy(discard_entry_slab);
88b88a66 3599 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3600}
Linux 6.x block layer and io_uring tree(s)