projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: split drop_inmem_pages from commit_inmem_pages
[deliverable/linux.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;
184a5cd2 29static struct kmem_cache *sit_entry_set_slab;
88b88a66 30static struct kmem_cache *inmem_entry_slab;
7fd9e544 31
f96999c3
JK		 32static unsigned long __reverse_ulong(unsigned char *str)
33{
34 unsigned long tmp = 0;
35 int shift = 24, idx = 0;
36
37#if BITS_PER_LONG == 64
38 shift = 56;
39#endif
40 while (shift >= 0) {
41 tmp |= (unsigned long)str[idx++] << shift;
42 shift -= BITS_PER_BYTE;
43 }
44 return tmp;
45}
46
9a7f143a
CL		 47/*
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
50 */
51static inline unsigned long __reverse_ffs(unsigned long word)
52{
53 int num = 0;
54
55#if BITS_PER_LONG == 64
f96999c3 56 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 57 num += 32;
f96999c3 58 else
9a7f143a 59 word >>= 32;
9a7f143a 60#endif
f96999c3 61 if ((word & 0xffff0000) == 0)
9a7f143a 62 num += 16;
f96999c3 63 else
9a7f143a 64 word >>= 16;
f96999c3
JK		 65
66 if ((word & 0xff00) == 0)
9a7f143a 67 num += 8;
f96999c3 68 else
9a7f143a 69 word >>= 8;
f96999c3 70
9a7f143a
CL		 71 if ((word & 0xf0) == 0)
72 num += 4;
73 else
74 word >>= 4;
f96999c3 75
9a7f143a
CL		 76 if ((word & 0xc) == 0)
77 num += 2;
78 else
79 word >>= 2;
f96999c3 80
9a7f143a
CL		 81 if ((word & 0x2) == 0)
82 num += 1;
83 return num;
84}
85
86/*
e1c42045 87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 89 * @size must be integral times of unsigned long.
9a7f143a 90 * Example:
f96999c3
JK		 91 * MSB <--> LSB
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 94 */
95static unsigned long __find_rev_next_bit(const unsigned long *addr,
96 unsigned long size, unsigned long offset)
97{
98 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 99 unsigned long result = size;
9a7f143a 100 unsigned long tmp;
9a7f143a
CL		 101
102 if (offset >= size)
103 return size;
104
692223d1 105 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 106 offset %= BITS_PER_LONG;
f96999c3 107
692223d1
FL		 108 while (1) {
109 if (*p == 0)
110 goto pass;
9a7f143a 111
f96999c3 112 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 113
114 tmp &= ~0UL >> offset;
115 if (size < BITS_PER_LONG)
116 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 117 if (tmp)
692223d1
FL		 118 goto found;
119pass:
120 if (size <= BITS_PER_LONG)
121 break;
9a7f143a 122 size -= BITS_PER_LONG;
692223d1 123 offset = 0;
f96999c3 124 p++;
9a7f143a 125 }
692223d1
FL		 126 return result;
127found:
128 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 129}
130
131static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
132 unsigned long size, unsigned long offset)
133{
134 const unsigned long *p = addr + BIT_WORD(offset);
80609448 135 unsigned long result = size;
9a7f143a 136 unsigned long tmp;
9a7f143a
CL		 137
138 if (offset >= size)
139 return size;
140
80609448 141 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 142 offset %= BITS_PER_LONG;
80609448
JK		 143
144 while (1) {
145 if (*p == ~0UL)
146 goto pass;
147
f96999c3 148 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 149
150 if (offset)
151 tmp |= ~0UL << (BITS_PER_LONG - offset);
152 if (size < BITS_PER_LONG)
153 tmp |= ~0UL >> size;
f96999c3 154 if (tmp != ~0UL)
80609448
JK		 155 goto found;
156pass:
157 if (size <= BITS_PER_LONG)
158 break;
9a7f143a 159 size -= BITS_PER_LONG;
80609448 160 offset = 0;
f96999c3 161 p++;
9a7f143a 162 }
80609448
JK		 163 return result;
164found:
165 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 166}
167
88b88a66
JK		 168void register_inmem_page(struct inode *inode, struct page *page)
169{
170 struct f2fs_inode_info *fi = F2FS_I(inode);
171 struct inmem_pages *new;
9be32d72 172
9e4ded3f 173 f2fs_trace_pid(page);
0722b101 174
decd36b6
CY		 175 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
176 SetPagePrivate(page);
177
88b88a66
JK		 178 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
179
180 /* add atomic page indices to the list */
181 new->page = page;
182 INIT_LIST_HEAD(&new->list);
decd36b6 183
88b88a66
JK		 184 /* increase reference count with clean state */
185 mutex_lock(&fi->inmem_lock);
186 get_page(page);
187 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 188 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 189 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 190
191 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 192}
193
29b96b54
CY		 194static void __revoke_inmem_pages(struct inode *inode,
195 struct list_head *head)
196{
197 struct inmem_pages *cur, *tmp;
198
199 list_for_each_entry_safe(cur, tmp, head, list) {
200 trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
201
202 lock_page(cur->page);
203 ClearPageUptodate(cur->page);
204 set_page_private(cur->page, 0);
205 ClearPagePrivate(cur->page);
206 f2fs_put_page(cur->page, 1);
207
208 list_del(&cur->list);
209 kmem_cache_free(inmem_entry_slab, cur);
210 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
211 }
212}
213
214void drop_inmem_pages(struct inode *inode)
215{
216 struct f2fs_inode_info *fi = F2FS_I(inode);
217
218 mutex_lock(&fi->inmem_lock);
219 __revoke_inmem_pages(inode, &fi->inmem_pages);
220 mutex_unlock(&fi->inmem_lock);
221}
222
223static int __commit_inmem_pages(struct inode *inode)
88b88a66
JK		 224{
225 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
226 struct f2fs_inode_info *fi = F2FS_I(inode);
227 struct inmem_pages *cur, *tmp;
88b88a66 228 struct f2fs_io_info fio = {
05ca3632 229 .sbi = sbi,
88b88a66 230 .type = DATA,
1e84371f 231 .rw = WRITE_SYNC | REQ_PRIO,
4375a336 232 .encrypted_page = NULL,
88b88a66 233 };
29b96b54 234 bool submit_bio = false;
edb27dee 235 int err = 0;
88b88a66 236
88b88a66 237 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
decd36b6 238 lock_page(cur->page);
29b96b54
CY		 239 if (cur->page->mapping == inode->i_mapping) {
240 set_page_dirty(cur->page);
241 f2fs_wait_on_page_writeback(cur->page, DATA, true);
242 if (clear_page_dirty_for_io(cur->page))
243 inode_dec_dirty_pages(inode);
244 trace_f2fs_commit_inmem_page(cur->page, INMEM);
245 fio.page = cur->page;
246 err = do_write_data_page(&fio);
247 if (err) {
248 unlock_page(cur->page);
249 break;
70c640b1 250 }
29b96b54
CY		 251 clear_cold_data(cur->page);
252 submit_bio = true;
88b88a66 253 }
29b96b54 254
decd36b6
CY		 255 set_page_private(cur->page, 0);
256 ClearPagePrivate(cur->page);
257 f2fs_put_page(cur->page, 1);
258
88b88a66
JK		 259 list_del(&cur->list);
260 kmem_cache_free(inmem_entry_slab, cur);
8dcf2ff7 261 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 262 }
29b96b54
CY		 263
264 if (submit_bio)
265 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
266 return err;
267}
268
269int commit_inmem_pages(struct inode *inode)
270{
271 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
272 struct f2fs_inode_info *fi = F2FS_I(inode);
273 int err = 0;
274
275 f2fs_balance_fs(sbi, true);
276 f2fs_lock_op(sbi);
277
278 mutex_lock(&fi->inmem_lock);
279 err = __commit_inmem_pages(inode);
88b88a66
JK		 280 mutex_unlock(&fi->inmem_lock);
281
29b96b54 282 f2fs_unlock_op(sbi);
edb27dee 283 return err;
88b88a66
JK		 284}
285
0a8165d7 286/*
351df4b2
JK		 287 * This function balances dirty node and dentry pages.
288 * In addition, it controls garbage collection.
289 */
2c4db1a6 290void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 291{
2c4db1a6
JK		 292 if (!need)
293 return;
351df4b2 294 /*
029cd28c
JK		 295 * We should do GC or end up with checkpoint, if there are so many dirty
296 * dir/node pages without enough free segments.
351df4b2 297 */
43727527 298 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 299 mutex_lock(&sbi->gc_mutex);
d530d4d8 300 f2fs_gc(sbi, false);
351df4b2
JK		 301 }
302}
303
4660f9c0
JK		 304void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
305{
1dcc336b 306 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 307 if (!available_free_memory(sbi, EXTENT_CACHE))
308 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 309
1b38dc8e
JK		 310 /* check the # of cached NAT entries */
311 if (!available_free_memory(sbi, NAT_ENTRIES))
312 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
313
31696580
CY		 314 if (!available_free_memory(sbi, FREE_NIDS))
315 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
316
1b38dc8e
JK		 317 /* checkpoint is the only way to shrink partial cached entries */
318 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 319 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 320 excess_prefree_segs(sbi) ||
321 excess_dirty_nats(sbi) ||
d0239e1b 322 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
36b35a0d
CY		 323 if (test_opt(sbi, DATA_FLUSH))
324 sync_dirty_inodes(sbi, FILE_INODE);
4660f9c0 325 f2fs_sync_fs(sbi->sb, true);
42190d2a 326 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 327 }
4660f9c0
JK		 328}
329
2163d198 330static int issue_flush_thread(void *data)
6b4afdd7
JK		 331{
332 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 333 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
334 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 335repeat:
336 if (kthread_should_stop())
337 return 0;
338
721bd4d5 339 if (!llist_empty(&fcc->issue_list)) {
740432f8 340 struct bio *bio;
6b4afdd7
JK		 341 struct flush_cmd *cmd, *next;
342 int ret;
343
740432f8
JK		 344 bio = f2fs_bio_alloc(0);
345
721bd4d5
GZ		 346 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
347 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
348
6b4afdd7
JK		 349 bio->bi_bdev = sbi->sb->s_bdev;
350 ret = submit_bio_wait(WRITE_FLUSH, bio);
351
721bd4d5
GZ		 352 llist_for_each_entry_safe(cmd, next,
353 fcc->dispatch_list, llnode) {
6b4afdd7 354 cmd->ret = ret;
6b4afdd7
JK		 355 complete(&cmd->wait);
356 }
a4ed23f2 357 bio_put(bio);
a688b9d9 358 fcc->dispatch_list = NULL;
6b4afdd7
JK		 359 }
360
a688b9d9 361 wait_event_interruptible(*q,
721bd4d5 362 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 363 goto repeat;
364}
365
366int f2fs_issue_flush(struct f2fs_sb_info *sbi)
367{
a688b9d9 368 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 369 struct flush_cmd cmd;
6b4afdd7 370
24a9ee0f
JK		 371 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
372 test_opt(sbi, FLUSH_MERGE));
373
0f7b2abd
JK		 374 if (test_opt(sbi, NOBARRIER))
375 return 0;
376
740432f8
JK		 377 if (!test_opt(sbi, FLUSH_MERGE)) {
378 struct bio *bio = f2fs_bio_alloc(0);
379 int ret;
380
381 bio->bi_bdev = sbi->sb->s_bdev;
382 ret = submit_bio_wait(WRITE_FLUSH, bio);
383 bio_put(bio);
384 return ret;
385 }
6b4afdd7 386
adf8d90b 387 init_completion(&cmd.wait);
6b4afdd7 388
721bd4d5 389 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 390
a688b9d9
GZ		 391 if (!fcc->dispatch_list)
392 wake_up(&fcc->flush_wait_queue);
6b4afdd7 393
adf8d90b
CY		 394 wait_for_completion(&cmd.wait);
395
396 return cmd.ret;
6b4afdd7
JK		 397}
398
2163d198
GZ		 399int create_flush_cmd_control(struct f2fs_sb_info *sbi)
400{
401 dev_t dev = sbi->sb->s_bdev->bd_dev;
402 struct flush_cmd_control *fcc;
403 int err = 0;
404
405 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
406 if (!fcc)
407 return -ENOMEM;
2163d198 408 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 409 init_llist_head(&fcc->issue_list);
6b2920a5 410 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 411 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
412 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
413 if (IS_ERR(fcc->f2fs_issue_flush)) {
414 err = PTR_ERR(fcc->f2fs_issue_flush);
415 kfree(fcc);
6b2920a5 416 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 417 return err;
418 }
2163d198
GZ		 419
420 return err;
421}
422
423void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
424{
6b2920a5 425 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 426
427 if (fcc && fcc->f2fs_issue_flush)
428 kthread_stop(fcc->f2fs_issue_flush);
429 kfree(fcc);
6b2920a5 430 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 431}
432
351df4b2
JK		 433static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
434 enum dirty_type dirty_type)
435{
436 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
437
438 /* need not be added */
439 if (IS_CURSEG(sbi, segno))
440 return;
441
442 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
443 dirty_i->nr_dirty[dirty_type]++;
444
445 if (dirty_type == DIRTY) {
446 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 447 enum dirty_type t = sentry->type;
b2f2c390 448
ec325b52
JK		 449 if (unlikely(t >= DIRTY)) {
450 f2fs_bug_on(sbi, 1);
451 return;
452 }
4625d6aa
CL		 453 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
454 dirty_i->nr_dirty[t]++;
351df4b2
JK		 455 }
456}
457
458static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
459 enum dirty_type dirty_type)
460{
461 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
462
463 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
464 dirty_i->nr_dirty[dirty_type]--;
465
466 if (dirty_type == DIRTY) {
4625d6aa
CL		 467 struct seg_entry *sentry = get_seg_entry(sbi, segno);
468 enum dirty_type t = sentry->type;
469
470 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
471 dirty_i->nr_dirty[t]--;
b2f2c390 472
5ec4e49f
JK		 473 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
474 clear_bit(GET_SECNO(sbi, segno),
475 dirty_i->victim_secmap);
351df4b2
JK		 476 }
477}
478
0a8165d7 479/*
351df4b2
JK		 480 * Should not occur error such as -ENOMEM.
481 * Adding dirty entry into seglist is not critical operation.
482 * If a given segment is one of current working segments, it won't be added.
483 */
8d8451af 484static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 485{
486 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
487 unsigned short valid_blocks;
488
489 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
490 return;
491
492 mutex_lock(&dirty_i->seglist_lock);
493
494 valid_blocks = get_valid_blocks(sbi, segno, 0);
495
496 if (valid_blocks == 0) {
497 __locate_dirty_segment(sbi, segno, PRE);
498 __remove_dirty_segment(sbi, segno, DIRTY);
499 } else if (valid_blocks < sbi->blocks_per_seg) {
500 __locate_dirty_segment(sbi, segno, DIRTY);
501 } else {
502 /* Recovery routine with SSR needs this */
503 __remove_dirty_segment(sbi, segno, DIRTY);
504 }
505
506 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 507}
508
1e87a78d 509static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 510 block_t blkstart, block_t blklen)
511{
55cf9cb6
CY		 512 sector_t start = SECTOR_FROM_BLOCK(blkstart);
513 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 514 struct seg_entry *se;
515 unsigned int offset;
516 block_t i;
517
518 for (i = blkstart; i < blkstart + blklen; i++) {
519 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
520 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
521
522 if (!f2fs_test_and_set_bit(offset, se->discard_map))
523 sbi->discard_blks--;
524 }
1661d07c 525 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 526 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
527}
528
e90c2d28 529bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 530{
60b286c4 531 int err = -EOPNOTSUPP;
40a02be1
JK		 532
533 if (test_opt(sbi, DISCARD)) {
534 struct seg_entry *se = get_seg_entry(sbi,
535 GET_SEGNO(sbi, blkaddr));
536 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
537
538 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 539 return false;
40a02be1
JK		 540
541 err = f2fs_issue_discard(sbi, blkaddr, 1);
542 }
543
e90c2d28 544 if (err) {
381722d2 545 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 546 return true;
547 }
548 return false;
37208879
JK		 549}
550
adf4983b 551static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 552 struct cp_control *cpc, struct seg_entry *se,
553 unsigned int start, unsigned int end)
b2955550
JK		 554{
555 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 556 struct discard_entry *new, *last;
557
558 if (!list_empty(head)) {
559 last = list_last_entry(head, struct discard_entry, list);
560 if (START_BLOCK(sbi, cpc->trim_start) + start ==
561 last->blkaddr + last->len) {
562 last->len += end - start;
563 goto done;
564 }
565 }
566
567 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
568 INIT_LIST_HEAD(&new->list);
569 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
570 new->len = end - start;
571 list_add_tail(&new->list, head);
572done:
573 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 574}
575
576static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
577{
b2955550
JK		 578 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
579 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 580 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 581 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
582 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 583 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 584 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 585 unsigned int start = 0, end = -1;
4b2fecc8 586 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 587 int i;
588
a66cdd98 589 if (se->valid_blocks == max_blocks)
b2955550
JK		 590 return;
591
a66cdd98
JK		 592 if (!force) {
593 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 594 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
595 return;
4b2fecc8
JK		 596 }
597
b2955550
JK		 598 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
599 for (i = 0; i < entries; i++)
a66cdd98 600 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 601 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 602
4b2fecc8 603 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 604 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
605 if (start >= max_blocks)
606 break;
607
608 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 609 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 610 }
611}
612
4b2fecc8
JK		 613void release_discard_addrs(struct f2fs_sb_info *sbi)
614{
615 struct list_head *head = &(SM_I(sbi)->discard_list);
616 struct discard_entry *entry, *this;
617
618 /* drop caches */
619 list_for_each_entry_safe(entry, this, head, list) {
620 list_del(&entry->list);
621 kmem_cache_free(discard_entry_slab, entry);
622 }
623}
624
0a8165d7 625/*
351df4b2
JK		 626 * Should call clear_prefree_segments after checkpoint is done.
627 */
628static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
629{
630 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 631 unsigned int segno;
351df4b2
JK		 632
633 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 634 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 635 __set_test_and_free(sbi, segno);
351df4b2
JK		 636 mutex_unlock(&dirty_i->seglist_lock);
637}
638
836b5a63 639void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 640{
b2955550 641 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 642 struct discard_entry *entry, *this;
351df4b2 643 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 644 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 645 unsigned int start = 0, end = -1;
351df4b2
JK		 646
647 mutex_lock(&dirty_i->seglist_lock);
29e59c14 648
351df4b2 649 while (1) {
29e59c14 650 int i;
7cd8558b
JK		 651 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
652 if (start >= MAIN_SEGS(sbi))
351df4b2 653 break;
7cd8558b
JK		 654 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
655 start + 1);
29e59c14
CL		 656
657 for (i = start; i < end; i++)
658 clear_bit(i, prefree_map);
659
660 dirty_i->nr_dirty[PRE] -= end - start;
661
662 if (!test_opt(sbi, DISCARD))
663 continue;
351df4b2 664
37208879
JK		 665 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
666 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 667 }
668 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 669
670 /* send small discards */
2d7b822a 671 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 672 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
673 goto skip;
37208879 674 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 675 cpc->trimmed += entry->len;
836b5a63 676skip:
b2955550
JK		 677 list_del(&entry->list);
678 SM_I(sbi)->nr_discards -= entry->len;
679 kmem_cache_free(discard_entry_slab, entry);
680 }
351df4b2
JK		 681}
682
184a5cd2 683static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 684{
685 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 686
687 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 688 sit_i->dirty_sentries++;
184a5cd2
CY		 689 return false;
690 }
691
692 return true;
351df4b2
JK		 693}
694
695static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
696 unsigned int segno, int modified)
697{
698 struct seg_entry *se = get_seg_entry(sbi, segno);
699 se->type = type;
700 if (modified)
701 __mark_sit_entry_dirty(sbi, segno);
702}
703
704static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
705{
706 struct seg_entry *se;
707 unsigned int segno, offset;
708 long int new_vblocks;
709
710 segno = GET_SEGNO(sbi, blkaddr);
711
712 se = get_seg_entry(sbi, segno);
713 new_vblocks = se->valid_blocks + del;
491c0854 714 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 715
9850cf4a 716 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 717 (new_vblocks > sbi->blocks_per_seg)));
718
719 se->valid_blocks = new_vblocks;
720 se->mtime = get_mtime(sbi);
721 SIT_I(sbi)->max_mtime = se->mtime;
722
723 /* Update valid block bitmap */
724 if (del > 0) {
52aca074 725 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 726 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 727 if (!f2fs_test_and_set_bit(offset, se->discard_map))
728 sbi->discard_blks--;
351df4b2 729 } else {
52aca074 730 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 731 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 732 if (f2fs_test_and_clear_bit(offset, se->discard_map))
733 sbi->discard_blks++;
351df4b2
JK		 734 }
735 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
736 se->ckpt_valid_blocks += del;
737
738 __mark_sit_entry_dirty(sbi, segno);
739
740 /* update total number of valid blocks to be written in ckpt area */
741 SIT_I(sbi)->written_valid_blocks += del;
742
743 if (sbi->segs_per_sec > 1)
744 get_sec_entry(sbi, segno)->valid_blocks += del;
745}
746
5e443818 747void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 748{
5e443818
JK		 749 update_sit_entry(sbi, new, 1);
750 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
751 update_sit_entry(sbi, old, -1);
752
753 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
754 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 755}
756
757void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
758{
759 unsigned int segno = GET_SEGNO(sbi, addr);
760 struct sit_info *sit_i = SIT_I(sbi);
761
9850cf4a 762 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 763 if (addr == NEW_ADDR)
764 return;
765
766 /* add it into sit main buffer */
767 mutex_lock(&sit_i->sentry_lock);
768
769 update_sit_entry(sbi, addr, -1);
770
771 /* add it into dirty seglist */
772 locate_dirty_segment(sbi, segno);
773
774 mutex_unlock(&sit_i->sentry_lock);
775}
776
6e2c64ad
JK		 777bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
778{
779 struct sit_info *sit_i = SIT_I(sbi);
780 unsigned int segno, offset;
781 struct seg_entry *se;
782 bool is_cp = false;
783
784 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
785 return true;
786
787 mutex_lock(&sit_i->sentry_lock);
788
789 segno = GET_SEGNO(sbi, blkaddr);
790 se = get_seg_entry(sbi, segno);
791 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
792
793 if (f2fs_test_bit(offset, se->ckpt_valid_map))
794 is_cp = true;
795
796 mutex_unlock(&sit_i->sentry_lock);
797
798 return is_cp;
799}
800
0a8165d7 801/*
351df4b2
JK		 802 * This function should be resided under the curseg_mutex lock
803 */
804static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 805 struct f2fs_summary *sum)
351df4b2
JK		 806{
807 struct curseg_info *curseg = CURSEG_I(sbi, type);
808 void *addr = curseg->sum_blk;
e79efe3b 809 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 810 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 811}
812
0a8165d7 813/*
351df4b2
JK		 814 * Calculate the number of current summary pages for writing
815 */
3fa06d7b 816int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 817{
351df4b2 818 int valid_sum_count = 0;
9a47938b 819 int i, sum_in_page;
351df4b2
JK		 820
821 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
822 if (sbi->ckpt->alloc_type[i] == SSR)
823 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 824 else {
825 if (for_ra)
826 valid_sum_count += le16_to_cpu(
827 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
828 else
829 valid_sum_count += curseg_blkoff(sbi, i);
830 }
351df4b2
JK		 831 }
832
9a47938b
FL		 833 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
834 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
835 if (valid_sum_count <= sum_in_page)
351df4b2 836 return 1;
9a47938b
FL		 837 else if ((valid_sum_count - sum_in_page) <=
838 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 839 return 2;
840 return 3;
841}
842
0a8165d7 843/*
351df4b2
JK		 844 * Caller should put this summary page
845 */
846struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
847{
848 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
849}
850
381722d2 851void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 852{
853 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 854 void *dst = page_address(page);
855
856 if (src)
857 memcpy(dst, src, PAGE_CACHE_SIZE);
858 else
859 memset(dst, 0, PAGE_CACHE_SIZE);
351df4b2
JK		 860 set_page_dirty(page);
861 f2fs_put_page(page, 1);
862}
863
381722d2
CY		 864static void write_sum_page(struct f2fs_sb_info *sbi,
865 struct f2fs_summary_block *sum_blk, block_t blk_addr)
866{
867 update_meta_page(sbi, (void *)sum_blk, blk_addr);
868}
869
60374688
JK		 870static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
871{
872 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 873 unsigned int segno = curseg->segno + 1;
60374688
JK		 874 struct free_segmap_info *free_i = FREE_I(sbi);
875
7cd8558b 876 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 877 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 878 return 0;
879}
880
0a8165d7 881/*
351df4b2
JK		 882 * Find a new segment from the free segments bitmap to right order
883 * This function should be returned with success, otherwise BUG
884 */
885static void get_new_segment(struct f2fs_sb_info *sbi,
886 unsigned int *newseg, bool new_sec, int dir)
887{
888 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 889 unsigned int segno, secno, zoneno;
7cd8558b 890 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 891 unsigned int hint = *newseg / sbi->segs_per_sec;
892 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
893 unsigned int left_start = hint;
894 bool init = true;
895 int go_left = 0;
896 int i;
897
1a118ccf 898 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 899
900 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
901 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 902 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
903 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 904 goto got_it;
905 }
906find_other_zone:
7cd8558b
JK		 907 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
908 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 909 if (dir == ALLOC_RIGHT) {
910 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 911 MAIN_SECS(sbi), 0);
912 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 913 } else {
914 go_left = 1;
915 left_start = hint - 1;
916 }
917 }
918 if (go_left == 0)
919 goto skip_left;
920
921 while (test_bit(left_start, free_i->free_secmap)) {
922 if (left_start > 0) {
923 left_start--;
924 continue;
925 }
926 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 927 MAIN_SECS(sbi), 0);
928 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 929 break;
930 }
931 secno = left_start;
932skip_left:
933 hint = secno;
934 segno = secno * sbi->segs_per_sec;
935 zoneno = secno / sbi->secs_per_zone;
936
937 /* give up on finding another zone */
938 if (!init)
939 goto got_it;
940 if (sbi->secs_per_zone == 1)
941 goto got_it;
942 if (zoneno == old_zoneno)
943 goto got_it;
944 if (dir == ALLOC_LEFT) {
945 if (!go_left && zoneno + 1 >= total_zones)
946 goto got_it;
947 if (go_left && zoneno == 0)
948 goto got_it;
949 }
950 for (i = 0; i < NR_CURSEG_TYPE; i++)
951 if (CURSEG_I(sbi, i)->zone == zoneno)
952 break;
953
954 if (i < NR_CURSEG_TYPE) {
955 /* zone is in user, try another */
956 if (go_left)
957 hint = zoneno * sbi->secs_per_zone - 1;
958 else if (zoneno + 1 >= total_zones)
959 hint = 0;
960 else
961 hint = (zoneno + 1) * sbi->secs_per_zone;
962 init = false;
963 goto find_other_zone;
964 }
965got_it:
966 /* set it as dirty segment in free segmap */
9850cf4a 967 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 968 __set_inuse(sbi, segno);
969 *newseg = segno;
1a118ccf 970 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 971}
972
973static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
974{
975 struct curseg_info *curseg = CURSEG_I(sbi, type);
976 struct summary_footer *sum_footer;
977
978 curseg->segno = curseg->next_segno;
979 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
980 curseg->next_blkoff = 0;
981 curseg->next_segno = NULL_SEGNO;
982
983 sum_footer = &(curseg->sum_blk->footer);
984 memset(sum_footer, 0, sizeof(struct summary_footer));
985 if (IS_DATASEG(type))
986 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
987 if (IS_NODESEG(type))
988 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
989 __set_sit_entry_type(sbi, type, curseg->segno, modified);
990}
991
0a8165d7 992/*
351df4b2
JK		 993 * Allocate a current working segment.
994 * This function always allocates a free segment in LFS manner.
995 */
996static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
997{
998 struct curseg_info *curseg = CURSEG_I(sbi, type);
999 unsigned int segno = curseg->segno;
1000 int dir = ALLOC_LEFT;
1001
1002 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1003 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1004 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1005 dir = ALLOC_RIGHT;
1006
1007 if (test_opt(sbi, NOHEAP))
1008 dir = ALLOC_RIGHT;
1009
1010 get_new_segment(sbi, &segno, new_sec, dir);
1011 curseg->next_segno = segno;
1012 reset_curseg(sbi, type, 1);
1013 curseg->alloc_type = LFS;
1014}
1015
1016static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1017 struct curseg_info *seg, block_t start)
1018{
1019 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1020 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1021 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1022 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1023 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1024 int i, pos;
1025
1026 for (i = 0; i < entries; i++)
1027 target_map[i] = ckpt_map[i] | cur_map[i];
1028
1029 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1030
1031 seg->next_blkoff = pos;
351df4b2
JK		 1032}
1033
0a8165d7 1034/*
351df4b2
JK		 1035 * If a segment is written by LFS manner, next block offset is just obtained
1036 * by increasing the current block offset. However, if a segment is written by
1037 * SSR manner, next block offset obtained by calling __next_free_blkoff
1038 */
1039static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1040 struct curseg_info *seg)
1041{
1042 if (seg->alloc_type == SSR)
1043 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1044 else
1045 seg->next_blkoff++;
1046}
1047
0a8165d7 1048/*
e1c42045 1049 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1050 * manner, so it should recover the existing segment information of valid blocks
1051 */
1052static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1053{
1054 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1055 struct curseg_info *curseg = CURSEG_I(sbi, type);
1056 unsigned int new_segno = curseg->next_segno;
1057 struct f2fs_summary_block *sum_node;
1058 struct page *sum_page;
1059
1060 write_sum_page(sbi, curseg->sum_blk,
1061 GET_SUM_BLOCK(sbi, curseg->segno));
1062 __set_test_and_inuse(sbi, new_segno);
1063
1064 mutex_lock(&dirty_i->seglist_lock);
1065 __remove_dirty_segment(sbi, new_segno, PRE);
1066 __remove_dirty_segment(sbi, new_segno, DIRTY);
1067 mutex_unlock(&dirty_i->seglist_lock);
1068
1069 reset_curseg(sbi, type, 1);
1070 curseg->alloc_type = SSR;
1071 __next_free_blkoff(sbi, curseg, 0);
1072
1073 if (reuse) {
1074 sum_page = get_sum_page(sbi, new_segno);
1075 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1076 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1077 f2fs_put_page(sum_page, 1);
1078 }
1079}
1080
43727527
JK		 1081static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1082{
1083 struct curseg_info *curseg = CURSEG_I(sbi, type);
1084 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1085
1086 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1087 return v_ops->get_victim(sbi,
1088 &(curseg)->next_segno, BG_GC, type, SSR);
1089
1090 /* For data segments, let's do SSR more intensively */
1091 for (; type >= CURSEG_HOT_DATA; type--)
1092 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1093 BG_GC, type, SSR))
1094 return 1;
1095 return 0;
1096}
1097
351df4b2
JK		 1098/*
1099 * flush out current segment and replace it with new segment
1100 * This function should be returned with success, otherwise BUG
1101 */
1102static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1103 int type, bool force)
1104{
1105 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1106
7b405275 1107 if (force)
351df4b2 1108 new_curseg(sbi, type, true);
7b405275 1109 else if (type == CURSEG_WARM_NODE)
351df4b2 1110 new_curseg(sbi, type, false);
60374688
JK		 1111 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1112 new_curseg(sbi, type, false);
351df4b2
JK		 1113 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1114 change_curseg(sbi, type, true);
1115 else
1116 new_curseg(sbi, type, false);
dcdfff65
JK		 1117
1118 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1119}
1120
38aa0889
JK		 1121static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1122{
1123 struct curseg_info *curseg = CURSEG_I(sbi, type);
1124 unsigned int old_segno;
1125
1126 old_segno = curseg->segno;
1127 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1128 locate_dirty_segment(sbi, old_segno);
1129}
1130
351df4b2
JK		 1131void allocate_new_segments(struct f2fs_sb_info *sbi)
1132{
351df4b2
JK		 1133 int i;
1134
38aa0889
JK		 1135 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1136 __allocate_new_segments(sbi, i);
351df4b2
JK		 1137}
1138
1139static const struct segment_allocation default_salloc_ops = {
1140 .allocate_segment = allocate_segment_by_default,
1141};
1142
4b2fecc8
JK		 1143int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1144{
f7ef9b83
JK		 1145 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1146 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1147 unsigned int start_segno, end_segno;
1148 struct cp_control cpc;
c34f42e2 1149 int err = 0;
4b2fecc8 1150
836b5a63 1151 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1152 return -EINVAL;
1153
9bd27ae4 1154 cpc.trimmed = 0;
7cd8558b 1155 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1156 goto out;
1157
1158 /* start/end segment number in main_area */
7cd8558b
JK		 1159 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1160 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1161 GET_SEGNO(sbi, end);
4b2fecc8 1162 cpc.reason = CP_DISCARD;
836b5a63 1163 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1164
1165 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1166 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1167 cpc.trim_start = start_segno;
a66cdd98
JK		 1168
1169 if (sbi->discard_blks == 0)
1170 break;
1171 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1172 cpc.trim_end = end_segno;
1173 else
1174 cpc.trim_end = min_t(unsigned int,
1175 rounddown(start_segno +
bba681cb
JK		 1176 BATCHED_TRIM_SEGMENTS(sbi),
1177 sbi->segs_per_sec) - 1, end_segno);
1178
1179 mutex_lock(&sbi->gc_mutex);
c34f42e2 1180 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1181 mutex_unlock(&sbi->gc_mutex);
1182 }
4b2fecc8 1183out:
f7ef9b83 1184 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1185 return err;
4b2fecc8
JK		 1186}
1187
351df4b2
JK		 1188static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1189{
1190 struct curseg_info *curseg = CURSEG_I(sbi, type);
1191 if (curseg->next_blkoff < sbi->blocks_per_seg)
1192 return true;
1193 return false;
1194}
1195
1196static int __get_segment_type_2(struct page *page, enum page_type p_type)
1197{
1198 if (p_type == DATA)
1199 return CURSEG_HOT_DATA;
1200 else
1201 return CURSEG_HOT_NODE;
1202}
1203
1204static int __get_segment_type_4(struct page *page, enum page_type p_type)
1205{
1206 if (p_type == DATA) {
1207 struct inode *inode = page->mapping->host;
1208
1209 if (S_ISDIR(inode->i_mode))
1210 return CURSEG_HOT_DATA;
1211 else
1212 return CURSEG_COLD_DATA;
1213 } else {
a344b9fd
JK		 1214 if (IS_DNODE(page) && is_cold_node(page))
1215 return CURSEG_WARM_NODE;
351df4b2
JK		 1216 else
1217 return CURSEG_COLD_NODE;
1218 }
1219}
1220
1221static int __get_segment_type_6(struct page *page, enum page_type p_type)
1222{
1223 if (p_type == DATA) {
1224 struct inode *inode = page->mapping->host;
1225
1226 if (S_ISDIR(inode->i_mode))
1227 return CURSEG_HOT_DATA;
354a3399 1228 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1229 return CURSEG_COLD_DATA;
1230 else
1231 return CURSEG_WARM_DATA;
1232 } else {
1233 if (IS_DNODE(page))
1234 return is_cold_node(page) ? CURSEG_WARM_NODE :
1235 CURSEG_HOT_NODE;
1236 else
1237 return CURSEG_COLD_NODE;
1238 }
1239}
1240
1241static int __get_segment_type(struct page *page, enum page_type p_type)
1242{
4081363f 1243 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1244 case 2:
1245 return __get_segment_type_2(page, p_type);
1246 case 4:
1247 return __get_segment_type_4(page, p_type);
351df4b2 1248 }
12a67146 1249 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1250 f2fs_bug_on(F2FS_P_SB(page),
1251 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1252 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1253}
1254
bfad7c2d
JK		 1255void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1256 block_t old_blkaddr, block_t *new_blkaddr,
1257 struct f2fs_summary *sum, int type)
351df4b2
JK		 1258{
1259 struct sit_info *sit_i = SIT_I(sbi);
1260 struct curseg_info *curseg;
38aa0889
JK		 1261 bool direct_io = (type == CURSEG_DIRECT_IO);
1262
1263 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1264
351df4b2
JK		 1265 curseg = CURSEG_I(sbi, type);
1266
1267 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1268 mutex_lock(&sit_i->sentry_lock);
351df4b2 1269
38aa0889 1270 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1271 if (direct_io && curseg->next_blkoff &&
1272 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1273 __allocate_new_segments(sbi, type);
1274
351df4b2 1275 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1276
1277 /*
1278 * __add_sum_entry should be resided under the curseg_mutex
1279 * because, this function updates a summary entry in the
1280 * current summary block.
1281 */
e79efe3b 1282 __add_sum_entry(sbi, type, sum);
351df4b2 1283
351df4b2 1284 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1285
1286 stat_inc_block_count(sbi, curseg);
351df4b2 1287
5e443818
JK		 1288 if (!__has_curseg_space(sbi, type))
1289 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1290 /*
1291 * SIT information should be updated before segment allocation,
1292 * since SSR needs latest valid block information.
1293 */
1294 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1295
351df4b2
JK		 1296 mutex_unlock(&sit_i->sentry_lock);
1297
bfad7c2d 1298 if (page && IS_NODESEG(type))
351df4b2
JK		 1299 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1300
bfad7c2d
JK		 1301 mutex_unlock(&curseg->curseg_mutex);
1302}
1303
05ca3632 1304static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1305{
05ca3632 1306 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1307
05ca3632
JK		 1308 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1309 &fio->blk_addr, sum, type);
bfad7c2d 1310
351df4b2 1311 /* writeout dirty page into bdev */
05ca3632 1312 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1313}
1314
577e3495 1315void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1316{
458e6197 1317 struct f2fs_io_info fio = {
05ca3632 1318 .sbi = sbi,
458e6197 1319 .type = META,
cf04e8eb
JK		 1320 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1321 .blk_addr = page->index,
05ca3632 1322 .page = page,
4375a336 1323 .encrypted_page = NULL,
458e6197
JK		 1324 };
1325
2b947003
CY		 1326 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1327 fio.rw &= ~REQ_META;
1328
351df4b2 1329 set_page_writeback(page);
05ca3632 1330 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1331}
1332
05ca3632 1333void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1334{
1335 struct f2fs_summary sum;
05ca3632 1336
351df4b2 1337 set_summary(&sum, nid, 0, 0);
05ca3632 1338 do_write_page(&sum, fio);
351df4b2
JK		 1339}
1340
05ca3632 1341void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1342{
05ca3632 1343 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1344 struct f2fs_summary sum;
1345 struct node_info ni;
1346
9850cf4a 1347 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1348 get_node_info(sbi, dn->nid, &ni);
1349 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1350 do_write_page(&sum, fio);
e1509cf2 1351 dn->data_blkaddr = fio->blk_addr;
351df4b2
JK		 1352}
1353
05ca3632 1354void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1355{
05ca3632
JK		 1356 stat_inc_inplace_blocks(fio->sbi);
1357 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1358}
1359
528e3459
CY		 1360static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1361 struct f2fs_summary *sum,
19f106bc
CY		 1362 block_t old_blkaddr, block_t new_blkaddr,
1363 bool recover_curseg)
351df4b2
JK		 1364{
1365 struct sit_info *sit_i = SIT_I(sbi);
1366 struct curseg_info *curseg;
1367 unsigned int segno, old_cursegno;
1368 struct seg_entry *se;
1369 int type;
19f106bc 1370 unsigned short old_blkoff;
351df4b2
JK		 1371
1372 segno = GET_SEGNO(sbi, new_blkaddr);
1373 se = get_seg_entry(sbi, segno);
1374 type = se->type;
1375
19f106bc
CY		 1376 if (!recover_curseg) {
1377 /* for recovery flow */
1378 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1379 if (old_blkaddr == NULL_ADDR)
1380 type = CURSEG_COLD_DATA;
1381 else
1382 type = CURSEG_WARM_DATA;
1383 }
1384 } else {
1385 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1386 type = CURSEG_WARM_DATA;
1387 }
19f106bc 1388
351df4b2
JK		 1389 curseg = CURSEG_I(sbi, type);
1390
1391 mutex_lock(&curseg->curseg_mutex);
1392 mutex_lock(&sit_i->sentry_lock);
1393
1394 old_cursegno = curseg->segno;
19f106bc 1395 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1396
1397 /* change the current segment */
1398 if (segno != curseg->segno) {
1399 curseg->next_segno = segno;
1400 change_curseg(sbi, type, true);
1401 }
1402
491c0854 1403 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1404 __add_sum_entry(sbi, type, sum);
351df4b2 1405
6e2c64ad
JK		 1406 if (!recover_curseg)
1407 update_sit_entry(sbi, new_blkaddr, 1);
1408 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1409 update_sit_entry(sbi, old_blkaddr, -1);
1410
1411 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1412 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1413
351df4b2 1414 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1415
19f106bc
CY		 1416 if (recover_curseg) {
1417 if (old_cursegno != curseg->segno) {
1418 curseg->next_segno = old_cursegno;
1419 change_curseg(sbi, type, true);
1420 }
1421 curseg->next_blkoff = old_blkoff;
1422 }
1423
351df4b2
JK		 1424 mutex_unlock(&sit_i->sentry_lock);
1425 mutex_unlock(&curseg->curseg_mutex);
1426}
1427
528e3459
CY		 1428void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1429 block_t old_addr, block_t new_addr,
1430 unsigned char version, bool recover_curseg)
1431{
1432 struct f2fs_summary sum;
1433
1434 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1435
1436 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1437
1438 dn->data_blkaddr = new_addr;
1439 set_data_blkaddr(dn);
1440 f2fs_update_extent_cache(dn);
1441}
1442
93dfe2ac 1443void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1444 enum page_type type, bool ordered)
93dfe2ac 1445{
93dfe2ac 1446 if (PageWriteback(page)) {
4081363f
JK		 1447 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1448
0c3a5797 1449 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1450 if (ordered)
1451 wait_on_page_writeback(page);
1452 else
1453 wait_for_stable_page(page);
93dfe2ac
JK		 1454 }
1455}
1456
08b39fbd
CY		 1457void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1458 block_t blkaddr)
1459{
1460 struct page *cpage;
1461
1462 if (blkaddr == NEW_ADDR)
1463 return;
1464
1465 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1466
1467 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1468 if (cpage) {
fec1d657 1469 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1470 f2fs_put_page(cpage, 1);
1471 }
1472}
1473
351df4b2
JK		 1474static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1475{
1476 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1477 struct curseg_info *seg_i;
1478 unsigned char *kaddr;
1479 struct page *page;
1480 block_t start;
1481 int i, j, offset;
1482
1483 start = start_sum_block(sbi);
1484
1485 page = get_meta_page(sbi, start++);
1486 kaddr = (unsigned char *)page_address(page);
1487
1488 /* Step 1: restore nat cache */
1489 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1490 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1491
1492 /* Step 2: restore sit cache */
1493 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1494 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1495 SUM_JOURNAL_SIZE);
1496 offset = 2 * SUM_JOURNAL_SIZE;
1497
1498 /* Step 3: restore summary entries */
1499 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1500 unsigned short blk_off;
1501 unsigned int segno;
1502
1503 seg_i = CURSEG_I(sbi, i);
1504 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1505 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1506 seg_i->next_segno = segno;
1507 reset_curseg(sbi, i, 0);
1508 seg_i->alloc_type = ckpt->alloc_type[i];
1509 seg_i->next_blkoff = blk_off;
1510
1511 if (seg_i->alloc_type == SSR)
1512 blk_off = sbi->blocks_per_seg;
1513
1514 for (j = 0; j < blk_off; j++) {
1515 struct f2fs_summary *s;
1516 s = (struct f2fs_summary *)(kaddr + offset);
1517 seg_i->sum_blk->entries[j] = *s;
1518 offset += SUMMARY_SIZE;
1519 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1520 SUM_FOOTER_SIZE)
1521 continue;
1522
1523 f2fs_put_page(page, 1);
1524 page = NULL;
1525
1526 page = get_meta_page(sbi, start++);
1527 kaddr = (unsigned char *)page_address(page);
1528 offset = 0;
1529 }
1530 }
1531 f2fs_put_page(page, 1);
1532 return 0;
1533}
1534
1535static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1536{
1537 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1538 struct f2fs_summary_block *sum;
1539 struct curseg_info *curseg;
1540 struct page *new;
1541 unsigned short blk_off;
1542 unsigned int segno = 0;
1543 block_t blk_addr = 0;
1544
1545 /* get segment number and block addr */
1546 if (IS_DATASEG(type)) {
1547 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1548 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1549 CURSEG_HOT_DATA]);
119ee914 1550 if (__exist_node_summaries(sbi))
351df4b2
JK		 1551 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1552 else
1553 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1554 } else {
1555 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1556 CURSEG_HOT_NODE]);
1557 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1558 CURSEG_HOT_NODE]);
119ee914 1559 if (__exist_node_summaries(sbi))
351df4b2
JK		 1560 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1561 type - CURSEG_HOT_NODE);
1562 else
1563 blk_addr = GET_SUM_BLOCK(sbi, segno);
1564 }
1565
1566 new = get_meta_page(sbi, blk_addr);
1567 sum = (struct f2fs_summary_block *)page_address(new);
1568
1569 if (IS_NODESEG(type)) {
119ee914 1570 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1571 struct f2fs_summary *ns = &sum->entries[0];
1572 int i;
1573 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1574 ns->version = 0;
1575 ns->ofs_in_node = 0;
1576 }
1577 } else {
d653788a
GZ		 1578 int err;
1579
1580 err = restore_node_summary(sbi, segno, sum);
1581 if (err) {
351df4b2 1582 f2fs_put_page(new, 1);
d653788a 1583 return err;
351df4b2
JK		 1584 }
1585 }
1586 }
1587
1588 /* set uncompleted segment to curseg */
1589 curseg = CURSEG_I(sbi, type);
1590 mutex_lock(&curseg->curseg_mutex);
1591 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1592 curseg->next_segno = segno;
1593 reset_curseg(sbi, type, 0);
1594 curseg->alloc_type = ckpt->alloc_type[type];
1595 curseg->next_blkoff = blk_off;
1596 mutex_unlock(&curseg->curseg_mutex);
1597 f2fs_put_page(new, 1);
1598 return 0;
1599}
1600
1601static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1602{
1603 int type = CURSEG_HOT_DATA;
e4fc5fbf 1604 int err;
351df4b2 1605
25ca923b 1606 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1607 int npages = npages_for_summary_flush(sbi, true);
1608
1609 if (npages >= 2)
1610 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1611 META_CP, true);
3fa06d7b 1612
351df4b2
JK		 1613 /* restore for compacted data summary */
1614 if (read_compacted_summaries(sbi))
1615 return -EINVAL;
1616 type = CURSEG_HOT_NODE;
1617 }
1618
119ee914 1619 if (__exist_node_summaries(sbi))
3fa06d7b 1620 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1621 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1622
e4fc5fbf
CY		 1623 for (; type <= CURSEG_COLD_NODE; type++) {
1624 err = read_normal_summaries(sbi, type);
1625 if (err)
1626 return err;
1627 }
1628
351df4b2
JK		 1629 return 0;
1630}
1631
1632static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1633{
1634 struct page *page;
1635 unsigned char *kaddr;
1636 struct f2fs_summary *summary;
1637 struct curseg_info *seg_i;
1638 int written_size = 0;
1639 int i, j;
1640
1641 page = grab_meta_page(sbi, blkaddr++);
1642 kaddr = (unsigned char *)page_address(page);
1643
1644 /* Step 1: write nat cache */
1645 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1646 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1647 written_size += SUM_JOURNAL_SIZE;
1648
1649 /* Step 2: write sit cache */
1650 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1651 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1652 SUM_JOURNAL_SIZE);
1653 written_size += SUM_JOURNAL_SIZE;
1654
351df4b2
JK		 1655 /* Step 3: write summary entries */
1656 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1657 unsigned short blkoff;
1658 seg_i = CURSEG_I(sbi, i);
1659 if (sbi->ckpt->alloc_type[i] == SSR)
1660 blkoff = sbi->blocks_per_seg;
1661 else
1662 blkoff = curseg_blkoff(sbi, i);
1663
1664 for (j = 0; j < blkoff; j++) {
1665 if (!page) {
1666 page = grab_meta_page(sbi, blkaddr++);
1667 kaddr = (unsigned char *)page_address(page);
1668 written_size = 0;
1669 }
1670 summary = (struct f2fs_summary *)(kaddr + written_size);
1671 *summary = seg_i->sum_blk->entries[j];
1672 written_size += SUMMARY_SIZE;
351df4b2
JK		 1673
1674 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1675 SUM_FOOTER_SIZE)
1676 continue;
1677
e8d61a74 1678 set_page_dirty(page);
351df4b2
JK		 1679 f2fs_put_page(page, 1);
1680 page = NULL;
1681 }
1682 }
e8d61a74
CY		 1683 if (page) {
1684 set_page_dirty(page);
351df4b2 1685 f2fs_put_page(page, 1);
e8d61a74 1686 }
351df4b2
JK		 1687}
1688
1689static void write_normal_summaries(struct f2fs_sb_info *sbi,
1690 block_t blkaddr, int type)
1691{
1692 int i, end;
1693 if (IS_DATASEG(type))
1694 end = type + NR_CURSEG_DATA_TYPE;
1695 else
1696 end = type + NR_CURSEG_NODE_TYPE;
1697
1698 for (i = type; i < end; i++) {
1699 struct curseg_info *sum = CURSEG_I(sbi, i);
1700 mutex_lock(&sum->curseg_mutex);
1701 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1702 mutex_unlock(&sum->curseg_mutex);
1703 }
1704}
1705
1706void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1707{
25ca923b 1708 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1709 write_compacted_summaries(sbi, start_blk);
1710 else
1711 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1712}
1713
1714void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1715{
119ee914 1716 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1717}
1718
1719int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1720 unsigned int val, int alloc)
1721{
1722 int i;
1723
1724 if (type == NAT_JOURNAL) {
1725 for (i = 0; i < nats_in_cursum(sum); i++) {
1726 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1727 return i;
1728 }
855639de 1729 if (alloc && __has_cursum_space(sum, 1, NAT_JOURNAL))
351df4b2
JK		 1730 return update_nats_in_cursum(sum, 1);
1731 } else if (type == SIT_JOURNAL) {
1732 for (i = 0; i < sits_in_cursum(sum); i++)
1733 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1734 return i;
855639de 1735 if (alloc && __has_cursum_space(sum, 1, SIT_JOURNAL))
351df4b2
JK		 1736 return update_sits_in_cursum(sum, 1);
1737 }
1738 return -1;
1739}
1740
1741static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1742 unsigned int segno)
1743{
2cc22186 1744 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1745}
1746
1747static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1748 unsigned int start)
1749{
1750 struct sit_info *sit_i = SIT_I(sbi);
1751 struct page *src_page, *dst_page;
1752 pgoff_t src_off, dst_off;
1753 void *src_addr, *dst_addr;
1754
1755 src_off = current_sit_addr(sbi, start);
1756 dst_off = next_sit_addr(sbi, src_off);
1757
1758 /* get current sit block page without lock */
1759 src_page = get_meta_page(sbi, src_off);
1760 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1761 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1762
1763 src_addr = page_address(src_page);
1764 dst_addr = page_address(dst_page);
1765 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1766
1767 set_page_dirty(dst_page);
1768 f2fs_put_page(src_page, 1);
1769
1770 set_to_next_sit(sit_i, start);
1771
1772 return dst_page;
1773}
1774
184a5cd2
CY		 1775static struct sit_entry_set *grab_sit_entry_set(void)
1776{
1777 struct sit_entry_set *ses =
80c54505 1778 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1779
1780 ses->entry_cnt = 0;
1781 INIT_LIST_HEAD(&ses->set_list);
1782 return ses;
1783}
1784
1785static void release_sit_entry_set(struct sit_entry_set *ses)
1786{
1787 list_del(&ses->set_list);
1788 kmem_cache_free(sit_entry_set_slab, ses);
1789}
1790
1791static void adjust_sit_entry_set(struct sit_entry_set *ses,
1792 struct list_head *head)
1793{
1794 struct sit_entry_set *next = ses;
1795
1796 if (list_is_last(&ses->set_list, head))
1797 return;
1798
1799 list_for_each_entry_continue(next, head, set_list)
1800 if (ses->entry_cnt <= next->entry_cnt)
1801 break;
1802
1803 list_move_tail(&ses->set_list, &next->set_list);
1804}
1805
1806static void add_sit_entry(unsigned int segno, struct list_head *head)
1807{
1808 struct sit_entry_set *ses;
1809 unsigned int start_segno = START_SEGNO(segno);
1810
1811 list_for_each_entry(ses, head, set_list) {
1812 if (ses->start_segno == start_segno) {
1813 ses->entry_cnt++;
1814 adjust_sit_entry_set(ses, head);
1815 return;
1816 }
1817 }
1818
1819 ses = grab_sit_entry_set();
1820
1821 ses->start_segno = start_segno;
1822 ses->entry_cnt++;
1823 list_add(&ses->set_list, head);
1824}
1825
1826static void add_sits_in_set(struct f2fs_sb_info *sbi)
1827{
1828 struct f2fs_sm_info *sm_info = SM_I(sbi);
1829 struct list_head *set_list = &sm_info->sit_entry_set;
1830 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1831 unsigned int segno;
1832
7cd8558b 1833 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1834 add_sit_entry(segno, set_list);
1835}
1836
1837static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1838{
1839 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1840 struct f2fs_summary_block *sum = curseg->sum_blk;
1841 int i;
1842
184a5cd2
CY		 1843 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1844 unsigned int segno;
1845 bool dirtied;
1846
1847 segno = le32_to_cpu(segno_in_journal(sum, i));
1848 dirtied = __mark_sit_entry_dirty(sbi, segno);
1849
1850 if (!dirtied)
1851 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1852 }
184a5cd2 1853 update_sits_in_cursum(sum, -sits_in_cursum(sum));
351df4b2
JK		 1854}
1855
0a8165d7 1856/*
351df4b2
JK		 1857 * CP calls this function, which flushes SIT entries including sit_journal,
1858 * and moves prefree segs to free segs.
1859 */
4b2fecc8 1860void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1861{
1862 struct sit_info *sit_i = SIT_I(sbi);
1863 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1864 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1865 struct f2fs_summary_block *sum = curseg->sum_blk;
184a5cd2
CY		 1866 struct sit_entry_set *ses, *tmp;
1867 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1868 bool to_journal = true;
4b2fecc8 1869 struct seg_entry *se;
351df4b2
JK		 1870
1871 mutex_lock(&curseg->curseg_mutex);
1872 mutex_lock(&sit_i->sentry_lock);
1873
2b11a74b
WL		 1874 if (!sit_i->dirty_sentries)
1875 goto out;
1876
351df4b2 1877 /*
184a5cd2
CY		 1878 * add and account sit entries of dirty bitmap in sit entry
1879 * set temporarily
351df4b2 1880 */
184a5cd2 1881 add_sits_in_set(sbi);
351df4b2 1882
184a5cd2
CY		 1883 /*
1884 * if there are no enough space in journal to store dirty sit
1885 * entries, remove all entries from journal and add and account
1886 * them in sit entry set.
1887 */
1888 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1889 remove_sits_in_journal(sbi);
b2955550 1890
184a5cd2
CY		 1891 /*
1892 * there are two steps to flush sit entries:
1893 * #1, flush sit entries to journal in current cold data summary block.
1894 * #2, flush sit entries to sit page.
1895 */
1896 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1897 struct page *page = NULL;
184a5cd2
CY		 1898 struct f2fs_sit_block *raw_sit = NULL;
1899 unsigned int start_segno = ses->start_segno;
1900 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1901 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1902 unsigned int segno = start_segno;
1903
1904 if (to_journal &&
1905 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1906 to_journal = false;
1907
1908 if (!to_journal) {
1909 page = get_next_sit_page(sbi, start_segno);
1910 raw_sit = page_address(page);
351df4b2 1911 }
351df4b2 1912
184a5cd2
CY		 1913 /* flush dirty sit entries in region of current sit set */
1914 for_each_set_bit_from(segno, bitmap, end) {
1915 int offset, sit_offset;
4b2fecc8
JK		 1916
1917 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 1918
1919 /* add discard candidates */
d7bc2484 1920 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 1921 cpc->trim_start = segno;
1922 add_discard_addrs(sbi, cpc);
1923 }
184a5cd2
CY		 1924
1925 if (to_journal) {
1926 offset = lookup_journal_in_cursum(sum,
1927 SIT_JOURNAL, segno, 1);
1928 f2fs_bug_on(sbi, offset < 0);
1929 segno_in_journal(sum, offset) =
1930 cpu_to_le32(segno);
1931 seg_info_to_raw_sit(se,
1932 &sit_in_journal(sum, offset));
1933 } else {
1934 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1935 seg_info_to_raw_sit(se,
1936 &raw_sit->entries[sit_offset]);
1937 }
351df4b2 1938
184a5cd2
CY		 1939 __clear_bit(segno, bitmap);
1940 sit_i->dirty_sentries--;
1941 ses->entry_cnt--;
351df4b2
JK		 1942 }
1943
184a5cd2
CY		 1944 if (!to_journal)
1945 f2fs_put_page(page, 1);
1946
1947 f2fs_bug_on(sbi, ses->entry_cnt);
1948 release_sit_entry_set(ses);
351df4b2 1949 }
184a5cd2
CY		 1950
1951 f2fs_bug_on(sbi, !list_empty(head));
1952 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 1953out:
4b2fecc8
JK		 1954 if (cpc->reason == CP_DISCARD) {
1955 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1956 add_discard_addrs(sbi, cpc);
1957 }
351df4b2
JK		 1958 mutex_unlock(&sit_i->sentry_lock);
1959 mutex_unlock(&curseg->curseg_mutex);
1960
351df4b2
JK		 1961 set_prefree_as_free_segments(sbi);
1962}
1963
1964static int build_sit_info(struct f2fs_sb_info *sbi)
1965{
1966 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1967 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1968 struct sit_info *sit_i;
1969 unsigned int sit_segs, start;
1970 char *src_bitmap, *dst_bitmap;
1971 unsigned int bitmap_size;
1972
1973 /* allocate memory for SIT information */
1974 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1975 if (!sit_i)
1976 return -ENOMEM;
1977
1978 SM_I(sbi)->sit_info = sit_i;
1979
39307a8e
JK		 1980 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
1981 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 1982 if (!sit_i->sentries)
1983 return -ENOMEM;
1984
7cd8558b 1985 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 1986 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 1987 if (!sit_i->dirty_sentries_bitmap)
1988 return -ENOMEM;
1989
7cd8558b 1990 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 1991 sit_i->sentries[start].cur_valid_map
1992 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1993 sit_i->sentries[start].ckpt_valid_map
1994 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 1995 sit_i->sentries[start].discard_map
1996 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1997 if (!sit_i->sentries[start].cur_valid_map ||
1998 !sit_i->sentries[start].ckpt_valid_map ||
1999 !sit_i->sentries[start].discard_map)
351df4b2
JK		 2000 return -ENOMEM;
2001 }
2002
60a3b782
JK		 2003 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2004 if (!sit_i->tmp_map)
2005 return -ENOMEM;
2006
351df4b2 2007 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2008 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2009 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2010 if (!sit_i->sec_entries)
2011 return -ENOMEM;
2012 }
2013
2014 /* get information related with SIT */
2015 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2016
2017 /* setup SIT bitmap from ckeckpoint pack */
2018 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2019 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2020
79b5793b 2021 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2022 if (!dst_bitmap)
2023 return -ENOMEM;
351df4b2
JK		 2024
2025 /* init SIT information */
2026 sit_i->s_ops = &default_salloc_ops;
2027
2028 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2029 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2030 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2031 sit_i->sit_bitmap = dst_bitmap;
2032 sit_i->bitmap_size = bitmap_size;
2033 sit_i->dirty_sentries = 0;
2034 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2035 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2036 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2037 mutex_init(&sit_i->sentry_lock);
2038 return 0;
2039}
2040
2041static int build_free_segmap(struct f2fs_sb_info *sbi)
2042{
351df4b2
JK		 2043 struct free_segmap_info *free_i;
2044 unsigned int bitmap_size, sec_bitmap_size;
2045
2046 /* allocate memory for free segmap information */
2047 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2048 if (!free_i)
2049 return -ENOMEM;
2050
2051 SM_I(sbi)->free_info = free_i;
2052
7cd8558b 2053 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2054 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2055 if (!free_i->free_segmap)
2056 return -ENOMEM;
2057
7cd8558b 2058 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2059 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2060 if (!free_i->free_secmap)
2061 return -ENOMEM;
2062
2063 /* set all segments as dirty temporarily */
2064 memset(free_i->free_segmap, 0xff, bitmap_size);
2065 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2066
2067 /* init free segmap information */
7cd8558b 2068 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2069 free_i->free_segments = 0;
2070 free_i->free_sections = 0;
1a118ccf 2071 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2072 return 0;
2073}
2074
2075static int build_curseg(struct f2fs_sb_info *sbi)
2076{
1042d60f 2077 struct curseg_info *array;
351df4b2
JK		 2078 int i;
2079
b434babf 2080 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2081 if (!array)
2082 return -ENOMEM;
2083
2084 SM_I(sbi)->curseg_array = array;
2085
2086 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2087 mutex_init(&array[i].curseg_mutex);
2088 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2089 if (!array[i].sum_blk)
2090 return -ENOMEM;
2091 array[i].segno = NULL_SEGNO;
2092 array[i].next_blkoff = 0;
2093 }
2094 return restore_curseg_summaries(sbi);
2095}
2096
2097static void build_sit_entries(struct f2fs_sb_info *sbi)
2098{
2099 struct sit_info *sit_i = SIT_I(sbi);
2100 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2101 struct f2fs_summary_block *sum = curseg->sum_blk;
74de593a
CY		 2102 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2103 unsigned int i, start, end;
2104 unsigned int readed, start_blk = 0;
90a893c7 2105 int nrpages = MAX_BIO_BLOCKS(sbi);
351df4b2 2106
74de593a 2107 do {
26879fb1 2108 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2109
2110 start = start_blk * sit_i->sents_per_block;
2111 end = (start_blk + readed) * sit_i->sents_per_block;
2112
7cd8558b 2113 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2114 struct seg_entry *se = &sit_i->sentries[start];
2115 struct f2fs_sit_block *sit_blk;
2116 struct f2fs_sit_entry sit;
2117 struct page *page;
2118
2119 mutex_lock(&curseg->curseg_mutex);
2120 for (i = 0; i < sits_in_cursum(sum); i++) {
6c311ec6
CF		 2121 if (le32_to_cpu(segno_in_journal(sum, i))
2122 == start) {
74de593a
CY		 2123 sit = sit_in_journal(sum, i);
2124 mutex_unlock(&curseg->curseg_mutex);
2125 goto got_it;
2126 }
351df4b2 2127 }
74de593a
CY		 2128 mutex_unlock(&curseg->curseg_mutex);
2129
2130 page = get_current_sit_page(sbi, start);
2131 sit_blk = (struct f2fs_sit_block *)page_address(page);
2132 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2133 f2fs_put_page(page, 1);
351df4b2 2134got_it:
74de593a
CY		 2135 check_block_count(sbi, start, &sit);
2136 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2137
2138 /* build discard map only one time */
2139 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2140 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2141
74de593a
CY		 2142 if (sbi->segs_per_sec > 1) {
2143 struct sec_entry *e = get_sec_entry(sbi, start);
2144 e->valid_blocks += se->valid_blocks;
2145 }
351df4b2 2146 }
74de593a
CY		 2147 start_blk += readed;
2148 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2149}
2150
2151static void init_free_segmap(struct f2fs_sb_info *sbi)
2152{
2153 unsigned int start;
2154 int type;
2155
7cd8558b 2156 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2157 struct seg_entry *sentry = get_seg_entry(sbi, start);
2158 if (!sentry->valid_blocks)
2159 __set_free(sbi, start);
2160 }
2161
2162 /* set use the current segments */
2163 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2164 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2165 __set_test_and_inuse(sbi, curseg_t->segno);
2166 }
2167}
2168
2169static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2170{
2171 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2172 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2173 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2174 unsigned short valid_blocks;
2175
8736fbf0 2176 while (1) {
351df4b2 2177 /* find dirty segment based on free segmap */
7cd8558b
JK		 2178 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2179 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2180 break;
2181 offset = segno + 1;
2182 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2183 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2184 continue;
ec325b52
JK		 2185 if (valid_blocks > sbi->blocks_per_seg) {
2186 f2fs_bug_on(sbi, 1);
2187 continue;
2188 }
351df4b2
JK		 2189 mutex_lock(&dirty_i->seglist_lock);
2190 __locate_dirty_segment(sbi, segno, DIRTY);
2191 mutex_unlock(&dirty_i->seglist_lock);
2192 }
2193}
2194
5ec4e49f 2195static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2196{
2197 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2198 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2199
39307a8e 2200 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2201 if (!dirty_i->victim_secmap)
351df4b2
JK		 2202 return -ENOMEM;
2203 return 0;
2204}
2205
2206static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2207{
2208 struct dirty_seglist_info *dirty_i;
2209 unsigned int bitmap_size, i;
2210
2211 /* allocate memory for dirty segments list information */
2212 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2213 if (!dirty_i)
2214 return -ENOMEM;
2215
2216 SM_I(sbi)->dirty_info = dirty_i;
2217 mutex_init(&dirty_i->seglist_lock);
2218
7cd8558b 2219 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2220
2221 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2222 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2223 if (!dirty_i->dirty_segmap[i])
2224 return -ENOMEM;
2225 }
2226
2227 init_dirty_segmap(sbi);
5ec4e49f 2228 return init_victim_secmap(sbi);
351df4b2
JK		 2229}
2230
0a8165d7 2231/*
351df4b2
JK		 2232 * Update min, max modified time for cost-benefit GC algorithm
2233 */
2234static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2235{
2236 struct sit_info *sit_i = SIT_I(sbi);
2237 unsigned int segno;
2238
2239 mutex_lock(&sit_i->sentry_lock);
2240
2241 sit_i->min_mtime = LLONG_MAX;
2242
7cd8558b 2243 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2244 unsigned int i;
2245 unsigned long long mtime = 0;
2246
2247 for (i = 0; i < sbi->segs_per_sec; i++)
2248 mtime += get_seg_entry(sbi, segno + i)->mtime;
2249
2250 mtime = div_u64(mtime, sbi->segs_per_sec);
2251
2252 if (sit_i->min_mtime > mtime)
2253 sit_i->min_mtime = mtime;
2254 }
2255 sit_i->max_mtime = get_mtime(sbi);
2256 mutex_unlock(&sit_i->sentry_lock);
2257}
2258
2259int build_segment_manager(struct f2fs_sb_info *sbi)
2260{
2261 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2262 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2263 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2264 int err;
2265
2266 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2267 if (!sm_info)
2268 return -ENOMEM;
2269
2270 /* init sm info */
2271 sbi->sm_info = sm_info;
351df4b2
JK		 2272 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2273 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2274 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2275 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2276 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2277 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2278 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2279 sm_info->rec_prefree_segments = sm_info->main_segments *
2280 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2281 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2282 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2283 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2284
7fd9e544
JK		 2285 INIT_LIST_HEAD(&sm_info->discard_list);
2286 sm_info->nr_discards = 0;
2287 sm_info->max_discards = 0;
2288
bba681cb
JK		 2289 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2290
184a5cd2
CY		 2291 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2292
b270ad6f 2293 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2294 err = create_flush_cmd_control(sbi);
2295 if (err)
a688b9d9 2296 return err;
6b4afdd7
JK		 2297 }
2298
351df4b2
JK		 2299 err = build_sit_info(sbi);
2300 if (err)
2301 return err;
2302 err = build_free_segmap(sbi);
2303 if (err)
2304 return err;
2305 err = build_curseg(sbi);
2306 if (err)
2307 return err;
2308
2309 /* reinit free segmap based on SIT */
2310 build_sit_entries(sbi);
2311
2312 init_free_segmap(sbi);
2313 err = build_dirty_segmap(sbi);
2314 if (err)
2315 return err;
2316
2317 init_min_max_mtime(sbi);
2318 return 0;
2319}
2320
2321static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2322 enum dirty_type dirty_type)
2323{
2324 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2325
2326 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2327 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2328 dirty_i->nr_dirty[dirty_type] = 0;
2329 mutex_unlock(&dirty_i->seglist_lock);
2330}
2331
5ec4e49f 2332static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2333{
2334 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2335 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2336}
2337
2338static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2339{
2340 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2341 int i;
2342
2343 if (!dirty_i)
2344 return;
2345
2346 /* discard pre-free/dirty segments list */
2347 for (i = 0; i < NR_DIRTY_TYPE; i++)
2348 discard_dirty_segmap(sbi, i);
2349
5ec4e49f 2350 destroy_victim_secmap(sbi);
351df4b2
JK		 2351 SM_I(sbi)->dirty_info = NULL;
2352 kfree(dirty_i);
2353}
2354
2355static void destroy_curseg(struct f2fs_sb_info *sbi)
2356{
2357 struct curseg_info *array = SM_I(sbi)->curseg_array;
2358 int i;
2359
2360 if (!array)
2361 return;
2362 SM_I(sbi)->curseg_array = NULL;
2363 for (i = 0; i < NR_CURSEG_TYPE; i++)
2364 kfree(array[i].sum_blk);
2365 kfree(array);
2366}
2367
2368static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2369{
2370 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2371 if (!free_i)
2372 return;
2373 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2374 kvfree(free_i->free_segmap);
2375 kvfree(free_i->free_secmap);
351df4b2
JK		 2376 kfree(free_i);
2377}
2378
2379static void destroy_sit_info(struct f2fs_sb_info *sbi)
2380{
2381 struct sit_info *sit_i = SIT_I(sbi);
2382 unsigned int start;
2383
2384 if (!sit_i)
2385 return;
2386
2387 if (sit_i->sentries) {
7cd8558b 2388 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2389 kfree(sit_i->sentries[start].cur_valid_map);
2390 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2391 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2392 }
2393 }
60a3b782
JK		 2394 kfree(sit_i->tmp_map);
2395
39307a8e
JK		 2396 kvfree(sit_i->sentries);
2397 kvfree(sit_i->sec_entries);
2398 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2399
2400 SM_I(sbi)->sit_info = NULL;
2401 kfree(sit_i->sit_bitmap);
2402 kfree(sit_i);
2403}
2404
2405void destroy_segment_manager(struct f2fs_sb_info *sbi)
2406{
2407 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2408
3b03f724
CY		 2409 if (!sm_info)
2410 return;
2163d198 2411 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2412 destroy_dirty_segmap(sbi);
2413 destroy_curseg(sbi);
2414 destroy_free_segmap(sbi);
2415 destroy_sit_info(sbi);
2416 sbi->sm_info = NULL;
2417 kfree(sm_info);
2418}
7fd9e544
JK		 2419
2420int __init create_segment_manager_caches(void)
2421{
2422 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2423 sizeof(struct discard_entry));
7fd9e544 2424 if (!discard_entry_slab)
184a5cd2
CY		 2425 goto fail;
2426
2427 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2428 sizeof(struct sit_entry_set));
184a5cd2
CY		 2429 if (!sit_entry_set_slab)
2430 goto destory_discard_entry;
88b88a66
JK		 2431
2432 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2433 sizeof(struct inmem_pages));
2434 if (!inmem_entry_slab)
2435 goto destroy_sit_entry_set;
7fd9e544 2436 return 0;
184a5cd2 2437
88b88a66
JK		 2438destroy_sit_entry_set:
2439 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2440destory_discard_entry:
2441 kmem_cache_destroy(discard_entry_slab);
2442fail:
2443 return -ENOMEM;
7fd9e544
JK		 2444}
2445
2446void destroy_segment_manager_caches(void)
2447{
184a5cd2 2448 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2449 kmem_cache_destroy(discard_entry_slab);
88b88a66 2450 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2451}
Linux kernel - Deliverables
This page took 0.258373 seconds and 5 git commands to generate.