]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: make raid6 rebuild retry more
[mirror_ubuntu-bionic-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
1d7be270 19#include <linux/freezer.h>
1eb1ef4a 20#include <linux/sched/signal.h>
351df4b2
JK		 21
22#include "f2fs.h"
23#include "segment.h"
24#include "node.h"
5f656541 25#include "gc.h"
9e4ded3f 26#include "trace.h"
6ec178da 27#include <trace/events/f2fs.h>
351df4b2 28
9a7f143a
CL		 29#define __reverse_ffz(x) __reverse_ffs(~(x))
30
7fd9e544 31static struct kmem_cache *discard_entry_slab;
b01a9201 32static struct kmem_cache *discard_cmd_slab;
184a5cd2 33static struct kmem_cache *sit_entry_set_slab;
88b88a66 34static struct kmem_cache *inmem_entry_slab;
7fd9e544 35
f96999c3
JK		 36static unsigned long __reverse_ulong(unsigned char *str)
37{
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41#if BITS_PER_LONG == 64
42 shift = 56;
43#endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49}
50
9a7f143a
CL		 51/*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55static inline unsigned long __reverse_ffs(unsigned long word)
56{
57 int num = 0;
58
59#if BITS_PER_LONG == 64
f96999c3 60 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 61 num += 32;
f96999c3 62 else
9a7f143a 63 word >>= 32;
9a7f143a 64#endif
f96999c3 65 if ((word & 0xffff0000) == 0)
9a7f143a 66 num += 16;
f96999c3 67 else
9a7f143a 68 word >>= 16;
f96999c3
JK		 69
70 if ((word & 0xff00) == 0)
9a7f143a 71 num += 8;
f96999c3 72 else
9a7f143a 73 word >>= 8;
f96999c3 74
9a7f143a
CL		 75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
f96999c3 79
9a7f143a
CL		 80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
f96999c3 84
9a7f143a
CL		 85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88}
89
90/*
e1c42045 91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 93 * @size must be integral times of unsigned long.
9a7f143a 94 * Example:
f96999c3
JK		 95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 98 */
99static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101{
102 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 103 unsigned long result = size;
9a7f143a 104 unsigned long tmp;
9a7f143a
CL		 105
106 if (offset >= size)
107 return size;
108
692223d1 109 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 110 offset %= BITS_PER_LONG;
f96999c3 111
692223d1
FL		 112 while (1) {
113 if (*p == 0)
114 goto pass;
9a7f143a 115
f96999c3 116 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 121 if (tmp)
692223d1
FL		 122 goto found;
123pass:
124 if (size <= BITS_PER_LONG)
125 break;
9a7f143a 126 size -= BITS_PER_LONG;
692223d1 127 offset = 0;
f96999c3 128 p++;
9a7f143a 129 }
692223d1
FL		 130 return result;
131found:
132 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 133}
134
135static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137{
138 const unsigned long *p = addr + BIT_WORD(offset);
80609448 139 unsigned long result = size;
9a7f143a 140 unsigned long tmp;
9a7f143a
CL		 141
142 if (offset >= size)
143 return size;
144
80609448 145 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 146 offset %= BITS_PER_LONG;
80609448
JK		 147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
f96999c3 152 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
f96999c3 158 if (tmp != ~0UL)
80609448
JK		 159 goto found;
160pass:
161 if (size <= BITS_PER_LONG)
162 break;
9a7f143a 163 size -= BITS_PER_LONG;
80609448 164 offset = 0;
f96999c3 165 p++;
9a7f143a 166 }
80609448
JK		 167 return result;
168found:
169 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 170}
171
b3a97a2a
JK		 172bool need_SSR(struct f2fs_sb_info *sbi)
173{
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177
178 if (test_opt(sbi, LFS))
179 return false;
180 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
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
28bc106b
CY		 233 if (recover) {
234 struct dnode_of_data dn;
235 struct node_info ni;
236
237 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 238retry:
28bc106b 239 set_new_dnode(&dn, inode, NULL, NULL, 0);
7f2b4e8e
CY		 240 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
241 if (err) {
242 if (err == -ENOMEM) {
243 congestion_wait(BLK_RW_ASYNC, HZ/50);
244 cond_resched();
245 goto retry;
246 }
28bc106b
CY		 247 err = -EAGAIN;
248 goto next;
249 }
250 get_node_info(sbi, dn.nid, &ni);
251 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
252 cur->old_addr, ni.version, true, true);
253 f2fs_put_dnode(&dn);
254 }
255next:
63c52d78
JK		 256 /* we don't need to invalidate this in the sccessful status */
257 if (drop || recover)
258 ClearPageUptodate(page);
28bc106b 259 set_page_private(page, 0);
c81ced05 260 ClearPagePrivate(page);
28bc106b 261 f2fs_put_page(page, 1);
29b96b54
CY		 262
263 list_del(&cur->list);
264 kmem_cache_free(inmem_entry_slab, cur);
265 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
266 }
28bc106b 267 return err;
29b96b54
CY		 268}
269
57864ae5
JK		 270void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
271{
272 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
273 struct inode *inode;
274 struct f2fs_inode_info *fi;
275next:
276 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
277 if (list_empty(head)) {
278 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
279 return;
280 }
281 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
282 inode = igrab(&fi->vfs_inode);
283 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
284
285 if (inode) {
286 drop_inmem_pages(inode);
287 iput(inode);
288 }
289 congestion_wait(BLK_RW_ASYNC, HZ/50);
290 cond_resched();
291 goto next;
292}
293
29b96b54
CY		 294void drop_inmem_pages(struct inode *inode)
295{
57864ae5 296 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54
CY		 297 struct f2fs_inode_info *fi = F2FS_I(inode);
298
299 mutex_lock(&fi->inmem_lock);
28bc106b 300 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
57864ae5
JK		 301 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
302 if (!list_empty(&fi->inmem_ilist))
303 list_del_init(&fi->inmem_ilist);
304 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
29b96b54 305 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 306
307 clear_inode_flag(inode, FI_ATOMIC_FILE);
84a23fbe 308 clear_inode_flag(inode, FI_HOT_DATA);
5fe45743 309 stat_dec_atomic_write(inode);
29b96b54
CY		 310}
311
8c242db9
JK		 312void drop_inmem_page(struct inode *inode, struct page *page)
313{
314 struct f2fs_inode_info *fi = F2FS_I(inode);
315 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
316 struct list_head *head = &fi->inmem_pages;
317 struct inmem_pages *cur = NULL;
318
319 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
320
321 mutex_lock(&fi->inmem_lock);
322 list_for_each_entry(cur, head, list) {
323 if (cur->page == page)
324 break;
325 }
326
327 f2fs_bug_on(sbi, !cur || cur->page != page);
328 list_del(&cur->list);
329 mutex_unlock(&fi->inmem_lock);
330
331 dec_page_count(sbi, F2FS_INMEM_PAGES);
332 kmem_cache_free(inmem_entry_slab, cur);
333
334 ClearPageUptodate(page);
335 set_page_private(page, 0);
336 ClearPagePrivate(page);
337 f2fs_put_page(page, 0);
338
339 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
340}
341
28bc106b
CY		 342static int __commit_inmem_pages(struct inode *inode,
343 struct list_head *revoke_list)
88b88a66
JK		 344{
345 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
346 struct f2fs_inode_info *fi = F2FS_I(inode);
347 struct inmem_pages *cur, *tmp;
88b88a66 348 struct f2fs_io_info fio = {
05ca3632 349 .sbi = sbi,
39d787be 350 .ino = inode->i_ino,
88b88a66 351 .type = DATA,
04d328de 352 .op = REQ_OP_WRITE,
70fd7614 353 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 354 .io_type = FS_DATA_IO,
88b88a66 355 };
942fd319 356 pgoff_t last_idx = ULONG_MAX;
edb27dee 357 int err = 0;
88b88a66 358
88b88a66 359 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 360 struct page *page = cur->page;
361
362 lock_page(page);
363 if (page->mapping == inode->i_mapping) {
364 trace_f2fs_commit_inmem_page(page, INMEM);
365
366 set_page_dirty(page);
367 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 368 if (clear_page_dirty_for_io(page)) {
29b96b54 369 inode_dec_dirty_pages(inode);
933439c8
CY		 370 remove_dirty_inode(inode);
371 }
640cc189 372retry:
28bc106b 373 fio.page = page;
e959c8f5 374 fio.old_blkaddr = NULL_ADDR;
4d978078 375 fio.encrypted_page = NULL;
cc15620b 376 fio.need_lock = LOCK_DONE;
29b96b54
CY		 377 err = do_write_data_page(&fio);
378 if (err) {
640cc189
JK		 379 if (err == -ENOMEM) {
380 congestion_wait(BLK_RW_ASYNC, HZ/50);
381 cond_resched();
382 goto retry;
383 }
28bc106b 384 unlock_page(page);
29b96b54 385 break;
70c640b1 386 }
28bc106b
CY		 387 /* record old blkaddr for revoking */
388 cur->old_addr = fio.old_blkaddr;
942fd319 389 last_idx = page->index;
28bc106b
CY		 390 }
391 unlock_page(page);
392 list_move_tail(&cur->list, revoke_list);
88b88a66 393 }
29b96b54 394
942fd319 395 if (last_idx != ULONG_MAX)
b9109b0e 396 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b
CY		 397
398 if (!err)
399 __revoke_inmem_pages(inode, revoke_list, false, false);
400
29b96b54
CY		 401 return err;
402}
403
404int commit_inmem_pages(struct inode *inode)
405{
406 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
407 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 408 struct list_head revoke_list;
409 int err;
29b96b54 410
28bc106b 411 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 412 f2fs_balance_fs(sbi, true);
413 f2fs_lock_op(sbi);
414
5fe45743
CY		 415 set_inode_flag(inode, FI_ATOMIC_COMMIT);
416
29b96b54 417 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 418 err = __commit_inmem_pages(inode, &revoke_list);
419 if (err) {
420 int ret;
421 /*
422 * try to revoke all committed pages, but still we could fail
423 * due to no memory or other reason, if that happened, EAGAIN
424 * will be returned, which means in such case, transaction is
425 * already not integrity, caller should use journal to do the
426 * recovery or rewrite & commit last transaction. For other
427 * error number, revoking was done by filesystem itself.
428 */
429 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
430 if (ret)
431 err = ret;
432
433 /* drop all uncommitted pages */
434 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
435 }
57864ae5
JK		 436 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
437 if (!list_empty(&fi->inmem_ilist))
438 list_del_init(&fi->inmem_ilist);
439 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
88b88a66
JK		 440 mutex_unlock(&fi->inmem_lock);
441
5fe45743
CY		 442 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
443
29b96b54 444 f2fs_unlock_op(sbi);
edb27dee 445 return err;
88b88a66
JK		 446}
447
0a8165d7 448/*
351df4b2
JK		 449 * This function balances dirty node and dentry pages.
450 * In addition, it controls garbage collection.
451 */
2c4db1a6 452void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 453{
0f348028 454#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 455 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
456 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 457 f2fs_stop_checkpoint(sbi, false);
55523519 458 }
0f348028
CY		 459#endif
460
e589c2c4 461 /* balance_fs_bg is able to be pending */
a7881893 462 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 463 f2fs_balance_fs_bg(sbi);
464
351df4b2 465 /*
029cd28c
JK		 466 * We should do GC or end up with checkpoint, if there are so many dirty
467 * dir/node pages without enough free segments.
351df4b2 468 */
7f3037a5 469 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 470 mutex_lock(&sbi->gc_mutex);
e066b83c 471 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 472 }
473}
474
4660f9c0
JK		 475void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
476{
1dcc336b 477 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 478 if (!available_free_memory(sbi, EXTENT_CACHE))
479 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 480
1b38dc8e
JK		 481 /* check the # of cached NAT entries */
482 if (!available_free_memory(sbi, NAT_ENTRIES))
483 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
484
31696580 485 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 486 try_to_free_nids(sbi, MAX_FREE_NIDS);
487 else
22ad0b6a 488 build_free_nids(sbi, false, false);
31696580 489
1c0f4bf5 490 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 491 return;
31696580 492
1b38dc8e
JK		 493 /* checkpoint is the only way to shrink partial cached entries */
494 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 495 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 496 excess_prefree_segs(sbi) ||
497 excess_dirty_nats(sbi) ||
f455c8a5 498 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 499 if (test_opt(sbi, DATA_FLUSH)) {
500 struct blk_plug plug;
501
502 blk_start_plug(&plug);
36b35a0d 503 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 504 blk_finish_plug(&plug);
505 }
4660f9c0 506 f2fs_sync_fs(sbi->sb, true);
42190d2a 507 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 508 }
4660f9c0
JK		 509}
510
20fda56b
KM		 511static int __submit_flush_wait(struct f2fs_sb_info *sbi,
512 struct block_device *bdev)
3c62be17 513{
d62fe971 514 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
3c62be17
JK		 515 int ret;
516
3adc5fcb 517 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 518 bio_set_dev(bio, bdev);
3c62be17
JK		 519 ret = submit_bio_wait(bio);
520 bio_put(bio);
20fda56b
KM		 521
522 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
523 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 524 return ret;
525}
526
39d787be 527static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 528{
39d787be 529 int ret = 0;
3c62be17
JK		 530 int i;
531
39d787be
CY		 532 if (!sbi->s_ndevs)
533 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 534
39d787be
CY		 535 for (i = 0; i < sbi->s_ndevs; i++) {
536 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
537 continue;
20fda56b
KM		 538 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
539 if (ret)
540 break;
3c62be17
JK		 541 }
542 return ret;
543}
544
2163d198 545static int issue_flush_thread(void *data)
6b4afdd7
JK		 546{
547 struct f2fs_sb_info *sbi = data;
b01a9201 548 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 549 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 550repeat:
551 if (kthread_should_stop())
552 return 0;
553
dc6febb6
CY		 554 sb_start_intwrite(sbi->sb);
555
721bd4d5 556 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 557 struct flush_cmd *cmd, *next;
558 int ret;
559
721bd4d5
GZ		 560 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
561 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
562
39d787be
CY		 563 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
564
565 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 566 atomic_inc(&fcc->issued_flush);
567
721bd4d5
GZ		 568 llist_for_each_entry_safe(cmd, next,
569 fcc->dispatch_list, llnode) {
6b4afdd7 570 cmd->ret = ret;
6b4afdd7
JK		 571 complete(&cmd->wait);
572 }
a688b9d9 573 fcc->dispatch_list = NULL;
6b4afdd7
JK		 574 }
575
dc6febb6
CY		 576 sb_end_intwrite(sbi->sb);
577
a688b9d9 578 wait_event_interruptible(*q,
721bd4d5 579 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 580 goto repeat;
581}
582
39d787be 583int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 584{
b01a9201 585 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 586 struct flush_cmd cmd;
8b8dd65f 587 int ret;
6b4afdd7 588
0f7b2abd
JK		 589 if (test_opt(sbi, NOBARRIER))
590 return 0;
591
8b8dd65f 592 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 593 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 594 atomic_inc(&fcc->issued_flush);
595 return ret;
596 }
740432f8 597
39d787be
CY		 598 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
599 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 600 atomic_dec(&fcc->issing_flush);
601
602 atomic_inc(&fcc->issued_flush);
740432f8
JK		 603 return ret;
604 }
6b4afdd7 605
39d787be 606 cmd.ino = ino;
adf8d90b 607 init_completion(&cmd.wait);
6b4afdd7 608
721bd4d5 609 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 610
6f890df0
CY		 611 /* update issue_list before we wake up issue_flush thread */
612 smp_mb();
613
614 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 615 wake_up(&fcc->flush_wait_queue);
6b4afdd7 616
5eba8c5d
JK		 617 if (fcc->f2fs_issue_flush) {
618 wait_for_completion(&cmd.wait);
8b8dd65f 619 atomic_dec(&fcc->issing_flush);
5eba8c5d 620 } else {
d3238691
CY		 621 struct llist_node *list;
622
623 list = llist_del_all(&fcc->issue_list);
624 if (!list) {
625 wait_for_completion(&cmd.wait);
626 atomic_dec(&fcc->issing_flush);
627 } else {
628 struct flush_cmd *tmp, *next;
629
39d787be 630 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 631
632 llist_for_each_entry_safe(tmp, next, list, llnode) {
633 if (tmp == &cmd) {
634 cmd.ret = ret;
635 atomic_dec(&fcc->issing_flush);
636 continue;
637 }
638 tmp->ret = ret;
639 complete(&tmp->wait);
640 }
641 }
5eba8c5d 642 }
adf8d90b
CY		 643
644 return cmd.ret;
6b4afdd7
JK		 645}
646
2163d198
GZ		 647int create_flush_cmd_control(struct f2fs_sb_info *sbi)
648{
649 dev_t dev = sbi->sb->s_bdev->bd_dev;
650 struct flush_cmd_control *fcc;
651 int err = 0;
652
b01a9201
JK		 653 if (SM_I(sbi)->fcc_info) {
654 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 655 if (fcc->f2fs_issue_flush)
656 return err;
5eba8c5d
JK		 657 goto init_thread;
658 }
659
2163d198
GZ		 660 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
661 if (!fcc)
662 return -ENOMEM;
8b8dd65f
CY		 663 atomic_set(&fcc->issued_flush, 0);
664 atomic_set(&fcc->issing_flush, 0);
2163d198 665 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 666 init_llist_head(&fcc->issue_list);
b01a9201 667 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 668 if (!test_opt(sbi, FLUSH_MERGE))
669 return err;
670
5eba8c5d 671init_thread:
2163d198
GZ		 672 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
673 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
674 if (IS_ERR(fcc->f2fs_issue_flush)) {
675 err = PTR_ERR(fcc->f2fs_issue_flush);
676 kfree(fcc);
b01a9201 677 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 678 return err;
679 }
2163d198
GZ		 680
681 return err;
682}
683
5eba8c5d 684void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 685{
b01a9201 686 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 687
5eba8c5d
JK		 688 if (fcc && fcc->f2fs_issue_flush) {
689 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
690
691 fcc->f2fs_issue_flush = NULL;
692 kthread_stop(flush_thread);
693 }
694 if (free) {
695 kfree(fcc);
b01a9201 696 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 697 }
2163d198
GZ		 698}
699
1228b482
CY		 700int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
701{
702 int ret = 0, i;
703
704 if (!sbi->s_ndevs)
705 return 0;
706
707 for (i = 1; i < sbi->s_ndevs; i++) {
708 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
709 continue;
710 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
711 if (ret)
712 break;
713
714 spin_lock(&sbi->dev_lock);
715 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
716 spin_unlock(&sbi->dev_lock);
717 }
718
719 return ret;
720}
721
351df4b2
JK		 722static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
723 enum dirty_type dirty_type)
724{
725 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
726
727 /* need not be added */
728 if (IS_CURSEG(sbi, segno))
729 return;
730
731 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
732 dirty_i->nr_dirty[dirty_type]++;
733
734 if (dirty_type == DIRTY) {
735 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 736 enum dirty_type t = sentry->type;
b2f2c390 737
ec325b52
JK		 738 if (unlikely(t >= DIRTY)) {
739 f2fs_bug_on(sbi, 1);
740 return;
741 }
4625d6aa
CL		 742 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
743 dirty_i->nr_dirty[t]++;
351df4b2
JK		 744 }
745}
746
747static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
748 enum dirty_type dirty_type)
749{
750 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
751
752 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
753 dirty_i->nr_dirty[dirty_type]--;
754
755 if (dirty_type == DIRTY) {
4625d6aa
CL		 756 struct seg_entry *sentry = get_seg_entry(sbi, segno);
757 enum dirty_type t = sentry->type;
758
759 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
760 dirty_i->nr_dirty[t]--;
b2f2c390 761
302bd348 762 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 763 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 764 dirty_i->victim_secmap);
351df4b2
JK		 765 }
766}
767
0a8165d7 768/*
351df4b2
JK		 769 * Should not occur error such as -ENOMEM.
770 * Adding dirty entry into seglist is not critical operation.
771 * If a given segment is one of current working segments, it won't be added.
772 */
8d8451af 773static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 774{
775 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
776 unsigned short valid_blocks;
777
778 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
779 return;
780
781 mutex_lock(&dirty_i->seglist_lock);
782
302bd348 783 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 784
785 if (valid_blocks == 0) {
786 __locate_dirty_segment(sbi, segno, PRE);
787 __remove_dirty_segment(sbi, segno, DIRTY);
788 } else if (valid_blocks < sbi->blocks_per_seg) {
789 __locate_dirty_segment(sbi, segno, DIRTY);
790 } else {
791 /* Recovery routine with SSR needs this */
792 __remove_dirty_segment(sbi, segno, DIRTY);
793 }
794
795 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 796}
797
004b6862 798static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 799 struct block_device *bdev, block_t lstart,
800 block_t start, block_t len)
275b66b0 801{
0b54fb84 802 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 803 struct list_head *pend_list;
b01a9201 804 struct discard_cmd *dc;
275b66b0 805
ba48a33e
CY		 806 f2fs_bug_on(sbi, !len);
807
808 pend_list = &dcc->pend_list[plist_idx(len)];
809
b01a9201
JK		 810 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
811 INIT_LIST_HEAD(&dc->list);
c81abe34 812 dc->bdev = bdev;
b01a9201 813 dc->lstart = lstart;
c81abe34 814 dc->start = start;
b01a9201 815 dc->len = len;
ec9895ad 816 dc->ref = 0;
15469963 817 dc->state = D_PREP;
c81abe34 818 dc->error = 0;
b01a9201 819 init_completion(&dc->wait);
22d375dd 820 list_add_tail(&dc->list, pend_list);
5f32366a 821 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 822 dcc->undiscard_blks += len;
004b6862
CY		 823
824 return dc;
825}
826
827static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
828 struct block_device *bdev, block_t lstart,
829 block_t start, block_t len,
830 struct rb_node *parent, struct rb_node **p)
831{
832 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
833 struct discard_cmd *dc;
834
835 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
836
837 rb_link_node(&dc->rb_node, parent, p);
838 rb_insert_color(&dc->rb_node, &dcc->root);
839
840 return dc;
15469963
JK		 841}
842
004b6862
CY		 843static void __detach_discard_cmd(struct discard_cmd_control *dcc,
844 struct discard_cmd *dc)
15469963 845{
dcc9165d 846 if (dc->state == D_DONE)
004b6862
CY		 847 atomic_dec(&dcc->issing_discard);
848
849 list_del(&dc->list);
850 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 851 dcc->undiscard_blks -= dc->len;
004b6862
CY		 852
853 kmem_cache_free(discard_cmd_slab, dc);
854
855 atomic_dec(&dcc->discard_cmd_cnt);
856}
857
858static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
859 struct discard_cmd *dc)
860{
861 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 862
2ec6f2ef
CY		 863 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
864
d9703d90
CY		 865 f2fs_bug_on(sbi, dc->ref);
866
c81abe34
JK		 867 if (dc->error == -EOPNOTSUPP)
868 dc->error = 0;
15469963 869
c81abe34 870 if (dc->error)
15469963 871 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 872 "Issue discard(%u, %u, %u) failed, ret: %d",
873 dc->lstart, dc->start, dc->len, dc->error);
004b6862 874 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 875}
876
c81abe34
JK		 877static void f2fs_submit_discard_endio(struct bio *bio)
878{
879 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
880
4e4cbee9 881 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 882 dc->state = D_DONE;
e31b9821 883 complete_all(&dc->wait);
c81abe34
JK		 884 bio_put(bio);
885}
886
6915ea9d
CY		 887void __check_sit_bitmap(struct f2fs_sb_info *sbi,
888 block_t start, block_t end)
889{
890#ifdef CONFIG_F2FS_CHECK_FS
891 struct seg_entry *sentry;
892 unsigned int segno;
893 block_t blk = start;
894 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
895 unsigned long *map;
896
897 while (blk < end) {
898 segno = GET_SEGNO(sbi, blk);
899 sentry = get_seg_entry(sbi, segno);
900 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
901
008396e1
YS		 902 if (end < START_BLOCK(sbi, segno + 1))
903 size = GET_BLKOFF_FROM_SEG0(sbi, end);
904 else
905 size = max_blocks;
6915ea9d
CY		 906 map = (unsigned long *)(sentry->cur_valid_map);
907 offset = __find_rev_next_bit(map, size, offset);
908 f2fs_bug_on(sbi, offset != size);
008396e1 909 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 910 }
911#endif
912}
913
c81abe34
JK		 914/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
915static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56
CY		 916 struct discard_policy *dpolicy,
917 struct discard_cmd *dc)
c81abe34
JK		 918{
919 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 920 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
921 &(dcc->fstrim_list) : &(dcc->wait_list);
c81abe34 922 struct bio *bio = NULL;
78997b56 923 int flag = dpolicy->sync ? REQ_SYNC : 0;
c81abe34
JK		 924
925 if (dc->state != D_PREP)
926 return;
927
0243a5f9
CY		 928 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
929
c81abe34
JK		 930 dc->error = __blkdev_issue_discard(dc->bdev,
931 SECTOR_FROM_BLOCK(dc->start),
932 SECTOR_FROM_BLOCK(dc->len),
933 GFP_NOFS, 0, &bio);
934 if (!dc->error) {
935 /* should keep before submission to avoid D_DONE right away */
936 dc->state = D_SUBMIT;
8b8dd65f
CY		 937 atomic_inc(&dcc->issued_discard);
938 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 939 if (bio) {
940 bio->bi_private = dc;
941 bio->bi_end_io = f2fs_submit_discard_endio;
ecc9aa00 942 bio->bi_opf |= flag;
c81abe34 943 submit_bio(bio);
8412663d 944 list_move_tail(&dc->list, wait_list);
6915ea9d 945 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 946
947 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 948 }
949 } else {
950 __remove_discard_cmd(sbi, dc);
951 }
952}
953
004b6862
CY		 954static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
955 struct block_device *bdev, block_t lstart,
956 block_t start, block_t len,
957 struct rb_node **insert_p,
958 struct rb_node *insert_parent)
c81abe34 959{
004b6862 960 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 961 struct rb_node **p;
004b6862
CY		 962 struct rb_node *parent = NULL;
963 struct discard_cmd *dc = NULL;
964
965 if (insert_p && insert_parent) {
966 parent = insert_parent;
967 p = insert_p;
968 goto do_insert;
969 }
c81abe34 970
004b6862
CY		 971 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
972do_insert:
973 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
974 if (!dc)
975 return NULL;
c81abe34 976
004b6862 977 return dc;
c81abe34
JK		 978}
979
ba48a33e
CY		 980static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
981 struct discard_cmd *dc)
982{
983 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
984}
985
3d6a650f
YH		 986static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
987 struct discard_cmd *dc, block_t blkaddr)
988{
ba48a33e 989 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 990 struct discard_info di = dc->di;
991 bool modified = false;
3d6a650f 992
004b6862 993 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 994 __remove_discard_cmd(sbi, dc);
995 return;
996 }
997
d84d1cbd
CY		 998 dcc->undiscard_blks -= di.len;
999
004b6862 1000 if (blkaddr > di.lstart) {
3d6a650f 1001 dc->len = blkaddr - dc->lstart;
d84d1cbd 1002 dcc->undiscard_blks += dc->len;
ba48a33e 1003 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1004 modified = true;
1005 }
1006
1007 if (blkaddr < di.lstart + di.len - 1) {
1008 if (modified) {
1009 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1010 di.start + blkaddr + 1 - di.lstart,
1011 di.lstart + di.len - 1 - blkaddr,
1012 NULL, NULL);
1013 } else {
1014 dc->lstart++;
1015 dc->len--;
1016 dc->start++;
d84d1cbd 1017 dcc->undiscard_blks += dc->len;
ba48a33e 1018 __relocate_discard_cmd(dcc, dc);
004b6862 1019 }
3d6a650f
YH		 1020 }
1021}
1022
004b6862
CY		 1023static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1024 struct block_device *bdev, block_t lstart,
1025 block_t start, block_t len)
275b66b0 1026{
0b54fb84 1027 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1028 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1029 struct discard_cmd *dc;
1030 struct discard_info di = {0};
1031 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1032 block_t end = lstart + len;
275b66b0 1033
15469963 1034 mutex_lock(&dcc->cmd_lock);
40465257 1035
004b6862
CY		 1036 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1037 NULL, lstart,
1038 (struct rb_entry **)&prev_dc,
1039 (struct rb_entry **)&next_dc,
1040 &insert_p, &insert_parent, true);
1041 if (dc)
1042 prev_dc = dc;
1043
1044 if (!prev_dc) {
1045 di.lstart = lstart;
1046 di.len = next_dc ? next_dc->lstart - lstart : len;
1047 di.len = min(di.len, len);
1048 di.start = start;
22d375dd 1049 }
15469963 1050
004b6862
CY		 1051 while (1) {
1052 struct rb_node *node;
1053 bool merged = false;
1054 struct discard_cmd *tdc = NULL;
1055
1056 if (prev_dc) {
1057 di.lstart = prev_dc->lstart + prev_dc->len;
1058 if (di.lstart < lstart)
1059 di.lstart = lstart;
1060 if (di.lstart >= end)
1061 break;
1062
1063 if (!next_dc || next_dc->lstart > end)
1064 di.len = end - di.lstart;
1065 else
1066 di.len = next_dc->lstart - di.lstart;
1067 di.start = start + di.lstart - lstart;
1068 }
1069
1070 if (!di.len)
1071 goto next;
1072
1073 if (prev_dc && prev_dc->state == D_PREP &&
1074 prev_dc->bdev == bdev &&
1075 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1076 prev_dc->di.len += di.len;
d84d1cbd 1077 dcc->undiscard_blks += di.len;
ba48a33e 1078 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1079 di = prev_dc->di;
1080 tdc = prev_dc;
1081 merged = true;
1082 }
1083
1084 if (next_dc && next_dc->state == D_PREP &&
1085 next_dc->bdev == bdev &&
1086 __is_discard_front_mergeable(&di, &next_dc->di)) {
1087 next_dc->di.lstart = di.lstart;
1088 next_dc->di.len += di.len;
1089 next_dc->di.start = di.start;
d84d1cbd 1090 dcc->undiscard_blks += di.len;
ba48a33e 1091 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1092 if (tdc)
1093 __remove_discard_cmd(sbi, tdc);
004b6862 1094 merged = true;
4e6a8d9b 1095 }
004b6862 1096
df0f6b44 1097 if (!merged) {
004b6862
CY		 1098 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1099 di.len, NULL, NULL);
df0f6b44 1100 }
004b6862
CY		 1101 next:
1102 prev_dc = next_dc;
1103 if (!prev_dc)
1104 break;
1105
1106 node = rb_next(&prev_dc->rb_node);
1107 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1108 }
1109
1110 mutex_unlock(&dcc->cmd_lock);
1111}
1112
1113static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1114 struct block_device *bdev, block_t blkstart, block_t blklen)
1115{
1116 block_t lblkstart = blkstart;
1117
0243a5f9 1118 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1119
1120 if (sbi->s_ndevs) {
1121 int devi = f2fs_target_device_index(sbi, blkstart);
1122
1123 blkstart -= FDEV(devi).start_blk;
1124 }
1125 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1126 return 0;
1127}
1128
8412663d 1129static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1130 struct discard_policy *dpolicy,
1131 unsigned int start, unsigned int end)
8412663d
CY		 1132{
1133 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1134 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1135 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1136 struct discard_cmd *dc;
1137 struct blk_plug plug;
1138 int issued;
1139
1140next:
1141 issued = 0;
1142
1143 mutex_lock(&dcc->cmd_lock);
1144 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1145
1146 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1147 NULL, start,
1148 (struct rb_entry **)&prev_dc,
1149 (struct rb_entry **)&next_dc,
1150 &insert_p, &insert_parent, true);
1151 if (!dc)
1152 dc = next_dc;
1153
1154 blk_start_plug(&plug);
1155
1156 while (dc && dc->lstart <= end) {
1157 struct rb_node *node;
1158
78997b56 1159 if (dc->len < dpolicy->granularity)
8412663d
CY		 1160 goto skip;
1161
1162 if (dc->state != D_PREP) {
1163 list_move_tail(&dc->list, &dcc->fstrim_list);
1164 goto skip;
1165 }
1166
78997b56 1167 __submit_discard_cmd(sbi, dpolicy, dc);
8412663d 1168
ecc9aa00 1169 if (++issued >= dpolicy->max_requests) {
8412663d
CY		 1170 start = dc->lstart + dc->len;
1171
1172 blk_finish_plug(&plug);
1173 mutex_unlock(&dcc->cmd_lock);
1174
1175 schedule();
1176
1177 goto next;
1178 }
1179skip:
1180 node = rb_next(&dc->rb_node);
1181 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1182
1183 if (fatal_signal_pending(current))
1184 break;
1185 }
1186
1187 blk_finish_plug(&plug);
1188 mutex_unlock(&dcc->cmd_lock);
1189}
1190
78997b56
CY		 1191static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1192 struct discard_policy *dpolicy)
bd5b0738
CY		 1193{
1194 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1195 struct list_head *pend_list;
1196 struct discard_cmd *dc, *tmp;
1197 struct blk_plug plug;
78997b56 1198 int i, iter = 0, issued = 0;
e6c6de18 1199 bool io_interrupted = false;
bd5b0738 1200
78997b56
CY		 1201 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1202 if (i + 1 < dpolicy->granularity)
1203 break;
bd5b0738 1204 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1205
1206 mutex_lock(&dcc->cmd_lock);
1207 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1208 blk_start_plug(&plug);
bd5b0738
CY		 1209 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1210 f2fs_bug_on(sbi, dc->state != D_PREP);
1211
ecc9aa00
CY		 1212 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1213 !is_idle(sbi)) {
e6c6de18 1214 io_interrupted = true;
ecc9aa00 1215 goto skip;
969d1b18 1216 }
e6c6de18 1217
78997b56 1218 __submit_discard_cmd(sbi, dpolicy, dc);
ecc9aa00
CY		 1219 issued++;
1220skip:
1221 if (++iter >= dpolicy->max_requests)
33da62cf 1222 break;
bd5b0738 1223 }
33da62cf
CY		 1224 blk_finish_plug(&plug);
1225 mutex_unlock(&dcc->cmd_lock);
1226
1227 if (iter >= dpolicy->max_requests)
1228 break;
bd5b0738 1229 }
969d1b18 1230
e6c6de18
CY		 1231 if (!issued && io_interrupted)
1232 issued = -1;
1233
969d1b18
CY		 1234 return issued;
1235}
1236
cf5c759f 1237static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1238{
1239 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1240 struct list_head *pend_list;
1241 struct discard_cmd *dc, *tmp;
1242 int i;
cf5c759f 1243 bool dropped = false;
969d1b18
CY		 1244
1245 mutex_lock(&dcc->cmd_lock);
1246 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1247 pend_list = &dcc->pend_list[i];
1248 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1249 f2fs_bug_on(sbi, dc->state != D_PREP);
1250 __remove_discard_cmd(sbi, dc);
cf5c759f 1251 dropped = true;
969d1b18
CY		 1252 }
1253 }
1254 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1255
1256 return dropped;
bd5b0738
CY		 1257}
1258
0ea80512 1259static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1260 struct discard_cmd *dc)
1261{
1262 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1263 unsigned int len = 0;
2a510c00
CY		 1264
1265 wait_for_completion_io(&dc->wait);
1266 mutex_lock(&dcc->cmd_lock);
1267 f2fs_bug_on(sbi, dc->state != D_DONE);
1268 dc->ref--;
0ea80512
CY		 1269 if (!dc->ref) {
1270 if (!dc->error)
1271 len = dc->len;
2a510c00 1272 __remove_discard_cmd(sbi, dc);
0ea80512 1273 }
2a510c00 1274 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1275
1276 return len;
2a510c00
CY		 1277}
1278
0ea80512 1279static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1280 struct discard_policy *dpolicy,
1281 block_t start, block_t end)
63a94fa1
CY		 1282{
1283 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1284 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1285 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1286 struct discard_cmd *dc, *tmp;
6afae633 1287 bool need_wait;
0ea80512 1288 unsigned int trimmed = 0;
6afae633
CY		 1289
1290next:
1291 need_wait = false;
63a94fa1
CY		 1292
1293 mutex_lock(&dcc->cmd_lock);
1294 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1295 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1296 continue;
78997b56 1297 if (dc->len < dpolicy->granularity)
8412663d 1298 continue;
78997b56 1299 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1300 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1301 if (!dc->error)
1302 trimmed += dc->len;
63a94fa1 1303 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1304 } else {
1305 dc->ref++;
1306 need_wait = true;
1307 break;
63a94fa1
CY		 1308 }
1309 }
1310 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1311
1312 if (need_wait) {
0ea80512 1313 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1314 goto next;
1315 }
0ea80512
CY		 1316
1317 return trimmed;
63a94fa1
CY		 1318}
1319
78997b56
CY		 1320static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1321 struct discard_policy *dpolicy)
8412663d 1322{
78997b56 1323 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
8412663d
CY		 1324}
1325
004b6862
CY		 1326/* This should be covered by global mutex, &sit_i->sentry_lock */
1327void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1328{
1329 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1330 struct discard_cmd *dc;
ec9895ad 1331 bool need_wait = false;
004b6862
CY		 1332
1333 mutex_lock(&dcc->cmd_lock);
004b6862
CY		 1334 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1335 if (dc) {
ec9895ad
CY		 1336 if (dc->state == D_PREP) {
1337 __punch_discard_cmd(sbi, dc, blkaddr);
1338 } else {
1339 dc->ref++;
1340 need_wait = true;
1341 }
275b66b0 1342 }
d431413f 1343 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1344
2a510c00
CY		 1345 if (need_wait)
1346 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1347}
1348
cce13252
CY		 1349void stop_discard_thread(struct f2fs_sb_info *sbi)
1350{
1351 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1352
1353 if (dcc && dcc->f2fs_issue_discard) {
1354 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1355
1356 dcc->f2fs_issue_discard = NULL;
1357 kthread_stop(discard_thread);
ec9895ad 1358 }
d431413f
CY		 1359}
1360
8412663d 1361/* This comes from f2fs_put_super */
cf5c759f 1362bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
969d1b18
CY		 1363{
1364 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1365 struct discard_policy dpolicy;
cf5c759f 1366 bool dropped;
969d1b18 1367
78997b56
CY		 1368 init_discard_policy(&dpolicy, DPOLICY_UMOUNT, dcc->discard_granularity);
1369 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1370 dropped = __drop_discard_cmd(sbi);
78997b56 1371 __wait_all_discard_cmd(sbi, &dpolicy);
cf5c759f
CY		 1372
1373 return dropped;
969d1b18
CY		 1374}
1375
15469963
JK		 1376static int issue_discard_thread(void *data)
1377{
1378 struct f2fs_sb_info *sbi = data;
1379 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1380 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1381 struct discard_policy dpolicy;
969d1b18
CY		 1382 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1383 int issued;
15469963 1384
1d7be270 1385 set_freezable();
15469963 1386
1d7be270 1387 do {
78997b56
CY		 1388 init_discard_policy(&dpolicy, DPOLICY_BG,
1389 dcc->discard_granularity);
1390
969d1b18
CY		 1391 wait_event_interruptible_timeout(*q,
1392 kthread_should_stop() || freezing(current) ||
1393 dcc->discard_wake,
1394 msecs_to_jiffies(wait_ms));
1d7be270
JK		 1395 if (try_to_freeze())
1396 continue;
1397 if (kthread_should_stop())
1398 return 0;
15469963 1399
5f656541 1400 if (dcc->discard_wake) {
969d1b18 1401 dcc->discard_wake = 0;
5f656541 1402 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
78997b56
CY		 1403 init_discard_policy(&dpolicy,
1404 DPOLICY_FORCE, 1);
5f656541 1405 }
969d1b18 1406
dc6febb6
CY		 1407 sb_start_intwrite(sbi->sb);
1408
78997b56 1409 issued = __issue_discard_cmd(sbi, &dpolicy);
969d1b18 1410 if (issued) {
78997b56
CY		 1411 __wait_all_discard_cmd(sbi, &dpolicy);
1412 wait_ms = dpolicy.min_interval;
969d1b18 1413 } else {
78997b56 1414 wait_ms = dpolicy.max_interval;
969d1b18 1415 }
1d7be270 1416
dc6febb6 1417 sb_end_intwrite(sbi->sb);
1d7be270 1418
1d7be270
JK		 1419 } while (!kthread_should_stop());
1420 return 0;
15469963
JK		 1421}
1422
f46e8809 1423#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1424static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1425 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1426{
92592285 1427 sector_t sector, nr_sects;
10a875f8 1428 block_t lblkstart = blkstart;
3c62be17
JK		 1429 int devi = 0;
1430
1431 if (sbi->s_ndevs) {
1432 devi = f2fs_target_device_index(sbi, blkstart);
1433 blkstart -= FDEV(devi).start_blk;
1434 }
f46e8809
DLM		 1435
1436 /*
1437 * We need to know the type of the zone: for conventional zones,
1438 * use regular discard if the drive supports it. For sequential
1439 * zones, reset the zone write pointer.
1440 */
3c62be17 1441 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1442
1443 case BLK_ZONE_TYPE_CONVENTIONAL:
1444 if (!blk_queue_discard(bdev_get_queue(bdev)))
1445 return 0;
c81abe34 1446 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1447 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1448 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1449 sector = SECTOR_FROM_BLOCK(blkstart);
1450 nr_sects = SECTOR_FROM_BLOCK(blklen);
1451
1452 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1453 nr_sects != bdev_zone_sectors(bdev)) {
1454 f2fs_msg(sbi->sb, KERN_INFO,
1455 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1456 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1457 blkstart, blklen);
1458 return -EIO;
1459 }
d50aaeec 1460 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1461 return blkdev_reset_zones(bdev, sector,
1462 nr_sects, GFP_NOFS);
1463 default:
1464 /* Unknown zone type: broken device ? */
1465 return -EIO;
1466 }
1467}
1468#endif
1469
3c62be17
JK		 1470static int __issue_discard_async(struct f2fs_sb_info *sbi,
1471 struct block_device *bdev, block_t blkstart, block_t blklen)
1472{
1473#ifdef CONFIG_BLK_DEV_ZONED
1474 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1475 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1476 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1477#endif
c81abe34 1478 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1479}
1480
1e87a78d 1481static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1482 block_t blkstart, block_t blklen)
1483{
3c62be17
JK		 1484 sector_t start = blkstart, len = 0;
1485 struct block_device *bdev;
a66cdd98
JK		 1486 struct seg_entry *se;
1487 unsigned int offset;
1488 block_t i;
3c62be17
JK		 1489 int err = 0;
1490
1491 bdev = f2fs_target_device(sbi, blkstart, NULL);
1492
1493 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1494 if (i != start) {
1495 struct block_device *bdev2 =
1496 f2fs_target_device(sbi, i, NULL);
1497
1498 if (bdev2 != bdev) {
1499 err = __issue_discard_async(sbi, bdev,
1500 start, len);
1501 if (err)
1502 return err;
1503 bdev = bdev2;
1504 start = i;
1505 len = 0;
1506 }
1507 }
a66cdd98 1508
a66cdd98
JK		 1509 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1510 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1511
1512 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1513 sbi->discard_blks--;
1514 }
f46e8809 1515
3c62be17
JK		 1516 if (len)
1517 err = __issue_discard_async(sbi, bdev, start, len);
1518 return err;
1e87a78d
JK		 1519}
1520
25290fa5
JK		 1521static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1522 bool check_only)
adf4983b 1523{
b2955550
JK		 1524 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1525 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1526 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1527 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1528 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1529 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1530 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1531 unsigned int start = 0, end = -1;
c473f1a9 1532 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1533 struct discard_entry *de = NULL;
46f84c2c 1534 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1535 int i;
1536
3e025740 1537 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1538 return false;
b2955550 1539
a66cdd98
JK		 1540 if (!force) {
1541 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1542 SM_I(sbi)->dcc_info->nr_discards >=
1543 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1544 return false;
4b2fecc8
JK		 1545 }
1546
b2955550
JK		 1547 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1548 for (i = 0; i < entries; i++)
a66cdd98 1549 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1550 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1551
0b54fb84
JK		 1552 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1553 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1554 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1555 if (start >= max_blocks)
1556 break;
1557
1558 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1559 if (force && start && end != max_blocks
1560 && (end - start) < cpc->trim_minlen)
1561 continue;
1562
25290fa5
JK		 1563 if (check_only)
1564 return true;
1565
a7eeb823
CY		 1566 if (!de) {
1567 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1568 GFP_F2FS_ZERO);
1569 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1570 list_add_tail(&de->list, head);
1571 }
1572
1573 for (i = start; i < end; i++)
1574 __set_bit_le(i, (void *)de->discard_map);
1575
1576 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1577 }
25290fa5 1578 return false;
b2955550
JK		 1579}
1580
4b2fecc8
JK		 1581void release_discard_addrs(struct f2fs_sb_info *sbi)
1582{
46f84c2c 1583 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1584 struct discard_entry *entry, *this;
1585
1586 /* drop caches */
1587 list_for_each_entry_safe(entry, this, head, list) {
1588 list_del(&entry->list);
1589 kmem_cache_free(discard_entry_slab, entry);
1590 }
1591}
1592
0a8165d7 1593/*
351df4b2
JK		 1594 * Should call clear_prefree_segments after checkpoint is done.
1595 */
1596static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1597{
1598 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1599 unsigned int segno;
351df4b2
JK		 1600
1601 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1602 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1603 __set_test_and_free(sbi, segno);
351df4b2
JK		 1604 mutex_unlock(&dirty_i->seglist_lock);
1605}
1606
836b5a63 1607void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1608{
969d1b18
CY		 1609 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1610 struct list_head *head = &dcc->entry_list;
2d7b822a 1611 struct discard_entry *entry, *this;
351df4b2 1612 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1613 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1614 unsigned int start = 0, end = -1;
36abef4e 1615 unsigned int secno, start_segno;
c473f1a9 1616 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK		 1617
1618 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1619
351df4b2 1620 while (1) {
29e59c14 1621 int i;
7cd8558b
JK		 1622 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1623 if (start >= MAIN_SEGS(sbi))
351df4b2 1624 break;
7cd8558b
JK		 1625 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1626 start + 1);
29e59c14
CL		 1627
1628 for (i = start; i < end; i++)
1629 clear_bit(i, prefree_map);
1630
1631 dirty_i->nr_dirty[PRE] -= end - start;
1632
650d3c4e 1633 if (!test_opt(sbi, DISCARD))
29e59c14 1634 continue;
351df4b2 1635
650d3c4e
YH		 1636 if (force && start >= cpc->trim_start &&
1637 (end - 1) <= cpc->trim_end)
1638 continue;
1639
36abef4e
JK		 1640 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1641 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1642 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1643 continue;
1644 }
1645next:
4ddb1a4d
JK		 1646 secno = GET_SEC_FROM_SEG(sbi, start);
1647 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1648 if (!IS_CURSEC(sbi, secno) &&
302bd348 1649 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1650 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1651 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1652
1653 start = start_segno + sbi->segs_per_sec;
1654 if (start < end)
1655 goto next;
8b107f5b
JK		 1656 else
1657 end = start - 1;
351df4b2
JK		 1658 }
1659 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1660
1661 /* send small discards */
2d7b822a 1662 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1663 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1664 bool is_valid = test_bit_le(0, entry->discard_map);
1665
1666find_next:
1667 if (is_valid) {
1668 next_pos = find_next_zero_bit_le(entry->discard_map,
1669 sbi->blocks_per_seg, cur_pos);
1670 len = next_pos - cur_pos;
1671
acfd2810
DLM		 1672 if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1673 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1674 goto skip;
1675
1676 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1677 len);
a7eeb823
CY		 1678 total_len += len;
1679 } else {
1680 next_pos = find_next_bit_le(entry->discard_map,
1681 sbi->blocks_per_seg, cur_pos);
1682 }
836b5a63 1683skip:
a7eeb823
CY		 1684 cur_pos = next_pos;
1685 is_valid = !is_valid;
1686
1687 if (cur_pos < sbi->blocks_per_seg)
1688 goto find_next;
1689
b2955550 1690 list_del(&entry->list);
969d1b18 1691 dcc->nr_discards -= total_len;
b2955550
JK		 1692 kmem_cache_free(discard_entry_slab, entry);
1693 }
34e159da 1694
01983c71 1695 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1696}
1697
78997b56
CY		 1698void init_discard_policy(struct discard_policy *dpolicy,
1699 int discard_type, unsigned int granularity)
ecc9aa00 1700{
78997b56
CY		 1701 /* common policy */
1702 dpolicy->type = discard_type;
ecc9aa00 1703 dpolicy->sync = true;
78997b56
CY		 1704 dpolicy->granularity = granularity;
1705
1706 if (discard_type == DPOLICY_BG) {
1707 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1708 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1709 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1710 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1711 dpolicy->io_aware = true;
1712 } else if (discard_type == DPOLICY_FORCE) {
1713 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1714 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1715 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1716 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1717 dpolicy->io_aware = true;
1718 } else if (discard_type == DPOLICY_FSTRIM) {
1719 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1720 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1721 dpolicy->io_aware = false;
1722 } else if (discard_type == DPOLICY_UMOUNT) {
1723 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1724 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1725 dpolicy->io_aware = false;
1726 }
ecc9aa00
CY		 1727}
1728
8ed59745 1729static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1730{
15469963 1731 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1732 struct discard_cmd_control *dcc;
ba48a33e 1733 int err = 0, i;
0b54fb84
JK		 1734
1735 if (SM_I(sbi)->dcc_info) {
1736 dcc = SM_I(sbi)->dcc_info;
1737 goto init_thread;
1738 }
1739
1740 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1741 if (!dcc)
1742 return -ENOMEM;
1743
969d1b18 1744 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1745 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 1746 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 1747 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 1748 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 1749 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 1750 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1751 atomic_set(&dcc->issued_discard, 0);
1752 atomic_set(&dcc->issing_discard, 0);
5f32366a 1753 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1754 dcc->nr_discards = 0;
d618ebaf 1755 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1756 dcc->undiscard_blks = 0;
004b6862 1757 dcc->root = RB_ROOT;
0b54fb84 1758
15469963 1759 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1760 SM_I(sbi)->dcc_info = dcc;
1761init_thread:
15469963
JK		 1762 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1763 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1764 if (IS_ERR(dcc->f2fs_issue_discard)) {
1765 err = PTR_ERR(dcc->f2fs_issue_discard);
1766 kfree(dcc);
1767 SM_I(sbi)->dcc_info = NULL;
1768 return err;
1769 }
1770
0b54fb84
JK		 1771 return err;
1772}
1773
f099405f 1774static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1775{
1776 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1777
f099405f
CY		 1778 if (!dcc)
1779 return;
1780
cce13252 1781 stop_discard_thread(sbi);
f099405f
CY		 1782
1783 kfree(dcc);
1784 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1785}
1786
184a5cd2 1787static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1788{
1789 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1790
1791 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1792 sit_i->dirty_sentries++;
184a5cd2
CY		 1793 return false;
1794 }
1795
1796 return true;
351df4b2
JK		 1797}
1798
1799static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1800 unsigned int segno, int modified)
1801{
1802 struct seg_entry *se = get_seg_entry(sbi, segno);
1803 se->type = type;
1804 if (modified)
1805 __mark_sit_entry_dirty(sbi, segno);
1806}
1807
1808static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1809{
1810 struct seg_entry *se;
1811 unsigned int segno, offset;
1812 long int new_vblocks;
6415fedc
YS		 1813 bool exist;
1814#ifdef CONFIG_F2FS_CHECK_FS
1815 bool mir_exist;
1816#endif
351df4b2
JK		 1817
1818 segno = GET_SEGNO(sbi, blkaddr);
1819
1820 se = get_seg_entry(sbi, segno);
1821 new_vblocks = se->valid_blocks + del;
491c0854 1822 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1823
9850cf4a 1824 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1825 (new_vblocks > sbi->blocks_per_seg)));
1826
1827 se->valid_blocks = new_vblocks;
1828 se->mtime = get_mtime(sbi);
1829 SIT_I(sbi)->max_mtime = se->mtime;
1830
1831 /* Update valid block bitmap */
1832 if (del > 0) {
6415fedc 1833 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1834#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1835 mir_exist = f2fs_test_and_set_bit(offset,
1836 se->cur_valid_map_mir);
1837 if (unlikely(exist != mir_exist)) {
1838 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1839 "when setting bitmap, blk:%u, old bit:%d",
1840 blkaddr, exist);
05796763 1841 f2fs_bug_on(sbi, 1);
6415fedc 1842 }
355e7891 1843#endif
6415fedc
YS		 1844 if (unlikely(exist)) {
1845 f2fs_msg(sbi->sb, KERN_ERR,
1846 "Bitmap was wrongly set, blk:%u", blkaddr);
1847 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1848 se->valid_blocks--;
1849 del = 0;
355e7891 1850 }
6415fedc 1851
3e025740
JK		 1852 if (f2fs_discard_en(sbi) &&
1853 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1854 sbi->discard_blks--;
720037f9
JK		 1855
1856 /* don't overwrite by SSR to keep node chain */
1857 if (se->type == CURSEG_WARM_NODE) {
1858 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1859 se->ckpt_valid_blocks++;
1860 }
351df4b2 1861 } else {
6415fedc 1862 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1863#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1864 mir_exist = f2fs_test_and_clear_bit(offset,
1865 se->cur_valid_map_mir);
1866 if (unlikely(exist != mir_exist)) {
1867 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1868 "when clearing bitmap, blk:%u, old bit:%d",
1869 blkaddr, exist);
05796763 1870 f2fs_bug_on(sbi, 1);
6415fedc 1871 }
355e7891 1872#endif
6415fedc
YS		 1873 if (unlikely(!exist)) {
1874 f2fs_msg(sbi->sb, KERN_ERR,
1875 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1876 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1877 se->valid_blocks++;
1878 del = 0;
355e7891 1879 }
6415fedc 1880
3e025740
JK		 1881 if (f2fs_discard_en(sbi) &&
1882 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1883 sbi->discard_blks++;
351df4b2
JK		 1884 }
1885 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1886 se->ckpt_valid_blocks += del;
1887
1888 __mark_sit_entry_dirty(sbi, segno);
1889
1890 /* update total number of valid blocks to be written in ckpt area */
1891 SIT_I(sbi)->written_valid_blocks += del;
1892
1893 if (sbi->segs_per_sec > 1)
1894 get_sec_entry(sbi, segno)->valid_blocks += del;
1895}
1896
351df4b2
JK		 1897void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1898{
1899 unsigned int segno = GET_SEGNO(sbi, addr);
1900 struct sit_info *sit_i = SIT_I(sbi);
1901
9850cf4a 1902 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1903 if (addr == NEW_ADDR)
1904 return;
1905
1906 /* add it into sit main buffer */
3d26fa6b 1907 down_write(&sit_i->sentry_lock);
351df4b2
JK		 1908
1909 update_sit_entry(sbi, addr, -1);
1910
1911 /* add it into dirty seglist */
1912 locate_dirty_segment(sbi, segno);
1913
3d26fa6b 1914 up_write(&sit_i->sentry_lock);
351df4b2
JK		 1915}
1916
6e2c64ad
JK		 1917bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1918{
1919 struct sit_info *sit_i = SIT_I(sbi);
1920 unsigned int segno, offset;
1921 struct seg_entry *se;
1922 bool is_cp = false;
1923
1924 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1925 return true;
1926
3d26fa6b 1927 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1928
1929 segno = GET_SEGNO(sbi, blkaddr);
1930 se = get_seg_entry(sbi, segno);
1931 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1932
1933 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1934 is_cp = true;
1935
3d26fa6b 1936 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1937
1938 return is_cp;
1939}
1940
0a8165d7 1941/*
351df4b2
JK		 1942 * This function should be resided under the curseg_mutex lock
1943 */
1944static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1945 struct f2fs_summary *sum)
351df4b2
JK		 1946{
1947 struct curseg_info *curseg = CURSEG_I(sbi, type);
1948 void *addr = curseg->sum_blk;
e79efe3b 1949 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1950 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1951}
1952
0a8165d7 1953/*
351df4b2
JK		 1954 * Calculate the number of current summary pages for writing
1955 */
3fa06d7b 1956int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1957{
351df4b2 1958 int valid_sum_count = 0;
9a47938b 1959 int i, sum_in_page;
351df4b2
JK		 1960
1961 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1962 if (sbi->ckpt->alloc_type[i] == SSR)
1963 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1964 else {
1965 if (for_ra)
1966 valid_sum_count += le16_to_cpu(
1967 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1968 else
1969 valid_sum_count += curseg_blkoff(sbi, i);
1970 }
351df4b2
JK		 1971 }
1972
09cbfeaf 1973 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1974 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1975 if (valid_sum_count <= sum_in_page)
351df4b2 1976 return 1;
9a47938b 1977 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1978 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1979 return 2;
1980 return 3;
1981}
1982
0a8165d7 1983/*
351df4b2
JK		 1984 * Caller should put this summary page
1985 */
1986struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1987{
1988 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1989}
1990
381722d2 1991void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1992{
1993 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2 1994
0537b811 1995 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 1996 set_page_dirty(page);
1997 f2fs_put_page(page, 1);
1998}
1999
381722d2
CY		 2000static void write_sum_page(struct f2fs_sb_info *sbi,
2001 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2002{
2003 update_meta_page(sbi, (void *)sum_blk, blk_addr);
2004}
2005
b7ad7512
CY		 2006static void write_current_sum_page(struct f2fs_sb_info *sbi,
2007 int type, block_t blk_addr)
2008{
2009 struct curseg_info *curseg = CURSEG_I(sbi, type);
2010 struct page *page = grab_meta_page(sbi, blk_addr);
2011 struct f2fs_summary_block *src = curseg->sum_blk;
2012 struct f2fs_summary_block *dst;
2013
2014 dst = (struct f2fs_summary_block *)page_address(page);
2015
2016 mutex_lock(&curseg->curseg_mutex);
2017
2018 down_read(&curseg->journal_rwsem);
2019 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2020 up_read(&curseg->journal_rwsem);
2021
2022 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2023 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2024
2025 mutex_unlock(&curseg->curseg_mutex);
2026
2027 set_page_dirty(page);
2028 f2fs_put_page(page, 1);
2029}
2030
a7881893
JK		 2031static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2032{
2033 struct curseg_info *curseg = CURSEG_I(sbi, type);
2034 unsigned int segno = curseg->segno + 1;
2035 struct free_segmap_info *free_i = FREE_I(sbi);
2036
2037 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2038 return !test_bit(segno, free_i->free_segmap);
2039 return 0;
2040}
2041
0a8165d7 2042/*
351df4b2
JK		 2043 * Find a new segment from the free segments bitmap to right order
2044 * This function should be returned with success, otherwise BUG
2045 */
2046static void get_new_segment(struct f2fs_sb_info *sbi,
2047 unsigned int *newseg, bool new_sec, int dir)
2048{
2049 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2050 unsigned int segno, secno, zoneno;
7cd8558b 2051 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2052 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2053 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2054 unsigned int left_start = hint;
2055 bool init = true;
2056 int go_left = 0;
2057 int i;
2058
1a118ccf 2059 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2060
2061 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2062 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2063 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2064 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2065 goto got_it;
2066 }
2067find_other_zone:
7cd8558b
JK		 2068 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2069 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2070 if (dir == ALLOC_RIGHT) {
2071 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2072 MAIN_SECS(sbi), 0);
2073 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2074 } else {
2075 go_left = 1;
2076 left_start = hint - 1;
2077 }
2078 }
2079 if (go_left == 0)
2080 goto skip_left;
2081
2082 while (test_bit(left_start, free_i->free_secmap)) {
2083 if (left_start > 0) {
2084 left_start--;
2085 continue;
2086 }
2087 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2088 MAIN_SECS(sbi), 0);
2089 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2090 break;
2091 }
2092 secno = left_start;
2093skip_left:
4ddb1a4d
JK		 2094 segno = GET_SEG_FROM_SEC(sbi, secno);
2095 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2096
2097 /* give up on finding another zone */
2098 if (!init)
2099 goto got_it;
2100 if (sbi->secs_per_zone == 1)
2101 goto got_it;
2102 if (zoneno == old_zoneno)
2103 goto got_it;
2104 if (dir == ALLOC_LEFT) {
2105 if (!go_left && zoneno + 1 >= total_zones)
2106 goto got_it;
2107 if (go_left && zoneno == 0)
2108 goto got_it;
2109 }
2110 for (i = 0; i < NR_CURSEG_TYPE; i++)
2111 if (CURSEG_I(sbi, i)->zone == zoneno)
2112 break;
2113
2114 if (i < NR_CURSEG_TYPE) {
2115 /* zone is in user, try another */
2116 if (go_left)
2117 hint = zoneno * sbi->secs_per_zone - 1;
2118 else if (zoneno + 1 >= total_zones)
2119 hint = 0;
2120 else
2121 hint = (zoneno + 1) * sbi->secs_per_zone;
2122 init = false;
2123 goto find_other_zone;
2124 }
2125got_it:
2126 /* set it as dirty segment in free segmap */
9850cf4a 2127 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2128 __set_inuse(sbi, segno);
2129 *newseg = segno;
1a118ccf 2130 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2131}
2132
2133static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2134{
2135 struct curseg_info *curseg = CURSEG_I(sbi, type);
2136 struct summary_footer *sum_footer;
2137
2138 curseg->segno = curseg->next_segno;
4ddb1a4d 2139 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2140 curseg->next_blkoff = 0;
2141 curseg->next_segno = NULL_SEGNO;
2142
2143 sum_footer = &(curseg->sum_blk->footer);
2144 memset(sum_footer, 0, sizeof(struct summary_footer));
2145 if (IS_DATASEG(type))
2146 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2147 if (IS_NODESEG(type))
2148 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2149 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2150}
2151
7a20b8a6
JK		 2152static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2153{
a7881893
JK		 2154 /* if segs_per_sec is large than 1, we need to keep original policy. */
2155 if (sbi->segs_per_sec != 1)
2156 return CURSEG_I(sbi, type)->segno;
2157
482eb99a
YS		 2158 if (test_opt(sbi, NOHEAP) &&
2159 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2160 return 0;
2161
e066b83c
JK		 2162 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2163 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
7a20b8a6
JK		 2164 return CURSEG_I(sbi, type)->segno;
2165}
2166
0a8165d7 2167/*
351df4b2
JK		 2168 * Allocate a current working segment.
2169 * This function always allocates a free segment in LFS manner.
2170 */
2171static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2172{
2173 struct curseg_info *curseg = CURSEG_I(sbi, type);
2174 unsigned int segno = curseg->segno;
2175 int dir = ALLOC_LEFT;
2176
2177 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2178 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2179 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2180 dir = ALLOC_RIGHT;
2181
2182 if (test_opt(sbi, NOHEAP))
2183 dir = ALLOC_RIGHT;
2184
7a20b8a6 2185 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2186 get_new_segment(sbi, &segno, new_sec, dir);
2187 curseg->next_segno = segno;
2188 reset_curseg(sbi, type, 1);
2189 curseg->alloc_type = LFS;
2190}
2191
2192static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2193 struct curseg_info *seg, block_t start)
2194{
2195 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2196 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2197 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2198 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2199 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2200 int i, pos;
2201
2202 for (i = 0; i < entries; i++)
2203 target_map[i] = ckpt_map[i] | cur_map[i];
2204
2205 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2206
2207 seg->next_blkoff = pos;
351df4b2
JK		 2208}
2209
0a8165d7 2210/*
351df4b2
JK		 2211 * If a segment is written by LFS manner, next block offset is just obtained
2212 * by increasing the current block offset. However, if a segment is written by
2213 * SSR manner, next block offset obtained by calling __next_free_blkoff
2214 */
2215static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2216 struct curseg_info *seg)
2217{
2218 if (seg->alloc_type == SSR)
2219 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2220 else
2221 seg->next_blkoff++;
2222}
2223
0a8165d7 2224/*
e1c42045 2225 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2226 * manner, so it should recover the existing segment information of valid blocks
2227 */
025d63a4 2228static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2229{
2230 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2231 struct curseg_info *curseg = CURSEG_I(sbi, type);
2232 unsigned int new_segno = curseg->next_segno;
2233 struct f2fs_summary_block *sum_node;
2234 struct page *sum_page;
2235
2236 write_sum_page(sbi, curseg->sum_blk,
2237 GET_SUM_BLOCK(sbi, curseg->segno));
2238 __set_test_and_inuse(sbi, new_segno);
2239
2240 mutex_lock(&dirty_i->seglist_lock);
2241 __remove_dirty_segment(sbi, new_segno, PRE);
2242 __remove_dirty_segment(sbi, new_segno, DIRTY);
2243 mutex_unlock(&dirty_i->seglist_lock);
2244
2245 reset_curseg(sbi, type, 1);
2246 curseg->alloc_type = SSR;
2247 __next_free_blkoff(sbi, curseg, 0);
2248
025d63a4
CY		 2249 sum_page = get_sum_page(sbi, new_segno);
2250 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2251 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2252 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2253}
2254
43727527
JK		 2255static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2256{
2257 struct curseg_info *curseg = CURSEG_I(sbi, type);
2258 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2259 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2260 int i, cnt;
2261 bool reversed = false;
c192f7a4
JK		 2262
2263 /* need_SSR() already forces to do this */
e066b83c
JK		 2264 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2265 curseg->next_segno = segno;
c192f7a4 2266 return 1;
e066b83c 2267 }
43727527 2268
70d625cb
JK		 2269 /* For node segments, let's do SSR more intensively */
2270 if (IS_NODESEG(type)) {
d27c3d89
CY		 2271 if (type >= CURSEG_WARM_NODE) {
2272 reversed = true;
2273 i = CURSEG_COLD_NODE;
2274 } else {
2275 i = CURSEG_HOT_NODE;
2276 }
2277 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2278 } else {
d27c3d89
CY		 2279 if (type >= CURSEG_WARM_DATA) {
2280 reversed = true;
2281 i = CURSEG_COLD_DATA;
2282 } else {
2283 i = CURSEG_HOT_DATA;
2284 }
2285 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2286 }
43727527 2287
d27c3d89 2288 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2289 if (i == type)
2290 continue;
e066b83c
JK		 2291 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2292 curseg->next_segno = segno;
43727527 2293 return 1;
e066b83c 2294 }
c192f7a4 2295 }
43727527
JK		 2296 return 0;
2297}
2298
351df4b2
JK		 2299/*
2300 * flush out current segment and replace it with new segment
2301 * This function should be returned with success, otherwise BUG
2302 */
2303static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2304 int type, bool force)
2305{
a7881893
JK		 2306 struct curseg_info *curseg = CURSEG_I(sbi, type);
2307
7b405275 2308 if (force)
351df4b2 2309 new_curseg(sbi, type, true);
5b6c6be2
JK		 2310 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2311 type == CURSEG_WARM_NODE)
351df4b2 2312 new_curseg(sbi, type, false);
a7881893
JK		 2313 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2314 new_curseg(sbi, type, false);
351df4b2 2315 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2316 change_curseg(sbi, type);
351df4b2
JK		 2317 else
2318 new_curseg(sbi, type, false);
dcdfff65 2319
a7881893 2320 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2321}
2322
2323void allocate_new_segments(struct f2fs_sb_info *sbi)
2324{
6ae1be13
JK		 2325 struct curseg_info *curseg;
2326 unsigned int old_segno;
351df4b2
JK		 2327 int i;
2328
3d26fa6b
CY		 2329 down_write(&SIT_I(sbi)->sentry_lock);
2330
6ae1be13
JK		 2331 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2332 curseg = CURSEG_I(sbi, i);
2333 old_segno = curseg->segno;
2334 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2335 locate_dirty_segment(sbi, old_segno);
2336 }
3d26fa6b
CY		 2337
2338 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2339}
2340
2341static const struct segment_allocation default_salloc_ops = {
2342 .allocate_segment = allocate_segment_by_default,
2343};
2344
25290fa5
JK		 2345bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2346{
2347 __u64 trim_start = cpc->trim_start;
2348 bool has_candidate = false;
2349
3d26fa6b 2350 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2351 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2352 if (add_discard_addrs(sbi, cpc, true)) {
2353 has_candidate = true;
2354 break;
2355 }
2356 }
3d26fa6b 2357 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2358
2359 cpc->trim_start = trim_start;
2360 return has_candidate;
2361}
2362
4b2fecc8
JK		 2363int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2364{
f7ef9b83
JK		 2365 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2366 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
8412663d
CY		 2367 unsigned int start_segno, end_segno, cur_segno;
2368 block_t start_block, end_block;
4b2fecc8 2369 struct cp_control cpc;
78997b56 2370 struct discard_policy dpolicy;
0ea80512 2371 unsigned long long trimmed = 0;
c34f42e2 2372 int err = 0;
4b2fecc8 2373
836b5a63 2374 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2375 return -EINVAL;
2376
7cd8558b 2377 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 2378 goto out;
2379
ed214a11
YH		 2380 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2381 f2fs_msg(sbi->sb, KERN_WARNING,
2382 "Found FS corruption, run fsck to fix.");
2383 goto out;
2384 }
2385
4b2fecc8 2386 /* start/end segment number in main_area */
7cd8558b
JK		 2387 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2388 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2389 GET_SEGNO(sbi, end);
8412663d 2390
4b2fecc8 2391 cpc.reason = CP_DISCARD;
836b5a63 2392 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 2393
2394 /* do checkpoint to issue discard commands safely */
8412663d
CY		 2395 for (cur_segno = start_segno; cur_segno <= end_segno;
2396 cur_segno = cpc.trim_end + 1) {
2397 cpc.trim_start = cur_segno;
a66cdd98
JK		 2398
2399 if (sbi->discard_blks == 0)
2400 break;
2401 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2402 cpc.trim_end = end_segno;
2403 else
2404 cpc.trim_end = min_t(unsigned int,
8412663d 2405 rounddown(cur_segno +
bba681cb
JK		 2406 BATCHED_TRIM_SEGMENTS(sbi),
2407 sbi->segs_per_sec) - 1, end_segno);
2408
2409 mutex_lock(&sbi->gc_mutex);
c34f42e2 2410 err = write_checkpoint(sbi, &cpc);
bba681cb 2411 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 2412 if (err)
2413 break;
74fa5f3d
CY		 2414
2415 schedule();
bba681cb 2416 }
8412663d
CY		 2417
2418 start_block = START_BLOCK(sbi, start_segno);
2419 end_block = START_BLOCK(sbi, min(cur_segno, end_segno) + 1);
2420
78997b56
CY		 2421 init_discard_policy(&dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2422 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
0ea80512
CY		 2423 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2424 start_block, end_block);
4b2fecc8 2425out:
0ea80512 2426 range->len = F2FS_BLK_TO_BYTES(trimmed);
c34f42e2 2427 return err;
4b2fecc8
JK		 2428}
2429
351df4b2
JK		 2430static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2431{
2432 struct curseg_info *curseg = CURSEG_I(sbi, type);
2433 if (curseg->next_blkoff < sbi->blocks_per_seg)
2434 return true;
2435 return false;
2436}
2437
4f0a03d3
HL		 2438int rw_hint_to_seg_type(enum rw_hint hint)
2439{
2440 switch (hint) {
2441 case WRITE_LIFE_SHORT:
2442 return CURSEG_HOT_DATA;
2443 case WRITE_LIFE_EXTREME:
2444 return CURSEG_COLD_DATA;
2445 default:
2446 return CURSEG_WARM_DATA;
2447 }
2448}
2449
81377bd6 2450static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2451{
81377bd6 2452 if (fio->type == DATA)
351df4b2
JK		 2453 return CURSEG_HOT_DATA;
2454 else
2455 return CURSEG_HOT_NODE;
2456}
2457
81377bd6 2458static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2459{
81377bd6
JK		 2460 if (fio->type == DATA) {
2461 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2462
2463 if (S_ISDIR(inode->i_mode))
2464 return CURSEG_HOT_DATA;
2465 else
2466 return CURSEG_COLD_DATA;
2467 } else {
81377bd6 2468 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2469 return CURSEG_WARM_NODE;
351df4b2
JK		 2470 else
2471 return CURSEG_COLD_NODE;
2472 }
2473}
2474
81377bd6 2475static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2476{
81377bd6
JK		 2477 if (fio->type == DATA) {
2478 struct inode *inode = fio->page->mapping->host;
351df4b2 2479
81377bd6 2480 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2481 return CURSEG_COLD_DATA;
ef095d19
JK		 2482 if (is_inode_flag_set(inode, FI_HOT_DATA))
2483 return CURSEG_HOT_DATA;
4f0a03d3 2484 return rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 2485 } else {
81377bd6
JK		 2486 if (IS_DNODE(fio->page))
2487 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2488 CURSEG_HOT_NODE;
ef095d19 2489 return CURSEG_COLD_NODE;
351df4b2
JK		 2490 }
2491}
2492
81377bd6 2493static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2494{
a912b54d
JK		 2495 int type = 0;
2496
81377bd6 2497 switch (fio->sbi->active_logs) {
351df4b2 2498 case 2:
a912b54d
JK		 2499 type = __get_segment_type_2(fio);
2500 break;
351df4b2 2501 case 4:
a912b54d
JK		 2502 type = __get_segment_type_4(fio);
2503 break;
2504 case 6:
2505 type = __get_segment_type_6(fio);
2506 break;
2507 default:
2508 f2fs_bug_on(fio->sbi, true);
351df4b2 2509 }
81377bd6 2510
a912b54d
JK		 2511 if (IS_HOT(type))
2512 fio->temp = HOT;
2513 else if (IS_WARM(type))
2514 fio->temp = WARM;
2515 else
2516 fio->temp = COLD;
2517 return type;
351df4b2
JK		 2518}
2519
bfad7c2d
JK		 2520void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2521 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2522 struct f2fs_summary *sum, int type,
2523 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2524{
2525 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2526 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 2527
2b60311d
CY		 2528 down_read(&SM_I(sbi)->curseg_lock);
2529
351df4b2 2530 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2531 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2532
2533 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2534
4e6a8d9b
JK		 2535 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2536
351df4b2
JK		 2537 /*
2538 * __add_sum_entry should be resided under the curseg_mutex
2539 * because, this function updates a summary entry in the
2540 * current summary block.
2541 */
e79efe3b 2542 __add_sum_entry(sbi, type, sum);
351df4b2 2543
351df4b2 2544 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2545
2546 stat_inc_block_count(sbi, curseg);
351df4b2 2547
65f1b80b
YS		 2548 /*
2549 * SIT information should be updated before segment allocation,
2550 * since SSR needs latest valid block information.
2551 */
2552 update_sit_entry(sbi, *new_blkaddr, 1);
2553 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2554 update_sit_entry(sbi, old_blkaddr, -1);
2555
c6f82fe9
JK		 2556 if (!__has_curseg_space(sbi, type))
2557 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 2558
351df4b2 2559 /*
65f1b80b
YS		 2560 * segment dirty status should be updated after segment allocation,
2561 * so we just need to update status only one time after previous
2562 * segment being closed.
351df4b2 2563 */
65f1b80b
YS		 2564 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2565 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 2566
3d26fa6b 2567 up_write(&sit_i->sentry_lock);
351df4b2 2568
704956ec 2569 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2570 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2571
704956ec
CY		 2572 f2fs_inode_chksum_set(sbi, page);
2573 }
2574
fb830fc5
CY		 2575 if (add_list) {
2576 struct f2fs_bio_info *io;
2577
2578 INIT_LIST_HEAD(&fio->list);
2579 fio->in_list = true;
2580 io = sbi->write_io[fio->type] + fio->temp;
2581 spin_lock(&io->io_lock);
2582 list_add_tail(&fio->list, &io->io_list);
2583 spin_unlock(&io->io_lock);
2584 }
2585
bfad7c2d 2586 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 2587
2588 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 2589}
2590
39d787be
CY		 2591static void update_device_state(struct f2fs_io_info *fio)
2592{
2593 struct f2fs_sb_info *sbi = fio->sbi;
2594 unsigned int devidx;
2595
2596 if (!sbi->s_ndevs)
2597 return;
2598
2599 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2600
2601 /* update device state for fsync */
2602 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 2603
2604 /* update device state for checkpoint */
2605 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2606 spin_lock(&sbi->dev_lock);
2607 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2608 spin_unlock(&sbi->dev_lock);
2609 }
39d787be
CY		 2610}
2611
05ca3632 2612static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2613{
81377bd6 2614 int type = __get_segment_type(fio);
0a595eba 2615 int err;
bfad7c2d 2616
0a595eba 2617reallocate:
7a9d7548 2618 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2619 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2620
351df4b2 2621 /* writeout dirty page into bdev */
b9109b0e 2622 err = f2fs_submit_page_write(fio);
0a595eba
JK		 2623 if (err == -EAGAIN) {
2624 fio->old_blkaddr = fio->new_blkaddr;
2625 goto reallocate;
39d787be
CY		 2626 } else if (!err) {
2627 update_device_state(fio);
0a595eba 2628 }
351df4b2
JK		 2629}
2630
b0af6d49
CY		 2631void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2632 enum iostat_type io_type)
351df4b2 2633{
458e6197 2634 struct f2fs_io_info fio = {
05ca3632 2635 .sbi = sbi,
458e6197 2636 .type = META,
04d328de 2637 .op = REQ_OP_WRITE,
70fd7614 2638 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2639 .old_blkaddr = page->index,
2640 .new_blkaddr = page->index,
05ca3632 2641 .page = page,
4375a336 2642 .encrypted_page = NULL,
fb830fc5 2643 .in_list = false,
458e6197
JK		 2644 };
2645
2b947003 2646 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2647 fio.op_flags &= ~REQ_META;
2b947003 2648
351df4b2 2649 set_page_writeback(page);
b9109b0e 2650 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2651
2652 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2653}
2654
05ca3632 2655void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2656{
2657 struct f2fs_summary sum;
05ca3632 2658
351df4b2 2659 set_summary(&sum, nid, 0, 0);
05ca3632 2660 do_write_page(&sum, fio);
b0af6d49
CY		 2661
2662 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2663}
2664
05ca3632 2665void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2666{
05ca3632 2667 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2668 struct f2fs_summary sum;
2669 struct node_info ni;
2670
9850cf4a 2671 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2672 get_node_info(sbi, dn->nid, &ni);
2673 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2674 do_write_page(&sum, fio);
f28b3434 2675 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2676
2677 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2678}
2679
d1b3e72d 2680int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2681{
b0af6d49
CY		 2682 int err;
2683
7a9d7548 2684 fio->new_blkaddr = fio->old_blkaddr;
05ca3632 2685 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2686
2687 err = f2fs_submit_page_bio(fio);
39d787be
CY		 2688 if (!err)
2689 update_device_state(fio);
b0af6d49
CY		 2690
2691 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2692
2693 return err;
351df4b2
JK		 2694}
2695
2b60311d
CY		 2696static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2697 unsigned int segno)
2698{
2699 int i;
2700
2701 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2702 if (CURSEG_I(sbi, i)->segno == segno)
2703 break;
2704 }
2705 return i;
2706}
2707
4356e48e 2708void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2709 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2710 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2711{
2712 struct sit_info *sit_i = SIT_I(sbi);
2713 struct curseg_info *curseg;
2714 unsigned int segno, old_cursegno;
2715 struct seg_entry *se;
2716 int type;
19f106bc 2717 unsigned short old_blkoff;
351df4b2
JK		 2718
2719 segno = GET_SEGNO(sbi, new_blkaddr);
2720 se = get_seg_entry(sbi, segno);
2721 type = se->type;
2722
2b60311d
CY		 2723 down_write(&SM_I(sbi)->curseg_lock);
2724
19f106bc
CY		 2725 if (!recover_curseg) {
2726 /* for recovery flow */
2727 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2728 if (old_blkaddr == NULL_ADDR)
2729 type = CURSEG_COLD_DATA;
2730 else
2731 type = CURSEG_WARM_DATA;
2732 }
2733 } else {
2b60311d
CY		 2734 if (IS_CURSEG(sbi, segno)) {
2735 /* se->type is volatile as SSR allocation */
2736 type = __f2fs_get_curseg(sbi, segno);
2737 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2738 } else {
351df4b2 2739 type = CURSEG_WARM_DATA;
2b60311d 2740 }
351df4b2 2741 }
19f106bc 2742
351df4b2
JK		 2743 curseg = CURSEG_I(sbi, type);
2744
2745 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2746 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2747
2748 old_cursegno = curseg->segno;
19f106bc 2749 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2750
2751 /* change the current segment */
2752 if (segno != curseg->segno) {
2753 curseg->next_segno = segno;
025d63a4 2754 change_curseg(sbi, type);
351df4b2
JK		 2755 }
2756
491c0854 2757 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2758 __add_sum_entry(sbi, type, sum);
351df4b2 2759
28bc106b 2760 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2761 update_sit_entry(sbi, new_blkaddr, 1);
2762 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2763 update_sit_entry(sbi, old_blkaddr, -1);
2764
2765 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2766 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2767
351df4b2 2768 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2769
19f106bc
CY		 2770 if (recover_curseg) {
2771 if (old_cursegno != curseg->segno) {
2772 curseg->next_segno = old_cursegno;
025d63a4 2773 change_curseg(sbi, type);
19f106bc
CY		 2774 }
2775 curseg->next_blkoff = old_blkoff;
2776 }
2777
3d26fa6b 2778 up_write(&sit_i->sentry_lock);
351df4b2 2779 mutex_unlock(&curseg->curseg_mutex);
2b60311d 2780 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 2781}
2782
528e3459
CY		 2783void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2784 block_t old_addr, block_t new_addr,
28bc106b
CY		 2785 unsigned char version, bool recover_curseg,
2786 bool recover_newaddr)
528e3459
CY		 2787{
2788 struct f2fs_summary sum;
2789
2790 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2791
28bc106b
CY		 2792 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2793 recover_curseg, recover_newaddr);
528e3459 2794
f28b3434 2795 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2796}
2797
93dfe2ac 2798void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2799 enum page_type type, bool ordered)
93dfe2ac 2800{
93dfe2ac 2801 if (PageWriteback(page)) {
4081363f
JK		 2802 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2803
b9109b0e
JK		 2804 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2805 0, page->index, type);
fec1d657
JK		 2806 if (ordered)
2807 wait_on_page_writeback(page);
2808 else
2809 wait_for_stable_page(page);
93dfe2ac
JK		 2810 }
2811}
2812
d4c759ee 2813void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 2814{
2815 struct page *cpage;
2816
5d4c0af4 2817 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2818 return;
2819
08b39fbd
CY		 2820 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2821 if (cpage) {
fec1d657 2822 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2823 f2fs_put_page(cpage, 1);
2824 }
2825}
2826
351df4b2
JK		 2827static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2828{
2829 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2830 struct curseg_info *seg_i;
2831 unsigned char *kaddr;
2832 struct page *page;
2833 block_t start;
2834 int i, j, offset;
2835
2836 start = start_sum_block(sbi);
2837
2838 page = get_meta_page(sbi, start++);
2839 kaddr = (unsigned char *)page_address(page);
2840
2841 /* Step 1: restore nat cache */
2842 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2843 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2844
2845 /* Step 2: restore sit cache */
2846 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2847 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2848 offset = 2 * SUM_JOURNAL_SIZE;
2849
2850 /* Step 3: restore summary entries */
2851 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2852 unsigned short blk_off;
2853 unsigned int segno;
2854
2855 seg_i = CURSEG_I(sbi, i);
2856 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2857 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2858 seg_i->next_segno = segno;
2859 reset_curseg(sbi, i, 0);
2860 seg_i->alloc_type = ckpt->alloc_type[i];
2861 seg_i->next_blkoff = blk_off;
2862
2863 if (seg_i->alloc_type == SSR)
2864 blk_off = sbi->blocks_per_seg;
2865
2866 for (j = 0; j < blk_off; j++) {
2867 struct f2fs_summary *s;
2868 s = (struct f2fs_summary *)(kaddr + offset);
2869 seg_i->sum_blk->entries[j] = *s;
2870 offset += SUMMARY_SIZE;
09cbfeaf 2871 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2872 SUM_FOOTER_SIZE)
2873 continue;
2874
2875 f2fs_put_page(page, 1);
2876 page = NULL;
2877
2878 page = get_meta_page(sbi, start++);
2879 kaddr = (unsigned char *)page_address(page);
2880 offset = 0;
2881 }
2882 }
2883 f2fs_put_page(page, 1);
2884 return 0;
2885}
2886
2887static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2888{
2889 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2890 struct f2fs_summary_block *sum;
2891 struct curseg_info *curseg;
2892 struct page *new;
2893 unsigned short blk_off;
2894 unsigned int segno = 0;
2895 block_t blk_addr = 0;
2896
2897 /* get segment number and block addr */
2898 if (IS_DATASEG(type)) {
2899 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2900 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2901 CURSEG_HOT_DATA]);
119ee914 2902 if (__exist_node_summaries(sbi))
351df4b2
JK		 2903 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2904 else
2905 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2906 } else {
2907 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2908 CURSEG_HOT_NODE]);
2909 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2910 CURSEG_HOT_NODE]);
119ee914 2911 if (__exist_node_summaries(sbi))
351df4b2
JK		 2912 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2913 type - CURSEG_HOT_NODE);
2914 else
2915 blk_addr = GET_SUM_BLOCK(sbi, segno);
2916 }
2917
2918 new = get_meta_page(sbi, blk_addr);
2919 sum = (struct f2fs_summary_block *)page_address(new);
2920
2921 if (IS_NODESEG(type)) {
119ee914 2922 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2923 struct f2fs_summary *ns = &sum->entries[0];
2924 int i;
2925 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2926 ns->version = 0;
2927 ns->ofs_in_node = 0;
2928 }
2929 } else {
d653788a
GZ		 2930 int err;
2931
2932 err = restore_node_summary(sbi, segno, sum);
2933 if (err) {
351df4b2 2934 f2fs_put_page(new, 1);
d653788a 2935 return err;
351df4b2
JK		 2936 }
2937 }
2938 }
2939
2940 /* set uncompleted segment to curseg */
2941 curseg = CURSEG_I(sbi, type);
2942 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2943
2944 /* update journal info */
2945 down_write(&curseg->journal_rwsem);
2946 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2947 up_write(&curseg->journal_rwsem);
2948
2949 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2950 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2951 curseg->next_segno = segno;
2952 reset_curseg(sbi, type, 0);
2953 curseg->alloc_type = ckpt->alloc_type[type];
2954 curseg->next_blkoff = blk_off;
2955 mutex_unlock(&curseg->curseg_mutex);
2956 f2fs_put_page(new, 1);
2957 return 0;
2958}
2959
2960static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2961{
21d3f8e1
JQ		 2962 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2963 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 2964 int type = CURSEG_HOT_DATA;
e4fc5fbf 2965 int err;
351df4b2 2966
aaec2b1d 2967 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2968 int npages = npages_for_summary_flush(sbi, true);
2969
2970 if (npages >= 2)
2971 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2972 META_CP, true);
3fa06d7b 2973
351df4b2
JK		 2974 /* restore for compacted data summary */
2975 if (read_compacted_summaries(sbi))
2976 return -EINVAL;
2977 type = CURSEG_HOT_NODE;
2978 }
2979
119ee914 2980 if (__exist_node_summaries(sbi))
3fa06d7b 2981 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2982 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2983
e4fc5fbf
CY		 2984 for (; type <= CURSEG_COLD_NODE; type++) {
2985 err = read_normal_summaries(sbi, type);
2986 if (err)
2987 return err;
2988 }
2989
21d3f8e1
JQ		 2990 /* sanity check for summary blocks */
2991 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2992 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2993 return -EINVAL;
2994
351df4b2
JK		 2995 return 0;
2996}
2997
2998static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2999{
3000 struct page *page;
3001 unsigned char *kaddr;
3002 struct f2fs_summary *summary;
3003 struct curseg_info *seg_i;
3004 int written_size = 0;
3005 int i, j;
3006
3007 page = grab_meta_page(sbi, blkaddr++);
3008 kaddr = (unsigned char *)page_address(page);
3009
3010 /* Step 1: write nat cache */
3011 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3012 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3013 written_size += SUM_JOURNAL_SIZE;
3014
3015 /* Step 2: write sit cache */
3016 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3017 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3018 written_size += SUM_JOURNAL_SIZE;
3019
351df4b2
JK		 3020 /* Step 3: write summary entries */
3021 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3022 unsigned short blkoff;
3023 seg_i = CURSEG_I(sbi, i);
3024 if (sbi->ckpt->alloc_type[i] == SSR)
3025 blkoff = sbi->blocks_per_seg;
3026 else
3027 blkoff = curseg_blkoff(sbi, i);
3028
3029 for (j = 0; j < blkoff; j++) {
3030 if (!page) {
3031 page = grab_meta_page(sbi, blkaddr++);
3032 kaddr = (unsigned char *)page_address(page);
3033 written_size = 0;
3034 }
3035 summary = (struct f2fs_summary *)(kaddr + written_size);
3036 *summary = seg_i->sum_blk->entries[j];
3037 written_size += SUMMARY_SIZE;
351df4b2 3038
09cbfeaf 3039 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3040 SUM_FOOTER_SIZE)
3041 continue;
3042
e8d61a74 3043 set_page_dirty(page);
351df4b2
JK		 3044 f2fs_put_page(page, 1);
3045 page = NULL;
3046 }
3047 }
e8d61a74
CY		 3048 if (page) {
3049 set_page_dirty(page);
351df4b2 3050 f2fs_put_page(page, 1);
e8d61a74 3051 }
351df4b2
JK		 3052}
3053
3054static void write_normal_summaries(struct f2fs_sb_info *sbi,
3055 block_t blkaddr, int type)
3056{
3057 int i, end;
3058 if (IS_DATASEG(type))
3059 end = type + NR_CURSEG_DATA_TYPE;
3060 else
3061 end = type + NR_CURSEG_NODE_TYPE;
3062
b7ad7512
CY		 3063 for (i = type; i < end; i++)
3064 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3065}
3066
3067void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3068{
aaec2b1d 3069 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3070 write_compacted_summaries(sbi, start_blk);
3071 else
3072 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3073}
3074
3075void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3076{
119ee914 3077 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3078}
3079
dfc08a12 3080int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3081 unsigned int val, int alloc)
3082{
3083 int i;
3084
3085 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3086 for (i = 0; i < nats_in_cursum(journal); i++) {
3087 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3088 return i;
3089 }
dfc08a12
CY		 3090 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3091 return update_nats_in_cursum(journal, 1);
351df4b2 3092 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3093 for (i = 0; i < sits_in_cursum(journal); i++)
3094 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3095 return i;
dfc08a12
CY		 3096 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3097 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3098 }
3099 return -1;
3100}
3101
3102static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3103 unsigned int segno)
3104{
2cc22186 3105 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3106}
3107
3108static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3109 unsigned int start)
3110{
3111 struct sit_info *sit_i = SIT_I(sbi);
3112 struct page *src_page, *dst_page;
3113 pgoff_t src_off, dst_off;
3114 void *src_addr, *dst_addr;
3115
3116 src_off = current_sit_addr(sbi, start);
3117 dst_off = next_sit_addr(sbi, src_off);
3118
3119 /* get current sit block page without lock */
3120 src_page = get_meta_page(sbi, src_off);
3121 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 3122 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 3123
3124 src_addr = page_address(src_page);
3125 dst_addr = page_address(dst_page);
09cbfeaf 3126 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 3127
3128 set_page_dirty(dst_page);
3129 f2fs_put_page(src_page, 1);
3130
3131 set_to_next_sit(sit_i, start);
3132
3133 return dst_page;
3134}
3135
184a5cd2
CY		 3136static struct sit_entry_set *grab_sit_entry_set(void)
3137{
3138 struct sit_entry_set *ses =
80c54505 3139 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3140
3141 ses->entry_cnt = 0;
3142 INIT_LIST_HEAD(&ses->set_list);
3143 return ses;
3144}
3145
3146static void release_sit_entry_set(struct sit_entry_set *ses)
3147{
3148 list_del(&ses->set_list);
3149 kmem_cache_free(sit_entry_set_slab, ses);
3150}
3151
3152static void adjust_sit_entry_set(struct sit_entry_set *ses,
3153 struct list_head *head)
3154{
3155 struct sit_entry_set *next = ses;
3156
3157 if (list_is_last(&ses->set_list, head))
3158 return;
3159
3160 list_for_each_entry_continue(next, head, set_list)
3161 if (ses->entry_cnt <= next->entry_cnt)
3162 break;
3163
3164 list_move_tail(&ses->set_list, &next->set_list);
3165}
3166
3167static void add_sit_entry(unsigned int segno, struct list_head *head)
3168{
3169 struct sit_entry_set *ses;
3170 unsigned int start_segno = START_SEGNO(segno);
3171
3172 list_for_each_entry(ses, head, set_list) {
3173 if (ses->start_segno == start_segno) {
3174 ses->entry_cnt++;
3175 adjust_sit_entry_set(ses, head);
3176 return;
3177 }
3178 }
3179
3180 ses = grab_sit_entry_set();
3181
3182 ses->start_segno = start_segno;
3183 ses->entry_cnt++;
3184 list_add(&ses->set_list, head);
3185}
3186
3187static void add_sits_in_set(struct f2fs_sb_info *sbi)
3188{
3189 struct f2fs_sm_info *sm_info = SM_I(sbi);
3190 struct list_head *set_list = &sm_info->sit_entry_set;
3191 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3192 unsigned int segno;
3193
7cd8558b 3194 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3195 add_sit_entry(segno, set_list);
3196}
3197
3198static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3199{
3200 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3201 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3202 int i;
3203
b7ad7512 3204 down_write(&curseg->journal_rwsem);
dfc08a12 3205 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3206 unsigned int segno;
3207 bool dirtied;
3208
dfc08a12 3209 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3210 dirtied = __mark_sit_entry_dirty(sbi, segno);
3211
3212 if (!dirtied)
3213 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3214 }
dfc08a12 3215 update_sits_in_cursum(journal, -i);
b7ad7512 3216 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3217}
3218
0a8165d7 3219/*
351df4b2
JK		 3220 * CP calls this function, which flushes SIT entries including sit_journal,
3221 * and moves prefree segs to free segs.
3222 */
4b2fecc8 3223void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3224{
3225 struct sit_info *sit_i = SIT_I(sbi);
3226 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3227 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3228 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3229 struct sit_entry_set *ses, *tmp;
3230 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3231 bool to_journal = true;
4b2fecc8 3232 struct seg_entry *se;
351df4b2 3233
3d26fa6b 3234 down_write(&sit_i->sentry_lock);
351df4b2 3235
2b11a74b
WL		 3236 if (!sit_i->dirty_sentries)
3237 goto out;
3238
351df4b2 3239 /*
184a5cd2
CY		 3240 * add and account sit entries of dirty bitmap in sit entry
3241 * set temporarily
351df4b2 3242 */
184a5cd2 3243 add_sits_in_set(sbi);
351df4b2 3244
184a5cd2
CY		 3245 /*
3246 * if there are no enough space in journal to store dirty sit
3247 * entries, remove all entries from journal and add and account
3248 * them in sit entry set.
3249 */
dfc08a12 3250 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3251 remove_sits_in_journal(sbi);
b2955550 3252
184a5cd2
CY		 3253 /*
3254 * there are two steps to flush sit entries:
3255 * #1, flush sit entries to journal in current cold data summary block.
3256 * #2, flush sit entries to sit page.
3257 */
3258 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3259 struct page *page = NULL;
184a5cd2
CY		 3260 struct f2fs_sit_block *raw_sit = NULL;
3261 unsigned int start_segno = ses->start_segno;
3262 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3263 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3264 unsigned int segno = start_segno;
3265
3266 if (to_journal &&
dfc08a12 3267 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3268 to_journal = false;
3269
b7ad7512
CY		 3270 if (to_journal) {
3271 down_write(&curseg->journal_rwsem);
3272 } else {
184a5cd2
CY		 3273 page = get_next_sit_page(sbi, start_segno);
3274 raw_sit = page_address(page);
351df4b2 3275 }
351df4b2 3276
184a5cd2
CY		 3277 /* flush dirty sit entries in region of current sit set */
3278 for_each_set_bit_from(segno, bitmap, end) {
3279 int offset, sit_offset;
4b2fecc8
JK		 3280
3281 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 3282
3283 /* add discard candidates */
c473f1a9 3284 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3285 cpc->trim_start = segno;
25290fa5 3286 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3287 }
184a5cd2
CY		 3288
3289 if (to_journal) {
dfc08a12 3290 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 3291 SIT_JOURNAL, segno, 1);
3292 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3293 segno_in_journal(journal, offset) =
184a5cd2
CY		 3294 cpu_to_le32(segno);
3295 seg_info_to_raw_sit(se,
dfc08a12 3296 &sit_in_journal(journal, offset));
184a5cd2
CY		 3297 } else {
3298 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3299 seg_info_to_raw_sit(se,
3300 &raw_sit->entries[sit_offset]);
3301 }
351df4b2 3302
184a5cd2
CY		 3303 __clear_bit(segno, bitmap);
3304 sit_i->dirty_sentries--;
3305 ses->entry_cnt--;
351df4b2
JK		 3306 }
3307
b7ad7512
CY		 3308 if (to_journal)
3309 up_write(&curseg->journal_rwsem);
3310 else
184a5cd2
CY		 3311 f2fs_put_page(page, 1);
3312
3313 f2fs_bug_on(sbi, ses->entry_cnt);
3314 release_sit_entry_set(ses);
351df4b2 3315 }
184a5cd2
CY		 3316
3317 f2fs_bug_on(sbi, !list_empty(head));
3318 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3319out:
c473f1a9 3320 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3321 __u64 trim_start = cpc->trim_start;
3322
4b2fecc8 3323 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3324 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3325
3326 cpc->trim_start = trim_start;
4b2fecc8 3327 }
3d26fa6b 3328 up_write(&sit_i->sentry_lock);
351df4b2 3329
351df4b2
JK		 3330 set_prefree_as_free_segments(sbi);
3331}
3332
3333static int build_sit_info(struct f2fs_sb_info *sbi)
3334{
3335 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3336 struct sit_info *sit_i;
3337 unsigned int sit_segs, start;
ae27d62e 3338 char *src_bitmap;
351df4b2
JK		 3339 unsigned int bitmap_size;
3340
3341 /* allocate memory for SIT information */
3342 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3343 if (!sit_i)
3344 return -ENOMEM;
3345
3346 SM_I(sbi)->sit_info = sit_i;
3347
a7c3e901 3348 sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
39307a8e 3349 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3350 if (!sit_i->sentries)
3351 return -ENOMEM;
3352
7cd8558b 3353 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3354 sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3355 if (!sit_i->dirty_sentries_bitmap)
3356 return -ENOMEM;
3357
7cd8558b 3358 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3359 sit_i->sentries[start].cur_valid_map
3360 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3361 sit_i->sentries[start].ckpt_valid_map
3362 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3363 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3364 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3365 return -ENOMEM;
3e025740 3366
355e7891
CY		 3367#ifdef CONFIG_F2FS_CHECK_FS
3368 sit_i->sentries[start].cur_valid_map_mir
3369 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3370 if (!sit_i->sentries[start].cur_valid_map_mir)
3371 return -ENOMEM;
3372#endif
3373
3e025740
JK		 3374 if (f2fs_discard_en(sbi)) {
3375 sit_i->sentries[start].discard_map
3376 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3377 if (!sit_i->sentries[start].discard_map)
3378 return -ENOMEM;
3379 }
351df4b2
JK		 3380 }
3381
60a3b782
JK		 3382 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3383 if (!sit_i->tmp_map)
3384 return -ENOMEM;
3385
351df4b2 3386 if (sbi->segs_per_sec > 1) {
a7c3e901 3387 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
39307a8e 3388 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3389 if (!sit_i->sec_entries)
3390 return -ENOMEM;
3391 }
3392
3393 /* get information related with SIT */
3394 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3395
3396 /* setup SIT bitmap from ckeckpoint pack */
3397 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3398 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3399
ae27d62e
CY		 3400 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3401 if (!sit_i->sit_bitmap)
351df4b2 3402 return -ENOMEM;
351df4b2 3403
ae27d62e
CY		 3404#ifdef CONFIG_F2FS_CHECK_FS
3405 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3406 if (!sit_i->sit_bitmap_mir)
3407 return -ENOMEM;
3408#endif
3409
351df4b2
JK		 3410 /* init SIT information */
3411 sit_i->s_ops = &default_salloc_ops;
3412
3413 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3414 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3415 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3416 sit_i->bitmap_size = bitmap_size;
3417 sit_i->dirty_sentries = 0;
3418 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3419 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3420 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 3421 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 3422 return 0;
3423}
3424
3425static int build_free_segmap(struct f2fs_sb_info *sbi)
3426{
351df4b2
JK		 3427 struct free_segmap_info *free_i;
3428 unsigned int bitmap_size, sec_bitmap_size;
3429
3430 /* allocate memory for free segmap information */
3431 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3432 if (!free_i)
3433 return -ENOMEM;
3434
3435 SM_I(sbi)->free_info = free_i;
3436
7cd8558b 3437 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3438 free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3439 if (!free_i->free_segmap)
3440 return -ENOMEM;
3441
7cd8558b 3442 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
a7c3e901 3443 free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3444 if (!free_i->free_secmap)
3445 return -ENOMEM;
3446
3447 /* set all segments as dirty temporarily */
3448 memset(free_i->free_segmap, 0xff, bitmap_size);
3449 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3450
3451 /* init free segmap information */
7cd8558b 3452 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3453 free_i->free_segments = 0;
3454 free_i->free_sections = 0;
1a118ccf 3455 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3456 return 0;
3457}
3458
3459static int build_curseg(struct f2fs_sb_info *sbi)
3460{
1042d60f 3461 struct curseg_info *array;
351df4b2
JK		 3462 int i;
3463
b434babf 3464 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 3465 if (!array)
3466 return -ENOMEM;
3467
3468 SM_I(sbi)->curseg_array = array;
3469
3470 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3471 mutex_init(&array[i].curseg_mutex);
09cbfeaf 3472 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3473 if (!array[i].sum_blk)
3474 return -ENOMEM;
b7ad7512
CY		 3475 init_rwsem(&array[i].journal_rwsem);
3476 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3477 GFP_KERNEL);
3478 if (!array[i].journal)
3479 return -ENOMEM;
351df4b2
JK		 3480 array[i].segno = NULL_SEGNO;
3481 array[i].next_blkoff = 0;
3482 }
3483 return restore_curseg_summaries(sbi);
3484}
3485
3486static void build_sit_entries(struct f2fs_sb_info *sbi)
3487{
3488 struct sit_info *sit_i = SIT_I(sbi);
3489 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3490 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3491 struct seg_entry *se;
3492 struct f2fs_sit_entry sit;
74de593a
CY		 3493 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3494 unsigned int i, start, end;
3495 unsigned int readed, start_blk = 0;
351df4b2 3496
74de593a 3497 do {
664ba972
JK		 3498 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3499 META_SIT, true);
74de593a
CY		 3500
3501 start = start_blk * sit_i->sents_per_block;
3502 end = (start_blk + readed) * sit_i->sents_per_block;
3503
7cd8558b 3504 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3505 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3506 struct page *page;
3507
9c094040 3508 se = &sit_i->sentries[start];
74de593a
CY		 3509 page = get_current_sit_page(sbi, start);
3510 sit_blk = (struct f2fs_sit_block *)page_address(page);
3511 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3512 f2fs_put_page(page, 1);
d600af23 3513
74de593a
CY		 3514 check_block_count(sbi, start, &sit);
3515 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 3516
3517 /* build discard map only one time */
3e025740 3518 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3519 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3520 memset(se->discard_map, 0xff,
3521 SIT_VBLOCK_MAP_SIZE);
3522 } else {
3523 memcpy(se->discard_map,
3524 se->cur_valid_map,
3525 SIT_VBLOCK_MAP_SIZE);
3526 sbi->discard_blks +=
3527 sbi->blocks_per_seg -
3528 se->valid_blocks;
3529 }
3e025740 3530 }
a66cdd98 3531
d600af23
CY		 3532 if (sbi->segs_per_sec > 1)
3533 get_sec_entry(sbi, start)->valid_blocks +=
3534 se->valid_blocks;
351df4b2 3535 }
74de593a
CY		 3536 start_blk += readed;
3537 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3538
3539 down_read(&curseg->journal_rwsem);
3540 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3541 unsigned int old_valid_blocks;
3542
3543 start = le32_to_cpu(segno_in_journal(journal, i));
3544 se = &sit_i->sentries[start];
3545 sit = sit_in_journal(journal, i);
3546
3547 old_valid_blocks = se->valid_blocks;
3548
3549 check_block_count(sbi, start, &sit);
3550 seg_info_from_raw_sit(se, &sit);
3551
3552 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3553 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3554 memset(se->discard_map, 0xff,
3555 SIT_VBLOCK_MAP_SIZE);
3556 } else {
3557 memcpy(se->discard_map, se->cur_valid_map,
3558 SIT_VBLOCK_MAP_SIZE);
3559 sbi->discard_blks += old_valid_blocks -
3560 se->valid_blocks;
3561 }
d600af23
CY		 3562 }
3563
3564 if (sbi->segs_per_sec > 1)
3565 get_sec_entry(sbi, start)->valid_blocks +=
3566 se->valid_blocks - old_valid_blocks;
3567 }
3568 up_read(&curseg->journal_rwsem);
351df4b2
JK		 3569}
3570
3571static void init_free_segmap(struct f2fs_sb_info *sbi)
3572{
3573 unsigned int start;
3574 int type;
3575
7cd8558b 3576 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3577 struct seg_entry *sentry = get_seg_entry(sbi, start);
3578 if (!sentry->valid_blocks)
3579 __set_free(sbi, start);
c79b7ff1
JK		 3580 else
3581 SIT_I(sbi)->written_valid_blocks +=
3582 sentry->valid_blocks;
351df4b2
JK		 3583 }
3584
3585 /* set use the current segments */
3586 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3587 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3588 __set_test_and_inuse(sbi, curseg_t->segno);
3589 }
3590}
3591
3592static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3593{
3594 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3595 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3596 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3597 unsigned short valid_blocks;
3598
8736fbf0 3599 while (1) {
351df4b2 3600 /* find dirty segment based on free segmap */
7cd8558b
JK		 3601 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3602 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3603 break;
3604 offset = segno + 1;
302bd348 3605 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3606 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3607 continue;
ec325b52
JK		 3608 if (valid_blocks > sbi->blocks_per_seg) {
3609 f2fs_bug_on(sbi, 1);
3610 continue;
3611 }
351df4b2
JK		 3612 mutex_lock(&dirty_i->seglist_lock);
3613 __locate_dirty_segment(sbi, segno, DIRTY);
3614 mutex_unlock(&dirty_i->seglist_lock);
3615 }
3616}
3617
5ec4e49f 3618static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3619{
3620 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3621 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3622
a7c3e901 3623 dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 3624 if (!dirty_i->victim_secmap)
351df4b2
JK		 3625 return -ENOMEM;
3626 return 0;
3627}
3628
3629static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3630{
3631 struct dirty_seglist_info *dirty_i;
3632 unsigned int bitmap_size, i;
3633
3634 /* allocate memory for dirty segments list information */
3635 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3636 if (!dirty_i)
3637 return -ENOMEM;
3638
3639 SM_I(sbi)->dirty_info = dirty_i;
3640 mutex_init(&dirty_i->seglist_lock);
3641
7cd8558b 3642 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3643
3644 for (i = 0; i < NR_DIRTY_TYPE; i++) {
a7c3e901 3645 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3646 if (!dirty_i->dirty_segmap[i])
3647 return -ENOMEM;
3648 }
3649
3650 init_dirty_segmap(sbi);
5ec4e49f 3651 return init_victim_secmap(sbi);
351df4b2
JK		 3652}
3653
0a8165d7 3654/*
351df4b2
JK		 3655 * Update min, max modified time for cost-benefit GC algorithm
3656 */
3657static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3658{
3659 struct sit_info *sit_i = SIT_I(sbi);
3660 unsigned int segno;
3661
3d26fa6b 3662 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3663
3664 sit_i->min_mtime = LLONG_MAX;
3665
7cd8558b 3666 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3667 unsigned int i;
3668 unsigned long long mtime = 0;
3669
3670 for (i = 0; i < sbi->segs_per_sec; i++)
3671 mtime += get_seg_entry(sbi, segno + i)->mtime;
3672
3673 mtime = div_u64(mtime, sbi->segs_per_sec);
3674
3675 if (sit_i->min_mtime > mtime)
3676 sit_i->min_mtime = mtime;
3677 }
3678 sit_i->max_mtime = get_mtime(sbi);
3d26fa6b 3679 up_write(&sit_i->sentry_lock);
351df4b2
JK		 3680}
3681
3682int build_segment_manager(struct f2fs_sb_info *sbi)
3683{
3684 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3685 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3686 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3687 int err;
3688
3689 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3690 if (!sm_info)
3691 return -ENOMEM;
3692
3693 /* init sm info */
3694 sbi->sm_info = sm_info;
351df4b2
JK		 3695 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3696 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3697 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3698 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3699 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3700 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3701 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3702 sm_info->rec_prefree_segments = sm_info->main_segments *
3703 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3704 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3705 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3706
52763a4b
JK		 3707 if (!test_opt(sbi, LFS))
3708 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3709 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3710 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3711 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 3712 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 3713
bba681cb
JK		 3714 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3715
184a5cd2
CY		 3716 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3717
2b60311d
CY		 3718 init_rwsem(&sm_info->curseg_lock);
3719
d4fdf8ba 3720 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3721 err = create_flush_cmd_control(sbi);
3722 if (err)
a688b9d9 3723 return err;
6b4afdd7
JK		 3724 }
3725
0b54fb84
JK		 3726 err = create_discard_cmd_control(sbi);
3727 if (err)
3728 return err;
3729
351df4b2
JK		 3730 err = build_sit_info(sbi);
3731 if (err)
3732 return err;
3733 err = build_free_segmap(sbi);
3734 if (err)
3735 return err;
3736 err = build_curseg(sbi);
3737 if (err)
3738 return err;
3739
3740 /* reinit free segmap based on SIT */
3741 build_sit_entries(sbi);
3742
3743 init_free_segmap(sbi);
3744 err = build_dirty_segmap(sbi);
3745 if (err)
3746 return err;
3747
3748 init_min_max_mtime(sbi);
3749 return 0;
3750}
3751
3752static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3753 enum dirty_type dirty_type)
3754{
3755 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3756
3757 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3758 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3759 dirty_i->nr_dirty[dirty_type] = 0;
3760 mutex_unlock(&dirty_i->seglist_lock);
3761}
3762
5ec4e49f 3763static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3764{
3765 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3766 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3767}
3768
3769static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3770{
3771 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3772 int i;
3773
3774 if (!dirty_i)
3775 return;
3776
3777 /* discard pre-free/dirty segments list */
3778 for (i = 0; i < NR_DIRTY_TYPE; i++)
3779 discard_dirty_segmap(sbi, i);
3780
5ec4e49f 3781 destroy_victim_secmap(sbi);
351df4b2
JK		 3782 SM_I(sbi)->dirty_info = NULL;
3783 kfree(dirty_i);
3784}
3785
3786static void destroy_curseg(struct f2fs_sb_info *sbi)
3787{
3788 struct curseg_info *array = SM_I(sbi)->curseg_array;
3789 int i;
3790
3791 if (!array)
3792 return;
3793 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3794 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3795 kfree(array[i].sum_blk);
b7ad7512
CY		 3796 kfree(array[i].journal);
3797 }
351df4b2
JK		 3798 kfree(array);
3799}
3800
3801static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3802{
3803 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3804 if (!free_i)
3805 return;
3806 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3807 kvfree(free_i->free_segmap);
3808 kvfree(free_i->free_secmap);
351df4b2
JK		 3809 kfree(free_i);
3810}
3811
3812static void destroy_sit_info(struct f2fs_sb_info *sbi)
3813{
3814 struct sit_info *sit_i = SIT_I(sbi);
3815 unsigned int start;
3816
3817 if (!sit_i)
3818 return;
3819
3820 if (sit_i->sentries) {
7cd8558b 3821 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3822 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 3823#ifdef CONFIG_F2FS_CHECK_FS
3824 kfree(sit_i->sentries[start].cur_valid_map_mir);
3825#endif
351df4b2 3826 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3827 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3828 }
3829 }
60a3b782
JK		 3830 kfree(sit_i->tmp_map);
3831
39307a8e
JK		 3832 kvfree(sit_i->sentries);
3833 kvfree(sit_i->sec_entries);
3834 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3835
3836 SM_I(sbi)->sit_info = NULL;
3837 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 3838#ifdef CONFIG_F2FS_CHECK_FS
3839 kfree(sit_i->sit_bitmap_mir);
3840#endif
351df4b2
JK		 3841 kfree(sit_i);
3842}
3843
3844void destroy_segment_manager(struct f2fs_sb_info *sbi)
3845{
3846 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3847
3b03f724
CY		 3848 if (!sm_info)
3849 return;
5eba8c5d 3850 destroy_flush_cmd_control(sbi, true);
f099405f 3851 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3852 destroy_dirty_segmap(sbi);
3853 destroy_curseg(sbi);
3854 destroy_free_segmap(sbi);
3855 destroy_sit_info(sbi);
3856 sbi->sm_info = NULL;
3857 kfree(sm_info);
3858}
7fd9e544
JK		 3859
3860int __init create_segment_manager_caches(void)
3861{
3862 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3863 sizeof(struct discard_entry));
7fd9e544 3864 if (!discard_entry_slab)
184a5cd2
CY		 3865 goto fail;
3866
b01a9201
JK		 3867 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3868 sizeof(struct discard_cmd));
3869 if (!discard_cmd_slab)
6ab2a308 3870 goto destroy_discard_entry;
275b66b0 3871
184a5cd2 3872 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3873 sizeof(struct sit_entry_set));
184a5cd2 3874 if (!sit_entry_set_slab)
b01a9201 3875 goto destroy_discard_cmd;
88b88a66
JK		 3876
3877 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3878 sizeof(struct inmem_pages));
3879 if (!inmem_entry_slab)
3880 goto destroy_sit_entry_set;
7fd9e544 3881 return 0;
184a5cd2 3882
88b88a66
JK		 3883destroy_sit_entry_set:
3884 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 3885destroy_discard_cmd:
3886 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3887destroy_discard_entry:
184a5cd2
CY		 3888 kmem_cache_destroy(discard_entry_slab);
3889fail:
3890 return -ENOMEM;
7fd9e544
JK		 3891}
3892
3893void destroy_segment_manager_caches(void)
3894{
184a5cd2 3895 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3896 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3897 kmem_cache_destroy(discard_entry_slab);
88b88a66 3898 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3899}
ubuntu-bionic-kernel mirror