]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: avoid race in between atomic_read & atomic_inc
[mirror_ubuntu-jammy-kernel.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);
84a23fbe 258 clear_inode_flag(inode, FI_HOT_DATA);
5fe45743 259 stat_dec_atomic_write(inode);
29b96b54
CY		 260}
261
8c242db9
JK		 262void drop_inmem_page(struct inode *inode, struct page *page)
263{
264 struct f2fs_inode_info *fi = F2FS_I(inode);
265 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
266 struct list_head *head = &fi->inmem_pages;
267 struct inmem_pages *cur = NULL;
268
269 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
270
271 mutex_lock(&fi->inmem_lock);
272 list_for_each_entry(cur, head, list) {
273 if (cur->page == page)
274 break;
275 }
276
277 f2fs_bug_on(sbi, !cur || cur->page != page);
278 list_del(&cur->list);
279 mutex_unlock(&fi->inmem_lock);
280
281 dec_page_count(sbi, F2FS_INMEM_PAGES);
282 kmem_cache_free(inmem_entry_slab, cur);
283
284 ClearPageUptodate(page);
285 set_page_private(page, 0);
286 ClearPagePrivate(page);
287 f2fs_put_page(page, 0);
288
289 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
290}
291
28bc106b
CY		 292static int __commit_inmem_pages(struct inode *inode,
293 struct list_head *revoke_list)
88b88a66
JK		 294{
295 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
296 struct f2fs_inode_info *fi = F2FS_I(inode);
297 struct inmem_pages *cur, *tmp;
88b88a66 298 struct f2fs_io_info fio = {
05ca3632 299 .sbi = sbi,
88b88a66 300 .type = DATA,
04d328de 301 .op = REQ_OP_WRITE,
70fd7614 302 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 303 .io_type = FS_DATA_IO,
88b88a66 304 };
942fd319 305 pgoff_t last_idx = ULONG_MAX;
edb27dee 306 int err = 0;
88b88a66 307
88b88a66 308 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 309 struct page *page = cur->page;
310
311 lock_page(page);
312 if (page->mapping == inode->i_mapping) {
313 trace_f2fs_commit_inmem_page(page, INMEM);
314
315 set_page_dirty(page);
316 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 317 if (clear_page_dirty_for_io(page)) {
29b96b54 318 inode_dec_dirty_pages(inode);
933439c8
CY		 319 remove_dirty_inode(inode);
320 }
640cc189 321retry:
28bc106b 322 fio.page = page;
e959c8f5 323 fio.old_blkaddr = NULL_ADDR;
4d978078 324 fio.encrypted_page = NULL;
cc15620b 325 fio.need_lock = LOCK_DONE;
29b96b54
CY		 326 err = do_write_data_page(&fio);
327 if (err) {
640cc189
JK		 328 if (err == -ENOMEM) {
329 congestion_wait(BLK_RW_ASYNC, HZ/50);
330 cond_resched();
331 goto retry;
332 }
28bc106b 333 unlock_page(page);
29b96b54 334 break;
70c640b1 335 }
28bc106b
CY		 336 /* record old blkaddr for revoking */
337 cur->old_addr = fio.old_blkaddr;
942fd319 338 last_idx = page->index;
28bc106b
CY		 339 }
340 unlock_page(page);
341 list_move_tail(&cur->list, revoke_list);
88b88a66 342 }
29b96b54 343
942fd319 344 if (last_idx != ULONG_MAX)
b9109b0e 345 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b
CY		 346
347 if (!err)
348 __revoke_inmem_pages(inode, revoke_list, false, false);
349
29b96b54
CY		 350 return err;
351}
352
353int commit_inmem_pages(struct inode *inode)
354{
355 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
356 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 357 struct list_head revoke_list;
358 int err;
29b96b54 359
28bc106b 360 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 361 f2fs_balance_fs(sbi, true);
362 f2fs_lock_op(sbi);
363
5fe45743
CY		 364 set_inode_flag(inode, FI_ATOMIC_COMMIT);
365
29b96b54 366 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 367 err = __commit_inmem_pages(inode, &revoke_list);
368 if (err) {
369 int ret;
370 /*
371 * try to revoke all committed pages, but still we could fail
372 * due to no memory or other reason, if that happened, EAGAIN
373 * will be returned, which means in such case, transaction is
374 * already not integrity, caller should use journal to do the
375 * recovery or rewrite & commit last transaction. For other
376 * error number, revoking was done by filesystem itself.
377 */
378 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
379 if (ret)
380 err = ret;
381
382 /* drop all uncommitted pages */
383 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
384 }
88b88a66
JK		 385 mutex_unlock(&fi->inmem_lock);
386
5fe45743
CY		 387 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
388
29b96b54 389 f2fs_unlock_op(sbi);
edb27dee 390 return err;
88b88a66
JK		 391}
392
0a8165d7 393/*
351df4b2
JK		 394 * This function balances dirty node and dentry pages.
395 * In addition, it controls garbage collection.
396 */
2c4db1a6 397void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 398{
0f348028 399#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 400 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
401 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 402 f2fs_stop_checkpoint(sbi, false);
55523519 403 }
0f348028
CY		 404#endif
405
e589c2c4 406 /* balance_fs_bg is able to be pending */
a7881893 407 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 408 f2fs_balance_fs_bg(sbi);
409
351df4b2 410 /*
029cd28c
JK		 411 * We should do GC or end up with checkpoint, if there are so many dirty
412 * dir/node pages without enough free segments.
351df4b2 413 */
7f3037a5 414 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 415 mutex_lock(&sbi->gc_mutex);
e066b83c 416 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 417 }
418}
419
4660f9c0
JK		 420void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
421{
1dcc336b 422 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 423 if (!available_free_memory(sbi, EXTENT_CACHE))
424 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 425
1b38dc8e
JK		 426 /* check the # of cached NAT entries */
427 if (!available_free_memory(sbi, NAT_ENTRIES))
428 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
429
31696580 430 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 431 try_to_free_nids(sbi, MAX_FREE_NIDS);
432 else
22ad0b6a 433 build_free_nids(sbi, false, false);
31696580 434
1c0f4bf5 435 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 436 return;
31696580 437
1b38dc8e
JK		 438 /* checkpoint is the only way to shrink partial cached entries */
439 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 440 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 441 excess_prefree_segs(sbi) ||
442 excess_dirty_nats(sbi) ||
f455c8a5 443 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 444 if (test_opt(sbi, DATA_FLUSH)) {
445 struct blk_plug plug;
446
447 blk_start_plug(&plug);
36b35a0d 448 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 449 blk_finish_plug(&plug);
450 }
4660f9c0 451 f2fs_sync_fs(sbi->sb, true);
42190d2a 452 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 453 }
4660f9c0
JK		 454}
455
20fda56b
KM		 456static int __submit_flush_wait(struct f2fs_sb_info *sbi,
457 struct block_device *bdev)
3c62be17
JK		 458{
459 struct bio *bio = f2fs_bio_alloc(0);
460 int ret;
461
3adc5fcb 462 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3c62be17
JK		 463 bio->bi_bdev = bdev;
464 ret = submit_bio_wait(bio);
465 bio_put(bio);
20fda56b
KM		 466
467 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
468 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 469 return ret;
470}
471
472static int submit_flush_wait(struct f2fs_sb_info *sbi)
473{
20fda56b 474 int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev);
3c62be17
JK		 475 int i;
476
20fda56b
KM		 477 if (!sbi->s_ndevs || ret)
478 return ret;
479
480 for (i = 1; i < sbi->s_ndevs; i++) {
481 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
482 if (ret)
483 break;
3c62be17
JK		 484 }
485 return ret;
486}
487
2163d198 488static int issue_flush_thread(void *data)
6b4afdd7
JK		 489{
490 struct f2fs_sb_info *sbi = data;
b01a9201 491 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 492 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 493repeat:
494 if (kthread_should_stop())
495 return 0;
496
dc6febb6
CY		 497 sb_start_intwrite(sbi->sb);
498
721bd4d5 499 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 500 struct flush_cmd *cmd, *next;
501 int ret;
502
721bd4d5
GZ		 503 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
504 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
505
3c62be17 506 ret = submit_flush_wait(sbi);
8b8dd65f
CY		 507 atomic_inc(&fcc->issued_flush);
508
721bd4d5
GZ		 509 llist_for_each_entry_safe(cmd, next,
510 fcc->dispatch_list, llnode) {
6b4afdd7 511 cmd->ret = ret;
6b4afdd7
JK		 512 complete(&cmd->wait);
513 }
a688b9d9 514 fcc->dispatch_list = NULL;
6b4afdd7
JK		 515 }
516
dc6febb6
CY		 517 sb_end_intwrite(sbi->sb);
518
a688b9d9 519 wait_event_interruptible(*q,
721bd4d5 520 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 521 goto repeat;
522}
523
524int f2fs_issue_flush(struct f2fs_sb_info *sbi)
525{
b01a9201 526 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 527 struct flush_cmd cmd;
8b8dd65f 528 int ret;
6b4afdd7 529
0f7b2abd
JK		 530 if (test_opt(sbi, NOBARRIER))
531 return 0;
532
8b8dd65f
CY		 533 if (!test_opt(sbi, FLUSH_MERGE)) {
534 ret = submit_flush_wait(sbi);
535 atomic_inc(&fcc->issued_flush);
536 return ret;
537 }
740432f8 538
edd748e6 539 if (atomic_inc_return(&fcc->issing_flush) == 1) {
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
721bd4d5 549 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 550
6f890df0
CY		 551 /* update issue_list before we wake up issue_flush thread */
552 smp_mb();
553
554 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 555 wake_up(&fcc->flush_wait_queue);
6b4afdd7 556
5eba8c5d
JK		 557 if (fcc->f2fs_issue_flush) {
558 wait_for_completion(&cmd.wait);
8b8dd65f 559 atomic_dec(&fcc->issing_flush);
5eba8c5d
JK		 560 } else {
561 llist_del_all(&fcc->issue_list);
8b8dd65f 562 atomic_set(&fcc->issing_flush, 0);
5eba8c5d 563 }
adf8d90b
CY		 564
565 return cmd.ret;
6b4afdd7
JK		 566}
567
2163d198
GZ		 568int create_flush_cmd_control(struct f2fs_sb_info *sbi)
569{
570 dev_t dev = sbi->sb->s_bdev->bd_dev;
571 struct flush_cmd_control *fcc;
572 int err = 0;
573
b01a9201
JK		 574 if (SM_I(sbi)->fcc_info) {
575 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 576 if (fcc->f2fs_issue_flush)
577 return err;
5eba8c5d
JK		 578 goto init_thread;
579 }
580
2163d198
GZ		 581 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
582 if (!fcc)
583 return -ENOMEM;
8b8dd65f
CY		 584 atomic_set(&fcc->issued_flush, 0);
585 atomic_set(&fcc->issing_flush, 0);
2163d198 586 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 587 init_llist_head(&fcc->issue_list);
b01a9201 588 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 589 if (!test_opt(sbi, FLUSH_MERGE))
590 return err;
591
5eba8c5d 592init_thread:
2163d198
GZ		 593 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
594 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
595 if (IS_ERR(fcc->f2fs_issue_flush)) {
596 err = PTR_ERR(fcc->f2fs_issue_flush);
597 kfree(fcc);
b01a9201 598 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 599 return err;
600 }
2163d198
GZ		 601
602 return err;
603}
604
5eba8c5d 605void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 606{
b01a9201 607 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 608
5eba8c5d
JK		 609 if (fcc && fcc->f2fs_issue_flush) {
610 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
611
612 fcc->f2fs_issue_flush = NULL;
613 kthread_stop(flush_thread);
614 }
615 if (free) {
616 kfree(fcc);
b01a9201 617 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 618 }
2163d198
GZ		 619}
620
351df4b2
JK		 621static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
622 enum dirty_type dirty_type)
623{
624 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
625
626 /* need not be added */
627 if (IS_CURSEG(sbi, segno))
628 return;
629
630 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
631 dirty_i->nr_dirty[dirty_type]++;
632
633 if (dirty_type == DIRTY) {
634 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 635 enum dirty_type t = sentry->type;
b2f2c390 636
ec325b52
JK		 637 if (unlikely(t >= DIRTY)) {
638 f2fs_bug_on(sbi, 1);
639 return;
640 }
4625d6aa
CL		 641 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
642 dirty_i->nr_dirty[t]++;
351df4b2
JK		 643 }
644}
645
646static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
647 enum dirty_type dirty_type)
648{
649 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
650
651 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
652 dirty_i->nr_dirty[dirty_type]--;
653
654 if (dirty_type == DIRTY) {
4625d6aa
CL		 655 struct seg_entry *sentry = get_seg_entry(sbi, segno);
656 enum dirty_type t = sentry->type;
657
658 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
659 dirty_i->nr_dirty[t]--;
b2f2c390 660
302bd348 661 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 662 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 663 dirty_i->victim_secmap);
351df4b2
JK		 664 }
665}
666
0a8165d7 667/*
351df4b2
JK		 668 * Should not occur error such as -ENOMEM.
669 * Adding dirty entry into seglist is not critical operation.
670 * If a given segment is one of current working segments, it won't be added.
671 */
8d8451af 672static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 673{
674 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
675 unsigned short valid_blocks;
676
677 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
678 return;
679
680 mutex_lock(&dirty_i->seglist_lock);
681
302bd348 682 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 683
684 if (valid_blocks == 0) {
685 __locate_dirty_segment(sbi, segno, PRE);
686 __remove_dirty_segment(sbi, segno, DIRTY);
687 } else if (valid_blocks < sbi->blocks_per_seg) {
688 __locate_dirty_segment(sbi, segno, DIRTY);
689 } else {
690 /* Recovery routine with SSR needs this */
691 __remove_dirty_segment(sbi, segno, DIRTY);
692 }
693
694 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 695}
696
004b6862 697static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 698 struct block_device *bdev, block_t lstart,
699 block_t start, block_t len)
275b66b0 700{
0b54fb84 701 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 702 struct list_head *pend_list;
b01a9201 703 struct discard_cmd *dc;
275b66b0 704
ba48a33e
CY		 705 f2fs_bug_on(sbi, !len);
706
707 pend_list = &dcc->pend_list[plist_idx(len)];
708
b01a9201
JK		 709 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
710 INIT_LIST_HEAD(&dc->list);
c81abe34 711 dc->bdev = bdev;
b01a9201 712 dc->lstart = lstart;
c81abe34 713 dc->start = start;
b01a9201 714 dc->len = len;
ec9895ad 715 dc->ref = 0;
15469963 716 dc->state = D_PREP;
c81abe34 717 dc->error = 0;
b01a9201 718 init_completion(&dc->wait);
22d375dd 719 list_add_tail(&dc->list, pend_list);
5f32366a 720 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 721 dcc->undiscard_blks += len;
004b6862
CY		 722
723 return dc;
724}
725
726static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
727 struct block_device *bdev, block_t lstart,
728 block_t start, block_t len,
729 struct rb_node *parent, struct rb_node **p)
730{
731 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
732 struct discard_cmd *dc;
733
734 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
735
736 rb_link_node(&dc->rb_node, parent, p);
737 rb_insert_color(&dc->rb_node, &dcc->root);
738
739 return dc;
15469963
JK		 740}
741
004b6862
CY		 742static void __detach_discard_cmd(struct discard_cmd_control *dcc,
743 struct discard_cmd *dc)
15469963 744{
dcc9165d 745 if (dc->state == D_DONE)
004b6862
CY		 746 atomic_dec(&dcc->issing_discard);
747
748 list_del(&dc->list);
749 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 750 dcc->undiscard_blks -= dc->len;
004b6862
CY		 751
752 kmem_cache_free(discard_cmd_slab, dc);
753
754 atomic_dec(&dcc->discard_cmd_cnt);
755}
756
757static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
758 struct discard_cmd *dc)
759{
760 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 761
d9703d90
CY		 762 f2fs_bug_on(sbi, dc->ref);
763
c81abe34
JK		 764 if (dc->error == -EOPNOTSUPP)
765 dc->error = 0;
15469963 766
c81abe34 767 if (dc->error)
15469963 768 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 769 "Issue discard(%u, %u, %u) failed, ret: %d",
770 dc->lstart, dc->start, dc->len, dc->error);
004b6862 771 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 772}
773
c81abe34
JK		 774static void f2fs_submit_discard_endio(struct bio *bio)
775{
776 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
777
4e4cbee9 778 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 779 dc->state = D_DONE;
e31b9821 780 complete_all(&dc->wait);
c81abe34
JK		 781 bio_put(bio);
782}
783
6915ea9d
CY		 784void __check_sit_bitmap(struct f2fs_sb_info *sbi,
785 block_t start, block_t end)
786{
787#ifdef CONFIG_F2FS_CHECK_FS
788 struct seg_entry *sentry;
789 unsigned int segno;
790 block_t blk = start;
791 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
792 unsigned long *map;
793
794 while (blk < end) {
795 segno = GET_SEGNO(sbi, blk);
796 sentry = get_seg_entry(sbi, segno);
797 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
798
008396e1
YS		 799 if (end < START_BLOCK(sbi, segno + 1))
800 size = GET_BLKOFF_FROM_SEG0(sbi, end);
801 else
802 size = max_blocks;
6915ea9d
CY		 803 map = (unsigned long *)(sentry->cur_valid_map);
804 offset = __find_rev_next_bit(map, size, offset);
805 f2fs_bug_on(sbi, offset != size);
008396e1 806 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 807 }
808#endif
809}
810
c81abe34
JK		 811/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
812static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
813 struct discard_cmd *dc)
814{
815 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
816 struct bio *bio = NULL;
817
818 if (dc->state != D_PREP)
819 return;
820
0243a5f9
CY		 821 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
822
c81abe34
JK		 823 dc->error = __blkdev_issue_discard(dc->bdev,
824 SECTOR_FROM_BLOCK(dc->start),
825 SECTOR_FROM_BLOCK(dc->len),
826 GFP_NOFS, 0, &bio);
827 if (!dc->error) {
828 /* should keep before submission to avoid D_DONE right away */
829 dc->state = D_SUBMIT;
8b8dd65f
CY		 830 atomic_inc(&dcc->issued_discard);
831 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 832 if (bio) {
833 bio->bi_private = dc;
834 bio->bi_end_io = f2fs_submit_discard_endio;
835 bio->bi_opf |= REQ_SYNC;
836 submit_bio(bio);
46f84c2c 837 list_move_tail(&dc->list, &dcc->wait_list);
6915ea9d 838 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 839
840 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 841 }
842 } else {
843 __remove_discard_cmd(sbi, dc);
844 }
845}
846
004b6862
CY		 847static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
848 struct block_device *bdev, block_t lstart,
849 block_t start, block_t len,
850 struct rb_node **insert_p,
851 struct rb_node *insert_parent)
c81abe34 852{
004b6862
CY		 853 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
854 struct rb_node **p = &dcc->root.rb_node;
855 struct rb_node *parent = NULL;
856 struct discard_cmd *dc = NULL;
857
858 if (insert_p && insert_parent) {
859 parent = insert_parent;
860 p = insert_p;
861 goto do_insert;
862 }
c81abe34 863
004b6862
CY		 864 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
865do_insert:
866 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
867 if (!dc)
868 return NULL;
c81abe34 869
004b6862 870 return dc;
c81abe34
JK		 871}
872
ba48a33e
CY		 873static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
874 struct discard_cmd *dc)
875{
876 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
877}
878
3d6a650f
YH		 879static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
880 struct discard_cmd *dc, block_t blkaddr)
881{
ba48a33e 882 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 883 struct discard_info di = dc->di;
884 bool modified = false;
3d6a650f 885
004b6862 886 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 887 __remove_discard_cmd(sbi, dc);
888 return;
889 }
890
d84d1cbd
CY		 891 dcc->undiscard_blks -= di.len;
892
004b6862 893 if (blkaddr > di.lstart) {
3d6a650f 894 dc->len = blkaddr - dc->lstart;
d84d1cbd 895 dcc->undiscard_blks += dc->len;
ba48a33e 896 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 897 modified = true;
898 }
899
900 if (blkaddr < di.lstart + di.len - 1) {
901 if (modified) {
902 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
903 di.start + blkaddr + 1 - di.lstart,
904 di.lstart + di.len - 1 - blkaddr,
905 NULL, NULL);
906 } else {
907 dc->lstart++;
908 dc->len--;
909 dc->start++;
d84d1cbd 910 dcc->undiscard_blks += dc->len;
ba48a33e 911 __relocate_discard_cmd(dcc, dc);
004b6862 912 }
3d6a650f
YH		 913 }
914}
915
004b6862
CY		 916static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
917 struct block_device *bdev, block_t lstart,
918 block_t start, block_t len)
275b66b0 919{
0b54fb84 920 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 921 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
922 struct discard_cmd *dc;
923 struct discard_info di = {0};
924 struct rb_node **insert_p = NULL, *insert_parent = NULL;
925 block_t end = lstart + len;
275b66b0 926
15469963 927 mutex_lock(&dcc->cmd_lock);
40465257 928
004b6862
CY		 929 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
930 NULL, lstart,
931 (struct rb_entry **)&prev_dc,
932 (struct rb_entry **)&next_dc,
933 &insert_p, &insert_parent, true);
934 if (dc)
935 prev_dc = dc;
936
937 if (!prev_dc) {
938 di.lstart = lstart;
939 di.len = next_dc ? next_dc->lstart - lstart : len;
940 di.len = min(di.len, len);
941 di.start = start;
22d375dd 942 }
15469963 943
004b6862
CY		 944 while (1) {
945 struct rb_node *node;
946 bool merged = false;
947 struct discard_cmd *tdc = NULL;
948
949 if (prev_dc) {
950 di.lstart = prev_dc->lstart + prev_dc->len;
951 if (di.lstart < lstart)
952 di.lstart = lstart;
953 if (di.lstart >= end)
954 break;
955
956 if (!next_dc || next_dc->lstart > end)
957 di.len = end - di.lstart;
958 else
959 di.len = next_dc->lstart - di.lstart;
960 di.start = start + di.lstart - lstart;
961 }
962
963 if (!di.len)
964 goto next;
965
966 if (prev_dc && prev_dc->state == D_PREP &&
967 prev_dc->bdev == bdev &&
968 __is_discard_back_mergeable(&di, &prev_dc->di)) {
969 prev_dc->di.len += di.len;
d84d1cbd 970 dcc->undiscard_blks += di.len;
ba48a33e 971 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 972 di = prev_dc->di;
973 tdc = prev_dc;
974 merged = true;
975 }
976
977 if (next_dc && next_dc->state == D_PREP &&
978 next_dc->bdev == bdev &&
979 __is_discard_front_mergeable(&di, &next_dc->di)) {
980 next_dc->di.lstart = di.lstart;
981 next_dc->di.len += di.len;
982 next_dc->di.start = di.start;
d84d1cbd 983 dcc->undiscard_blks += di.len;
ba48a33e 984 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 985 if (tdc)
986 __remove_discard_cmd(sbi, tdc);
004b6862 987 merged = true;
4e6a8d9b 988 }
004b6862 989
df0f6b44 990 if (!merged) {
004b6862
CY		 991 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
992 di.len, NULL, NULL);
df0f6b44 993 }
004b6862
CY		 994 next:
995 prev_dc = next_dc;
996 if (!prev_dc)
997 break;
998
999 node = rb_next(&prev_dc->rb_node);
1000 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1001 }
1002
1003 mutex_unlock(&dcc->cmd_lock);
1004}
1005
1006static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1007 struct block_device *bdev, block_t blkstart, block_t blklen)
1008{
1009 block_t lblkstart = blkstart;
1010
0243a5f9 1011 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1012
1013 if (sbi->s_ndevs) {
1014 int devi = f2fs_target_device_index(sbi, blkstart);
1015
1016 blkstart -= FDEV(devi).start_blk;
1017 }
1018 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1019 return 0;
1020}
1021
969d1b18 1022static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
bd5b0738
CY		 1023{
1024 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1025 struct list_head *pend_list;
1026 struct discard_cmd *dc, *tmp;
1027 struct blk_plug plug;
969d1b18
CY		 1028 int iter = 0, issued = 0;
1029 int i;
bd5b0738
CY		 1030
1031 mutex_lock(&dcc->cmd_lock);
963932a9
YH		 1032 f2fs_bug_on(sbi,
1033 !__check_rb_tree_consistence(sbi, &dcc->root));
bd5b0738 1034 blk_start_plug(&plug);
969d1b18
CY		 1035 for (i = MAX_PLIST_NUM - 1;
1036 i >= 0 && plist_issue(dcc->pend_list_tag[i]); i--) {
bd5b0738
CY		 1037 pend_list = &dcc->pend_list[i];
1038 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1039 f2fs_bug_on(sbi, dc->state != D_PREP);
1040
969d1b18
CY		 1041 /* Hurry up to finish fstrim */
1042 if (dcc->pend_list_tag[i] & P_TRIM) {
1043 __submit_discard_cmd(sbi, dc);
1044 issued++;
1045 continue;
1046 }
1047
1048 if (!issue_cond || is_idle(sbi)) {
1049 issued++;
bd5b0738 1050 __submit_discard_cmd(sbi, dc);
969d1b18 1051 }
bd5b0738
CY		 1052 if (issue_cond && iter++ > DISCARD_ISSUE_RATE)
1053 goto out;
1054 }
969d1b18
CY		 1055 if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM)
1056 dcc->pend_list_tag[i] &= (~P_TRIM);
bd5b0738
CY		 1057 }
1058out:
1059 blk_finish_plug(&plug);
1060 mutex_unlock(&dcc->cmd_lock);
969d1b18
CY		 1061
1062 return issued;
1063}
1064
1065static void __drop_discard_cmd(struct f2fs_sb_info *sbi)
1066{
1067 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1068 struct list_head *pend_list;
1069 struct discard_cmd *dc, *tmp;
1070 int i;
1071
1072 mutex_lock(&dcc->cmd_lock);
1073 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1074 pend_list = &dcc->pend_list[i];
1075 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1076 f2fs_bug_on(sbi, dc->state != D_PREP);
1077 __remove_discard_cmd(sbi, dc);
1078 }
1079 }
1080 mutex_unlock(&dcc->cmd_lock);
bd5b0738
CY		 1081}
1082
2a510c00
CY		 1083static void __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1084 struct discard_cmd *dc)
1085{
1086 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1087
1088 wait_for_completion_io(&dc->wait);
1089 mutex_lock(&dcc->cmd_lock);
1090 f2fs_bug_on(sbi, dc->state != D_DONE);
1091 dc->ref--;
1092 if (!dc->ref)
1093 __remove_discard_cmd(sbi, dc);
1094 mutex_unlock(&dcc->cmd_lock);
1095}
1096
63a94fa1
CY		 1097static void __wait_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1098{
1099 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1100 struct list_head *wait_list = &(dcc->wait_list);
1101 struct discard_cmd *dc, *tmp;
6afae633
CY		 1102 bool need_wait;
1103
1104next:
1105 need_wait = false;
63a94fa1
CY		 1106
1107 mutex_lock(&dcc->cmd_lock);
1108 list_for_each_entry_safe(dc, tmp, wait_list, list) {
6afae633 1109 if (!wait_cond || (dc->state == D_DONE && !dc->ref)) {
63a94fa1
CY		 1110 wait_for_completion_io(&dc->wait);
1111 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1112 } else {
1113 dc->ref++;
1114 need_wait = true;
1115 break;
63a94fa1
CY		 1116 }
1117 }
1118 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1119
1120 if (need_wait) {
2a510c00 1121 __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1122 goto next;
1123 }
63a94fa1
CY		 1124}
1125
004b6862
CY		 1126/* This should be covered by global mutex, &sit_i->sentry_lock */
1127void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1128{
1129 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1130 struct discard_cmd *dc;
ec9895ad 1131 bool need_wait = false;
004b6862
CY		 1132
1133 mutex_lock(&dcc->cmd_lock);
004b6862
CY		 1134 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1135 if (dc) {
ec9895ad
CY		 1136 if (dc->state == D_PREP) {
1137 __punch_discard_cmd(sbi, dc, blkaddr);
1138 } else {
1139 dc->ref++;
1140 need_wait = true;
1141 }
275b66b0 1142 }
d431413f 1143 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1144
2a510c00
CY		 1145 if (need_wait)
1146 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1147}
1148
cce13252
CY		 1149void stop_discard_thread(struct f2fs_sb_info *sbi)
1150{
1151 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1152
1153 if (dcc && dcc->f2fs_issue_discard) {
1154 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1155
1156 dcc->f2fs_issue_discard = NULL;
1157 kthread_stop(discard_thread);
ec9895ad 1158 }
d431413f
CY		 1159}
1160
1161/* This comes from f2fs_put_super */
1162void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1163{
bd5b0738 1164 __issue_discard_cmd(sbi, false);
969d1b18 1165 __drop_discard_cmd(sbi);
63a94fa1 1166 __wait_discard_cmd(sbi, false);
275b66b0
CY		 1167}
1168
969d1b18
CY		 1169static void mark_discard_range_all(struct f2fs_sb_info *sbi)
1170{
1171 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1172 int i;
1173
1174 mutex_lock(&dcc->cmd_lock);
1175 for (i = 0; i < MAX_PLIST_NUM; i++)
1176 dcc->pend_list_tag[i] |= P_TRIM;
1177 mutex_unlock(&dcc->cmd_lock);
1178}
1179
15469963
JK		 1180static int issue_discard_thread(void *data)
1181{
1182 struct f2fs_sb_info *sbi = data;
1183 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1184 wait_queue_head_t *q = &dcc->discard_wait_queue;
969d1b18
CY		 1185 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1186 int issued;
15469963 1187
1d7be270 1188 set_freezable();
15469963 1189
1d7be270 1190 do {
969d1b18
CY		 1191 wait_event_interruptible_timeout(*q,
1192 kthread_should_stop() || freezing(current) ||
1193 dcc->discard_wake,
1194 msecs_to_jiffies(wait_ms));
1d7be270
JK		 1195 if (try_to_freeze())
1196 continue;
1197 if (kthread_should_stop())
1198 return 0;
15469963 1199
5f656541 1200 if (dcc->discard_wake) {
969d1b18 1201 dcc->discard_wake = 0;
5f656541
JK		 1202 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1203 mark_discard_range_all(sbi);
1204 }
969d1b18 1205
dc6febb6
CY		 1206 sb_start_intwrite(sbi->sb);
1207
969d1b18
CY		 1208 issued = __issue_discard_cmd(sbi, true);
1209 if (issued) {
1210 __wait_discard_cmd(sbi, true);
1211 wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1212 } else {
1213 wait_ms = DEF_MAX_DISCARD_ISSUE_TIME;
1214 }
1d7be270 1215
dc6febb6
CY		 1216 sb_end_intwrite(sbi->sb);
1217
1d7be270
JK		 1218 } while (!kthread_should_stop());
1219 return 0;
15469963
JK		 1220}
1221
f46e8809 1222#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1223static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1224 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1225{
92592285 1226 sector_t sector, nr_sects;
10a875f8 1227 block_t lblkstart = blkstart;
3c62be17
JK		 1228 int devi = 0;
1229
1230 if (sbi->s_ndevs) {
1231 devi = f2fs_target_device_index(sbi, blkstart);
1232 blkstart -= FDEV(devi).start_blk;
1233 }
f46e8809
DLM		 1234
1235 /*
1236 * We need to know the type of the zone: for conventional zones,
1237 * use regular discard if the drive supports it. For sequential
1238 * zones, reset the zone write pointer.
1239 */
3c62be17 1240 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1241
1242 case BLK_ZONE_TYPE_CONVENTIONAL:
1243 if (!blk_queue_discard(bdev_get_queue(bdev)))
1244 return 0;
c81abe34 1245 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1246 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1247 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1248 sector = SECTOR_FROM_BLOCK(blkstart);
1249 nr_sects = SECTOR_FROM_BLOCK(blklen);
1250
1251 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1252 nr_sects != bdev_zone_sectors(bdev)) {
1253 f2fs_msg(sbi->sb, KERN_INFO,
1254 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1255 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1256 blkstart, blklen);
1257 return -EIO;
1258 }
d50aaeec 1259 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1260 return blkdev_reset_zones(bdev, sector,
1261 nr_sects, GFP_NOFS);
1262 default:
1263 /* Unknown zone type: broken device ? */
1264 return -EIO;
1265 }
1266}
1267#endif
1268
3c62be17
JK		 1269static int __issue_discard_async(struct f2fs_sb_info *sbi,
1270 struct block_device *bdev, block_t blkstart, block_t blklen)
1271{
1272#ifdef CONFIG_BLK_DEV_ZONED
1273 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1274 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1275 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1276#endif
c81abe34 1277 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1278}
1279
1e87a78d 1280static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1281 block_t blkstart, block_t blklen)
1282{
3c62be17
JK		 1283 sector_t start = blkstart, len = 0;
1284 struct block_device *bdev;
a66cdd98
JK		 1285 struct seg_entry *se;
1286 unsigned int offset;
1287 block_t i;
3c62be17
JK		 1288 int err = 0;
1289
1290 bdev = f2fs_target_device(sbi, blkstart, NULL);
1291
1292 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1293 if (i != start) {
1294 struct block_device *bdev2 =
1295 f2fs_target_device(sbi, i, NULL);
1296
1297 if (bdev2 != bdev) {
1298 err = __issue_discard_async(sbi, bdev,
1299 start, len);
1300 if (err)
1301 return err;
1302 bdev = bdev2;
1303 start = i;
1304 len = 0;
1305 }
1306 }
a66cdd98 1307
a66cdd98
JK		 1308 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1309 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1310
1311 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1312 sbi->discard_blks--;
1313 }
f46e8809 1314
3c62be17
JK		 1315 if (len)
1316 err = __issue_discard_async(sbi, bdev, start, len);
1317 return err;
1e87a78d
JK		 1318}
1319
25290fa5
JK		 1320static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1321 bool check_only)
adf4983b 1322{
b2955550
JK		 1323 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1324 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1325 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1326 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1327 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1328 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1329 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1330 unsigned int start = 0, end = -1;
c473f1a9 1331 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1332 struct discard_entry *de = NULL;
46f84c2c 1333 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1334 int i;
1335
3e025740 1336 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1337 return false;
b2955550 1338
a66cdd98
JK		 1339 if (!force) {
1340 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1341 SM_I(sbi)->dcc_info->nr_discards >=
1342 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1343 return false;
4b2fecc8
JK		 1344 }
1345
b2955550
JK		 1346 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1347 for (i = 0; i < entries; i++)
a66cdd98 1348 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1349 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1350
0b54fb84
JK		 1351 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1352 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1353 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1354 if (start >= max_blocks)
1355 break;
1356
1357 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1358 if (force && start && end != max_blocks
1359 && (end - start) < cpc->trim_minlen)
1360 continue;
1361
25290fa5
JK		 1362 if (check_only)
1363 return true;
1364
a7eeb823
CY		 1365 if (!de) {
1366 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1367 GFP_F2FS_ZERO);
1368 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1369 list_add_tail(&de->list, head);
1370 }
1371
1372 for (i = start; i < end; i++)
1373 __set_bit_le(i, (void *)de->discard_map);
1374
1375 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1376 }
25290fa5 1377 return false;
b2955550
JK		 1378}
1379
4b2fecc8
JK		 1380void release_discard_addrs(struct f2fs_sb_info *sbi)
1381{
46f84c2c 1382 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1383 struct discard_entry *entry, *this;
1384
1385 /* drop caches */
1386 list_for_each_entry_safe(entry, this, head, list) {
1387 list_del(&entry->list);
1388 kmem_cache_free(discard_entry_slab, entry);
1389 }
1390}
1391
0a8165d7 1392/*
351df4b2
JK		 1393 * Should call clear_prefree_segments after checkpoint is done.
1394 */
1395static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1396{
1397 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1398 unsigned int segno;
351df4b2
JK		 1399
1400 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1401 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1402 __set_test_and_free(sbi, segno);
351df4b2
JK		 1403 mutex_unlock(&dirty_i->seglist_lock);
1404}
1405
836b5a63 1406void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1407{
969d1b18
CY		 1408 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1409 struct list_head *head = &dcc->entry_list;
2d7b822a 1410 struct discard_entry *entry, *this;
351df4b2 1411 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1412 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1413 unsigned int start = 0, end = -1;
36abef4e 1414 unsigned int secno, start_segno;
c473f1a9 1415 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK		 1416
1417 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1418
351df4b2 1419 while (1) {
29e59c14 1420 int i;
7cd8558b
JK		 1421 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1422 if (start >= MAIN_SEGS(sbi))
351df4b2 1423 break;
7cd8558b
JK		 1424 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1425 start + 1);
29e59c14
CL		 1426
1427 for (i = start; i < end; i++)
1428 clear_bit(i, prefree_map);
1429
1430 dirty_i->nr_dirty[PRE] -= end - start;
1431
650d3c4e 1432 if (!test_opt(sbi, DISCARD))
29e59c14 1433 continue;
351df4b2 1434
650d3c4e
YH		 1435 if (force && start >= cpc->trim_start &&
1436 (end - 1) <= cpc->trim_end)
1437 continue;
1438
36abef4e
JK		 1439 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1440 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1441 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1442 continue;
1443 }
1444next:
4ddb1a4d
JK		 1445 secno = GET_SEC_FROM_SEG(sbi, start);
1446 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1447 if (!IS_CURSEC(sbi, secno) &&
302bd348 1448 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1449 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1450 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1451
1452 start = start_segno + sbi->segs_per_sec;
1453 if (start < end)
1454 goto next;
8b107f5b
JK		 1455 else
1456 end = start - 1;
351df4b2
JK		 1457 }
1458 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1459
1460 /* send small discards */
2d7b822a 1461 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1462 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1463 bool is_valid = test_bit_le(0, entry->discard_map);
1464
1465find_next:
1466 if (is_valid) {
1467 next_pos = find_next_zero_bit_le(entry->discard_map,
1468 sbi->blocks_per_seg, cur_pos);
1469 len = next_pos - cur_pos;
1470
acfd2810
DLM		 1471 if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1472 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1473 goto skip;
1474
1475 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1476 len);
1477 cpc->trimmed += len;
1478 total_len += len;
1479 } else {
1480 next_pos = find_next_bit_le(entry->discard_map,
1481 sbi->blocks_per_seg, cur_pos);
1482 }
836b5a63 1483skip:
a7eeb823
CY		 1484 cur_pos = next_pos;
1485 is_valid = !is_valid;
1486
1487 if (cur_pos < sbi->blocks_per_seg)
1488 goto find_next;
1489
b2955550 1490 list_del(&entry->list);
969d1b18 1491 dcc->nr_discards -= total_len;
b2955550
JK		 1492 kmem_cache_free(discard_entry_slab, entry);
1493 }
34e159da 1494
01983c71 1495 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1496}
1497
8ed59745 1498static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1499{
15469963 1500 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1501 struct discard_cmd_control *dcc;
ba48a33e 1502 int err = 0, i;
0b54fb84
JK		 1503
1504 if (SM_I(sbi)->dcc_info) {
1505 dcc = SM_I(sbi)->dcc_info;
1506 goto init_thread;
1507 }
1508
1509 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1510 if (!dcc)
1511 return -ENOMEM;
1512
969d1b18 1513 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1514 INIT_LIST_HEAD(&dcc->entry_list);
969d1b18 1515 for (i = 0; i < MAX_PLIST_NUM; i++) {
ba48a33e 1516 INIT_LIST_HEAD(&dcc->pend_list[i]);
969d1b18
CY		 1517 if (i >= dcc->discard_granularity - 1)
1518 dcc->pend_list_tag[i] |= P_ACTIVE;
1519 }
46f84c2c 1520 INIT_LIST_HEAD(&dcc->wait_list);
15469963 1521 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1522 atomic_set(&dcc->issued_discard, 0);
1523 atomic_set(&dcc->issing_discard, 0);
5f32366a 1524 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1525 dcc->nr_discards = 0;
d618ebaf 1526 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1527 dcc->undiscard_blks = 0;
004b6862 1528 dcc->root = RB_ROOT;
0b54fb84 1529
15469963 1530 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1531 SM_I(sbi)->dcc_info = dcc;
1532init_thread:
15469963
JK		 1533 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1534 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1535 if (IS_ERR(dcc->f2fs_issue_discard)) {
1536 err = PTR_ERR(dcc->f2fs_issue_discard);
1537 kfree(dcc);
1538 SM_I(sbi)->dcc_info = NULL;
1539 return err;
1540 }
1541
0b54fb84
JK		 1542 return err;
1543}
1544
f099405f 1545static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1546{
1547 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1548
f099405f
CY		 1549 if (!dcc)
1550 return;
1551
cce13252 1552 stop_discard_thread(sbi);
f099405f
CY		 1553
1554 kfree(dcc);
1555 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1556}
1557
184a5cd2 1558static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1559{
1560 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1561
1562 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1563 sit_i->dirty_sentries++;
184a5cd2
CY		 1564 return false;
1565 }
1566
1567 return true;
351df4b2
JK		 1568}
1569
1570static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1571 unsigned int segno, int modified)
1572{
1573 struct seg_entry *se = get_seg_entry(sbi, segno);
1574 se->type = type;
1575 if (modified)
1576 __mark_sit_entry_dirty(sbi, segno);
1577}
1578
1579static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1580{
1581 struct seg_entry *se;
1582 unsigned int segno, offset;
1583 long int new_vblocks;
6415fedc
YS		 1584 bool exist;
1585#ifdef CONFIG_F2FS_CHECK_FS
1586 bool mir_exist;
1587#endif
351df4b2
JK		 1588
1589 segno = GET_SEGNO(sbi, blkaddr);
1590
1591 se = get_seg_entry(sbi, segno);
1592 new_vblocks = se->valid_blocks + del;
491c0854 1593 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1594
9850cf4a 1595 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1596 (new_vblocks > sbi->blocks_per_seg)));
1597
1598 se->valid_blocks = new_vblocks;
1599 se->mtime = get_mtime(sbi);
1600 SIT_I(sbi)->max_mtime = se->mtime;
1601
1602 /* Update valid block bitmap */
1603 if (del > 0) {
6415fedc 1604 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1605#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1606 mir_exist = f2fs_test_and_set_bit(offset,
1607 se->cur_valid_map_mir);
1608 if (unlikely(exist != mir_exist)) {
1609 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1610 "when setting bitmap, blk:%u, old bit:%d",
1611 blkaddr, exist);
05796763 1612 f2fs_bug_on(sbi, 1);
6415fedc 1613 }
355e7891 1614#endif
6415fedc
YS		 1615 if (unlikely(exist)) {
1616 f2fs_msg(sbi->sb, KERN_ERR,
1617 "Bitmap was wrongly set, blk:%u", blkaddr);
1618 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1619 se->valid_blocks--;
1620 del = 0;
355e7891 1621 }
6415fedc 1622
3e025740
JK		 1623 if (f2fs_discard_en(sbi) &&
1624 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1625 sbi->discard_blks--;
720037f9
JK		 1626
1627 /* don't overwrite by SSR to keep node chain */
1628 if (se->type == CURSEG_WARM_NODE) {
1629 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1630 se->ckpt_valid_blocks++;
1631 }
351df4b2 1632 } else {
6415fedc 1633 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1634#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1635 mir_exist = f2fs_test_and_clear_bit(offset,
1636 se->cur_valid_map_mir);
1637 if (unlikely(exist != mir_exist)) {
1638 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1639 "when clearing bitmap, blk:%u, old bit:%d",
1640 blkaddr, exist);
05796763 1641 f2fs_bug_on(sbi, 1);
6415fedc 1642 }
355e7891 1643#endif
6415fedc
YS		 1644 if (unlikely(!exist)) {
1645 f2fs_msg(sbi->sb, KERN_ERR,
1646 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1647 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1648 se->valid_blocks++;
1649 del = 0;
355e7891 1650 }
6415fedc 1651
3e025740
JK		 1652 if (f2fs_discard_en(sbi) &&
1653 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1654 sbi->discard_blks++;
351df4b2
JK		 1655 }
1656 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1657 se->ckpt_valid_blocks += del;
1658
1659 __mark_sit_entry_dirty(sbi, segno);
1660
1661 /* update total number of valid blocks to be written in ckpt area */
1662 SIT_I(sbi)->written_valid_blocks += del;
1663
1664 if (sbi->segs_per_sec > 1)
1665 get_sec_entry(sbi, segno)->valid_blocks += del;
1666}
1667
5e443818 1668void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1669{
5e443818
JK		 1670 update_sit_entry(sbi, new, 1);
1671 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1672 update_sit_entry(sbi, old, -1);
1673
1674 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1675 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 1676}
1677
1678void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1679{
1680 unsigned int segno = GET_SEGNO(sbi, addr);
1681 struct sit_info *sit_i = SIT_I(sbi);
1682
9850cf4a 1683 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1684 if (addr == NEW_ADDR)
1685 return;
1686
1687 /* add it into sit main buffer */
1688 mutex_lock(&sit_i->sentry_lock);
1689
1690 update_sit_entry(sbi, addr, -1);
1691
1692 /* add it into dirty seglist */
1693 locate_dirty_segment(sbi, segno);
1694
1695 mutex_unlock(&sit_i->sentry_lock);
1696}
1697
6e2c64ad
JK		 1698bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1699{
1700 struct sit_info *sit_i = SIT_I(sbi);
1701 unsigned int segno, offset;
1702 struct seg_entry *se;
1703 bool is_cp = false;
1704
1705 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1706 return true;
1707
1708 mutex_lock(&sit_i->sentry_lock);
1709
1710 segno = GET_SEGNO(sbi, blkaddr);
1711 se = get_seg_entry(sbi, segno);
1712 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1713
1714 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1715 is_cp = true;
1716
1717 mutex_unlock(&sit_i->sentry_lock);
1718
1719 return is_cp;
1720}
1721
0a8165d7 1722/*
351df4b2
JK		 1723 * This function should be resided under the curseg_mutex lock
1724 */
1725static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1726 struct f2fs_summary *sum)
351df4b2
JK		 1727{
1728 struct curseg_info *curseg = CURSEG_I(sbi, type);
1729 void *addr = curseg->sum_blk;
e79efe3b 1730 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1731 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1732}
1733
0a8165d7 1734/*
351df4b2
JK		 1735 * Calculate the number of current summary pages for writing
1736 */
3fa06d7b 1737int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1738{
351df4b2 1739 int valid_sum_count = 0;
9a47938b 1740 int i, sum_in_page;
351df4b2
JK		 1741
1742 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1743 if (sbi->ckpt->alloc_type[i] == SSR)
1744 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1745 else {
1746 if (for_ra)
1747 valid_sum_count += le16_to_cpu(
1748 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1749 else
1750 valid_sum_count += curseg_blkoff(sbi, i);
1751 }
351df4b2
JK		 1752 }
1753
09cbfeaf 1754 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1755 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1756 if (valid_sum_count <= sum_in_page)
351df4b2 1757 return 1;
9a47938b 1758 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1759 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1760 return 2;
1761 return 3;
1762}
1763
0a8165d7 1764/*
351df4b2
JK		 1765 * Caller should put this summary page
1766 */
1767struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1768{
1769 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1770}
1771
381722d2 1772void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1773{
1774 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1775 void *dst = page_address(page);
1776
1777 if (src)
09cbfeaf 1778 memcpy(dst, src, PAGE_SIZE);
381722d2 1779 else
09cbfeaf 1780 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1781 set_page_dirty(page);
1782 f2fs_put_page(page, 1);
1783}
1784
381722d2
CY		 1785static void write_sum_page(struct f2fs_sb_info *sbi,
1786 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1787{
1788 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1789}
1790
b7ad7512
CY		 1791static void write_current_sum_page(struct f2fs_sb_info *sbi,
1792 int type, block_t blk_addr)
1793{
1794 struct curseg_info *curseg = CURSEG_I(sbi, type);
1795 struct page *page = grab_meta_page(sbi, blk_addr);
1796 struct f2fs_summary_block *src = curseg->sum_blk;
1797 struct f2fs_summary_block *dst;
1798
1799 dst = (struct f2fs_summary_block *)page_address(page);
1800
1801 mutex_lock(&curseg->curseg_mutex);
1802
1803 down_read(&curseg->journal_rwsem);
1804 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1805 up_read(&curseg->journal_rwsem);
1806
1807 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1808 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1809
1810 mutex_unlock(&curseg->curseg_mutex);
1811
1812 set_page_dirty(page);
1813 f2fs_put_page(page, 1);
1814}
1815
a7881893
JK		 1816static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1817{
1818 struct curseg_info *curseg = CURSEG_I(sbi, type);
1819 unsigned int segno = curseg->segno + 1;
1820 struct free_segmap_info *free_i = FREE_I(sbi);
1821
1822 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1823 return !test_bit(segno, free_i->free_segmap);
1824 return 0;
1825}
1826
0a8165d7 1827/*
351df4b2
JK		 1828 * Find a new segment from the free segments bitmap to right order
1829 * This function should be returned with success, otherwise BUG
1830 */
1831static void get_new_segment(struct f2fs_sb_info *sbi,
1832 unsigned int *newseg, bool new_sec, int dir)
1833{
1834 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1835 unsigned int segno, secno, zoneno;
7cd8558b 1836 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 1837 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1838 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 1839 unsigned int left_start = hint;
1840 bool init = true;
1841 int go_left = 0;
1842 int i;
1843
1a118ccf 1844 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1845
1846 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1847 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 1848 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
1849 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 1850 goto got_it;
1851 }
1852find_other_zone:
7cd8558b
JK		 1853 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1854 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1855 if (dir == ALLOC_RIGHT) {
1856 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1857 MAIN_SECS(sbi), 0);
1858 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1859 } else {
1860 go_left = 1;
1861 left_start = hint - 1;
1862 }
1863 }
1864 if (go_left == 0)
1865 goto skip_left;
1866
1867 while (test_bit(left_start, free_i->free_secmap)) {
1868 if (left_start > 0) {
1869 left_start--;
1870 continue;
1871 }
1872 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1873 MAIN_SECS(sbi), 0);
1874 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1875 break;
1876 }
1877 secno = left_start;
1878skip_left:
1879 hint = secno;
4ddb1a4d
JK		 1880 segno = GET_SEG_FROM_SEC(sbi, secno);
1881 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 1882
1883 /* give up on finding another zone */
1884 if (!init)
1885 goto got_it;
1886 if (sbi->secs_per_zone == 1)
1887 goto got_it;
1888 if (zoneno == old_zoneno)
1889 goto got_it;
1890 if (dir == ALLOC_LEFT) {
1891 if (!go_left && zoneno + 1 >= total_zones)
1892 goto got_it;
1893 if (go_left && zoneno == 0)
1894 goto got_it;
1895 }
1896 for (i = 0; i < NR_CURSEG_TYPE; i++)
1897 if (CURSEG_I(sbi, i)->zone == zoneno)
1898 break;
1899
1900 if (i < NR_CURSEG_TYPE) {
1901 /* zone is in user, try another */
1902 if (go_left)
1903 hint = zoneno * sbi->secs_per_zone - 1;
1904 else if (zoneno + 1 >= total_zones)
1905 hint = 0;
1906 else
1907 hint = (zoneno + 1) * sbi->secs_per_zone;
1908 init = false;
1909 goto find_other_zone;
1910 }
1911got_it:
1912 /* set it as dirty segment in free segmap */
9850cf4a 1913 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1914 __set_inuse(sbi, segno);
1915 *newseg = segno;
1a118ccf 1916 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1917}
1918
1919static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1920{
1921 struct curseg_info *curseg = CURSEG_I(sbi, type);
1922 struct summary_footer *sum_footer;
1923
1924 curseg->segno = curseg->next_segno;
4ddb1a4d 1925 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 1926 curseg->next_blkoff = 0;
1927 curseg->next_segno = NULL_SEGNO;
1928
1929 sum_footer = &(curseg->sum_blk->footer);
1930 memset(sum_footer, 0, sizeof(struct summary_footer));
1931 if (IS_DATASEG(type))
1932 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1933 if (IS_NODESEG(type))
1934 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1935 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1936}
1937
7a20b8a6
JK		 1938static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
1939{
a7881893
JK		 1940 /* if segs_per_sec is large than 1, we need to keep original policy. */
1941 if (sbi->segs_per_sec != 1)
1942 return CURSEG_I(sbi, type)->segno;
1943
7a20b8a6
JK		 1944 if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
1945 return 0;
1946
e066b83c
JK		 1947 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
1948 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
7a20b8a6
JK		 1949 return CURSEG_I(sbi, type)->segno;
1950}
1951
0a8165d7 1952/*
351df4b2
JK		 1953 * Allocate a current working segment.
1954 * This function always allocates a free segment in LFS manner.
1955 */
1956static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1957{
1958 struct curseg_info *curseg = CURSEG_I(sbi, type);
1959 unsigned int segno = curseg->segno;
1960 int dir = ALLOC_LEFT;
1961
1962 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1963 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1964 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1965 dir = ALLOC_RIGHT;
1966
1967 if (test_opt(sbi, NOHEAP))
1968 dir = ALLOC_RIGHT;
1969
7a20b8a6 1970 segno = __get_next_segno(sbi, type);
351df4b2
JK		 1971 get_new_segment(sbi, &segno, new_sec, dir);
1972 curseg->next_segno = segno;
1973 reset_curseg(sbi, type, 1);
1974 curseg->alloc_type = LFS;
1975}
1976
1977static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1978 struct curseg_info *seg, block_t start)
1979{
1980 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1981 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1982 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1983 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1984 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1985 int i, pos;
1986
1987 for (i = 0; i < entries; i++)
1988 target_map[i] = ckpt_map[i] | cur_map[i];
1989
1990 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1991
1992 seg->next_blkoff = pos;
351df4b2
JK		 1993}
1994
0a8165d7 1995/*
351df4b2
JK		 1996 * If a segment is written by LFS manner, next block offset is just obtained
1997 * by increasing the current block offset. However, if a segment is written by
1998 * SSR manner, next block offset obtained by calling __next_free_blkoff
1999 */
2000static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2001 struct curseg_info *seg)
2002{
2003 if (seg->alloc_type == SSR)
2004 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2005 else
2006 seg->next_blkoff++;
2007}
2008
0a8165d7 2009/*
e1c42045 2010 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2011 * manner, so it should recover the existing segment information of valid blocks
2012 */
025d63a4 2013static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2014{
2015 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2016 struct curseg_info *curseg = CURSEG_I(sbi, type);
2017 unsigned int new_segno = curseg->next_segno;
2018 struct f2fs_summary_block *sum_node;
2019 struct page *sum_page;
2020
2021 write_sum_page(sbi, curseg->sum_blk,
2022 GET_SUM_BLOCK(sbi, curseg->segno));
2023 __set_test_and_inuse(sbi, new_segno);
2024
2025 mutex_lock(&dirty_i->seglist_lock);
2026 __remove_dirty_segment(sbi, new_segno, PRE);
2027 __remove_dirty_segment(sbi, new_segno, DIRTY);
2028 mutex_unlock(&dirty_i->seglist_lock);
2029
2030 reset_curseg(sbi, type, 1);
2031 curseg->alloc_type = SSR;
2032 __next_free_blkoff(sbi, curseg, 0);
2033
025d63a4
CY		 2034 sum_page = get_sum_page(sbi, new_segno);
2035 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2036 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2037 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2038}
2039
43727527
JK		 2040static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2041{
2042 struct curseg_info *curseg = CURSEG_I(sbi, type);
2043 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2044 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2045 int i, cnt;
2046 bool reversed = false;
c192f7a4
JK		 2047
2048 /* need_SSR() already forces to do this */
e066b83c
JK		 2049 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2050 curseg->next_segno = segno;
c192f7a4 2051 return 1;
e066b83c 2052 }
43727527 2053
70d625cb
JK		 2054 /* For node segments, let's do SSR more intensively */
2055 if (IS_NODESEG(type)) {
d27c3d89
CY		 2056 if (type >= CURSEG_WARM_NODE) {
2057 reversed = true;
2058 i = CURSEG_COLD_NODE;
2059 } else {
2060 i = CURSEG_HOT_NODE;
2061 }
2062 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2063 } else {
d27c3d89
CY		 2064 if (type >= CURSEG_WARM_DATA) {
2065 reversed = true;
2066 i = CURSEG_COLD_DATA;
2067 } else {
2068 i = CURSEG_HOT_DATA;
2069 }
2070 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2071 }
43727527 2072
d27c3d89 2073 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2074 if (i == type)
2075 continue;
e066b83c
JK		 2076 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2077 curseg->next_segno = segno;
43727527 2078 return 1;
e066b83c 2079 }
c192f7a4 2080 }
43727527
JK		 2081 return 0;
2082}
2083
351df4b2
JK		 2084/*
2085 * flush out current segment and replace it with new segment
2086 * This function should be returned with success, otherwise BUG
2087 */
2088static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2089 int type, bool force)
2090{
a7881893
JK		 2091 struct curseg_info *curseg = CURSEG_I(sbi, type);
2092
7b405275 2093 if (force)
351df4b2 2094 new_curseg(sbi, type, true);
5b6c6be2
JK		 2095 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2096 type == CURSEG_WARM_NODE)
351df4b2 2097 new_curseg(sbi, type, false);
a7881893
JK		 2098 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2099 new_curseg(sbi, type, false);
351df4b2 2100 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2101 change_curseg(sbi, type);
351df4b2
JK		 2102 else
2103 new_curseg(sbi, type, false);
dcdfff65 2104
a7881893 2105 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2106}
2107
2108void allocate_new_segments(struct f2fs_sb_info *sbi)
2109{
6ae1be13
JK		 2110 struct curseg_info *curseg;
2111 unsigned int old_segno;
351df4b2
JK		 2112 int i;
2113
6ae1be13
JK		 2114 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2115 curseg = CURSEG_I(sbi, i);
2116 old_segno = curseg->segno;
2117 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2118 locate_dirty_segment(sbi, old_segno);
2119 }
351df4b2
JK		 2120}
2121
2122static const struct segment_allocation default_salloc_ops = {
2123 .allocate_segment = allocate_segment_by_default,
2124};
2125
25290fa5
JK		 2126bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2127{
2128 __u64 trim_start = cpc->trim_start;
2129 bool has_candidate = false;
2130
2131 mutex_lock(&SIT_I(sbi)->sentry_lock);
2132 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2133 if (add_discard_addrs(sbi, cpc, true)) {
2134 has_candidate = true;
2135 break;
2136 }
2137 }
2138 mutex_unlock(&SIT_I(sbi)->sentry_lock);
2139
2140 cpc->trim_start = trim_start;
2141 return has_candidate;
2142}
2143
4b2fecc8
JK		 2144int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2145{
f7ef9b83
JK		 2146 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2147 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 2148 unsigned int start_segno, end_segno;
2149 struct cp_control cpc;
c34f42e2 2150 int err = 0;
4b2fecc8 2151
836b5a63 2152 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2153 return -EINVAL;
2154
9bd27ae4 2155 cpc.trimmed = 0;
7cd8558b 2156 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 2157 goto out;
2158
ed214a11
YH		 2159 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2160 f2fs_msg(sbi->sb, KERN_WARNING,
2161 "Found FS corruption, run fsck to fix.");
2162 goto out;
2163 }
2164
4b2fecc8 2165 /* start/end segment number in main_area */
7cd8558b
JK		 2166 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2167 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2168 GET_SEGNO(sbi, end);
4b2fecc8 2169 cpc.reason = CP_DISCARD;
836b5a63 2170 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 2171
2172 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 2173 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
2174 cpc.trim_start = start_segno;
a66cdd98
JK		 2175
2176 if (sbi->discard_blks == 0)
2177 break;
2178 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2179 cpc.trim_end = end_segno;
2180 else
2181 cpc.trim_end = min_t(unsigned int,
2182 rounddown(start_segno +
bba681cb
JK		 2183 BATCHED_TRIM_SEGMENTS(sbi),
2184 sbi->segs_per_sec) - 1, end_segno);
2185
2186 mutex_lock(&sbi->gc_mutex);
c34f42e2 2187 err = write_checkpoint(sbi, &cpc);
bba681cb 2188 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 2189 if (err)
2190 break;
74fa5f3d
CY		 2191
2192 schedule();
bba681cb 2193 }
969d1b18
CY		 2194 /* It's time to issue all the filed discards */
2195 mark_discard_range_all(sbi);
4b2fecc8 2196out:
f7ef9b83 2197 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 2198 return err;
4b2fecc8
JK		 2199}
2200
351df4b2
JK		 2201static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2202{
2203 struct curseg_info *curseg = CURSEG_I(sbi, type);
2204 if (curseg->next_blkoff < sbi->blocks_per_seg)
2205 return true;
2206 return false;
2207}
2208
81377bd6 2209static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2210{
81377bd6 2211 if (fio->type == DATA)
351df4b2
JK		 2212 return CURSEG_HOT_DATA;
2213 else
2214 return CURSEG_HOT_NODE;
2215}
2216
81377bd6 2217static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2218{
81377bd6
JK		 2219 if (fio->type == DATA) {
2220 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2221
2222 if (S_ISDIR(inode->i_mode))
2223 return CURSEG_HOT_DATA;
2224 else
2225 return CURSEG_COLD_DATA;
2226 } else {
81377bd6 2227 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2228 return CURSEG_WARM_NODE;
351df4b2
JK		 2229 else
2230 return CURSEG_COLD_NODE;
2231 }
2232}
2233
81377bd6 2234static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2235{
81377bd6
JK		 2236 if (fio->type == DATA) {
2237 struct inode *inode = fio->page->mapping->host;
351df4b2 2238
81377bd6 2239 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2240 return CURSEG_COLD_DATA;
ef095d19
JK		 2241 if (is_inode_flag_set(inode, FI_HOT_DATA))
2242 return CURSEG_HOT_DATA;
2243 return CURSEG_WARM_DATA;
351df4b2 2244 } else {
81377bd6
JK		 2245 if (IS_DNODE(fio->page))
2246 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2247 CURSEG_HOT_NODE;
ef095d19 2248 return CURSEG_COLD_NODE;
351df4b2
JK		 2249 }
2250}
2251
81377bd6 2252static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2253{
a912b54d
JK		 2254 int type = 0;
2255
81377bd6 2256 switch (fio->sbi->active_logs) {
351df4b2 2257 case 2:
a912b54d
JK		 2258 type = __get_segment_type_2(fio);
2259 break;
351df4b2 2260 case 4:
a912b54d
JK		 2261 type = __get_segment_type_4(fio);
2262 break;
2263 case 6:
2264 type = __get_segment_type_6(fio);
2265 break;
2266 default:
2267 f2fs_bug_on(fio->sbi, true);
351df4b2 2268 }
81377bd6 2269
a912b54d
JK		 2270 if (IS_HOT(type))
2271 fio->temp = HOT;
2272 else if (IS_WARM(type))
2273 fio->temp = WARM;
2274 else
2275 fio->temp = COLD;
2276 return type;
351df4b2
JK		 2277}
2278
bfad7c2d
JK		 2279void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2280 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2281 struct f2fs_summary *sum, int type,
2282 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2283{
2284 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2285 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 2286
2287 mutex_lock(&curseg->curseg_mutex);
21cb1d99 2288 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK		 2289
2290 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2291
4e6a8d9b
JK		 2292 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2293
351df4b2
JK		 2294 /*
2295 * __add_sum_entry should be resided under the curseg_mutex
2296 * because, this function updates a summary entry in the
2297 * current summary block.
2298 */
e79efe3b 2299 __add_sum_entry(sbi, type, sum);
351df4b2 2300
351df4b2 2301 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2302
2303 stat_inc_block_count(sbi, curseg);
351df4b2 2304
c6f82fe9
JK		 2305 if (!__has_curseg_space(sbi, type))
2306 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2 2307 /*
c6f82fe9
JK		 2308 * SIT information should be updated after segment allocation,
2309 * since we need to keep dirty segments precisely under SSR.
351df4b2
JK		 2310 */
2311 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 2312
351df4b2
JK		 2313 mutex_unlock(&sit_i->sentry_lock);
2314
704956ec 2315 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2316 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2317
704956ec
CY		 2318 f2fs_inode_chksum_set(sbi, page);
2319 }
2320
fb830fc5
CY		 2321 if (add_list) {
2322 struct f2fs_bio_info *io;
2323
2324 INIT_LIST_HEAD(&fio->list);
2325 fio->in_list = true;
2326 io = sbi->write_io[fio->type] + fio->temp;
2327 spin_lock(&io->io_lock);
2328 list_add_tail(&fio->list, &io->io_list);
2329 spin_unlock(&io->io_lock);
2330 }
2331
bfad7c2d
JK		 2332 mutex_unlock(&curseg->curseg_mutex);
2333}
2334
05ca3632 2335static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2336{
81377bd6 2337 int type = __get_segment_type(fio);
0a595eba 2338 int err;
bfad7c2d 2339
0a595eba 2340reallocate:
7a9d7548 2341 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2342 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2343
351df4b2 2344 /* writeout dirty page into bdev */
b9109b0e 2345 err = f2fs_submit_page_write(fio);
0a595eba
JK		 2346 if (err == -EAGAIN) {
2347 fio->old_blkaddr = fio->new_blkaddr;
2348 goto reallocate;
2349 }
351df4b2
JK		 2350}
2351
b0af6d49
CY		 2352void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2353 enum iostat_type io_type)
351df4b2 2354{
458e6197 2355 struct f2fs_io_info fio = {
05ca3632 2356 .sbi = sbi,
458e6197 2357 .type = META,
04d328de 2358 .op = REQ_OP_WRITE,
70fd7614 2359 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2360 .old_blkaddr = page->index,
2361 .new_blkaddr = page->index,
05ca3632 2362 .page = page,
4375a336 2363 .encrypted_page = NULL,
fb830fc5 2364 .in_list = false,
458e6197
JK		 2365 };
2366
2b947003 2367 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2368 fio.op_flags &= ~REQ_META;
2b947003 2369
351df4b2 2370 set_page_writeback(page);
b9109b0e 2371 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2372
2373 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2374}
2375
05ca3632 2376void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2377{
2378 struct f2fs_summary sum;
05ca3632 2379
351df4b2 2380 set_summary(&sum, nid, 0, 0);
05ca3632 2381 do_write_page(&sum, fio);
b0af6d49
CY		 2382
2383 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2384}
2385
05ca3632 2386void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2387{
05ca3632 2388 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2389 struct f2fs_summary sum;
2390 struct node_info ni;
2391
9850cf4a 2392 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2393 get_node_info(sbi, dn->nid, &ni);
2394 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2395 do_write_page(&sum, fio);
f28b3434 2396 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2397
2398 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2399}
2400
d1b3e72d 2401int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2402{
b0af6d49
CY		 2403 int err;
2404
7a9d7548 2405 fio->new_blkaddr = fio->old_blkaddr;
05ca3632 2406 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2407
2408 err = f2fs_submit_page_bio(fio);
2409
2410 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2411
2412 return err;
351df4b2
JK		 2413}
2414
4356e48e 2415void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2416 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2417 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2418{
2419 struct sit_info *sit_i = SIT_I(sbi);
2420 struct curseg_info *curseg;
2421 unsigned int segno, old_cursegno;
2422 struct seg_entry *se;
2423 int type;
19f106bc 2424 unsigned short old_blkoff;
351df4b2
JK		 2425
2426 segno = GET_SEGNO(sbi, new_blkaddr);
2427 se = get_seg_entry(sbi, segno);
2428 type = se->type;
2429
19f106bc
CY		 2430 if (!recover_curseg) {
2431 /* for recovery flow */
2432 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2433 if (old_blkaddr == NULL_ADDR)
2434 type = CURSEG_COLD_DATA;
2435 else
2436 type = CURSEG_WARM_DATA;
2437 }
2438 } else {
2439 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 2440 type = CURSEG_WARM_DATA;
2441 }
19f106bc 2442
351df4b2
JK		 2443 curseg = CURSEG_I(sbi, type);
2444
2445 mutex_lock(&curseg->curseg_mutex);
2446 mutex_lock(&sit_i->sentry_lock);
2447
2448 old_cursegno = curseg->segno;
19f106bc 2449 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2450
2451 /* change the current segment */
2452 if (segno != curseg->segno) {
2453 curseg->next_segno = segno;
025d63a4 2454 change_curseg(sbi, type);
351df4b2
JK		 2455 }
2456
491c0854 2457 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2458 __add_sum_entry(sbi, type, sum);
351df4b2 2459
28bc106b 2460 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2461 update_sit_entry(sbi, new_blkaddr, 1);
2462 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2463 update_sit_entry(sbi, old_blkaddr, -1);
2464
2465 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2466 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2467
351df4b2 2468 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2469
19f106bc
CY		 2470 if (recover_curseg) {
2471 if (old_cursegno != curseg->segno) {
2472 curseg->next_segno = old_cursegno;
025d63a4 2473 change_curseg(sbi, type);
19f106bc
CY		 2474 }
2475 curseg->next_blkoff = old_blkoff;
2476 }
2477
351df4b2
JK		 2478 mutex_unlock(&sit_i->sentry_lock);
2479 mutex_unlock(&curseg->curseg_mutex);
2480}
2481
528e3459
CY		 2482void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2483 block_t old_addr, block_t new_addr,
28bc106b
CY		 2484 unsigned char version, bool recover_curseg,
2485 bool recover_newaddr)
528e3459
CY		 2486{
2487 struct f2fs_summary sum;
2488
2489 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2490
28bc106b
CY		 2491 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2492 recover_curseg, recover_newaddr);
528e3459 2493
f28b3434 2494 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2495}
2496
93dfe2ac 2497void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2498 enum page_type type, bool ordered)
93dfe2ac 2499{
93dfe2ac 2500 if (PageWriteback(page)) {
4081363f
JK		 2501 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2502
b9109b0e
JK		 2503 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2504 0, page->index, type);
fec1d657
JK		 2505 if (ordered)
2506 wait_on_page_writeback(page);
2507 else
2508 wait_for_stable_page(page);
93dfe2ac
JK		 2509 }
2510}
2511
08b39fbd
CY		 2512void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
2513 block_t blkaddr)
2514{
2515 struct page *cpage;
2516
5d4c0af4 2517 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2518 return;
2519
08b39fbd
CY		 2520 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2521 if (cpage) {
fec1d657 2522 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2523 f2fs_put_page(cpage, 1);
2524 }
2525}
2526
351df4b2
JK		 2527static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2528{
2529 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2530 struct curseg_info *seg_i;
2531 unsigned char *kaddr;
2532 struct page *page;
2533 block_t start;
2534 int i, j, offset;
2535
2536 start = start_sum_block(sbi);
2537
2538 page = get_meta_page(sbi, start++);
2539 kaddr = (unsigned char *)page_address(page);
2540
2541 /* Step 1: restore nat cache */
2542 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2543 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2544
2545 /* Step 2: restore sit cache */
2546 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2547 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2548 offset = 2 * SUM_JOURNAL_SIZE;
2549
2550 /* Step 3: restore summary entries */
2551 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2552 unsigned short blk_off;
2553 unsigned int segno;
2554
2555 seg_i = CURSEG_I(sbi, i);
2556 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2557 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2558 seg_i->next_segno = segno;
2559 reset_curseg(sbi, i, 0);
2560 seg_i->alloc_type = ckpt->alloc_type[i];
2561 seg_i->next_blkoff = blk_off;
2562
2563 if (seg_i->alloc_type == SSR)
2564 blk_off = sbi->blocks_per_seg;
2565
2566 for (j = 0; j < blk_off; j++) {
2567 struct f2fs_summary *s;
2568 s = (struct f2fs_summary *)(kaddr + offset);
2569 seg_i->sum_blk->entries[j] = *s;
2570 offset += SUMMARY_SIZE;
09cbfeaf 2571 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2572 SUM_FOOTER_SIZE)
2573 continue;
2574
2575 f2fs_put_page(page, 1);
2576 page = NULL;
2577
2578 page = get_meta_page(sbi, start++);
2579 kaddr = (unsigned char *)page_address(page);
2580 offset = 0;
2581 }
2582 }
2583 f2fs_put_page(page, 1);
2584 return 0;
2585}
2586
2587static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2588{
2589 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2590 struct f2fs_summary_block *sum;
2591 struct curseg_info *curseg;
2592 struct page *new;
2593 unsigned short blk_off;
2594 unsigned int segno = 0;
2595 block_t blk_addr = 0;
2596
2597 /* get segment number and block addr */
2598 if (IS_DATASEG(type)) {
2599 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2600 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2601 CURSEG_HOT_DATA]);
119ee914 2602 if (__exist_node_summaries(sbi))
351df4b2
JK		 2603 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2604 else
2605 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2606 } else {
2607 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2608 CURSEG_HOT_NODE]);
2609 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2610 CURSEG_HOT_NODE]);
119ee914 2611 if (__exist_node_summaries(sbi))
351df4b2
JK		 2612 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2613 type - CURSEG_HOT_NODE);
2614 else
2615 blk_addr = GET_SUM_BLOCK(sbi, segno);
2616 }
2617
2618 new = get_meta_page(sbi, blk_addr);
2619 sum = (struct f2fs_summary_block *)page_address(new);
2620
2621 if (IS_NODESEG(type)) {
119ee914 2622 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2623 struct f2fs_summary *ns = &sum->entries[0];
2624 int i;
2625 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2626 ns->version = 0;
2627 ns->ofs_in_node = 0;
2628 }
2629 } else {
d653788a
GZ		 2630 int err;
2631
2632 err = restore_node_summary(sbi, segno, sum);
2633 if (err) {
351df4b2 2634 f2fs_put_page(new, 1);
d653788a 2635 return err;
351df4b2
JK		 2636 }
2637 }
2638 }
2639
2640 /* set uncompleted segment to curseg */
2641 curseg = CURSEG_I(sbi, type);
2642 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2643
2644 /* update journal info */
2645 down_write(&curseg->journal_rwsem);
2646 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2647 up_write(&curseg->journal_rwsem);
2648
2649 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2650 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2651 curseg->next_segno = segno;
2652 reset_curseg(sbi, type, 0);
2653 curseg->alloc_type = ckpt->alloc_type[type];
2654 curseg->next_blkoff = blk_off;
2655 mutex_unlock(&curseg->curseg_mutex);
2656 f2fs_put_page(new, 1);
2657 return 0;
2658}
2659
2660static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2661{
21d3f8e1
JQ		 2662 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2663 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 2664 int type = CURSEG_HOT_DATA;
e4fc5fbf 2665 int err;
351df4b2 2666
aaec2b1d 2667 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2668 int npages = npages_for_summary_flush(sbi, true);
2669
2670 if (npages >= 2)
2671 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2672 META_CP, true);
3fa06d7b 2673
351df4b2
JK		 2674 /* restore for compacted data summary */
2675 if (read_compacted_summaries(sbi))
2676 return -EINVAL;
2677 type = CURSEG_HOT_NODE;
2678 }
2679
119ee914 2680 if (__exist_node_summaries(sbi))
3fa06d7b 2681 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2682 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2683
e4fc5fbf
CY		 2684 for (; type <= CURSEG_COLD_NODE; type++) {
2685 err = read_normal_summaries(sbi, type);
2686 if (err)
2687 return err;
2688 }
2689
21d3f8e1
JQ		 2690 /* sanity check for summary blocks */
2691 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2692 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2693 return -EINVAL;
2694
351df4b2
JK		 2695 return 0;
2696}
2697
2698static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2699{
2700 struct page *page;
2701 unsigned char *kaddr;
2702 struct f2fs_summary *summary;
2703 struct curseg_info *seg_i;
2704 int written_size = 0;
2705 int i, j;
2706
2707 page = grab_meta_page(sbi, blkaddr++);
2708 kaddr = (unsigned char *)page_address(page);
2709
2710 /* Step 1: write nat cache */
2711 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2712 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2713 written_size += SUM_JOURNAL_SIZE;
2714
2715 /* Step 2: write sit cache */
2716 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2717 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2718 written_size += SUM_JOURNAL_SIZE;
2719
351df4b2
JK		 2720 /* Step 3: write summary entries */
2721 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2722 unsigned short blkoff;
2723 seg_i = CURSEG_I(sbi, i);
2724 if (sbi->ckpt->alloc_type[i] == SSR)
2725 blkoff = sbi->blocks_per_seg;
2726 else
2727 blkoff = curseg_blkoff(sbi, i);
2728
2729 for (j = 0; j < blkoff; j++) {
2730 if (!page) {
2731 page = grab_meta_page(sbi, blkaddr++);
2732 kaddr = (unsigned char *)page_address(page);
2733 written_size = 0;
2734 }
2735 summary = (struct f2fs_summary *)(kaddr + written_size);
2736 *summary = seg_i->sum_blk->entries[j];
2737 written_size += SUMMARY_SIZE;
351df4b2 2738
09cbfeaf 2739 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2740 SUM_FOOTER_SIZE)
2741 continue;
2742
e8d61a74 2743 set_page_dirty(page);
351df4b2
JK		 2744 f2fs_put_page(page, 1);
2745 page = NULL;
2746 }
2747 }
e8d61a74
CY		 2748 if (page) {
2749 set_page_dirty(page);
351df4b2 2750 f2fs_put_page(page, 1);
e8d61a74 2751 }
351df4b2
JK		 2752}
2753
2754static void write_normal_summaries(struct f2fs_sb_info *sbi,
2755 block_t blkaddr, int type)
2756{
2757 int i, end;
2758 if (IS_DATASEG(type))
2759 end = type + NR_CURSEG_DATA_TYPE;
2760 else
2761 end = type + NR_CURSEG_NODE_TYPE;
2762
b7ad7512
CY		 2763 for (i = type; i < end; i++)
2764 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 2765}
2766
2767void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2768{
aaec2b1d 2769 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 2770 write_compacted_summaries(sbi, start_blk);
2771 else
2772 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2773}
2774
2775void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2776{
119ee914 2777 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 2778}
2779
dfc08a12 2780int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 2781 unsigned int val, int alloc)
2782{
2783 int i;
2784
2785 if (type == NAT_JOURNAL) {
dfc08a12
CY		 2786 for (i = 0; i < nats_in_cursum(journal); i++) {
2787 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 2788 return i;
2789 }
dfc08a12
CY		 2790 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2791 return update_nats_in_cursum(journal, 1);
351df4b2 2792 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 2793 for (i = 0; i < sits_in_cursum(journal); i++)
2794 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 2795 return i;
dfc08a12
CY		 2796 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2797 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 2798 }
2799 return -1;
2800}
2801
2802static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2803 unsigned int segno)
2804{
2cc22186 2805 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 2806}
2807
2808static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2809 unsigned int start)
2810{
2811 struct sit_info *sit_i = SIT_I(sbi);
2812 struct page *src_page, *dst_page;
2813 pgoff_t src_off, dst_off;
2814 void *src_addr, *dst_addr;
2815
2816 src_off = current_sit_addr(sbi, start);
2817 dst_off = next_sit_addr(sbi, src_off);
2818
2819 /* get current sit block page without lock */
2820 src_page = get_meta_page(sbi, src_off);
2821 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 2822 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 2823
2824 src_addr = page_address(src_page);
2825 dst_addr = page_address(dst_page);
09cbfeaf 2826 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 2827
2828 set_page_dirty(dst_page);
2829 f2fs_put_page(src_page, 1);
2830
2831 set_to_next_sit(sit_i, start);
2832
2833 return dst_page;
2834}
2835
184a5cd2
CY		 2836static struct sit_entry_set *grab_sit_entry_set(void)
2837{
2838 struct sit_entry_set *ses =
80c54505 2839 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 2840
2841 ses->entry_cnt = 0;
2842 INIT_LIST_HEAD(&ses->set_list);
2843 return ses;
2844}
2845
2846static void release_sit_entry_set(struct sit_entry_set *ses)
2847{
2848 list_del(&ses->set_list);
2849 kmem_cache_free(sit_entry_set_slab, ses);
2850}
2851
2852static void adjust_sit_entry_set(struct sit_entry_set *ses,
2853 struct list_head *head)
2854{
2855 struct sit_entry_set *next = ses;
2856
2857 if (list_is_last(&ses->set_list, head))
2858 return;
2859
2860 list_for_each_entry_continue(next, head, set_list)
2861 if (ses->entry_cnt <= next->entry_cnt)
2862 break;
2863
2864 list_move_tail(&ses->set_list, &next->set_list);
2865}
2866
2867static void add_sit_entry(unsigned int segno, struct list_head *head)
2868{
2869 struct sit_entry_set *ses;
2870 unsigned int start_segno = START_SEGNO(segno);
2871
2872 list_for_each_entry(ses, head, set_list) {
2873 if (ses->start_segno == start_segno) {
2874 ses->entry_cnt++;
2875 adjust_sit_entry_set(ses, head);
2876 return;
2877 }
2878 }
2879
2880 ses = grab_sit_entry_set();
2881
2882 ses->start_segno = start_segno;
2883 ses->entry_cnt++;
2884 list_add(&ses->set_list, head);
2885}
2886
2887static void add_sits_in_set(struct f2fs_sb_info *sbi)
2888{
2889 struct f2fs_sm_info *sm_info = SM_I(sbi);
2890 struct list_head *set_list = &sm_info->sit_entry_set;
2891 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 2892 unsigned int segno;
2893
7cd8558b 2894 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 2895 add_sit_entry(segno, set_list);
2896}
2897
2898static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 2899{
2900 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2901 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 2902 int i;
2903
b7ad7512 2904 down_write(&curseg->journal_rwsem);
dfc08a12 2905 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 2906 unsigned int segno;
2907 bool dirtied;
2908
dfc08a12 2909 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 2910 dirtied = __mark_sit_entry_dirty(sbi, segno);
2911
2912 if (!dirtied)
2913 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 2914 }
dfc08a12 2915 update_sits_in_cursum(journal, -i);
b7ad7512 2916 up_write(&curseg->journal_rwsem);
351df4b2
JK		 2917}
2918
0a8165d7 2919/*
351df4b2
JK		 2920 * CP calls this function, which flushes SIT entries including sit_journal,
2921 * and moves prefree segs to free segs.
2922 */
4b2fecc8 2923void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 2924{
2925 struct sit_info *sit_i = SIT_I(sbi);
2926 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2927 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2928 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 2929 struct sit_entry_set *ses, *tmp;
2930 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 2931 bool to_journal = true;
4b2fecc8 2932 struct seg_entry *se;
351df4b2 2933
351df4b2
JK		 2934 mutex_lock(&sit_i->sentry_lock);
2935
2b11a74b
WL		 2936 if (!sit_i->dirty_sentries)
2937 goto out;
2938
351df4b2 2939 /*
184a5cd2
CY		 2940 * add and account sit entries of dirty bitmap in sit entry
2941 * set temporarily
351df4b2 2942 */
184a5cd2 2943 add_sits_in_set(sbi);
351df4b2 2944
184a5cd2
CY		 2945 /*
2946 * if there are no enough space in journal to store dirty sit
2947 * entries, remove all entries from journal and add and account
2948 * them in sit entry set.
2949 */
dfc08a12 2950 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2951 remove_sits_in_journal(sbi);
b2955550 2952
184a5cd2
CY		 2953 /*
2954 * there are two steps to flush sit entries:
2955 * #1, flush sit entries to journal in current cold data summary block.
2956 * #2, flush sit entries to sit page.
2957 */
2958 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2959 struct page *page = NULL;
184a5cd2
CY		 2960 struct f2fs_sit_block *raw_sit = NULL;
2961 unsigned int start_segno = ses->start_segno;
2962 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2963 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2964 unsigned int segno = start_segno;
2965
2966 if (to_journal &&
dfc08a12 2967 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2968 to_journal = false;
2969
b7ad7512
CY		 2970 if (to_journal) {
2971 down_write(&curseg->journal_rwsem);
2972 } else {
184a5cd2
CY		 2973 page = get_next_sit_page(sbi, start_segno);
2974 raw_sit = page_address(page);
351df4b2 2975 }
351df4b2 2976
184a5cd2
CY		 2977 /* flush dirty sit entries in region of current sit set */
2978 for_each_set_bit_from(segno, bitmap, end) {
2979 int offset, sit_offset;
4b2fecc8
JK		 2980
2981 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2982
2983 /* add discard candidates */
c473f1a9 2984 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 2985 cpc->trim_start = segno;
25290fa5 2986 add_discard_addrs(sbi, cpc, false);
4b2fecc8 2987 }
184a5cd2
CY		 2988
2989 if (to_journal) {
dfc08a12 2990 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2991 SIT_JOURNAL, segno, 1);
2992 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2993 segno_in_journal(journal, offset) =
184a5cd2
CY		 2994 cpu_to_le32(segno);
2995 seg_info_to_raw_sit(se,
dfc08a12 2996 &sit_in_journal(journal, offset));
184a5cd2
CY		 2997 } else {
2998 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2999 seg_info_to_raw_sit(se,
3000 &raw_sit->entries[sit_offset]);
3001 }
351df4b2 3002
184a5cd2
CY		 3003 __clear_bit(segno, bitmap);
3004 sit_i->dirty_sentries--;
3005 ses->entry_cnt--;
351df4b2
JK		 3006 }
3007
b7ad7512
CY		 3008 if (to_journal)
3009 up_write(&curseg->journal_rwsem);
3010 else
184a5cd2
CY		 3011 f2fs_put_page(page, 1);
3012
3013 f2fs_bug_on(sbi, ses->entry_cnt);
3014 release_sit_entry_set(ses);
351df4b2 3015 }
184a5cd2
CY		 3016
3017 f2fs_bug_on(sbi, !list_empty(head));
3018 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3019out:
c473f1a9 3020 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3021 __u64 trim_start = cpc->trim_start;
3022
4b2fecc8 3023 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3024 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3025
3026 cpc->trim_start = trim_start;
4b2fecc8 3027 }
351df4b2 3028 mutex_unlock(&sit_i->sentry_lock);
351df4b2 3029
351df4b2
JK		 3030 set_prefree_as_free_segments(sbi);
3031}
3032
3033static int build_sit_info(struct f2fs_sb_info *sbi)
3034{
3035 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3036 struct sit_info *sit_i;
3037 unsigned int sit_segs, start;
ae27d62e 3038 char *src_bitmap;
351df4b2
JK		 3039 unsigned int bitmap_size;
3040
3041 /* allocate memory for SIT information */
3042 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3043 if (!sit_i)
3044 return -ENOMEM;
3045
3046 SM_I(sbi)->sit_info = sit_i;
3047
a7c3e901 3048 sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
39307a8e 3049 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3050 if (!sit_i->sentries)
3051 return -ENOMEM;
3052
7cd8558b 3053 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3054 sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3055 if (!sit_i->dirty_sentries_bitmap)
3056 return -ENOMEM;
3057
7cd8558b 3058 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3059 sit_i->sentries[start].cur_valid_map
3060 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3061 sit_i->sentries[start].ckpt_valid_map
3062 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3063 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3064 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3065 return -ENOMEM;
3e025740 3066
355e7891
CY		 3067#ifdef CONFIG_F2FS_CHECK_FS
3068 sit_i->sentries[start].cur_valid_map_mir
3069 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3070 if (!sit_i->sentries[start].cur_valid_map_mir)
3071 return -ENOMEM;
3072#endif
3073
3e025740
JK		 3074 if (f2fs_discard_en(sbi)) {
3075 sit_i->sentries[start].discard_map
3076 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3077 if (!sit_i->sentries[start].discard_map)
3078 return -ENOMEM;
3079 }
351df4b2
JK		 3080 }
3081
60a3b782
JK		 3082 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3083 if (!sit_i->tmp_map)
3084 return -ENOMEM;
3085
351df4b2 3086 if (sbi->segs_per_sec > 1) {
a7c3e901 3087 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
39307a8e 3088 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3089 if (!sit_i->sec_entries)
3090 return -ENOMEM;
3091 }
3092
3093 /* get information related with SIT */
3094 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3095
3096 /* setup SIT bitmap from ckeckpoint pack */
3097 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3098 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3099
ae27d62e
CY		 3100 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3101 if (!sit_i->sit_bitmap)
351df4b2 3102 return -ENOMEM;
351df4b2 3103
ae27d62e
CY		 3104#ifdef CONFIG_F2FS_CHECK_FS
3105 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3106 if (!sit_i->sit_bitmap_mir)
3107 return -ENOMEM;
3108#endif
3109
351df4b2
JK		 3110 /* init SIT information */
3111 sit_i->s_ops = &default_salloc_ops;
3112
3113 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3114 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3115 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3116 sit_i->bitmap_size = bitmap_size;
3117 sit_i->dirty_sentries = 0;
3118 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3119 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3120 sit_i->mounted_time = ktime_get_real_seconds();
351df4b2
JK		 3121 mutex_init(&sit_i->sentry_lock);
3122 return 0;
3123}
3124
3125static int build_free_segmap(struct f2fs_sb_info *sbi)
3126{
351df4b2
JK		 3127 struct free_segmap_info *free_i;
3128 unsigned int bitmap_size, sec_bitmap_size;
3129
3130 /* allocate memory for free segmap information */
3131 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3132 if (!free_i)
3133 return -ENOMEM;
3134
3135 SM_I(sbi)->free_info = free_i;
3136
7cd8558b 3137 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3138 free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3139 if (!free_i->free_segmap)
3140 return -ENOMEM;
3141
7cd8558b 3142 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
a7c3e901 3143 free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3144 if (!free_i->free_secmap)
3145 return -ENOMEM;
3146
3147 /* set all segments as dirty temporarily */
3148 memset(free_i->free_segmap, 0xff, bitmap_size);
3149 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3150
3151 /* init free segmap information */
7cd8558b 3152 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3153 free_i->free_segments = 0;
3154 free_i->free_sections = 0;
1a118ccf 3155 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3156 return 0;
3157}
3158
3159static int build_curseg(struct f2fs_sb_info *sbi)
3160{
1042d60f 3161 struct curseg_info *array;
351df4b2
JK		 3162 int i;
3163
b434babf 3164 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 3165 if (!array)
3166 return -ENOMEM;
3167
3168 SM_I(sbi)->curseg_array = array;
3169
3170 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3171 mutex_init(&array[i].curseg_mutex);
09cbfeaf 3172 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3173 if (!array[i].sum_blk)
3174 return -ENOMEM;
b7ad7512
CY		 3175 init_rwsem(&array[i].journal_rwsem);
3176 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3177 GFP_KERNEL);
3178 if (!array[i].journal)
3179 return -ENOMEM;
351df4b2
JK		 3180 array[i].segno = NULL_SEGNO;
3181 array[i].next_blkoff = 0;
3182 }
3183 return restore_curseg_summaries(sbi);
3184}
3185
3186static void build_sit_entries(struct f2fs_sb_info *sbi)
3187{
3188 struct sit_info *sit_i = SIT_I(sbi);
3189 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3190 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3191 struct seg_entry *se;
3192 struct f2fs_sit_entry sit;
74de593a
CY		 3193 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3194 unsigned int i, start, end;
3195 unsigned int readed, start_blk = 0;
351df4b2 3196
74de593a 3197 do {
664ba972
JK		 3198 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3199 META_SIT, true);
74de593a
CY		 3200
3201 start = start_blk * sit_i->sents_per_block;
3202 end = (start_blk + readed) * sit_i->sents_per_block;
3203
7cd8558b 3204 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3205 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3206 struct page *page;
3207
9c094040 3208 se = &sit_i->sentries[start];
74de593a
CY		 3209 page = get_current_sit_page(sbi, start);
3210 sit_blk = (struct f2fs_sit_block *)page_address(page);
3211 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3212 f2fs_put_page(page, 1);
d600af23 3213
74de593a
CY		 3214 check_block_count(sbi, start, &sit);
3215 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 3216
3217 /* build discard map only one time */
3e025740 3218 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3219 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3220 memset(se->discard_map, 0xff,
3221 SIT_VBLOCK_MAP_SIZE);
3222 } else {
3223 memcpy(se->discard_map,
3224 se->cur_valid_map,
3225 SIT_VBLOCK_MAP_SIZE);
3226 sbi->discard_blks +=
3227 sbi->blocks_per_seg -
3228 se->valid_blocks;
3229 }
3e025740 3230 }
a66cdd98 3231
d600af23
CY		 3232 if (sbi->segs_per_sec > 1)
3233 get_sec_entry(sbi, start)->valid_blocks +=
3234 se->valid_blocks;
351df4b2 3235 }
74de593a
CY		 3236 start_blk += readed;
3237 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3238
3239 down_read(&curseg->journal_rwsem);
3240 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3241 unsigned int old_valid_blocks;
3242
3243 start = le32_to_cpu(segno_in_journal(journal, i));
3244 se = &sit_i->sentries[start];
3245 sit = sit_in_journal(journal, i);
3246
3247 old_valid_blocks = se->valid_blocks;
3248
3249 check_block_count(sbi, start, &sit);
3250 seg_info_from_raw_sit(se, &sit);
3251
3252 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3253 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3254 memset(se->discard_map, 0xff,
3255 SIT_VBLOCK_MAP_SIZE);
3256 } else {
3257 memcpy(se->discard_map, se->cur_valid_map,
3258 SIT_VBLOCK_MAP_SIZE);
3259 sbi->discard_blks += old_valid_blocks -
3260 se->valid_blocks;
3261 }
d600af23
CY		 3262 }
3263
3264 if (sbi->segs_per_sec > 1)
3265 get_sec_entry(sbi, start)->valid_blocks +=
3266 se->valid_blocks - old_valid_blocks;
3267 }
3268 up_read(&curseg->journal_rwsem);
351df4b2
JK		 3269}
3270
3271static void init_free_segmap(struct f2fs_sb_info *sbi)
3272{
3273 unsigned int start;
3274 int type;
3275
7cd8558b 3276 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3277 struct seg_entry *sentry = get_seg_entry(sbi, start);
3278 if (!sentry->valid_blocks)
3279 __set_free(sbi, start);
c79b7ff1
JK		 3280 else
3281 SIT_I(sbi)->written_valid_blocks +=
3282 sentry->valid_blocks;
351df4b2
JK		 3283 }
3284
3285 /* set use the current segments */
3286 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3287 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3288 __set_test_and_inuse(sbi, curseg_t->segno);
3289 }
3290}
3291
3292static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3293{
3294 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3295 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3296 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3297 unsigned short valid_blocks;
3298
8736fbf0 3299 while (1) {
351df4b2 3300 /* find dirty segment based on free segmap */
7cd8558b
JK		 3301 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3302 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3303 break;
3304 offset = segno + 1;
302bd348 3305 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3306 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3307 continue;
ec325b52
JK		 3308 if (valid_blocks > sbi->blocks_per_seg) {
3309 f2fs_bug_on(sbi, 1);
3310 continue;
3311 }
351df4b2
JK		 3312 mutex_lock(&dirty_i->seglist_lock);
3313 __locate_dirty_segment(sbi, segno, DIRTY);
3314 mutex_unlock(&dirty_i->seglist_lock);
3315 }
3316}
3317
5ec4e49f 3318static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3319{
3320 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3321 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3322
a7c3e901 3323 dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 3324 if (!dirty_i->victim_secmap)
351df4b2
JK		 3325 return -ENOMEM;
3326 return 0;
3327}
3328
3329static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3330{
3331 struct dirty_seglist_info *dirty_i;
3332 unsigned int bitmap_size, i;
3333
3334 /* allocate memory for dirty segments list information */
3335 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3336 if (!dirty_i)
3337 return -ENOMEM;
3338
3339 SM_I(sbi)->dirty_info = dirty_i;
3340 mutex_init(&dirty_i->seglist_lock);
3341
7cd8558b 3342 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3343
3344 for (i = 0; i < NR_DIRTY_TYPE; i++) {
a7c3e901 3345 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3346 if (!dirty_i->dirty_segmap[i])
3347 return -ENOMEM;
3348 }
3349
3350 init_dirty_segmap(sbi);
5ec4e49f 3351 return init_victim_secmap(sbi);
351df4b2
JK		 3352}
3353
0a8165d7 3354/*
351df4b2
JK		 3355 * Update min, max modified time for cost-benefit GC algorithm
3356 */
3357static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3358{
3359 struct sit_info *sit_i = SIT_I(sbi);
3360 unsigned int segno;
3361
3362 mutex_lock(&sit_i->sentry_lock);
3363
3364 sit_i->min_mtime = LLONG_MAX;
3365
7cd8558b 3366 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3367 unsigned int i;
3368 unsigned long long mtime = 0;
3369
3370 for (i = 0; i < sbi->segs_per_sec; i++)
3371 mtime += get_seg_entry(sbi, segno + i)->mtime;
3372
3373 mtime = div_u64(mtime, sbi->segs_per_sec);
3374
3375 if (sit_i->min_mtime > mtime)
3376 sit_i->min_mtime = mtime;
3377 }
3378 sit_i->max_mtime = get_mtime(sbi);
3379 mutex_unlock(&sit_i->sentry_lock);
3380}
3381
3382int build_segment_manager(struct f2fs_sb_info *sbi)
3383{
3384 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3385 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3386 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3387 int err;
3388
3389 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3390 if (!sm_info)
3391 return -ENOMEM;
3392
3393 /* init sm info */
3394 sbi->sm_info = sm_info;
351df4b2
JK		 3395 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3396 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3397 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3398 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3399 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3400 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3401 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3402 sm_info->rec_prefree_segments = sm_info->main_segments *
3403 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3404 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3405 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3406
52763a4b
JK		 3407 if (!test_opt(sbi, LFS))
3408 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3409 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3410 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3411 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
351df4b2 3412
bba681cb
JK		 3413 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3414
184a5cd2
CY		 3415 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3416
d4fdf8ba 3417 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3418 err = create_flush_cmd_control(sbi);
3419 if (err)
a688b9d9 3420 return err;
6b4afdd7
JK		 3421 }
3422
0b54fb84
JK		 3423 err = create_discard_cmd_control(sbi);
3424 if (err)
3425 return err;
3426
351df4b2
JK		 3427 err = build_sit_info(sbi);
3428 if (err)
3429 return err;
3430 err = build_free_segmap(sbi);
3431 if (err)
3432 return err;
3433 err = build_curseg(sbi);
3434 if (err)
3435 return err;
3436
3437 /* reinit free segmap based on SIT */
3438 build_sit_entries(sbi);
3439
3440 init_free_segmap(sbi);
3441 err = build_dirty_segmap(sbi);
3442 if (err)
3443 return err;
3444
3445 init_min_max_mtime(sbi);
3446 return 0;
3447}
3448
3449static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3450 enum dirty_type dirty_type)
3451{
3452 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3453
3454 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3455 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3456 dirty_i->nr_dirty[dirty_type] = 0;
3457 mutex_unlock(&dirty_i->seglist_lock);
3458}
3459
5ec4e49f 3460static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3461{
3462 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3463 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3464}
3465
3466static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3467{
3468 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3469 int i;
3470
3471 if (!dirty_i)
3472 return;
3473
3474 /* discard pre-free/dirty segments list */
3475 for (i = 0; i < NR_DIRTY_TYPE; i++)
3476 discard_dirty_segmap(sbi, i);
3477
5ec4e49f 3478 destroy_victim_secmap(sbi);
351df4b2
JK		 3479 SM_I(sbi)->dirty_info = NULL;
3480 kfree(dirty_i);
3481}
3482
3483static void destroy_curseg(struct f2fs_sb_info *sbi)
3484{
3485 struct curseg_info *array = SM_I(sbi)->curseg_array;
3486 int i;
3487
3488 if (!array)
3489 return;
3490 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3491 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3492 kfree(array[i].sum_blk);
b7ad7512
CY		 3493 kfree(array[i].journal);
3494 }
351df4b2
JK		 3495 kfree(array);
3496}
3497
3498static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3499{
3500 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3501 if (!free_i)
3502 return;
3503 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3504 kvfree(free_i->free_segmap);
3505 kvfree(free_i->free_secmap);
351df4b2
JK		 3506 kfree(free_i);
3507}
3508
3509static void destroy_sit_info(struct f2fs_sb_info *sbi)
3510{
3511 struct sit_info *sit_i = SIT_I(sbi);
3512 unsigned int start;
3513
3514 if (!sit_i)
3515 return;
3516
3517 if (sit_i->sentries) {
7cd8558b 3518 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3519 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 3520#ifdef CONFIG_F2FS_CHECK_FS
3521 kfree(sit_i->sentries[start].cur_valid_map_mir);
3522#endif
351df4b2 3523 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3524 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3525 }
3526 }
60a3b782
JK		 3527 kfree(sit_i->tmp_map);
3528
39307a8e
JK		 3529 kvfree(sit_i->sentries);
3530 kvfree(sit_i->sec_entries);
3531 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3532
3533 SM_I(sbi)->sit_info = NULL;
3534 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 3535#ifdef CONFIG_F2FS_CHECK_FS
3536 kfree(sit_i->sit_bitmap_mir);
3537#endif
351df4b2
JK		 3538 kfree(sit_i);
3539}
3540
3541void destroy_segment_manager(struct f2fs_sb_info *sbi)
3542{
3543 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3544
3b03f724
CY		 3545 if (!sm_info)
3546 return;
5eba8c5d 3547 destroy_flush_cmd_control(sbi, true);
f099405f 3548 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3549 destroy_dirty_segmap(sbi);
3550 destroy_curseg(sbi);
3551 destroy_free_segmap(sbi);
3552 destroy_sit_info(sbi);
3553 sbi->sm_info = NULL;
3554 kfree(sm_info);
3555}
7fd9e544
JK		 3556
3557int __init create_segment_manager_caches(void)
3558{
3559 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3560 sizeof(struct discard_entry));
7fd9e544 3561 if (!discard_entry_slab)
184a5cd2
CY		 3562 goto fail;
3563
b01a9201
JK		 3564 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3565 sizeof(struct discard_cmd));
3566 if (!discard_cmd_slab)
6ab2a308 3567 goto destroy_discard_entry;
275b66b0 3568
184a5cd2 3569 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3570 sizeof(struct sit_entry_set));
184a5cd2 3571 if (!sit_entry_set_slab)
b01a9201 3572 goto destroy_discard_cmd;
88b88a66
JK		 3573
3574 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3575 sizeof(struct inmem_pages));
3576 if (!inmem_entry_slab)
3577 goto destroy_sit_entry_set;
7fd9e544 3578 return 0;
184a5cd2 3579
88b88a66
JK		 3580destroy_sit_entry_set:
3581 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 3582destroy_discard_cmd:
3583 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3584destroy_discard_entry:
184a5cd2
CY		 3585 kmem_cache_destroy(discard_entry_slab);
3586fail:
3587 return -ENOMEM;
7fd9e544
JK		 3588}
3589
3590void destroy_segment_manager_caches(void)
3591{
184a5cd2 3592 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3593 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3594 kmem_cache_destroy(discard_entry_slab);
88b88a66 3595 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3596}
Ubuntu Jammy Linux kernel mirror