]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: fix to flush multiple device in checkpoint
[mirror_ubuntu-jammy-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
1d7be270 19#include <linux/freezer.h>
1eb1ef4a 20#include <linux/sched/signal.h>
351df4b2
JK		 21
22#include "f2fs.h"
23#include "segment.h"
24#include "node.h"
5f656541 25#include "gc.h"
9e4ded3f 26#include "trace.h"
6ec178da 27#include <trace/events/f2fs.h>
351df4b2 28
9a7f143a
CL		 29#define __reverse_ffz(x) __reverse_ffs(~(x))
30
7fd9e544 31static struct kmem_cache *discard_entry_slab;
b01a9201 32static struct kmem_cache *discard_cmd_slab;
184a5cd2 33static struct kmem_cache *sit_entry_set_slab;
88b88a66 34static struct kmem_cache *inmem_entry_slab;
7fd9e544 35
f96999c3
JK		 36static unsigned long __reverse_ulong(unsigned char *str)
37{
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41#if BITS_PER_LONG == 64
42 shift = 56;
43#endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49}
50
9a7f143a
CL		 51/*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55static inline unsigned long __reverse_ffs(unsigned long word)
56{
57 int num = 0;
58
59#if BITS_PER_LONG == 64
f96999c3 60 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 61 num += 32;
f96999c3 62 else
9a7f143a 63 word >>= 32;
9a7f143a 64#endif
f96999c3 65 if ((word & 0xffff0000) == 0)
9a7f143a 66 num += 16;
f96999c3 67 else
9a7f143a 68 word >>= 16;
f96999c3
JK		 69
70 if ((word & 0xff00) == 0)
9a7f143a 71 num += 8;
f96999c3 72 else
9a7f143a 73 word >>= 8;
f96999c3 74
9a7f143a
CL		 75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
f96999c3 79
9a7f143a
CL		 80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
f96999c3 84
9a7f143a
CL		 85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88}
89
90/*
e1c42045 91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 93 * @size must be integral times of unsigned long.
9a7f143a 94 * Example:
f96999c3
JK		 95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 98 */
99static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101{
102 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 103 unsigned long result = size;
9a7f143a 104 unsigned long tmp;
9a7f143a
CL		 105
106 if (offset >= size)
107 return size;
108
692223d1 109 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 110 offset %= BITS_PER_LONG;
f96999c3 111
692223d1
FL		 112 while (1) {
113 if (*p == 0)
114 goto pass;
9a7f143a 115
f96999c3 116 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 121 if (tmp)
692223d1
FL		 122 goto found;
123pass:
124 if (size <= BITS_PER_LONG)
125 break;
9a7f143a 126 size -= BITS_PER_LONG;
692223d1 127 offset = 0;
f96999c3 128 p++;
9a7f143a 129 }
692223d1
FL		 130 return result;
131found:
132 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 133}
134
135static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137{
138 const unsigned long *p = addr + BIT_WORD(offset);
80609448 139 unsigned long result = size;
9a7f143a 140 unsigned long tmp;
9a7f143a
CL		 141
142 if (offset >= size)
143 return size;
144
80609448 145 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 146 offset %= BITS_PER_LONG;
80609448
JK		 147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
f96999c3 152 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
f96999c3 158 if (tmp != ~0UL)
80609448
JK		 159 goto found;
160pass:
161 if (size <= BITS_PER_LONG)
162 break;
9a7f143a 163 size -= BITS_PER_LONG;
80609448 164 offset = 0;
f96999c3 165 p++;
9a7f143a 166 }
80609448
JK		 167 return result;
168found:
169 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 170}
171
b3a97a2a
JK		 172bool need_SSR(struct f2fs_sb_info *sbi)
173{
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177
178 if (test_opt(sbi, LFS))
179 return false;
180 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
184 2 * reserved_sections(sbi));
185}
186
88b88a66
JK		 187void register_inmem_page(struct inode *inode, struct page *page)
188{
189 struct f2fs_inode_info *fi = F2FS_I(inode);
190 struct inmem_pages *new;
9be32d72 191
9e4ded3f 192 f2fs_trace_pid(page);
0722b101 193
decd36b6
CY		 194 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
195 SetPagePrivate(page);
196
88b88a66
JK		 197 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
198
199 /* add atomic page indices to the list */
200 new->page = page;
201 INIT_LIST_HEAD(&new->list);
decd36b6 202
88b88a66
JK		 203 /* increase reference count with clean state */
204 mutex_lock(&fi->inmem_lock);
205 get_page(page);
206 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 207 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 208 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 209
210 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 211}
212
28bc106b
CY		 213static int __revoke_inmem_pages(struct inode *inode,
214 struct list_head *head, bool drop, bool recover)
29b96b54 215{
28bc106b 216 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 217 struct inmem_pages *cur, *tmp;
28bc106b 218 int err = 0;
29b96b54
CY		 219
220 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 221 struct page *page = cur->page;
222
223 if (drop)
224 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
225
226 lock_page(page);
29b96b54 227
28bc106b
CY		 228 if (recover) {
229 struct dnode_of_data dn;
230 struct node_info ni;
231
232 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 233retry:
28bc106b 234 set_new_dnode(&dn, inode, NULL, NULL, 0);
7f2b4e8e
CY		 235 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
236 if (err) {
237 if (err == -ENOMEM) {
238 congestion_wait(BLK_RW_ASYNC, HZ/50);
239 cond_resched();
240 goto retry;
241 }
28bc106b
CY		 242 err = -EAGAIN;
243 goto next;
244 }
245 get_node_info(sbi, dn.nid, &ni);
246 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
247 cur->old_addr, ni.version, true, true);
248 f2fs_put_dnode(&dn);
249 }
250next:
63c52d78
JK		 251 /* we don't need to invalidate this in the sccessful status */
252 if (drop || recover)
253 ClearPageUptodate(page);
28bc106b 254 set_page_private(page, 0);
c81ced05 255 ClearPagePrivate(page);
28bc106b 256 f2fs_put_page(page, 1);
29b96b54
CY		 257
258 list_del(&cur->list);
259 kmem_cache_free(inmem_entry_slab, cur);
260 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
261 }
28bc106b 262 return err;
29b96b54
CY		 263}
264
265void drop_inmem_pages(struct inode *inode)
266{
267 struct f2fs_inode_info *fi = F2FS_I(inode);
268
269 mutex_lock(&fi->inmem_lock);
28bc106b 270 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54 271 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 272
273 clear_inode_flag(inode, FI_ATOMIC_FILE);
84a23fbe 274 clear_inode_flag(inode, FI_HOT_DATA);
5fe45743 275 stat_dec_atomic_write(inode);
29b96b54
CY		 276}
277
8c242db9
JK		 278void drop_inmem_page(struct inode *inode, struct page *page)
279{
280 struct f2fs_inode_info *fi = F2FS_I(inode);
281 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
282 struct list_head *head = &fi->inmem_pages;
283 struct inmem_pages *cur = NULL;
284
285 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
286
287 mutex_lock(&fi->inmem_lock);
288 list_for_each_entry(cur, head, list) {
289 if (cur->page == page)
290 break;
291 }
292
293 f2fs_bug_on(sbi, !cur || cur->page != page);
294 list_del(&cur->list);
295 mutex_unlock(&fi->inmem_lock);
296
297 dec_page_count(sbi, F2FS_INMEM_PAGES);
298 kmem_cache_free(inmem_entry_slab, cur);
299
300 ClearPageUptodate(page);
301 set_page_private(page, 0);
302 ClearPagePrivate(page);
303 f2fs_put_page(page, 0);
304
305 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
306}
307
28bc106b
CY		 308static int __commit_inmem_pages(struct inode *inode,
309 struct list_head *revoke_list)
88b88a66
JK		 310{
311 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
312 struct f2fs_inode_info *fi = F2FS_I(inode);
313 struct inmem_pages *cur, *tmp;
88b88a66 314 struct f2fs_io_info fio = {
05ca3632 315 .sbi = sbi,
39d787be 316 .ino = inode->i_ino,
88b88a66 317 .type = DATA,
04d328de 318 .op = REQ_OP_WRITE,
70fd7614 319 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 320 .io_type = FS_DATA_IO,
88b88a66 321 };
942fd319 322 pgoff_t last_idx = ULONG_MAX;
edb27dee 323 int err = 0;
88b88a66 324
88b88a66 325 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 326 struct page *page = cur->page;
327
328 lock_page(page);
329 if (page->mapping == inode->i_mapping) {
330 trace_f2fs_commit_inmem_page(page, INMEM);
331
332 set_page_dirty(page);
333 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 334 if (clear_page_dirty_for_io(page)) {
29b96b54 335 inode_dec_dirty_pages(inode);
933439c8
CY		 336 remove_dirty_inode(inode);
337 }
640cc189 338retry:
28bc106b 339 fio.page = page;
e959c8f5 340 fio.old_blkaddr = NULL_ADDR;
4d978078 341 fio.encrypted_page = NULL;
cc15620b 342 fio.need_lock = LOCK_DONE;
29b96b54
CY		 343 err = do_write_data_page(&fio);
344 if (err) {
640cc189
JK		 345 if (err == -ENOMEM) {
346 congestion_wait(BLK_RW_ASYNC, HZ/50);
347 cond_resched();
348 goto retry;
349 }
28bc106b 350 unlock_page(page);
29b96b54 351 break;
70c640b1 352 }
28bc106b
CY		 353 /* record old blkaddr for revoking */
354 cur->old_addr = fio.old_blkaddr;
942fd319 355 last_idx = page->index;
28bc106b
CY		 356 }
357 unlock_page(page);
358 list_move_tail(&cur->list, revoke_list);
88b88a66 359 }
29b96b54 360
942fd319 361 if (last_idx != ULONG_MAX)
b9109b0e 362 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b
CY		 363
364 if (!err)
365 __revoke_inmem_pages(inode, revoke_list, false, false);
366
29b96b54
CY		 367 return err;
368}
369
370int commit_inmem_pages(struct inode *inode)
371{
372 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
373 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 374 struct list_head revoke_list;
375 int err;
29b96b54 376
28bc106b 377 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 378 f2fs_balance_fs(sbi, true);
379 f2fs_lock_op(sbi);
380
5fe45743
CY		 381 set_inode_flag(inode, FI_ATOMIC_COMMIT);
382
29b96b54 383 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 384 err = __commit_inmem_pages(inode, &revoke_list);
385 if (err) {
386 int ret;
387 /*
388 * try to revoke all committed pages, but still we could fail
389 * due to no memory or other reason, if that happened, EAGAIN
390 * will be returned, which means in such case, transaction is
391 * already not integrity, caller should use journal to do the
392 * recovery or rewrite & commit last transaction. For other
393 * error number, revoking was done by filesystem itself.
394 */
395 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
396 if (ret)
397 err = ret;
398
399 /* drop all uncommitted pages */
400 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
401 }
88b88a66
JK		 402 mutex_unlock(&fi->inmem_lock);
403
5fe45743
CY		 404 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
405
29b96b54 406 f2fs_unlock_op(sbi);
edb27dee 407 return err;
88b88a66
JK		 408}
409
0a8165d7 410/*
351df4b2
JK		 411 * This function balances dirty node and dentry pages.
412 * In addition, it controls garbage collection.
413 */
2c4db1a6 414void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 415{
0f348028 416#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 417 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
418 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 419 f2fs_stop_checkpoint(sbi, false);
55523519 420 }
0f348028
CY		 421#endif
422
e589c2c4 423 /* balance_fs_bg is able to be pending */
a7881893 424 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 425 f2fs_balance_fs_bg(sbi);
426
351df4b2 427 /*
029cd28c
JK		 428 * We should do GC or end up with checkpoint, if there are so many dirty
429 * dir/node pages without enough free segments.
351df4b2 430 */
7f3037a5 431 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 432 mutex_lock(&sbi->gc_mutex);
e066b83c 433 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 434 }
435}
436
4660f9c0
JK		 437void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
438{
1dcc336b 439 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 440 if (!available_free_memory(sbi, EXTENT_CACHE))
441 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 442
1b38dc8e
JK		 443 /* check the # of cached NAT entries */
444 if (!available_free_memory(sbi, NAT_ENTRIES))
445 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
446
31696580 447 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 448 try_to_free_nids(sbi, MAX_FREE_NIDS);
449 else
22ad0b6a 450 build_free_nids(sbi, false, false);
31696580 451
1c0f4bf5 452 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 453 return;
31696580 454
1b38dc8e
JK		 455 /* checkpoint is the only way to shrink partial cached entries */
456 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 457 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 458 excess_prefree_segs(sbi) ||
459 excess_dirty_nats(sbi) ||
f455c8a5 460 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 461 if (test_opt(sbi, DATA_FLUSH)) {
462 struct blk_plug plug;
463
464 blk_start_plug(&plug);
36b35a0d 465 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 466 blk_finish_plug(&plug);
467 }
4660f9c0 468 f2fs_sync_fs(sbi->sb, true);
42190d2a 469 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 470 }
4660f9c0
JK		 471}
472
20fda56b
KM		 473static int __submit_flush_wait(struct f2fs_sb_info *sbi,
474 struct block_device *bdev)
3c62be17
JK		 475{
476 struct bio *bio = f2fs_bio_alloc(0);
477 int ret;
478
3adc5fcb 479 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 480 bio_set_dev(bio, bdev);
3c62be17
JK		 481 ret = submit_bio_wait(bio);
482 bio_put(bio);
20fda56b
KM		 483
484 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
485 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 486 return ret;
487}
488
39d787be 489static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 490{
39d787be 491 int ret = 0;
3c62be17
JK		 492 int i;
493
39d787be
CY		 494 if (!sbi->s_ndevs)
495 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 496
39d787be
CY		 497 for (i = 0; i < sbi->s_ndevs; i++) {
498 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
499 continue;
20fda56b
KM		 500 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
501 if (ret)
502 break;
3c62be17
JK		 503 }
504 return ret;
505}
506
2163d198 507static int issue_flush_thread(void *data)
6b4afdd7
JK		 508{
509 struct f2fs_sb_info *sbi = data;
b01a9201 510 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 511 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 512repeat:
513 if (kthread_should_stop())
514 return 0;
515
dc6febb6
CY		 516 sb_start_intwrite(sbi->sb);
517
721bd4d5 518 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 519 struct flush_cmd *cmd, *next;
520 int ret;
521
721bd4d5
GZ		 522 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
523 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
524
39d787be
CY		 525 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
526
527 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 528 atomic_inc(&fcc->issued_flush);
529
721bd4d5
GZ		 530 llist_for_each_entry_safe(cmd, next,
531 fcc->dispatch_list, llnode) {
6b4afdd7 532 cmd->ret = ret;
6b4afdd7
JK		 533 complete(&cmd->wait);
534 }
a688b9d9 535 fcc->dispatch_list = NULL;
6b4afdd7
JK		 536 }
537
dc6febb6
CY		 538 sb_end_intwrite(sbi->sb);
539
a688b9d9 540 wait_event_interruptible(*q,
721bd4d5 541 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 542 goto repeat;
543}
544
39d787be 545int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 546{
b01a9201 547 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 548 struct flush_cmd cmd;
8b8dd65f 549 int ret;
6b4afdd7 550
0f7b2abd
JK		 551 if (test_opt(sbi, NOBARRIER))
552 return 0;
553
8b8dd65f 554 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 555 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 556 atomic_inc(&fcc->issued_flush);
557 return ret;
558 }
740432f8 559
39d787be
CY		 560 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
561 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 562 atomic_dec(&fcc->issing_flush);
563
564 atomic_inc(&fcc->issued_flush);
740432f8
JK		 565 return ret;
566 }
6b4afdd7 567
39d787be 568 cmd.ino = ino;
adf8d90b 569 init_completion(&cmd.wait);
6b4afdd7 570
721bd4d5 571 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 572
6f890df0
CY		 573 /* update issue_list before we wake up issue_flush thread */
574 smp_mb();
575
576 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 577 wake_up(&fcc->flush_wait_queue);
6b4afdd7 578
5eba8c5d
JK		 579 if (fcc->f2fs_issue_flush) {
580 wait_for_completion(&cmd.wait);
8b8dd65f 581 atomic_dec(&fcc->issing_flush);
5eba8c5d 582 } else {
d3238691
CY		 583 struct llist_node *list;
584
585 list = llist_del_all(&fcc->issue_list);
586 if (!list) {
587 wait_for_completion(&cmd.wait);
588 atomic_dec(&fcc->issing_flush);
589 } else {
590 struct flush_cmd *tmp, *next;
591
39d787be 592 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 593
594 llist_for_each_entry_safe(tmp, next, list, llnode) {
595 if (tmp == &cmd) {
596 cmd.ret = ret;
597 atomic_dec(&fcc->issing_flush);
598 continue;
599 }
600 tmp->ret = ret;
601 complete(&tmp->wait);
602 }
603 }
5eba8c5d 604 }
adf8d90b
CY		 605
606 return cmd.ret;
6b4afdd7
JK		 607}
608
2163d198
GZ		 609int create_flush_cmd_control(struct f2fs_sb_info *sbi)
610{
611 dev_t dev = sbi->sb->s_bdev->bd_dev;
612 struct flush_cmd_control *fcc;
613 int err = 0;
614
b01a9201
JK		 615 if (SM_I(sbi)->fcc_info) {
616 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 617 if (fcc->f2fs_issue_flush)
618 return err;
5eba8c5d
JK		 619 goto init_thread;
620 }
621
2163d198
GZ		 622 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
623 if (!fcc)
624 return -ENOMEM;
8b8dd65f
CY		 625 atomic_set(&fcc->issued_flush, 0);
626 atomic_set(&fcc->issing_flush, 0);
2163d198 627 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 628 init_llist_head(&fcc->issue_list);
b01a9201 629 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 630 if (!test_opt(sbi, FLUSH_MERGE))
631 return err;
632
5eba8c5d 633init_thread:
2163d198
GZ		 634 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
635 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
636 if (IS_ERR(fcc->f2fs_issue_flush)) {
637 err = PTR_ERR(fcc->f2fs_issue_flush);
638 kfree(fcc);
b01a9201 639 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 640 return err;
641 }
2163d198
GZ		 642
643 return err;
644}
645
5eba8c5d 646void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 647{
b01a9201 648 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 649
5eba8c5d
JK		 650 if (fcc && fcc->f2fs_issue_flush) {
651 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
652
653 fcc->f2fs_issue_flush = NULL;
654 kthread_stop(flush_thread);
655 }
656 if (free) {
657 kfree(fcc);
b01a9201 658 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 659 }
2163d198
GZ		 660}
661
1228b482
CY		 662int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
663{
664 int ret = 0, i;
665
666 if (!sbi->s_ndevs)
667 return 0;
668
669 for (i = 1; i < sbi->s_ndevs; i++) {
670 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
671 continue;
672 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
673 if (ret)
674 break;
675
676 spin_lock(&sbi->dev_lock);
677 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
678 spin_unlock(&sbi->dev_lock);
679 }
680
681 return ret;
682}
683
351df4b2
JK		 684static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
685 enum dirty_type dirty_type)
686{
687 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
688
689 /* need not be added */
690 if (IS_CURSEG(sbi, segno))
691 return;
692
693 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
694 dirty_i->nr_dirty[dirty_type]++;
695
696 if (dirty_type == DIRTY) {
697 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 698 enum dirty_type t = sentry->type;
b2f2c390 699
ec325b52
JK		 700 if (unlikely(t >= DIRTY)) {
701 f2fs_bug_on(sbi, 1);
702 return;
703 }
4625d6aa
CL		 704 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
705 dirty_i->nr_dirty[t]++;
351df4b2
JK		 706 }
707}
708
709static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
710 enum dirty_type dirty_type)
711{
712 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
713
714 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
715 dirty_i->nr_dirty[dirty_type]--;
716
717 if (dirty_type == DIRTY) {
4625d6aa
CL		 718 struct seg_entry *sentry = get_seg_entry(sbi, segno);
719 enum dirty_type t = sentry->type;
720
721 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
722 dirty_i->nr_dirty[t]--;
b2f2c390 723
302bd348 724 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 725 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 726 dirty_i->victim_secmap);
351df4b2
JK		 727 }
728}
729
0a8165d7 730/*
351df4b2
JK		 731 * Should not occur error such as -ENOMEM.
732 * Adding dirty entry into seglist is not critical operation.
733 * If a given segment is one of current working segments, it won't be added.
734 */
8d8451af 735static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 736{
737 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
738 unsigned short valid_blocks;
739
740 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
741 return;
742
743 mutex_lock(&dirty_i->seglist_lock);
744
302bd348 745 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 746
747 if (valid_blocks == 0) {
748 __locate_dirty_segment(sbi, segno, PRE);
749 __remove_dirty_segment(sbi, segno, DIRTY);
750 } else if (valid_blocks < sbi->blocks_per_seg) {
751 __locate_dirty_segment(sbi, segno, DIRTY);
752 } else {
753 /* Recovery routine with SSR needs this */
754 __remove_dirty_segment(sbi, segno, DIRTY);
755 }
756
757 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 758}
759
004b6862 760static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 761 struct block_device *bdev, block_t lstart,
762 block_t start, block_t len)
275b66b0 763{
0b54fb84 764 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 765 struct list_head *pend_list;
b01a9201 766 struct discard_cmd *dc;
275b66b0 767
ba48a33e
CY		 768 f2fs_bug_on(sbi, !len);
769
770 pend_list = &dcc->pend_list[plist_idx(len)];
771
b01a9201
JK		 772 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
773 INIT_LIST_HEAD(&dc->list);
c81abe34 774 dc->bdev = bdev;
b01a9201 775 dc->lstart = lstart;
c81abe34 776 dc->start = start;
b01a9201 777 dc->len = len;
ec9895ad 778 dc->ref = 0;
15469963 779 dc->state = D_PREP;
c81abe34 780 dc->error = 0;
b01a9201 781 init_completion(&dc->wait);
22d375dd 782 list_add_tail(&dc->list, pend_list);
5f32366a 783 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 784 dcc->undiscard_blks += len;
004b6862
CY		 785
786 return dc;
787}
788
789static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
790 struct block_device *bdev, block_t lstart,
791 block_t start, block_t len,
792 struct rb_node *parent, struct rb_node **p)
793{
794 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
795 struct discard_cmd *dc;
796
797 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
798
799 rb_link_node(&dc->rb_node, parent, p);
800 rb_insert_color(&dc->rb_node, &dcc->root);
801
802 return dc;
15469963
JK		 803}
804
004b6862
CY		 805static void __detach_discard_cmd(struct discard_cmd_control *dcc,
806 struct discard_cmd *dc)
15469963 807{
dcc9165d 808 if (dc->state == D_DONE)
004b6862
CY		 809 atomic_dec(&dcc->issing_discard);
810
811 list_del(&dc->list);
812 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 813 dcc->undiscard_blks -= dc->len;
004b6862
CY		 814
815 kmem_cache_free(discard_cmd_slab, dc);
816
817 atomic_dec(&dcc->discard_cmd_cnt);
818}
819
820static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
821 struct discard_cmd *dc)
822{
823 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 824
d9703d90
CY		 825 f2fs_bug_on(sbi, dc->ref);
826
c81abe34
JK		 827 if (dc->error == -EOPNOTSUPP)
828 dc->error = 0;
15469963 829
c81abe34 830 if (dc->error)
15469963 831 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 832 "Issue discard(%u, %u, %u) failed, ret: %d",
833 dc->lstart, dc->start, dc->len, dc->error);
004b6862 834 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 835}
836
c81abe34
JK		 837static void f2fs_submit_discard_endio(struct bio *bio)
838{
839 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
840
4e4cbee9 841 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 842 dc->state = D_DONE;
e31b9821 843 complete_all(&dc->wait);
c81abe34
JK		 844 bio_put(bio);
845}
846
6915ea9d
CY		 847void __check_sit_bitmap(struct f2fs_sb_info *sbi,
848 block_t start, block_t end)
849{
850#ifdef CONFIG_F2FS_CHECK_FS
851 struct seg_entry *sentry;
852 unsigned int segno;
853 block_t blk = start;
854 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
855 unsigned long *map;
856
857 while (blk < end) {
858 segno = GET_SEGNO(sbi, blk);
859 sentry = get_seg_entry(sbi, segno);
860 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
861
008396e1
YS		 862 if (end < START_BLOCK(sbi, segno + 1))
863 size = GET_BLKOFF_FROM_SEG0(sbi, end);
864 else
865 size = max_blocks;
6915ea9d
CY		 866 map = (unsigned long *)(sentry->cur_valid_map);
867 offset = __find_rev_next_bit(map, size, offset);
868 f2fs_bug_on(sbi, offset != size);
008396e1 869 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 870 }
871#endif
872}
873
c81abe34
JK		 874/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
875static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
876 struct discard_cmd *dc)
877{
878 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
879 struct bio *bio = NULL;
880
881 if (dc->state != D_PREP)
882 return;
883
0243a5f9
CY		 884 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
885
c81abe34
JK		 886 dc->error = __blkdev_issue_discard(dc->bdev,
887 SECTOR_FROM_BLOCK(dc->start),
888 SECTOR_FROM_BLOCK(dc->len),
889 GFP_NOFS, 0, &bio);
890 if (!dc->error) {
891 /* should keep before submission to avoid D_DONE right away */
892 dc->state = D_SUBMIT;
8b8dd65f
CY		 893 atomic_inc(&dcc->issued_discard);
894 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 895 if (bio) {
896 bio->bi_private = dc;
897 bio->bi_end_io = f2fs_submit_discard_endio;
898 bio->bi_opf |= REQ_SYNC;
899 submit_bio(bio);
46f84c2c 900 list_move_tail(&dc->list, &dcc->wait_list);
6915ea9d 901 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 902
903 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 904 }
905 } else {
906 __remove_discard_cmd(sbi, dc);
907 }
908}
909
004b6862
CY		 910static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
911 struct block_device *bdev, block_t lstart,
912 block_t start, block_t len,
913 struct rb_node **insert_p,
914 struct rb_node *insert_parent)
c81abe34 915{
004b6862
CY		 916 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
917 struct rb_node **p = &dcc->root.rb_node;
918 struct rb_node *parent = NULL;
919 struct discard_cmd *dc = NULL;
920
921 if (insert_p && insert_parent) {
922 parent = insert_parent;
923 p = insert_p;
924 goto do_insert;
925 }
c81abe34 926
004b6862
CY		 927 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
928do_insert:
929 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
930 if (!dc)
931 return NULL;
c81abe34 932
004b6862 933 return dc;
c81abe34
JK		 934}
935
ba48a33e
CY		 936static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
937 struct discard_cmd *dc)
938{
939 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
940}
941
3d6a650f
YH		 942static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
943 struct discard_cmd *dc, block_t blkaddr)
944{
ba48a33e 945 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 946 struct discard_info di = dc->di;
947 bool modified = false;
3d6a650f 948
004b6862 949 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 950 __remove_discard_cmd(sbi, dc);
951 return;
952 }
953
d84d1cbd
CY		 954 dcc->undiscard_blks -= di.len;
955
004b6862 956 if (blkaddr > di.lstart) {
3d6a650f 957 dc->len = blkaddr - dc->lstart;
d84d1cbd 958 dcc->undiscard_blks += dc->len;
ba48a33e 959 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 960 modified = true;
961 }
962
963 if (blkaddr < di.lstart + di.len - 1) {
964 if (modified) {
965 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
966 di.start + blkaddr + 1 - di.lstart,
967 di.lstart + di.len - 1 - blkaddr,
968 NULL, NULL);
969 } else {
970 dc->lstart++;
971 dc->len--;
972 dc->start++;
d84d1cbd 973 dcc->undiscard_blks += dc->len;
ba48a33e 974 __relocate_discard_cmd(dcc, dc);
004b6862 975 }
3d6a650f
YH		 976 }
977}
978
004b6862
CY		 979static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
980 struct block_device *bdev, block_t lstart,
981 block_t start, block_t len)
275b66b0 982{
0b54fb84 983 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 984 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
985 struct discard_cmd *dc;
986 struct discard_info di = {0};
987 struct rb_node **insert_p = NULL, *insert_parent = NULL;
988 block_t end = lstart + len;
275b66b0 989
15469963 990 mutex_lock(&dcc->cmd_lock);
40465257 991
004b6862
CY		 992 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
993 NULL, lstart,
994 (struct rb_entry **)&prev_dc,
995 (struct rb_entry **)&next_dc,
996 &insert_p, &insert_parent, true);
997 if (dc)
998 prev_dc = dc;
999
1000 if (!prev_dc) {
1001 di.lstart = lstart;
1002 di.len = next_dc ? next_dc->lstart - lstart : len;
1003 di.len = min(di.len, len);
1004 di.start = start;
22d375dd 1005 }
15469963 1006
004b6862
CY		 1007 while (1) {
1008 struct rb_node *node;
1009 bool merged = false;
1010 struct discard_cmd *tdc = NULL;
1011
1012 if (prev_dc) {
1013 di.lstart = prev_dc->lstart + prev_dc->len;
1014 if (di.lstart < lstart)
1015 di.lstart = lstart;
1016 if (di.lstart >= end)
1017 break;
1018
1019 if (!next_dc || next_dc->lstart > end)
1020 di.len = end - di.lstart;
1021 else
1022 di.len = next_dc->lstart - di.lstart;
1023 di.start = start + di.lstart - lstart;
1024 }
1025
1026 if (!di.len)
1027 goto next;
1028
1029 if (prev_dc && prev_dc->state == D_PREP &&
1030 prev_dc->bdev == bdev &&
1031 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1032 prev_dc->di.len += di.len;
d84d1cbd 1033 dcc->undiscard_blks += di.len;
ba48a33e 1034 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1035 di = prev_dc->di;
1036 tdc = prev_dc;
1037 merged = true;
1038 }
1039
1040 if (next_dc && next_dc->state == D_PREP &&
1041 next_dc->bdev == bdev &&
1042 __is_discard_front_mergeable(&di, &next_dc->di)) {
1043 next_dc->di.lstart = di.lstart;
1044 next_dc->di.len += di.len;
1045 next_dc->di.start = di.start;
d84d1cbd 1046 dcc->undiscard_blks += di.len;
ba48a33e 1047 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1048 if (tdc)
1049 __remove_discard_cmd(sbi, tdc);
004b6862 1050 merged = true;
4e6a8d9b 1051 }
004b6862 1052
df0f6b44 1053 if (!merged) {
004b6862
CY		 1054 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1055 di.len, NULL, NULL);
df0f6b44 1056 }
004b6862
CY		 1057 next:
1058 prev_dc = next_dc;
1059 if (!prev_dc)
1060 break;
1061
1062 node = rb_next(&prev_dc->rb_node);
1063 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1064 }
1065
1066 mutex_unlock(&dcc->cmd_lock);
1067}
1068
1069static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1070 struct block_device *bdev, block_t blkstart, block_t blklen)
1071{
1072 block_t lblkstart = blkstart;
1073
0243a5f9 1074 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1075
1076 if (sbi->s_ndevs) {
1077 int devi = f2fs_target_device_index(sbi, blkstart);
1078
1079 blkstart -= FDEV(devi).start_blk;
1080 }
1081 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1082 return 0;
1083}
1084
969d1b18 1085static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
bd5b0738
CY		 1086{
1087 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1088 struct list_head *pend_list;
1089 struct discard_cmd *dc, *tmp;
1090 struct blk_plug plug;
969d1b18
CY		 1091 int iter = 0, issued = 0;
1092 int i;
e6c6de18 1093 bool io_interrupted = false;
bd5b0738
CY		 1094
1095 mutex_lock(&dcc->cmd_lock);
963932a9
YH		 1096 f2fs_bug_on(sbi,
1097 !__check_rb_tree_consistence(sbi, &dcc->root));
bd5b0738 1098 blk_start_plug(&plug);
969d1b18
CY		 1099 for (i = MAX_PLIST_NUM - 1;
1100 i >= 0 && plist_issue(dcc->pend_list_tag[i]); i--) {
bd5b0738
CY		 1101 pend_list = &dcc->pend_list[i];
1102 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1103 f2fs_bug_on(sbi, dc->state != D_PREP);
1104
969d1b18
CY		 1105 /* Hurry up to finish fstrim */
1106 if (dcc->pend_list_tag[i] & P_TRIM) {
bd5b0738 1107 __submit_discard_cmd(sbi, dc);
969d1b18 1108 issued++;
1eb1ef4a
JK		 1109
1110 if (fatal_signal_pending(current))
1111 break;
969d1b18
CY		 1112 continue;
1113 }
1114
e6c6de18
CY		 1115 if (!issue_cond) {
1116 __submit_discard_cmd(sbi, dc);
969d1b18 1117 issued++;
e6c6de18
CY		 1118 continue;
1119 }
1120
1121 if (is_idle(sbi)) {
bd5b0738 1122 __submit_discard_cmd(sbi, dc);
e6c6de18
CY		 1123 issued++;
1124 } else {
1125 io_interrupted = true;
969d1b18 1126 }
e6c6de18
CY		 1127
1128 if (++iter >= DISCARD_ISSUE_RATE)
bd5b0738
CY		 1129 goto out;
1130 }
969d1b18
CY		 1131 if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM)
1132 dcc->pend_list_tag[i] &= (~P_TRIM);
bd5b0738
CY		 1133 }
1134out:
1135 blk_finish_plug(&plug);
1136 mutex_unlock(&dcc->cmd_lock);
969d1b18 1137
e6c6de18
CY		 1138 if (!issued && io_interrupted)
1139 issued = -1;
1140
969d1b18
CY		 1141 return issued;
1142}
1143
1144static void __drop_discard_cmd(struct f2fs_sb_info *sbi)
1145{
1146 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1147 struct list_head *pend_list;
1148 struct discard_cmd *dc, *tmp;
1149 int i;
1150
1151 mutex_lock(&dcc->cmd_lock);
1152 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1153 pend_list = &dcc->pend_list[i];
1154 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1155 f2fs_bug_on(sbi, dc->state != D_PREP);
1156 __remove_discard_cmd(sbi, dc);
1157 }
1158 }
1159 mutex_unlock(&dcc->cmd_lock);
bd5b0738
CY		 1160}
1161
2a510c00
CY		 1162static void __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1163 struct discard_cmd *dc)
1164{
1165 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1166
1167 wait_for_completion_io(&dc->wait);
1168 mutex_lock(&dcc->cmd_lock);
1169 f2fs_bug_on(sbi, dc->state != D_DONE);
1170 dc->ref--;
1171 if (!dc->ref)
1172 __remove_discard_cmd(sbi, dc);
1173 mutex_unlock(&dcc->cmd_lock);
1174}
1175
63a94fa1
CY		 1176static void __wait_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1177{
1178 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1179 struct list_head *wait_list = &(dcc->wait_list);
1180 struct discard_cmd *dc, *tmp;
6afae633
CY		 1181 bool need_wait;
1182
1183next:
1184 need_wait = false;
63a94fa1
CY		 1185
1186 mutex_lock(&dcc->cmd_lock);
1187 list_for_each_entry_safe(dc, tmp, wait_list, list) {
6afae633 1188 if (!wait_cond || (dc->state == D_DONE && !dc->ref)) {
63a94fa1
CY		 1189 wait_for_completion_io(&dc->wait);
1190 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1191 } else {
1192 dc->ref++;
1193 need_wait = true;
1194 break;
63a94fa1
CY		 1195 }
1196 }
1197 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1198
1199 if (need_wait) {
2a510c00 1200 __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1201 goto next;
1202 }
63a94fa1
CY		 1203}
1204
004b6862
CY		 1205/* This should be covered by global mutex, &sit_i->sentry_lock */
1206void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1207{
1208 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1209 struct discard_cmd *dc;
ec9895ad 1210 bool need_wait = false;
004b6862
CY		 1211
1212 mutex_lock(&dcc->cmd_lock);
004b6862
CY		 1213 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1214 if (dc) {
ec9895ad
CY		 1215 if (dc->state == D_PREP) {
1216 __punch_discard_cmd(sbi, dc, blkaddr);
1217 } else {
1218 dc->ref++;
1219 need_wait = true;
1220 }
275b66b0 1221 }
d431413f 1222 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1223
2a510c00
CY		 1224 if (need_wait)
1225 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1226}
1227
cce13252
CY		 1228void stop_discard_thread(struct f2fs_sb_info *sbi)
1229{
1230 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1231
1232 if (dcc && dcc->f2fs_issue_discard) {
1233 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1234
1235 dcc->f2fs_issue_discard = NULL;
1236 kthread_stop(discard_thread);
ec9895ad 1237 }
d431413f
CY		 1238}
1239
1eb1ef4a 1240/* This comes from f2fs_put_super and f2fs_trim_fs */
638164a2 1241void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi, bool umount)
d431413f 1242{
bd5b0738 1243 __issue_discard_cmd(sbi, false);
969d1b18 1244 __drop_discard_cmd(sbi);
638164a2 1245 __wait_discard_cmd(sbi, !umount);
275b66b0
CY		 1246}
1247
969d1b18
CY		 1248static void mark_discard_range_all(struct f2fs_sb_info *sbi)
1249{
1250 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1251 int i;
1252
1253 mutex_lock(&dcc->cmd_lock);
1254 for (i = 0; i < MAX_PLIST_NUM; i++)
1255 dcc->pend_list_tag[i] |= P_TRIM;
1256 mutex_unlock(&dcc->cmd_lock);
1257}
1258
15469963
JK		 1259static int issue_discard_thread(void *data)
1260{
1261 struct f2fs_sb_info *sbi = data;
1262 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1263 wait_queue_head_t *q = &dcc->discard_wait_queue;
969d1b18
CY		 1264 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1265 int issued;
15469963 1266
1d7be270 1267 set_freezable();
15469963 1268
1d7be270 1269 do {
969d1b18
CY		 1270 wait_event_interruptible_timeout(*q,
1271 kthread_should_stop() || freezing(current) ||
1272 dcc->discard_wake,
1273 msecs_to_jiffies(wait_ms));
1d7be270
JK		 1274 if (try_to_freeze())
1275 continue;
1276 if (kthread_should_stop())
1277 return 0;
15469963 1278
5f656541 1279 if (dcc->discard_wake) {
969d1b18 1280 dcc->discard_wake = 0;
5f656541
JK		 1281 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1282 mark_discard_range_all(sbi);
1283 }
969d1b18 1284
dc6febb6
CY		 1285 sb_start_intwrite(sbi->sb);
1286
969d1b18
CY		 1287 issued = __issue_discard_cmd(sbi, true);
1288 if (issued) {
1289 __wait_discard_cmd(sbi, true);
1290 wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1291 } else {
1292 wait_ms = DEF_MAX_DISCARD_ISSUE_TIME;
1293 }
1d7be270 1294
dc6febb6 1295 sb_end_intwrite(sbi->sb);
1d7be270 1296
1d7be270
JK		 1297 } while (!kthread_should_stop());
1298 return 0;
15469963
JK		 1299}
1300
f46e8809 1301#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1302static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1303 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1304{
92592285 1305 sector_t sector, nr_sects;
10a875f8 1306 block_t lblkstart = blkstart;
3c62be17
JK		 1307 int devi = 0;
1308
1309 if (sbi->s_ndevs) {
1310 devi = f2fs_target_device_index(sbi, blkstart);
1311 blkstart -= FDEV(devi).start_blk;
1312 }
f46e8809
DLM		 1313
1314 /*
1315 * We need to know the type of the zone: for conventional zones,
1316 * use regular discard if the drive supports it. For sequential
1317 * zones, reset the zone write pointer.
1318 */
3c62be17 1319 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1320
1321 case BLK_ZONE_TYPE_CONVENTIONAL:
1322 if (!blk_queue_discard(bdev_get_queue(bdev)))
1323 return 0;
c81abe34 1324 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1325 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1326 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1327 sector = SECTOR_FROM_BLOCK(blkstart);
1328 nr_sects = SECTOR_FROM_BLOCK(blklen);
1329
1330 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1331 nr_sects != bdev_zone_sectors(bdev)) {
1332 f2fs_msg(sbi->sb, KERN_INFO,
1333 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1334 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1335 blkstart, blklen);
1336 return -EIO;
1337 }
d50aaeec 1338 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1339 return blkdev_reset_zones(bdev, sector,
1340 nr_sects, GFP_NOFS);
1341 default:
1342 /* Unknown zone type: broken device ? */
1343 return -EIO;
1344 }
1345}
1346#endif
1347
3c62be17
JK		 1348static int __issue_discard_async(struct f2fs_sb_info *sbi,
1349 struct block_device *bdev, block_t blkstart, block_t blklen)
1350{
1351#ifdef CONFIG_BLK_DEV_ZONED
1352 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1353 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1354 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1355#endif
c81abe34 1356 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1357}
1358
1e87a78d 1359static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1360 block_t blkstart, block_t blklen)
1361{
3c62be17
JK		 1362 sector_t start = blkstart, len = 0;
1363 struct block_device *bdev;
a66cdd98
JK		 1364 struct seg_entry *se;
1365 unsigned int offset;
1366 block_t i;
3c62be17
JK		 1367 int err = 0;
1368
1369 bdev = f2fs_target_device(sbi, blkstart, NULL);
1370
1371 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1372 if (i != start) {
1373 struct block_device *bdev2 =
1374 f2fs_target_device(sbi, i, NULL);
1375
1376 if (bdev2 != bdev) {
1377 err = __issue_discard_async(sbi, bdev,
1378 start, len);
1379 if (err)
1380 return err;
1381 bdev = bdev2;
1382 start = i;
1383 len = 0;
1384 }
1385 }
a66cdd98 1386
a66cdd98
JK		 1387 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1388 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1389
1390 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1391 sbi->discard_blks--;
1392 }
f46e8809 1393
3c62be17
JK		 1394 if (len)
1395 err = __issue_discard_async(sbi, bdev, start, len);
1396 return err;
1e87a78d
JK		 1397}
1398
25290fa5
JK		 1399static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1400 bool check_only)
adf4983b 1401{
b2955550
JK		 1402 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1403 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1404 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1405 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1406 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1407 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1408 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1409 unsigned int start = 0, end = -1;
c473f1a9 1410 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1411 struct discard_entry *de = NULL;
46f84c2c 1412 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1413 int i;
1414
3e025740 1415 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1416 return false;
b2955550 1417
a66cdd98
JK		 1418 if (!force) {
1419 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1420 SM_I(sbi)->dcc_info->nr_discards >=
1421 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1422 return false;
4b2fecc8
JK		 1423 }
1424
b2955550
JK		 1425 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1426 for (i = 0; i < entries; i++)
a66cdd98 1427 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1428 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1429
0b54fb84
JK		 1430 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1431 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1432 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1433 if (start >= max_blocks)
1434 break;
1435
1436 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1437 if (force && start && end != max_blocks
1438 && (end - start) < cpc->trim_minlen)
1439 continue;
1440
25290fa5
JK		 1441 if (check_only)
1442 return true;
1443
a7eeb823
CY		 1444 if (!de) {
1445 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1446 GFP_F2FS_ZERO);
1447 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1448 list_add_tail(&de->list, head);
1449 }
1450
1451 for (i = start; i < end; i++)
1452 __set_bit_le(i, (void *)de->discard_map);
1453
1454 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1455 }
25290fa5 1456 return false;
b2955550
JK		 1457}
1458
4b2fecc8
JK		 1459void release_discard_addrs(struct f2fs_sb_info *sbi)
1460{
46f84c2c 1461 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1462 struct discard_entry *entry, *this;
1463
1464 /* drop caches */
1465 list_for_each_entry_safe(entry, this, head, list) {
1466 list_del(&entry->list);
1467 kmem_cache_free(discard_entry_slab, entry);
1468 }
1469}
1470
0a8165d7 1471/*
351df4b2
JK		 1472 * Should call clear_prefree_segments after checkpoint is done.
1473 */
1474static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1475{
1476 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1477 unsigned int segno;
351df4b2
JK		 1478
1479 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1480 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1481 __set_test_and_free(sbi, segno);
351df4b2
JK		 1482 mutex_unlock(&dirty_i->seglist_lock);
1483}
1484
836b5a63 1485void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1486{
969d1b18
CY		 1487 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1488 struct list_head *head = &dcc->entry_list;
2d7b822a 1489 struct discard_entry *entry, *this;
351df4b2 1490 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1491 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1492 unsigned int start = 0, end = -1;
36abef4e 1493 unsigned int secno, start_segno;
c473f1a9 1494 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK		 1495
1496 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1497
351df4b2 1498 while (1) {
29e59c14 1499 int i;
7cd8558b
JK		 1500 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1501 if (start >= MAIN_SEGS(sbi))
351df4b2 1502 break;
7cd8558b
JK		 1503 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1504 start + 1);
29e59c14
CL		 1505
1506 for (i = start; i < end; i++)
1507 clear_bit(i, prefree_map);
1508
1509 dirty_i->nr_dirty[PRE] -= end - start;
1510
650d3c4e 1511 if (!test_opt(sbi, DISCARD))
29e59c14 1512 continue;
351df4b2 1513
650d3c4e
YH		 1514 if (force && start >= cpc->trim_start &&
1515 (end - 1) <= cpc->trim_end)
1516 continue;
1517
36abef4e
JK		 1518 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1519 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1520 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1521 continue;
1522 }
1523next:
4ddb1a4d
JK		 1524 secno = GET_SEC_FROM_SEG(sbi, start);
1525 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1526 if (!IS_CURSEC(sbi, secno) &&
302bd348 1527 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1528 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1529 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1530
1531 start = start_segno + sbi->segs_per_sec;
1532 if (start < end)
1533 goto next;
8b107f5b
JK		 1534 else
1535 end = start - 1;
351df4b2
JK		 1536 }
1537 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1538
1539 /* send small discards */
2d7b822a 1540 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1541 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1542 bool is_valid = test_bit_le(0, entry->discard_map);
1543
1544find_next:
1545 if (is_valid) {
1546 next_pos = find_next_zero_bit_le(entry->discard_map,
1547 sbi->blocks_per_seg, cur_pos);
1548 len = next_pos - cur_pos;
1549
acfd2810
DLM		 1550 if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1551 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1552 goto skip;
1553
1554 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1555 len);
1556 cpc->trimmed += len;
1557 total_len += len;
1558 } else {
1559 next_pos = find_next_bit_le(entry->discard_map,
1560 sbi->blocks_per_seg, cur_pos);
1561 }
836b5a63 1562skip:
a7eeb823
CY		 1563 cur_pos = next_pos;
1564 is_valid = !is_valid;
1565
1566 if (cur_pos < sbi->blocks_per_seg)
1567 goto find_next;
1568
b2955550 1569 list_del(&entry->list);
969d1b18 1570 dcc->nr_discards -= total_len;
b2955550
JK		 1571 kmem_cache_free(discard_entry_slab, entry);
1572 }
34e159da 1573
01983c71 1574 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1575}
1576
8ed59745 1577static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1578{
15469963 1579 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1580 struct discard_cmd_control *dcc;
ba48a33e 1581 int err = 0, i;
0b54fb84
JK		 1582
1583 if (SM_I(sbi)->dcc_info) {
1584 dcc = SM_I(sbi)->dcc_info;
1585 goto init_thread;
1586 }
1587
1588 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1589 if (!dcc)
1590 return -ENOMEM;
1591
969d1b18 1592 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1593 INIT_LIST_HEAD(&dcc->entry_list);
969d1b18 1594 for (i = 0; i < MAX_PLIST_NUM; i++) {
ba48a33e 1595 INIT_LIST_HEAD(&dcc->pend_list[i]);
969d1b18
CY		 1596 if (i >= dcc->discard_granularity - 1)
1597 dcc->pend_list_tag[i] |= P_ACTIVE;
1598 }
46f84c2c 1599 INIT_LIST_HEAD(&dcc->wait_list);
15469963 1600 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1601 atomic_set(&dcc->issued_discard, 0);
1602 atomic_set(&dcc->issing_discard, 0);
5f32366a 1603 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1604 dcc->nr_discards = 0;
d618ebaf 1605 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1606 dcc->undiscard_blks = 0;
004b6862 1607 dcc->root = RB_ROOT;
0b54fb84 1608
15469963 1609 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1610 SM_I(sbi)->dcc_info = dcc;
1611init_thread:
15469963
JK		 1612 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1613 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1614 if (IS_ERR(dcc->f2fs_issue_discard)) {
1615 err = PTR_ERR(dcc->f2fs_issue_discard);
1616 kfree(dcc);
1617 SM_I(sbi)->dcc_info = NULL;
1618 return err;
1619 }
1620
0b54fb84
JK		 1621 return err;
1622}
1623
f099405f 1624static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1625{
1626 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1627
f099405f
CY		 1628 if (!dcc)
1629 return;
1630
cce13252 1631 stop_discard_thread(sbi);
f099405f
CY		 1632
1633 kfree(dcc);
1634 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1635}
1636
184a5cd2 1637static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1638{
1639 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1640
1641 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1642 sit_i->dirty_sentries++;
184a5cd2
CY		 1643 return false;
1644 }
1645
1646 return true;
351df4b2
JK		 1647}
1648
1649static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1650 unsigned int segno, int modified)
1651{
1652 struct seg_entry *se = get_seg_entry(sbi, segno);
1653 se->type = type;
1654 if (modified)
1655 __mark_sit_entry_dirty(sbi, segno);
1656}
1657
1658static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1659{
1660 struct seg_entry *se;
1661 unsigned int segno, offset;
1662 long int new_vblocks;
6415fedc
YS		 1663 bool exist;
1664#ifdef CONFIG_F2FS_CHECK_FS
1665 bool mir_exist;
1666#endif
351df4b2
JK		 1667
1668 segno = GET_SEGNO(sbi, blkaddr);
1669
1670 se = get_seg_entry(sbi, segno);
1671 new_vblocks = se->valid_blocks + del;
491c0854 1672 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1673
9850cf4a 1674 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1675 (new_vblocks > sbi->blocks_per_seg)));
1676
1677 se->valid_blocks = new_vblocks;
1678 se->mtime = get_mtime(sbi);
1679 SIT_I(sbi)->max_mtime = se->mtime;
1680
1681 /* Update valid block bitmap */
1682 if (del > 0) {
6415fedc 1683 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1684#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1685 mir_exist = f2fs_test_and_set_bit(offset,
1686 se->cur_valid_map_mir);
1687 if (unlikely(exist != mir_exist)) {
1688 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1689 "when setting bitmap, blk:%u, old bit:%d",
1690 blkaddr, exist);
05796763 1691 f2fs_bug_on(sbi, 1);
6415fedc 1692 }
355e7891 1693#endif
6415fedc
YS		 1694 if (unlikely(exist)) {
1695 f2fs_msg(sbi->sb, KERN_ERR,
1696 "Bitmap was wrongly set, blk:%u", blkaddr);
1697 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1698 se->valid_blocks--;
1699 del = 0;
355e7891 1700 }
6415fedc 1701
3e025740
JK		 1702 if (f2fs_discard_en(sbi) &&
1703 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1704 sbi->discard_blks--;
720037f9
JK		 1705
1706 /* don't overwrite by SSR to keep node chain */
1707 if (se->type == CURSEG_WARM_NODE) {
1708 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1709 se->ckpt_valid_blocks++;
1710 }
351df4b2 1711 } else {
6415fedc 1712 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1713#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1714 mir_exist = f2fs_test_and_clear_bit(offset,
1715 se->cur_valid_map_mir);
1716 if (unlikely(exist != mir_exist)) {
1717 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1718 "when clearing bitmap, blk:%u, old bit:%d",
1719 blkaddr, exist);
05796763 1720 f2fs_bug_on(sbi, 1);
6415fedc 1721 }
355e7891 1722#endif
6415fedc
YS		 1723 if (unlikely(!exist)) {
1724 f2fs_msg(sbi->sb, KERN_ERR,
1725 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1726 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1727 se->valid_blocks++;
1728 del = 0;
355e7891 1729 }
6415fedc 1730
3e025740
JK		 1731 if (f2fs_discard_en(sbi) &&
1732 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1733 sbi->discard_blks++;
351df4b2
JK		 1734 }
1735 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1736 se->ckpt_valid_blocks += del;
1737
1738 __mark_sit_entry_dirty(sbi, segno);
1739
1740 /* update total number of valid blocks to be written in ckpt area */
1741 SIT_I(sbi)->written_valid_blocks += del;
1742
1743 if (sbi->segs_per_sec > 1)
1744 get_sec_entry(sbi, segno)->valid_blocks += del;
1745}
1746
5e443818 1747void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1748{
5e443818
JK		 1749 update_sit_entry(sbi, new, 1);
1750 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1751 update_sit_entry(sbi, old, -1);
1752
1753 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1754 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 1755}
1756
1757void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1758{
1759 unsigned int segno = GET_SEGNO(sbi, addr);
1760 struct sit_info *sit_i = SIT_I(sbi);
1761
9850cf4a 1762 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1763 if (addr == NEW_ADDR)
1764 return;
1765
1766 /* add it into sit main buffer */
1767 mutex_lock(&sit_i->sentry_lock);
1768
1769 update_sit_entry(sbi, addr, -1);
1770
1771 /* add it into dirty seglist */
1772 locate_dirty_segment(sbi, segno);
1773
1774 mutex_unlock(&sit_i->sentry_lock);
1775}
1776
6e2c64ad
JK		 1777bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1778{
1779 struct sit_info *sit_i = SIT_I(sbi);
1780 unsigned int segno, offset;
1781 struct seg_entry *se;
1782 bool is_cp = false;
1783
1784 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1785 return true;
1786
1787 mutex_lock(&sit_i->sentry_lock);
1788
1789 segno = GET_SEGNO(sbi, blkaddr);
1790 se = get_seg_entry(sbi, segno);
1791 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1792
1793 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1794 is_cp = true;
1795
1796 mutex_unlock(&sit_i->sentry_lock);
1797
1798 return is_cp;
1799}
1800
0a8165d7 1801/*
351df4b2
JK		 1802 * This function should be resided under the curseg_mutex lock
1803 */
1804static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1805 struct f2fs_summary *sum)
351df4b2
JK		 1806{
1807 struct curseg_info *curseg = CURSEG_I(sbi, type);
1808 void *addr = curseg->sum_blk;
e79efe3b 1809 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1810 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1811}
1812
0a8165d7 1813/*
351df4b2
JK		 1814 * Calculate the number of current summary pages for writing
1815 */
3fa06d7b 1816int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1817{
351df4b2 1818 int valid_sum_count = 0;
9a47938b 1819 int i, sum_in_page;
351df4b2
JK		 1820
1821 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1822 if (sbi->ckpt->alloc_type[i] == SSR)
1823 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1824 else {
1825 if (for_ra)
1826 valid_sum_count += le16_to_cpu(
1827 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1828 else
1829 valid_sum_count += curseg_blkoff(sbi, i);
1830 }
351df4b2
JK		 1831 }
1832
09cbfeaf 1833 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1834 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1835 if (valid_sum_count <= sum_in_page)
351df4b2 1836 return 1;
9a47938b 1837 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1838 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1839 return 2;
1840 return 3;
1841}
1842
0a8165d7 1843/*
351df4b2
JK		 1844 * Caller should put this summary page
1845 */
1846struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1847{
1848 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1849}
1850
381722d2 1851void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1852{
1853 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1854 void *dst = page_address(page);
1855
1856 if (src)
09cbfeaf 1857 memcpy(dst, src, PAGE_SIZE);
381722d2 1858 else
09cbfeaf 1859 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1860 set_page_dirty(page);
1861 f2fs_put_page(page, 1);
1862}
1863
381722d2
CY		 1864static void write_sum_page(struct f2fs_sb_info *sbi,
1865 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1866{
1867 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1868}
1869
b7ad7512
CY		 1870static void write_current_sum_page(struct f2fs_sb_info *sbi,
1871 int type, block_t blk_addr)
1872{
1873 struct curseg_info *curseg = CURSEG_I(sbi, type);
1874 struct page *page = grab_meta_page(sbi, blk_addr);
1875 struct f2fs_summary_block *src = curseg->sum_blk;
1876 struct f2fs_summary_block *dst;
1877
1878 dst = (struct f2fs_summary_block *)page_address(page);
1879
1880 mutex_lock(&curseg->curseg_mutex);
1881
1882 down_read(&curseg->journal_rwsem);
1883 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1884 up_read(&curseg->journal_rwsem);
1885
1886 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1887 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1888
1889 mutex_unlock(&curseg->curseg_mutex);
1890
1891 set_page_dirty(page);
1892 f2fs_put_page(page, 1);
1893}
1894
a7881893
JK		 1895static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1896{
1897 struct curseg_info *curseg = CURSEG_I(sbi, type);
1898 unsigned int segno = curseg->segno + 1;
1899 struct free_segmap_info *free_i = FREE_I(sbi);
1900
1901 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1902 return !test_bit(segno, free_i->free_segmap);
1903 return 0;
1904}
1905
0a8165d7 1906/*
351df4b2
JK		 1907 * Find a new segment from the free segments bitmap to right order
1908 * This function should be returned with success, otherwise BUG
1909 */
1910static void get_new_segment(struct f2fs_sb_info *sbi,
1911 unsigned int *newseg, bool new_sec, int dir)
1912{
1913 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1914 unsigned int segno, secno, zoneno;
7cd8558b 1915 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 1916 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1917 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 1918 unsigned int left_start = hint;
1919 bool init = true;
1920 int go_left = 0;
1921 int i;
1922
1a118ccf 1923 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1924
1925 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1926 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 1927 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
1928 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 1929 goto got_it;
1930 }
1931find_other_zone:
7cd8558b
JK		 1932 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1933 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1934 if (dir == ALLOC_RIGHT) {
1935 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1936 MAIN_SECS(sbi), 0);
1937 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1938 } else {
1939 go_left = 1;
1940 left_start = hint - 1;
1941 }
1942 }
1943 if (go_left == 0)
1944 goto skip_left;
1945
1946 while (test_bit(left_start, free_i->free_secmap)) {
1947 if (left_start > 0) {
1948 left_start--;
1949 continue;
1950 }
1951 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1952 MAIN_SECS(sbi), 0);
1953 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1954 break;
1955 }
1956 secno = left_start;
1957skip_left:
1958 hint = secno;
4ddb1a4d
JK		 1959 segno = GET_SEG_FROM_SEC(sbi, secno);
1960 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 1961
1962 /* give up on finding another zone */
1963 if (!init)
1964 goto got_it;
1965 if (sbi->secs_per_zone == 1)
1966 goto got_it;
1967 if (zoneno == old_zoneno)
1968 goto got_it;
1969 if (dir == ALLOC_LEFT) {
1970 if (!go_left && zoneno + 1 >= total_zones)
1971 goto got_it;
1972 if (go_left && zoneno == 0)
1973 goto got_it;
1974 }
1975 for (i = 0; i < NR_CURSEG_TYPE; i++)
1976 if (CURSEG_I(sbi, i)->zone == zoneno)
1977 break;
1978
1979 if (i < NR_CURSEG_TYPE) {
1980 /* zone is in user, try another */
1981 if (go_left)
1982 hint = zoneno * sbi->secs_per_zone - 1;
1983 else if (zoneno + 1 >= total_zones)
1984 hint = 0;
1985 else
1986 hint = (zoneno + 1) * sbi->secs_per_zone;
1987 init = false;
1988 goto find_other_zone;
1989 }
1990got_it:
1991 /* set it as dirty segment in free segmap */
9850cf4a 1992 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1993 __set_inuse(sbi, segno);
1994 *newseg = segno;
1a118ccf 1995 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1996}
1997
1998static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1999{
2000 struct curseg_info *curseg = CURSEG_I(sbi, type);
2001 struct summary_footer *sum_footer;
2002
2003 curseg->segno = curseg->next_segno;
4ddb1a4d 2004 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2005 curseg->next_blkoff = 0;
2006 curseg->next_segno = NULL_SEGNO;
2007
2008 sum_footer = &(curseg->sum_blk->footer);
2009 memset(sum_footer, 0, sizeof(struct summary_footer));
2010 if (IS_DATASEG(type))
2011 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2012 if (IS_NODESEG(type))
2013 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2014 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2015}
2016
7a20b8a6
JK		 2017static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2018{
a7881893
JK		 2019 /* if segs_per_sec is large than 1, we need to keep original policy. */
2020 if (sbi->segs_per_sec != 1)
2021 return CURSEG_I(sbi, type)->segno;
2022
7a20b8a6
JK		 2023 if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
2024 return 0;
2025
e066b83c
JK		 2026 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2027 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
7a20b8a6
JK		 2028 return CURSEG_I(sbi, type)->segno;
2029}
2030
0a8165d7 2031/*
351df4b2
JK		 2032 * Allocate a current working segment.
2033 * This function always allocates a free segment in LFS manner.
2034 */
2035static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2036{
2037 struct curseg_info *curseg = CURSEG_I(sbi, type);
2038 unsigned int segno = curseg->segno;
2039 int dir = ALLOC_LEFT;
2040
2041 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2042 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2043 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2044 dir = ALLOC_RIGHT;
2045
2046 if (test_opt(sbi, NOHEAP))
2047 dir = ALLOC_RIGHT;
2048
7a20b8a6 2049 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2050 get_new_segment(sbi, &segno, new_sec, dir);
2051 curseg->next_segno = segno;
2052 reset_curseg(sbi, type, 1);
2053 curseg->alloc_type = LFS;
2054}
2055
2056static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2057 struct curseg_info *seg, block_t start)
2058{
2059 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2060 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2061 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2062 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2063 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2064 int i, pos;
2065
2066 for (i = 0; i < entries; i++)
2067 target_map[i] = ckpt_map[i] | cur_map[i];
2068
2069 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2070
2071 seg->next_blkoff = pos;
351df4b2
JK		 2072}
2073
0a8165d7 2074/*
351df4b2
JK		 2075 * If a segment is written by LFS manner, next block offset is just obtained
2076 * by increasing the current block offset. However, if a segment is written by
2077 * SSR manner, next block offset obtained by calling __next_free_blkoff
2078 */
2079static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2080 struct curseg_info *seg)
2081{
2082 if (seg->alloc_type == SSR)
2083 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2084 else
2085 seg->next_blkoff++;
2086}
2087
0a8165d7 2088/*
e1c42045 2089 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2090 * manner, so it should recover the existing segment information of valid blocks
2091 */
025d63a4 2092static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2093{
2094 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2095 struct curseg_info *curseg = CURSEG_I(sbi, type);
2096 unsigned int new_segno = curseg->next_segno;
2097 struct f2fs_summary_block *sum_node;
2098 struct page *sum_page;
2099
2100 write_sum_page(sbi, curseg->sum_blk,
2101 GET_SUM_BLOCK(sbi, curseg->segno));
2102 __set_test_and_inuse(sbi, new_segno);
2103
2104 mutex_lock(&dirty_i->seglist_lock);
2105 __remove_dirty_segment(sbi, new_segno, PRE);
2106 __remove_dirty_segment(sbi, new_segno, DIRTY);
2107 mutex_unlock(&dirty_i->seglist_lock);
2108
2109 reset_curseg(sbi, type, 1);
2110 curseg->alloc_type = SSR;
2111 __next_free_blkoff(sbi, curseg, 0);
2112
025d63a4
CY		 2113 sum_page = get_sum_page(sbi, new_segno);
2114 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2115 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2116 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2117}
2118
43727527
JK		 2119static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2120{
2121 struct curseg_info *curseg = CURSEG_I(sbi, type);
2122 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2123 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2124 int i, cnt;
2125 bool reversed = false;
c192f7a4
JK		 2126
2127 /* need_SSR() already forces to do this */
e066b83c
JK		 2128 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2129 curseg->next_segno = segno;
c192f7a4 2130 return 1;
e066b83c 2131 }
43727527 2132
70d625cb
JK		 2133 /* For node segments, let's do SSR more intensively */
2134 if (IS_NODESEG(type)) {
d27c3d89
CY		 2135 if (type >= CURSEG_WARM_NODE) {
2136 reversed = true;
2137 i = CURSEG_COLD_NODE;
2138 } else {
2139 i = CURSEG_HOT_NODE;
2140 }
2141 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2142 } else {
d27c3d89
CY		 2143 if (type >= CURSEG_WARM_DATA) {
2144 reversed = true;
2145 i = CURSEG_COLD_DATA;
2146 } else {
2147 i = CURSEG_HOT_DATA;
2148 }
2149 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2150 }
43727527 2151
d27c3d89 2152 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2153 if (i == type)
2154 continue;
e066b83c
JK		 2155 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2156 curseg->next_segno = segno;
43727527 2157 return 1;
e066b83c 2158 }
c192f7a4 2159 }
43727527
JK		 2160 return 0;
2161}
2162
351df4b2
JK		 2163/*
2164 * flush out current segment and replace it with new segment
2165 * This function should be returned with success, otherwise BUG
2166 */
2167static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2168 int type, bool force)
2169{
a7881893
JK		 2170 struct curseg_info *curseg = CURSEG_I(sbi, type);
2171
7b405275 2172 if (force)
351df4b2 2173 new_curseg(sbi, type, true);
5b6c6be2
JK		 2174 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2175 type == CURSEG_WARM_NODE)
351df4b2 2176 new_curseg(sbi, type, false);
a7881893
JK		 2177 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2178 new_curseg(sbi, type, false);
351df4b2 2179 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2180 change_curseg(sbi, type);
351df4b2
JK		 2181 else
2182 new_curseg(sbi, type, false);
dcdfff65 2183
a7881893 2184 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2185}
2186
2187void allocate_new_segments(struct f2fs_sb_info *sbi)
2188{
6ae1be13
JK		 2189 struct curseg_info *curseg;
2190 unsigned int old_segno;
351df4b2
JK		 2191 int i;
2192
6ae1be13
JK		 2193 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2194 curseg = CURSEG_I(sbi, i);
2195 old_segno = curseg->segno;
2196 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2197 locate_dirty_segment(sbi, old_segno);
2198 }
351df4b2
JK		 2199}
2200
2201static const struct segment_allocation default_salloc_ops = {
2202 .allocate_segment = allocate_segment_by_default,
2203};
2204
25290fa5
JK		 2205bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2206{
2207 __u64 trim_start = cpc->trim_start;
2208 bool has_candidate = false;
2209
2210 mutex_lock(&SIT_I(sbi)->sentry_lock);
2211 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2212 if (add_discard_addrs(sbi, cpc, true)) {
2213 has_candidate = true;
2214 break;
2215 }
2216 }
2217 mutex_unlock(&SIT_I(sbi)->sentry_lock);
2218
2219 cpc->trim_start = trim_start;
2220 return has_candidate;
2221}
2222
4b2fecc8
JK		 2223int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2224{
f7ef9b83
JK		 2225 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2226 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 2227 unsigned int start_segno, end_segno;
2228 struct cp_control cpc;
c34f42e2 2229 int err = 0;
4b2fecc8 2230
836b5a63 2231 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2232 return -EINVAL;
2233
9bd27ae4 2234 cpc.trimmed = 0;
7cd8558b 2235 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 2236 goto out;
2237
ed214a11
YH		 2238 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2239 f2fs_msg(sbi->sb, KERN_WARNING,
2240 "Found FS corruption, run fsck to fix.");
2241 goto out;
2242 }
2243
4b2fecc8 2244 /* start/end segment number in main_area */
7cd8558b
JK		 2245 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2246 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2247 GET_SEGNO(sbi, end);
4b2fecc8 2248 cpc.reason = CP_DISCARD;
836b5a63 2249 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 2250
2251 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 2252 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
2253 cpc.trim_start = start_segno;
a66cdd98
JK		 2254
2255 if (sbi->discard_blks == 0)
2256 break;
2257 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2258 cpc.trim_end = end_segno;
2259 else
2260 cpc.trim_end = min_t(unsigned int,
2261 rounddown(start_segno +
bba681cb
JK		 2262 BATCHED_TRIM_SEGMENTS(sbi),
2263 sbi->segs_per_sec) - 1, end_segno);
2264
2265 mutex_lock(&sbi->gc_mutex);
c34f42e2 2266 err = write_checkpoint(sbi, &cpc);
bba681cb 2267 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 2268 if (err)
2269 break;
74fa5f3d
CY		 2270
2271 schedule();
bba681cb 2272 }
969d1b18
CY		 2273 /* It's time to issue all the filed discards */
2274 mark_discard_range_all(sbi);
638164a2 2275 f2fs_wait_discard_bios(sbi, false);
4b2fecc8 2276out:
f7ef9b83 2277 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 2278 return err;
4b2fecc8
JK		 2279}
2280
351df4b2
JK		 2281static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2282{
2283 struct curseg_info *curseg = CURSEG_I(sbi, type);
2284 if (curseg->next_blkoff < sbi->blocks_per_seg)
2285 return true;
2286 return false;
2287}
2288
81377bd6 2289static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2290{
81377bd6 2291 if (fio->type == DATA)
351df4b2
JK		 2292 return CURSEG_HOT_DATA;
2293 else
2294 return CURSEG_HOT_NODE;
2295}
2296
81377bd6 2297static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2298{
81377bd6
JK		 2299 if (fio->type == DATA) {
2300 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2301
2302 if (S_ISDIR(inode->i_mode))
2303 return CURSEG_HOT_DATA;
2304 else
2305 return CURSEG_COLD_DATA;
2306 } else {
81377bd6 2307 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2308 return CURSEG_WARM_NODE;
351df4b2
JK		 2309 else
2310 return CURSEG_COLD_NODE;
2311 }
2312}
2313
81377bd6 2314static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2315{
81377bd6
JK		 2316 if (fio->type == DATA) {
2317 struct inode *inode = fio->page->mapping->host;
351df4b2 2318
81377bd6 2319 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2320 return CURSEG_COLD_DATA;
ef095d19
JK		 2321 if (is_inode_flag_set(inode, FI_HOT_DATA))
2322 return CURSEG_HOT_DATA;
2323 return CURSEG_WARM_DATA;
351df4b2 2324 } else {
81377bd6
JK		 2325 if (IS_DNODE(fio->page))
2326 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2327 CURSEG_HOT_NODE;
ef095d19 2328 return CURSEG_COLD_NODE;
351df4b2
JK		 2329 }
2330}
2331
81377bd6 2332static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2333{
a912b54d
JK		 2334 int type = 0;
2335
81377bd6 2336 switch (fio->sbi->active_logs) {
351df4b2 2337 case 2:
a912b54d
JK		 2338 type = __get_segment_type_2(fio);
2339 break;
351df4b2 2340 case 4:
a912b54d
JK		 2341 type = __get_segment_type_4(fio);
2342 break;
2343 case 6:
2344 type = __get_segment_type_6(fio);
2345 break;
2346 default:
2347 f2fs_bug_on(fio->sbi, true);
351df4b2 2348 }
81377bd6 2349
a912b54d
JK		 2350 if (IS_HOT(type))
2351 fio->temp = HOT;
2352 else if (IS_WARM(type))
2353 fio->temp = WARM;
2354 else
2355 fio->temp = COLD;
2356 return type;
351df4b2
JK		 2357}
2358
bfad7c2d
JK		 2359void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2360 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2361 struct f2fs_summary *sum, int type,
2362 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2363{
2364 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2365 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 2366
2367 mutex_lock(&curseg->curseg_mutex);
21cb1d99 2368 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK		 2369
2370 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2371
4e6a8d9b
JK		 2372 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2373
351df4b2
JK		 2374 /*
2375 * __add_sum_entry should be resided under the curseg_mutex
2376 * because, this function updates a summary entry in the
2377 * current summary block.
2378 */
e79efe3b 2379 __add_sum_entry(sbi, type, sum);
351df4b2 2380
351df4b2 2381 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2382
2383 stat_inc_block_count(sbi, curseg);
351df4b2 2384
c6f82fe9
JK		 2385 if (!__has_curseg_space(sbi, type))
2386 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2 2387 /*
c6f82fe9
JK		 2388 * SIT information should be updated after segment allocation,
2389 * since we need to keep dirty segments precisely under SSR.
351df4b2
JK		 2390 */
2391 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 2392
351df4b2
JK		 2393 mutex_unlock(&sit_i->sentry_lock);
2394
704956ec 2395 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2396 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2397
704956ec
CY		 2398 f2fs_inode_chksum_set(sbi, page);
2399 }
2400
fb830fc5
CY		 2401 if (add_list) {
2402 struct f2fs_bio_info *io;
2403
2404 INIT_LIST_HEAD(&fio->list);
2405 fio->in_list = true;
2406 io = sbi->write_io[fio->type] + fio->temp;
2407 spin_lock(&io->io_lock);
2408 list_add_tail(&fio->list, &io->io_list);
2409 spin_unlock(&io->io_lock);
2410 }
2411
bfad7c2d
JK		 2412 mutex_unlock(&curseg->curseg_mutex);
2413}
2414
39d787be
CY		 2415static void update_device_state(struct f2fs_io_info *fio)
2416{
2417 struct f2fs_sb_info *sbi = fio->sbi;
2418 unsigned int devidx;
2419
2420 if (!sbi->s_ndevs)
2421 return;
2422
2423 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2424
2425 /* update device state for fsync */
2426 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 2427
2428 /* update device state for checkpoint */
2429 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2430 spin_lock(&sbi->dev_lock);
2431 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2432 spin_unlock(&sbi->dev_lock);
2433 }
39d787be
CY		 2434}
2435
05ca3632 2436static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2437{
81377bd6 2438 int type = __get_segment_type(fio);
0a595eba 2439 int err;
bfad7c2d 2440
0a595eba 2441reallocate:
7a9d7548 2442 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2443 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2444
351df4b2 2445 /* writeout dirty page into bdev */
b9109b0e 2446 err = f2fs_submit_page_write(fio);
0a595eba
JK		 2447 if (err == -EAGAIN) {
2448 fio->old_blkaddr = fio->new_blkaddr;
2449 goto reallocate;
39d787be
CY		 2450 } else if (!err) {
2451 update_device_state(fio);
0a595eba 2452 }
351df4b2
JK		 2453}
2454
b0af6d49
CY		 2455void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2456 enum iostat_type io_type)
351df4b2 2457{
458e6197 2458 struct f2fs_io_info fio = {
05ca3632 2459 .sbi = sbi,
458e6197 2460 .type = META,
04d328de 2461 .op = REQ_OP_WRITE,
70fd7614 2462 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2463 .old_blkaddr = page->index,
2464 .new_blkaddr = page->index,
05ca3632 2465 .page = page,
4375a336 2466 .encrypted_page = NULL,
fb830fc5 2467 .in_list = false,
458e6197
JK		 2468 };
2469
2b947003 2470 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2471 fio.op_flags &= ~REQ_META;
2b947003 2472
351df4b2 2473 set_page_writeback(page);
b9109b0e 2474 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2475
2476 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2477}
2478
05ca3632 2479void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2480{
2481 struct f2fs_summary sum;
05ca3632 2482
351df4b2 2483 set_summary(&sum, nid, 0, 0);
05ca3632 2484 do_write_page(&sum, fio);
b0af6d49
CY		 2485
2486 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2487}
2488
05ca3632 2489void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2490{
05ca3632 2491 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2492 struct f2fs_summary sum;
2493 struct node_info ni;
2494
9850cf4a 2495 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2496 get_node_info(sbi, dn->nid, &ni);
2497 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2498 do_write_page(&sum, fio);
f28b3434 2499 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2500
2501 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2502}
2503
d1b3e72d 2504int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2505{
b0af6d49
CY		 2506 int err;
2507
7a9d7548 2508 fio->new_blkaddr = fio->old_blkaddr;
05ca3632 2509 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2510
2511 err = f2fs_submit_page_bio(fio);
39d787be
CY		 2512 if (!err)
2513 update_device_state(fio);
b0af6d49
CY		 2514
2515 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2516
2517 return err;
351df4b2
JK		 2518}
2519
4356e48e 2520void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2521 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2522 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2523{
2524 struct sit_info *sit_i = SIT_I(sbi);
2525 struct curseg_info *curseg;
2526 unsigned int segno, old_cursegno;
2527 struct seg_entry *se;
2528 int type;
19f106bc 2529 unsigned short old_blkoff;
351df4b2
JK		 2530
2531 segno = GET_SEGNO(sbi, new_blkaddr);
2532 se = get_seg_entry(sbi, segno);
2533 type = se->type;
2534
19f106bc
CY		 2535 if (!recover_curseg) {
2536 /* for recovery flow */
2537 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2538 if (old_blkaddr == NULL_ADDR)
2539 type = CURSEG_COLD_DATA;
2540 else
2541 type = CURSEG_WARM_DATA;
2542 }
2543 } else {
2544 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 2545 type = CURSEG_WARM_DATA;
2546 }
19f106bc 2547
351df4b2
JK		 2548 curseg = CURSEG_I(sbi, type);
2549
2550 mutex_lock(&curseg->curseg_mutex);
2551 mutex_lock(&sit_i->sentry_lock);
2552
2553 old_cursegno = curseg->segno;
19f106bc 2554 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2555
2556 /* change the current segment */
2557 if (segno != curseg->segno) {
2558 curseg->next_segno = segno;
025d63a4 2559 change_curseg(sbi, type);
351df4b2
JK		 2560 }
2561
491c0854 2562 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2563 __add_sum_entry(sbi, type, sum);
351df4b2 2564
28bc106b 2565 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2566 update_sit_entry(sbi, new_blkaddr, 1);
2567 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2568 update_sit_entry(sbi, old_blkaddr, -1);
2569
2570 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2571 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2572
351df4b2 2573 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2574
19f106bc
CY		 2575 if (recover_curseg) {
2576 if (old_cursegno != curseg->segno) {
2577 curseg->next_segno = old_cursegno;
025d63a4 2578 change_curseg(sbi, type);
19f106bc
CY		 2579 }
2580 curseg->next_blkoff = old_blkoff;
2581 }
2582
351df4b2
JK		 2583 mutex_unlock(&sit_i->sentry_lock);
2584 mutex_unlock(&curseg->curseg_mutex);
2585}
2586
528e3459
CY		 2587void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2588 block_t old_addr, block_t new_addr,
28bc106b
CY		 2589 unsigned char version, bool recover_curseg,
2590 bool recover_newaddr)
528e3459
CY		 2591{
2592 struct f2fs_summary sum;
2593
2594 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2595
28bc106b
CY		 2596 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2597 recover_curseg, recover_newaddr);
528e3459 2598
f28b3434 2599 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2600}
2601
93dfe2ac 2602void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2603 enum page_type type, bool ordered)
93dfe2ac 2604{
93dfe2ac 2605 if (PageWriteback(page)) {
4081363f
JK		 2606 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2607
b9109b0e
JK		 2608 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2609 0, page->index, type);
fec1d657
JK		 2610 if (ordered)
2611 wait_on_page_writeback(page);
2612 else
2613 wait_for_stable_page(page);
93dfe2ac
JK		 2614 }
2615}
2616
d4c759ee 2617void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 2618{
2619 struct page *cpage;
2620
5d4c0af4 2621 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2622 return;
2623
08b39fbd
CY		 2624 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2625 if (cpage) {
fec1d657 2626 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2627 f2fs_put_page(cpage, 1);
2628 }
2629}
2630
351df4b2
JK		 2631static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2632{
2633 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2634 struct curseg_info *seg_i;
2635 unsigned char *kaddr;
2636 struct page *page;
2637 block_t start;
2638 int i, j, offset;
2639
2640 start = start_sum_block(sbi);
2641
2642 page = get_meta_page(sbi, start++);
2643 kaddr = (unsigned char *)page_address(page);
2644
2645 /* Step 1: restore nat cache */
2646 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2647 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2648
2649 /* Step 2: restore sit cache */
2650 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2651 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2652 offset = 2 * SUM_JOURNAL_SIZE;
2653
2654 /* Step 3: restore summary entries */
2655 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2656 unsigned short blk_off;
2657 unsigned int segno;
2658
2659 seg_i = CURSEG_I(sbi, i);
2660 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2661 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2662 seg_i->next_segno = segno;
2663 reset_curseg(sbi, i, 0);
2664 seg_i->alloc_type = ckpt->alloc_type[i];
2665 seg_i->next_blkoff = blk_off;
2666
2667 if (seg_i->alloc_type == SSR)
2668 blk_off = sbi->blocks_per_seg;
2669
2670 for (j = 0; j < blk_off; j++) {
2671 struct f2fs_summary *s;
2672 s = (struct f2fs_summary *)(kaddr + offset);
2673 seg_i->sum_blk->entries[j] = *s;
2674 offset += SUMMARY_SIZE;
09cbfeaf 2675 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2676 SUM_FOOTER_SIZE)
2677 continue;
2678
2679 f2fs_put_page(page, 1);
2680 page = NULL;
2681
2682 page = get_meta_page(sbi, start++);
2683 kaddr = (unsigned char *)page_address(page);
2684 offset = 0;
2685 }
2686 }
2687 f2fs_put_page(page, 1);
2688 return 0;
2689}
2690
2691static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2692{
2693 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2694 struct f2fs_summary_block *sum;
2695 struct curseg_info *curseg;
2696 struct page *new;
2697 unsigned short blk_off;
2698 unsigned int segno = 0;
2699 block_t blk_addr = 0;
2700
2701 /* get segment number and block addr */
2702 if (IS_DATASEG(type)) {
2703 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2704 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2705 CURSEG_HOT_DATA]);
119ee914 2706 if (__exist_node_summaries(sbi))
351df4b2
JK		 2707 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2708 else
2709 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2710 } else {
2711 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2712 CURSEG_HOT_NODE]);
2713 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2714 CURSEG_HOT_NODE]);
119ee914 2715 if (__exist_node_summaries(sbi))
351df4b2
JK		 2716 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2717 type - CURSEG_HOT_NODE);
2718 else
2719 blk_addr = GET_SUM_BLOCK(sbi, segno);
2720 }
2721
2722 new = get_meta_page(sbi, blk_addr);
2723 sum = (struct f2fs_summary_block *)page_address(new);
2724
2725 if (IS_NODESEG(type)) {
119ee914 2726 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2727 struct f2fs_summary *ns = &sum->entries[0];
2728 int i;
2729 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2730 ns->version = 0;
2731 ns->ofs_in_node = 0;
2732 }
2733 } else {
d653788a
GZ		 2734 int err;
2735
2736 err = restore_node_summary(sbi, segno, sum);
2737 if (err) {
351df4b2 2738 f2fs_put_page(new, 1);
d653788a 2739 return err;
351df4b2
JK		 2740 }
2741 }
2742 }
2743
2744 /* set uncompleted segment to curseg */
2745 curseg = CURSEG_I(sbi, type);
2746 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2747
2748 /* update journal info */
2749 down_write(&curseg->journal_rwsem);
2750 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2751 up_write(&curseg->journal_rwsem);
2752
2753 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2754 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2755 curseg->next_segno = segno;
2756 reset_curseg(sbi, type, 0);
2757 curseg->alloc_type = ckpt->alloc_type[type];
2758 curseg->next_blkoff = blk_off;
2759 mutex_unlock(&curseg->curseg_mutex);
2760 f2fs_put_page(new, 1);
2761 return 0;
2762}
2763
2764static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2765{
21d3f8e1
JQ		 2766 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2767 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 2768 int type = CURSEG_HOT_DATA;
e4fc5fbf 2769 int err;
351df4b2 2770
aaec2b1d 2771 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2772 int npages = npages_for_summary_flush(sbi, true);
2773
2774 if (npages >= 2)
2775 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2776 META_CP, true);
3fa06d7b 2777
351df4b2
JK		 2778 /* restore for compacted data summary */
2779 if (read_compacted_summaries(sbi))
2780 return -EINVAL;
2781 type = CURSEG_HOT_NODE;
2782 }
2783
119ee914 2784 if (__exist_node_summaries(sbi))
3fa06d7b 2785 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2786 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2787
e4fc5fbf
CY		 2788 for (; type <= CURSEG_COLD_NODE; type++) {
2789 err = read_normal_summaries(sbi, type);
2790 if (err)
2791 return err;
2792 }
2793
21d3f8e1
JQ		 2794 /* sanity check for summary blocks */
2795 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2796 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2797 return -EINVAL;
2798
351df4b2
JK		 2799 return 0;
2800}
2801
2802static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2803{
2804 struct page *page;
2805 unsigned char *kaddr;
2806 struct f2fs_summary *summary;
2807 struct curseg_info *seg_i;
2808 int written_size = 0;
2809 int i, j;
2810
2811 page = grab_meta_page(sbi, blkaddr++);
2812 kaddr = (unsigned char *)page_address(page);
2813
2814 /* Step 1: write nat cache */
2815 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2816 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2817 written_size += SUM_JOURNAL_SIZE;
2818
2819 /* Step 2: write sit cache */
2820 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2821 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2822 written_size += SUM_JOURNAL_SIZE;
2823
351df4b2
JK		 2824 /* Step 3: write summary entries */
2825 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2826 unsigned short blkoff;
2827 seg_i = CURSEG_I(sbi, i);
2828 if (sbi->ckpt->alloc_type[i] == SSR)
2829 blkoff = sbi->blocks_per_seg;
2830 else
2831 blkoff = curseg_blkoff(sbi, i);
2832
2833 for (j = 0; j < blkoff; j++) {
2834 if (!page) {
2835 page = grab_meta_page(sbi, blkaddr++);
2836 kaddr = (unsigned char *)page_address(page);
2837 written_size = 0;
2838 }
2839 summary = (struct f2fs_summary *)(kaddr + written_size);
2840 *summary = seg_i->sum_blk->entries[j];
2841 written_size += SUMMARY_SIZE;
351df4b2 2842
09cbfeaf 2843 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2844 SUM_FOOTER_SIZE)
2845 continue;
2846
e8d61a74 2847 set_page_dirty(page);
351df4b2
JK		 2848 f2fs_put_page(page, 1);
2849 page = NULL;
2850 }
2851 }
e8d61a74
CY		 2852 if (page) {
2853 set_page_dirty(page);
351df4b2 2854 f2fs_put_page(page, 1);
e8d61a74 2855 }
351df4b2
JK		 2856}
2857
2858static void write_normal_summaries(struct f2fs_sb_info *sbi,
2859 block_t blkaddr, int type)
2860{
2861 int i, end;
2862 if (IS_DATASEG(type))
2863 end = type + NR_CURSEG_DATA_TYPE;
2864 else
2865 end = type + NR_CURSEG_NODE_TYPE;
2866
b7ad7512
CY		 2867 for (i = type; i < end; i++)
2868 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 2869}
2870
2871void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2872{
aaec2b1d 2873 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 2874 write_compacted_summaries(sbi, start_blk);
2875 else
2876 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2877}
2878
2879void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2880{
119ee914 2881 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 2882}
2883
dfc08a12 2884int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 2885 unsigned int val, int alloc)
2886{
2887 int i;
2888
2889 if (type == NAT_JOURNAL) {
dfc08a12
CY		 2890 for (i = 0; i < nats_in_cursum(journal); i++) {
2891 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 2892 return i;
2893 }
dfc08a12
CY		 2894 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2895 return update_nats_in_cursum(journal, 1);
351df4b2 2896 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 2897 for (i = 0; i < sits_in_cursum(journal); i++)
2898 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 2899 return i;
dfc08a12
CY		 2900 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2901 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 2902 }
2903 return -1;
2904}
2905
2906static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2907 unsigned int segno)
2908{
2cc22186 2909 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 2910}
2911
2912static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2913 unsigned int start)
2914{
2915 struct sit_info *sit_i = SIT_I(sbi);
2916 struct page *src_page, *dst_page;
2917 pgoff_t src_off, dst_off;
2918 void *src_addr, *dst_addr;
2919
2920 src_off = current_sit_addr(sbi, start);
2921 dst_off = next_sit_addr(sbi, src_off);
2922
2923 /* get current sit block page without lock */
2924 src_page = get_meta_page(sbi, src_off);
2925 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 2926 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 2927
2928 src_addr = page_address(src_page);
2929 dst_addr = page_address(dst_page);
09cbfeaf 2930 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 2931
2932 set_page_dirty(dst_page);
2933 f2fs_put_page(src_page, 1);
2934
2935 set_to_next_sit(sit_i, start);
2936
2937 return dst_page;
2938}
2939
184a5cd2
CY		 2940static struct sit_entry_set *grab_sit_entry_set(void)
2941{
2942 struct sit_entry_set *ses =
80c54505 2943 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 2944
2945 ses->entry_cnt = 0;
2946 INIT_LIST_HEAD(&ses->set_list);
2947 return ses;
2948}
2949
2950static void release_sit_entry_set(struct sit_entry_set *ses)
2951{
2952 list_del(&ses->set_list);
2953 kmem_cache_free(sit_entry_set_slab, ses);
2954}
2955
2956static void adjust_sit_entry_set(struct sit_entry_set *ses,
2957 struct list_head *head)
2958{
2959 struct sit_entry_set *next = ses;
2960
2961 if (list_is_last(&ses->set_list, head))
2962 return;
2963
2964 list_for_each_entry_continue(next, head, set_list)
2965 if (ses->entry_cnt <= next->entry_cnt)
2966 break;
2967
2968 list_move_tail(&ses->set_list, &next->set_list);
2969}
2970
2971static void add_sit_entry(unsigned int segno, struct list_head *head)
2972{
2973 struct sit_entry_set *ses;
2974 unsigned int start_segno = START_SEGNO(segno);
2975
2976 list_for_each_entry(ses, head, set_list) {
2977 if (ses->start_segno == start_segno) {
2978 ses->entry_cnt++;
2979 adjust_sit_entry_set(ses, head);
2980 return;
2981 }
2982 }
2983
2984 ses = grab_sit_entry_set();
2985
2986 ses->start_segno = start_segno;
2987 ses->entry_cnt++;
2988 list_add(&ses->set_list, head);
2989}
2990
2991static void add_sits_in_set(struct f2fs_sb_info *sbi)
2992{
2993 struct f2fs_sm_info *sm_info = SM_I(sbi);
2994 struct list_head *set_list = &sm_info->sit_entry_set;
2995 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 2996 unsigned int segno;
2997
7cd8558b 2998 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 2999 add_sit_entry(segno, set_list);
3000}
3001
3002static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3003{
3004 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3005 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3006 int i;
3007
b7ad7512 3008 down_write(&curseg->journal_rwsem);
dfc08a12 3009 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3010 unsigned int segno;
3011 bool dirtied;
3012
dfc08a12 3013 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3014 dirtied = __mark_sit_entry_dirty(sbi, segno);
3015
3016 if (!dirtied)
3017 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3018 }
dfc08a12 3019 update_sits_in_cursum(journal, -i);
b7ad7512 3020 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3021}
3022
0a8165d7 3023/*
351df4b2
JK		 3024 * CP calls this function, which flushes SIT entries including sit_journal,
3025 * and moves prefree segs to free segs.
3026 */
4b2fecc8 3027void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3028{
3029 struct sit_info *sit_i = SIT_I(sbi);
3030 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3031 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3032 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3033 struct sit_entry_set *ses, *tmp;
3034 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3035 bool to_journal = true;
4b2fecc8 3036 struct seg_entry *se;
351df4b2 3037
351df4b2
JK		 3038 mutex_lock(&sit_i->sentry_lock);
3039
2b11a74b
WL		 3040 if (!sit_i->dirty_sentries)
3041 goto out;
3042
351df4b2 3043 /*
184a5cd2
CY		 3044 * add and account sit entries of dirty bitmap in sit entry
3045 * set temporarily
351df4b2 3046 */
184a5cd2 3047 add_sits_in_set(sbi);
351df4b2 3048
184a5cd2
CY		 3049 /*
3050 * if there are no enough space in journal to store dirty sit
3051 * entries, remove all entries from journal and add and account
3052 * them in sit entry set.
3053 */
dfc08a12 3054 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3055 remove_sits_in_journal(sbi);
b2955550 3056
184a5cd2
CY		 3057 /*
3058 * there are two steps to flush sit entries:
3059 * #1, flush sit entries to journal in current cold data summary block.
3060 * #2, flush sit entries to sit page.
3061 */
3062 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3063 struct page *page = NULL;
184a5cd2
CY		 3064 struct f2fs_sit_block *raw_sit = NULL;
3065 unsigned int start_segno = ses->start_segno;
3066 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3067 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3068 unsigned int segno = start_segno;
3069
3070 if (to_journal &&
dfc08a12 3071 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3072 to_journal = false;
3073
b7ad7512
CY		 3074 if (to_journal) {
3075 down_write(&curseg->journal_rwsem);
3076 } else {
184a5cd2
CY		 3077 page = get_next_sit_page(sbi, start_segno);
3078 raw_sit = page_address(page);
351df4b2 3079 }
351df4b2 3080
184a5cd2
CY		 3081 /* flush dirty sit entries in region of current sit set */
3082 for_each_set_bit_from(segno, bitmap, end) {
3083 int offset, sit_offset;
4b2fecc8
JK		 3084
3085 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 3086
3087 /* add discard candidates */
c473f1a9 3088 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3089 cpc->trim_start = segno;
25290fa5 3090 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3091 }
184a5cd2
CY		 3092
3093 if (to_journal) {
dfc08a12 3094 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 3095 SIT_JOURNAL, segno, 1);
3096 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3097 segno_in_journal(journal, offset) =
184a5cd2
CY		 3098 cpu_to_le32(segno);
3099 seg_info_to_raw_sit(se,
dfc08a12 3100 &sit_in_journal(journal, offset));
184a5cd2
CY		 3101 } else {
3102 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3103 seg_info_to_raw_sit(se,
3104 &raw_sit->entries[sit_offset]);
3105 }
351df4b2 3106
184a5cd2
CY		 3107 __clear_bit(segno, bitmap);
3108 sit_i->dirty_sentries--;
3109 ses->entry_cnt--;
351df4b2
JK		 3110 }
3111
b7ad7512
CY		 3112 if (to_journal)
3113 up_write(&curseg->journal_rwsem);
3114 else
184a5cd2
CY		 3115 f2fs_put_page(page, 1);
3116
3117 f2fs_bug_on(sbi, ses->entry_cnt);
3118 release_sit_entry_set(ses);
351df4b2 3119 }
184a5cd2
CY		 3120
3121 f2fs_bug_on(sbi, !list_empty(head));
3122 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3123out:
c473f1a9 3124 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3125 __u64 trim_start = cpc->trim_start;
3126
4b2fecc8 3127 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3128 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3129
3130 cpc->trim_start = trim_start;
4b2fecc8 3131 }
351df4b2 3132 mutex_unlock(&sit_i->sentry_lock);
351df4b2 3133
351df4b2
JK		 3134 set_prefree_as_free_segments(sbi);
3135}
3136
3137static int build_sit_info(struct f2fs_sb_info *sbi)
3138{
3139 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3140 struct sit_info *sit_i;
3141 unsigned int sit_segs, start;
ae27d62e 3142 char *src_bitmap;
351df4b2
JK		 3143 unsigned int bitmap_size;
3144
3145 /* allocate memory for SIT information */
3146 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3147 if (!sit_i)
3148 return -ENOMEM;
3149
3150 SM_I(sbi)->sit_info = sit_i;
3151
a7c3e901 3152 sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
39307a8e 3153 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3154 if (!sit_i->sentries)
3155 return -ENOMEM;
3156
7cd8558b 3157 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3158 sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3159 if (!sit_i->dirty_sentries_bitmap)
3160 return -ENOMEM;
3161
7cd8558b 3162 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3163 sit_i->sentries[start].cur_valid_map
3164 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3165 sit_i->sentries[start].ckpt_valid_map
3166 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3167 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3168 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3169 return -ENOMEM;
3e025740 3170
355e7891
CY		 3171#ifdef CONFIG_F2FS_CHECK_FS
3172 sit_i->sentries[start].cur_valid_map_mir
3173 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3174 if (!sit_i->sentries[start].cur_valid_map_mir)
3175 return -ENOMEM;
3176#endif
3177
3e025740
JK		 3178 if (f2fs_discard_en(sbi)) {
3179 sit_i->sentries[start].discard_map
3180 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3181 if (!sit_i->sentries[start].discard_map)
3182 return -ENOMEM;
3183 }
351df4b2
JK		 3184 }
3185
60a3b782
JK		 3186 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3187 if (!sit_i->tmp_map)
3188 return -ENOMEM;
3189
351df4b2 3190 if (sbi->segs_per_sec > 1) {
a7c3e901 3191 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
39307a8e 3192 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3193 if (!sit_i->sec_entries)
3194 return -ENOMEM;
3195 }
3196
3197 /* get information related with SIT */
3198 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3199
3200 /* setup SIT bitmap from ckeckpoint pack */
3201 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3202 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3203
ae27d62e
CY		 3204 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3205 if (!sit_i->sit_bitmap)
351df4b2 3206 return -ENOMEM;
351df4b2 3207
ae27d62e
CY		 3208#ifdef CONFIG_F2FS_CHECK_FS
3209 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3210 if (!sit_i->sit_bitmap_mir)
3211 return -ENOMEM;
3212#endif
3213
351df4b2
JK		 3214 /* init SIT information */
3215 sit_i->s_ops = &default_salloc_ops;
3216
3217 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3218 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3219 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3220 sit_i->bitmap_size = bitmap_size;
3221 sit_i->dirty_sentries = 0;
3222 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3223 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3224 sit_i->mounted_time = ktime_get_real_seconds();
351df4b2
JK		 3225 mutex_init(&sit_i->sentry_lock);
3226 return 0;
3227}
3228
3229static int build_free_segmap(struct f2fs_sb_info *sbi)
3230{
351df4b2
JK		 3231 struct free_segmap_info *free_i;
3232 unsigned int bitmap_size, sec_bitmap_size;
3233
3234 /* allocate memory for free segmap information */
3235 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3236 if (!free_i)
3237 return -ENOMEM;
3238
3239 SM_I(sbi)->free_info = free_i;
3240
7cd8558b 3241 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3242 free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3243 if (!free_i->free_segmap)
3244 return -ENOMEM;
3245
7cd8558b 3246 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
a7c3e901 3247 free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3248 if (!free_i->free_secmap)
3249 return -ENOMEM;
3250
3251 /* set all segments as dirty temporarily */
3252 memset(free_i->free_segmap, 0xff, bitmap_size);
3253 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3254
3255 /* init free segmap information */
7cd8558b 3256 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3257 free_i->free_segments = 0;
3258 free_i->free_sections = 0;
1a118ccf 3259 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3260 return 0;
3261}
3262
3263static int build_curseg(struct f2fs_sb_info *sbi)
3264{
1042d60f 3265 struct curseg_info *array;
351df4b2
JK		 3266 int i;
3267
b434babf 3268 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 3269 if (!array)
3270 return -ENOMEM;
3271
3272 SM_I(sbi)->curseg_array = array;
3273
3274 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3275 mutex_init(&array[i].curseg_mutex);
09cbfeaf 3276 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3277 if (!array[i].sum_blk)
3278 return -ENOMEM;
b7ad7512
CY		 3279 init_rwsem(&array[i].journal_rwsem);
3280 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3281 GFP_KERNEL);
3282 if (!array[i].journal)
3283 return -ENOMEM;
351df4b2
JK		 3284 array[i].segno = NULL_SEGNO;
3285 array[i].next_blkoff = 0;
3286 }
3287 return restore_curseg_summaries(sbi);
3288}
3289
3290static void build_sit_entries(struct f2fs_sb_info *sbi)
3291{
3292 struct sit_info *sit_i = SIT_I(sbi);
3293 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3294 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3295 struct seg_entry *se;
3296 struct f2fs_sit_entry sit;
74de593a
CY		 3297 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3298 unsigned int i, start, end;
3299 unsigned int readed, start_blk = 0;
351df4b2 3300
74de593a 3301 do {
664ba972
JK		 3302 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3303 META_SIT, true);
74de593a
CY		 3304
3305 start = start_blk * sit_i->sents_per_block;
3306 end = (start_blk + readed) * sit_i->sents_per_block;
3307
7cd8558b 3308 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3309 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3310 struct page *page;
3311
9c094040 3312 se = &sit_i->sentries[start];
74de593a
CY		 3313 page = get_current_sit_page(sbi, start);
3314 sit_blk = (struct f2fs_sit_block *)page_address(page);
3315 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3316 f2fs_put_page(page, 1);
d600af23 3317
74de593a
CY		 3318 check_block_count(sbi, start, &sit);
3319 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 3320
3321 /* build discard map only one time */
3e025740 3322 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3323 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3324 memset(se->discard_map, 0xff,
3325 SIT_VBLOCK_MAP_SIZE);
3326 } else {
3327 memcpy(se->discard_map,
3328 se->cur_valid_map,
3329 SIT_VBLOCK_MAP_SIZE);
3330 sbi->discard_blks +=
3331 sbi->blocks_per_seg -
3332 se->valid_blocks;
3333 }
3e025740 3334 }
a66cdd98 3335
d600af23
CY		 3336 if (sbi->segs_per_sec > 1)
3337 get_sec_entry(sbi, start)->valid_blocks +=
3338 se->valid_blocks;
351df4b2 3339 }
74de593a
CY		 3340 start_blk += readed;
3341 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3342
3343 down_read(&curseg->journal_rwsem);
3344 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3345 unsigned int old_valid_blocks;
3346
3347 start = le32_to_cpu(segno_in_journal(journal, i));
3348 se = &sit_i->sentries[start];
3349 sit = sit_in_journal(journal, i);
3350
3351 old_valid_blocks = se->valid_blocks;
3352
3353 check_block_count(sbi, start, &sit);
3354 seg_info_from_raw_sit(se, &sit);
3355
3356 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3357 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3358 memset(se->discard_map, 0xff,
3359 SIT_VBLOCK_MAP_SIZE);
3360 } else {
3361 memcpy(se->discard_map, se->cur_valid_map,
3362 SIT_VBLOCK_MAP_SIZE);
3363 sbi->discard_blks += old_valid_blocks -
3364 se->valid_blocks;
3365 }
d600af23
CY		 3366 }
3367
3368 if (sbi->segs_per_sec > 1)
3369 get_sec_entry(sbi, start)->valid_blocks +=
3370 se->valid_blocks - old_valid_blocks;
3371 }
3372 up_read(&curseg->journal_rwsem);
351df4b2
JK		 3373}
3374
3375static void init_free_segmap(struct f2fs_sb_info *sbi)
3376{
3377 unsigned int start;
3378 int type;
3379
7cd8558b 3380 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3381 struct seg_entry *sentry = get_seg_entry(sbi, start);
3382 if (!sentry->valid_blocks)
3383 __set_free(sbi, start);
c79b7ff1
JK		 3384 else
3385 SIT_I(sbi)->written_valid_blocks +=
3386 sentry->valid_blocks;
351df4b2
JK		 3387 }
3388
3389 /* set use the current segments */
3390 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3391 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3392 __set_test_and_inuse(sbi, curseg_t->segno);
3393 }
3394}
3395
3396static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3397{
3398 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3399 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3400 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3401 unsigned short valid_blocks;
3402
8736fbf0 3403 while (1) {
351df4b2 3404 /* find dirty segment based on free segmap */
7cd8558b
JK		 3405 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3406 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3407 break;
3408 offset = segno + 1;
302bd348 3409 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3410 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3411 continue;
ec325b52
JK		 3412 if (valid_blocks > sbi->blocks_per_seg) {
3413 f2fs_bug_on(sbi, 1);
3414 continue;
3415 }
351df4b2
JK		 3416 mutex_lock(&dirty_i->seglist_lock);
3417 __locate_dirty_segment(sbi, segno, DIRTY);
3418 mutex_unlock(&dirty_i->seglist_lock);
3419 }
3420}
3421
5ec4e49f 3422static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3423{
3424 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3425 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3426
a7c3e901 3427 dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 3428 if (!dirty_i->victim_secmap)
351df4b2
JK		 3429 return -ENOMEM;
3430 return 0;
3431}
3432
3433static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3434{
3435 struct dirty_seglist_info *dirty_i;
3436 unsigned int bitmap_size, i;
3437
3438 /* allocate memory for dirty segments list information */
3439 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3440 if (!dirty_i)
3441 return -ENOMEM;
3442
3443 SM_I(sbi)->dirty_info = dirty_i;
3444 mutex_init(&dirty_i->seglist_lock);
3445
7cd8558b 3446 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3447
3448 for (i = 0; i < NR_DIRTY_TYPE; i++) {
a7c3e901 3449 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3450 if (!dirty_i->dirty_segmap[i])
3451 return -ENOMEM;
3452 }
3453
3454 init_dirty_segmap(sbi);
5ec4e49f 3455 return init_victim_secmap(sbi);
351df4b2
JK		 3456}
3457
0a8165d7 3458/*
351df4b2
JK		 3459 * Update min, max modified time for cost-benefit GC algorithm
3460 */
3461static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3462{
3463 struct sit_info *sit_i = SIT_I(sbi);
3464 unsigned int segno;
3465
3466 mutex_lock(&sit_i->sentry_lock);
3467
3468 sit_i->min_mtime = LLONG_MAX;
3469
7cd8558b 3470 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3471 unsigned int i;
3472 unsigned long long mtime = 0;
3473
3474 for (i = 0; i < sbi->segs_per_sec; i++)
3475 mtime += get_seg_entry(sbi, segno + i)->mtime;
3476
3477 mtime = div_u64(mtime, sbi->segs_per_sec);
3478
3479 if (sit_i->min_mtime > mtime)
3480 sit_i->min_mtime = mtime;
3481 }
3482 sit_i->max_mtime = get_mtime(sbi);
3483 mutex_unlock(&sit_i->sentry_lock);
3484}
3485
3486int build_segment_manager(struct f2fs_sb_info *sbi)
3487{
3488 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3489 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3490 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3491 int err;
3492
3493 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3494 if (!sm_info)
3495 return -ENOMEM;
3496
3497 /* init sm info */
3498 sbi->sm_info = sm_info;
351df4b2
JK		 3499 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3500 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3501 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3502 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3503 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3504 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3505 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3506 sm_info->rec_prefree_segments = sm_info->main_segments *
3507 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3508 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3509 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3510
52763a4b
JK		 3511 if (!test_opt(sbi, LFS))
3512 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3513 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3514 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3515 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
351df4b2 3516
bba681cb
JK		 3517 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3518
184a5cd2
CY		 3519 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3520
d4fdf8ba 3521 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3522 err = create_flush_cmd_control(sbi);
3523 if (err)
a688b9d9 3524 return err;
6b4afdd7
JK		 3525 }
3526
0b54fb84
JK		 3527 err = create_discard_cmd_control(sbi);
3528 if (err)
3529 return err;
3530
351df4b2
JK		 3531 err = build_sit_info(sbi);
3532 if (err)
3533 return err;
3534 err = build_free_segmap(sbi);
3535 if (err)
3536 return err;
3537 err = build_curseg(sbi);
3538 if (err)
3539 return err;
3540
3541 /* reinit free segmap based on SIT */
3542 build_sit_entries(sbi);
3543
3544 init_free_segmap(sbi);
3545 err = build_dirty_segmap(sbi);
3546 if (err)
3547 return err;
3548
3549 init_min_max_mtime(sbi);
3550 return 0;
3551}
3552
3553static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3554 enum dirty_type dirty_type)
3555{
3556 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3557
3558 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3559 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3560 dirty_i->nr_dirty[dirty_type] = 0;
3561 mutex_unlock(&dirty_i->seglist_lock);
3562}
3563
5ec4e49f 3564static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3565{
3566 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3567 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3568}
3569
3570static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3571{
3572 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3573 int i;
3574
3575 if (!dirty_i)
3576 return;
3577
3578 /* discard pre-free/dirty segments list */
3579 for (i = 0; i < NR_DIRTY_TYPE; i++)
3580 discard_dirty_segmap(sbi, i);
3581
5ec4e49f 3582 destroy_victim_secmap(sbi);
351df4b2
JK		 3583 SM_I(sbi)->dirty_info = NULL;
3584 kfree(dirty_i);
3585}
3586
3587static void destroy_curseg(struct f2fs_sb_info *sbi)
3588{
3589 struct curseg_info *array = SM_I(sbi)->curseg_array;
3590 int i;
3591
3592 if (!array)
3593 return;
3594 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3595 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3596 kfree(array[i].sum_blk);
b7ad7512
CY		 3597 kfree(array[i].journal);
3598 }
351df4b2
JK		 3599 kfree(array);
3600}
3601
3602static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3603{
3604 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3605 if (!free_i)
3606 return;
3607 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3608 kvfree(free_i->free_segmap);
3609 kvfree(free_i->free_secmap);
351df4b2
JK		 3610 kfree(free_i);
3611}
3612
3613static void destroy_sit_info(struct f2fs_sb_info *sbi)
3614{
3615 struct sit_info *sit_i = SIT_I(sbi);
3616 unsigned int start;
3617
3618 if (!sit_i)
3619 return;
3620
3621 if (sit_i->sentries) {
7cd8558b 3622 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3623 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 3624#ifdef CONFIG_F2FS_CHECK_FS
3625 kfree(sit_i->sentries[start].cur_valid_map_mir);
3626#endif
351df4b2 3627 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3628 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3629 }
3630 }
60a3b782
JK		 3631 kfree(sit_i->tmp_map);
3632
39307a8e
JK		 3633 kvfree(sit_i->sentries);
3634 kvfree(sit_i->sec_entries);
3635 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3636
3637 SM_I(sbi)->sit_info = NULL;
3638 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 3639#ifdef CONFIG_F2FS_CHECK_FS
3640 kfree(sit_i->sit_bitmap_mir);
3641#endif
351df4b2
JK		 3642 kfree(sit_i);
3643}
3644
3645void destroy_segment_manager(struct f2fs_sb_info *sbi)
3646{
3647 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3648
3b03f724
CY		 3649 if (!sm_info)
3650 return;
5eba8c5d 3651 destroy_flush_cmd_control(sbi, true);
f099405f 3652 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3653 destroy_dirty_segmap(sbi);
3654 destroy_curseg(sbi);
3655 destroy_free_segmap(sbi);
3656 destroy_sit_info(sbi);
3657 sbi->sm_info = NULL;
3658 kfree(sm_info);
3659}
7fd9e544
JK		 3660
3661int __init create_segment_manager_caches(void)
3662{
3663 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3664 sizeof(struct discard_entry));
7fd9e544 3665 if (!discard_entry_slab)
184a5cd2
CY		 3666 goto fail;
3667
b01a9201
JK		 3668 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3669 sizeof(struct discard_cmd));
3670 if (!discard_cmd_slab)
6ab2a308 3671 goto destroy_discard_entry;
275b66b0 3672
184a5cd2 3673 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3674 sizeof(struct sit_entry_set));
184a5cd2 3675 if (!sit_entry_set_slab)
b01a9201 3676 goto destroy_discard_cmd;
88b88a66
JK		 3677
3678 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3679 sizeof(struct inmem_pages));
3680 if (!inmem_entry_slab)
3681 goto destroy_sit_entry_set;
7fd9e544 3682 return 0;
184a5cd2 3683
88b88a66
JK		 3684destroy_sit_entry_set:
3685 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 3686destroy_discard_cmd:
3687 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3688destroy_discard_entry:
184a5cd2
CY		 3689 kmem_cache_destroy(discard_entry_slab);
3690fail:
3691 return -ENOMEM;
7fd9e544
JK		 3692}
3693
3694void destroy_segment_manager_caches(void)
3695{
184a5cd2 3696 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3697 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3698 kmem_cache_destroy(discard_entry_slab);
88b88a66 3699 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3700}
Ubuntu Jammy Linux kernel mirror