projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: remove the redundant variable definition
[linux-2.6-block.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
351df4b2
JK		 19
20#include "f2fs.h"
21#include "segment.h"
22#include "node.h"
9e4ded3f 23#include "trace.h"
6ec178da 24#include <trace/events/f2fs.h>
351df4b2 25
9a7f143a
CL		 26#define __reverse_ffz(x) __reverse_ffs(~(x))
27
7fd9e544 28static struct kmem_cache *discard_entry_slab;
b01a9201 29static struct kmem_cache *discard_cmd_slab;
184a5cd2 30static struct kmem_cache *sit_entry_set_slab;
88b88a66 31static struct kmem_cache *inmem_entry_slab;
7fd9e544 32
f96999c3
JK		 33static unsigned long __reverse_ulong(unsigned char *str)
34{
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
37
38#if BITS_PER_LONG == 64
39 shift = 56;
40#endif
41 while (shift >= 0) {
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
44 }
45 return tmp;
46}
47
9a7f143a
CL		 48/*
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 */
52static inline unsigned long __reverse_ffs(unsigned long word)
53{
54 int num = 0;
55
56#if BITS_PER_LONG == 64
f96999c3 57 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 58 num += 32;
f96999c3 59 else
9a7f143a 60 word >>= 32;
9a7f143a 61#endif
f96999c3 62 if ((word & 0xffff0000) == 0)
9a7f143a 63 num += 16;
f96999c3 64 else
9a7f143a 65 word >>= 16;
f96999c3
JK		 66
67 if ((word & 0xff00) == 0)
9a7f143a 68 num += 8;
f96999c3 69 else
9a7f143a 70 word >>= 8;
f96999c3 71
9a7f143a
CL		 72 if ((word & 0xf0) == 0)
73 num += 4;
74 else
75 word >>= 4;
f96999c3 76
9a7f143a
CL		 77 if ((word & 0xc) == 0)
78 num += 2;
79 else
80 word >>= 2;
f96999c3 81
9a7f143a
CL		 82 if ((word & 0x2) == 0)
83 num += 1;
84 return num;
85}
86
87/*
e1c42045 88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 90 * @size must be integral times of unsigned long.
9a7f143a 91 * Example:
f96999c3
JK		 92 * MSB <--> LSB
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 95 */
96static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
98{
99 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 100 unsigned long result = size;
9a7f143a 101 unsigned long tmp;
9a7f143a
CL		 102
103 if (offset >= size)
104 return size;
105
692223d1 106 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 107 offset %= BITS_PER_LONG;
f96999c3 108
692223d1
FL		 109 while (1) {
110 if (*p == 0)
111 goto pass;
9a7f143a 112
f96999c3 113 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 114
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 118 if (tmp)
692223d1
FL		 119 goto found;
120pass:
121 if (size <= BITS_PER_LONG)
122 break;
9a7f143a 123 size -= BITS_PER_LONG;
692223d1 124 offset = 0;
f96999c3 125 p++;
9a7f143a 126 }
692223d1
FL		 127 return result;
128found:
129 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 130}
131
132static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
134{
135 const unsigned long *p = addr + BIT_WORD(offset);
80609448 136 unsigned long result = size;
9a7f143a 137 unsigned long tmp;
9a7f143a
CL		 138
139 if (offset >= size)
140 return size;
141
80609448 142 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 143 offset %= BITS_PER_LONG;
80609448
JK		 144
145 while (1) {
146 if (*p == ~0UL)
147 goto pass;
148
f96999c3 149 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 150
151 if (offset)
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
154 tmp |= ~0UL >> size;
f96999c3 155 if (tmp != ~0UL)
80609448
JK		 156 goto found;
157pass:
158 if (size <= BITS_PER_LONG)
159 break;
9a7f143a 160 size -= BITS_PER_LONG;
80609448 161 offset = 0;
f96999c3 162 p++;
9a7f143a 163 }
80609448
JK		 164 return result;
165found:
166 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 167}
168
88b88a66
JK		 169void register_inmem_page(struct inode *inode, struct page *page)
170{
171 struct f2fs_inode_info *fi = F2FS_I(inode);
172 struct inmem_pages *new;
9be32d72 173
9e4ded3f 174 f2fs_trace_pid(page);
0722b101 175
decd36b6
CY		 176 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
177 SetPagePrivate(page);
178
88b88a66
JK		 179 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
180
181 /* add atomic page indices to the list */
182 new->page = page;
183 INIT_LIST_HEAD(&new->list);
decd36b6 184
88b88a66
JK		 185 /* increase reference count with clean state */
186 mutex_lock(&fi->inmem_lock);
187 get_page(page);
188 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 189 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 190 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 191
192 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 193}
194
28bc106b
CY		 195static int __revoke_inmem_pages(struct inode *inode,
196 struct list_head *head, bool drop, bool recover)
29b96b54 197{
28bc106b 198 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 199 struct inmem_pages *cur, *tmp;
28bc106b 200 int err = 0;
29b96b54
CY		 201
202 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 203 struct page *page = cur->page;
204
205 if (drop)
206 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
207
208 lock_page(page);
29b96b54 209
28bc106b
CY		 210 if (recover) {
211 struct dnode_of_data dn;
212 struct node_info ni;
213
214 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
215
216 set_new_dnode(&dn, inode, NULL, NULL, 0);
217 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
218 err = -EAGAIN;
219 goto next;
220 }
221 get_node_info(sbi, dn.nid, &ni);
222 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
223 cur->old_addr, ni.version, true, true);
224 f2fs_put_dnode(&dn);
225 }
226next:
63c52d78
JK		 227 /* we don't need to invalidate this in the sccessful status */
228 if (drop || recover)
229 ClearPageUptodate(page);
28bc106b 230 set_page_private(page, 0);
c81ced05 231 ClearPagePrivate(page);
28bc106b 232 f2fs_put_page(page, 1);
29b96b54
CY		 233
234 list_del(&cur->list);
235 kmem_cache_free(inmem_entry_slab, cur);
236 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
237 }
28bc106b 238 return err;
29b96b54
CY		 239}
240
241void drop_inmem_pages(struct inode *inode)
242{
243 struct f2fs_inode_info *fi = F2FS_I(inode);
244
245 mutex_lock(&fi->inmem_lock);
28bc106b 246 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54 247 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 248
249 clear_inode_flag(inode, FI_ATOMIC_FILE);
250 stat_dec_atomic_write(inode);
29b96b54
CY		 251}
252
8c242db9
JK		 253void drop_inmem_page(struct inode *inode, struct page *page)
254{
255 struct f2fs_inode_info *fi = F2FS_I(inode);
256 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
257 struct list_head *head = &fi->inmem_pages;
258 struct inmem_pages *cur = NULL;
259
260 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
261
262 mutex_lock(&fi->inmem_lock);
263 list_for_each_entry(cur, head, list) {
264 if (cur->page == page)
265 break;
266 }
267
268 f2fs_bug_on(sbi, !cur || cur->page != page);
269 list_del(&cur->list);
270 mutex_unlock(&fi->inmem_lock);
271
272 dec_page_count(sbi, F2FS_INMEM_PAGES);
273 kmem_cache_free(inmem_entry_slab, cur);
274
275 ClearPageUptodate(page);
276 set_page_private(page, 0);
277 ClearPagePrivate(page);
278 f2fs_put_page(page, 0);
279
280 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
281}
282
28bc106b
CY		 283static int __commit_inmem_pages(struct inode *inode,
284 struct list_head *revoke_list)
88b88a66
JK		 285{
286 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
287 struct f2fs_inode_info *fi = F2FS_I(inode);
288 struct inmem_pages *cur, *tmp;
88b88a66 289 struct f2fs_io_info fio = {
05ca3632 290 .sbi = sbi,
88b88a66 291 .type = DATA,
04d328de 292 .op = REQ_OP_WRITE,
70fd7614 293 .op_flags = REQ_SYNC | REQ_PRIO,
4375a336 294 .encrypted_page = NULL,
88b88a66 295 };
942fd319 296 pgoff_t last_idx = ULONG_MAX;
edb27dee 297 int err = 0;
88b88a66 298
88b88a66 299 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 300 struct page *page = cur->page;
301
302 lock_page(page);
303 if (page->mapping == inode->i_mapping) {
304 trace_f2fs_commit_inmem_page(page, INMEM);
305
306 set_page_dirty(page);
307 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 308 if (clear_page_dirty_for_io(page)) {
29b96b54 309 inode_dec_dirty_pages(inode);
933439c8
CY		 310 remove_dirty_inode(inode);
311 }
28bc106b
CY		 312
313 fio.page = page;
29b96b54
CY		 314 err = do_write_data_page(&fio);
315 if (err) {
28bc106b 316 unlock_page(page);
29b96b54 317 break;
70c640b1 318 }
29b96b54 319
28bc106b
CY		 320 /* record old blkaddr for revoking */
321 cur->old_addr = fio.old_blkaddr;
942fd319 322 last_idx = page->index;
28bc106b
CY		 323 }
324 unlock_page(page);
325 list_move_tail(&cur->list, revoke_list);
88b88a66 326 }
29b96b54 327
942fd319
JK		 328 if (last_idx != ULONG_MAX)
329 f2fs_submit_merged_bio_cond(sbi, inode, 0, last_idx,
330 DATA, WRITE);
28bc106b
CY		 331
332 if (!err)
333 __revoke_inmem_pages(inode, revoke_list, false, false);
334
29b96b54
CY		 335 return err;
336}
337
338int commit_inmem_pages(struct inode *inode)
339{
340 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
341 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 342 struct list_head revoke_list;
343 int err;
29b96b54 344
28bc106b 345 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 346 f2fs_balance_fs(sbi, true);
347 f2fs_lock_op(sbi);
348
5fe45743
CY		 349 set_inode_flag(inode, FI_ATOMIC_COMMIT);
350
29b96b54 351 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 352 err = __commit_inmem_pages(inode, &revoke_list);
353 if (err) {
354 int ret;
355 /*
356 * try to revoke all committed pages, but still we could fail
357 * due to no memory or other reason, if that happened, EAGAIN
358 * will be returned, which means in such case, transaction is
359 * already not integrity, caller should use journal to do the
360 * recovery or rewrite & commit last transaction. For other
361 * error number, revoking was done by filesystem itself.
362 */
363 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
364 if (ret)
365 err = ret;
366
367 /* drop all uncommitted pages */
368 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
369 }
88b88a66
JK		 370 mutex_unlock(&fi->inmem_lock);
371
5fe45743
CY		 372 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
373
29b96b54 374 f2fs_unlock_op(sbi);
edb27dee 375 return err;
88b88a66
JK		 376}
377
0a8165d7 378/*
351df4b2
JK		 379 * This function balances dirty node and dentry pages.
380 * In addition, it controls garbage collection.
381 */
2c4db1a6 382void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 383{
0f348028 384#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 385 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
386 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 387 f2fs_stop_checkpoint(sbi, false);
55523519 388 }
0f348028
CY		 389#endif
390
2c4db1a6
JK		 391 if (!need)
392 return;
e589c2c4
JK		 393
394 /* balance_fs_bg is able to be pending */
395 if (excess_cached_nats(sbi))
396 f2fs_balance_fs_bg(sbi);
397
351df4b2 398 /*
029cd28c
JK		 399 * We should do GC or end up with checkpoint, if there are so many dirty
400 * dir/node pages without enough free segments.
351df4b2 401 */
7f3037a5 402 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 403 mutex_lock(&sbi->gc_mutex);
7702bdbe 404 f2fs_gc(sbi, false, false);
351df4b2
JK		 405 }
406}
407
4660f9c0
JK		 408void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
409{
1dcc336b 410 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 411 if (!available_free_memory(sbi, EXTENT_CACHE))
412 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 413
1b38dc8e
JK		 414 /* check the # of cached NAT entries */
415 if (!available_free_memory(sbi, NAT_ENTRIES))
416 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
417
31696580 418 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 419 try_to_free_nids(sbi, MAX_FREE_NIDS);
420 else
22ad0b6a 421 build_free_nids(sbi, false, false);
31696580 422
f455c8a5
JK		 423 if (!is_idle(sbi))
424 return;
31696580 425
1b38dc8e
JK		 426 /* checkpoint is the only way to shrink partial cached entries */
427 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 428 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 429 excess_prefree_segs(sbi) ||
430 excess_dirty_nats(sbi) ||
f455c8a5 431 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 432 if (test_opt(sbi, DATA_FLUSH)) {
433 struct blk_plug plug;
434
435 blk_start_plug(&plug);
36b35a0d 436 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 437 blk_finish_plug(&plug);
438 }
4660f9c0 439 f2fs_sync_fs(sbi->sb, true);
42190d2a 440 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 441 }
4660f9c0
JK		 442}
443
20fda56b
KM		 444static int __submit_flush_wait(struct f2fs_sb_info *sbi,
445 struct block_device *bdev)
3c62be17
JK		 446{
447 struct bio *bio = f2fs_bio_alloc(0);
448 int ret;
449
09cb6464 450 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
3c62be17
JK		 451 bio->bi_bdev = bdev;
452 ret = submit_bio_wait(bio);
453 bio_put(bio);
20fda56b
KM		 454
455 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
456 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 457 return ret;
458}
459
460static int submit_flush_wait(struct f2fs_sb_info *sbi)
461{
20fda56b 462 int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev);
3c62be17
JK		 463 int i;
464
20fda56b
KM		 465 if (!sbi->s_ndevs || ret)
466 return ret;
467
468 for (i = 1; i < sbi->s_ndevs; i++) {
469 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
470 if (ret)
471 break;
3c62be17
JK		 472 }
473 return ret;
474}
475
2163d198 476static int issue_flush_thread(void *data)
6b4afdd7
JK		 477{
478 struct f2fs_sb_info *sbi = data;
b01a9201 479 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 480 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 481repeat:
482 if (kthread_should_stop())
483 return 0;
484
721bd4d5 485 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 486 struct flush_cmd *cmd, *next;
487 int ret;
488
721bd4d5
GZ		 489 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
490 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
491
3c62be17 492 ret = submit_flush_wait(sbi);
8b8dd65f
CY		 493 atomic_inc(&fcc->issued_flush);
494
721bd4d5
GZ		 495 llist_for_each_entry_safe(cmd, next,
496 fcc->dispatch_list, llnode) {
6b4afdd7 497 cmd->ret = ret;
6b4afdd7
JK		 498 complete(&cmd->wait);
499 }
a688b9d9 500 fcc->dispatch_list = NULL;
6b4afdd7
JK		 501 }
502
a688b9d9 503 wait_event_interruptible(*q,
721bd4d5 504 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 505 goto repeat;
506}
507
508int f2fs_issue_flush(struct f2fs_sb_info *sbi)
509{
b01a9201 510 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 511 struct flush_cmd cmd;
8b8dd65f 512 int ret;
6b4afdd7 513
0f7b2abd
JK		 514 if (test_opt(sbi, NOBARRIER))
515 return 0;
516
8b8dd65f
CY		 517 if (!test_opt(sbi, FLUSH_MERGE)) {
518 ret = submit_flush_wait(sbi);
519 atomic_inc(&fcc->issued_flush);
520 return ret;
521 }
740432f8 522
8b8dd65f
CY		 523 if (!atomic_read(&fcc->issing_flush)) {
524 atomic_inc(&fcc->issing_flush);
3c62be17 525 ret = submit_flush_wait(sbi);
8b8dd65f
CY		 526 atomic_dec(&fcc->issing_flush);
527
528 atomic_inc(&fcc->issued_flush);
740432f8
JK		 529 return ret;
530 }
6b4afdd7 531
adf8d90b 532 init_completion(&cmd.wait);
6b4afdd7 533
8b8dd65f 534 atomic_inc(&fcc->issing_flush);
721bd4d5 535 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 536
a688b9d9
GZ		 537 if (!fcc->dispatch_list)
538 wake_up(&fcc->flush_wait_queue);
6b4afdd7 539
5eba8c5d
JK		 540 if (fcc->f2fs_issue_flush) {
541 wait_for_completion(&cmd.wait);
8b8dd65f 542 atomic_dec(&fcc->issing_flush);
5eba8c5d
JK		 543 } else {
544 llist_del_all(&fcc->issue_list);
8b8dd65f 545 atomic_set(&fcc->issing_flush, 0);
5eba8c5d 546 }
adf8d90b
CY		 547
548 return cmd.ret;
6b4afdd7
JK		 549}
550
2163d198
GZ		 551int create_flush_cmd_control(struct f2fs_sb_info *sbi)
552{
553 dev_t dev = sbi->sb->s_bdev->bd_dev;
554 struct flush_cmd_control *fcc;
555 int err = 0;
556
b01a9201
JK		 557 if (SM_I(sbi)->fcc_info) {
558 fcc = SM_I(sbi)->fcc_info;
5eba8c5d
JK		 559 goto init_thread;
560 }
561
2163d198
GZ		 562 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
563 if (!fcc)
564 return -ENOMEM;
8b8dd65f
CY		 565 atomic_set(&fcc->issued_flush, 0);
566 atomic_set(&fcc->issing_flush, 0);
2163d198 567 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 568 init_llist_head(&fcc->issue_list);
b01a9201 569 SM_I(sbi)->fcc_info = fcc;
5eba8c5d 570init_thread:
2163d198
GZ		 571 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
572 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
573 if (IS_ERR(fcc->f2fs_issue_flush)) {
574 err = PTR_ERR(fcc->f2fs_issue_flush);
575 kfree(fcc);
b01a9201 576 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 577 return err;
578 }
2163d198
GZ		 579
580 return err;
581}
582
5eba8c5d 583void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 584{
b01a9201 585 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 586
5eba8c5d
JK		 587 if (fcc && fcc->f2fs_issue_flush) {
588 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
589
590 fcc->f2fs_issue_flush = NULL;
591 kthread_stop(flush_thread);
592 }
593 if (free) {
594 kfree(fcc);
b01a9201 595 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 596 }
2163d198
GZ		 597}
598
351df4b2
JK		 599static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
600 enum dirty_type dirty_type)
601{
602 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
603
604 /* need not be added */
605 if (IS_CURSEG(sbi, segno))
606 return;
607
608 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
609 dirty_i->nr_dirty[dirty_type]++;
610
611 if (dirty_type == DIRTY) {
612 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 613 enum dirty_type t = sentry->type;
b2f2c390 614
ec325b52
JK		 615 if (unlikely(t >= DIRTY)) {
616 f2fs_bug_on(sbi, 1);
617 return;
618 }
4625d6aa
CL		 619 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
620 dirty_i->nr_dirty[t]++;
351df4b2
JK		 621 }
622}
623
624static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
625 enum dirty_type dirty_type)
626{
627 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
628
629 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
630 dirty_i->nr_dirty[dirty_type]--;
631
632 if (dirty_type == DIRTY) {
4625d6aa
CL		 633 struct seg_entry *sentry = get_seg_entry(sbi, segno);
634 enum dirty_type t = sentry->type;
635
636 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
637 dirty_i->nr_dirty[t]--;
b2f2c390 638
5ec4e49f
JK		 639 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
640 clear_bit(GET_SECNO(sbi, segno),
641 dirty_i->victim_secmap);
351df4b2
JK		 642 }
643}
644
0a8165d7 645/*
351df4b2
JK		 646 * Should not occur error such as -ENOMEM.
647 * Adding dirty entry into seglist is not critical operation.
648 * If a given segment is one of current working segments, it won't be added.
649 */
8d8451af 650static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 651{
652 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
653 unsigned short valid_blocks;
654
655 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
656 return;
657
658 mutex_lock(&dirty_i->seglist_lock);
659
660 valid_blocks = get_valid_blocks(sbi, segno, 0);
661
662 if (valid_blocks == 0) {
663 __locate_dirty_segment(sbi, segno, PRE);
664 __remove_dirty_segment(sbi, segno, DIRTY);
665 } else if (valid_blocks < sbi->blocks_per_seg) {
666 __locate_dirty_segment(sbi, segno, DIRTY);
667 } else {
668 /* Recovery routine with SSR needs this */
669 __remove_dirty_segment(sbi, segno, DIRTY);
670 }
671
672 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 673}
674
15469963 675static void __add_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 676 struct block_device *bdev, block_t lstart,
677 block_t start, block_t len)
275b66b0 678{
0b54fb84
JK		 679 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
680 struct list_head *cmd_list = &(dcc->discard_cmd_list);
b01a9201 681 struct discard_cmd *dc;
275b66b0 682
b01a9201
JK		 683 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
684 INIT_LIST_HEAD(&dc->list);
c81abe34 685 dc->bdev = bdev;
b01a9201 686 dc->lstart = lstart;
c81abe34 687 dc->start = start;
b01a9201 688 dc->len = len;
15469963 689 dc->state = D_PREP;
c81abe34 690 dc->error = 0;
b01a9201 691 init_completion(&dc->wait);
15469963
JK		 692
693 mutex_lock(&dcc->cmd_lock);
0b54fb84 694 list_add_tail(&dc->list, cmd_list);
15469963 695 mutex_unlock(&dcc->cmd_lock);
5f32366a
CY		 696
697 atomic_inc(&dcc->discard_cmd_cnt);
15469963
JK		 698}
699
700static void __remove_discard_cmd(struct f2fs_sb_info *sbi, struct discard_cmd *dc)
701{
dcc9165d 702 if (dc->state == D_DONE)
8b8dd65f 703 atomic_dec(&(SM_I(sbi)->dcc_info->issing_discard));
dcc9165d 704
c81abe34
JK		 705 if (dc->error == -EOPNOTSUPP)
706 dc->error = 0;
15469963 707
c81abe34 708 if (dc->error)
15469963 709 f2fs_msg(sbi->sb, KERN_INFO,
c81abe34 710 "Issue discard failed, ret: %d", dc->error);
15469963
JK		 711 list_del(&dc->list);
712 kmem_cache_free(discard_cmd_slab, dc);
5f32366a 713 atomic_dec(&SM_I(sbi)->dcc_info->discard_cmd_cnt);
275b66b0
CY		 714}
715
c81abe34
JK		 716static void f2fs_submit_discard_endio(struct bio *bio)
717{
718 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
719
720 complete(&dc->wait);
721 dc->error = bio->bi_error;
722 dc->state = D_DONE;
723 bio_put(bio);
724}
725
726/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
727static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
728 struct discard_cmd *dc)
729{
730 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
731 struct bio *bio = NULL;
732
733 if (dc->state != D_PREP)
734 return;
735
736 dc->error = __blkdev_issue_discard(dc->bdev,
737 SECTOR_FROM_BLOCK(dc->start),
738 SECTOR_FROM_BLOCK(dc->len),
739 GFP_NOFS, 0, &bio);
740 if (!dc->error) {
741 /* should keep before submission to avoid D_DONE right away */
742 dc->state = D_SUBMIT;
8b8dd65f
CY		 743 atomic_inc(&dcc->issued_discard);
744 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 745 if (bio) {
746 bio->bi_private = dc;
747 bio->bi_end_io = f2fs_submit_discard_endio;
748 bio->bi_opf |= REQ_SYNC;
749 submit_bio(bio);
750 }
751 } else {
752 __remove_discard_cmd(sbi, dc);
753 }
754}
755
756static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
757 struct block_device *bdev, block_t blkstart, block_t blklen)
758{
759 block_t lblkstart = blkstart;
760
761 trace_f2fs_issue_discard(bdev, blkstart, blklen);
762
763 if (sbi->s_ndevs) {
764 int devi = f2fs_target_device_index(sbi, blkstart);
765
766 blkstart -= FDEV(devi).start_blk;
767 }
768 __add_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
769 wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue);
770 return 0;
771}
772
3d6a650f
YH		 773static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
774 struct discard_cmd *dc, block_t blkaddr)
775{
776 block_t end_block = START_BLOCK(sbi, GET_SEGNO(sbi, blkaddr) + 1);
777
778 if (dc->state == D_DONE || dc->lstart + dc->len <= end_block) {
779 __remove_discard_cmd(sbi, dc);
780 return;
781 }
782
783 if (blkaddr - dc->lstart < dc->lstart + dc->len - end_block) {
784 dc->start += (end_block - dc->lstart);
785 dc->len -= (end_block - dc->lstart);
786 dc->lstart = end_block;
787 } else {
788 dc->len = blkaddr - dc->lstart;
789 }
790}
791
4e6a8d9b
JK		 792/* This should be covered by global mutex, &sit_i->sentry_lock */
793void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
275b66b0 794{
0b54fb84
JK		 795 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
796 struct list_head *wait_list = &(dcc->discard_cmd_list);
b01a9201 797 struct discard_cmd *dc, *tmp;
40465257 798 struct blk_plug plug;
275b66b0 799
15469963 800 mutex_lock(&dcc->cmd_lock);
40465257
JK		 801
802 blk_start_plug(&plug);
803
b01a9201 804 list_for_each_entry_safe(dc, tmp, wait_list, list) {
275b66b0 805
15469963 806 if (blkaddr == NULL_ADDR) {
c81abe34 807 __submit_discard_cmd(sbi, dc);
15469963
JK		 808 continue;
809 }
810
811 if (dc->lstart <= blkaddr && blkaddr < dc->lstart + dc->len) {
812 if (dc->state == D_SUBMIT)
b01a9201 813 wait_for_completion_io(&dc->wait);
3d6a650f 814 __punch_discard_cmd(sbi, dc, blkaddr);
4e6a8d9b 815 }
275b66b0 816 }
40465257
JK		 817 blk_finish_plug(&plug);
818
819 /* this comes from f2fs_put_super */
820 if (blkaddr == NULL_ADDR) {
821 list_for_each_entry_safe(dc, tmp, wait_list, list) {
822 wait_for_completion_io(&dc->wait);
823 __remove_discard_cmd(sbi, dc);
824 }
825 }
15469963 826 mutex_unlock(&dcc->cmd_lock);
275b66b0
CY		 827}
828
15469963
JK		 829static int issue_discard_thread(void *data)
830{
831 struct f2fs_sb_info *sbi = data;
832 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
833 wait_queue_head_t *q = &dcc->discard_wait_queue;
834 struct list_head *cmd_list = &dcc->discard_cmd_list;
835 struct discard_cmd *dc, *tmp;
836 struct blk_plug plug;
837 int iter = 0;
838repeat:
839 if (kthread_should_stop())
840 return 0;
841
842 blk_start_plug(&plug);
843
844 mutex_lock(&dcc->cmd_lock);
845 list_for_each_entry_safe(dc, tmp, cmd_list, list) {
c81abe34
JK		 846
847 if (is_idle(sbi))
848 __submit_discard_cmd(sbi, dc);
849
850 if (dc->state == D_PREP && iter++ > DISCARD_ISSUE_RATE)
851 break;
852 if (dc->state == D_DONE)
15469963 853 __remove_discard_cmd(sbi, dc);
15469963
JK		 854 }
855 mutex_unlock(&dcc->cmd_lock);
856
857 blk_finish_plug(&plug);
858
859 iter = 0;
860 congestion_wait(BLK_RW_SYNC, HZ/50);
861
862 wait_event_interruptible(*q,
863 kthread_should_stop() || !list_empty(&dcc->discard_cmd_list));
864 goto repeat;
865}
866
f46e8809 867#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 868static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
869 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 870{
92592285 871 sector_t sector, nr_sects;
10a875f8 872 block_t lblkstart = blkstart;
3c62be17
JK		 873 int devi = 0;
874
875 if (sbi->s_ndevs) {
876 devi = f2fs_target_device_index(sbi, blkstart);
877 blkstart -= FDEV(devi).start_blk;
878 }
f46e8809
DLM		 879
880 /*
881 * We need to know the type of the zone: for conventional zones,
882 * use regular discard if the drive supports it. For sequential
883 * zones, reset the zone write pointer.
884 */
3c62be17 885 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 886
887 case BLK_ZONE_TYPE_CONVENTIONAL:
888 if (!blk_queue_discard(bdev_get_queue(bdev)))
889 return 0;
c81abe34 890 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 891 case BLK_ZONE_TYPE_SEQWRITE_REQ:
892 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 893 sector = SECTOR_FROM_BLOCK(blkstart);
894 nr_sects = SECTOR_FROM_BLOCK(blklen);
895
896 if (sector & (bdev_zone_sectors(bdev) - 1) ||
897 nr_sects != bdev_zone_sectors(bdev)) {
898 f2fs_msg(sbi->sb, KERN_INFO,
899 "(%d) %s: Unaligned discard attempted (block %x + %x)",
900 devi, sbi->s_ndevs ? FDEV(devi).path: "",
901 blkstart, blklen);
902 return -EIO;
903 }
d50aaeec 904 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 905 return blkdev_reset_zones(bdev, sector,
906 nr_sects, GFP_NOFS);
907 default:
908 /* Unknown zone type: broken device ? */
909 return -EIO;
910 }
911}
912#endif
913
3c62be17
JK		 914static int __issue_discard_async(struct f2fs_sb_info *sbi,
915 struct block_device *bdev, block_t blkstart, block_t blklen)
916{
917#ifdef CONFIG_BLK_DEV_ZONED
918 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
919 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
920 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
921#endif
c81abe34 922 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 923}
924
1e87a78d 925static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 926 block_t blkstart, block_t blklen)
927{
3c62be17
JK		 928 sector_t start = blkstart, len = 0;
929 struct block_device *bdev;
a66cdd98
JK		 930 struct seg_entry *se;
931 unsigned int offset;
932 block_t i;
3c62be17
JK		 933 int err = 0;
934
935 bdev = f2fs_target_device(sbi, blkstart, NULL);
936
937 for (i = blkstart; i < blkstart + blklen; i++, len++) {
938 if (i != start) {
939 struct block_device *bdev2 =
940 f2fs_target_device(sbi, i, NULL);
941
942 if (bdev2 != bdev) {
943 err = __issue_discard_async(sbi, bdev,
944 start, len);
945 if (err)
946 return err;
947 bdev = bdev2;
948 start = i;
949 len = 0;
950 }
951 }
a66cdd98 952
a66cdd98
JK		 953 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
954 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
955
956 if (!f2fs_test_and_set_bit(offset, se->discard_map))
957 sbi->discard_blks--;
958 }
f46e8809 959
3c62be17
JK		 960 if (len)
961 err = __issue_discard_async(sbi, bdev, start, len);
962 return err;
1e87a78d
JK		 963}
964
25290fa5
JK		 965static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
966 bool check_only)
adf4983b 967{
b2955550
JK		 968 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
969 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 970 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 971 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
972 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 973 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 974 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 975 unsigned int start = 0, end = -1;
4b2fecc8 976 bool force = (cpc->reason == CP_DISCARD);
a7eeb823
CY		 977 struct discard_entry *de = NULL;
978 struct list_head *head = &SM_I(sbi)->dcc_info->discard_entry_list;
b2955550
JK		 979 int i;
980
3e025740 981 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 982 return false;
b2955550 983
a66cdd98
JK		 984 if (!force) {
985 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 986 SM_I(sbi)->dcc_info->nr_discards >=
987 SM_I(sbi)->dcc_info->max_discards)
25290fa5 988 return false;
4b2fecc8
JK		 989 }
990
b2955550
JK		 991 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
992 for (i = 0; i < entries; i++)
a66cdd98 993 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 994 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 995
0b54fb84
JK		 996 while (force || SM_I(sbi)->dcc_info->nr_discards <=
997 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 998 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
999 if (start >= max_blocks)
1000 break;
1001
1002 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1003 if (force && start && end != max_blocks
1004 && (end - start) < cpc->trim_minlen)
1005 continue;
1006
25290fa5
JK		 1007 if (check_only)
1008 return true;
1009
a7eeb823
CY		 1010 if (!de) {
1011 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1012 GFP_F2FS_ZERO);
1013 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1014 list_add_tail(&de->list, head);
1015 }
1016
1017 for (i = start; i < end; i++)
1018 __set_bit_le(i, (void *)de->discard_map);
1019
1020 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1021 }
25290fa5 1022 return false;
b2955550
JK		 1023}
1024
4b2fecc8
JK		 1025void release_discard_addrs(struct f2fs_sb_info *sbi)
1026{
0b54fb84 1027 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
4b2fecc8
JK		 1028 struct discard_entry *entry, *this;
1029
1030 /* drop caches */
1031 list_for_each_entry_safe(entry, this, head, list) {
1032 list_del(&entry->list);
1033 kmem_cache_free(discard_entry_slab, entry);
1034 }
1035}
1036
0a8165d7 1037/*
351df4b2
JK		 1038 * Should call clear_prefree_segments after checkpoint is done.
1039 */
1040static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1041{
1042 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1043 unsigned int segno;
351df4b2
JK		 1044
1045 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1046 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1047 __set_test_and_free(sbi, segno);
351df4b2
JK		 1048 mutex_unlock(&dirty_i->seglist_lock);
1049}
1050
836b5a63 1051void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1052{
0b54fb84 1053 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
2d7b822a 1054 struct discard_entry *entry, *this;
351df4b2 1055 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1056 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1057 unsigned int start = 0, end = -1;
36abef4e 1058 unsigned int secno, start_segno;
c24a0fd6 1059 bool force = (cpc->reason == CP_DISCARD);
351df4b2
JK		 1060
1061 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1062
351df4b2 1063 while (1) {
29e59c14 1064 int i;
7cd8558b
JK		 1065 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1066 if (start >= MAIN_SEGS(sbi))
351df4b2 1067 break;
7cd8558b
JK		 1068 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1069 start + 1);
29e59c14
CL		 1070
1071 for (i = start; i < end; i++)
1072 clear_bit(i, prefree_map);
1073
1074 dirty_i->nr_dirty[PRE] -= end - start;
1075
650d3c4e 1076 if (!test_opt(sbi, DISCARD))
29e59c14 1077 continue;
351df4b2 1078
650d3c4e
YH		 1079 if (force && start >= cpc->trim_start &&
1080 (end - 1) <= cpc->trim_end)
1081 continue;
1082
36abef4e
JK		 1083 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1084 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1085 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1086 continue;
1087 }
1088next:
1089 secno = GET_SECNO(sbi, start);
1090 start_segno = secno * sbi->segs_per_sec;
1091 if (!IS_CURSEC(sbi, secno) &&
1092 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
1093 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1094 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1095
1096 start = start_segno + sbi->segs_per_sec;
1097 if (start < end)
1098 goto next;
8b107f5b
JK		 1099 else
1100 end = start - 1;
351df4b2
JK		 1101 }
1102 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1103
1104 /* send small discards */
2d7b822a 1105 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1106 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1107 bool is_valid = test_bit_le(0, entry->discard_map);
1108
1109find_next:
1110 if (is_valid) {
1111 next_pos = find_next_zero_bit_le(entry->discard_map,
1112 sbi->blocks_per_seg, cur_pos);
1113 len = next_pos - cur_pos;
1114
1115 if (force && len < cpc->trim_minlen)
1116 goto skip;
1117
1118 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1119 len);
1120 cpc->trimmed += len;
1121 total_len += len;
1122 } else {
1123 next_pos = find_next_bit_le(entry->discard_map,
1124 sbi->blocks_per_seg, cur_pos);
1125 }
836b5a63 1126skip:
a7eeb823
CY		 1127 cur_pos = next_pos;
1128 is_valid = !is_valid;
1129
1130 if (cur_pos < sbi->blocks_per_seg)
1131 goto find_next;
1132
b2955550 1133 list_del(&entry->list);
a7eeb823 1134 SM_I(sbi)->dcc_info->nr_discards -= total_len;
b2955550
JK		 1135 kmem_cache_free(discard_entry_slab, entry);
1136 }
351df4b2
JK		 1137}
1138
8ed59745 1139static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1140{
15469963 1141 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84
JK		 1142 struct discard_cmd_control *dcc;
1143 int err = 0;
1144
1145 if (SM_I(sbi)->dcc_info) {
1146 dcc = SM_I(sbi)->dcc_info;
1147 goto init_thread;
1148 }
1149
1150 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1151 if (!dcc)
1152 return -ENOMEM;
1153
1154 INIT_LIST_HEAD(&dcc->discard_entry_list);
1155 INIT_LIST_HEAD(&dcc->discard_cmd_list);
15469963 1156 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1157 atomic_set(&dcc->issued_discard, 0);
1158 atomic_set(&dcc->issing_discard, 0);
5f32366a 1159 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84
JK		 1160 dcc->nr_discards = 0;
1161 dcc->max_discards = 0;
1162
15469963 1163 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1164 SM_I(sbi)->dcc_info = dcc;
1165init_thread:
15469963
JK		 1166 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1167 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1168 if (IS_ERR(dcc->f2fs_issue_discard)) {
1169 err = PTR_ERR(dcc->f2fs_issue_discard);
1170 kfree(dcc);
1171 SM_I(sbi)->dcc_info = NULL;
1172 return err;
1173 }
1174
0b54fb84
JK		 1175 return err;
1176}
1177
f099405f 1178static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1179{
1180 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1181
f099405f
CY		 1182 if (!dcc)
1183 return;
1184
1185 if (dcc->f2fs_issue_discard) {
15469963
JK		 1186 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1187
1188 dcc->f2fs_issue_discard = NULL;
1189 kthread_stop(discard_thread);
1190 }
f099405f
CY		 1191
1192 kfree(dcc);
1193 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1194}
1195
184a5cd2 1196static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1197{
1198 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1199
1200 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1201 sit_i->dirty_sentries++;
184a5cd2
CY		 1202 return false;
1203 }
1204
1205 return true;
351df4b2
JK		 1206}
1207
1208static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1209 unsigned int segno, int modified)
1210{
1211 struct seg_entry *se = get_seg_entry(sbi, segno);
1212 se->type = type;
1213 if (modified)
1214 __mark_sit_entry_dirty(sbi, segno);
1215}
1216
1217static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1218{
1219 struct seg_entry *se;
1220 unsigned int segno, offset;
1221 long int new_vblocks;
1222
1223 segno = GET_SEGNO(sbi, blkaddr);
1224
1225 se = get_seg_entry(sbi, segno);
1226 new_vblocks = se->valid_blocks + del;
491c0854 1227 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1228
9850cf4a 1229 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1230 (new_vblocks > sbi->blocks_per_seg)));
1231
1232 se->valid_blocks = new_vblocks;
1233 se->mtime = get_mtime(sbi);
1234 SIT_I(sbi)->max_mtime = se->mtime;
1235
1236 /* Update valid block bitmap */
1237 if (del > 0) {
355e7891
CY		 1238 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1239#ifdef CONFIG_F2FS_CHECK_FS
1240 if (f2fs_test_and_set_bit(offset,
1241 se->cur_valid_map_mir))
1242 f2fs_bug_on(sbi, 1);
1243 else
1244 WARN_ON(1);
1245#else
05796763 1246 f2fs_bug_on(sbi, 1);
355e7891
CY		 1247#endif
1248 }
3e025740
JK		 1249 if (f2fs_discard_en(sbi) &&
1250 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1251 sbi->discard_blks--;
720037f9
JK		 1252
1253 /* don't overwrite by SSR to keep node chain */
1254 if (se->type == CURSEG_WARM_NODE) {
1255 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1256 se->ckpt_valid_blocks++;
1257 }
351df4b2 1258 } else {
355e7891
CY		 1259 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1260#ifdef CONFIG_F2FS_CHECK_FS
1261 if (!f2fs_test_and_clear_bit(offset,
1262 se->cur_valid_map_mir))
1263 f2fs_bug_on(sbi, 1);
1264 else
1265 WARN_ON(1);
1266#else
05796763 1267 f2fs_bug_on(sbi, 1);
355e7891
CY		 1268#endif
1269 }
3e025740
JK		 1270 if (f2fs_discard_en(sbi) &&
1271 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1272 sbi->discard_blks++;
351df4b2
JK		 1273 }
1274 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1275 se->ckpt_valid_blocks += del;
1276
1277 __mark_sit_entry_dirty(sbi, segno);
1278
1279 /* update total number of valid blocks to be written in ckpt area */
1280 SIT_I(sbi)->written_valid_blocks += del;
1281
1282 if (sbi->segs_per_sec > 1)
1283 get_sec_entry(sbi, segno)->valid_blocks += del;
1284}
1285
5e443818 1286void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1287{
5e443818
JK		 1288 update_sit_entry(sbi, new, 1);
1289 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1290 update_sit_entry(sbi, old, -1);
1291
1292 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1293 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 1294}
1295
1296void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1297{
1298 unsigned int segno = GET_SEGNO(sbi, addr);
1299 struct sit_info *sit_i = SIT_I(sbi);
1300
9850cf4a 1301 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1302 if (addr == NEW_ADDR)
1303 return;
1304
1305 /* add it into sit main buffer */
1306 mutex_lock(&sit_i->sentry_lock);
1307
1308 update_sit_entry(sbi, addr, -1);
1309
1310 /* add it into dirty seglist */
1311 locate_dirty_segment(sbi, segno);
1312
1313 mutex_unlock(&sit_i->sentry_lock);
1314}
1315
6e2c64ad
JK		 1316bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1317{
1318 struct sit_info *sit_i = SIT_I(sbi);
1319 unsigned int segno, offset;
1320 struct seg_entry *se;
1321 bool is_cp = false;
1322
1323 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1324 return true;
1325
1326 mutex_lock(&sit_i->sentry_lock);
1327
1328 segno = GET_SEGNO(sbi, blkaddr);
1329 se = get_seg_entry(sbi, segno);
1330 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1331
1332 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1333 is_cp = true;
1334
1335 mutex_unlock(&sit_i->sentry_lock);
1336
1337 return is_cp;
1338}
1339
0a8165d7 1340/*
351df4b2
JK		 1341 * This function should be resided under the curseg_mutex lock
1342 */
1343static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1344 struct f2fs_summary *sum)
351df4b2
JK		 1345{
1346 struct curseg_info *curseg = CURSEG_I(sbi, type);
1347 void *addr = curseg->sum_blk;
e79efe3b 1348 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1349 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1350}
1351
0a8165d7 1352/*
351df4b2
JK		 1353 * Calculate the number of current summary pages for writing
1354 */
3fa06d7b 1355int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1356{
351df4b2 1357 int valid_sum_count = 0;
9a47938b 1358 int i, sum_in_page;
351df4b2
JK		 1359
1360 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1361 if (sbi->ckpt->alloc_type[i] == SSR)
1362 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1363 else {
1364 if (for_ra)
1365 valid_sum_count += le16_to_cpu(
1366 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1367 else
1368 valid_sum_count += curseg_blkoff(sbi, i);
1369 }
351df4b2
JK		 1370 }
1371
09cbfeaf 1372 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1373 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1374 if (valid_sum_count <= sum_in_page)
351df4b2 1375 return 1;
9a47938b 1376 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1377 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1378 return 2;
1379 return 3;
1380}
1381
0a8165d7 1382/*
351df4b2
JK		 1383 * Caller should put this summary page
1384 */
1385struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1386{
1387 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1388}
1389
381722d2 1390void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1391{
1392 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1393 void *dst = page_address(page);
1394
1395 if (src)
09cbfeaf 1396 memcpy(dst, src, PAGE_SIZE);
381722d2 1397 else
09cbfeaf 1398 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1399 set_page_dirty(page);
1400 f2fs_put_page(page, 1);
1401}
1402
381722d2
CY		 1403static void write_sum_page(struct f2fs_sb_info *sbi,
1404 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1405{
1406 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1407}
1408
b7ad7512
CY		 1409static void write_current_sum_page(struct f2fs_sb_info *sbi,
1410 int type, block_t blk_addr)
1411{
1412 struct curseg_info *curseg = CURSEG_I(sbi, type);
1413 struct page *page = grab_meta_page(sbi, blk_addr);
1414 struct f2fs_summary_block *src = curseg->sum_blk;
1415 struct f2fs_summary_block *dst;
1416
1417 dst = (struct f2fs_summary_block *)page_address(page);
1418
1419 mutex_lock(&curseg->curseg_mutex);
1420
1421 down_read(&curseg->journal_rwsem);
1422 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1423 up_read(&curseg->journal_rwsem);
1424
1425 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1426 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1427
1428 mutex_unlock(&curseg->curseg_mutex);
1429
1430 set_page_dirty(page);
1431 f2fs_put_page(page, 1);
1432}
1433
0a8165d7 1434/*
351df4b2
JK		 1435 * Find a new segment from the free segments bitmap to right order
1436 * This function should be returned with success, otherwise BUG
1437 */
1438static void get_new_segment(struct f2fs_sb_info *sbi,
1439 unsigned int *newseg, bool new_sec, int dir)
1440{
1441 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1442 unsigned int segno, secno, zoneno;
7cd8558b 1443 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 1444 unsigned int hint = *newseg / sbi->segs_per_sec;
1445 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1446 unsigned int left_start = hint;
1447 bool init = true;
1448 int go_left = 0;
1449 int i;
1450
1a118ccf 1451 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1452
1453 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1454 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1455 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1456 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1457 goto got_it;
1458 }
1459find_other_zone:
7cd8558b
JK		 1460 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1461 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1462 if (dir == ALLOC_RIGHT) {
1463 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1464 MAIN_SECS(sbi), 0);
1465 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1466 } else {
1467 go_left = 1;
1468 left_start = hint - 1;
1469 }
1470 }
1471 if (go_left == 0)
1472 goto skip_left;
1473
1474 while (test_bit(left_start, free_i->free_secmap)) {
1475 if (left_start > 0) {
1476 left_start--;
1477 continue;
1478 }
1479 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1480 MAIN_SECS(sbi), 0);
1481 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1482 break;
1483 }
1484 secno = left_start;
1485skip_left:
1486 hint = secno;
1487 segno = secno * sbi->segs_per_sec;
1488 zoneno = secno / sbi->secs_per_zone;
1489
1490 /* give up on finding another zone */
1491 if (!init)
1492 goto got_it;
1493 if (sbi->secs_per_zone == 1)
1494 goto got_it;
1495 if (zoneno == old_zoneno)
1496 goto got_it;
1497 if (dir == ALLOC_LEFT) {
1498 if (!go_left && zoneno + 1 >= total_zones)
1499 goto got_it;
1500 if (go_left && zoneno == 0)
1501 goto got_it;
1502 }
1503 for (i = 0; i < NR_CURSEG_TYPE; i++)
1504 if (CURSEG_I(sbi, i)->zone == zoneno)
1505 break;
1506
1507 if (i < NR_CURSEG_TYPE) {
1508 /* zone is in user, try another */
1509 if (go_left)
1510 hint = zoneno * sbi->secs_per_zone - 1;
1511 else if (zoneno + 1 >= total_zones)
1512 hint = 0;
1513 else
1514 hint = (zoneno + 1) * sbi->secs_per_zone;
1515 init = false;
1516 goto find_other_zone;
1517 }
1518got_it:
1519 /* set it as dirty segment in free segmap */
9850cf4a 1520 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1521 __set_inuse(sbi, segno);
1522 *newseg = segno;
1a118ccf 1523 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1524}
1525
1526static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1527{
1528 struct curseg_info *curseg = CURSEG_I(sbi, type);
1529 struct summary_footer *sum_footer;
1530
1531 curseg->segno = curseg->next_segno;
1532 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1533 curseg->next_blkoff = 0;
1534 curseg->next_segno = NULL_SEGNO;
1535
1536 sum_footer = &(curseg->sum_blk->footer);
1537 memset(sum_footer, 0, sizeof(struct summary_footer));
1538 if (IS_DATASEG(type))
1539 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1540 if (IS_NODESEG(type))
1541 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1542 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1543}
1544
7a20b8a6
JK		 1545static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
1546{
1547 if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
1548 return 0;
1549
1550 return CURSEG_I(sbi, type)->segno;
1551}
1552
0a8165d7 1553/*
351df4b2
JK		 1554 * Allocate a current working segment.
1555 * This function always allocates a free segment in LFS manner.
1556 */
1557static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1558{
1559 struct curseg_info *curseg = CURSEG_I(sbi, type);
1560 unsigned int segno = curseg->segno;
1561 int dir = ALLOC_LEFT;
1562
1563 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1564 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1565 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1566 dir = ALLOC_RIGHT;
1567
1568 if (test_opt(sbi, NOHEAP))
1569 dir = ALLOC_RIGHT;
1570
7a20b8a6 1571 segno = __get_next_segno(sbi, type);
351df4b2
JK		 1572 get_new_segment(sbi, &segno, new_sec, dir);
1573 curseg->next_segno = segno;
1574 reset_curseg(sbi, type, 1);
1575 curseg->alloc_type = LFS;
1576}
1577
1578static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1579 struct curseg_info *seg, block_t start)
1580{
1581 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1582 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1583 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1584 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1585 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1586 int i, pos;
1587
1588 for (i = 0; i < entries; i++)
1589 target_map[i] = ckpt_map[i] | cur_map[i];
1590
1591 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1592
1593 seg->next_blkoff = pos;
351df4b2
JK		 1594}
1595
0a8165d7 1596/*
351df4b2
JK		 1597 * If a segment is written by LFS manner, next block offset is just obtained
1598 * by increasing the current block offset. However, if a segment is written by
1599 * SSR manner, next block offset obtained by calling __next_free_blkoff
1600 */
1601static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1602 struct curseg_info *seg)
1603{
1604 if (seg->alloc_type == SSR)
1605 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1606 else
1607 seg->next_blkoff++;
1608}
1609
0a8165d7 1610/*
e1c42045 1611 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1612 * manner, so it should recover the existing segment information of valid blocks
1613 */
1614static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1615{
1616 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1617 struct curseg_info *curseg = CURSEG_I(sbi, type);
1618 unsigned int new_segno = curseg->next_segno;
1619 struct f2fs_summary_block *sum_node;
1620 struct page *sum_page;
1621
1622 write_sum_page(sbi, curseg->sum_blk,
1623 GET_SUM_BLOCK(sbi, curseg->segno));
1624 __set_test_and_inuse(sbi, new_segno);
1625
1626 mutex_lock(&dirty_i->seglist_lock);
1627 __remove_dirty_segment(sbi, new_segno, PRE);
1628 __remove_dirty_segment(sbi, new_segno, DIRTY);
1629 mutex_unlock(&dirty_i->seglist_lock);
1630
1631 reset_curseg(sbi, type, 1);
1632 curseg->alloc_type = SSR;
1633 __next_free_blkoff(sbi, curseg, 0);
1634
1635 if (reuse) {
1636 sum_page = get_sum_page(sbi, new_segno);
1637 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1638 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1639 f2fs_put_page(sum_page, 1);
1640 }
1641}
1642
43727527
JK		 1643static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1644{
1645 struct curseg_info *curseg = CURSEG_I(sbi, type);
1646 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
d27c3d89
CY		 1647 int i, cnt;
1648 bool reversed = false;
c192f7a4
JK		 1649
1650 /* need_SSR() already forces to do this */
1651 if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR))
1652 return 1;
43727527 1653
70d625cb
JK		 1654 /* For node segments, let's do SSR more intensively */
1655 if (IS_NODESEG(type)) {
d27c3d89
CY		 1656 if (type >= CURSEG_WARM_NODE) {
1657 reversed = true;
1658 i = CURSEG_COLD_NODE;
1659 } else {
1660 i = CURSEG_HOT_NODE;
1661 }
1662 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 1663 } else {
d27c3d89
CY		 1664 if (type >= CURSEG_WARM_DATA) {
1665 reversed = true;
1666 i = CURSEG_COLD_DATA;
1667 } else {
1668 i = CURSEG_HOT_DATA;
1669 }
1670 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 1671 }
43727527 1672
d27c3d89 1673 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 1674 if (i == type)
1675 continue;
43727527 1676 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
c192f7a4 1677 BG_GC, i, SSR))
43727527 1678 return 1;
c192f7a4 1679 }
43727527
JK		 1680 return 0;
1681}
1682
351df4b2
JK		 1683/*
1684 * flush out current segment and replace it with new segment
1685 * This function should be returned with success, otherwise BUG
1686 */
1687static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1688 int type, bool force)
1689{
7b405275 1690 if (force)
351df4b2 1691 new_curseg(sbi, type, true);
5b6c6be2
JK		 1692 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1693 type == CURSEG_WARM_NODE)
351df4b2
JK		 1694 new_curseg(sbi, type, false);
1695 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1696 change_curseg(sbi, type, true);
1697 else
1698 new_curseg(sbi, type, false);
dcdfff65 1699
d0db7703 1700 stat_inc_seg_type(sbi, CURSEG_I(sbi, type));
351df4b2
JK		 1701}
1702
1703void allocate_new_segments(struct f2fs_sb_info *sbi)
1704{
6ae1be13
JK		 1705 struct curseg_info *curseg;
1706 unsigned int old_segno;
351df4b2
JK		 1707 int i;
1708
6ae1be13
JK		 1709 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1710 curseg = CURSEG_I(sbi, i);
1711 old_segno = curseg->segno;
1712 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1713 locate_dirty_segment(sbi, old_segno);
1714 }
351df4b2
JK		 1715}
1716
1717static const struct segment_allocation default_salloc_ops = {
1718 .allocate_segment = allocate_segment_by_default,
1719};
1720
25290fa5
JK		 1721bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1722{
1723 __u64 trim_start = cpc->trim_start;
1724 bool has_candidate = false;
1725
1726 mutex_lock(&SIT_I(sbi)->sentry_lock);
1727 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1728 if (add_discard_addrs(sbi, cpc, true)) {
1729 has_candidate = true;
1730 break;
1731 }
1732 }
1733 mutex_unlock(&SIT_I(sbi)->sentry_lock);
1734
1735 cpc->trim_start = trim_start;
1736 return has_candidate;
1737}
1738
4b2fecc8
JK		 1739int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1740{
f7ef9b83
JK		 1741 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1742 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1743 unsigned int start_segno, end_segno;
1744 struct cp_control cpc;
c34f42e2 1745 int err = 0;
4b2fecc8 1746
836b5a63 1747 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1748 return -EINVAL;
1749
9bd27ae4 1750 cpc.trimmed = 0;
7cd8558b 1751 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1752 goto out;
1753
ed214a11
YH		 1754 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1755 f2fs_msg(sbi->sb, KERN_WARNING,
1756 "Found FS corruption, run fsck to fix.");
1757 goto out;
1758 }
1759
4b2fecc8 1760 /* start/end segment number in main_area */
7cd8558b
JK		 1761 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1762 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1763 GET_SEGNO(sbi, end);
4b2fecc8 1764 cpc.reason = CP_DISCARD;
836b5a63 1765 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1766
1767 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1768 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1769 cpc.trim_start = start_segno;
a66cdd98
JK		 1770
1771 if (sbi->discard_blks == 0)
1772 break;
1773 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1774 cpc.trim_end = end_segno;
1775 else
1776 cpc.trim_end = min_t(unsigned int,
1777 rounddown(start_segno +
bba681cb
JK		 1778 BATCHED_TRIM_SEGMENTS(sbi),
1779 sbi->segs_per_sec) - 1, end_segno);
1780
1781 mutex_lock(&sbi->gc_mutex);
c34f42e2 1782 err = write_checkpoint(sbi, &cpc);
bba681cb 1783 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 1784 if (err)
1785 break;
74fa5f3d
CY		 1786
1787 schedule();
bba681cb 1788 }
4b2fecc8 1789out:
f7ef9b83 1790 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1791 return err;
4b2fecc8
JK		 1792}
1793
351df4b2
JK		 1794static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1795{
1796 struct curseg_info *curseg = CURSEG_I(sbi, type);
1797 if (curseg->next_blkoff < sbi->blocks_per_seg)
1798 return true;
1799 return false;
1800}
1801
1802static int __get_segment_type_2(struct page *page, enum page_type p_type)
1803{
1804 if (p_type == DATA)
1805 return CURSEG_HOT_DATA;
1806 else
1807 return CURSEG_HOT_NODE;
1808}
1809
1810static int __get_segment_type_4(struct page *page, enum page_type p_type)
1811{
1812 if (p_type == DATA) {
1813 struct inode *inode = page->mapping->host;
1814
1815 if (S_ISDIR(inode->i_mode))
1816 return CURSEG_HOT_DATA;
1817 else
1818 return CURSEG_COLD_DATA;
1819 } else {
a344b9fd
JK		 1820 if (IS_DNODE(page) && is_cold_node(page))
1821 return CURSEG_WARM_NODE;
351df4b2
JK		 1822 else
1823 return CURSEG_COLD_NODE;
1824 }
1825}
1826
1827static int __get_segment_type_6(struct page *page, enum page_type p_type)
1828{
1829 if (p_type == DATA) {
1830 struct inode *inode = page->mapping->host;
1831
ef095d19 1832 if (is_cold_data(page) || file_is_cold(inode))
351df4b2 1833 return CURSEG_COLD_DATA;
ef095d19
JK		 1834 if (is_inode_flag_set(inode, FI_HOT_DATA))
1835 return CURSEG_HOT_DATA;
1836 return CURSEG_WARM_DATA;
351df4b2
JK		 1837 } else {
1838 if (IS_DNODE(page))
1839 return is_cold_node(page) ? CURSEG_WARM_NODE :
1840 CURSEG_HOT_NODE;
ef095d19 1841 return CURSEG_COLD_NODE;
351df4b2
JK		 1842 }
1843}
1844
1845static int __get_segment_type(struct page *page, enum page_type p_type)
1846{
4081363f 1847 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1848 case 2:
1849 return __get_segment_type_2(page, p_type);
1850 case 4:
1851 return __get_segment_type_4(page, p_type);
351df4b2 1852 }
12a67146 1853 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1854 f2fs_bug_on(F2FS_P_SB(page),
1855 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1856 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1857}
1858
bfad7c2d
JK		 1859void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1860 block_t old_blkaddr, block_t *new_blkaddr,
1861 struct f2fs_summary *sum, int type)
351df4b2
JK		 1862{
1863 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 1864 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 1865
1866 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1867 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK		 1868
1869 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 1870
4e6a8d9b
JK		 1871 f2fs_wait_discard_bio(sbi, *new_blkaddr);
1872
351df4b2
JK		 1873 /*
1874 * __add_sum_entry should be resided under the curseg_mutex
1875 * because, this function updates a summary entry in the
1876 * current summary block.
1877 */
e79efe3b 1878 __add_sum_entry(sbi, type, sum);
351df4b2 1879
351df4b2 1880 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1881
1882 stat_inc_block_count(sbi, curseg);
351df4b2
JK		 1883
1884 /*
1885 * SIT information should be updated before segment allocation,
1886 * since SSR needs latest valid block information.
1887 */
1888 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1889
3436c4bd
YS		 1890 if (!__has_curseg_space(sbi, type))
1891 sit_i->s_ops->allocate_segment(sbi, type, false);
1892
351df4b2
JK		 1893 mutex_unlock(&sit_i->sentry_lock);
1894
bfad7c2d 1895 if (page && IS_NODESEG(type))
351df4b2
JK		 1896 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1897
bfad7c2d
JK		 1898 mutex_unlock(&curseg->curseg_mutex);
1899}
1900
05ca3632 1901static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1902{
05ca3632 1903 int type = __get_segment_type(fio->page, fio->type);
0a595eba 1904 int err;
bfad7c2d 1905
7dfeaa32
JK		 1906 if (fio->type == NODE || fio->type == DATA)
1907 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
0a595eba 1908reallocate:
7a9d7548
CY		 1909 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1910 &fio->new_blkaddr, sum, type);
bfad7c2d 1911
351df4b2 1912 /* writeout dirty page into bdev */
0a595eba
JK		 1913 err = f2fs_submit_page_mbio(fio);
1914 if (err == -EAGAIN) {
1915 fio->old_blkaddr = fio->new_blkaddr;
1916 goto reallocate;
1917 }
7dfeaa32
JK		 1918
1919 if (fio->type == NODE || fio->type == DATA)
1920 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1921}
1922
577e3495 1923void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1924{
458e6197 1925 struct f2fs_io_info fio = {
05ca3632 1926 .sbi = sbi,
458e6197 1927 .type = META,
04d328de 1928 .op = REQ_OP_WRITE,
70fd7614 1929 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1930 .old_blkaddr = page->index,
1931 .new_blkaddr = page->index,
05ca3632 1932 .page = page,
4375a336 1933 .encrypted_page = NULL,
458e6197
JK		 1934 };
1935
2b947003 1936 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 1937 fio.op_flags &= ~REQ_META;
2b947003 1938
351df4b2 1939 set_page_writeback(page);
05ca3632 1940 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1941}
1942
05ca3632 1943void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1944{
1945 struct f2fs_summary sum;
05ca3632 1946
351df4b2 1947 set_summary(&sum, nid, 0, 0);
05ca3632 1948 do_write_page(&sum, fio);
351df4b2
JK		 1949}
1950
05ca3632 1951void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1952{
05ca3632 1953 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1954 struct f2fs_summary sum;
1955 struct node_info ni;
1956
9850cf4a 1957 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1958 get_node_info(sbi, dn->nid, &ni);
1959 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1960 do_write_page(&sum, fio);
f28b3434 1961 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1962}
1963
05ca3632 1964void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1965{
7a9d7548 1966 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1967 stat_inc_inplace_blocks(fio->sbi);
1968 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1969}
1970
4356e48e 1971void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1972 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1973 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1974{
1975 struct sit_info *sit_i = SIT_I(sbi);
1976 struct curseg_info *curseg;
1977 unsigned int segno, old_cursegno;
1978 struct seg_entry *se;
1979 int type;
19f106bc 1980 unsigned short old_blkoff;
351df4b2
JK		 1981
1982 segno = GET_SEGNO(sbi, new_blkaddr);
1983 se = get_seg_entry(sbi, segno);
1984 type = se->type;
1985
19f106bc
CY		 1986 if (!recover_curseg) {
1987 /* for recovery flow */
1988 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1989 if (old_blkaddr == NULL_ADDR)
1990 type = CURSEG_COLD_DATA;
1991 else
1992 type = CURSEG_WARM_DATA;
1993 }
1994 } else {
1995 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1996 type = CURSEG_WARM_DATA;
1997 }
19f106bc 1998
351df4b2
JK		 1999 curseg = CURSEG_I(sbi, type);
2000
2001 mutex_lock(&curseg->curseg_mutex);
2002 mutex_lock(&sit_i->sentry_lock);
2003
2004 old_cursegno = curseg->segno;
19f106bc 2005 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2006
2007 /* change the current segment */
2008 if (segno != curseg->segno) {
2009 curseg->next_segno = segno;
2010 change_curseg(sbi, type, true);
2011 }
2012
491c0854 2013 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2014 __add_sum_entry(sbi, type, sum);
351df4b2 2015
28bc106b 2016 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2017 update_sit_entry(sbi, new_blkaddr, 1);
2018 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2019 update_sit_entry(sbi, old_blkaddr, -1);
2020
2021 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2022 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2023
351df4b2 2024 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2025
19f106bc
CY		 2026 if (recover_curseg) {
2027 if (old_cursegno != curseg->segno) {
2028 curseg->next_segno = old_cursegno;
2029 change_curseg(sbi, type, true);
2030 }
2031 curseg->next_blkoff = old_blkoff;
2032 }
2033
351df4b2
JK		 2034 mutex_unlock(&sit_i->sentry_lock);
2035 mutex_unlock(&curseg->curseg_mutex);
2036}
2037
528e3459
CY		 2038void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2039 block_t old_addr, block_t new_addr,
28bc106b
CY		 2040 unsigned char version, bool recover_curseg,
2041 bool recover_newaddr)
528e3459
CY		 2042{
2043 struct f2fs_summary sum;
2044
2045 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2046
28bc106b
CY		 2047 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2048 recover_curseg, recover_newaddr);
528e3459 2049
f28b3434 2050 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2051}
2052
93dfe2ac 2053void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2054 enum page_type type, bool ordered)
93dfe2ac 2055{
93dfe2ac 2056 if (PageWriteback(page)) {
4081363f
JK		 2057 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2058
942fd319
JK		 2059 f2fs_submit_merged_bio_cond(sbi, page->mapping->host,
2060 0, page->index, type, WRITE);
fec1d657
JK		 2061 if (ordered)
2062 wait_on_page_writeback(page);
2063 else
2064 wait_for_stable_page(page);
93dfe2ac
JK		 2065 }
2066}
2067
08b39fbd
CY		 2068void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
2069 block_t blkaddr)
2070{
2071 struct page *cpage;
2072
5d4c0af4 2073 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2074 return;
2075
08b39fbd
CY		 2076 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2077 if (cpage) {
fec1d657 2078 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2079 f2fs_put_page(cpage, 1);
2080 }
2081}
2082
351df4b2
JK		 2083static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2084{
2085 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2086 struct curseg_info *seg_i;
2087 unsigned char *kaddr;
2088 struct page *page;
2089 block_t start;
2090 int i, j, offset;
2091
2092 start = start_sum_block(sbi);
2093
2094 page = get_meta_page(sbi, start++);
2095 kaddr = (unsigned char *)page_address(page);
2096
2097 /* Step 1: restore nat cache */
2098 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2099 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2100
2101 /* Step 2: restore sit cache */
2102 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2103 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2104 offset = 2 * SUM_JOURNAL_SIZE;
2105
2106 /* Step 3: restore summary entries */
2107 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2108 unsigned short blk_off;
2109 unsigned int segno;
2110
2111 seg_i = CURSEG_I(sbi, i);
2112 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2113 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2114 seg_i->next_segno = segno;
2115 reset_curseg(sbi, i, 0);
2116 seg_i->alloc_type = ckpt->alloc_type[i];
2117 seg_i->next_blkoff = blk_off;
2118
2119 if (seg_i->alloc_type == SSR)
2120 blk_off = sbi->blocks_per_seg;
2121
2122 for (j = 0; j < blk_off; j++) {
2123 struct f2fs_summary *s;
2124 s = (struct f2fs_summary *)(kaddr + offset);
2125 seg_i->sum_blk->entries[j] = *s;
2126 offset += SUMMARY_SIZE;
09cbfeaf 2127 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2128 SUM_FOOTER_SIZE)
2129 continue;
2130
2131 f2fs_put_page(page, 1);
2132 page = NULL;
2133
2134 page = get_meta_page(sbi, start++);
2135 kaddr = (unsigned char *)page_address(page);
2136 offset = 0;
2137 }
2138 }
2139 f2fs_put_page(page, 1);
2140 return 0;
2141}
2142
2143static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2144{
2145 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2146 struct f2fs_summary_block *sum;
2147 struct curseg_info *curseg;
2148 struct page *new;
2149 unsigned short blk_off;
2150 unsigned int segno = 0;
2151 block_t blk_addr = 0;
2152
2153 /* get segment number and block addr */
2154 if (IS_DATASEG(type)) {
2155 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2156 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2157 CURSEG_HOT_DATA]);
119ee914 2158 if (__exist_node_summaries(sbi))
351df4b2
JK		 2159 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2160 else
2161 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2162 } else {
2163 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2164 CURSEG_HOT_NODE]);
2165 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2166 CURSEG_HOT_NODE]);
119ee914 2167 if (__exist_node_summaries(sbi))
351df4b2
JK		 2168 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2169 type - CURSEG_HOT_NODE);
2170 else
2171 blk_addr = GET_SUM_BLOCK(sbi, segno);
2172 }
2173
2174 new = get_meta_page(sbi, blk_addr);
2175 sum = (struct f2fs_summary_block *)page_address(new);
2176
2177 if (IS_NODESEG(type)) {
119ee914 2178 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2179 struct f2fs_summary *ns = &sum->entries[0];
2180 int i;
2181 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2182 ns->version = 0;
2183 ns->ofs_in_node = 0;
2184 }
2185 } else {
d653788a
GZ		 2186 int err;
2187
2188 err = restore_node_summary(sbi, segno, sum);
2189 if (err) {
351df4b2 2190 f2fs_put_page(new, 1);
d653788a 2191 return err;
351df4b2
JK		 2192 }
2193 }
2194 }
2195
2196 /* set uncompleted segment to curseg */
2197 curseg = CURSEG_I(sbi, type);
2198 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2199
2200 /* update journal info */
2201 down_write(&curseg->journal_rwsem);
2202 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2203 up_write(&curseg->journal_rwsem);
2204
2205 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2206 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2207 curseg->next_segno = segno;
2208 reset_curseg(sbi, type, 0);
2209 curseg->alloc_type = ckpt->alloc_type[type];
2210 curseg->next_blkoff = blk_off;
2211 mutex_unlock(&curseg->curseg_mutex);
2212 f2fs_put_page(new, 1);
2213 return 0;
2214}
2215
2216static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2217{
2218 int type = CURSEG_HOT_DATA;
e4fc5fbf 2219 int err;
351df4b2 2220
aaec2b1d 2221 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2222 int npages = npages_for_summary_flush(sbi, true);
2223
2224 if (npages >= 2)
2225 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2226 META_CP, true);
3fa06d7b 2227
351df4b2
JK		 2228 /* restore for compacted data summary */
2229 if (read_compacted_summaries(sbi))
2230 return -EINVAL;
2231 type = CURSEG_HOT_NODE;
2232 }
2233
119ee914 2234 if (__exist_node_summaries(sbi))
3fa06d7b 2235 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2236 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2237
e4fc5fbf
CY		 2238 for (; type <= CURSEG_COLD_NODE; type++) {
2239 err = read_normal_summaries(sbi, type);
2240 if (err)
2241 return err;
2242 }
2243
351df4b2
JK		 2244 return 0;
2245}
2246
2247static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2248{
2249 struct page *page;
2250 unsigned char *kaddr;
2251 struct f2fs_summary *summary;
2252 struct curseg_info *seg_i;
2253 int written_size = 0;
2254 int i, j;
2255
2256 page = grab_meta_page(sbi, blkaddr++);
2257 kaddr = (unsigned char *)page_address(page);
2258
2259 /* Step 1: write nat cache */
2260 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2261 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2262 written_size += SUM_JOURNAL_SIZE;
2263
2264 /* Step 2: write sit cache */
2265 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2266 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2267 written_size += SUM_JOURNAL_SIZE;
2268
351df4b2
JK		 2269 /* Step 3: write summary entries */
2270 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2271 unsigned short blkoff;
2272 seg_i = CURSEG_I(sbi, i);
2273 if (sbi->ckpt->alloc_type[i] == SSR)
2274 blkoff = sbi->blocks_per_seg;
2275 else
2276 blkoff = curseg_blkoff(sbi, i);
2277
2278 for (j = 0; j < blkoff; j++) {
2279 if (!page) {
2280 page = grab_meta_page(sbi, blkaddr++);
2281 kaddr = (unsigned char *)page_address(page);
2282 written_size = 0;
2283 }
2284 summary = (struct f2fs_summary *)(kaddr + written_size);
2285 *summary = seg_i->sum_blk->entries[j];
2286 written_size += SUMMARY_SIZE;
351df4b2 2287
09cbfeaf 2288 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2289 SUM_FOOTER_SIZE)
2290 continue;
2291
e8d61a74 2292 set_page_dirty(page);
351df4b2
JK		 2293 f2fs_put_page(page, 1);
2294 page = NULL;
2295 }
2296 }
e8d61a74
CY		 2297 if (page) {
2298 set_page_dirty(page);
351df4b2 2299 f2fs_put_page(page, 1);
e8d61a74 2300 }
351df4b2
JK		 2301}
2302
2303static void write_normal_summaries(struct f2fs_sb_info *sbi,
2304 block_t blkaddr, int type)
2305{
2306 int i, end;
2307 if (IS_DATASEG(type))
2308 end = type + NR_CURSEG_DATA_TYPE;
2309 else
2310 end = type + NR_CURSEG_NODE_TYPE;
2311
b7ad7512
CY		 2312 for (i = type; i < end; i++)
2313 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 2314}
2315
2316void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2317{
aaec2b1d 2318 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 2319 write_compacted_summaries(sbi, start_blk);
2320 else
2321 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2322}
2323
2324void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2325{
119ee914 2326 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 2327}
2328
dfc08a12 2329int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 2330 unsigned int val, int alloc)
2331{
2332 int i;
2333
2334 if (type == NAT_JOURNAL) {
dfc08a12
CY		 2335 for (i = 0; i < nats_in_cursum(journal); i++) {
2336 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 2337 return i;
2338 }
dfc08a12
CY		 2339 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2340 return update_nats_in_cursum(journal, 1);
351df4b2 2341 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 2342 for (i = 0; i < sits_in_cursum(journal); i++)
2343 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 2344 return i;
dfc08a12
CY		 2345 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2346 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 2347 }
2348 return -1;
2349}
2350
2351static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2352 unsigned int segno)
2353{
2cc22186 2354 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 2355}
2356
2357static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2358 unsigned int start)
2359{
2360 struct sit_info *sit_i = SIT_I(sbi);
2361 struct page *src_page, *dst_page;
2362 pgoff_t src_off, dst_off;
2363 void *src_addr, *dst_addr;
2364
2365 src_off = current_sit_addr(sbi, start);
2366 dst_off = next_sit_addr(sbi, src_off);
2367
2368 /* get current sit block page without lock */
2369 src_page = get_meta_page(sbi, src_off);
2370 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 2371 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 2372
2373 src_addr = page_address(src_page);
2374 dst_addr = page_address(dst_page);
09cbfeaf 2375 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 2376
2377 set_page_dirty(dst_page);
2378 f2fs_put_page(src_page, 1);
2379
2380 set_to_next_sit(sit_i, start);
2381
2382 return dst_page;
2383}
2384
184a5cd2
CY		 2385static struct sit_entry_set *grab_sit_entry_set(void)
2386{
2387 struct sit_entry_set *ses =
80c54505 2388 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 2389
2390 ses->entry_cnt = 0;
2391 INIT_LIST_HEAD(&ses->set_list);
2392 return ses;
2393}
2394
2395static void release_sit_entry_set(struct sit_entry_set *ses)
2396{
2397 list_del(&ses->set_list);
2398 kmem_cache_free(sit_entry_set_slab, ses);
2399}
2400
2401static void adjust_sit_entry_set(struct sit_entry_set *ses,
2402 struct list_head *head)
2403{
2404 struct sit_entry_set *next = ses;
2405
2406 if (list_is_last(&ses->set_list, head))
2407 return;
2408
2409 list_for_each_entry_continue(next, head, set_list)
2410 if (ses->entry_cnt <= next->entry_cnt)
2411 break;
2412
2413 list_move_tail(&ses->set_list, &next->set_list);
2414}
2415
2416static void add_sit_entry(unsigned int segno, struct list_head *head)
2417{
2418 struct sit_entry_set *ses;
2419 unsigned int start_segno = START_SEGNO(segno);
2420
2421 list_for_each_entry(ses, head, set_list) {
2422 if (ses->start_segno == start_segno) {
2423 ses->entry_cnt++;
2424 adjust_sit_entry_set(ses, head);
2425 return;
2426 }
2427 }
2428
2429 ses = grab_sit_entry_set();
2430
2431 ses->start_segno = start_segno;
2432 ses->entry_cnt++;
2433 list_add(&ses->set_list, head);
2434}
2435
2436static void add_sits_in_set(struct f2fs_sb_info *sbi)
2437{
2438 struct f2fs_sm_info *sm_info = SM_I(sbi);
2439 struct list_head *set_list = &sm_info->sit_entry_set;
2440 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 2441 unsigned int segno;
2442
7cd8558b 2443 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 2444 add_sit_entry(segno, set_list);
2445}
2446
2447static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 2448{
2449 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2450 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 2451 int i;
2452
b7ad7512 2453 down_write(&curseg->journal_rwsem);
dfc08a12 2454 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 2455 unsigned int segno;
2456 bool dirtied;
2457
dfc08a12 2458 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 2459 dirtied = __mark_sit_entry_dirty(sbi, segno);
2460
2461 if (!dirtied)
2462 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 2463 }
dfc08a12 2464 update_sits_in_cursum(journal, -i);
b7ad7512 2465 up_write(&curseg->journal_rwsem);
351df4b2
JK		 2466}
2467
0a8165d7 2468/*
351df4b2
JK		 2469 * CP calls this function, which flushes SIT entries including sit_journal,
2470 * and moves prefree segs to free segs.
2471 */
4b2fecc8 2472void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 2473{
2474 struct sit_info *sit_i = SIT_I(sbi);
2475 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2476 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2477 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 2478 struct sit_entry_set *ses, *tmp;
2479 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 2480 bool to_journal = true;
4b2fecc8 2481 struct seg_entry *se;
351df4b2 2482
351df4b2
JK		 2483 mutex_lock(&sit_i->sentry_lock);
2484
2b11a74b
WL		 2485 if (!sit_i->dirty_sentries)
2486 goto out;
2487
351df4b2 2488 /*
184a5cd2
CY		 2489 * add and account sit entries of dirty bitmap in sit entry
2490 * set temporarily
351df4b2 2491 */
184a5cd2 2492 add_sits_in_set(sbi);
351df4b2 2493
184a5cd2
CY		 2494 /*
2495 * if there are no enough space in journal to store dirty sit
2496 * entries, remove all entries from journal and add and account
2497 * them in sit entry set.
2498 */
dfc08a12 2499 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2500 remove_sits_in_journal(sbi);
b2955550 2501
184a5cd2
CY		 2502 /*
2503 * there are two steps to flush sit entries:
2504 * #1, flush sit entries to journal in current cold data summary block.
2505 * #2, flush sit entries to sit page.
2506 */
2507 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2508 struct page *page = NULL;
184a5cd2
CY		 2509 struct f2fs_sit_block *raw_sit = NULL;
2510 unsigned int start_segno = ses->start_segno;
2511 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2512 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2513 unsigned int segno = start_segno;
2514
2515 if (to_journal &&
dfc08a12 2516 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2517 to_journal = false;
2518
b7ad7512
CY		 2519 if (to_journal) {
2520 down_write(&curseg->journal_rwsem);
2521 } else {
184a5cd2
CY		 2522 page = get_next_sit_page(sbi, start_segno);
2523 raw_sit = page_address(page);
351df4b2 2524 }
351df4b2 2525
184a5cd2
CY		 2526 /* flush dirty sit entries in region of current sit set */
2527 for_each_set_bit_from(segno, bitmap, end) {
2528 int offset, sit_offset;
4b2fecc8
JK		 2529
2530 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2531
2532 /* add discard candidates */
d7bc2484 2533 if (cpc->reason != CP_DISCARD) {
4b2fecc8 2534 cpc->trim_start = segno;
25290fa5 2535 add_discard_addrs(sbi, cpc, false);
4b2fecc8 2536 }
184a5cd2
CY		 2537
2538 if (to_journal) {
dfc08a12 2539 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2540 SIT_JOURNAL, segno, 1);
2541 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2542 segno_in_journal(journal, offset) =
184a5cd2
CY		 2543 cpu_to_le32(segno);
2544 seg_info_to_raw_sit(se,
dfc08a12 2545 &sit_in_journal(journal, offset));
184a5cd2
CY		 2546 } else {
2547 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2548 seg_info_to_raw_sit(se,
2549 &raw_sit->entries[sit_offset]);
2550 }
351df4b2 2551
184a5cd2
CY		 2552 __clear_bit(segno, bitmap);
2553 sit_i->dirty_sentries--;
2554 ses->entry_cnt--;
351df4b2
JK		 2555 }
2556
b7ad7512
CY		 2557 if (to_journal)
2558 up_write(&curseg->journal_rwsem);
2559 else
184a5cd2
CY		 2560 f2fs_put_page(page, 1);
2561
2562 f2fs_bug_on(sbi, ses->entry_cnt);
2563 release_sit_entry_set(ses);
351df4b2 2564 }
184a5cd2
CY		 2565
2566 f2fs_bug_on(sbi, !list_empty(head));
2567 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2568out:
4b2fecc8 2569 if (cpc->reason == CP_DISCARD) {
650d3c4e
YH		 2570 __u64 trim_start = cpc->trim_start;
2571
4b2fecc8 2572 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 2573 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 2574
2575 cpc->trim_start = trim_start;
4b2fecc8 2576 }
351df4b2 2577 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2578
351df4b2
JK		 2579 set_prefree_as_free_segments(sbi);
2580}
2581
2582static int build_sit_info(struct f2fs_sb_info *sbi)
2583{
2584 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 2585 struct sit_info *sit_i;
2586 unsigned int sit_segs, start;
ae27d62e 2587 char *src_bitmap;
351df4b2
JK		 2588 unsigned int bitmap_size;
2589
2590 /* allocate memory for SIT information */
2591 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2592 if (!sit_i)
2593 return -ENOMEM;
2594
2595 SM_I(sbi)->sit_info = sit_i;
2596
39307a8e
JK		 2597 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2598 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2599 if (!sit_i->sentries)
2600 return -ENOMEM;
2601
7cd8558b 2602 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2603 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2604 if (!sit_i->dirty_sentries_bitmap)
2605 return -ENOMEM;
2606
7cd8558b 2607 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2608 sit_i->sentries[start].cur_valid_map
2609 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2610 sit_i->sentries[start].ckpt_valid_map
2611 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 2612 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 2613 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 2614 return -ENOMEM;
3e025740 2615
355e7891
CY		 2616#ifdef CONFIG_F2FS_CHECK_FS
2617 sit_i->sentries[start].cur_valid_map_mir
2618 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2619 if (!sit_i->sentries[start].cur_valid_map_mir)
2620 return -ENOMEM;
2621#endif
2622
3e025740
JK		 2623 if (f2fs_discard_en(sbi)) {
2624 sit_i->sentries[start].discard_map
2625 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2626 if (!sit_i->sentries[start].discard_map)
2627 return -ENOMEM;
2628 }
351df4b2
JK		 2629 }
2630
60a3b782
JK		 2631 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2632 if (!sit_i->tmp_map)
2633 return -ENOMEM;
2634
351df4b2 2635 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2636 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2637 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2638 if (!sit_i->sec_entries)
2639 return -ENOMEM;
2640 }
2641
2642 /* get information related with SIT */
2643 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2644
2645 /* setup SIT bitmap from ckeckpoint pack */
2646 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2647 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2648
ae27d62e
CY		 2649 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2650 if (!sit_i->sit_bitmap)
351df4b2 2651 return -ENOMEM;
351df4b2 2652
ae27d62e
CY		 2653#ifdef CONFIG_F2FS_CHECK_FS
2654 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2655 if (!sit_i->sit_bitmap_mir)
2656 return -ENOMEM;
2657#endif
2658
351df4b2
JK		 2659 /* init SIT information */
2660 sit_i->s_ops = &default_salloc_ops;
2661
2662 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2663 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 2664 sit_i->written_valid_blocks = 0;
351df4b2
JK		 2665 sit_i->bitmap_size = bitmap_size;
2666 sit_i->dirty_sentries = 0;
2667 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2668 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2669 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2670 mutex_init(&sit_i->sentry_lock);
2671 return 0;
2672}
2673
2674static int build_free_segmap(struct f2fs_sb_info *sbi)
2675{
351df4b2
JK		 2676 struct free_segmap_info *free_i;
2677 unsigned int bitmap_size, sec_bitmap_size;
2678
2679 /* allocate memory for free segmap information */
2680 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2681 if (!free_i)
2682 return -ENOMEM;
2683
2684 SM_I(sbi)->free_info = free_i;
2685
7cd8558b 2686 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2687 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2688 if (!free_i->free_segmap)
2689 return -ENOMEM;
2690
7cd8558b 2691 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2692 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2693 if (!free_i->free_secmap)
2694 return -ENOMEM;
2695
2696 /* set all segments as dirty temporarily */
2697 memset(free_i->free_segmap, 0xff, bitmap_size);
2698 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2699
2700 /* init free segmap information */
7cd8558b 2701 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2702 free_i->free_segments = 0;
2703 free_i->free_sections = 0;
1a118ccf 2704 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2705 return 0;
2706}
2707
2708static int build_curseg(struct f2fs_sb_info *sbi)
2709{
1042d60f 2710 struct curseg_info *array;
351df4b2
JK		 2711 int i;
2712
b434babf 2713 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2714 if (!array)
2715 return -ENOMEM;
2716
2717 SM_I(sbi)->curseg_array = array;
2718
2719 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2720 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2721 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2722 if (!array[i].sum_blk)
2723 return -ENOMEM;
b7ad7512
CY		 2724 init_rwsem(&array[i].journal_rwsem);
2725 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2726 GFP_KERNEL);
2727 if (!array[i].journal)
2728 return -ENOMEM;
351df4b2
JK		 2729 array[i].segno = NULL_SEGNO;
2730 array[i].next_blkoff = 0;
2731 }
2732 return restore_curseg_summaries(sbi);
2733}
2734
2735static void build_sit_entries(struct f2fs_sb_info *sbi)
2736{
2737 struct sit_info *sit_i = SIT_I(sbi);
2738 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2739 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 2740 struct seg_entry *se;
2741 struct f2fs_sit_entry sit;
74de593a
CY		 2742 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2743 unsigned int i, start, end;
2744 unsigned int readed, start_blk = 0;
351df4b2 2745
74de593a 2746 do {
664ba972
JK		 2747 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2748 META_SIT, true);
74de593a
CY		 2749
2750 start = start_blk * sit_i->sents_per_block;
2751 end = (start_blk + readed) * sit_i->sents_per_block;
2752
7cd8558b 2753 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 2754 struct f2fs_sit_block *sit_blk;
74de593a
CY		 2755 struct page *page;
2756
9c094040 2757 se = &sit_i->sentries[start];
74de593a
CY		 2758 page = get_current_sit_page(sbi, start);
2759 sit_blk = (struct f2fs_sit_block *)page_address(page);
2760 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2761 f2fs_put_page(page, 1);
d600af23 2762
74de593a
CY		 2763 check_block_count(sbi, start, &sit);
2764 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2765
2766 /* build discard map only one time */
3e025740
JK		 2767 if (f2fs_discard_en(sbi)) {
2768 memcpy(se->discard_map, se->cur_valid_map,
2769 SIT_VBLOCK_MAP_SIZE);
2770 sbi->discard_blks += sbi->blocks_per_seg -
2771 se->valid_blocks;
2772 }
a66cdd98 2773
d600af23
CY		 2774 if (sbi->segs_per_sec > 1)
2775 get_sec_entry(sbi, start)->valid_blocks +=
2776 se->valid_blocks;
351df4b2 2777 }
74de593a
CY		 2778 start_blk += readed;
2779 } while (start_blk < sit_blk_cnt);
d600af23
CY		 2780
2781 down_read(&curseg->journal_rwsem);
2782 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 2783 unsigned int old_valid_blocks;
2784
2785 start = le32_to_cpu(segno_in_journal(journal, i));
2786 se = &sit_i->sentries[start];
2787 sit = sit_in_journal(journal, i);
2788
2789 old_valid_blocks = se->valid_blocks;
2790
2791 check_block_count(sbi, start, &sit);
2792 seg_info_from_raw_sit(se, &sit);
2793
2794 if (f2fs_discard_en(sbi)) {
2795 memcpy(se->discard_map, se->cur_valid_map,
2796 SIT_VBLOCK_MAP_SIZE);
2797 sbi->discard_blks += old_valid_blocks -
2798 se->valid_blocks;
2799 }
2800
2801 if (sbi->segs_per_sec > 1)
2802 get_sec_entry(sbi, start)->valid_blocks +=
2803 se->valid_blocks - old_valid_blocks;
2804 }
2805 up_read(&curseg->journal_rwsem);
351df4b2
JK		 2806}
2807
2808static void init_free_segmap(struct f2fs_sb_info *sbi)
2809{
2810 unsigned int start;
2811 int type;
2812
7cd8558b 2813 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2814 struct seg_entry *sentry = get_seg_entry(sbi, start);
2815 if (!sentry->valid_blocks)
2816 __set_free(sbi, start);
c79b7ff1
JK		 2817 else
2818 SIT_I(sbi)->written_valid_blocks +=
2819 sentry->valid_blocks;
351df4b2
JK		 2820 }
2821
2822 /* set use the current segments */
2823 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2824 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2825 __set_test_and_inuse(sbi, curseg_t->segno);
2826 }
2827}
2828
2829static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2830{
2831 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2832 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2833 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2834 unsigned short valid_blocks;
2835
8736fbf0 2836 while (1) {
351df4b2 2837 /* find dirty segment based on free segmap */
7cd8558b
JK		 2838 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2839 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2840 break;
2841 offset = segno + 1;
2842 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2843 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2844 continue;
ec325b52
JK		 2845 if (valid_blocks > sbi->blocks_per_seg) {
2846 f2fs_bug_on(sbi, 1);
2847 continue;
2848 }
351df4b2
JK		 2849 mutex_lock(&dirty_i->seglist_lock);
2850 __locate_dirty_segment(sbi, segno, DIRTY);
2851 mutex_unlock(&dirty_i->seglist_lock);
2852 }
2853}
2854
5ec4e49f 2855static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2856{
2857 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2858 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2859
39307a8e 2860 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2861 if (!dirty_i->victim_secmap)
351df4b2
JK		 2862 return -ENOMEM;
2863 return 0;
2864}
2865
2866static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2867{
2868 struct dirty_seglist_info *dirty_i;
2869 unsigned int bitmap_size, i;
2870
2871 /* allocate memory for dirty segments list information */
2872 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2873 if (!dirty_i)
2874 return -ENOMEM;
2875
2876 SM_I(sbi)->dirty_info = dirty_i;
2877 mutex_init(&dirty_i->seglist_lock);
2878
7cd8558b 2879 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2880
2881 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2882 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2883 if (!dirty_i->dirty_segmap[i])
2884 return -ENOMEM;
2885 }
2886
2887 init_dirty_segmap(sbi);
5ec4e49f 2888 return init_victim_secmap(sbi);
351df4b2
JK		 2889}
2890
0a8165d7 2891/*
351df4b2
JK		 2892 * Update min, max modified time for cost-benefit GC algorithm
2893 */
2894static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2895{
2896 struct sit_info *sit_i = SIT_I(sbi);
2897 unsigned int segno;
2898
2899 mutex_lock(&sit_i->sentry_lock);
2900
2901 sit_i->min_mtime = LLONG_MAX;
2902
7cd8558b 2903 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2904 unsigned int i;
2905 unsigned long long mtime = 0;
2906
2907 for (i = 0; i < sbi->segs_per_sec; i++)
2908 mtime += get_seg_entry(sbi, segno + i)->mtime;
2909
2910 mtime = div_u64(mtime, sbi->segs_per_sec);
2911
2912 if (sit_i->min_mtime > mtime)
2913 sit_i->min_mtime = mtime;
2914 }
2915 sit_i->max_mtime = get_mtime(sbi);
2916 mutex_unlock(&sit_i->sentry_lock);
2917}
2918
2919int build_segment_manager(struct f2fs_sb_info *sbi)
2920{
2921 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2922 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2923 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2924 int err;
2925
2926 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2927 if (!sm_info)
2928 return -ENOMEM;
2929
2930 /* init sm info */
2931 sbi->sm_info = sm_info;
351df4b2
JK		 2932 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2933 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2934 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2935 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2936 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2937 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2938 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2939 sm_info->rec_prefree_segments = sm_info->main_segments *
2940 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 2941 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2942 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2943
52763a4b
JK		 2944 if (!test_opt(sbi, LFS))
2945 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2946 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2947 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 2948 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
351df4b2 2949
bba681cb
JK		 2950 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2951
184a5cd2
CY		 2952 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2953
b270ad6f 2954 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2955 err = create_flush_cmd_control(sbi);
2956 if (err)
a688b9d9 2957 return err;
6b4afdd7
JK		 2958 }
2959
0b54fb84
JK		 2960 err = create_discard_cmd_control(sbi);
2961 if (err)
2962 return err;
2963
351df4b2
JK		 2964 err = build_sit_info(sbi);
2965 if (err)
2966 return err;
2967 err = build_free_segmap(sbi);
2968 if (err)
2969 return err;
2970 err = build_curseg(sbi);
2971 if (err)
2972 return err;
2973
2974 /* reinit free segmap based on SIT */
2975 build_sit_entries(sbi);
2976
2977 init_free_segmap(sbi);
2978 err = build_dirty_segmap(sbi);
2979 if (err)
2980 return err;
2981
2982 init_min_max_mtime(sbi);
2983 return 0;
2984}
2985
2986static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2987 enum dirty_type dirty_type)
2988{
2989 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2990
2991 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2992 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2993 dirty_i->nr_dirty[dirty_type] = 0;
2994 mutex_unlock(&dirty_i->seglist_lock);
2995}
2996
5ec4e49f 2997static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2998{
2999 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3000 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3001}
3002
3003static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3004{
3005 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3006 int i;
3007
3008 if (!dirty_i)
3009 return;
3010
3011 /* discard pre-free/dirty segments list */
3012 for (i = 0; i < NR_DIRTY_TYPE; i++)
3013 discard_dirty_segmap(sbi, i);
3014
5ec4e49f 3015 destroy_victim_secmap(sbi);
351df4b2
JK		 3016 SM_I(sbi)->dirty_info = NULL;
3017 kfree(dirty_i);
3018}
3019
3020static void destroy_curseg(struct f2fs_sb_info *sbi)
3021{
3022 struct curseg_info *array = SM_I(sbi)->curseg_array;
3023 int i;
3024
3025 if (!array)
3026 return;
3027 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3028 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3029 kfree(array[i].sum_blk);
b7ad7512
CY		 3030 kfree(array[i].journal);
3031 }
351df4b2
JK		 3032 kfree(array);
3033}
3034
3035static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3036{
3037 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3038 if (!free_i)
3039 return;
3040 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3041 kvfree(free_i->free_segmap);
3042 kvfree(free_i->free_secmap);
351df4b2
JK		 3043 kfree(free_i);
3044}
3045
3046static void destroy_sit_info(struct f2fs_sb_info *sbi)
3047{
3048 struct sit_info *sit_i = SIT_I(sbi);
3049 unsigned int start;
3050
3051 if (!sit_i)
3052 return;
3053
3054 if (sit_i->sentries) {
7cd8558b 3055 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3056 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 3057#ifdef CONFIG_F2FS_CHECK_FS
3058 kfree(sit_i->sentries[start].cur_valid_map_mir);
3059#endif
351df4b2 3060 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3061 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3062 }
3063 }
60a3b782
JK		 3064 kfree(sit_i->tmp_map);
3065
39307a8e
JK		 3066 kvfree(sit_i->sentries);
3067 kvfree(sit_i->sec_entries);
3068 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3069
3070 SM_I(sbi)->sit_info = NULL;
3071 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 3072#ifdef CONFIG_F2FS_CHECK_FS
3073 kfree(sit_i->sit_bitmap_mir);
3074#endif
351df4b2
JK		 3075 kfree(sit_i);
3076}
3077
3078void destroy_segment_manager(struct f2fs_sb_info *sbi)
3079{
3080 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3081
3b03f724
CY		 3082 if (!sm_info)
3083 return;
5eba8c5d 3084 destroy_flush_cmd_control(sbi, true);
f099405f 3085 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3086 destroy_dirty_segmap(sbi);
3087 destroy_curseg(sbi);
3088 destroy_free_segmap(sbi);
3089 destroy_sit_info(sbi);
3090 sbi->sm_info = NULL;
3091 kfree(sm_info);
3092}
7fd9e544
JK		 3093
3094int __init create_segment_manager_caches(void)
3095{
3096 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3097 sizeof(struct discard_entry));
7fd9e544 3098 if (!discard_entry_slab)
184a5cd2
CY		 3099 goto fail;
3100
b01a9201
JK		 3101 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3102 sizeof(struct discard_cmd));
3103 if (!discard_cmd_slab)
6ab2a308 3104 goto destroy_discard_entry;
275b66b0 3105
184a5cd2 3106 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3107 sizeof(struct sit_entry_set));
184a5cd2 3108 if (!sit_entry_set_slab)
b01a9201 3109 goto destroy_discard_cmd;
88b88a66
JK		 3110
3111 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3112 sizeof(struct inmem_pages));
3113 if (!inmem_entry_slab)
3114 goto destroy_sit_entry_set;
7fd9e544 3115 return 0;
184a5cd2 3116
88b88a66
JK		 3117destroy_sit_entry_set:
3118 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 3119destroy_discard_cmd:
3120 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3121destroy_discard_entry:
184a5cd2
CY		 3122 kmem_cache_destroy(discard_entry_slab);
3123fail:
3124 return -ENOMEM;
7fd9e544
JK		 3125}
3126
3127void destroy_segment_manager_caches(void)
3128{
184a5cd2 3129 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3130 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3131 kmem_cache_destroy(discard_entry_slab);
88b88a66 3132 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3133}
Linux 6.x block layer and io_uring tree(s)