projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: let discard thread wait a little longer if dev is busy
[linux-2.6-block.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
1d7be270 19#include <linux/freezer.h>
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;
5b0e9539 180 if (sbi->gc_mode == GC_URGENT)
b3a97a2a
JK		 181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
a2a12b67 184 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
b3a97a2a
JK		 185}
186
88b88a66
JK		 187void register_inmem_page(struct inode *inode, struct page *page)
188{
57864ae5 189 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
88b88a66
JK		 190 struct f2fs_inode_info *fi = F2FS_I(inode);
191 struct inmem_pages *new;
9be32d72 192
9e4ded3f 193 f2fs_trace_pid(page);
0722b101 194
decd36b6
CY		 195 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 SetPagePrivate(page);
197
88b88a66
JK		 198 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
199
200 /* add atomic page indices to the list */
201 new->page = page;
202 INIT_LIST_HEAD(&new->list);
decd36b6 203
88b88a66
JK		 204 /* increase reference count with clean state */
205 mutex_lock(&fi->inmem_lock);
206 get_page(page);
207 list_add_tail(&new->list, &fi->inmem_pages);
57864ae5
JK		 208 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
209 if (list_empty(&fi->inmem_ilist))
210 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
211 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
8dcf2ff7 212 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 213 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 214
215 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 216}
217
28bc106b
CY		 218static int __revoke_inmem_pages(struct inode *inode,
219 struct list_head *head, bool drop, bool recover)
29b96b54 220{
28bc106b 221 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 222 struct inmem_pages *cur, *tmp;
28bc106b 223 int err = 0;
29b96b54
CY		 224
225 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 226 struct page *page = cur->page;
227
228 if (drop)
229 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
230
231 lock_page(page);
29b96b54 232
e5e5732d
CY		 233 f2fs_wait_on_page_writeback(page, DATA, true);
234
28bc106b
CY		 235 if (recover) {
236 struct dnode_of_data dn;
237 struct node_info ni;
238
239 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 240retry:
28bc106b 241 set_new_dnode(&dn, inode, NULL, NULL, 0);
7f2b4e8e
CY		 242 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
243 if (err) {
244 if (err == -ENOMEM) {
245 congestion_wait(BLK_RW_ASYNC, HZ/50);
246 cond_resched();
247 goto retry;
248 }
28bc106b
CY		 249 err = -EAGAIN;
250 goto next;
251 }
252 get_node_info(sbi, dn.nid, &ni);
f1d2564a
DJ		 253 if (cur->old_addr == NEW_ADDR) {
254 invalidate_blocks(sbi, dn.data_blkaddr);
255 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
256 } else
257 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
28bc106b
CY		 258 cur->old_addr, ni.version, true, true);
259 f2fs_put_dnode(&dn);
260 }
261next:
63c52d78
JK		 262 /* we don't need to invalidate this in the sccessful status */
263 if (drop || recover)
264 ClearPageUptodate(page);
28bc106b 265 set_page_private(page, 0);
c81ced05 266 ClearPagePrivate(page);
28bc106b 267 f2fs_put_page(page, 1);
29b96b54
CY		 268
269 list_del(&cur->list);
270 kmem_cache_free(inmem_entry_slab, cur);
271 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
272 }
28bc106b 273 return err;
29b96b54
CY		 274}
275
2ef79ecb 276void drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
57864ae5
JK		 277{
278 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
279 struct inode *inode;
280 struct f2fs_inode_info *fi;
281next:
282 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
283 if (list_empty(head)) {
284 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
285 return;
286 }
287 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
288 inode = igrab(&fi->vfs_inode);
289 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
290
291 if (inode) {
2ef79ecb
CY		 292 if (gc_failure) {
293 if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
294 goto drop;
295 goto skip;
296 }
297drop:
298 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
57864ae5
JK		 299 drop_inmem_pages(inode);
300 iput(inode);
301 }
2ef79ecb 302skip:
57864ae5
JK		 303 congestion_wait(BLK_RW_ASYNC, HZ/50);
304 cond_resched();
305 goto next;
306}
307
29b96b54
CY		 308void drop_inmem_pages(struct inode *inode)
309{
57864ae5 310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54
CY		 311 struct f2fs_inode_info *fi = F2FS_I(inode);
312
313 mutex_lock(&fi->inmem_lock);
28bc106b 314 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
57864ae5
JK		 315 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
316 if (!list_empty(&fi->inmem_ilist))
317 list_del_init(&fi->inmem_ilist);
318 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
29b96b54 319 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 320
321 clear_inode_flag(inode, FI_ATOMIC_FILE);
2ef79ecb 322 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
5fe45743 323 stat_dec_atomic_write(inode);
29b96b54
CY		 324}
325
8c242db9
JK		 326void drop_inmem_page(struct inode *inode, struct page *page)
327{
328 struct f2fs_inode_info *fi = F2FS_I(inode);
329 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
330 struct list_head *head = &fi->inmem_pages;
331 struct inmem_pages *cur = NULL;
332
333 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
334
335 mutex_lock(&fi->inmem_lock);
336 list_for_each_entry(cur, head, list) {
337 if (cur->page == page)
338 break;
339 }
340
d0891e84 341 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
8c242db9
JK		 342 list_del(&cur->list);
343 mutex_unlock(&fi->inmem_lock);
344
345 dec_page_count(sbi, F2FS_INMEM_PAGES);
346 kmem_cache_free(inmem_entry_slab, cur);
347
348 ClearPageUptodate(page);
349 set_page_private(page, 0);
350 ClearPagePrivate(page);
351 f2fs_put_page(page, 0);
352
353 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
354}
355
cf52b27a 356static int __commit_inmem_pages(struct inode *inode)
88b88a66
JK		 357{
358 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
359 struct f2fs_inode_info *fi = F2FS_I(inode);
360 struct inmem_pages *cur, *tmp;
88b88a66 361 struct f2fs_io_info fio = {
05ca3632 362 .sbi = sbi,
39d787be 363 .ino = inode->i_ino,
88b88a66 364 .type = DATA,
04d328de 365 .op = REQ_OP_WRITE,
70fd7614 366 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 367 .io_type = FS_DATA_IO,
88b88a66 368 };
cf52b27a 369 struct list_head revoke_list;
942fd319 370 pgoff_t last_idx = ULONG_MAX;
edb27dee 371 int err = 0;
88b88a66 372
cf52b27a
CY		 373 INIT_LIST_HEAD(&revoke_list);
374
88b88a66 375 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 376 struct page *page = cur->page;
377
378 lock_page(page);
379 if (page->mapping == inode->i_mapping) {
380 trace_f2fs_commit_inmem_page(page, INMEM);
381
382 set_page_dirty(page);
383 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 384 if (clear_page_dirty_for_io(page)) {
29b96b54 385 inode_dec_dirty_pages(inode);
933439c8
CY		 386 remove_dirty_inode(inode);
387 }
640cc189 388retry:
28bc106b 389 fio.page = page;
e959c8f5 390 fio.old_blkaddr = NULL_ADDR;
4d978078 391 fio.encrypted_page = NULL;
cc15620b 392 fio.need_lock = LOCK_DONE;
29b96b54
CY		 393 err = do_write_data_page(&fio);
394 if (err) {
640cc189
JK		 395 if (err == -ENOMEM) {
396 congestion_wait(BLK_RW_ASYNC, HZ/50);
397 cond_resched();
398 goto retry;
399 }
28bc106b 400 unlock_page(page);
29b96b54 401 break;
70c640b1 402 }
28bc106b
CY		 403 /* record old blkaddr for revoking */
404 cur->old_addr = fio.old_blkaddr;
942fd319 405 last_idx = page->index;
28bc106b
CY		 406 }
407 unlock_page(page);
cf52b27a 408 list_move_tail(&cur->list, &revoke_list);
88b88a66 409 }
29b96b54 410
942fd319 411 if (last_idx != ULONG_MAX)
b9109b0e 412 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b 413
cf52b27a
CY		 414 if (err) {
415 /*
416 * try to revoke all committed pages, but still we could fail
417 * due to no memory or other reason, if that happened, EAGAIN
418 * will be returned, which means in such case, transaction is
419 * already not integrity, caller should use journal to do the
420 * recovery or rewrite & commit last transaction. For other
421 * error number, revoking was done by filesystem itself.
422 */
423 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
424
425 /* drop all uncommitted pages */
426 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
427 } else {
428 __revoke_inmem_pages(inode, &revoke_list, false, false);
429 }
28bc106b 430
29b96b54
CY		 431 return err;
432}
433
434int commit_inmem_pages(struct inode *inode)
435{
436 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
437 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b 438 int err;
29b96b54
CY		 439
440 f2fs_balance_fs(sbi, true);
441 f2fs_lock_op(sbi);
442
5fe45743
CY		 443 set_inode_flag(inode, FI_ATOMIC_COMMIT);
444
29b96b54 445 mutex_lock(&fi->inmem_lock);
cf52b27a 446 err = __commit_inmem_pages(inode);
28bc106b 447
57864ae5
JK		 448 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
449 if (!list_empty(&fi->inmem_ilist))
450 list_del_init(&fi->inmem_ilist);
451 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
88b88a66
JK		 452 mutex_unlock(&fi->inmem_lock);
453
5fe45743
CY		 454 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
455
29b96b54 456 f2fs_unlock_op(sbi);
edb27dee 457 return err;
88b88a66
JK		 458}
459
0a8165d7 460/*
351df4b2
JK		 461 * This function balances dirty node and dentry pages.
462 * In addition, it controls garbage collection.
463 */
2c4db1a6 464void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 465{
0f348028 466#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 467 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
468 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 469 f2fs_stop_checkpoint(sbi, false);
55523519 470 }
0f348028
CY		 471#endif
472
e589c2c4 473 /* balance_fs_bg is able to be pending */
a7881893 474 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 475 f2fs_balance_fs_bg(sbi);
476
351df4b2 477 /*
029cd28c
JK		 478 * We should do GC or end up with checkpoint, if there are so many dirty
479 * dir/node pages without enough free segments.
351df4b2 480 */
7f3037a5 481 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 482 mutex_lock(&sbi->gc_mutex);
e066b83c 483 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 484 }
485}
486
4660f9c0
JK		 487void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
488{
1dcc336b 489 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 490 if (!available_free_memory(sbi, EXTENT_CACHE))
491 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 492
1b38dc8e
JK		 493 /* check the # of cached NAT entries */
494 if (!available_free_memory(sbi, NAT_ENTRIES))
495 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
496
31696580 497 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 498 try_to_free_nids(sbi, MAX_FREE_NIDS);
499 else
22ad0b6a 500 build_free_nids(sbi, false, false);
31696580 501
1c0f4bf5 502 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 503 return;
31696580 504
1b38dc8e
JK		 505 /* checkpoint is the only way to shrink partial cached entries */
506 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 507 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 508 excess_prefree_segs(sbi) ||
509 excess_dirty_nats(sbi) ||
f455c8a5 510 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 511 if (test_opt(sbi, DATA_FLUSH)) {
512 struct blk_plug plug;
513
514 blk_start_plug(&plug);
36b35a0d 515 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 516 blk_finish_plug(&plug);
517 }
4660f9c0 518 f2fs_sync_fs(sbi->sb, true);
42190d2a 519 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 520 }
4660f9c0
JK		 521}
522
20fda56b
KM		 523static int __submit_flush_wait(struct f2fs_sb_info *sbi,
524 struct block_device *bdev)
3c62be17 525{
d62fe971 526 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
3c62be17
JK		 527 int ret;
528
3adc5fcb 529 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 530 bio_set_dev(bio, bdev);
3c62be17
JK		 531 ret = submit_bio_wait(bio);
532 bio_put(bio);
20fda56b
KM		 533
534 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
535 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 536 return ret;
537}
538
39d787be 539static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 540{
39d787be 541 int ret = 0;
3c62be17
JK		 542 int i;
543
39d787be
CY		 544 if (!sbi->s_ndevs)
545 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 546
39d787be
CY		 547 for (i = 0; i < sbi->s_ndevs; i++) {
548 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
549 continue;
20fda56b
KM		 550 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
551 if (ret)
552 break;
3c62be17
JK		 553 }
554 return ret;
555}
556
2163d198 557static int issue_flush_thread(void *data)
6b4afdd7
JK		 558{
559 struct f2fs_sb_info *sbi = data;
b01a9201 560 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 561 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 562repeat:
563 if (kthread_should_stop())
564 return 0;
565
dc6febb6
CY		 566 sb_start_intwrite(sbi->sb);
567
721bd4d5 568 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 569 struct flush_cmd *cmd, *next;
570 int ret;
571
721bd4d5
GZ		 572 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
573 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
574
39d787be
CY		 575 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
576
577 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 578 atomic_inc(&fcc->issued_flush);
579
721bd4d5
GZ		 580 llist_for_each_entry_safe(cmd, next,
581 fcc->dispatch_list, llnode) {
6b4afdd7 582 cmd->ret = ret;
6b4afdd7
JK		 583 complete(&cmd->wait);
584 }
a688b9d9 585 fcc->dispatch_list = NULL;
6b4afdd7
JK		 586 }
587
dc6febb6
CY		 588 sb_end_intwrite(sbi->sb);
589
a688b9d9 590 wait_event_interruptible(*q,
721bd4d5 591 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 592 goto repeat;
593}
594
39d787be 595int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 596{
b01a9201 597 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 598 struct flush_cmd cmd;
8b8dd65f 599 int ret;
6b4afdd7 600
0f7b2abd
JK		 601 if (test_opt(sbi, NOBARRIER))
602 return 0;
603
8b8dd65f 604 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 605 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 606 atomic_inc(&fcc->issued_flush);
607 return ret;
608 }
740432f8 609
39d787be
CY		 610 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
611 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 612 atomic_dec(&fcc->issing_flush);
613
614 atomic_inc(&fcc->issued_flush);
740432f8
JK		 615 return ret;
616 }
6b4afdd7 617
39d787be 618 cmd.ino = ino;
adf8d90b 619 init_completion(&cmd.wait);
6b4afdd7 620
721bd4d5 621 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 622
6f890df0
CY		 623 /* update issue_list before we wake up issue_flush thread */
624 smp_mb();
625
626 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 627 wake_up(&fcc->flush_wait_queue);
6b4afdd7 628
5eba8c5d
JK		 629 if (fcc->f2fs_issue_flush) {
630 wait_for_completion(&cmd.wait);
8b8dd65f 631 atomic_dec(&fcc->issing_flush);
5eba8c5d 632 } else {
d3238691
CY		 633 struct llist_node *list;
634
635 list = llist_del_all(&fcc->issue_list);
636 if (!list) {
637 wait_for_completion(&cmd.wait);
638 atomic_dec(&fcc->issing_flush);
639 } else {
640 struct flush_cmd *tmp, *next;
641
39d787be 642 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 643
644 llist_for_each_entry_safe(tmp, next, list, llnode) {
645 if (tmp == &cmd) {
646 cmd.ret = ret;
647 atomic_dec(&fcc->issing_flush);
648 continue;
649 }
650 tmp->ret = ret;
651 complete(&tmp->wait);
652 }
653 }
5eba8c5d 654 }
adf8d90b
CY		 655
656 return cmd.ret;
6b4afdd7
JK		 657}
658
2163d198
GZ		 659int create_flush_cmd_control(struct f2fs_sb_info *sbi)
660{
661 dev_t dev = sbi->sb->s_bdev->bd_dev;
662 struct flush_cmd_control *fcc;
663 int err = 0;
664
b01a9201
JK		 665 if (SM_I(sbi)->fcc_info) {
666 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 667 if (fcc->f2fs_issue_flush)
668 return err;
5eba8c5d
JK		 669 goto init_thread;
670 }
671
acbf054d 672 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
2163d198
GZ		 673 if (!fcc)
674 return -ENOMEM;
8b8dd65f
CY		 675 atomic_set(&fcc->issued_flush, 0);
676 atomic_set(&fcc->issing_flush, 0);
2163d198 677 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 678 init_llist_head(&fcc->issue_list);
b01a9201 679 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 680 if (!test_opt(sbi, FLUSH_MERGE))
681 return err;
682
5eba8c5d 683init_thread:
2163d198
GZ		 684 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
685 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
686 if (IS_ERR(fcc->f2fs_issue_flush)) {
687 err = PTR_ERR(fcc->f2fs_issue_flush);
688 kfree(fcc);
b01a9201 689 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 690 return err;
691 }
2163d198
GZ		 692
693 return err;
694}
695
5eba8c5d 696void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 697{
b01a9201 698 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 699
5eba8c5d
JK		 700 if (fcc && fcc->f2fs_issue_flush) {
701 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
702
703 fcc->f2fs_issue_flush = NULL;
704 kthread_stop(flush_thread);
705 }
706 if (free) {
707 kfree(fcc);
b01a9201 708 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 709 }
2163d198
GZ		 710}
711
1228b482
CY		 712int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
713{
714 int ret = 0, i;
715
716 if (!sbi->s_ndevs)
717 return 0;
718
719 for (i = 1; i < sbi->s_ndevs; i++) {
720 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
721 continue;
722 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
723 if (ret)
724 break;
725
726 spin_lock(&sbi->dev_lock);
727 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
728 spin_unlock(&sbi->dev_lock);
729 }
730
731 return ret;
732}
733
351df4b2
JK		 734static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
735 enum dirty_type dirty_type)
736{
737 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
738
739 /* need not be added */
740 if (IS_CURSEG(sbi, segno))
741 return;
742
743 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
744 dirty_i->nr_dirty[dirty_type]++;
745
746 if (dirty_type == DIRTY) {
747 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 748 enum dirty_type t = sentry->type;
b2f2c390 749
ec325b52
JK		 750 if (unlikely(t >= DIRTY)) {
751 f2fs_bug_on(sbi, 1);
752 return;
753 }
4625d6aa
CL		 754 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
755 dirty_i->nr_dirty[t]++;
351df4b2
JK		 756 }
757}
758
759static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
760 enum dirty_type dirty_type)
761{
762 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
763
764 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
765 dirty_i->nr_dirty[dirty_type]--;
766
767 if (dirty_type == DIRTY) {
4625d6aa
CL		 768 struct seg_entry *sentry = get_seg_entry(sbi, segno);
769 enum dirty_type t = sentry->type;
770
771 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
772 dirty_i->nr_dirty[t]--;
b2f2c390 773
302bd348 774 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 775 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 776 dirty_i->victim_secmap);
351df4b2
JK		 777 }
778}
779
0a8165d7 780/*
351df4b2
JK		 781 * Should not occur error such as -ENOMEM.
782 * Adding dirty entry into seglist is not critical operation.
783 * If a given segment is one of current working segments, it won't be added.
784 */
8d8451af 785static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 786{
787 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
788 unsigned short valid_blocks;
789
790 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
791 return;
792
793 mutex_lock(&dirty_i->seglist_lock);
794
302bd348 795 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 796
797 if (valid_blocks == 0) {
798 __locate_dirty_segment(sbi, segno, PRE);
799 __remove_dirty_segment(sbi, segno, DIRTY);
800 } else if (valid_blocks < sbi->blocks_per_seg) {
801 __locate_dirty_segment(sbi, segno, DIRTY);
802 } else {
803 /* Recovery routine with SSR needs this */
804 __remove_dirty_segment(sbi, segno, DIRTY);
805 }
806
807 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 808}
809
004b6862 810static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 811 struct block_device *bdev, block_t lstart,
812 block_t start, block_t len)
275b66b0 813{
0b54fb84 814 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 815 struct list_head *pend_list;
b01a9201 816 struct discard_cmd *dc;
275b66b0 817
ba48a33e
CY		 818 f2fs_bug_on(sbi, !len);
819
820 pend_list = &dcc->pend_list[plist_idx(len)];
821
b01a9201
JK		 822 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
823 INIT_LIST_HEAD(&dc->list);
c81abe34 824 dc->bdev = bdev;
b01a9201 825 dc->lstart = lstart;
c81abe34 826 dc->start = start;
b01a9201 827 dc->len = len;
ec9895ad 828 dc->ref = 0;
15469963 829 dc->state = D_PREP;
c81abe34 830 dc->error = 0;
b01a9201 831 init_completion(&dc->wait);
22d375dd 832 list_add_tail(&dc->list, pend_list);
5f32366a 833 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 834 dcc->undiscard_blks += len;
004b6862
CY		 835
836 return dc;
837}
838
839static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
840 struct block_device *bdev, block_t lstart,
841 block_t start, block_t len,
842 struct rb_node *parent, struct rb_node **p)
843{
844 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
845 struct discard_cmd *dc;
846
847 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
848
849 rb_link_node(&dc->rb_node, parent, p);
850 rb_insert_color(&dc->rb_node, &dcc->root);
851
852 return dc;
15469963
JK		 853}
854
004b6862
CY		 855static void __detach_discard_cmd(struct discard_cmd_control *dcc,
856 struct discard_cmd *dc)
15469963 857{
dcc9165d 858 if (dc->state == D_DONE)
004b6862
CY		 859 atomic_dec(&dcc->issing_discard);
860
861 list_del(&dc->list);
862 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 863 dcc->undiscard_blks -= dc->len;
004b6862
CY		 864
865 kmem_cache_free(discard_cmd_slab, dc);
866
867 atomic_dec(&dcc->discard_cmd_cnt);
868}
869
870static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
871 struct discard_cmd *dc)
872{
873 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 874
2ec6f2ef
CY		 875 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
876
d9703d90
CY		 877 f2fs_bug_on(sbi, dc->ref);
878
c81abe34
JK		 879 if (dc->error == -EOPNOTSUPP)
880 dc->error = 0;
15469963 881
c81abe34 882 if (dc->error)
15469963 883 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 884 "Issue discard(%u, %u, %u) failed, ret: %d",
885 dc->lstart, dc->start, dc->len, dc->error);
004b6862 886 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 887}
888
c81abe34
JK		 889static void f2fs_submit_discard_endio(struct bio *bio)
890{
891 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
892
4e4cbee9 893 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 894 dc->state = D_DONE;
e31b9821 895 complete_all(&dc->wait);
c81abe34
JK		 896 bio_put(bio);
897}
898
94b1e10e 899static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
6915ea9d
CY		 900 block_t start, block_t end)
901{
902#ifdef CONFIG_F2FS_CHECK_FS
903 struct seg_entry *sentry;
904 unsigned int segno;
905 block_t blk = start;
906 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
907 unsigned long *map;
908
909 while (blk < end) {
910 segno = GET_SEGNO(sbi, blk);
911 sentry = get_seg_entry(sbi, segno);
912 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
913
008396e1
YS		 914 if (end < START_BLOCK(sbi, segno + 1))
915 size = GET_BLKOFF_FROM_SEG0(sbi, end);
916 else
917 size = max_blocks;
6915ea9d
CY		 918 map = (unsigned long *)(sentry->cur_valid_map);
919 offset = __find_rev_next_bit(map, size, offset);
920 f2fs_bug_on(sbi, offset != size);
008396e1 921 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 922 }
923#endif
924}
925
8bb4f253
JK		 926static void __init_discard_policy(struct f2fs_sb_info *sbi,
927 struct discard_policy *dpolicy,
928 int discard_type, unsigned int granularity)
929{
930 /* common policy */
931 dpolicy->type = discard_type;
932 dpolicy->sync = true;
933 dpolicy->granularity = granularity;
934
935 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
936 dpolicy->io_aware_gran = MAX_PLIST_NUM;
937
938 if (discard_type == DPOLICY_BG) {
939 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 940 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 941 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
942 dpolicy->io_aware = true;
cba60849 943 dpolicy->sync = false;
8bb4f253
JK		 944 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
945 dpolicy->granularity = 1;
946 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
947 }
948 } else if (discard_type == DPOLICY_FORCE) {
949 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 950 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 951 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
952 dpolicy->io_aware = false;
953 } else if (discard_type == DPOLICY_FSTRIM) {
954 dpolicy->io_aware = false;
955 } else if (discard_type == DPOLICY_UMOUNT) {
241b493d 956 dpolicy->max_requests = UINT_MAX;
8bb4f253
JK		 957 dpolicy->io_aware = false;
958 }
959}
960
961
c81abe34
JK		 962/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
963static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56
CY		 964 struct discard_policy *dpolicy,
965 struct discard_cmd *dc)
c81abe34
JK		 966{
967 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 968 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
969 &(dcc->fstrim_list) : &(dcc->wait_list);
c81abe34 970 struct bio *bio = NULL;
78997b56 971 int flag = dpolicy->sync ? REQ_SYNC : 0;
c81abe34
JK		 972
973 if (dc->state != D_PREP)
974 return;
975
d6184774
YH		 976 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
977 return;
978
0243a5f9
CY		 979 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
980
c81abe34
JK		 981 dc->error = __blkdev_issue_discard(dc->bdev,
982 SECTOR_FROM_BLOCK(dc->start),
983 SECTOR_FROM_BLOCK(dc->len),
984 GFP_NOFS, 0, &bio);
985 if (!dc->error) {
986 /* should keep before submission to avoid D_DONE right away */
987 dc->state = D_SUBMIT;
8b8dd65f
CY		 988 atomic_inc(&dcc->issued_discard);
989 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 990 if (bio) {
991 bio->bi_private = dc;
992 bio->bi_end_io = f2fs_submit_discard_endio;
ecc9aa00 993 bio->bi_opf |= flag;
c81abe34 994 submit_bio(bio);
8412663d 995 list_move_tail(&dc->list, wait_list);
6915ea9d 996 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 997
998 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 999 }
1000 } else {
1001 __remove_discard_cmd(sbi, dc);
1002 }
1003}
1004
004b6862
CY		 1005static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1006 struct block_device *bdev, block_t lstart,
1007 block_t start, block_t len,
1008 struct rb_node **insert_p,
1009 struct rb_node *insert_parent)
c81abe34 1010{
004b6862 1011 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 1012 struct rb_node **p;
004b6862
CY		 1013 struct rb_node *parent = NULL;
1014 struct discard_cmd *dc = NULL;
1015
1016 if (insert_p && insert_parent) {
1017 parent = insert_parent;
1018 p = insert_p;
1019 goto do_insert;
1020 }
c81abe34 1021
004b6862
CY		 1022 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1023do_insert:
1024 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1025 if (!dc)
1026 return NULL;
c81abe34 1027
004b6862 1028 return dc;
c81abe34
JK		 1029}
1030
ba48a33e
CY		 1031static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1032 struct discard_cmd *dc)
1033{
1034 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1035}
1036
3d6a650f
YH		 1037static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1038 struct discard_cmd *dc, block_t blkaddr)
1039{
ba48a33e 1040 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1041 struct discard_info di = dc->di;
1042 bool modified = false;
3d6a650f 1043
004b6862 1044 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 1045 __remove_discard_cmd(sbi, dc);
1046 return;
1047 }
1048
d84d1cbd
CY		 1049 dcc->undiscard_blks -= di.len;
1050
004b6862 1051 if (blkaddr > di.lstart) {
3d6a650f 1052 dc->len = blkaddr - dc->lstart;
d84d1cbd 1053 dcc->undiscard_blks += dc->len;
ba48a33e 1054 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1055 modified = true;
1056 }
1057
1058 if (blkaddr < di.lstart + di.len - 1) {
1059 if (modified) {
1060 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1061 di.start + blkaddr + 1 - di.lstart,
1062 di.lstart + di.len - 1 - blkaddr,
1063 NULL, NULL);
1064 } else {
1065 dc->lstart++;
1066 dc->len--;
1067 dc->start++;
d84d1cbd 1068 dcc->undiscard_blks += dc->len;
ba48a33e 1069 __relocate_discard_cmd(dcc, dc);
004b6862 1070 }
3d6a650f
YH		 1071 }
1072}
1073
004b6862
CY		 1074static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1075 struct block_device *bdev, block_t lstart,
1076 block_t start, block_t len)
275b66b0 1077{
0b54fb84 1078 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1079 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1080 struct discard_cmd *dc;
1081 struct discard_info di = {0};
1082 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1083 block_t end = lstart + len;
275b66b0 1084
15469963 1085 mutex_lock(&dcc->cmd_lock);
40465257 1086
004b6862
CY		 1087 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1088 NULL, lstart,
1089 (struct rb_entry **)&prev_dc,
1090 (struct rb_entry **)&next_dc,
1091 &insert_p, &insert_parent, true);
1092 if (dc)
1093 prev_dc = dc;
1094
1095 if (!prev_dc) {
1096 di.lstart = lstart;
1097 di.len = next_dc ? next_dc->lstart - lstart : len;
1098 di.len = min(di.len, len);
1099 di.start = start;
22d375dd 1100 }
15469963 1101
004b6862
CY		 1102 while (1) {
1103 struct rb_node *node;
1104 bool merged = false;
1105 struct discard_cmd *tdc = NULL;
1106
1107 if (prev_dc) {
1108 di.lstart = prev_dc->lstart + prev_dc->len;
1109 if (di.lstart < lstart)
1110 di.lstart = lstart;
1111 if (di.lstart >= end)
1112 break;
1113
1114 if (!next_dc || next_dc->lstart > end)
1115 di.len = end - di.lstart;
1116 else
1117 di.len = next_dc->lstart - di.lstart;
1118 di.start = start + di.lstart - lstart;
1119 }
1120
1121 if (!di.len)
1122 goto next;
1123
1124 if (prev_dc && prev_dc->state == D_PREP &&
1125 prev_dc->bdev == bdev &&
1126 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1127 prev_dc->di.len += di.len;
d84d1cbd 1128 dcc->undiscard_blks += di.len;
ba48a33e 1129 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1130 di = prev_dc->di;
1131 tdc = prev_dc;
1132 merged = true;
1133 }
1134
1135 if (next_dc && next_dc->state == D_PREP &&
1136 next_dc->bdev == bdev &&
1137 __is_discard_front_mergeable(&di, &next_dc->di)) {
1138 next_dc->di.lstart = di.lstart;
1139 next_dc->di.len += di.len;
1140 next_dc->di.start = di.start;
d84d1cbd 1141 dcc->undiscard_blks += di.len;
ba48a33e 1142 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1143 if (tdc)
1144 __remove_discard_cmd(sbi, tdc);
004b6862 1145 merged = true;
4e6a8d9b 1146 }
004b6862 1147
df0f6b44 1148 if (!merged) {
004b6862
CY		 1149 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1150 di.len, NULL, NULL);
df0f6b44 1151 }
004b6862
CY		 1152 next:
1153 prev_dc = next_dc;
1154 if (!prev_dc)
1155 break;
1156
1157 node = rb_next(&prev_dc->rb_node);
1158 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1159 }
1160
1161 mutex_unlock(&dcc->cmd_lock);
1162}
1163
1164static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1165 struct block_device *bdev, block_t blkstart, block_t blklen)
1166{
1167 block_t lblkstart = blkstart;
1168
0243a5f9 1169 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1170
1171 if (sbi->s_ndevs) {
1172 int devi = f2fs_target_device_index(sbi, blkstart);
1173
1174 blkstart -= FDEV(devi).start_blk;
1175 }
1176 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1177 return 0;
1178}
1179
78997b56
CY		 1180static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1181 struct discard_policy *dpolicy)
bd5b0738
CY		 1182{
1183 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1184 struct list_head *pend_list;
1185 struct discard_cmd *dc, *tmp;
1186 struct blk_plug plug;
78997b56 1187 int i, iter = 0, issued = 0;
e6c6de18 1188 bool io_interrupted = false;
bd5b0738 1189
78997b56
CY		 1190 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1191 if (i + 1 < dpolicy->granularity)
1192 break;
bd5b0738 1193 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1194
1195 mutex_lock(&dcc->cmd_lock);
49c60c67
CY		 1196 if (list_empty(pend_list))
1197 goto next;
33da62cf
CY		 1198 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1199 blk_start_plug(&plug);
bd5b0738
CY		 1200 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1201 f2fs_bug_on(sbi, dc->state != D_PREP);
1202
ecc9aa00
CY		 1203 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1204 !is_idle(sbi)) {
e6c6de18 1205 io_interrupted = true;
ecc9aa00 1206 goto skip;
969d1b18 1207 }
e6c6de18 1208
78997b56 1209 __submit_discard_cmd(sbi, dpolicy, dc);
ecc9aa00
CY		 1210 issued++;
1211skip:
1212 if (++iter >= dpolicy->max_requests)
33da62cf 1213 break;
bd5b0738 1214 }
33da62cf 1215 blk_finish_plug(&plug);
49c60c67 1216next:
33da62cf
CY		 1217 mutex_unlock(&dcc->cmd_lock);
1218
1219 if (iter >= dpolicy->max_requests)
1220 break;
bd5b0738 1221 }
969d1b18 1222
e6c6de18
CY		 1223 if (!issued && io_interrupted)
1224 issued = -1;
1225
969d1b18
CY		 1226 return issued;
1227}
1228
cf5c759f 1229static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1230{
1231 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1232 struct list_head *pend_list;
1233 struct discard_cmd *dc, *tmp;
1234 int i;
cf5c759f 1235 bool dropped = false;
969d1b18
CY		 1236
1237 mutex_lock(&dcc->cmd_lock);
1238 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1239 pend_list = &dcc->pend_list[i];
1240 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1241 f2fs_bug_on(sbi, dc->state != D_PREP);
1242 __remove_discard_cmd(sbi, dc);
cf5c759f 1243 dropped = true;
969d1b18
CY		 1244 }
1245 }
1246 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1247
1248 return dropped;
bd5b0738
CY		 1249}
1250
7950e9ac
CY		 1251void drop_discard_cmd(struct f2fs_sb_info *sbi)
1252{
1253 __drop_discard_cmd(sbi);
1254}
1255
0ea80512 1256static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1257 struct discard_cmd *dc)
1258{
1259 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1260 unsigned int len = 0;
2a510c00
CY		 1261
1262 wait_for_completion_io(&dc->wait);
1263 mutex_lock(&dcc->cmd_lock);
1264 f2fs_bug_on(sbi, dc->state != D_DONE);
1265 dc->ref--;
0ea80512
CY		 1266 if (!dc->ref) {
1267 if (!dc->error)
1268 len = dc->len;
2a510c00 1269 __remove_discard_cmd(sbi, dc);
0ea80512 1270 }
2a510c00 1271 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1272
1273 return len;
2a510c00
CY		 1274}
1275
0ea80512 1276static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1277 struct discard_policy *dpolicy,
1278 block_t start, block_t end)
63a94fa1
CY		 1279{
1280 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1281 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1282 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1283 struct discard_cmd *dc, *tmp;
6afae633 1284 bool need_wait;
0ea80512 1285 unsigned int trimmed = 0;
6afae633
CY		 1286
1287next:
1288 need_wait = false;
63a94fa1
CY		 1289
1290 mutex_lock(&dcc->cmd_lock);
1291 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1292 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1293 continue;
78997b56 1294 if (dc->len < dpolicy->granularity)
8412663d 1295 continue;
78997b56 1296 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1297 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1298 if (!dc->error)
1299 trimmed += dc->len;
63a94fa1 1300 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1301 } else {
1302 dc->ref++;
1303 need_wait = true;
1304 break;
63a94fa1
CY		 1305 }
1306 }
1307 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1308
1309 if (need_wait) {
0ea80512 1310 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1311 goto next;
1312 }
0ea80512
CY		 1313
1314 return trimmed;
63a94fa1
CY		 1315}
1316
78997b56
CY		 1317static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1318 struct discard_policy *dpolicy)
8412663d 1319{
9a997188
JK		 1320 struct discard_policy dp;
1321
1322 if (dpolicy) {
1323 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1324 return;
1325 }
1326
1327 /* wait all */
8bb4f253 1328 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
9a997188 1329 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8bb4f253 1330 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
9a997188 1331 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8412663d
CY		 1332}
1333
004b6862 1334/* This should be covered by global mutex, &sit_i->sentry_lock */
94b1e10e 1335static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
004b6862
CY		 1336{
1337 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1338 struct discard_cmd *dc;
ec9895ad 1339 bool need_wait = false;
004b6862
CY		 1340
1341 mutex_lock(&dcc->cmd_lock);
004b6862
CY		 1342 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1343 if (dc) {
ec9895ad
CY		 1344 if (dc->state == D_PREP) {
1345 __punch_discard_cmd(sbi, dc, blkaddr);
1346 } else {
1347 dc->ref++;
1348 need_wait = true;
1349 }
275b66b0 1350 }
d431413f 1351 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1352
2a510c00
CY		 1353 if (need_wait)
1354 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1355}
1356
cce13252
CY		 1357void stop_discard_thread(struct f2fs_sb_info *sbi)
1358{
1359 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1360
1361 if (dcc && dcc->f2fs_issue_discard) {
1362 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1363
1364 dcc->f2fs_issue_discard = NULL;
1365 kthread_stop(discard_thread);
ec9895ad 1366 }
d431413f
CY		 1367}
1368
8412663d 1369/* This comes from f2fs_put_super */
cf5c759f 1370bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
969d1b18
CY		 1371{
1372 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1373 struct discard_policy dpolicy;
cf5c759f 1374 bool dropped;
969d1b18 1375
8bb4f253
JK		 1376 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1377 dcc->discard_granularity);
78997b56 1378 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1379 dropped = __drop_discard_cmd(sbi);
cf5c759f 1380
9a997188
JK		 1381 /* just to make sure there is no pending discard commands */
1382 __wait_all_discard_cmd(sbi, NULL);
cf5c759f 1383 return dropped;
969d1b18
CY		 1384}
1385
15469963
JK		 1386static int issue_discard_thread(void *data)
1387{
1388 struct f2fs_sb_info *sbi = data;
1389 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1390 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1391 struct discard_policy dpolicy;
969d1b18
CY		 1392 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1393 int issued;
15469963 1394
1d7be270 1395 set_freezable();
15469963 1396
1d7be270 1397 do {
8bb4f253 1398 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
78997b56
CY		 1399 dcc->discard_granularity);
1400
969d1b18
CY		 1401 wait_event_interruptible_timeout(*q,
1402 kthread_should_stop() || freezing(current) ||
1403 dcc->discard_wake,
1404 msecs_to_jiffies(wait_ms));
1d7be270
JK		 1405 if (try_to_freeze())
1406 continue;
3b60d802
CY		 1407 if (f2fs_readonly(sbi->sb))
1408 continue;
1d7be270
JK		 1409 if (kthread_should_stop())
1410 return 0;
d6184774
YH		 1411 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1412 wait_ms = dpolicy.max_interval;
1413 continue;
1414 }
15469963 1415
dee02f0d 1416 if (dcc->discard_wake)
969d1b18 1417 dcc->discard_wake = 0;
dee02f0d 1418
5b0e9539 1419 if (sbi->gc_mode == GC_URGENT)
8bb4f253 1420 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
969d1b18 1421
dc6febb6
CY		 1422 sb_start_intwrite(sbi->sb);
1423
78997b56 1424 issued = __issue_discard_cmd(sbi, &dpolicy);
f9d1dced 1425 if (issued > 0) {
78997b56
CY		 1426 __wait_all_discard_cmd(sbi, &dpolicy);
1427 wait_ms = dpolicy.min_interval;
f9d1dced
YH		 1428 } else if (issued == -1){
1429 wait_ms = dpolicy.mid_interval;
969d1b18 1430 } else {
78997b56 1431 wait_ms = dpolicy.max_interval;
969d1b18 1432 }
1d7be270 1433
dc6febb6 1434 sb_end_intwrite(sbi->sb);
1d7be270 1435
1d7be270
JK		 1436 } while (!kthread_should_stop());
1437 return 0;
15469963
JK		 1438}
1439
f46e8809 1440#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1441static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1442 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1443{
92592285 1444 sector_t sector, nr_sects;
10a875f8 1445 block_t lblkstart = blkstart;
3c62be17
JK		 1446 int devi = 0;
1447
1448 if (sbi->s_ndevs) {
1449 devi = f2fs_target_device_index(sbi, blkstart);
1450 blkstart -= FDEV(devi).start_blk;
1451 }
f46e8809
DLM		 1452
1453 /*
1454 * We need to know the type of the zone: for conventional zones,
1455 * use regular discard if the drive supports it. For sequential
1456 * zones, reset the zone write pointer.
1457 */
3c62be17 1458 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1459
1460 case BLK_ZONE_TYPE_CONVENTIONAL:
1461 if (!blk_queue_discard(bdev_get_queue(bdev)))
1462 return 0;
c81abe34 1463 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1464 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1465 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1466 sector = SECTOR_FROM_BLOCK(blkstart);
1467 nr_sects = SECTOR_FROM_BLOCK(blklen);
1468
1469 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1470 nr_sects != bdev_zone_sectors(bdev)) {
1471 f2fs_msg(sbi->sb, KERN_INFO,
1472 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1473 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1474 blkstart, blklen);
1475 return -EIO;
1476 }
d50aaeec 1477 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1478 return blkdev_reset_zones(bdev, sector,
1479 nr_sects, GFP_NOFS);
1480 default:
1481 /* Unknown zone type: broken device ? */
1482 return -EIO;
1483 }
1484}
1485#endif
1486
3c62be17
JK		 1487static int __issue_discard_async(struct f2fs_sb_info *sbi,
1488 struct block_device *bdev, block_t blkstart, block_t blklen)
1489{
1490#ifdef CONFIG_BLK_DEV_ZONED
ccd31cb2 1491 if (f2fs_sb_has_blkzoned(sbi->sb) &&
3c62be17
JK		 1492 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1493 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1494#endif
c81abe34 1495 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1496}
1497
1e87a78d 1498static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1499 block_t blkstart, block_t blklen)
1500{
3c62be17
JK		 1501 sector_t start = blkstart, len = 0;
1502 struct block_device *bdev;
a66cdd98
JK		 1503 struct seg_entry *se;
1504 unsigned int offset;
1505 block_t i;
3c62be17
JK		 1506 int err = 0;
1507
1508 bdev = f2fs_target_device(sbi, blkstart, NULL);
1509
1510 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1511 if (i != start) {
1512 struct block_device *bdev2 =
1513 f2fs_target_device(sbi, i, NULL);
1514
1515 if (bdev2 != bdev) {
1516 err = __issue_discard_async(sbi, bdev,
1517 start, len);
1518 if (err)
1519 return err;
1520 bdev = bdev2;
1521 start = i;
1522 len = 0;
1523 }
1524 }
a66cdd98 1525
a66cdd98
JK		 1526 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1527 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1528
1529 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1530 sbi->discard_blks--;
1531 }
f46e8809 1532
3c62be17
JK		 1533 if (len)
1534 err = __issue_discard_async(sbi, bdev, start, len);
1535 return err;
1e87a78d
JK		 1536}
1537
25290fa5
JK		 1538static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1539 bool check_only)
adf4983b 1540{
b2955550
JK		 1541 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1542 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1543 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1544 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1545 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1546 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1547 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1548 unsigned int start = 0, end = -1;
c473f1a9 1549 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1550 struct discard_entry *de = NULL;
46f84c2c 1551 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1552 int i;
1553
3e025740 1554 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1555 return false;
b2955550 1556
a66cdd98
JK		 1557 if (!force) {
1558 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1559 SM_I(sbi)->dcc_info->nr_discards >=
1560 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1561 return false;
4b2fecc8
JK		 1562 }
1563
b2955550
JK		 1564 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1565 for (i = 0; i < entries; i++)
a66cdd98 1566 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1567 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1568
0b54fb84
JK		 1569 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1570 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1571 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1572 if (start >= max_blocks)
1573 break;
1574
1575 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1576 if (force && start && end != max_blocks
1577 && (end - start) < cpc->trim_minlen)
1578 continue;
1579
25290fa5
JK		 1580 if (check_only)
1581 return true;
1582
a7eeb823
CY		 1583 if (!de) {
1584 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1585 GFP_F2FS_ZERO);
1586 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1587 list_add_tail(&de->list, head);
1588 }
1589
1590 for (i = start; i < end; i++)
1591 __set_bit_le(i, (void *)de->discard_map);
1592
1593 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1594 }
25290fa5 1595 return false;
b2955550
JK		 1596}
1597
af8ff65b
CY		 1598static void release_discard_addr(struct discard_entry *entry)
1599{
1600 list_del(&entry->list);
1601 kmem_cache_free(discard_entry_slab, entry);
1602}
1603
4b2fecc8
JK		 1604void release_discard_addrs(struct f2fs_sb_info *sbi)
1605{
46f84c2c 1606 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1607 struct discard_entry *entry, *this;
1608
1609 /* drop caches */
af8ff65b
CY		 1610 list_for_each_entry_safe(entry, this, head, list)
1611 release_discard_addr(entry);
4b2fecc8
JK		 1612}
1613
0a8165d7 1614/*
351df4b2
JK		 1615 * Should call clear_prefree_segments after checkpoint is done.
1616 */
1617static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1618{
1619 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1620 unsigned int segno;
351df4b2
JK		 1621
1622 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1623 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1624 __set_test_and_free(sbi, segno);
351df4b2
JK		 1625 mutex_unlock(&dirty_i->seglist_lock);
1626}
1627
836b5a63 1628void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1629{
969d1b18
CY		 1630 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1631 struct list_head *head = &dcc->entry_list;
2d7b822a 1632 struct discard_entry *entry, *this;
351df4b2 1633 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1634 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1635 unsigned int start = 0, end = -1;
36abef4e 1636 unsigned int secno, start_segno;
c473f1a9 1637 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK		 1638
1639 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1640
351df4b2 1641 while (1) {
29e59c14 1642 int i;
7cd8558b
JK		 1643 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1644 if (start >= MAIN_SEGS(sbi))
351df4b2 1645 break;
7cd8558b
JK		 1646 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1647 start + 1);
29e59c14
CL		 1648
1649 for (i = start; i < end; i++)
1650 clear_bit(i, prefree_map);
1651
1652 dirty_i->nr_dirty[PRE] -= end - start;
1653
650d3c4e 1654 if (!test_opt(sbi, DISCARD))
29e59c14 1655 continue;
351df4b2 1656
650d3c4e
YH		 1657 if (force && start >= cpc->trim_start &&
1658 (end - 1) <= cpc->trim_end)
1659 continue;
1660
36abef4e
JK		 1661 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1662 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1663 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1664 continue;
1665 }
1666next:
4ddb1a4d
JK		 1667 secno = GET_SEC_FROM_SEG(sbi, start);
1668 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1669 if (!IS_CURSEC(sbi, secno) &&
302bd348 1670 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1671 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1672 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1673
1674 start = start_segno + sbi->segs_per_sec;
1675 if (start < end)
1676 goto next;
8b107f5b
JK		 1677 else
1678 end = start - 1;
351df4b2
JK		 1679 }
1680 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1681
1682 /* send small discards */
2d7b822a 1683 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1684 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1685 bool is_valid = test_bit_le(0, entry->discard_map);
1686
1687find_next:
1688 if (is_valid) {
1689 next_pos = find_next_zero_bit_le(entry->discard_map,
1690 sbi->blocks_per_seg, cur_pos);
1691 len = next_pos - cur_pos;
1692
ccd31cb2 1693 if (f2fs_sb_has_blkzoned(sbi->sb) ||
acfd2810 1694 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1695 goto skip;
1696
1697 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1698 len);
a7eeb823
CY		 1699 total_len += len;
1700 } else {
1701 next_pos = find_next_bit_le(entry->discard_map,
1702 sbi->blocks_per_seg, cur_pos);
1703 }
836b5a63 1704skip:
a7eeb823
CY		 1705 cur_pos = next_pos;
1706 is_valid = !is_valid;
1707
1708 if (cur_pos < sbi->blocks_per_seg)
1709 goto find_next;
1710
af8ff65b 1711 release_discard_addr(entry);
969d1b18 1712 dcc->nr_discards -= total_len;
b2955550 1713 }
34e159da 1714
01983c71 1715 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1716}
1717
8ed59745 1718static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1719{
15469963 1720 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1721 struct discard_cmd_control *dcc;
ba48a33e 1722 int err = 0, i;
0b54fb84
JK		 1723
1724 if (SM_I(sbi)->dcc_info) {
1725 dcc = SM_I(sbi)->dcc_info;
1726 goto init_thread;
1727 }
1728
acbf054d 1729 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 1730 if (!dcc)
1731 return -ENOMEM;
1732
969d1b18 1733 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1734 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 1735 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 1736 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 1737 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 1738 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 1739 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1740 atomic_set(&dcc->issued_discard, 0);
1741 atomic_set(&dcc->issing_discard, 0);
5f32366a 1742 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1743 dcc->nr_discards = 0;
d618ebaf 1744 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1745 dcc->undiscard_blks = 0;
004b6862 1746 dcc->root = RB_ROOT;
0b54fb84 1747
15469963 1748 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1749 SM_I(sbi)->dcc_info = dcc;
1750init_thread:
15469963
JK		 1751 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1752 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1753 if (IS_ERR(dcc->f2fs_issue_discard)) {
1754 err = PTR_ERR(dcc->f2fs_issue_discard);
1755 kfree(dcc);
1756 SM_I(sbi)->dcc_info = NULL;
1757 return err;
1758 }
1759
0b54fb84
JK		 1760 return err;
1761}
1762
f099405f 1763static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1764{
1765 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1766
f099405f
CY		 1767 if (!dcc)
1768 return;
1769
cce13252 1770 stop_discard_thread(sbi);
f099405f
CY		 1771
1772 kfree(dcc);
1773 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1774}
1775
184a5cd2 1776static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1777{
1778 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1779
1780 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1781 sit_i->dirty_sentries++;
184a5cd2
CY		 1782 return false;
1783 }
1784
1785 return true;
351df4b2
JK		 1786}
1787
1788static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1789 unsigned int segno, int modified)
1790{
1791 struct seg_entry *se = get_seg_entry(sbi, segno);
1792 se->type = type;
1793 if (modified)
1794 __mark_sit_entry_dirty(sbi, segno);
1795}
1796
1797static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1798{
1799 struct seg_entry *se;
1800 unsigned int segno, offset;
1801 long int new_vblocks;
6415fedc
YS		 1802 bool exist;
1803#ifdef CONFIG_F2FS_CHECK_FS
1804 bool mir_exist;
1805#endif
351df4b2
JK		 1806
1807 segno = GET_SEGNO(sbi, blkaddr);
1808
1809 se = get_seg_entry(sbi, segno);
1810 new_vblocks = se->valid_blocks + del;
491c0854 1811 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1812
9850cf4a 1813 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1814 (new_vblocks > sbi->blocks_per_seg)));
1815
1816 se->valid_blocks = new_vblocks;
1817 se->mtime = get_mtime(sbi);
1818 SIT_I(sbi)->max_mtime = se->mtime;
1819
1820 /* Update valid block bitmap */
1821 if (del > 0) {
6415fedc 1822 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1823#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1824 mir_exist = f2fs_test_and_set_bit(offset,
1825 se->cur_valid_map_mir);
1826 if (unlikely(exist != mir_exist)) {
1827 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1828 "when setting bitmap, blk:%u, old bit:%d",
1829 blkaddr, exist);
05796763 1830 f2fs_bug_on(sbi, 1);
6415fedc 1831 }
355e7891 1832#endif
6415fedc
YS		 1833 if (unlikely(exist)) {
1834 f2fs_msg(sbi->sb, KERN_ERR,
1835 "Bitmap was wrongly set, blk:%u", blkaddr);
1836 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1837 se->valid_blocks--;
1838 del = 0;
355e7891 1839 }
6415fedc 1840
3e025740
JK		 1841 if (f2fs_discard_en(sbi) &&
1842 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1843 sbi->discard_blks--;
720037f9
JK		 1844
1845 /* don't overwrite by SSR to keep node chain */
5d7881ca 1846 if (IS_NODESEG(se->type)) {
720037f9
JK		 1847 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1848 se->ckpt_valid_blocks++;
1849 }
351df4b2 1850 } else {
6415fedc 1851 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1852#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1853 mir_exist = f2fs_test_and_clear_bit(offset,
1854 se->cur_valid_map_mir);
1855 if (unlikely(exist != mir_exist)) {
1856 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1857 "when clearing bitmap, blk:%u, old bit:%d",
1858 blkaddr, exist);
05796763 1859 f2fs_bug_on(sbi, 1);
6415fedc 1860 }
355e7891 1861#endif
6415fedc
YS		 1862 if (unlikely(!exist)) {
1863 f2fs_msg(sbi->sb, KERN_ERR,
1864 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1865 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1866 se->valid_blocks++;
1867 del = 0;
355e7891 1868 }
6415fedc 1869
3e025740
JK		 1870 if (f2fs_discard_en(sbi) &&
1871 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1872 sbi->discard_blks++;
351df4b2
JK		 1873 }
1874 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1875 se->ckpt_valid_blocks += del;
1876
1877 __mark_sit_entry_dirty(sbi, segno);
1878
1879 /* update total number of valid blocks to be written in ckpt area */
1880 SIT_I(sbi)->written_valid_blocks += del;
1881
1882 if (sbi->segs_per_sec > 1)
1883 get_sec_entry(sbi, segno)->valid_blocks += del;
1884}
1885
351df4b2
JK		 1886void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1887{
1888 unsigned int segno = GET_SEGNO(sbi, addr);
1889 struct sit_info *sit_i = SIT_I(sbi);
1890
9850cf4a 1891 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1892 if (addr == NEW_ADDR)
1893 return;
1894
1895 /* add it into sit main buffer */
3d26fa6b 1896 down_write(&sit_i->sentry_lock);
351df4b2
JK		 1897
1898 update_sit_entry(sbi, addr, -1);
1899
1900 /* add it into dirty seglist */
1901 locate_dirty_segment(sbi, segno);
1902
3d26fa6b 1903 up_write(&sit_i->sentry_lock);
351df4b2
JK		 1904}
1905
6e2c64ad
JK		 1906bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1907{
1908 struct sit_info *sit_i = SIT_I(sbi);
1909 unsigned int segno, offset;
1910 struct seg_entry *se;
1911 bool is_cp = false;
1912
7b525dd0 1913 if (!is_valid_blkaddr(blkaddr))
6e2c64ad
JK		 1914 return true;
1915
3d26fa6b 1916 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1917
1918 segno = GET_SEGNO(sbi, blkaddr);
1919 se = get_seg_entry(sbi, segno);
1920 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1921
1922 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1923 is_cp = true;
1924
3d26fa6b 1925 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1926
1927 return is_cp;
1928}
1929
0a8165d7 1930/*
351df4b2
JK		 1931 * This function should be resided under the curseg_mutex lock
1932 */
1933static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1934 struct f2fs_summary *sum)
351df4b2
JK		 1935{
1936 struct curseg_info *curseg = CURSEG_I(sbi, type);
1937 void *addr = curseg->sum_blk;
e79efe3b 1938 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1939 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1940}
1941
0a8165d7 1942/*
351df4b2
JK		 1943 * Calculate the number of current summary pages for writing
1944 */
3fa06d7b 1945int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1946{
351df4b2 1947 int valid_sum_count = 0;
9a47938b 1948 int i, sum_in_page;
351df4b2
JK		 1949
1950 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1951 if (sbi->ckpt->alloc_type[i] == SSR)
1952 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1953 else {
1954 if (for_ra)
1955 valid_sum_count += le16_to_cpu(
1956 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1957 else
1958 valid_sum_count += curseg_blkoff(sbi, i);
1959 }
351df4b2
JK		 1960 }
1961
09cbfeaf 1962 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1963 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1964 if (valid_sum_count <= sum_in_page)
351df4b2 1965 return 1;
9a47938b 1966 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1967 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1968 return 2;
1969 return 3;
1970}
1971
0a8165d7 1972/*
351df4b2
JK		 1973 * Caller should put this summary page
1974 */
1975struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1976{
1977 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1978}
1979
381722d2 1980void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1981{
1982 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2 1983
0537b811 1984 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 1985 set_page_dirty(page);
1986 f2fs_put_page(page, 1);
1987}
1988
381722d2
CY		 1989static void write_sum_page(struct f2fs_sb_info *sbi,
1990 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1991{
1992 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1993}
1994
b7ad7512
CY		 1995static void write_current_sum_page(struct f2fs_sb_info *sbi,
1996 int type, block_t blk_addr)
1997{
1998 struct curseg_info *curseg = CURSEG_I(sbi, type);
1999 struct page *page = grab_meta_page(sbi, blk_addr);
2000 struct f2fs_summary_block *src = curseg->sum_blk;
2001 struct f2fs_summary_block *dst;
2002
2003 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 2004 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 2005
2006 mutex_lock(&curseg->curseg_mutex);
2007
2008 down_read(&curseg->journal_rwsem);
2009 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2010 up_read(&curseg->journal_rwsem);
2011
2012 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2013 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2014
2015 mutex_unlock(&curseg->curseg_mutex);
2016
2017 set_page_dirty(page);
2018 f2fs_put_page(page, 1);
2019}
2020
a7881893
JK		 2021static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2022{
2023 struct curseg_info *curseg = CURSEG_I(sbi, type);
2024 unsigned int segno = curseg->segno + 1;
2025 struct free_segmap_info *free_i = FREE_I(sbi);
2026
2027 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2028 return !test_bit(segno, free_i->free_segmap);
2029 return 0;
2030}
2031
0a8165d7 2032/*
351df4b2
JK		 2033 * Find a new segment from the free segments bitmap to right order
2034 * This function should be returned with success, otherwise BUG
2035 */
2036static void get_new_segment(struct f2fs_sb_info *sbi,
2037 unsigned int *newseg, bool new_sec, int dir)
2038{
2039 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2040 unsigned int segno, secno, zoneno;
7cd8558b 2041 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2042 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2043 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2044 unsigned int left_start = hint;
2045 bool init = true;
2046 int go_left = 0;
2047 int i;
2048
1a118ccf 2049 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2050
2051 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2052 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2053 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2054 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2055 goto got_it;
2056 }
2057find_other_zone:
7cd8558b
JK		 2058 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2059 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2060 if (dir == ALLOC_RIGHT) {
2061 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2062 MAIN_SECS(sbi), 0);
2063 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2064 } else {
2065 go_left = 1;
2066 left_start = hint - 1;
2067 }
2068 }
2069 if (go_left == 0)
2070 goto skip_left;
2071
2072 while (test_bit(left_start, free_i->free_secmap)) {
2073 if (left_start > 0) {
2074 left_start--;
2075 continue;
2076 }
2077 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2078 MAIN_SECS(sbi), 0);
2079 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2080 break;
2081 }
2082 secno = left_start;
2083skip_left:
4ddb1a4d
JK		 2084 segno = GET_SEG_FROM_SEC(sbi, secno);
2085 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2086
2087 /* give up on finding another zone */
2088 if (!init)
2089 goto got_it;
2090 if (sbi->secs_per_zone == 1)
2091 goto got_it;
2092 if (zoneno == old_zoneno)
2093 goto got_it;
2094 if (dir == ALLOC_LEFT) {
2095 if (!go_left && zoneno + 1 >= total_zones)
2096 goto got_it;
2097 if (go_left && zoneno == 0)
2098 goto got_it;
2099 }
2100 for (i = 0; i < NR_CURSEG_TYPE; i++)
2101 if (CURSEG_I(sbi, i)->zone == zoneno)
2102 break;
2103
2104 if (i < NR_CURSEG_TYPE) {
2105 /* zone is in user, try another */
2106 if (go_left)
2107 hint = zoneno * sbi->secs_per_zone - 1;
2108 else if (zoneno + 1 >= total_zones)
2109 hint = 0;
2110 else
2111 hint = (zoneno + 1) * sbi->secs_per_zone;
2112 init = false;
2113 goto find_other_zone;
2114 }
2115got_it:
2116 /* set it as dirty segment in free segmap */
9850cf4a 2117 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2118 __set_inuse(sbi, segno);
2119 *newseg = segno;
1a118ccf 2120 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2121}
2122
2123static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2124{
2125 struct curseg_info *curseg = CURSEG_I(sbi, type);
2126 struct summary_footer *sum_footer;
2127
2128 curseg->segno = curseg->next_segno;
4ddb1a4d 2129 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2130 curseg->next_blkoff = 0;
2131 curseg->next_segno = NULL_SEGNO;
2132
2133 sum_footer = &(curseg->sum_blk->footer);
2134 memset(sum_footer, 0, sizeof(struct summary_footer));
2135 if (IS_DATASEG(type))
2136 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2137 if (IS_NODESEG(type))
2138 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2139 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2140}
2141
7a20b8a6
JK		 2142static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2143{
a7881893
JK		 2144 /* if segs_per_sec is large than 1, we need to keep original policy. */
2145 if (sbi->segs_per_sec != 1)
2146 return CURSEG_I(sbi, type)->segno;
2147
b94929d9
YS		 2148 if (test_opt(sbi, NOHEAP) &&
2149 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2150 return 0;
2151
e066b83c
JK		 2152 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2153 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2154
2155 /* find segments from 0 to reuse freed segments */
63189b78 2156 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2157 return 0;
2158
7a20b8a6
JK		 2159 return CURSEG_I(sbi, type)->segno;
2160}
2161
0a8165d7 2162/*
351df4b2
JK		 2163 * Allocate a current working segment.
2164 * This function always allocates a free segment in LFS manner.
2165 */
2166static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2167{
2168 struct curseg_info *curseg = CURSEG_I(sbi, type);
2169 unsigned int segno = curseg->segno;
2170 int dir = ALLOC_LEFT;
2171
2172 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2173 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2174 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2175 dir = ALLOC_RIGHT;
2176
2177 if (test_opt(sbi, NOHEAP))
2178 dir = ALLOC_RIGHT;
2179
7a20b8a6 2180 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2181 get_new_segment(sbi, &segno, new_sec, dir);
2182 curseg->next_segno = segno;
2183 reset_curseg(sbi, type, 1);
2184 curseg->alloc_type = LFS;
2185}
2186
2187static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2188 struct curseg_info *seg, block_t start)
2189{
2190 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2191 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2192 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2193 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2194 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2195 int i, pos;
2196
2197 for (i = 0; i < entries; i++)
2198 target_map[i] = ckpt_map[i] | cur_map[i];
2199
2200 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2201
2202 seg->next_blkoff = pos;
351df4b2
JK		 2203}
2204
0a8165d7 2205/*
351df4b2
JK		 2206 * If a segment is written by LFS manner, next block offset is just obtained
2207 * by increasing the current block offset. However, if a segment is written by
2208 * SSR manner, next block offset obtained by calling __next_free_blkoff
2209 */
2210static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2211 struct curseg_info *seg)
2212{
2213 if (seg->alloc_type == SSR)
2214 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2215 else
2216 seg->next_blkoff++;
2217}
2218
0a8165d7 2219/*
e1c42045 2220 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2221 * manner, so it should recover the existing segment information of valid blocks
2222 */
025d63a4 2223static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2224{
2225 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2226 struct curseg_info *curseg = CURSEG_I(sbi, type);
2227 unsigned int new_segno = curseg->next_segno;
2228 struct f2fs_summary_block *sum_node;
2229 struct page *sum_page;
2230
2231 write_sum_page(sbi, curseg->sum_blk,
2232 GET_SUM_BLOCK(sbi, curseg->segno));
2233 __set_test_and_inuse(sbi, new_segno);
2234
2235 mutex_lock(&dirty_i->seglist_lock);
2236 __remove_dirty_segment(sbi, new_segno, PRE);
2237 __remove_dirty_segment(sbi, new_segno, DIRTY);
2238 mutex_unlock(&dirty_i->seglist_lock);
2239
2240 reset_curseg(sbi, type, 1);
2241 curseg->alloc_type = SSR;
2242 __next_free_blkoff(sbi, curseg, 0);
2243
025d63a4
CY		 2244 sum_page = get_sum_page(sbi, new_segno);
2245 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2246 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2247 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2248}
2249
43727527
JK		 2250static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2251{
2252 struct curseg_info *curseg = CURSEG_I(sbi, type);
2253 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2254 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2255 int i, cnt;
2256 bool reversed = false;
c192f7a4
JK		 2257
2258 /* need_SSR() already forces to do this */
e066b83c
JK		 2259 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2260 curseg->next_segno = segno;
c192f7a4 2261 return 1;
e066b83c 2262 }
43727527 2263
70d625cb
JK		 2264 /* For node segments, let's do SSR more intensively */
2265 if (IS_NODESEG(type)) {
d27c3d89
CY		 2266 if (type >= CURSEG_WARM_NODE) {
2267 reversed = true;
2268 i = CURSEG_COLD_NODE;
2269 } else {
2270 i = CURSEG_HOT_NODE;
2271 }
2272 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2273 } else {
d27c3d89
CY		 2274 if (type >= CURSEG_WARM_DATA) {
2275 reversed = true;
2276 i = CURSEG_COLD_DATA;
2277 } else {
2278 i = CURSEG_HOT_DATA;
2279 }
2280 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2281 }
43727527 2282
d27c3d89 2283 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2284 if (i == type)
2285 continue;
e066b83c
JK		 2286 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2287 curseg->next_segno = segno;
43727527 2288 return 1;
e066b83c 2289 }
c192f7a4 2290 }
43727527
JK		 2291 return 0;
2292}
2293
351df4b2
JK		 2294/*
2295 * flush out current segment and replace it with new segment
2296 * This function should be returned with success, otherwise BUG
2297 */
2298static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2299 int type, bool force)
2300{
a7881893
JK		 2301 struct curseg_info *curseg = CURSEG_I(sbi, type);
2302
7b405275 2303 if (force)
351df4b2 2304 new_curseg(sbi, type, true);
5b6c6be2
JK		 2305 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2306 type == CURSEG_WARM_NODE)
351df4b2 2307 new_curseg(sbi, type, false);
a7881893
JK		 2308 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2309 new_curseg(sbi, type, false);
351df4b2 2310 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2311 change_curseg(sbi, type);
351df4b2
JK		 2312 else
2313 new_curseg(sbi, type, false);
dcdfff65 2314
a7881893 2315 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2316}
2317
2318void allocate_new_segments(struct f2fs_sb_info *sbi)
2319{
6ae1be13
JK		 2320 struct curseg_info *curseg;
2321 unsigned int old_segno;
351df4b2
JK		 2322 int i;
2323
3d26fa6b
CY		 2324 down_write(&SIT_I(sbi)->sentry_lock);
2325
6ae1be13
JK		 2326 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2327 curseg = CURSEG_I(sbi, i);
2328 old_segno = curseg->segno;
2329 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2330 locate_dirty_segment(sbi, old_segno);
2331 }
3d26fa6b
CY		 2332
2333 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2334}
2335
2336static const struct segment_allocation default_salloc_ops = {
2337 .allocate_segment = allocate_segment_by_default,
2338};
2339
25290fa5
JK		 2340bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2341{
2342 __u64 trim_start = cpc->trim_start;
2343 bool has_candidate = false;
2344
3d26fa6b 2345 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2346 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2347 if (add_discard_addrs(sbi, cpc, true)) {
2348 has_candidate = true;
2349 break;
2350 }
2351 }
3d26fa6b 2352 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2353
2354 cpc->trim_start = trim_start;
2355 return has_candidate;
2356}
2357
9a997188
JK		 2358static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2359 struct discard_policy *dpolicy,
2360 unsigned int start, unsigned int end)
2361{
2362 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2363 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2364 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2365 struct discard_cmd *dc;
2366 struct blk_plug plug;
2367 int issued;
2368
2369next:
2370 issued = 0;
2371
2372 mutex_lock(&dcc->cmd_lock);
2373 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
2374
2375 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
2376 NULL, start,
2377 (struct rb_entry **)&prev_dc,
2378 (struct rb_entry **)&next_dc,
2379 &insert_p, &insert_parent, true);
2380 if (!dc)
2381 dc = next_dc;
2382
2383 blk_start_plug(&plug);
2384
2385 while (dc && dc->lstart <= end) {
2386 struct rb_node *node;
2387
2388 if (dc->len < dpolicy->granularity)
2389 goto skip;
2390
2391 if (dc->state != D_PREP) {
2392 list_move_tail(&dc->list, &dcc->fstrim_list);
2393 goto skip;
2394 }
2395
2396 __submit_discard_cmd(sbi, dpolicy, dc);
2397
2398 if (++issued >= dpolicy->max_requests) {
2399 start = dc->lstart + dc->len;
2400
2401 blk_finish_plug(&plug);
2402 mutex_unlock(&dcc->cmd_lock);
2403 __wait_all_discard_cmd(sbi, NULL);
2404 congestion_wait(BLK_RW_ASYNC, HZ/50);
2405 goto next;
2406 }
2407skip:
2408 node = rb_next(&dc->rb_node);
2409 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2410
2411 if (fatal_signal_pending(current))
2412 break;
2413 }
2414
2415 blk_finish_plug(&plug);
2416 mutex_unlock(&dcc->cmd_lock);
2417}
2418
4b2fecc8
JK		 2419int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2420{
f7ef9b83
JK		 2421 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2422 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 2423 unsigned int start_segno, end_segno;
8412663d 2424 block_t start_block, end_block;
4b2fecc8 2425 struct cp_control cpc;
78997b56 2426 struct discard_policy dpolicy;
0ea80512 2427 unsigned long long trimmed = 0;
c34f42e2 2428 int err = 0;
4b2fecc8 2429
836b5a63 2430 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2431 return -EINVAL;
2432
7cd8558b 2433 if (end <= MAIN_BLKADDR(sbi))
3d165dc3 2434 return -EINVAL;
4b2fecc8 2435
ed214a11
YH		 2436 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2437 f2fs_msg(sbi->sb, KERN_WARNING,
2438 "Found FS corruption, run fsck to fix.");
3d165dc3 2439 return -EIO;
ed214a11
YH		 2440 }
2441
4b2fecc8 2442 /* start/end segment number in main_area */
7cd8558b
JK		 2443 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2444 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2445 GET_SEGNO(sbi, end);
8412663d 2446
4b2fecc8 2447 cpc.reason = CP_DISCARD;
836b5a63 2448 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 2449 cpc.trim_start = start_segno;
2450 cpc.trim_end = end_segno;
4b2fecc8 2451
377224c4
CY		 2452 if (sbi->discard_blks == 0)
2453 goto out;
74fa5f3d 2454
377224c4
CY		 2455 mutex_lock(&sbi->gc_mutex);
2456 err = write_checkpoint(sbi, &cpc);
2457 mutex_unlock(&sbi->gc_mutex);
2458 if (err)
2459 goto out;
8412663d
CY		 2460
2461 start_block = START_BLOCK(sbi, start_segno);
377224c4 2462 end_block = START_BLOCK(sbi, end_segno + 1);
8412663d 2463
8bb4f253 2464 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
78997b56 2465 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
e555da9f
JK		 2466
2467 /*
2468 * We filed discard candidates, but actually we don't need to wait for
2469 * all of them, since they'll be issued in idle time along with runtime
2470 * discard option. User configuration looks like using runtime discard
2471 * or periodic fstrim instead of it.
2472 */
2473 if (!test_opt(sbi, DISCARD)) {
2474 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 2475 start_block, end_block);
e555da9f
JK		 2476 range->len = F2FS_BLK_TO_BYTES(trimmed);
2477 }
377224c4 2478out:
c34f42e2 2479 return err;
4b2fecc8
JK		 2480}
2481
351df4b2
JK		 2482static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2483{
2484 struct curseg_info *curseg = CURSEG_I(sbi, type);
2485 if (curseg->next_blkoff < sbi->blocks_per_seg)
2486 return true;
2487 return false;
2488}
2489
4f0a03d3
HL		 2490int rw_hint_to_seg_type(enum rw_hint hint)
2491{
2492 switch (hint) {
2493 case WRITE_LIFE_SHORT:
2494 return CURSEG_HOT_DATA;
2495 case WRITE_LIFE_EXTREME:
2496 return CURSEG_COLD_DATA;
2497 default:
2498 return CURSEG_WARM_DATA;
2499 }
2500}
2501
0cdd3195
HL		 2502/* This returns write hints for each segment type. This hints will be
2503 * passed down to block layer. There are mapping tables which depend on
2504 * the mount option 'whint_mode'.
2505 *
2506 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2507 *
2508 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2509 *
2510 * User F2FS Block
2511 * ---- ---- -----
2512 * META WRITE_LIFE_NOT_SET
2513 * HOT_NODE "
2514 * WARM_NODE "
2515 * COLD_NODE "
2516 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2517 * extension list " "
2518 *
2519 * -- buffered io
2520 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2521 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2522 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2523 * WRITE_LIFE_NONE " "
2524 * WRITE_LIFE_MEDIUM " "
2525 * WRITE_LIFE_LONG " "
2526 *
2527 * -- direct io
2528 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2529 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2530 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2531 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2532 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2533 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2534 *
f2e703f9
HL		 2535 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2536 *
2537 * User F2FS Block
2538 * ---- ---- -----
2539 * META WRITE_LIFE_MEDIUM;
2540 * HOT_NODE WRITE_LIFE_NOT_SET
2541 * WARM_NODE "
2542 * COLD_NODE WRITE_LIFE_NONE
2543 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2544 * extension list " "
2545 *
2546 * -- buffered io
2547 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2548 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2549 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2550 * WRITE_LIFE_NONE " "
2551 * WRITE_LIFE_MEDIUM " "
2552 * WRITE_LIFE_LONG " "
2553 *
2554 * -- direct io
2555 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2556 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2557 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2558 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2559 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2560 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 2561 */
2562
2563enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2564 enum page_type type, enum temp_type temp)
2565{
63189b78 2566 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 2567 if (type == DATA) {
f2e703f9 2568 if (temp == WARM)
0cdd3195 2569 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 2570 else if (temp == HOT)
2571 return WRITE_LIFE_SHORT;
2572 else if (temp == COLD)
2573 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 2574 } else {
2575 return WRITE_LIFE_NOT_SET;
2576 }
63189b78 2577 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 2578 if (type == DATA) {
2579 if (temp == WARM)
2580 return WRITE_LIFE_LONG;
2581 else if (temp == HOT)
2582 return WRITE_LIFE_SHORT;
2583 else if (temp == COLD)
2584 return WRITE_LIFE_EXTREME;
2585 } else if (type == NODE) {
2586 if (temp == WARM || temp == HOT)
2587 return WRITE_LIFE_NOT_SET;
2588 else if (temp == COLD)
2589 return WRITE_LIFE_NONE;
2590 } else if (type == META) {
2591 return WRITE_LIFE_MEDIUM;
2592 }
0cdd3195 2593 }
f2e703f9 2594 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 2595}
2596
81377bd6 2597static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2598{
81377bd6 2599 if (fio->type == DATA)
351df4b2
JK		 2600 return CURSEG_HOT_DATA;
2601 else
2602 return CURSEG_HOT_NODE;
2603}
2604
81377bd6 2605static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2606{
81377bd6
JK		 2607 if (fio->type == DATA) {
2608 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2609
2610 if (S_ISDIR(inode->i_mode))
2611 return CURSEG_HOT_DATA;
2612 else
2613 return CURSEG_COLD_DATA;
2614 } else {
81377bd6 2615 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2616 return CURSEG_WARM_NODE;
351df4b2
JK		 2617 else
2618 return CURSEG_COLD_NODE;
2619 }
2620}
2621
81377bd6 2622static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2623{
81377bd6
JK		 2624 if (fio->type == DATA) {
2625 struct inode *inode = fio->page->mapping->host;
351df4b2 2626
81377bd6 2627 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2628 return CURSEG_COLD_DATA;
b6a06cbb 2629 if (file_is_hot(inode) ||
b4c3ca8b
CY		 2630 is_inode_flag_set(inode, FI_HOT_DATA) ||
2631 is_inode_flag_set(inode, FI_ATOMIC_FILE) ||
2632 is_inode_flag_set(inode, FI_VOLATILE_FILE))
ef095d19 2633 return CURSEG_HOT_DATA;
4f0a03d3 2634 return rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 2635 } else {
81377bd6
JK		 2636 if (IS_DNODE(fio->page))
2637 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2638 CURSEG_HOT_NODE;
ef095d19 2639 return CURSEG_COLD_NODE;
351df4b2
JK		 2640 }
2641}
2642
81377bd6 2643static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2644{
a912b54d
JK		 2645 int type = 0;
2646
63189b78 2647 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 2648 case 2:
a912b54d
JK		 2649 type = __get_segment_type_2(fio);
2650 break;
351df4b2 2651 case 4:
a912b54d
JK		 2652 type = __get_segment_type_4(fio);
2653 break;
2654 case 6:
2655 type = __get_segment_type_6(fio);
2656 break;
2657 default:
2658 f2fs_bug_on(fio->sbi, true);
351df4b2 2659 }
81377bd6 2660
a912b54d
JK		 2661 if (IS_HOT(type))
2662 fio->temp = HOT;
2663 else if (IS_WARM(type))
2664 fio->temp = WARM;
2665 else
2666 fio->temp = COLD;
2667 return type;
351df4b2
JK		 2668}
2669
bfad7c2d
JK		 2670void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2671 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2672 struct f2fs_summary *sum, int type,
2673 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2674{
2675 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2676 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 2677
2b60311d
CY		 2678 down_read(&SM_I(sbi)->curseg_lock);
2679
351df4b2 2680 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2681 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2682
2683 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2684
4e6a8d9b
JK		 2685 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2686
351df4b2
JK		 2687 /*
2688 * __add_sum_entry should be resided under the curseg_mutex
2689 * because, this function updates a summary entry in the
2690 * current summary block.
2691 */
e79efe3b 2692 __add_sum_entry(sbi, type, sum);
351df4b2 2693
351df4b2 2694 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2695
2696 stat_inc_block_count(sbi, curseg);
351df4b2 2697
65f1b80b
YS		 2698 /*
2699 * SIT information should be updated before segment allocation,
2700 * since SSR needs latest valid block information.
2701 */
2702 update_sit_entry(sbi, *new_blkaddr, 1);
2703 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2704 update_sit_entry(sbi, old_blkaddr, -1);
2705
c6f82fe9
JK		 2706 if (!__has_curseg_space(sbi, type))
2707 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 2708
351df4b2 2709 /*
65f1b80b
YS		 2710 * segment dirty status should be updated after segment allocation,
2711 * so we just need to update status only one time after previous
2712 * segment being closed.
351df4b2 2713 */
65f1b80b
YS		 2714 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2715 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 2716
3d26fa6b 2717 up_write(&sit_i->sentry_lock);
351df4b2 2718
704956ec 2719 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2720 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2721
704956ec
CY		 2722 f2fs_inode_chksum_set(sbi, page);
2723 }
2724
fb830fc5
CY		 2725 if (add_list) {
2726 struct f2fs_bio_info *io;
2727
2728 INIT_LIST_HEAD(&fio->list);
2729 fio->in_list = true;
2730 io = sbi->write_io[fio->type] + fio->temp;
2731 spin_lock(&io->io_lock);
2732 list_add_tail(&fio->list, &io->io_list);
2733 spin_unlock(&io->io_lock);
2734 }
2735
bfad7c2d 2736 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 2737
2738 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 2739}
2740
39d787be
CY		 2741static void update_device_state(struct f2fs_io_info *fio)
2742{
2743 struct f2fs_sb_info *sbi = fio->sbi;
2744 unsigned int devidx;
2745
2746 if (!sbi->s_ndevs)
2747 return;
2748
2749 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2750
2751 /* update device state for fsync */
2752 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 2753
2754 /* update device state for checkpoint */
2755 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2756 spin_lock(&sbi->dev_lock);
2757 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2758 spin_unlock(&sbi->dev_lock);
2759 }
39d787be
CY		 2760}
2761
05ca3632 2762static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2763{
81377bd6 2764 int type = __get_segment_type(fio);
0a595eba 2765 int err;
107a805d 2766 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
bfad7c2d 2767
107a805d
CY		 2768 if (keep_order)
2769 down_read(&fio->sbi->io_order_lock);
0a595eba 2770reallocate:
7a9d7548 2771 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2772 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2773
351df4b2 2774 /* writeout dirty page into bdev */
b9109b0e 2775 err = f2fs_submit_page_write(fio);
0a595eba
JK		 2776 if (err == -EAGAIN) {
2777 fio->old_blkaddr = fio->new_blkaddr;
2778 goto reallocate;
39d787be
CY		 2779 } else if (!err) {
2780 update_device_state(fio);
0a595eba 2781 }
107a805d
CY		 2782 if (keep_order)
2783 up_read(&fio->sbi->io_order_lock);
351df4b2
JK		 2784}
2785
b0af6d49
CY		 2786void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2787 enum iostat_type io_type)
351df4b2 2788{
458e6197 2789 struct f2fs_io_info fio = {
05ca3632 2790 .sbi = sbi,
458e6197 2791 .type = META,
0cdd3195 2792 .temp = HOT,
04d328de 2793 .op = REQ_OP_WRITE,
70fd7614 2794 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2795 .old_blkaddr = page->index,
2796 .new_blkaddr = page->index,
05ca3632 2797 .page = page,
4375a336 2798 .encrypted_page = NULL,
fb830fc5 2799 .in_list = false,
458e6197
JK		 2800 };
2801
2b947003 2802 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2803 fio.op_flags &= ~REQ_META;
2b947003 2804
351df4b2 2805 set_page_writeback(page);
17c50035 2806 ClearPageError(page);
b9109b0e 2807 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2808
2809 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2810}
2811
05ca3632 2812void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2813{
2814 struct f2fs_summary sum;
05ca3632 2815
351df4b2 2816 set_summary(&sum, nid, 0, 0);
05ca3632 2817 do_write_page(&sum, fio);
b0af6d49
CY		 2818
2819 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2820}
2821
05ca3632 2822void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2823{
05ca3632 2824 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2825 struct f2fs_summary sum;
2826 struct node_info ni;
2827
9850cf4a 2828 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2829 get_node_info(sbi, dn->nid, &ni);
2830 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2831 do_write_page(&sum, fio);
f28b3434 2832 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2833
2834 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2835}
2836
d1b3e72d 2837int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2838{
b0af6d49 2839 int err;
d21b0f23 2840 struct f2fs_sb_info *sbi = fio->sbi;
b0af6d49 2841
7a9d7548 2842 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 2843 /* i/o temperature is needed for passing down write hints */
2844 __get_segment_type(fio);
d21b0f23
YH		 2845
2846 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2847 GET_SEGNO(sbi, fio->new_blkaddr))->type));
2848
05ca3632 2849 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2850
2851 err = f2fs_submit_page_bio(fio);
39d787be
CY		 2852 if (!err)
2853 update_device_state(fio);
b0af6d49
CY		 2854
2855 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2856
2857 return err;
351df4b2
JK		 2858}
2859
2b60311d
CY		 2860static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2861 unsigned int segno)
2862{
2863 int i;
2864
2865 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2866 if (CURSEG_I(sbi, i)->segno == segno)
2867 break;
2868 }
2869 return i;
2870}
2871
4356e48e 2872void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2873 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2874 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2875{
2876 struct sit_info *sit_i = SIT_I(sbi);
2877 struct curseg_info *curseg;
2878 unsigned int segno, old_cursegno;
2879 struct seg_entry *se;
2880 int type;
19f106bc 2881 unsigned short old_blkoff;
351df4b2
JK		 2882
2883 segno = GET_SEGNO(sbi, new_blkaddr);
2884 se = get_seg_entry(sbi, segno);
2885 type = se->type;
2886
2b60311d
CY		 2887 down_write(&SM_I(sbi)->curseg_lock);
2888
19f106bc
CY		 2889 if (!recover_curseg) {
2890 /* for recovery flow */
2891 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2892 if (old_blkaddr == NULL_ADDR)
2893 type = CURSEG_COLD_DATA;
2894 else
2895 type = CURSEG_WARM_DATA;
2896 }
2897 } else {
2b60311d
CY		 2898 if (IS_CURSEG(sbi, segno)) {
2899 /* se->type is volatile as SSR allocation */
2900 type = __f2fs_get_curseg(sbi, segno);
2901 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2902 } else {
351df4b2 2903 type = CURSEG_WARM_DATA;
2b60311d 2904 }
351df4b2 2905 }
19f106bc 2906
2c190504 2907 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 2908 curseg = CURSEG_I(sbi, type);
2909
2910 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2911 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2912
2913 old_cursegno = curseg->segno;
19f106bc 2914 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2915
2916 /* change the current segment */
2917 if (segno != curseg->segno) {
2918 curseg->next_segno = segno;
025d63a4 2919 change_curseg(sbi, type);
351df4b2
JK		 2920 }
2921
491c0854 2922 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2923 __add_sum_entry(sbi, type, sum);
351df4b2 2924
28bc106b 2925 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2926 update_sit_entry(sbi, new_blkaddr, 1);
2927 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2928 update_sit_entry(sbi, old_blkaddr, -1);
2929
2930 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2931 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2932
351df4b2 2933 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2934
19f106bc
CY		 2935 if (recover_curseg) {
2936 if (old_cursegno != curseg->segno) {
2937 curseg->next_segno = old_cursegno;
025d63a4 2938 change_curseg(sbi, type);
19f106bc
CY		 2939 }
2940 curseg->next_blkoff = old_blkoff;
2941 }
2942
3d26fa6b 2943 up_write(&sit_i->sentry_lock);
351df4b2 2944 mutex_unlock(&curseg->curseg_mutex);
2b60311d 2945 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 2946}
2947
528e3459
CY		 2948void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2949 block_t old_addr, block_t new_addr,
28bc106b
CY		 2950 unsigned char version, bool recover_curseg,
2951 bool recover_newaddr)
528e3459
CY		 2952{
2953 struct f2fs_summary sum;
2954
2955 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2956
28bc106b
CY		 2957 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2958 recover_curseg, recover_newaddr);
528e3459 2959
f28b3434 2960 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2961}
2962
93dfe2ac 2963void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2964 enum page_type type, bool ordered)
93dfe2ac 2965{
93dfe2ac 2966 if (PageWriteback(page)) {
4081363f
JK		 2967 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2968
b9109b0e
JK		 2969 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2970 0, page->index, type);
fec1d657
JK		 2971 if (ordered)
2972 wait_on_page_writeback(page);
2973 else
2974 wait_for_stable_page(page);
93dfe2ac
JK		 2975 }
2976}
2977
d4c759ee 2978void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 2979{
2980 struct page *cpage;
2981
7b525dd0 2982 if (!is_valid_blkaddr(blkaddr))
08b39fbd
CY		 2983 return;
2984
08b39fbd
CY		 2985 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2986 if (cpage) {
fec1d657 2987 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2988 f2fs_put_page(cpage, 1);
2989 }
2990}
2991
c376fc0f 2992static void read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 2993{
2994 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2995 struct curseg_info *seg_i;
2996 unsigned char *kaddr;
2997 struct page *page;
2998 block_t start;
2999 int i, j, offset;
3000
3001 start = start_sum_block(sbi);
3002
3003 page = get_meta_page(sbi, start++);
3004 kaddr = (unsigned char *)page_address(page);
3005
3006 /* Step 1: restore nat cache */
3007 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3008 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 3009
3010 /* Step 2: restore sit cache */
3011 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3012 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 3013 offset = 2 * SUM_JOURNAL_SIZE;
3014
3015 /* Step 3: restore summary entries */
3016 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3017 unsigned short blk_off;
3018 unsigned int segno;
3019
3020 seg_i = CURSEG_I(sbi, i);
3021 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3022 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3023 seg_i->next_segno = segno;
3024 reset_curseg(sbi, i, 0);
3025 seg_i->alloc_type = ckpt->alloc_type[i];
3026 seg_i->next_blkoff = blk_off;
3027
3028 if (seg_i->alloc_type == SSR)
3029 blk_off = sbi->blocks_per_seg;
3030
3031 for (j = 0; j < blk_off; j++) {
3032 struct f2fs_summary *s;
3033 s = (struct f2fs_summary *)(kaddr + offset);
3034 seg_i->sum_blk->entries[j] = *s;
3035 offset += SUMMARY_SIZE;
09cbfeaf 3036 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3037 SUM_FOOTER_SIZE)
3038 continue;
3039
3040 f2fs_put_page(page, 1);
3041 page = NULL;
3042
3043 page = get_meta_page(sbi, start++);
3044 kaddr = (unsigned char *)page_address(page);
3045 offset = 0;
3046 }
3047 }
3048 f2fs_put_page(page, 1);
351df4b2
JK		 3049}
3050
3051static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3052{
3053 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3054 struct f2fs_summary_block *sum;
3055 struct curseg_info *curseg;
3056 struct page *new;
3057 unsigned short blk_off;
3058 unsigned int segno = 0;
3059 block_t blk_addr = 0;
3060
3061 /* get segment number and block addr */
3062 if (IS_DATASEG(type)) {
3063 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3064 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3065 CURSEG_HOT_DATA]);
119ee914 3066 if (__exist_node_summaries(sbi))
351df4b2
JK		 3067 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3068 else
3069 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3070 } else {
3071 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3072 CURSEG_HOT_NODE]);
3073 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3074 CURSEG_HOT_NODE]);
119ee914 3075 if (__exist_node_summaries(sbi))
351df4b2
JK		 3076 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3077 type - CURSEG_HOT_NODE);
3078 else
3079 blk_addr = GET_SUM_BLOCK(sbi, segno);
3080 }
3081
3082 new = get_meta_page(sbi, blk_addr);
3083 sum = (struct f2fs_summary_block *)page_address(new);
3084
3085 if (IS_NODESEG(type)) {
119ee914 3086 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3087 struct f2fs_summary *ns = &sum->entries[0];
3088 int i;
3089 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3090 ns->version = 0;
3091 ns->ofs_in_node = 0;
3092 }
3093 } else {
c376fc0f 3094 restore_node_summary(sbi, segno, sum);
351df4b2
JK		 3095 }
3096 }
3097
3098 /* set uncompleted segment to curseg */
3099 curseg = CURSEG_I(sbi, type);
3100 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3101
3102 /* update journal info */
3103 down_write(&curseg->journal_rwsem);
3104 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3105 up_write(&curseg->journal_rwsem);
3106
3107 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3108 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3109 curseg->next_segno = segno;
3110 reset_curseg(sbi, type, 0);
3111 curseg->alloc_type = ckpt->alloc_type[type];
3112 curseg->next_blkoff = blk_off;
3113 mutex_unlock(&curseg->curseg_mutex);
3114 f2fs_put_page(new, 1);
3115 return 0;
3116}
3117
3118static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3119{
21d3f8e1
JQ		 3120 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3121 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3122 int type = CURSEG_HOT_DATA;
e4fc5fbf 3123 int err;
351df4b2 3124
aaec2b1d 3125 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 3126 int npages = npages_for_summary_flush(sbi, true);
3127
3128 if (npages >= 2)
3129 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3130 META_CP, true);
3fa06d7b 3131
351df4b2 3132 /* restore for compacted data summary */
c376fc0f 3133 read_compacted_summaries(sbi);
351df4b2
JK		 3134 type = CURSEG_HOT_NODE;
3135 }
3136
119ee914 3137 if (__exist_node_summaries(sbi))
3fa06d7b 3138 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3139 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3140
e4fc5fbf
CY		 3141 for (; type <= CURSEG_COLD_NODE; type++) {
3142 err = read_normal_summaries(sbi, type);
3143 if (err)
3144 return err;
3145 }
3146
21d3f8e1
JQ		 3147 /* sanity check for summary blocks */
3148 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3149 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3150 return -EINVAL;
3151
351df4b2
JK		 3152 return 0;
3153}
3154
3155static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3156{
3157 struct page *page;
3158 unsigned char *kaddr;
3159 struct f2fs_summary *summary;
3160 struct curseg_info *seg_i;
3161 int written_size = 0;
3162 int i, j;
3163
3164 page = grab_meta_page(sbi, blkaddr++);
3165 kaddr = (unsigned char *)page_address(page);
81114baa 3166 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3167
3168 /* Step 1: write nat cache */
3169 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3170 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3171 written_size += SUM_JOURNAL_SIZE;
3172
3173 /* Step 2: write sit cache */
3174 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3175 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3176 written_size += SUM_JOURNAL_SIZE;
3177
351df4b2
JK		 3178 /* Step 3: write summary entries */
3179 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3180 unsigned short blkoff;
3181 seg_i = CURSEG_I(sbi, i);
3182 if (sbi->ckpt->alloc_type[i] == SSR)
3183 blkoff = sbi->blocks_per_seg;
3184 else
3185 blkoff = curseg_blkoff(sbi, i);
3186
3187 for (j = 0; j < blkoff; j++) {
3188 if (!page) {
3189 page = grab_meta_page(sbi, blkaddr++);
3190 kaddr = (unsigned char *)page_address(page);
81114baa 3191 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3192 written_size = 0;
3193 }
3194 summary = (struct f2fs_summary *)(kaddr + written_size);
3195 *summary = seg_i->sum_blk->entries[j];
3196 written_size += SUMMARY_SIZE;
351df4b2 3197
09cbfeaf 3198 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3199 SUM_FOOTER_SIZE)
3200 continue;
3201
e8d61a74 3202 set_page_dirty(page);
351df4b2
JK		 3203 f2fs_put_page(page, 1);
3204 page = NULL;
3205 }
3206 }
e8d61a74
CY		 3207 if (page) {
3208 set_page_dirty(page);
351df4b2 3209 f2fs_put_page(page, 1);
e8d61a74 3210 }
351df4b2
JK		 3211}
3212
3213static void write_normal_summaries(struct f2fs_sb_info *sbi,
3214 block_t blkaddr, int type)
3215{
3216 int i, end;
3217 if (IS_DATASEG(type))
3218 end = type + NR_CURSEG_DATA_TYPE;
3219 else
3220 end = type + NR_CURSEG_NODE_TYPE;
3221
b7ad7512
CY		 3222 for (i = type; i < end; i++)
3223 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3224}
3225
3226void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3227{
aaec2b1d 3228 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3229 write_compacted_summaries(sbi, start_blk);
3230 else
3231 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3232}
3233
3234void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3235{
119ee914 3236 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3237}
3238
dfc08a12 3239int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3240 unsigned int val, int alloc)
3241{
3242 int i;
3243
3244 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3245 for (i = 0; i < nats_in_cursum(journal); i++) {
3246 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3247 return i;
3248 }
dfc08a12
CY		 3249 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3250 return update_nats_in_cursum(journal, 1);
351df4b2 3251 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3252 for (i = 0; i < sits_in_cursum(journal); i++)
3253 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3254 return i;
dfc08a12
CY		 3255 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3256 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3257 }
3258 return -1;
3259}
3260
3261static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3262 unsigned int segno)
3263{
2cc22186 3264 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3265}
3266
3267static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3268 unsigned int start)
3269{
3270 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 3271 struct page *page;
351df4b2 3272 pgoff_t src_off, dst_off;
351df4b2
JK		 3273
3274 src_off = current_sit_addr(sbi, start);
3275 dst_off = next_sit_addr(sbi, src_off);
3276
068c3cd8
YH		 3277 page = grab_meta_page(sbi, dst_off);
3278 seg_info_to_sit_page(sbi, page, start);
351df4b2 3279
068c3cd8 3280 set_page_dirty(page);
351df4b2
JK		 3281 set_to_next_sit(sit_i, start);
3282
068c3cd8 3283 return page;
351df4b2
JK		 3284}
3285
184a5cd2
CY		 3286static struct sit_entry_set *grab_sit_entry_set(void)
3287{
3288 struct sit_entry_set *ses =
80c54505 3289 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3290
3291 ses->entry_cnt = 0;
3292 INIT_LIST_HEAD(&ses->set_list);
3293 return ses;
3294}
3295
3296static void release_sit_entry_set(struct sit_entry_set *ses)
3297{
3298 list_del(&ses->set_list);
3299 kmem_cache_free(sit_entry_set_slab, ses);
3300}
3301
3302static void adjust_sit_entry_set(struct sit_entry_set *ses,
3303 struct list_head *head)
3304{
3305 struct sit_entry_set *next = ses;
3306
3307 if (list_is_last(&ses->set_list, head))
3308 return;
3309
3310 list_for_each_entry_continue(next, head, set_list)
3311 if (ses->entry_cnt <= next->entry_cnt)
3312 break;
3313
3314 list_move_tail(&ses->set_list, &next->set_list);
3315}
3316
3317static void add_sit_entry(unsigned int segno, struct list_head *head)
3318{
3319 struct sit_entry_set *ses;
3320 unsigned int start_segno = START_SEGNO(segno);
3321
3322 list_for_each_entry(ses, head, set_list) {
3323 if (ses->start_segno == start_segno) {
3324 ses->entry_cnt++;
3325 adjust_sit_entry_set(ses, head);
3326 return;
3327 }
3328 }
3329
3330 ses = grab_sit_entry_set();
3331
3332 ses->start_segno = start_segno;
3333 ses->entry_cnt++;
3334 list_add(&ses->set_list, head);
3335}
3336
3337static void add_sits_in_set(struct f2fs_sb_info *sbi)
3338{
3339 struct f2fs_sm_info *sm_info = SM_I(sbi);
3340 struct list_head *set_list = &sm_info->sit_entry_set;
3341 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3342 unsigned int segno;
3343
7cd8558b 3344 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3345 add_sit_entry(segno, set_list);
3346}
3347
3348static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3349{
3350 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3351 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3352 int i;
3353
b7ad7512 3354 down_write(&curseg->journal_rwsem);
dfc08a12 3355 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3356 unsigned int segno;
3357 bool dirtied;
3358
dfc08a12 3359 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3360 dirtied = __mark_sit_entry_dirty(sbi, segno);
3361
3362 if (!dirtied)
3363 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3364 }
dfc08a12 3365 update_sits_in_cursum(journal, -i);
b7ad7512 3366 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3367}
3368
0a8165d7 3369/*
351df4b2
JK		 3370 * CP calls this function, which flushes SIT entries including sit_journal,
3371 * and moves prefree segs to free segs.
3372 */
4b2fecc8 3373void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3374{
3375 struct sit_info *sit_i = SIT_I(sbi);
3376 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3377 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3378 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3379 struct sit_entry_set *ses, *tmp;
3380 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3381 bool to_journal = true;
4b2fecc8 3382 struct seg_entry *se;
351df4b2 3383
3d26fa6b 3384 down_write(&sit_i->sentry_lock);
351df4b2 3385
2b11a74b
WL		 3386 if (!sit_i->dirty_sentries)
3387 goto out;
3388
351df4b2 3389 /*
184a5cd2
CY		 3390 * add and account sit entries of dirty bitmap in sit entry
3391 * set temporarily
351df4b2 3392 */
184a5cd2 3393 add_sits_in_set(sbi);
351df4b2 3394
184a5cd2
CY		 3395 /*
3396 * if there are no enough space in journal to store dirty sit
3397 * entries, remove all entries from journal and add and account
3398 * them in sit entry set.
3399 */
dfc08a12 3400 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3401 remove_sits_in_journal(sbi);
b2955550 3402
184a5cd2
CY		 3403 /*
3404 * there are two steps to flush sit entries:
3405 * #1, flush sit entries to journal in current cold data summary block.
3406 * #2, flush sit entries to sit page.
3407 */
3408 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3409 struct page *page = NULL;
184a5cd2
CY		 3410 struct f2fs_sit_block *raw_sit = NULL;
3411 unsigned int start_segno = ses->start_segno;
3412 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3413 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3414 unsigned int segno = start_segno;
3415
3416 if (to_journal &&
dfc08a12 3417 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3418 to_journal = false;
3419
b7ad7512
CY		 3420 if (to_journal) {
3421 down_write(&curseg->journal_rwsem);
3422 } else {
184a5cd2
CY		 3423 page = get_next_sit_page(sbi, start_segno);
3424 raw_sit = page_address(page);
351df4b2 3425 }
351df4b2 3426
184a5cd2
CY		 3427 /* flush dirty sit entries in region of current sit set */
3428 for_each_set_bit_from(segno, bitmap, end) {
3429 int offset, sit_offset;
4b2fecc8
JK		 3430
3431 se = get_seg_entry(sbi, segno);
56b07e7e
ZZ		 3432#ifdef CONFIG_F2FS_CHECK_FS
3433 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3434 SIT_VBLOCK_MAP_SIZE))
3435 f2fs_bug_on(sbi, 1);
3436#endif
184a5cd2
CY		 3437
3438 /* add discard candidates */
c473f1a9 3439 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3440 cpc->trim_start = segno;
25290fa5 3441 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3442 }
184a5cd2
CY		 3443
3444 if (to_journal) {
dfc08a12 3445 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 3446 SIT_JOURNAL, segno, 1);
3447 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3448 segno_in_journal(journal, offset) =
184a5cd2
CY		 3449 cpu_to_le32(segno);
3450 seg_info_to_raw_sit(se,
dfc08a12 3451 &sit_in_journal(journal, offset));
56b07e7e
ZZ		 3452 check_block_count(sbi, segno,
3453 &sit_in_journal(journal, offset));
184a5cd2
CY		 3454 } else {
3455 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3456 seg_info_to_raw_sit(se,
3457 &raw_sit->entries[sit_offset]);
56b07e7e
ZZ		 3458 check_block_count(sbi, segno,
3459 &raw_sit->entries[sit_offset]);
184a5cd2 3460 }
351df4b2 3461
184a5cd2
CY		 3462 __clear_bit(segno, bitmap);
3463 sit_i->dirty_sentries--;
3464 ses->entry_cnt--;
351df4b2
JK		 3465 }
3466
b7ad7512
CY		 3467 if (to_journal)
3468 up_write(&curseg->journal_rwsem);
3469 else
184a5cd2
CY		 3470 f2fs_put_page(page, 1);
3471
3472 f2fs_bug_on(sbi, ses->entry_cnt);
3473 release_sit_entry_set(ses);
351df4b2 3474 }
184a5cd2
CY		 3475
3476 f2fs_bug_on(sbi, !list_empty(head));
3477 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3478out:
c473f1a9 3479 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3480 __u64 trim_start = cpc->trim_start;
3481
4b2fecc8 3482 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3483 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3484
3485 cpc->trim_start = trim_start;
4b2fecc8 3486 }
3d26fa6b 3487 up_write(&sit_i->sentry_lock);
351df4b2 3488
351df4b2
JK		 3489 set_prefree_as_free_segments(sbi);
3490}
3491
3492static int build_sit_info(struct f2fs_sb_info *sbi)
3493{
3494 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3495 struct sit_info *sit_i;
3496 unsigned int sit_segs, start;
ae27d62e 3497 char *src_bitmap;
351df4b2
JK		 3498 unsigned int bitmap_size;
3499
3500 /* allocate memory for SIT information */
acbf054d 3501 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 3502 if (!sit_i)
3503 return -ENOMEM;
3504
3505 SM_I(sbi)->sit_info = sit_i;
3506
628b3d14 3507 sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
39307a8e 3508 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3509 if (!sit_i->sentries)
3510 return -ENOMEM;
3511
7cd8558b 3512 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14
CY		 3513 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3514 GFP_KERNEL);
351df4b2
JK		 3515 if (!sit_i->dirty_sentries_bitmap)
3516 return -ENOMEM;
3517
7cd8558b 3518 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3519 sit_i->sentries[start].cur_valid_map
acbf054d 3520 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
351df4b2 3521 sit_i->sentries[start].ckpt_valid_map
acbf054d 3522 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3523 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3524 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3525 return -ENOMEM;
3e025740 3526
355e7891
CY		 3527#ifdef CONFIG_F2FS_CHECK_FS
3528 sit_i->sentries[start].cur_valid_map_mir
acbf054d 3529 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
355e7891
CY		 3530 if (!sit_i->sentries[start].cur_valid_map_mir)
3531 return -ENOMEM;
3532#endif
3533
3e025740
JK		 3534 if (f2fs_discard_en(sbi)) {
3535 sit_i->sentries[start].discard_map
acbf054d
CY		 3536 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3537 GFP_KERNEL);
3e025740
JK		 3538 if (!sit_i->sentries[start].discard_map)
3539 return -ENOMEM;
3540 }
351df4b2
JK		 3541 }
3542
acbf054d 3543 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 3544 if (!sit_i->tmp_map)
3545 return -ENOMEM;
3546
351df4b2 3547 if (sbi->segs_per_sec > 1) {
628b3d14 3548 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
39307a8e 3549 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3550 if (!sit_i->sec_entries)
3551 return -ENOMEM;
3552 }
3553
3554 /* get information related with SIT */
3555 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3556
3557 /* setup SIT bitmap from ckeckpoint pack */
3558 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3559 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3560
ae27d62e
CY		 3561 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3562 if (!sit_i->sit_bitmap)
351df4b2 3563 return -ENOMEM;
351df4b2 3564
ae27d62e
CY		 3565#ifdef CONFIG_F2FS_CHECK_FS
3566 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3567 if (!sit_i->sit_bitmap_mir)
3568 return -ENOMEM;
3569#endif
3570
351df4b2
JK		 3571 /* init SIT information */
3572 sit_i->s_ops = &default_salloc_ops;
3573
3574 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3575 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3576 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3577 sit_i->bitmap_size = bitmap_size;
3578 sit_i->dirty_sentries = 0;
3579 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3580 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3581 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 3582 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 3583 return 0;
3584}
3585
3586static int build_free_segmap(struct f2fs_sb_info *sbi)
3587{
351df4b2
JK		 3588 struct free_segmap_info *free_i;
3589 unsigned int bitmap_size, sec_bitmap_size;
3590
3591 /* allocate memory for free segmap information */
acbf054d 3592 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 3593 if (!free_i)
3594 return -ENOMEM;
3595
3596 SM_I(sbi)->free_info = free_i;
3597
7cd8558b 3598 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 3599 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 3600 if (!free_i->free_segmap)
3601 return -ENOMEM;
3602
7cd8558b 3603 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 3604 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3605 if (!free_i->free_secmap)
3606 return -ENOMEM;
3607
3608 /* set all segments as dirty temporarily */
3609 memset(free_i->free_segmap, 0xff, bitmap_size);
3610 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3611
3612 /* init free segmap information */
7cd8558b 3613 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3614 free_i->free_segments = 0;
3615 free_i->free_sections = 0;
1a118ccf 3616 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3617 return 0;
3618}
3619
3620static int build_curseg(struct f2fs_sb_info *sbi)
3621{
1042d60f 3622 struct curseg_info *array;
351df4b2
JK		 3623 int i;
3624
4e6aad29 3625 array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
351df4b2
JK		 3626 if (!array)
3627 return -ENOMEM;
3628
3629 SM_I(sbi)->curseg_array = array;
3630
3631 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3632 mutex_init(&array[i].curseg_mutex);
acbf054d 3633 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3634 if (!array[i].sum_blk)
3635 return -ENOMEM;
b7ad7512 3636 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 3637 array[i].journal = f2fs_kzalloc(sbi,
3638 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 3639 if (!array[i].journal)
3640 return -ENOMEM;
351df4b2
JK		 3641 array[i].segno = NULL_SEGNO;
3642 array[i].next_blkoff = 0;
3643 }
3644 return restore_curseg_summaries(sbi);
3645}
3646
c39a1b34 3647static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3648{
3649 struct sit_info *sit_i = SIT_I(sbi);
3650 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3651 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3652 struct seg_entry *se;
3653 struct f2fs_sit_entry sit;
74de593a
CY		 3654 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3655 unsigned int i, start, end;
3656 unsigned int readed, start_blk = 0;
c39a1b34 3657 int err = 0;
8a29c126 3658 block_t total_node_blocks = 0;
351df4b2 3659
74de593a 3660 do {
664ba972
JK		 3661 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3662 META_SIT, true);
74de593a
CY		 3663
3664 start = start_blk * sit_i->sents_per_block;
3665 end = (start_blk + readed) * sit_i->sents_per_block;
3666
7cd8558b 3667 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3668 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3669 struct page *page;
3670
9c094040 3671 se = &sit_i->sentries[start];
74de593a
CY		 3672 page = get_current_sit_page(sbi, start);
3673 sit_blk = (struct f2fs_sit_block *)page_address(page);
3674 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3675 f2fs_put_page(page, 1);
d600af23 3676
c39a1b34
JK		 3677 err = check_block_count(sbi, start, &sit);
3678 if (err)
3679 return err;
74de593a 3680 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 3681 if (IS_NODESEG(se->type))
3682 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 3683
3684 /* build discard map only one time */
3e025740 3685 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3686 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3687 memset(se->discard_map, 0xff,
3688 SIT_VBLOCK_MAP_SIZE);
3689 } else {
3690 memcpy(se->discard_map,
3691 se->cur_valid_map,
3692 SIT_VBLOCK_MAP_SIZE);
3693 sbi->discard_blks +=
3694 sbi->blocks_per_seg -
3695 se->valid_blocks;
3696 }
3e025740 3697 }
a66cdd98 3698
d600af23
CY		 3699 if (sbi->segs_per_sec > 1)
3700 get_sec_entry(sbi, start)->valid_blocks +=
3701 se->valid_blocks;
351df4b2 3702 }
74de593a
CY		 3703 start_blk += readed;
3704 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3705
3706 down_read(&curseg->journal_rwsem);
3707 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3708 unsigned int old_valid_blocks;
3709
3710 start = le32_to_cpu(segno_in_journal(journal, i));
b2ca374f
JK		 3711 if (start >= MAIN_SEGS(sbi)) {
3712 f2fs_msg(sbi->sb, KERN_ERR,
3713 "Wrong journal entry on segno %u",
3714 start);
3715 set_sbi_flag(sbi, SBI_NEED_FSCK);
3716 err = -EINVAL;
3717 break;
3718 }
3719
d600af23
CY		 3720 se = &sit_i->sentries[start];
3721 sit = sit_in_journal(journal, i);
3722
3723 old_valid_blocks = se->valid_blocks;
8a29c126
JK		 3724 if (IS_NODESEG(se->type))
3725 total_node_blocks -= old_valid_blocks;
d600af23 3726
c39a1b34
JK		 3727 err = check_block_count(sbi, start, &sit);
3728 if (err)
3729 break;
d600af23 3730 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 3731 if (IS_NODESEG(se->type))
3732 total_node_blocks += se->valid_blocks;
d600af23
CY		 3733
3734 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3735 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3736 memset(se->discard_map, 0xff,
3737 SIT_VBLOCK_MAP_SIZE);
3738 } else {
3739 memcpy(se->discard_map, se->cur_valid_map,
3740 SIT_VBLOCK_MAP_SIZE);
a9af3fdc
CY		 3741 sbi->discard_blks += old_valid_blocks;
3742 sbi->discard_blks -= se->valid_blocks;
1f43e2ad 3743 }
d600af23
CY		 3744 }
3745
a9af3fdc 3746 if (sbi->segs_per_sec > 1) {
d600af23 3747 get_sec_entry(sbi, start)->valid_blocks +=
a9af3fdc
CY		 3748 se->valid_blocks;
3749 get_sec_entry(sbi, start)->valid_blocks -=
3750 old_valid_blocks;
3751 }
d600af23
CY		 3752 }
3753 up_read(&curseg->journal_rwsem);
8a29c126
JK		 3754
3755 if (!err && total_node_blocks != valid_node_count(sbi)) {
3756 f2fs_msg(sbi->sb, KERN_ERR,
3757 "SIT is corrupted node# %u vs %u",
3758 total_node_blocks, valid_node_count(sbi));
3759 set_sbi_flag(sbi, SBI_NEED_FSCK);
3760 err = -EINVAL;
3761 }
3762
c39a1b34 3763 return err;
351df4b2
JK		 3764}
3765
3766static void init_free_segmap(struct f2fs_sb_info *sbi)
3767{
3768 unsigned int start;
3769 int type;
3770
7cd8558b 3771 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3772 struct seg_entry *sentry = get_seg_entry(sbi, start);
3773 if (!sentry->valid_blocks)
3774 __set_free(sbi, start);
c79b7ff1
JK		 3775 else
3776 SIT_I(sbi)->written_valid_blocks +=
3777 sentry->valid_blocks;
351df4b2
JK		 3778 }
3779
3780 /* set use the current segments */
3781 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3782 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3783 __set_test_and_inuse(sbi, curseg_t->segno);
3784 }
3785}
3786
3787static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3788{
3789 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3790 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3791 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3792 unsigned short valid_blocks;
3793
8736fbf0 3794 while (1) {
351df4b2 3795 /* find dirty segment based on free segmap */
7cd8558b
JK		 3796 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3797 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3798 break;
3799 offset = segno + 1;
302bd348 3800 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3801 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3802 continue;
ec325b52
JK		 3803 if (valid_blocks > sbi->blocks_per_seg) {
3804 f2fs_bug_on(sbi, 1);
3805 continue;
3806 }
351df4b2
JK		 3807 mutex_lock(&dirty_i->seglist_lock);
3808 __locate_dirty_segment(sbi, segno, DIRTY);
3809 mutex_unlock(&dirty_i->seglist_lock);
3810 }
3811}
3812
5ec4e49f 3813static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3814{
3815 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3816 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3817
628b3d14 3818 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 3819 if (!dirty_i->victim_secmap)
351df4b2
JK		 3820 return -ENOMEM;
3821 return 0;
3822}
3823
3824static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3825{
3826 struct dirty_seglist_info *dirty_i;
3827 unsigned int bitmap_size, i;
3828
3829 /* allocate memory for dirty segments list information */
acbf054d
CY		 3830 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3831 GFP_KERNEL);
351df4b2
JK		 3832 if (!dirty_i)
3833 return -ENOMEM;
3834
3835 SM_I(sbi)->dirty_info = dirty_i;
3836 mutex_init(&dirty_i->seglist_lock);
3837
7cd8558b 3838 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3839
3840 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 3841 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3842 GFP_KERNEL);
351df4b2
JK		 3843 if (!dirty_i->dirty_segmap[i])
3844 return -ENOMEM;
3845 }
3846
3847 init_dirty_segmap(sbi);
5ec4e49f 3848 return init_victim_secmap(sbi);
351df4b2
JK		 3849}
3850
0a8165d7 3851/*
351df4b2
JK		 3852 * Update min, max modified time for cost-benefit GC algorithm
3853 */
3854static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3855{
3856 struct sit_info *sit_i = SIT_I(sbi);
3857 unsigned int segno;
3858
3d26fa6b 3859 down_write(&sit_i->sentry_lock);
351df4b2 3860
5ad25442 3861 sit_i->min_mtime = ULLONG_MAX;
351df4b2 3862
7cd8558b 3863 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3864 unsigned int i;
3865 unsigned long long mtime = 0;
3866
3867 for (i = 0; i < sbi->segs_per_sec; i++)
3868 mtime += get_seg_entry(sbi, segno + i)->mtime;
3869
3870 mtime = div_u64(mtime, sbi->segs_per_sec);
3871
3872 if (sit_i->min_mtime > mtime)
3873 sit_i->min_mtime = mtime;
3874 }
3875 sit_i->max_mtime = get_mtime(sbi);
3d26fa6b 3876 up_write(&sit_i->sentry_lock);
351df4b2
JK		 3877}
3878
3879int build_segment_manager(struct f2fs_sb_info *sbi)
3880{
3881 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3882 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3883 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3884 int err;
3885
acbf054d 3886 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 3887 if (!sm_info)
3888 return -ENOMEM;
3889
3890 /* init sm info */
3891 sbi->sm_info = sm_info;
351df4b2
JK		 3892 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3893 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3894 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3895 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3896 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3897 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3898 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3899 sm_info->rec_prefree_segments = sm_info->main_segments *
3900 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3901 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3902 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3903
52763a4b
JK		 3904 if (!test_opt(sbi, LFS))
3905 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3906 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3907 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3908 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 3909 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 3910
184a5cd2
CY		 3911 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3912
2b60311d
CY		 3913 init_rwsem(&sm_info->curseg_lock);
3914
d4fdf8ba 3915 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3916 err = create_flush_cmd_control(sbi);
3917 if (err)
a688b9d9 3918 return err;
6b4afdd7
JK		 3919 }
3920
0b54fb84
JK		 3921 err = create_discard_cmd_control(sbi);
3922 if (err)
3923 return err;
3924
351df4b2
JK		 3925 err = build_sit_info(sbi);
3926 if (err)
3927 return err;
3928 err = build_free_segmap(sbi);
3929 if (err)
3930 return err;
3931 err = build_curseg(sbi);
3932 if (err)
3933 return err;
3934
3935 /* reinit free segmap based on SIT */
c39a1b34
JK		 3936 err = build_sit_entries(sbi);
3937 if (err)
3938 return err;
351df4b2
JK		 3939
3940 init_free_segmap(sbi);
3941 err = build_dirty_segmap(sbi);
3942 if (err)
3943 return err;
3944
3945 init_min_max_mtime(sbi);
3946 return 0;
3947}
3948
3949static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3950 enum dirty_type dirty_type)
3951{
3952 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3953
3954 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3955 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3956 dirty_i->nr_dirty[dirty_type] = 0;
3957 mutex_unlock(&dirty_i->seglist_lock);
3958}
3959
5ec4e49f 3960static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3961{
3962 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3963 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3964}
3965
3966static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3967{
3968 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3969 int i;
3970
3971 if (!dirty_i)
3972 return;
3973
3974 /* discard pre-free/dirty segments list */
3975 for (i = 0; i < NR_DIRTY_TYPE; i++)
3976 discard_dirty_segmap(sbi, i);
3977
5ec4e49f 3978 destroy_victim_secmap(sbi);
351df4b2
JK		 3979 SM_I(sbi)->dirty_info = NULL;
3980 kfree(dirty_i);
3981}
3982
3983static void destroy_curseg(struct f2fs_sb_info *sbi)
3984{
3985 struct curseg_info *array = SM_I(sbi)->curseg_array;
3986 int i;
3987
3988 if (!array)
3989 return;
3990 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3991 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3992 kfree(array[i].sum_blk);
b7ad7512
CY		 3993 kfree(array[i].journal);
3994 }
351df4b2
JK		 3995 kfree(array);
3996}
3997
3998static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3999{
4000 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4001 if (!free_i)
4002 return;
4003 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 4004 kvfree(free_i->free_segmap);
4005 kvfree(free_i->free_secmap);
351df4b2
JK		 4006 kfree(free_i);
4007}
4008
4009static void destroy_sit_info(struct f2fs_sb_info *sbi)
4010{
4011 struct sit_info *sit_i = SIT_I(sbi);
4012 unsigned int start;
4013
4014 if (!sit_i)
4015 return;
4016
4017 if (sit_i->sentries) {
7cd8558b 4018 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 4019 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 4020#ifdef CONFIG_F2FS_CHECK_FS
4021 kfree(sit_i->sentries[start].cur_valid_map_mir);
4022#endif
351df4b2 4023 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 4024 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 4025 }
4026 }
60a3b782
JK		 4027 kfree(sit_i->tmp_map);
4028
39307a8e
JK		 4029 kvfree(sit_i->sentries);
4030 kvfree(sit_i->sec_entries);
4031 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 4032
4033 SM_I(sbi)->sit_info = NULL;
4034 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 4035#ifdef CONFIG_F2FS_CHECK_FS
4036 kfree(sit_i->sit_bitmap_mir);
4037#endif
351df4b2
JK		 4038 kfree(sit_i);
4039}
4040
4041void destroy_segment_manager(struct f2fs_sb_info *sbi)
4042{
4043 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 4044
3b03f724
CY		 4045 if (!sm_info)
4046 return;
5eba8c5d 4047 destroy_flush_cmd_control(sbi, true);
f099405f 4048 destroy_discard_cmd_control(sbi);
351df4b2
JK		 4049 destroy_dirty_segmap(sbi);
4050 destroy_curseg(sbi);
4051 destroy_free_segmap(sbi);
4052 destroy_sit_info(sbi);
4053 sbi->sm_info = NULL;
4054 kfree(sm_info);
4055}
7fd9e544
JK		 4056
4057int __init create_segment_manager_caches(void)
4058{
4059 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 4060 sizeof(struct discard_entry));
7fd9e544 4061 if (!discard_entry_slab)
184a5cd2
CY		 4062 goto fail;
4063
b01a9201
JK		 4064 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4065 sizeof(struct discard_cmd));
4066 if (!discard_cmd_slab)
6ab2a308 4067 goto destroy_discard_entry;
275b66b0 4068
184a5cd2 4069 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 4070 sizeof(struct sit_entry_set));
184a5cd2 4071 if (!sit_entry_set_slab)
b01a9201 4072 goto destroy_discard_cmd;
88b88a66
JK		 4073
4074 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4075 sizeof(struct inmem_pages));
4076 if (!inmem_entry_slab)
4077 goto destroy_sit_entry_set;
7fd9e544 4078 return 0;
184a5cd2 4079
88b88a66
JK		 4080destroy_sit_entry_set:
4081 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 4082destroy_discard_cmd:
4083 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 4084destroy_discard_entry:
184a5cd2
CY		 4085 kmem_cache_destroy(discard_entry_slab);
4086fail:
4087 return -ENOMEM;
7fd9e544
JK		 4088}
4089
4090void destroy_segment_manager_caches(void)
4091{
184a5cd2 4092 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 4093 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 4094 kmem_cache_destroy(discard_entry_slab);
88b88a66 4095 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 4096}
Linux 6.x block layer and io_uring tree(s)