]> 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: enhance multiple device flush
[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>
1eb1ef4a 20#include <linux/sched/signal.h>
351df4b2
JK		 21
22#include "f2fs.h"
23#include "segment.h"
24#include "node.h"
5f656541 25#include "gc.h"
9e4ded3f 26#include "trace.h"
6ec178da 27#include <trace/events/f2fs.h>
351df4b2 28
9a7f143a
CL		 29#define __reverse_ffz(x) __reverse_ffs(~(x))
30
7fd9e544 31static struct kmem_cache *discard_entry_slab;
b01a9201 32static struct kmem_cache *discard_cmd_slab;
184a5cd2 33static struct kmem_cache *sit_entry_set_slab;
88b88a66 34static struct kmem_cache *inmem_entry_slab;
7fd9e544 35
f96999c3
JK		 36static unsigned long __reverse_ulong(unsigned char *str)
37{
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41#if BITS_PER_LONG == 64
42 shift = 56;
43#endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49}
50
9a7f143a
CL		 51/*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55static inline unsigned long __reverse_ffs(unsigned long word)
56{
57 int num = 0;
58
59#if BITS_PER_LONG == 64
f96999c3 60 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 61 num += 32;
f96999c3 62 else
9a7f143a 63 word >>= 32;
9a7f143a 64#endif
f96999c3 65 if ((word & 0xffff0000) == 0)
9a7f143a 66 num += 16;
f96999c3 67 else
9a7f143a 68 word >>= 16;
f96999c3
JK		 69
70 if ((word & 0xff00) == 0)
9a7f143a 71 num += 8;
f96999c3 72 else
9a7f143a 73 word >>= 8;
f96999c3 74
9a7f143a
CL		 75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
f96999c3 79
9a7f143a
CL		 80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
f96999c3 84
9a7f143a
CL		 85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88}
89
90/*
e1c42045 91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 93 * @size must be integral times of unsigned long.
9a7f143a 94 * Example:
f96999c3
JK		 95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 98 */
99static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101{
102 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 103 unsigned long result = size;
9a7f143a 104 unsigned long tmp;
9a7f143a
CL		 105
106 if (offset >= size)
107 return size;
108
692223d1 109 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 110 offset %= BITS_PER_LONG;
f96999c3 111
692223d1
FL		 112 while (1) {
113 if (*p == 0)
114 goto pass;
9a7f143a 115
f96999c3 116 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 121 if (tmp)
692223d1
FL		 122 goto found;
123pass:
124 if (size <= BITS_PER_LONG)
125 break;
9a7f143a 126 size -= BITS_PER_LONG;
692223d1 127 offset = 0;
f96999c3 128 p++;
9a7f143a 129 }
692223d1
FL		 130 return result;
131found:
132 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 133}
134
135static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137{
138 const unsigned long *p = addr + BIT_WORD(offset);
80609448 139 unsigned long result = size;
9a7f143a 140 unsigned long tmp;
9a7f143a
CL		 141
142 if (offset >= size)
143 return size;
144
80609448 145 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 146 offset %= BITS_PER_LONG;
80609448
JK		 147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
f96999c3 152 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
f96999c3 158 if (tmp != ~0UL)
80609448
JK		 159 goto found;
160pass:
161 if (size <= BITS_PER_LONG)
162 break;
9a7f143a 163 size -= BITS_PER_LONG;
80609448 164 offset = 0;
f96999c3 165 p++;
9a7f143a 166 }
80609448
JK		 167 return result;
168found:
169 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 170}
171
b3a97a2a
JK		 172bool need_SSR(struct f2fs_sb_info *sbi)
173{
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177
178 if (test_opt(sbi, LFS))
179 return false;
180 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
184 2 * reserved_sections(sbi));
185}
186
88b88a66
JK		 187void register_inmem_page(struct inode *inode, struct page *page)
188{
189 struct f2fs_inode_info *fi = F2FS_I(inode);
190 struct inmem_pages *new;
9be32d72 191
9e4ded3f 192 f2fs_trace_pid(page);
0722b101 193
decd36b6
CY		 194 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
195 SetPagePrivate(page);
196
88b88a66
JK		 197 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
198
199 /* add atomic page indices to the list */
200 new->page = page;
201 INIT_LIST_HEAD(&new->list);
decd36b6 202
88b88a66
JK		 203 /* increase reference count with clean state */
204 mutex_lock(&fi->inmem_lock);
205 get_page(page);
206 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 207 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 208 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 209
210 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 211}
212
28bc106b
CY		 213static int __revoke_inmem_pages(struct inode *inode,
214 struct list_head *head, bool drop, bool recover)
29b96b54 215{
28bc106b 216 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 217 struct inmem_pages *cur, *tmp;
28bc106b 218 int err = 0;
29b96b54
CY		 219
220 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 221 struct page *page = cur->page;
222
223 if (drop)
224 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
225
226 lock_page(page);
29b96b54 227
28bc106b
CY		 228 if (recover) {
229 struct dnode_of_data dn;
230 struct node_info ni;
231
232 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 233retry:
28bc106b 234 set_new_dnode(&dn, inode, NULL, NULL, 0);
7f2b4e8e
CY		 235 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
236 if (err) {
237 if (err == -ENOMEM) {
238 congestion_wait(BLK_RW_ASYNC, HZ/50);
239 cond_resched();
240 goto retry;
241 }
28bc106b
CY		 242 err = -EAGAIN;
243 goto next;
244 }
245 get_node_info(sbi, dn.nid, &ni);
246 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
247 cur->old_addr, ni.version, true, true);
248 f2fs_put_dnode(&dn);
249 }
250next:
63c52d78
JK		 251 /* we don't need to invalidate this in the sccessful status */
252 if (drop || recover)
253 ClearPageUptodate(page);
28bc106b 254 set_page_private(page, 0);
c81ced05 255 ClearPagePrivate(page);
28bc106b 256 f2fs_put_page(page, 1);
29b96b54
CY		 257
258 list_del(&cur->list);
259 kmem_cache_free(inmem_entry_slab, cur);
260 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
261 }
28bc106b 262 return err;
29b96b54
CY		 263}
264
265void drop_inmem_pages(struct inode *inode)
266{
267 struct f2fs_inode_info *fi = F2FS_I(inode);
268
269 mutex_lock(&fi->inmem_lock);
28bc106b 270 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54 271 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 272
273 clear_inode_flag(inode, FI_ATOMIC_FILE);
84a23fbe 274 clear_inode_flag(inode, FI_HOT_DATA);
5fe45743 275 stat_dec_atomic_write(inode);
29b96b54
CY		 276}
277
8c242db9
JK		 278void drop_inmem_page(struct inode *inode, struct page *page)
279{
280 struct f2fs_inode_info *fi = F2FS_I(inode);
281 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
282 struct list_head *head = &fi->inmem_pages;
283 struct inmem_pages *cur = NULL;
284
285 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
286
287 mutex_lock(&fi->inmem_lock);
288 list_for_each_entry(cur, head, list) {
289 if (cur->page == page)
290 break;
291 }
292
293 f2fs_bug_on(sbi, !cur || cur->page != page);
294 list_del(&cur->list);
295 mutex_unlock(&fi->inmem_lock);
296
297 dec_page_count(sbi, F2FS_INMEM_PAGES);
298 kmem_cache_free(inmem_entry_slab, cur);
299
300 ClearPageUptodate(page);
301 set_page_private(page, 0);
302 ClearPagePrivate(page);
303 f2fs_put_page(page, 0);
304
305 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
306}
307
28bc106b
CY		 308static int __commit_inmem_pages(struct inode *inode,
309 struct list_head *revoke_list)
88b88a66
JK		 310{
311 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
312 struct f2fs_inode_info *fi = F2FS_I(inode);
313 struct inmem_pages *cur, *tmp;
88b88a66 314 struct f2fs_io_info fio = {
05ca3632 315 .sbi = sbi,
39d787be 316 .ino = inode->i_ino,
88b88a66 317 .type = DATA,
04d328de 318 .op = REQ_OP_WRITE,
70fd7614 319 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 320 .io_type = FS_DATA_IO,
88b88a66 321 };
942fd319 322 pgoff_t last_idx = ULONG_MAX;
edb27dee 323 int err = 0;
88b88a66 324
88b88a66 325 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 326 struct page *page = cur->page;
327
328 lock_page(page);
329 if (page->mapping == inode->i_mapping) {
330 trace_f2fs_commit_inmem_page(page, INMEM);
331
332 set_page_dirty(page);
333 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 334 if (clear_page_dirty_for_io(page)) {
29b96b54 335 inode_dec_dirty_pages(inode);
933439c8
CY		 336 remove_dirty_inode(inode);
337 }
640cc189 338retry:
28bc106b 339 fio.page = page;
e959c8f5 340 fio.old_blkaddr = NULL_ADDR;
4d978078 341 fio.encrypted_page = NULL;
cc15620b 342 fio.need_lock = LOCK_DONE;
29b96b54
CY		 343 err = do_write_data_page(&fio);
344 if (err) {
640cc189
JK		 345 if (err == -ENOMEM) {
346 congestion_wait(BLK_RW_ASYNC, HZ/50);
347 cond_resched();
348 goto retry;
349 }
28bc106b 350 unlock_page(page);
29b96b54 351 break;
70c640b1 352 }
28bc106b
CY		 353 /* record old blkaddr for revoking */
354 cur->old_addr = fio.old_blkaddr;
942fd319 355 last_idx = page->index;
28bc106b
CY		 356 }
357 unlock_page(page);
358 list_move_tail(&cur->list, revoke_list);
88b88a66 359 }
29b96b54 360
942fd319 361 if (last_idx != ULONG_MAX)
b9109b0e 362 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b
CY		 363
364 if (!err)
365 __revoke_inmem_pages(inode, revoke_list, false, false);
366
29b96b54
CY		 367 return err;
368}
369
370int commit_inmem_pages(struct inode *inode)
371{
372 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
373 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 374 struct list_head revoke_list;
375 int err;
29b96b54 376
28bc106b 377 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 378 f2fs_balance_fs(sbi, true);
379 f2fs_lock_op(sbi);
380
5fe45743
CY		 381 set_inode_flag(inode, FI_ATOMIC_COMMIT);
382
29b96b54 383 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 384 err = __commit_inmem_pages(inode, &revoke_list);
385 if (err) {
386 int ret;
387 /*
388 * try to revoke all committed pages, but still we could fail
389 * due to no memory or other reason, if that happened, EAGAIN
390 * will be returned, which means in such case, transaction is
391 * already not integrity, caller should use journal to do the
392 * recovery or rewrite & commit last transaction. For other
393 * error number, revoking was done by filesystem itself.
394 */
395 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
396 if (ret)
397 err = ret;
398
399 /* drop all uncommitted pages */
400 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
401 }
88b88a66
JK		 402 mutex_unlock(&fi->inmem_lock);
403
5fe45743
CY		 404 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
405
29b96b54 406 f2fs_unlock_op(sbi);
edb27dee 407 return err;
88b88a66
JK		 408}
409
0a8165d7 410/*
351df4b2
JK		 411 * This function balances dirty node and dentry pages.
412 * In addition, it controls garbage collection.
413 */
2c4db1a6 414void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 415{
0f348028 416#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 417 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
418 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 419 f2fs_stop_checkpoint(sbi, false);
55523519 420 }
0f348028
CY		 421#endif
422
e589c2c4 423 /* balance_fs_bg is able to be pending */
a7881893 424 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 425 f2fs_balance_fs_bg(sbi);
426
351df4b2 427 /*
029cd28c
JK		 428 * We should do GC or end up with checkpoint, if there are so many dirty
429 * dir/node pages without enough free segments.
351df4b2 430 */
7f3037a5 431 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 432 mutex_lock(&sbi->gc_mutex);
e066b83c 433 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 434 }
435}
436
4660f9c0
JK		 437void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
438{
1dcc336b 439 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 440 if (!available_free_memory(sbi, EXTENT_CACHE))
441 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 442
1b38dc8e
JK		 443 /* check the # of cached NAT entries */
444 if (!available_free_memory(sbi, NAT_ENTRIES))
445 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
446
31696580 447 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 448 try_to_free_nids(sbi, MAX_FREE_NIDS);
449 else
22ad0b6a 450 build_free_nids(sbi, false, false);
31696580 451
1c0f4bf5 452 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 453 return;
31696580 454
1b38dc8e
JK		 455 /* checkpoint is the only way to shrink partial cached entries */
456 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 457 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 458 excess_prefree_segs(sbi) ||
459 excess_dirty_nats(sbi) ||
f455c8a5 460 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 461 if (test_opt(sbi, DATA_FLUSH)) {
462 struct blk_plug plug;
463
464 blk_start_plug(&plug);
36b35a0d 465 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 466 blk_finish_plug(&plug);
467 }
4660f9c0 468 f2fs_sync_fs(sbi->sb, true);
42190d2a 469 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 470 }
4660f9c0
JK		 471}
472
20fda56b
KM		 473static int __submit_flush_wait(struct f2fs_sb_info *sbi,
474 struct block_device *bdev)
3c62be17
JK		 475{
476 struct bio *bio = f2fs_bio_alloc(0);
477 int ret;
478
3adc5fcb 479 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 480 bio_set_dev(bio, bdev);
3c62be17
JK		 481 ret = submit_bio_wait(bio);
482 bio_put(bio);
20fda56b
KM		 483
484 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
485 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 486 return ret;
487}
488
39d787be 489static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 490{
39d787be 491 int ret = 0;
3c62be17
JK		 492 int i;
493
39d787be
CY		 494 if (!sbi->s_ndevs)
495 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 496
39d787be
CY		 497 for (i = 0; i < sbi->s_ndevs; i++) {
498 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
499 continue;
20fda56b
KM		 500 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
501 if (ret)
502 break;
3c62be17
JK		 503 }
504 return ret;
505}
506
2163d198 507static int issue_flush_thread(void *data)
6b4afdd7
JK		 508{
509 struct f2fs_sb_info *sbi = data;
b01a9201 510 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 511 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 512repeat:
513 if (kthread_should_stop())
514 return 0;
515
dc6febb6
CY		 516 sb_start_intwrite(sbi->sb);
517
721bd4d5 518 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 519 struct flush_cmd *cmd, *next;
520 int ret;
521
721bd4d5
GZ		 522 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
523 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
524
39d787be
CY		 525 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
526
527 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 528 atomic_inc(&fcc->issued_flush);
529
721bd4d5
GZ		 530 llist_for_each_entry_safe(cmd, next,
531 fcc->dispatch_list, llnode) {
6b4afdd7 532 cmd->ret = ret;
6b4afdd7
JK		 533 complete(&cmd->wait);
534 }
a688b9d9 535 fcc->dispatch_list = NULL;
6b4afdd7
JK		 536 }
537
dc6febb6
CY		 538 sb_end_intwrite(sbi->sb);
539
a688b9d9 540 wait_event_interruptible(*q,
721bd4d5 541 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 542 goto repeat;
543}
544
39d787be 545int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 546{
b01a9201 547 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 548 struct flush_cmd cmd;
8b8dd65f 549 int ret;
6b4afdd7 550
0f7b2abd
JK		 551 if (test_opt(sbi, NOBARRIER))
552 return 0;
553
8b8dd65f 554 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 555 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 556 atomic_inc(&fcc->issued_flush);
557 return ret;
558 }
740432f8 559
39d787be
CY		 560 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
561 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 562 atomic_dec(&fcc->issing_flush);
563
564 atomic_inc(&fcc->issued_flush);
740432f8
JK		 565 return ret;
566 }
6b4afdd7 567
39d787be 568 cmd.ino = ino;
adf8d90b 569 init_completion(&cmd.wait);
6b4afdd7 570
721bd4d5 571 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 572
6f890df0
CY		 573 /* update issue_list before we wake up issue_flush thread */
574 smp_mb();
575
576 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 577 wake_up(&fcc->flush_wait_queue);
6b4afdd7 578
5eba8c5d
JK		 579 if (fcc->f2fs_issue_flush) {
580 wait_for_completion(&cmd.wait);
8b8dd65f 581 atomic_dec(&fcc->issing_flush);
5eba8c5d 582 } else {
d3238691
CY		 583 struct llist_node *list;
584
585 list = llist_del_all(&fcc->issue_list);
586 if (!list) {
587 wait_for_completion(&cmd.wait);
588 atomic_dec(&fcc->issing_flush);
589 } else {
590 struct flush_cmd *tmp, *next;
591
39d787be 592 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 593
594 llist_for_each_entry_safe(tmp, next, list, llnode) {
595 if (tmp == &cmd) {
596 cmd.ret = ret;
597 atomic_dec(&fcc->issing_flush);
598 continue;
599 }
600 tmp->ret = ret;
601 complete(&tmp->wait);
602 }
603 }
5eba8c5d 604 }
adf8d90b
CY		 605
606 return cmd.ret;
6b4afdd7
JK		 607}
608
2163d198
GZ		 609int create_flush_cmd_control(struct f2fs_sb_info *sbi)
610{
611 dev_t dev = sbi->sb->s_bdev->bd_dev;
612 struct flush_cmd_control *fcc;
613 int err = 0;
614
b01a9201
JK		 615 if (SM_I(sbi)->fcc_info) {
616 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 617 if (fcc->f2fs_issue_flush)
618 return err;
5eba8c5d
JK		 619 goto init_thread;
620 }
621
2163d198
GZ		 622 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
623 if (!fcc)
624 return -ENOMEM;
8b8dd65f
CY		 625 atomic_set(&fcc->issued_flush, 0);
626 atomic_set(&fcc->issing_flush, 0);
2163d198 627 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 628 init_llist_head(&fcc->issue_list);
b01a9201 629 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 630 if (!test_opt(sbi, FLUSH_MERGE))
631 return err;
632
5eba8c5d 633init_thread:
2163d198
GZ		 634 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
635 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
636 if (IS_ERR(fcc->f2fs_issue_flush)) {
637 err = PTR_ERR(fcc->f2fs_issue_flush);
638 kfree(fcc);
b01a9201 639 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 640 return err;
641 }
2163d198
GZ		 642
643 return err;
644}
645
5eba8c5d 646void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 647{
b01a9201 648 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 649
5eba8c5d
JK		 650 if (fcc && fcc->f2fs_issue_flush) {
651 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
652
653 fcc->f2fs_issue_flush = NULL;
654 kthread_stop(flush_thread);
655 }
656 if (free) {
657 kfree(fcc);
b01a9201 658 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 659 }
2163d198
GZ		 660}
661
351df4b2
JK		 662static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
663 enum dirty_type dirty_type)
664{
665 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
666
667 /* need not be added */
668 if (IS_CURSEG(sbi, segno))
669 return;
670
671 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
672 dirty_i->nr_dirty[dirty_type]++;
673
674 if (dirty_type == DIRTY) {
675 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 676 enum dirty_type t = sentry->type;
b2f2c390 677
ec325b52
JK		 678 if (unlikely(t >= DIRTY)) {
679 f2fs_bug_on(sbi, 1);
680 return;
681 }
4625d6aa
CL		 682 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
683 dirty_i->nr_dirty[t]++;
351df4b2
JK		 684 }
685}
686
687static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
688 enum dirty_type dirty_type)
689{
690 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
691
692 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
693 dirty_i->nr_dirty[dirty_type]--;
694
695 if (dirty_type == DIRTY) {
4625d6aa
CL		 696 struct seg_entry *sentry = get_seg_entry(sbi, segno);
697 enum dirty_type t = sentry->type;
698
699 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
700 dirty_i->nr_dirty[t]--;
b2f2c390 701
302bd348 702 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 703 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 704 dirty_i->victim_secmap);
351df4b2
JK		 705 }
706}
707
0a8165d7 708/*
351df4b2
JK		 709 * Should not occur error such as -ENOMEM.
710 * Adding dirty entry into seglist is not critical operation.
711 * If a given segment is one of current working segments, it won't be added.
712 */
8d8451af 713static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 714{
715 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
716 unsigned short valid_blocks;
717
718 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
719 return;
720
721 mutex_lock(&dirty_i->seglist_lock);
722
302bd348 723 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 724
725 if (valid_blocks == 0) {
726 __locate_dirty_segment(sbi, segno, PRE);
727 __remove_dirty_segment(sbi, segno, DIRTY);
728 } else if (valid_blocks < sbi->blocks_per_seg) {
729 __locate_dirty_segment(sbi, segno, DIRTY);
730 } else {
731 /* Recovery routine with SSR needs this */
732 __remove_dirty_segment(sbi, segno, DIRTY);
733 }
734
735 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 736}
737
004b6862 738static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 739 struct block_device *bdev, block_t lstart,
740 block_t start, block_t len)
275b66b0 741{
0b54fb84 742 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 743 struct list_head *pend_list;
b01a9201 744 struct discard_cmd *dc;
275b66b0 745
ba48a33e
CY		 746 f2fs_bug_on(sbi, !len);
747
748 pend_list = &dcc->pend_list[plist_idx(len)];
749
b01a9201
JK		 750 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
751 INIT_LIST_HEAD(&dc->list);
c81abe34 752 dc->bdev = bdev;
b01a9201 753 dc->lstart = lstart;
c81abe34 754 dc->start = start;
b01a9201 755 dc->len = len;
ec9895ad 756 dc->ref = 0;
15469963 757 dc->state = D_PREP;
c81abe34 758 dc->error = 0;
b01a9201 759 init_completion(&dc->wait);
22d375dd 760 list_add_tail(&dc->list, pend_list);
5f32366a 761 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 762 dcc->undiscard_blks += len;
004b6862
CY		 763
764 return dc;
765}
766
767static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
768 struct block_device *bdev, block_t lstart,
769 block_t start, block_t len,
770 struct rb_node *parent, struct rb_node **p)
771{
772 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
773 struct discard_cmd *dc;
774
775 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
776
777 rb_link_node(&dc->rb_node, parent, p);
778 rb_insert_color(&dc->rb_node, &dcc->root);
779
780 return dc;
15469963
JK		 781}
782
004b6862
CY		 783static void __detach_discard_cmd(struct discard_cmd_control *dcc,
784 struct discard_cmd *dc)
15469963 785{
dcc9165d 786 if (dc->state == D_DONE)
004b6862
CY		 787 atomic_dec(&dcc->issing_discard);
788
789 list_del(&dc->list);
790 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 791 dcc->undiscard_blks -= dc->len;
004b6862
CY		 792
793 kmem_cache_free(discard_cmd_slab, dc);
794
795 atomic_dec(&dcc->discard_cmd_cnt);
796}
797
798static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
799 struct discard_cmd *dc)
800{
801 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 802
d9703d90
CY		 803 f2fs_bug_on(sbi, dc->ref);
804
c81abe34
JK		 805 if (dc->error == -EOPNOTSUPP)
806 dc->error = 0;
15469963 807
c81abe34 808 if (dc->error)
15469963 809 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 810 "Issue discard(%u, %u, %u) failed, ret: %d",
811 dc->lstart, dc->start, dc->len, dc->error);
004b6862 812 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 813}
814
c81abe34
JK		 815static void f2fs_submit_discard_endio(struct bio *bio)
816{
817 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
818
4e4cbee9 819 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 820 dc->state = D_DONE;
e31b9821 821 complete_all(&dc->wait);
c81abe34
JK		 822 bio_put(bio);
823}
824
6915ea9d
CY		 825void __check_sit_bitmap(struct f2fs_sb_info *sbi,
826 block_t start, block_t end)
827{
828#ifdef CONFIG_F2FS_CHECK_FS
829 struct seg_entry *sentry;
830 unsigned int segno;
831 block_t blk = start;
832 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
833 unsigned long *map;
834
835 while (blk < end) {
836 segno = GET_SEGNO(sbi, blk);
837 sentry = get_seg_entry(sbi, segno);
838 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
839
008396e1
YS		 840 if (end < START_BLOCK(sbi, segno + 1))
841 size = GET_BLKOFF_FROM_SEG0(sbi, end);
842 else
843 size = max_blocks;
6915ea9d
CY		 844 map = (unsigned long *)(sentry->cur_valid_map);
845 offset = __find_rev_next_bit(map, size, offset);
846 f2fs_bug_on(sbi, offset != size);
008396e1 847 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 848 }
849#endif
850}
851
c81abe34
JK		 852/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
853static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
854 struct discard_cmd *dc)
855{
856 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
857 struct bio *bio = NULL;
858
859 if (dc->state != D_PREP)
860 return;
861
0243a5f9
CY		 862 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
863
c81abe34
JK		 864 dc->error = __blkdev_issue_discard(dc->bdev,
865 SECTOR_FROM_BLOCK(dc->start),
866 SECTOR_FROM_BLOCK(dc->len),
867 GFP_NOFS, 0, &bio);
868 if (!dc->error) {
869 /* should keep before submission to avoid D_DONE right away */
870 dc->state = D_SUBMIT;
8b8dd65f
CY		 871 atomic_inc(&dcc->issued_discard);
872 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 873 if (bio) {
874 bio->bi_private = dc;
875 bio->bi_end_io = f2fs_submit_discard_endio;
876 bio->bi_opf |= REQ_SYNC;
877 submit_bio(bio);
46f84c2c 878 list_move_tail(&dc->list, &dcc->wait_list);
6915ea9d 879 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 880
881 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 882 }
883 } else {
884 __remove_discard_cmd(sbi, dc);
885 }
886}
887
004b6862
CY		 888static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
889 struct block_device *bdev, block_t lstart,
890 block_t start, block_t len,
891 struct rb_node **insert_p,
892 struct rb_node *insert_parent)
c81abe34 893{
004b6862
CY		 894 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
895 struct rb_node **p = &dcc->root.rb_node;
896 struct rb_node *parent = NULL;
897 struct discard_cmd *dc = NULL;
898
899 if (insert_p && insert_parent) {
900 parent = insert_parent;
901 p = insert_p;
902 goto do_insert;
903 }
c81abe34 904
004b6862
CY		 905 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
906do_insert:
907 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
908 if (!dc)
909 return NULL;
c81abe34 910
004b6862 911 return dc;
c81abe34
JK		 912}
913
ba48a33e
CY		 914static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
915 struct discard_cmd *dc)
916{
917 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
918}
919
3d6a650f
YH		 920static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
921 struct discard_cmd *dc, block_t blkaddr)
922{
ba48a33e 923 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 924 struct discard_info di = dc->di;
925 bool modified = false;
3d6a650f 926
004b6862 927 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 928 __remove_discard_cmd(sbi, dc);
929 return;
930 }
931
d84d1cbd
CY		 932 dcc->undiscard_blks -= di.len;
933
004b6862 934 if (blkaddr > di.lstart) {
3d6a650f 935 dc->len = blkaddr - dc->lstart;
d84d1cbd 936 dcc->undiscard_blks += dc->len;
ba48a33e 937 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 938 modified = true;
939 }
940
941 if (blkaddr < di.lstart + di.len - 1) {
942 if (modified) {
943 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
944 di.start + blkaddr + 1 - di.lstart,
945 di.lstart + di.len - 1 - blkaddr,
946 NULL, NULL);
947 } else {
948 dc->lstart++;
949 dc->len--;
950 dc->start++;
d84d1cbd 951 dcc->undiscard_blks += dc->len;
ba48a33e 952 __relocate_discard_cmd(dcc, dc);
004b6862 953 }
3d6a650f
YH		 954 }
955}
956
004b6862
CY		 957static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
958 struct block_device *bdev, block_t lstart,
959 block_t start, block_t len)
275b66b0 960{
0b54fb84 961 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 962 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
963 struct discard_cmd *dc;
964 struct discard_info di = {0};
965 struct rb_node **insert_p = NULL, *insert_parent = NULL;
966 block_t end = lstart + len;
275b66b0 967
15469963 968 mutex_lock(&dcc->cmd_lock);
40465257 969
004b6862
CY		 970 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
971 NULL, lstart,
972 (struct rb_entry **)&prev_dc,
973 (struct rb_entry **)&next_dc,
974 &insert_p, &insert_parent, true);
975 if (dc)
976 prev_dc = dc;
977
978 if (!prev_dc) {
979 di.lstart = lstart;
980 di.len = next_dc ? next_dc->lstart - lstart : len;
981 di.len = min(di.len, len);
982 di.start = start;
22d375dd 983 }
15469963 984
004b6862
CY		 985 while (1) {
986 struct rb_node *node;
987 bool merged = false;
988 struct discard_cmd *tdc = NULL;
989
990 if (prev_dc) {
991 di.lstart = prev_dc->lstart + prev_dc->len;
992 if (di.lstart < lstart)
993 di.lstart = lstart;
994 if (di.lstart >= end)
995 break;
996
997 if (!next_dc || next_dc->lstart > end)
998 di.len = end - di.lstart;
999 else
1000 di.len = next_dc->lstart - di.lstart;
1001 di.start = start + di.lstart - lstart;
1002 }
1003
1004 if (!di.len)
1005 goto next;
1006
1007 if (prev_dc && prev_dc->state == D_PREP &&
1008 prev_dc->bdev == bdev &&
1009 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1010 prev_dc->di.len += di.len;
d84d1cbd 1011 dcc->undiscard_blks += di.len;
ba48a33e 1012 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1013 di = prev_dc->di;
1014 tdc = prev_dc;
1015 merged = true;
1016 }
1017
1018 if (next_dc && next_dc->state == D_PREP &&
1019 next_dc->bdev == bdev &&
1020 __is_discard_front_mergeable(&di, &next_dc->di)) {
1021 next_dc->di.lstart = di.lstart;
1022 next_dc->di.len += di.len;
1023 next_dc->di.start = di.start;
d84d1cbd 1024 dcc->undiscard_blks += di.len;
ba48a33e 1025 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1026 if (tdc)
1027 __remove_discard_cmd(sbi, tdc);
004b6862 1028 merged = true;
4e6a8d9b 1029 }
004b6862 1030
df0f6b44 1031 if (!merged) {
004b6862
CY		 1032 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1033 di.len, NULL, NULL);
df0f6b44 1034 }
004b6862
CY		 1035 next:
1036 prev_dc = next_dc;
1037 if (!prev_dc)
1038 break;
1039
1040 node = rb_next(&prev_dc->rb_node);
1041 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1042 }
1043
1044 mutex_unlock(&dcc->cmd_lock);
1045}
1046
1047static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1048 struct block_device *bdev, block_t blkstart, block_t blklen)
1049{
1050 block_t lblkstart = blkstart;
1051
0243a5f9 1052 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1053
1054 if (sbi->s_ndevs) {
1055 int devi = f2fs_target_device_index(sbi, blkstart);
1056
1057 blkstart -= FDEV(devi).start_blk;
1058 }
1059 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1060 return 0;
1061}
1062
969d1b18 1063static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
bd5b0738
CY		 1064{
1065 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1066 struct list_head *pend_list;
1067 struct discard_cmd *dc, *tmp;
1068 struct blk_plug plug;
969d1b18
CY		 1069 int iter = 0, issued = 0;
1070 int i;
e6c6de18 1071 bool io_interrupted = false;
bd5b0738
CY		 1072
1073 mutex_lock(&dcc->cmd_lock);
963932a9
YH		 1074 f2fs_bug_on(sbi,
1075 !__check_rb_tree_consistence(sbi, &dcc->root));
bd5b0738 1076 blk_start_plug(&plug);
969d1b18
CY		 1077 for (i = MAX_PLIST_NUM - 1;
1078 i >= 0 && plist_issue(dcc->pend_list_tag[i]); i--) {
bd5b0738
CY		 1079 pend_list = &dcc->pend_list[i];
1080 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1081 f2fs_bug_on(sbi, dc->state != D_PREP);
1082
969d1b18
CY		 1083 /* Hurry up to finish fstrim */
1084 if (dcc->pend_list_tag[i] & P_TRIM) {
bd5b0738 1085 __submit_discard_cmd(sbi, dc);
969d1b18 1086 issued++;
1eb1ef4a
JK		 1087
1088 if (fatal_signal_pending(current))
1089 break;
969d1b18
CY		 1090 continue;
1091 }
1092
e6c6de18
CY		 1093 if (!issue_cond) {
1094 __submit_discard_cmd(sbi, dc);
969d1b18 1095 issued++;
e6c6de18
CY		 1096 continue;
1097 }
1098
1099 if (is_idle(sbi)) {
bd5b0738 1100 __submit_discard_cmd(sbi, dc);
e6c6de18
CY		 1101 issued++;
1102 } else {
1103 io_interrupted = true;
969d1b18 1104 }
e6c6de18
CY		 1105
1106 if (++iter >= DISCARD_ISSUE_RATE)
bd5b0738
CY		 1107 goto out;
1108 }
969d1b18
CY		 1109 if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM)
1110 dcc->pend_list_tag[i] &= (~P_TRIM);
bd5b0738
CY		 1111 }
1112out:
1113 blk_finish_plug(&plug);
1114 mutex_unlock(&dcc->cmd_lock);
969d1b18 1115
e6c6de18
CY		 1116 if (!issued && io_interrupted)
1117 issued = -1;
1118
969d1b18
CY		 1119 return issued;
1120}
1121
1122static void __drop_discard_cmd(struct f2fs_sb_info *sbi)
1123{
1124 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1125 struct list_head *pend_list;
1126 struct discard_cmd *dc, *tmp;
1127 int i;
1128
1129 mutex_lock(&dcc->cmd_lock);
1130 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1131 pend_list = &dcc->pend_list[i];
1132 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1133 f2fs_bug_on(sbi, dc->state != D_PREP);
1134 __remove_discard_cmd(sbi, dc);
1135 }
1136 }
1137 mutex_unlock(&dcc->cmd_lock);
bd5b0738
CY		 1138}
1139
2a510c00
CY		 1140static void __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1141 struct discard_cmd *dc)
1142{
1143 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1144
1145 wait_for_completion_io(&dc->wait);
1146 mutex_lock(&dcc->cmd_lock);
1147 f2fs_bug_on(sbi, dc->state != D_DONE);
1148 dc->ref--;
1149 if (!dc->ref)
1150 __remove_discard_cmd(sbi, dc);
1151 mutex_unlock(&dcc->cmd_lock);
1152}
1153
63a94fa1
CY		 1154static void __wait_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1155{
1156 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1157 struct list_head *wait_list = &(dcc->wait_list);
1158 struct discard_cmd *dc, *tmp;
6afae633
CY		 1159 bool need_wait;
1160
1161next:
1162 need_wait = false;
63a94fa1
CY		 1163
1164 mutex_lock(&dcc->cmd_lock);
1165 list_for_each_entry_safe(dc, tmp, wait_list, list) {
6afae633 1166 if (!wait_cond || (dc->state == D_DONE && !dc->ref)) {
63a94fa1
CY		 1167 wait_for_completion_io(&dc->wait);
1168 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1169 } else {
1170 dc->ref++;
1171 need_wait = true;
1172 break;
63a94fa1
CY		 1173 }
1174 }
1175 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1176
1177 if (need_wait) {
2a510c00 1178 __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1179 goto next;
1180 }
63a94fa1
CY		 1181}
1182
004b6862
CY		 1183/* This should be covered by global mutex, &sit_i->sentry_lock */
1184void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1185{
1186 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1187 struct discard_cmd *dc;
ec9895ad 1188 bool need_wait = false;
004b6862
CY		 1189
1190 mutex_lock(&dcc->cmd_lock);
004b6862
CY		 1191 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1192 if (dc) {
ec9895ad
CY		 1193 if (dc->state == D_PREP) {
1194 __punch_discard_cmd(sbi, dc, blkaddr);
1195 } else {
1196 dc->ref++;
1197 need_wait = true;
1198 }
275b66b0 1199 }
d431413f 1200 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1201
2a510c00
CY		 1202 if (need_wait)
1203 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1204}
1205
cce13252
CY		 1206void stop_discard_thread(struct f2fs_sb_info *sbi)
1207{
1208 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1209
1210 if (dcc && dcc->f2fs_issue_discard) {
1211 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1212
1213 dcc->f2fs_issue_discard = NULL;
1214 kthread_stop(discard_thread);
ec9895ad 1215 }
d431413f
CY		 1216}
1217
1eb1ef4a 1218/* This comes from f2fs_put_super and f2fs_trim_fs */
638164a2 1219void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi, bool umount)
d431413f 1220{
bd5b0738 1221 __issue_discard_cmd(sbi, false);
969d1b18 1222 __drop_discard_cmd(sbi);
638164a2 1223 __wait_discard_cmd(sbi, !umount);
275b66b0
CY		 1224}
1225
969d1b18
CY		 1226static void mark_discard_range_all(struct f2fs_sb_info *sbi)
1227{
1228 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1229 int i;
1230
1231 mutex_lock(&dcc->cmd_lock);
1232 for (i = 0; i < MAX_PLIST_NUM; i++)
1233 dcc->pend_list_tag[i] |= P_TRIM;
1234 mutex_unlock(&dcc->cmd_lock);
1235}
1236
15469963
JK		 1237static int issue_discard_thread(void *data)
1238{
1239 struct f2fs_sb_info *sbi = data;
1240 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1241 wait_queue_head_t *q = &dcc->discard_wait_queue;
969d1b18
CY		 1242 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1243 int issued;
15469963 1244
1d7be270 1245 set_freezable();
15469963 1246
1d7be270 1247 do {
969d1b18
CY		 1248 wait_event_interruptible_timeout(*q,
1249 kthread_should_stop() || freezing(current) ||
1250 dcc->discard_wake,
1251 msecs_to_jiffies(wait_ms));
1d7be270
JK		 1252 if (try_to_freeze())
1253 continue;
1254 if (kthread_should_stop())
1255 return 0;
15469963 1256
5f656541 1257 if (dcc->discard_wake) {
969d1b18 1258 dcc->discard_wake = 0;
5f656541
JK		 1259 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1260 mark_discard_range_all(sbi);
1261 }
969d1b18 1262
dc6febb6
CY		 1263 sb_start_intwrite(sbi->sb);
1264
969d1b18
CY		 1265 issued = __issue_discard_cmd(sbi, true);
1266 if (issued) {
1267 __wait_discard_cmd(sbi, true);
1268 wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1269 } else {
1270 wait_ms = DEF_MAX_DISCARD_ISSUE_TIME;
1271 }
1d7be270 1272
dc6febb6 1273 sb_end_intwrite(sbi->sb);
1d7be270 1274
1d7be270
JK		 1275 } while (!kthread_should_stop());
1276 return 0;
15469963
JK		 1277}
1278
f46e8809 1279#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1280static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1281 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1282{
92592285 1283 sector_t sector, nr_sects;
10a875f8 1284 block_t lblkstart = blkstart;
3c62be17
JK		 1285 int devi = 0;
1286
1287 if (sbi->s_ndevs) {
1288 devi = f2fs_target_device_index(sbi, blkstart);
1289 blkstart -= FDEV(devi).start_blk;
1290 }
f46e8809
DLM		 1291
1292 /*
1293 * We need to know the type of the zone: for conventional zones,
1294 * use regular discard if the drive supports it. For sequential
1295 * zones, reset the zone write pointer.
1296 */
3c62be17 1297 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1298
1299 case BLK_ZONE_TYPE_CONVENTIONAL:
1300 if (!blk_queue_discard(bdev_get_queue(bdev)))
1301 return 0;
c81abe34 1302 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1303 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1304 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1305 sector = SECTOR_FROM_BLOCK(blkstart);
1306 nr_sects = SECTOR_FROM_BLOCK(blklen);
1307
1308 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1309 nr_sects != bdev_zone_sectors(bdev)) {
1310 f2fs_msg(sbi->sb, KERN_INFO,
1311 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1312 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1313 blkstart, blklen);
1314 return -EIO;
1315 }
d50aaeec 1316 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1317 return blkdev_reset_zones(bdev, sector,
1318 nr_sects, GFP_NOFS);
1319 default:
1320 /* Unknown zone type: broken device ? */
1321 return -EIO;
1322 }
1323}
1324#endif
1325
3c62be17
JK		 1326static int __issue_discard_async(struct f2fs_sb_info *sbi,
1327 struct block_device *bdev, block_t blkstart, block_t blklen)
1328{
1329#ifdef CONFIG_BLK_DEV_ZONED
1330 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1331 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1332 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1333#endif
c81abe34 1334 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1335}
1336
1e87a78d 1337static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1338 block_t blkstart, block_t blklen)
1339{
3c62be17
JK		 1340 sector_t start = blkstart, len = 0;
1341 struct block_device *bdev;
a66cdd98
JK		 1342 struct seg_entry *se;
1343 unsigned int offset;
1344 block_t i;
3c62be17
JK		 1345 int err = 0;
1346
1347 bdev = f2fs_target_device(sbi, blkstart, NULL);
1348
1349 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1350 if (i != start) {
1351 struct block_device *bdev2 =
1352 f2fs_target_device(sbi, i, NULL);
1353
1354 if (bdev2 != bdev) {
1355 err = __issue_discard_async(sbi, bdev,
1356 start, len);
1357 if (err)
1358 return err;
1359 bdev = bdev2;
1360 start = i;
1361 len = 0;
1362 }
1363 }
a66cdd98 1364
a66cdd98
JK		 1365 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1366 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1367
1368 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1369 sbi->discard_blks--;
1370 }
f46e8809 1371
3c62be17
JK		 1372 if (len)
1373 err = __issue_discard_async(sbi, bdev, start, len);
1374 return err;
1e87a78d
JK		 1375}
1376
25290fa5
JK		 1377static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1378 bool check_only)
adf4983b 1379{
b2955550
JK		 1380 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1381 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1382 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1383 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1384 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1385 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1386 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1387 unsigned int start = 0, end = -1;
c473f1a9 1388 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1389 struct discard_entry *de = NULL;
46f84c2c 1390 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1391 int i;
1392
3e025740 1393 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1394 return false;
b2955550 1395
a66cdd98
JK		 1396 if (!force) {
1397 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1398 SM_I(sbi)->dcc_info->nr_discards >=
1399 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1400 return false;
4b2fecc8
JK		 1401 }
1402
b2955550
JK		 1403 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1404 for (i = 0; i < entries; i++)
a66cdd98 1405 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1406 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1407
0b54fb84
JK		 1408 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1409 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1410 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1411 if (start >= max_blocks)
1412 break;
1413
1414 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1415 if (force && start && end != max_blocks
1416 && (end - start) < cpc->trim_minlen)
1417 continue;
1418
25290fa5
JK		 1419 if (check_only)
1420 return true;
1421
a7eeb823
CY		 1422 if (!de) {
1423 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1424 GFP_F2FS_ZERO);
1425 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1426 list_add_tail(&de->list, head);
1427 }
1428
1429 for (i = start; i < end; i++)
1430 __set_bit_le(i, (void *)de->discard_map);
1431
1432 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1433 }
25290fa5 1434 return false;
b2955550
JK		 1435}
1436
4b2fecc8
JK		 1437void release_discard_addrs(struct f2fs_sb_info *sbi)
1438{
46f84c2c 1439 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1440 struct discard_entry *entry, *this;
1441
1442 /* drop caches */
1443 list_for_each_entry_safe(entry, this, head, list) {
1444 list_del(&entry->list);
1445 kmem_cache_free(discard_entry_slab, entry);
1446 }
1447}
1448
0a8165d7 1449/*
351df4b2
JK		 1450 * Should call clear_prefree_segments after checkpoint is done.
1451 */
1452static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1453{
1454 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1455 unsigned int segno;
351df4b2
JK		 1456
1457 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1458 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1459 __set_test_and_free(sbi, segno);
351df4b2
JK		 1460 mutex_unlock(&dirty_i->seglist_lock);
1461}
1462
836b5a63 1463void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1464{
969d1b18
CY		 1465 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1466 struct list_head *head = &dcc->entry_list;
2d7b822a 1467 struct discard_entry *entry, *this;
351df4b2 1468 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1469 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1470 unsigned int start = 0, end = -1;
36abef4e 1471 unsigned int secno, start_segno;
c473f1a9 1472 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK		 1473
1474 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1475
351df4b2 1476 while (1) {
29e59c14 1477 int i;
7cd8558b
JK		 1478 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1479 if (start >= MAIN_SEGS(sbi))
351df4b2 1480 break;
7cd8558b
JK		 1481 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1482 start + 1);
29e59c14
CL		 1483
1484 for (i = start; i < end; i++)
1485 clear_bit(i, prefree_map);
1486
1487 dirty_i->nr_dirty[PRE] -= end - start;
1488
650d3c4e 1489 if (!test_opt(sbi, DISCARD))
29e59c14 1490 continue;
351df4b2 1491
650d3c4e
YH		 1492 if (force && start >= cpc->trim_start &&
1493 (end - 1) <= cpc->trim_end)
1494 continue;
1495
36abef4e
JK		 1496 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1497 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1498 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1499 continue;
1500 }
1501next:
4ddb1a4d
JK		 1502 secno = GET_SEC_FROM_SEG(sbi, start);
1503 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1504 if (!IS_CURSEC(sbi, secno) &&
302bd348 1505 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1506 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1507 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1508
1509 start = start_segno + sbi->segs_per_sec;
1510 if (start < end)
1511 goto next;
8b107f5b
JK		 1512 else
1513 end = start - 1;
351df4b2
JK		 1514 }
1515 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1516
1517 /* send small discards */
2d7b822a 1518 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1519 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1520 bool is_valid = test_bit_le(0, entry->discard_map);
1521
1522find_next:
1523 if (is_valid) {
1524 next_pos = find_next_zero_bit_le(entry->discard_map,
1525 sbi->blocks_per_seg, cur_pos);
1526 len = next_pos - cur_pos;
1527
acfd2810
DLM		 1528 if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1529 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1530 goto skip;
1531
1532 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1533 len);
1534 cpc->trimmed += len;
1535 total_len += len;
1536 } else {
1537 next_pos = find_next_bit_le(entry->discard_map,
1538 sbi->blocks_per_seg, cur_pos);
1539 }
836b5a63 1540skip:
a7eeb823
CY		 1541 cur_pos = next_pos;
1542 is_valid = !is_valid;
1543
1544 if (cur_pos < sbi->blocks_per_seg)
1545 goto find_next;
1546
b2955550 1547 list_del(&entry->list);
969d1b18 1548 dcc->nr_discards -= total_len;
b2955550
JK		 1549 kmem_cache_free(discard_entry_slab, entry);
1550 }
34e159da 1551
01983c71 1552 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1553}
1554
8ed59745 1555static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1556{
15469963 1557 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1558 struct discard_cmd_control *dcc;
ba48a33e 1559 int err = 0, i;
0b54fb84
JK		 1560
1561 if (SM_I(sbi)->dcc_info) {
1562 dcc = SM_I(sbi)->dcc_info;
1563 goto init_thread;
1564 }
1565
1566 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1567 if (!dcc)
1568 return -ENOMEM;
1569
969d1b18 1570 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1571 INIT_LIST_HEAD(&dcc->entry_list);
969d1b18 1572 for (i = 0; i < MAX_PLIST_NUM; i++) {
ba48a33e 1573 INIT_LIST_HEAD(&dcc->pend_list[i]);
969d1b18
CY		 1574 if (i >= dcc->discard_granularity - 1)
1575 dcc->pend_list_tag[i] |= P_ACTIVE;
1576 }
46f84c2c 1577 INIT_LIST_HEAD(&dcc->wait_list);
15469963 1578 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1579 atomic_set(&dcc->issued_discard, 0);
1580 atomic_set(&dcc->issing_discard, 0);
5f32366a 1581 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1582 dcc->nr_discards = 0;
d618ebaf 1583 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1584 dcc->undiscard_blks = 0;
004b6862 1585 dcc->root = RB_ROOT;
0b54fb84 1586
15469963 1587 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1588 SM_I(sbi)->dcc_info = dcc;
1589init_thread:
15469963
JK		 1590 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1591 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1592 if (IS_ERR(dcc->f2fs_issue_discard)) {
1593 err = PTR_ERR(dcc->f2fs_issue_discard);
1594 kfree(dcc);
1595 SM_I(sbi)->dcc_info = NULL;
1596 return err;
1597 }
1598
0b54fb84
JK		 1599 return err;
1600}
1601
f099405f 1602static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1603{
1604 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1605
f099405f
CY		 1606 if (!dcc)
1607 return;
1608
cce13252 1609 stop_discard_thread(sbi);
f099405f
CY		 1610
1611 kfree(dcc);
1612 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1613}
1614
184a5cd2 1615static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1616{
1617 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1618
1619 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1620 sit_i->dirty_sentries++;
184a5cd2
CY		 1621 return false;
1622 }
1623
1624 return true;
351df4b2
JK		 1625}
1626
1627static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1628 unsigned int segno, int modified)
1629{
1630 struct seg_entry *se = get_seg_entry(sbi, segno);
1631 se->type = type;
1632 if (modified)
1633 __mark_sit_entry_dirty(sbi, segno);
1634}
1635
1636static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1637{
1638 struct seg_entry *se;
1639 unsigned int segno, offset;
1640 long int new_vblocks;
6415fedc
YS		 1641 bool exist;
1642#ifdef CONFIG_F2FS_CHECK_FS
1643 bool mir_exist;
1644#endif
351df4b2
JK		 1645
1646 segno = GET_SEGNO(sbi, blkaddr);
1647
1648 se = get_seg_entry(sbi, segno);
1649 new_vblocks = se->valid_blocks + del;
491c0854 1650 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1651
9850cf4a 1652 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1653 (new_vblocks > sbi->blocks_per_seg)));
1654
1655 se->valid_blocks = new_vblocks;
1656 se->mtime = get_mtime(sbi);
1657 SIT_I(sbi)->max_mtime = se->mtime;
1658
1659 /* Update valid block bitmap */
1660 if (del > 0) {
6415fedc 1661 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1662#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1663 mir_exist = f2fs_test_and_set_bit(offset,
1664 se->cur_valid_map_mir);
1665 if (unlikely(exist != mir_exist)) {
1666 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1667 "when setting bitmap, blk:%u, old bit:%d",
1668 blkaddr, exist);
05796763 1669 f2fs_bug_on(sbi, 1);
6415fedc 1670 }
355e7891 1671#endif
6415fedc
YS		 1672 if (unlikely(exist)) {
1673 f2fs_msg(sbi->sb, KERN_ERR,
1674 "Bitmap was wrongly set, blk:%u", blkaddr);
1675 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1676 se->valid_blocks--;
1677 del = 0;
355e7891 1678 }
6415fedc 1679
3e025740
JK		 1680 if (f2fs_discard_en(sbi) &&
1681 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1682 sbi->discard_blks--;
720037f9
JK		 1683
1684 /* don't overwrite by SSR to keep node chain */
1685 if (se->type == CURSEG_WARM_NODE) {
1686 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1687 se->ckpt_valid_blocks++;
1688 }
351df4b2 1689 } else {
6415fedc 1690 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1691#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1692 mir_exist = f2fs_test_and_clear_bit(offset,
1693 se->cur_valid_map_mir);
1694 if (unlikely(exist != mir_exist)) {
1695 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1696 "when clearing bitmap, blk:%u, old bit:%d",
1697 blkaddr, exist);
05796763 1698 f2fs_bug_on(sbi, 1);
6415fedc 1699 }
355e7891 1700#endif
6415fedc
YS		 1701 if (unlikely(!exist)) {
1702 f2fs_msg(sbi->sb, KERN_ERR,
1703 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1704 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1705 se->valid_blocks++;
1706 del = 0;
355e7891 1707 }
6415fedc 1708
3e025740
JK		 1709 if (f2fs_discard_en(sbi) &&
1710 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1711 sbi->discard_blks++;
351df4b2
JK		 1712 }
1713 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1714 se->ckpt_valid_blocks += del;
1715
1716 __mark_sit_entry_dirty(sbi, segno);
1717
1718 /* update total number of valid blocks to be written in ckpt area */
1719 SIT_I(sbi)->written_valid_blocks += del;
1720
1721 if (sbi->segs_per_sec > 1)
1722 get_sec_entry(sbi, segno)->valid_blocks += del;
1723}
1724
5e443818 1725void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1726{
5e443818
JK		 1727 update_sit_entry(sbi, new, 1);
1728 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1729 update_sit_entry(sbi, old, -1);
1730
1731 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1732 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 1733}
1734
1735void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1736{
1737 unsigned int segno = GET_SEGNO(sbi, addr);
1738 struct sit_info *sit_i = SIT_I(sbi);
1739
9850cf4a 1740 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1741 if (addr == NEW_ADDR)
1742 return;
1743
1744 /* add it into sit main buffer */
1745 mutex_lock(&sit_i->sentry_lock);
1746
1747 update_sit_entry(sbi, addr, -1);
1748
1749 /* add it into dirty seglist */
1750 locate_dirty_segment(sbi, segno);
1751
1752 mutex_unlock(&sit_i->sentry_lock);
1753}
1754
6e2c64ad
JK		 1755bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1756{
1757 struct sit_info *sit_i = SIT_I(sbi);
1758 unsigned int segno, offset;
1759 struct seg_entry *se;
1760 bool is_cp = false;
1761
1762 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1763 return true;
1764
1765 mutex_lock(&sit_i->sentry_lock);
1766
1767 segno = GET_SEGNO(sbi, blkaddr);
1768 se = get_seg_entry(sbi, segno);
1769 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1770
1771 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1772 is_cp = true;
1773
1774 mutex_unlock(&sit_i->sentry_lock);
1775
1776 return is_cp;
1777}
1778
0a8165d7 1779/*
351df4b2
JK		 1780 * This function should be resided under the curseg_mutex lock
1781 */
1782static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1783 struct f2fs_summary *sum)
351df4b2
JK		 1784{
1785 struct curseg_info *curseg = CURSEG_I(sbi, type);
1786 void *addr = curseg->sum_blk;
e79efe3b 1787 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1788 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1789}
1790
0a8165d7 1791/*
351df4b2
JK		 1792 * Calculate the number of current summary pages for writing
1793 */
3fa06d7b 1794int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1795{
351df4b2 1796 int valid_sum_count = 0;
9a47938b 1797 int i, sum_in_page;
351df4b2
JK		 1798
1799 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1800 if (sbi->ckpt->alloc_type[i] == SSR)
1801 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1802 else {
1803 if (for_ra)
1804 valid_sum_count += le16_to_cpu(
1805 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1806 else
1807 valid_sum_count += curseg_blkoff(sbi, i);
1808 }
351df4b2
JK		 1809 }
1810
09cbfeaf 1811 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1812 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1813 if (valid_sum_count <= sum_in_page)
351df4b2 1814 return 1;
9a47938b 1815 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1816 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1817 return 2;
1818 return 3;
1819}
1820
0a8165d7 1821/*
351df4b2
JK		 1822 * Caller should put this summary page
1823 */
1824struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1825{
1826 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1827}
1828
381722d2 1829void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1830{
1831 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1832 void *dst = page_address(page);
1833
1834 if (src)
09cbfeaf 1835 memcpy(dst, src, PAGE_SIZE);
381722d2 1836 else
09cbfeaf 1837 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1838 set_page_dirty(page);
1839 f2fs_put_page(page, 1);
1840}
1841
381722d2
CY		 1842static void write_sum_page(struct f2fs_sb_info *sbi,
1843 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1844{
1845 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1846}
1847
b7ad7512
CY		 1848static void write_current_sum_page(struct f2fs_sb_info *sbi,
1849 int type, block_t blk_addr)
1850{
1851 struct curseg_info *curseg = CURSEG_I(sbi, type);
1852 struct page *page = grab_meta_page(sbi, blk_addr);
1853 struct f2fs_summary_block *src = curseg->sum_blk;
1854 struct f2fs_summary_block *dst;
1855
1856 dst = (struct f2fs_summary_block *)page_address(page);
1857
1858 mutex_lock(&curseg->curseg_mutex);
1859
1860 down_read(&curseg->journal_rwsem);
1861 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1862 up_read(&curseg->journal_rwsem);
1863
1864 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1865 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1866
1867 mutex_unlock(&curseg->curseg_mutex);
1868
1869 set_page_dirty(page);
1870 f2fs_put_page(page, 1);
1871}
1872
a7881893
JK		 1873static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1874{
1875 struct curseg_info *curseg = CURSEG_I(sbi, type);
1876 unsigned int segno = curseg->segno + 1;
1877 struct free_segmap_info *free_i = FREE_I(sbi);
1878
1879 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1880 return !test_bit(segno, free_i->free_segmap);
1881 return 0;
1882}
1883
0a8165d7 1884/*
351df4b2
JK		 1885 * Find a new segment from the free segments bitmap to right order
1886 * This function should be returned with success, otherwise BUG
1887 */
1888static void get_new_segment(struct f2fs_sb_info *sbi,
1889 unsigned int *newseg, bool new_sec, int dir)
1890{
1891 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1892 unsigned int segno, secno, zoneno;
7cd8558b 1893 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 1894 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1895 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 1896 unsigned int left_start = hint;
1897 bool init = true;
1898 int go_left = 0;
1899 int i;
1900
1a118ccf 1901 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1902
1903 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1904 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 1905 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
1906 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 1907 goto got_it;
1908 }
1909find_other_zone:
7cd8558b
JK		 1910 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1911 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1912 if (dir == ALLOC_RIGHT) {
1913 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1914 MAIN_SECS(sbi), 0);
1915 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1916 } else {
1917 go_left = 1;
1918 left_start = hint - 1;
1919 }
1920 }
1921 if (go_left == 0)
1922 goto skip_left;
1923
1924 while (test_bit(left_start, free_i->free_secmap)) {
1925 if (left_start > 0) {
1926 left_start--;
1927 continue;
1928 }
1929 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1930 MAIN_SECS(sbi), 0);
1931 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1932 break;
1933 }
1934 secno = left_start;
1935skip_left:
1936 hint = secno;
4ddb1a4d
JK		 1937 segno = GET_SEG_FROM_SEC(sbi, secno);
1938 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 1939
1940 /* give up on finding another zone */
1941 if (!init)
1942 goto got_it;
1943 if (sbi->secs_per_zone == 1)
1944 goto got_it;
1945 if (zoneno == old_zoneno)
1946 goto got_it;
1947 if (dir == ALLOC_LEFT) {
1948 if (!go_left && zoneno + 1 >= total_zones)
1949 goto got_it;
1950 if (go_left && zoneno == 0)
1951 goto got_it;
1952 }
1953 for (i = 0; i < NR_CURSEG_TYPE; i++)
1954 if (CURSEG_I(sbi, i)->zone == zoneno)
1955 break;
1956
1957 if (i < NR_CURSEG_TYPE) {
1958 /* zone is in user, try another */
1959 if (go_left)
1960 hint = zoneno * sbi->secs_per_zone - 1;
1961 else if (zoneno + 1 >= total_zones)
1962 hint = 0;
1963 else
1964 hint = (zoneno + 1) * sbi->secs_per_zone;
1965 init = false;
1966 goto find_other_zone;
1967 }
1968got_it:
1969 /* set it as dirty segment in free segmap */
9850cf4a 1970 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1971 __set_inuse(sbi, segno);
1972 *newseg = segno;
1a118ccf 1973 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1974}
1975
1976static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1977{
1978 struct curseg_info *curseg = CURSEG_I(sbi, type);
1979 struct summary_footer *sum_footer;
1980
1981 curseg->segno = curseg->next_segno;
4ddb1a4d 1982 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 1983 curseg->next_blkoff = 0;
1984 curseg->next_segno = NULL_SEGNO;
1985
1986 sum_footer = &(curseg->sum_blk->footer);
1987 memset(sum_footer, 0, sizeof(struct summary_footer));
1988 if (IS_DATASEG(type))
1989 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1990 if (IS_NODESEG(type))
1991 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1992 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1993}
1994
7a20b8a6
JK		 1995static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
1996{
a7881893
JK		 1997 /* if segs_per_sec is large than 1, we need to keep original policy. */
1998 if (sbi->segs_per_sec != 1)
1999 return CURSEG_I(sbi, type)->segno;
2000
7a20b8a6
JK		 2001 if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
2002 return 0;
2003
e066b83c
JK		 2004 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2005 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
7a20b8a6
JK		 2006 return CURSEG_I(sbi, type)->segno;
2007}
2008
0a8165d7 2009/*
351df4b2
JK		 2010 * Allocate a current working segment.
2011 * This function always allocates a free segment in LFS manner.
2012 */
2013static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2014{
2015 struct curseg_info *curseg = CURSEG_I(sbi, type);
2016 unsigned int segno = curseg->segno;
2017 int dir = ALLOC_LEFT;
2018
2019 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2020 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2021 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2022 dir = ALLOC_RIGHT;
2023
2024 if (test_opt(sbi, NOHEAP))
2025 dir = ALLOC_RIGHT;
2026
7a20b8a6 2027 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2028 get_new_segment(sbi, &segno, new_sec, dir);
2029 curseg->next_segno = segno;
2030 reset_curseg(sbi, type, 1);
2031 curseg->alloc_type = LFS;
2032}
2033
2034static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2035 struct curseg_info *seg, block_t start)
2036{
2037 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2038 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2039 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2040 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2041 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2042 int i, pos;
2043
2044 for (i = 0; i < entries; i++)
2045 target_map[i] = ckpt_map[i] | cur_map[i];
2046
2047 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2048
2049 seg->next_blkoff = pos;
351df4b2
JK		 2050}
2051
0a8165d7 2052/*
351df4b2
JK		 2053 * If a segment is written by LFS manner, next block offset is just obtained
2054 * by increasing the current block offset. However, if a segment is written by
2055 * SSR manner, next block offset obtained by calling __next_free_blkoff
2056 */
2057static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2058 struct curseg_info *seg)
2059{
2060 if (seg->alloc_type == SSR)
2061 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2062 else
2063 seg->next_blkoff++;
2064}
2065
0a8165d7 2066/*
e1c42045 2067 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2068 * manner, so it should recover the existing segment information of valid blocks
2069 */
025d63a4 2070static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2071{
2072 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2073 struct curseg_info *curseg = CURSEG_I(sbi, type);
2074 unsigned int new_segno = curseg->next_segno;
2075 struct f2fs_summary_block *sum_node;
2076 struct page *sum_page;
2077
2078 write_sum_page(sbi, curseg->sum_blk,
2079 GET_SUM_BLOCK(sbi, curseg->segno));
2080 __set_test_and_inuse(sbi, new_segno);
2081
2082 mutex_lock(&dirty_i->seglist_lock);
2083 __remove_dirty_segment(sbi, new_segno, PRE);
2084 __remove_dirty_segment(sbi, new_segno, DIRTY);
2085 mutex_unlock(&dirty_i->seglist_lock);
2086
2087 reset_curseg(sbi, type, 1);
2088 curseg->alloc_type = SSR;
2089 __next_free_blkoff(sbi, curseg, 0);
2090
025d63a4
CY		 2091 sum_page = get_sum_page(sbi, new_segno);
2092 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2093 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2094 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2095}
2096
43727527
JK		 2097static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2098{
2099 struct curseg_info *curseg = CURSEG_I(sbi, type);
2100 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2101 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2102 int i, cnt;
2103 bool reversed = false;
c192f7a4
JK		 2104
2105 /* need_SSR() already forces to do this */
e066b83c
JK		 2106 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2107 curseg->next_segno = segno;
c192f7a4 2108 return 1;
e066b83c 2109 }
43727527 2110
70d625cb
JK		 2111 /* For node segments, let's do SSR more intensively */
2112 if (IS_NODESEG(type)) {
d27c3d89
CY		 2113 if (type >= CURSEG_WARM_NODE) {
2114 reversed = true;
2115 i = CURSEG_COLD_NODE;
2116 } else {
2117 i = CURSEG_HOT_NODE;
2118 }
2119 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2120 } else {
d27c3d89
CY		 2121 if (type >= CURSEG_WARM_DATA) {
2122 reversed = true;
2123 i = CURSEG_COLD_DATA;
2124 } else {
2125 i = CURSEG_HOT_DATA;
2126 }
2127 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2128 }
43727527 2129
d27c3d89 2130 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2131 if (i == type)
2132 continue;
e066b83c
JK		 2133 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2134 curseg->next_segno = segno;
43727527 2135 return 1;
e066b83c 2136 }
c192f7a4 2137 }
43727527
JK		 2138 return 0;
2139}
2140
351df4b2
JK		 2141/*
2142 * flush out current segment and replace it with new segment
2143 * This function should be returned with success, otherwise BUG
2144 */
2145static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2146 int type, bool force)
2147{
a7881893
JK		 2148 struct curseg_info *curseg = CURSEG_I(sbi, type);
2149
7b405275 2150 if (force)
351df4b2 2151 new_curseg(sbi, type, true);
5b6c6be2
JK		 2152 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2153 type == CURSEG_WARM_NODE)
351df4b2 2154 new_curseg(sbi, type, false);
a7881893
JK		 2155 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2156 new_curseg(sbi, type, false);
351df4b2 2157 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2158 change_curseg(sbi, type);
351df4b2
JK		 2159 else
2160 new_curseg(sbi, type, false);
dcdfff65 2161
a7881893 2162 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2163}
2164
2165void allocate_new_segments(struct f2fs_sb_info *sbi)
2166{
6ae1be13
JK		 2167 struct curseg_info *curseg;
2168 unsigned int old_segno;
351df4b2
JK		 2169 int i;
2170
6ae1be13
JK		 2171 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2172 curseg = CURSEG_I(sbi, i);
2173 old_segno = curseg->segno;
2174 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2175 locate_dirty_segment(sbi, old_segno);
2176 }
351df4b2
JK		 2177}
2178
2179static const struct segment_allocation default_salloc_ops = {
2180 .allocate_segment = allocate_segment_by_default,
2181};
2182
25290fa5
JK		 2183bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2184{
2185 __u64 trim_start = cpc->trim_start;
2186 bool has_candidate = false;
2187
2188 mutex_lock(&SIT_I(sbi)->sentry_lock);
2189 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2190 if (add_discard_addrs(sbi, cpc, true)) {
2191 has_candidate = true;
2192 break;
2193 }
2194 }
2195 mutex_unlock(&SIT_I(sbi)->sentry_lock);
2196
2197 cpc->trim_start = trim_start;
2198 return has_candidate;
2199}
2200
4b2fecc8
JK		 2201int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2202{
f7ef9b83
JK		 2203 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2204 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 2205 unsigned int start_segno, end_segno;
2206 struct cp_control cpc;
c34f42e2 2207 int err = 0;
4b2fecc8 2208
836b5a63 2209 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2210 return -EINVAL;
2211
9bd27ae4 2212 cpc.trimmed = 0;
7cd8558b 2213 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 2214 goto out;
2215
ed214a11
YH		 2216 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2217 f2fs_msg(sbi->sb, KERN_WARNING,
2218 "Found FS corruption, run fsck to fix.");
2219 goto out;
2220 }
2221
4b2fecc8 2222 /* start/end segment number in main_area */
7cd8558b
JK		 2223 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2224 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2225 GET_SEGNO(sbi, end);
4b2fecc8 2226 cpc.reason = CP_DISCARD;
836b5a63 2227 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 2228
2229 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 2230 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
2231 cpc.trim_start = start_segno;
a66cdd98
JK		 2232
2233 if (sbi->discard_blks == 0)
2234 break;
2235 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2236 cpc.trim_end = end_segno;
2237 else
2238 cpc.trim_end = min_t(unsigned int,
2239 rounddown(start_segno +
bba681cb
JK		 2240 BATCHED_TRIM_SEGMENTS(sbi),
2241 sbi->segs_per_sec) - 1, end_segno);
2242
2243 mutex_lock(&sbi->gc_mutex);
c34f42e2 2244 err = write_checkpoint(sbi, &cpc);
bba681cb 2245 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 2246 if (err)
2247 break;
74fa5f3d
CY		 2248
2249 schedule();
bba681cb 2250 }
969d1b18
CY		 2251 /* It's time to issue all the filed discards */
2252 mark_discard_range_all(sbi);
638164a2 2253 f2fs_wait_discard_bios(sbi, false);
4b2fecc8 2254out:
f7ef9b83 2255 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 2256 return err;
4b2fecc8
JK		 2257}
2258
351df4b2
JK		 2259static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2260{
2261 struct curseg_info *curseg = CURSEG_I(sbi, type);
2262 if (curseg->next_blkoff < sbi->blocks_per_seg)
2263 return true;
2264 return false;
2265}
2266
81377bd6 2267static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2268{
81377bd6 2269 if (fio->type == DATA)
351df4b2
JK		 2270 return CURSEG_HOT_DATA;
2271 else
2272 return CURSEG_HOT_NODE;
2273}
2274
81377bd6 2275static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2276{
81377bd6
JK		 2277 if (fio->type == DATA) {
2278 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2279
2280 if (S_ISDIR(inode->i_mode))
2281 return CURSEG_HOT_DATA;
2282 else
2283 return CURSEG_COLD_DATA;
2284 } else {
81377bd6 2285 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2286 return CURSEG_WARM_NODE;
351df4b2
JK		 2287 else
2288 return CURSEG_COLD_NODE;
2289 }
2290}
2291
81377bd6 2292static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2293{
81377bd6
JK		 2294 if (fio->type == DATA) {
2295 struct inode *inode = fio->page->mapping->host;
351df4b2 2296
81377bd6 2297 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2298 return CURSEG_COLD_DATA;
ef095d19
JK		 2299 if (is_inode_flag_set(inode, FI_HOT_DATA))
2300 return CURSEG_HOT_DATA;
2301 return CURSEG_WARM_DATA;
351df4b2 2302 } else {
81377bd6
JK		 2303 if (IS_DNODE(fio->page))
2304 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2305 CURSEG_HOT_NODE;
ef095d19 2306 return CURSEG_COLD_NODE;
351df4b2
JK		 2307 }
2308}
2309
81377bd6 2310static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2311{
a912b54d
JK		 2312 int type = 0;
2313
81377bd6 2314 switch (fio->sbi->active_logs) {
351df4b2 2315 case 2:
a912b54d
JK		 2316 type = __get_segment_type_2(fio);
2317 break;
351df4b2 2318 case 4:
a912b54d
JK		 2319 type = __get_segment_type_4(fio);
2320 break;
2321 case 6:
2322 type = __get_segment_type_6(fio);
2323 break;
2324 default:
2325 f2fs_bug_on(fio->sbi, true);
351df4b2 2326 }
81377bd6 2327
a912b54d
JK		 2328 if (IS_HOT(type))
2329 fio->temp = HOT;
2330 else if (IS_WARM(type))
2331 fio->temp = WARM;
2332 else
2333 fio->temp = COLD;
2334 return type;
351df4b2
JK		 2335}
2336
bfad7c2d
JK		 2337void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2338 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2339 struct f2fs_summary *sum, int type,
2340 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2341{
2342 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2343 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 2344
2345 mutex_lock(&curseg->curseg_mutex);
21cb1d99 2346 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK		 2347
2348 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2349
4e6a8d9b
JK		 2350 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2351
351df4b2
JK		 2352 /*
2353 * __add_sum_entry should be resided under the curseg_mutex
2354 * because, this function updates a summary entry in the
2355 * current summary block.
2356 */
e79efe3b 2357 __add_sum_entry(sbi, type, sum);
351df4b2 2358
351df4b2 2359 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2360
2361 stat_inc_block_count(sbi, curseg);
351df4b2 2362
c6f82fe9
JK		 2363 if (!__has_curseg_space(sbi, type))
2364 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2 2365 /*
c6f82fe9
JK		 2366 * SIT information should be updated after segment allocation,
2367 * since we need to keep dirty segments precisely under SSR.
351df4b2
JK		 2368 */
2369 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 2370
351df4b2
JK		 2371 mutex_unlock(&sit_i->sentry_lock);
2372
704956ec 2373 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2374 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2375
704956ec
CY		 2376 f2fs_inode_chksum_set(sbi, page);
2377 }
2378
fb830fc5
CY		 2379 if (add_list) {
2380 struct f2fs_bio_info *io;
2381
2382 INIT_LIST_HEAD(&fio->list);
2383 fio->in_list = true;
2384 io = sbi->write_io[fio->type] + fio->temp;
2385 spin_lock(&io->io_lock);
2386 list_add_tail(&fio->list, &io->io_list);
2387 spin_unlock(&io->io_lock);
2388 }
2389
bfad7c2d
JK		 2390 mutex_unlock(&curseg->curseg_mutex);
2391}
2392
39d787be
CY		 2393static void update_device_state(struct f2fs_io_info *fio)
2394{
2395 struct f2fs_sb_info *sbi = fio->sbi;
2396 unsigned int devidx;
2397
2398 if (!sbi->s_ndevs)
2399 return;
2400
2401 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2402
2403 /* update device state for fsync */
2404 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2405}
2406
05ca3632 2407static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2408{
81377bd6 2409 int type = __get_segment_type(fio);
0a595eba 2410 int err;
bfad7c2d 2411
0a595eba 2412reallocate:
7a9d7548 2413 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2414 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2415
351df4b2 2416 /* writeout dirty page into bdev */
b9109b0e 2417 err = f2fs_submit_page_write(fio);
0a595eba
JK		 2418 if (err == -EAGAIN) {
2419 fio->old_blkaddr = fio->new_blkaddr;
2420 goto reallocate;
39d787be
CY		 2421 } else if (!err) {
2422 update_device_state(fio);
0a595eba 2423 }
351df4b2
JK		 2424}
2425
b0af6d49
CY		 2426void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2427 enum iostat_type io_type)
351df4b2 2428{
458e6197 2429 struct f2fs_io_info fio = {
05ca3632 2430 .sbi = sbi,
458e6197 2431 .type = META,
04d328de 2432 .op = REQ_OP_WRITE,
70fd7614 2433 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2434 .old_blkaddr = page->index,
2435 .new_blkaddr = page->index,
05ca3632 2436 .page = page,
4375a336 2437 .encrypted_page = NULL,
fb830fc5 2438 .in_list = false,
458e6197
JK		 2439 };
2440
2b947003 2441 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2442 fio.op_flags &= ~REQ_META;
2b947003 2443
351df4b2 2444 set_page_writeback(page);
b9109b0e 2445 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2446
2447 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2448}
2449
05ca3632 2450void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2451{
2452 struct f2fs_summary sum;
05ca3632 2453
351df4b2 2454 set_summary(&sum, nid, 0, 0);
05ca3632 2455 do_write_page(&sum, fio);
b0af6d49
CY		 2456
2457 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2458}
2459
05ca3632 2460void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2461{
05ca3632 2462 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2463 struct f2fs_summary sum;
2464 struct node_info ni;
2465
9850cf4a 2466 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2467 get_node_info(sbi, dn->nid, &ni);
2468 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2469 do_write_page(&sum, fio);
f28b3434 2470 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2471
2472 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2473}
2474
d1b3e72d 2475int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2476{
b0af6d49
CY		 2477 int err;
2478
7a9d7548 2479 fio->new_blkaddr = fio->old_blkaddr;
05ca3632 2480 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2481
2482 err = f2fs_submit_page_bio(fio);
39d787be
CY		 2483 if (!err)
2484 update_device_state(fio);
b0af6d49
CY		 2485
2486 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2487
2488 return err;
351df4b2
JK		 2489}
2490
4356e48e 2491void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2492 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2493 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2494{
2495 struct sit_info *sit_i = SIT_I(sbi);
2496 struct curseg_info *curseg;
2497 unsigned int segno, old_cursegno;
2498 struct seg_entry *se;
2499 int type;
19f106bc 2500 unsigned short old_blkoff;
351df4b2
JK		 2501
2502 segno = GET_SEGNO(sbi, new_blkaddr);
2503 se = get_seg_entry(sbi, segno);
2504 type = se->type;
2505
19f106bc
CY		 2506 if (!recover_curseg) {
2507 /* for recovery flow */
2508 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2509 if (old_blkaddr == NULL_ADDR)
2510 type = CURSEG_COLD_DATA;
2511 else
2512 type = CURSEG_WARM_DATA;
2513 }
2514 } else {
2515 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 2516 type = CURSEG_WARM_DATA;
2517 }
19f106bc 2518
351df4b2
JK		 2519 curseg = CURSEG_I(sbi, type);
2520
2521 mutex_lock(&curseg->curseg_mutex);
2522 mutex_lock(&sit_i->sentry_lock);
2523
2524 old_cursegno = curseg->segno;
19f106bc 2525 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2526
2527 /* change the current segment */
2528 if (segno != curseg->segno) {
2529 curseg->next_segno = segno;
025d63a4 2530 change_curseg(sbi, type);
351df4b2
JK		 2531 }
2532
491c0854 2533 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2534 __add_sum_entry(sbi, type, sum);
351df4b2 2535
28bc106b 2536 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2537 update_sit_entry(sbi, new_blkaddr, 1);
2538 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2539 update_sit_entry(sbi, old_blkaddr, -1);
2540
2541 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2542 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2543
351df4b2 2544 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2545
19f106bc
CY		 2546 if (recover_curseg) {
2547 if (old_cursegno != curseg->segno) {
2548 curseg->next_segno = old_cursegno;
025d63a4 2549 change_curseg(sbi, type);
19f106bc
CY		 2550 }
2551 curseg->next_blkoff = old_blkoff;
2552 }
2553
351df4b2
JK		 2554 mutex_unlock(&sit_i->sentry_lock);
2555 mutex_unlock(&curseg->curseg_mutex);
2556}
2557
528e3459
CY		 2558void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2559 block_t old_addr, block_t new_addr,
28bc106b
CY		 2560 unsigned char version, bool recover_curseg,
2561 bool recover_newaddr)
528e3459
CY		 2562{
2563 struct f2fs_summary sum;
2564
2565 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2566
28bc106b
CY		 2567 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2568 recover_curseg, recover_newaddr);
528e3459 2569
f28b3434 2570 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2571}
2572
93dfe2ac 2573void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2574 enum page_type type, bool ordered)
93dfe2ac 2575{
93dfe2ac 2576 if (PageWriteback(page)) {
4081363f
JK		 2577 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2578
b9109b0e
JK		 2579 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2580 0, page->index, type);
fec1d657
JK		 2581 if (ordered)
2582 wait_on_page_writeback(page);
2583 else
2584 wait_for_stable_page(page);
93dfe2ac
JK		 2585 }
2586}
2587
d4c759ee 2588void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 2589{
2590 struct page *cpage;
2591
5d4c0af4 2592 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2593 return;
2594
08b39fbd
CY		 2595 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2596 if (cpage) {
fec1d657 2597 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2598 f2fs_put_page(cpage, 1);
2599 }
2600}
2601
351df4b2
JK		 2602static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2603{
2604 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2605 struct curseg_info *seg_i;
2606 unsigned char *kaddr;
2607 struct page *page;
2608 block_t start;
2609 int i, j, offset;
2610
2611 start = start_sum_block(sbi);
2612
2613 page = get_meta_page(sbi, start++);
2614 kaddr = (unsigned char *)page_address(page);
2615
2616 /* Step 1: restore nat cache */
2617 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2618 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2619
2620 /* Step 2: restore sit cache */
2621 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2622 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2623 offset = 2 * SUM_JOURNAL_SIZE;
2624
2625 /* Step 3: restore summary entries */
2626 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2627 unsigned short blk_off;
2628 unsigned int segno;
2629
2630 seg_i = CURSEG_I(sbi, i);
2631 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2632 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2633 seg_i->next_segno = segno;
2634 reset_curseg(sbi, i, 0);
2635 seg_i->alloc_type = ckpt->alloc_type[i];
2636 seg_i->next_blkoff = blk_off;
2637
2638 if (seg_i->alloc_type == SSR)
2639 blk_off = sbi->blocks_per_seg;
2640
2641 for (j = 0; j < blk_off; j++) {
2642 struct f2fs_summary *s;
2643 s = (struct f2fs_summary *)(kaddr + offset);
2644 seg_i->sum_blk->entries[j] = *s;
2645 offset += SUMMARY_SIZE;
09cbfeaf 2646 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2647 SUM_FOOTER_SIZE)
2648 continue;
2649
2650 f2fs_put_page(page, 1);
2651 page = NULL;
2652
2653 page = get_meta_page(sbi, start++);
2654 kaddr = (unsigned char *)page_address(page);
2655 offset = 0;
2656 }
2657 }
2658 f2fs_put_page(page, 1);
2659 return 0;
2660}
2661
2662static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2663{
2664 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2665 struct f2fs_summary_block *sum;
2666 struct curseg_info *curseg;
2667 struct page *new;
2668 unsigned short blk_off;
2669 unsigned int segno = 0;
2670 block_t blk_addr = 0;
2671
2672 /* get segment number and block addr */
2673 if (IS_DATASEG(type)) {
2674 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2675 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2676 CURSEG_HOT_DATA]);
119ee914 2677 if (__exist_node_summaries(sbi))
351df4b2
JK		 2678 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2679 else
2680 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2681 } else {
2682 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2683 CURSEG_HOT_NODE]);
2684 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2685 CURSEG_HOT_NODE]);
119ee914 2686 if (__exist_node_summaries(sbi))
351df4b2
JK		 2687 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2688 type - CURSEG_HOT_NODE);
2689 else
2690 blk_addr = GET_SUM_BLOCK(sbi, segno);
2691 }
2692
2693 new = get_meta_page(sbi, blk_addr);
2694 sum = (struct f2fs_summary_block *)page_address(new);
2695
2696 if (IS_NODESEG(type)) {
119ee914 2697 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2698 struct f2fs_summary *ns = &sum->entries[0];
2699 int i;
2700 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2701 ns->version = 0;
2702 ns->ofs_in_node = 0;
2703 }
2704 } else {
d653788a
GZ		 2705 int err;
2706
2707 err = restore_node_summary(sbi, segno, sum);
2708 if (err) {
351df4b2 2709 f2fs_put_page(new, 1);
d653788a 2710 return err;
351df4b2
JK		 2711 }
2712 }
2713 }
2714
2715 /* set uncompleted segment to curseg */
2716 curseg = CURSEG_I(sbi, type);
2717 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2718
2719 /* update journal info */
2720 down_write(&curseg->journal_rwsem);
2721 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2722 up_write(&curseg->journal_rwsem);
2723
2724 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2725 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2726 curseg->next_segno = segno;
2727 reset_curseg(sbi, type, 0);
2728 curseg->alloc_type = ckpt->alloc_type[type];
2729 curseg->next_blkoff = blk_off;
2730 mutex_unlock(&curseg->curseg_mutex);
2731 f2fs_put_page(new, 1);
2732 return 0;
2733}
2734
2735static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2736{
21d3f8e1
JQ		 2737 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2738 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 2739 int type = CURSEG_HOT_DATA;
e4fc5fbf 2740 int err;
351df4b2 2741
aaec2b1d 2742 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2743 int npages = npages_for_summary_flush(sbi, true);
2744
2745 if (npages >= 2)
2746 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2747 META_CP, true);
3fa06d7b 2748
351df4b2
JK		 2749 /* restore for compacted data summary */
2750 if (read_compacted_summaries(sbi))
2751 return -EINVAL;
2752 type = CURSEG_HOT_NODE;
2753 }
2754
119ee914 2755 if (__exist_node_summaries(sbi))
3fa06d7b 2756 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2757 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2758
e4fc5fbf
CY		 2759 for (; type <= CURSEG_COLD_NODE; type++) {
2760 err = read_normal_summaries(sbi, type);
2761 if (err)
2762 return err;
2763 }
2764
21d3f8e1
JQ		 2765 /* sanity check for summary blocks */
2766 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2767 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2768 return -EINVAL;
2769
351df4b2
JK		 2770 return 0;
2771}
2772
2773static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2774{
2775 struct page *page;
2776 unsigned char *kaddr;
2777 struct f2fs_summary *summary;
2778 struct curseg_info *seg_i;
2779 int written_size = 0;
2780 int i, j;
2781
2782 page = grab_meta_page(sbi, blkaddr++);
2783 kaddr = (unsigned char *)page_address(page);
2784
2785 /* Step 1: write nat cache */
2786 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2787 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2788 written_size += SUM_JOURNAL_SIZE;
2789
2790 /* Step 2: write sit cache */
2791 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2792 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2793 written_size += SUM_JOURNAL_SIZE;
2794
351df4b2
JK		 2795 /* Step 3: write summary entries */
2796 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2797 unsigned short blkoff;
2798 seg_i = CURSEG_I(sbi, i);
2799 if (sbi->ckpt->alloc_type[i] == SSR)
2800 blkoff = sbi->blocks_per_seg;
2801 else
2802 blkoff = curseg_blkoff(sbi, i);
2803
2804 for (j = 0; j < blkoff; j++) {
2805 if (!page) {
2806 page = grab_meta_page(sbi, blkaddr++);
2807 kaddr = (unsigned char *)page_address(page);
2808 written_size = 0;
2809 }
2810 summary = (struct f2fs_summary *)(kaddr + written_size);
2811 *summary = seg_i->sum_blk->entries[j];
2812 written_size += SUMMARY_SIZE;
351df4b2 2813
09cbfeaf 2814 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2815 SUM_FOOTER_SIZE)
2816 continue;
2817
e8d61a74 2818 set_page_dirty(page);
351df4b2
JK		 2819 f2fs_put_page(page, 1);
2820 page = NULL;
2821 }
2822 }
e8d61a74
CY		 2823 if (page) {
2824 set_page_dirty(page);
351df4b2 2825 f2fs_put_page(page, 1);
e8d61a74 2826 }
351df4b2
JK		 2827}
2828
2829static void write_normal_summaries(struct f2fs_sb_info *sbi,
2830 block_t blkaddr, int type)
2831{
2832 int i, end;
2833 if (IS_DATASEG(type))
2834 end = type + NR_CURSEG_DATA_TYPE;
2835 else
2836 end = type + NR_CURSEG_NODE_TYPE;
2837
b7ad7512
CY		 2838 for (i = type; i < end; i++)
2839 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 2840}
2841
2842void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2843{
aaec2b1d 2844 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 2845 write_compacted_summaries(sbi, start_blk);
2846 else
2847 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2848}
2849
2850void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2851{
119ee914 2852 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 2853}
2854
dfc08a12 2855int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 2856 unsigned int val, int alloc)
2857{
2858 int i;
2859
2860 if (type == NAT_JOURNAL) {
dfc08a12
CY		 2861 for (i = 0; i < nats_in_cursum(journal); i++) {
2862 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 2863 return i;
2864 }
dfc08a12
CY		 2865 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2866 return update_nats_in_cursum(journal, 1);
351df4b2 2867 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 2868 for (i = 0; i < sits_in_cursum(journal); i++)
2869 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 2870 return i;
dfc08a12
CY		 2871 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2872 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 2873 }
2874 return -1;
2875}
2876
2877static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2878 unsigned int segno)
2879{
2cc22186 2880 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 2881}
2882
2883static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2884 unsigned int start)
2885{
2886 struct sit_info *sit_i = SIT_I(sbi);
2887 struct page *src_page, *dst_page;
2888 pgoff_t src_off, dst_off;
2889 void *src_addr, *dst_addr;
2890
2891 src_off = current_sit_addr(sbi, start);
2892 dst_off = next_sit_addr(sbi, src_off);
2893
2894 /* get current sit block page without lock */
2895 src_page = get_meta_page(sbi, src_off);
2896 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 2897 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 2898
2899 src_addr = page_address(src_page);
2900 dst_addr = page_address(dst_page);
09cbfeaf 2901 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 2902
2903 set_page_dirty(dst_page);
2904 f2fs_put_page(src_page, 1);
2905
2906 set_to_next_sit(sit_i, start);
2907
2908 return dst_page;
2909}
2910
184a5cd2
CY		 2911static struct sit_entry_set *grab_sit_entry_set(void)
2912{
2913 struct sit_entry_set *ses =
80c54505 2914 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 2915
2916 ses->entry_cnt = 0;
2917 INIT_LIST_HEAD(&ses->set_list);
2918 return ses;
2919}
2920
2921static void release_sit_entry_set(struct sit_entry_set *ses)
2922{
2923 list_del(&ses->set_list);
2924 kmem_cache_free(sit_entry_set_slab, ses);
2925}
2926
2927static void adjust_sit_entry_set(struct sit_entry_set *ses,
2928 struct list_head *head)
2929{
2930 struct sit_entry_set *next = ses;
2931
2932 if (list_is_last(&ses->set_list, head))
2933 return;
2934
2935 list_for_each_entry_continue(next, head, set_list)
2936 if (ses->entry_cnt <= next->entry_cnt)
2937 break;
2938
2939 list_move_tail(&ses->set_list, &next->set_list);
2940}
2941
2942static void add_sit_entry(unsigned int segno, struct list_head *head)
2943{
2944 struct sit_entry_set *ses;
2945 unsigned int start_segno = START_SEGNO(segno);
2946
2947 list_for_each_entry(ses, head, set_list) {
2948 if (ses->start_segno == start_segno) {
2949 ses->entry_cnt++;
2950 adjust_sit_entry_set(ses, head);
2951 return;
2952 }
2953 }
2954
2955 ses = grab_sit_entry_set();
2956
2957 ses->start_segno = start_segno;
2958 ses->entry_cnt++;
2959 list_add(&ses->set_list, head);
2960}
2961
2962static void add_sits_in_set(struct f2fs_sb_info *sbi)
2963{
2964 struct f2fs_sm_info *sm_info = SM_I(sbi);
2965 struct list_head *set_list = &sm_info->sit_entry_set;
2966 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 2967 unsigned int segno;
2968
7cd8558b 2969 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 2970 add_sit_entry(segno, set_list);
2971}
2972
2973static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 2974{
2975 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2976 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 2977 int i;
2978
b7ad7512 2979 down_write(&curseg->journal_rwsem);
dfc08a12 2980 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 2981 unsigned int segno;
2982 bool dirtied;
2983
dfc08a12 2984 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 2985 dirtied = __mark_sit_entry_dirty(sbi, segno);
2986
2987 if (!dirtied)
2988 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 2989 }
dfc08a12 2990 update_sits_in_cursum(journal, -i);
b7ad7512 2991 up_write(&curseg->journal_rwsem);
351df4b2
JK		 2992}
2993
0a8165d7 2994/*
351df4b2
JK		 2995 * CP calls this function, which flushes SIT entries including sit_journal,
2996 * and moves prefree segs to free segs.
2997 */
4b2fecc8 2998void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 2999{
3000 struct sit_info *sit_i = SIT_I(sbi);
3001 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3002 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3003 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3004 struct sit_entry_set *ses, *tmp;
3005 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3006 bool to_journal = true;
4b2fecc8 3007 struct seg_entry *se;
351df4b2 3008
351df4b2
JK		 3009 mutex_lock(&sit_i->sentry_lock);
3010
2b11a74b
WL		 3011 if (!sit_i->dirty_sentries)
3012 goto out;
3013
351df4b2 3014 /*
184a5cd2
CY		 3015 * add and account sit entries of dirty bitmap in sit entry
3016 * set temporarily
351df4b2 3017 */
184a5cd2 3018 add_sits_in_set(sbi);
351df4b2 3019
184a5cd2
CY		 3020 /*
3021 * if there are no enough space in journal to store dirty sit
3022 * entries, remove all entries from journal and add and account
3023 * them in sit entry set.
3024 */
dfc08a12 3025 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3026 remove_sits_in_journal(sbi);
b2955550 3027
184a5cd2
CY		 3028 /*
3029 * there are two steps to flush sit entries:
3030 * #1, flush sit entries to journal in current cold data summary block.
3031 * #2, flush sit entries to sit page.
3032 */
3033 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3034 struct page *page = NULL;
184a5cd2
CY		 3035 struct f2fs_sit_block *raw_sit = NULL;
3036 unsigned int start_segno = ses->start_segno;
3037 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3038 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3039 unsigned int segno = start_segno;
3040
3041 if (to_journal &&
dfc08a12 3042 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3043 to_journal = false;
3044
b7ad7512
CY		 3045 if (to_journal) {
3046 down_write(&curseg->journal_rwsem);
3047 } else {
184a5cd2
CY		 3048 page = get_next_sit_page(sbi, start_segno);
3049 raw_sit = page_address(page);
351df4b2 3050 }
351df4b2 3051
184a5cd2
CY		 3052 /* flush dirty sit entries in region of current sit set */
3053 for_each_set_bit_from(segno, bitmap, end) {
3054 int offset, sit_offset;
4b2fecc8
JK		 3055
3056 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 3057
3058 /* add discard candidates */
c473f1a9 3059 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3060 cpc->trim_start = segno;
25290fa5 3061 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3062 }
184a5cd2
CY		 3063
3064 if (to_journal) {
dfc08a12 3065 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 3066 SIT_JOURNAL, segno, 1);
3067 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3068 segno_in_journal(journal, offset) =
184a5cd2
CY		 3069 cpu_to_le32(segno);
3070 seg_info_to_raw_sit(se,
dfc08a12 3071 &sit_in_journal(journal, offset));
184a5cd2
CY		 3072 } else {
3073 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3074 seg_info_to_raw_sit(se,
3075 &raw_sit->entries[sit_offset]);
3076 }
351df4b2 3077
184a5cd2
CY		 3078 __clear_bit(segno, bitmap);
3079 sit_i->dirty_sentries--;
3080 ses->entry_cnt--;
351df4b2
JK		 3081 }
3082
b7ad7512
CY		 3083 if (to_journal)
3084 up_write(&curseg->journal_rwsem);
3085 else
184a5cd2
CY		 3086 f2fs_put_page(page, 1);
3087
3088 f2fs_bug_on(sbi, ses->entry_cnt);
3089 release_sit_entry_set(ses);
351df4b2 3090 }
184a5cd2
CY		 3091
3092 f2fs_bug_on(sbi, !list_empty(head));
3093 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3094out:
c473f1a9 3095 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3096 __u64 trim_start = cpc->trim_start;
3097
4b2fecc8 3098 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3099 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3100
3101 cpc->trim_start = trim_start;
4b2fecc8 3102 }
351df4b2 3103 mutex_unlock(&sit_i->sentry_lock);
351df4b2 3104
351df4b2
JK		 3105 set_prefree_as_free_segments(sbi);
3106}
3107
3108static int build_sit_info(struct f2fs_sb_info *sbi)
3109{
3110 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3111 struct sit_info *sit_i;
3112 unsigned int sit_segs, start;
ae27d62e 3113 char *src_bitmap;
351df4b2
JK		 3114 unsigned int bitmap_size;
3115
3116 /* allocate memory for SIT information */
3117 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3118 if (!sit_i)
3119 return -ENOMEM;
3120
3121 SM_I(sbi)->sit_info = sit_i;
3122
a7c3e901 3123 sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
39307a8e 3124 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3125 if (!sit_i->sentries)
3126 return -ENOMEM;
3127
7cd8558b 3128 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3129 sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3130 if (!sit_i->dirty_sentries_bitmap)
3131 return -ENOMEM;
3132
7cd8558b 3133 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3134 sit_i->sentries[start].cur_valid_map
3135 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3136 sit_i->sentries[start].ckpt_valid_map
3137 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3138 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3139 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3140 return -ENOMEM;
3e025740 3141
355e7891
CY		 3142#ifdef CONFIG_F2FS_CHECK_FS
3143 sit_i->sentries[start].cur_valid_map_mir
3144 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3145 if (!sit_i->sentries[start].cur_valid_map_mir)
3146 return -ENOMEM;
3147#endif
3148
3e025740
JK		 3149 if (f2fs_discard_en(sbi)) {
3150 sit_i->sentries[start].discard_map
3151 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3152 if (!sit_i->sentries[start].discard_map)
3153 return -ENOMEM;
3154 }
351df4b2
JK		 3155 }
3156
60a3b782
JK		 3157 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3158 if (!sit_i->tmp_map)
3159 return -ENOMEM;
3160
351df4b2 3161 if (sbi->segs_per_sec > 1) {
a7c3e901 3162 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
39307a8e 3163 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3164 if (!sit_i->sec_entries)
3165 return -ENOMEM;
3166 }
3167
3168 /* get information related with SIT */
3169 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3170
3171 /* setup SIT bitmap from ckeckpoint pack */
3172 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3173 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3174
ae27d62e
CY		 3175 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3176 if (!sit_i->sit_bitmap)
351df4b2 3177 return -ENOMEM;
351df4b2 3178
ae27d62e
CY		 3179#ifdef CONFIG_F2FS_CHECK_FS
3180 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3181 if (!sit_i->sit_bitmap_mir)
3182 return -ENOMEM;
3183#endif
3184
351df4b2
JK		 3185 /* init SIT information */
3186 sit_i->s_ops = &default_salloc_ops;
3187
3188 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3189 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3190 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3191 sit_i->bitmap_size = bitmap_size;
3192 sit_i->dirty_sentries = 0;
3193 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3194 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3195 sit_i->mounted_time = ktime_get_real_seconds();
351df4b2
JK		 3196 mutex_init(&sit_i->sentry_lock);
3197 return 0;
3198}
3199
3200static int build_free_segmap(struct f2fs_sb_info *sbi)
3201{
351df4b2
JK		 3202 struct free_segmap_info *free_i;
3203 unsigned int bitmap_size, sec_bitmap_size;
3204
3205 /* allocate memory for free segmap information */
3206 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3207 if (!free_i)
3208 return -ENOMEM;
3209
3210 SM_I(sbi)->free_info = free_i;
3211
7cd8558b 3212 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3213 free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3214 if (!free_i->free_segmap)
3215 return -ENOMEM;
3216
7cd8558b 3217 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
a7c3e901 3218 free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3219 if (!free_i->free_secmap)
3220 return -ENOMEM;
3221
3222 /* set all segments as dirty temporarily */
3223 memset(free_i->free_segmap, 0xff, bitmap_size);
3224 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3225
3226 /* init free segmap information */
7cd8558b 3227 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3228 free_i->free_segments = 0;
3229 free_i->free_sections = 0;
1a118ccf 3230 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3231 return 0;
3232}
3233
3234static int build_curseg(struct f2fs_sb_info *sbi)
3235{
1042d60f 3236 struct curseg_info *array;
351df4b2
JK		 3237 int i;
3238
b434babf 3239 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 3240 if (!array)
3241 return -ENOMEM;
3242
3243 SM_I(sbi)->curseg_array = array;
3244
3245 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3246 mutex_init(&array[i].curseg_mutex);
09cbfeaf 3247 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3248 if (!array[i].sum_blk)
3249 return -ENOMEM;
b7ad7512
CY		 3250 init_rwsem(&array[i].journal_rwsem);
3251 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3252 GFP_KERNEL);
3253 if (!array[i].journal)
3254 return -ENOMEM;
351df4b2
JK		 3255 array[i].segno = NULL_SEGNO;
3256 array[i].next_blkoff = 0;
3257 }
3258 return restore_curseg_summaries(sbi);
3259}
3260
3261static void build_sit_entries(struct f2fs_sb_info *sbi)
3262{
3263 struct sit_info *sit_i = SIT_I(sbi);
3264 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3265 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3266 struct seg_entry *se;
3267 struct f2fs_sit_entry sit;
74de593a
CY		 3268 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3269 unsigned int i, start, end;
3270 unsigned int readed, start_blk = 0;
351df4b2 3271
74de593a 3272 do {
664ba972
JK		 3273 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3274 META_SIT, true);
74de593a
CY		 3275
3276 start = start_blk * sit_i->sents_per_block;
3277 end = (start_blk + readed) * sit_i->sents_per_block;
3278
7cd8558b 3279 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3280 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3281 struct page *page;
3282
9c094040 3283 se = &sit_i->sentries[start];
74de593a
CY		 3284 page = get_current_sit_page(sbi, start);
3285 sit_blk = (struct f2fs_sit_block *)page_address(page);
3286 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3287 f2fs_put_page(page, 1);
d600af23 3288
74de593a
CY		 3289 check_block_count(sbi, start, &sit);
3290 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 3291
3292 /* build discard map only one time */
3e025740 3293 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3294 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3295 memset(se->discard_map, 0xff,
3296 SIT_VBLOCK_MAP_SIZE);
3297 } else {
3298 memcpy(se->discard_map,
3299 se->cur_valid_map,
3300 SIT_VBLOCK_MAP_SIZE);
3301 sbi->discard_blks +=
3302 sbi->blocks_per_seg -
3303 se->valid_blocks;
3304 }
3e025740 3305 }
a66cdd98 3306
d600af23
CY		 3307 if (sbi->segs_per_sec > 1)
3308 get_sec_entry(sbi, start)->valid_blocks +=
3309 se->valid_blocks;
351df4b2 3310 }
74de593a
CY		 3311 start_blk += readed;
3312 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3313
3314 down_read(&curseg->journal_rwsem);
3315 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3316 unsigned int old_valid_blocks;
3317
3318 start = le32_to_cpu(segno_in_journal(journal, i));
3319 se = &sit_i->sentries[start];
3320 sit = sit_in_journal(journal, i);
3321
3322 old_valid_blocks = se->valid_blocks;
3323
3324 check_block_count(sbi, start, &sit);
3325 seg_info_from_raw_sit(se, &sit);
3326
3327 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3328 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3329 memset(se->discard_map, 0xff,
3330 SIT_VBLOCK_MAP_SIZE);
3331 } else {
3332 memcpy(se->discard_map, se->cur_valid_map,
3333 SIT_VBLOCK_MAP_SIZE);
3334 sbi->discard_blks += old_valid_blocks -
3335 se->valid_blocks;
3336 }
d600af23
CY		 3337 }
3338
3339 if (sbi->segs_per_sec > 1)
3340 get_sec_entry(sbi, start)->valid_blocks +=
3341 se->valid_blocks - old_valid_blocks;
3342 }
3343 up_read(&curseg->journal_rwsem);
351df4b2
JK		 3344}
3345
3346static void init_free_segmap(struct f2fs_sb_info *sbi)
3347{
3348 unsigned int start;
3349 int type;
3350
7cd8558b 3351 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3352 struct seg_entry *sentry = get_seg_entry(sbi, start);
3353 if (!sentry->valid_blocks)
3354 __set_free(sbi, start);
c79b7ff1
JK		 3355 else
3356 SIT_I(sbi)->written_valid_blocks +=
3357 sentry->valid_blocks;
351df4b2
JK		 3358 }
3359
3360 /* set use the current segments */
3361 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3362 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3363 __set_test_and_inuse(sbi, curseg_t->segno);
3364 }
3365}
3366
3367static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3368{
3369 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3370 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3371 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3372 unsigned short valid_blocks;
3373
8736fbf0 3374 while (1) {
351df4b2 3375 /* find dirty segment based on free segmap */
7cd8558b
JK		 3376 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3377 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3378 break;
3379 offset = segno + 1;
302bd348 3380 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3381 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3382 continue;
ec325b52
JK		 3383 if (valid_blocks > sbi->blocks_per_seg) {
3384 f2fs_bug_on(sbi, 1);
3385 continue;
3386 }
351df4b2
JK		 3387 mutex_lock(&dirty_i->seglist_lock);
3388 __locate_dirty_segment(sbi, segno, DIRTY);
3389 mutex_unlock(&dirty_i->seglist_lock);
3390 }
3391}
3392
5ec4e49f 3393static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3394{
3395 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3396 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3397
a7c3e901 3398 dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 3399 if (!dirty_i->victim_secmap)
351df4b2
JK		 3400 return -ENOMEM;
3401 return 0;
3402}
3403
3404static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3405{
3406 struct dirty_seglist_info *dirty_i;
3407 unsigned int bitmap_size, i;
3408
3409 /* allocate memory for dirty segments list information */
3410 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3411 if (!dirty_i)
3412 return -ENOMEM;
3413
3414 SM_I(sbi)->dirty_info = dirty_i;
3415 mutex_init(&dirty_i->seglist_lock);
3416
7cd8558b 3417 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3418
3419 for (i = 0; i < NR_DIRTY_TYPE; i++) {
a7c3e901 3420 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3421 if (!dirty_i->dirty_segmap[i])
3422 return -ENOMEM;
3423 }
3424
3425 init_dirty_segmap(sbi);
5ec4e49f 3426 return init_victim_secmap(sbi);
351df4b2
JK		 3427}
3428
0a8165d7 3429/*
351df4b2
JK		 3430 * Update min, max modified time for cost-benefit GC algorithm
3431 */
3432static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3433{
3434 struct sit_info *sit_i = SIT_I(sbi);
3435 unsigned int segno;
3436
3437 mutex_lock(&sit_i->sentry_lock);
3438
3439 sit_i->min_mtime = LLONG_MAX;
3440
7cd8558b 3441 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3442 unsigned int i;
3443 unsigned long long mtime = 0;
3444
3445 for (i = 0; i < sbi->segs_per_sec; i++)
3446 mtime += get_seg_entry(sbi, segno + i)->mtime;
3447
3448 mtime = div_u64(mtime, sbi->segs_per_sec);
3449
3450 if (sit_i->min_mtime > mtime)
3451 sit_i->min_mtime = mtime;
3452 }
3453 sit_i->max_mtime = get_mtime(sbi);
3454 mutex_unlock(&sit_i->sentry_lock);
3455}
3456
3457int build_segment_manager(struct f2fs_sb_info *sbi)
3458{
3459 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3460 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3461 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3462 int err;
3463
3464 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3465 if (!sm_info)
3466 return -ENOMEM;
3467
3468 /* init sm info */
3469 sbi->sm_info = sm_info;
351df4b2
JK		 3470 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3471 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3472 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3473 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3474 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3475 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3476 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3477 sm_info->rec_prefree_segments = sm_info->main_segments *
3478 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3479 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3480 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3481
52763a4b
JK		 3482 if (!test_opt(sbi, LFS))
3483 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3484 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3485 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3486 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
351df4b2 3487
bba681cb
JK		 3488 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3489
184a5cd2
CY		 3490 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3491
d4fdf8ba 3492 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3493 err = create_flush_cmd_control(sbi);
3494 if (err)
a688b9d9 3495 return err;
6b4afdd7
JK		 3496 }
3497
0b54fb84
JK		 3498 err = create_discard_cmd_control(sbi);
3499 if (err)
3500 return err;
3501
351df4b2
JK		 3502 err = build_sit_info(sbi);
3503 if (err)
3504 return err;
3505 err = build_free_segmap(sbi);
3506 if (err)
3507 return err;
3508 err = build_curseg(sbi);
3509 if (err)
3510 return err;
3511
3512 /* reinit free segmap based on SIT */
3513 build_sit_entries(sbi);
3514
3515 init_free_segmap(sbi);
3516 err = build_dirty_segmap(sbi);
3517 if (err)
3518 return err;
3519
3520 init_min_max_mtime(sbi);
3521 return 0;
3522}
3523
3524static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3525 enum dirty_type dirty_type)
3526{
3527 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3528
3529 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3530 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3531 dirty_i->nr_dirty[dirty_type] = 0;
3532 mutex_unlock(&dirty_i->seglist_lock);
3533}
3534
5ec4e49f 3535static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3536{
3537 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3538 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3539}
3540
3541static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3542{
3543 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3544 int i;
3545
3546 if (!dirty_i)
3547 return;
3548
3549 /* discard pre-free/dirty segments list */
3550 for (i = 0; i < NR_DIRTY_TYPE; i++)
3551 discard_dirty_segmap(sbi, i);
3552
5ec4e49f 3553 destroy_victim_secmap(sbi);
351df4b2
JK		 3554 SM_I(sbi)->dirty_info = NULL;
3555 kfree(dirty_i);
3556}
3557
3558static void destroy_curseg(struct f2fs_sb_info *sbi)
3559{
3560 struct curseg_info *array = SM_I(sbi)->curseg_array;
3561 int i;
3562
3563 if (!array)
3564 return;
3565 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3566 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3567 kfree(array[i].sum_blk);
b7ad7512
CY		 3568 kfree(array[i].journal);
3569 }
351df4b2
JK		 3570 kfree(array);
3571}
3572
3573static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3574{
3575 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3576 if (!free_i)
3577 return;
3578 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3579 kvfree(free_i->free_segmap);
3580 kvfree(free_i->free_secmap);
351df4b2
JK		 3581 kfree(free_i);
3582}
3583
3584static void destroy_sit_info(struct f2fs_sb_info *sbi)
3585{
3586 struct sit_info *sit_i = SIT_I(sbi);
3587 unsigned int start;
3588
3589 if (!sit_i)
3590 return;
3591
3592 if (sit_i->sentries) {
7cd8558b 3593 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3594 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 3595#ifdef CONFIG_F2FS_CHECK_FS
3596 kfree(sit_i->sentries[start].cur_valid_map_mir);
3597#endif
351df4b2 3598 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3599 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3600 }
3601 }
60a3b782
JK		 3602 kfree(sit_i->tmp_map);
3603
39307a8e
JK		 3604 kvfree(sit_i->sentries);
3605 kvfree(sit_i->sec_entries);
3606 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3607
3608 SM_I(sbi)->sit_info = NULL;
3609 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 3610#ifdef CONFIG_F2FS_CHECK_FS
3611 kfree(sit_i->sit_bitmap_mir);
3612#endif
351df4b2
JK		 3613 kfree(sit_i);
3614}
3615
3616void destroy_segment_manager(struct f2fs_sb_info *sbi)
3617{
3618 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3619
3b03f724
CY		 3620 if (!sm_info)
3621 return;
5eba8c5d 3622 destroy_flush_cmd_control(sbi, true);
f099405f 3623 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3624 destroy_dirty_segmap(sbi);
3625 destroy_curseg(sbi);
3626 destroy_free_segmap(sbi);
3627 destroy_sit_info(sbi);
3628 sbi->sm_info = NULL;
3629 kfree(sm_info);
3630}
7fd9e544
JK		 3631
3632int __init create_segment_manager_caches(void)
3633{
3634 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3635 sizeof(struct discard_entry));
7fd9e544 3636 if (!discard_entry_slab)
184a5cd2
CY		 3637 goto fail;
3638
b01a9201
JK		 3639 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3640 sizeof(struct discard_cmd));
3641 if (!discard_cmd_slab)
6ab2a308 3642 goto destroy_discard_entry;
275b66b0 3643
184a5cd2 3644 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3645 sizeof(struct sit_entry_set));
184a5cd2 3646 if (!sit_entry_set_slab)
b01a9201 3647 goto destroy_discard_cmd;
88b88a66
JK		 3648
3649 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3650 sizeof(struct inmem_pages));
3651 if (!inmem_entry_slab)
3652 goto destroy_sit_entry_set;
7fd9e544 3653 return 0;
184a5cd2 3654
88b88a66
JK		 3655destroy_sit_entry_set:
3656 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 3657destroy_discard_cmd:
3658 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3659destroy_discard_entry:
184a5cd2
CY		 3660 kmem_cache_destroy(discard_entry_slab);
3661fail:
3662 return -ENOMEM;
7fd9e544
JK		 3663}
3664
3665void destroy_segment_manager_caches(void)
3666{
184a5cd2 3667 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3668 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3669 kmem_cache_destroy(discard_entry_slab);
88b88a66 3670 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3671}
Ubuntu Jammy Linux kernel mirror