]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/f2fs/segment.c
f2fs: introduce the number of inode entries
[mirror_ubuntu-artful-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK
2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
351df4b2 17#include <linux/vmalloc.h>
74de593a 18#include <linux/swap.h>
351df4b2
JK
19
20#include "f2fs.h"
21#include "segment.h"
22#include "node.h"
6ec178da 23#include <trace/events/f2fs.h>
351df4b2 24
9a7f143a
CL
25#define __reverse_ffz(x) __reverse_ffs(~(x))
26
7fd9e544 27static struct kmem_cache *discard_entry_slab;
184a5cd2 28static struct kmem_cache *sit_entry_set_slab;
88b88a66 29static struct kmem_cache *inmem_entry_slab;
7fd9e544 30
9a7f143a
CL
31/*
32 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
33 * MSB and LSB are reversed in a byte by f2fs_set_bit.
34 */
35static inline unsigned long __reverse_ffs(unsigned long word)
36{
37 int num = 0;
38
39#if BITS_PER_LONG == 64
40 if ((word & 0xffffffff) == 0) {
41 num += 32;
42 word >>= 32;
43 }
44#endif
45 if ((word & 0xffff) == 0) {
46 num += 16;
47 word >>= 16;
48 }
49 if ((word & 0xff) == 0) {
50 num += 8;
51 word >>= 8;
52 }
53 if ((word & 0xf0) == 0)
54 num += 4;
55 else
56 word >>= 4;
57 if ((word & 0xc) == 0)
58 num += 2;
59 else
60 word >>= 2;
61 if ((word & 0x2) == 0)
62 num += 1;
63 return num;
64}
65
66/*
e1c42045 67 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a
CL
68 * f2fs_set_bit makes MSB and LSB reversed in a byte.
69 * Example:
70 * LSB <--> MSB
71 * f2fs_set_bit(0, bitmap) => 0000 0001
72 * f2fs_set_bit(7, bitmap) => 1000 0000
73 */
74static unsigned long __find_rev_next_bit(const unsigned long *addr,
75 unsigned long size, unsigned long offset)
76{
77 const unsigned long *p = addr + BIT_WORD(offset);
78 unsigned long result = offset & ~(BITS_PER_LONG - 1);
79 unsigned long tmp;
80 unsigned long mask, submask;
81 unsigned long quot, rest;
82
83 if (offset >= size)
84 return size;
85
86 size -= result;
87 offset %= BITS_PER_LONG;
88 if (!offset)
89 goto aligned;
90
91 tmp = *(p++);
92 quot = (offset >> 3) << 3;
93 rest = offset & 0x7;
94 mask = ~0UL << quot;
95 submask = (unsigned char)(0xff << rest) >> rest;
96 submask <<= quot;
97 mask &= submask;
98 tmp &= mask;
99 if (size < BITS_PER_LONG)
100 goto found_first;
101 if (tmp)
102 goto found_middle;
103
104 size -= BITS_PER_LONG;
105 result += BITS_PER_LONG;
106aligned:
107 while (size & ~(BITS_PER_LONG-1)) {
108 tmp = *(p++);
109 if (tmp)
110 goto found_middle;
111 result += BITS_PER_LONG;
112 size -= BITS_PER_LONG;
113 }
114 if (!size)
115 return result;
116 tmp = *p;
117found_first:
118 tmp &= (~0UL >> (BITS_PER_LONG - size));
119 if (tmp == 0UL) /* Are any bits set? */
120 return result + size; /* Nope. */
121found_middle:
122 return result + __reverse_ffs(tmp);
123}
124
125static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
126 unsigned long size, unsigned long offset)
127{
128 const unsigned long *p = addr + BIT_WORD(offset);
129 unsigned long result = offset & ~(BITS_PER_LONG - 1);
130 unsigned long tmp;
131 unsigned long mask, submask;
132 unsigned long quot, rest;
133
134 if (offset >= size)
135 return size;
136
137 size -= result;
138 offset %= BITS_PER_LONG;
139 if (!offset)
140 goto aligned;
141
142 tmp = *(p++);
143 quot = (offset >> 3) << 3;
144 rest = offset & 0x7;
145 mask = ~(~0UL << quot);
146 submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
147 submask <<= quot;
148 mask += submask;
149 tmp |= mask;
150 if (size < BITS_PER_LONG)
151 goto found_first;
152 if (~tmp)
153 goto found_middle;
154
155 size -= BITS_PER_LONG;
156 result += BITS_PER_LONG;
157aligned:
158 while (size & ~(BITS_PER_LONG - 1)) {
159 tmp = *(p++);
160 if (~tmp)
161 goto found_middle;
162 result += BITS_PER_LONG;
163 size -= BITS_PER_LONG;
164 }
165 if (!size)
166 return result;
167 tmp = *p;
168
169found_first:
170 tmp |= ~0UL << size;
171 if (tmp == ~0UL) /* Are any bits zero? */
172 return result + size; /* Nope. */
173found_middle:
174 return result + __reverse_ffz(tmp);
175}
176
88b88a66
JK
177void register_inmem_page(struct inode *inode, struct page *page)
178{
179 struct f2fs_inode_info *fi = F2FS_I(inode);
180 struct inmem_pages *new;
34ba94ba
JK
181 int err;
182retry:
88b88a66
JK
183 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
184
185 /* add atomic page indices to the list */
186 new->page = page;
187 INIT_LIST_HEAD(&new->list);
188
189 /* increase reference count with clean state */
190 mutex_lock(&fi->inmem_lock);
34ba94ba
JK
191 err = radix_tree_insert(&fi->inmem_root, page->index, new);
192 if (err == -EEXIST) {
193 mutex_unlock(&fi->inmem_lock);
194 kmem_cache_free(inmem_entry_slab, new);
195 return;
196 } else if (err) {
197 mutex_unlock(&fi->inmem_lock);
198 kmem_cache_free(inmem_entry_slab, new);
199 goto retry;
200 }
88b88a66
JK
201 get_page(page);
202 list_add_tail(&new->list, &fi->inmem_pages);
203 mutex_unlock(&fi->inmem_lock);
204}
205
cbcb2872
JK
206void invalidate_inmem_page(struct inode *inode, struct page *page)
207{
208 struct f2fs_inode_info *fi = F2FS_I(inode);
209 struct inmem_pages *cur;
210
211 mutex_lock(&fi->inmem_lock);
212 cur = radix_tree_lookup(&fi->inmem_root, page->index);
213 if (cur) {
214 radix_tree_delete(&fi->inmem_root, cur->page->index);
215 f2fs_put_page(cur->page, 0);
216 list_del(&cur->list);
217 kmem_cache_free(inmem_entry_slab, cur);
218 }
219 mutex_unlock(&fi->inmem_lock);
220}
221
88b88a66
JK
222void commit_inmem_pages(struct inode *inode, bool abort)
223{
224 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
225 struct f2fs_inode_info *fi = F2FS_I(inode);
226 struct inmem_pages *cur, *tmp;
227 bool submit_bio = false;
228 struct f2fs_io_info fio = {
229 .type = DATA,
230 .rw = WRITE_SYNC,
231 };
232
233 f2fs_balance_fs(sbi);
234 f2fs_lock_op(sbi);
235
236 mutex_lock(&fi->inmem_lock);
237 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
238 lock_page(cur->page);
239 if (!abort && cur->page->mapping == inode->i_mapping) {
240 f2fs_wait_on_page_writeback(cur->page, DATA);
241 if (clear_page_dirty_for_io(cur->page))
242 inode_dec_dirty_pages(inode);
243 do_write_data_page(cur->page, &fio);
244 submit_bio = true;
245 }
34ba94ba 246 radix_tree_delete(&fi->inmem_root, cur->page->index);
88b88a66
JK
247 f2fs_put_page(cur->page, 1);
248 list_del(&cur->list);
249 kmem_cache_free(inmem_entry_slab, cur);
250 }
251 if (submit_bio)
252 f2fs_submit_merged_bio(sbi, DATA, WRITE);
253 mutex_unlock(&fi->inmem_lock);
254
255 filemap_fdatawait_range(inode->i_mapping, 0, LLONG_MAX);
256 f2fs_unlock_op(sbi);
257}
258
0a8165d7 259/*
351df4b2
JK
260 * This function balances dirty node and dentry pages.
261 * In addition, it controls garbage collection.
262 */
263void f2fs_balance_fs(struct f2fs_sb_info *sbi)
264{
351df4b2 265 /*
029cd28c
JK
266 * We should do GC or end up with checkpoint, if there are so many dirty
267 * dir/node pages without enough free segments.
351df4b2 268 */
43727527 269 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 270 mutex_lock(&sbi->gc_mutex);
408e9375 271 f2fs_gc(sbi);
351df4b2
JK
272 }
273}
274
4660f9c0
JK
275void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
276{
277 /* check the # of cached NAT entries and prefree segments */
278 if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
279 excess_prefree_segs(sbi))
280 f2fs_sync_fs(sbi->sb, true);
281}
282
2163d198 283static int issue_flush_thread(void *data)
6b4afdd7
JK
284{
285 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ
286 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
287 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK
288repeat:
289 if (kthread_should_stop())
290 return 0;
291
721bd4d5 292 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK
293 struct bio *bio = bio_alloc(GFP_NOIO, 0);
294 struct flush_cmd *cmd, *next;
295 int ret;
296
721bd4d5
GZ
297 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
298 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
299
6b4afdd7
JK
300 bio->bi_bdev = sbi->sb->s_bdev;
301 ret = submit_bio_wait(WRITE_FLUSH, bio);
302
721bd4d5
GZ
303 llist_for_each_entry_safe(cmd, next,
304 fcc->dispatch_list, llnode) {
6b4afdd7 305 cmd->ret = ret;
6b4afdd7
JK
306 complete(&cmd->wait);
307 }
a4ed23f2 308 bio_put(bio);
a688b9d9 309 fcc->dispatch_list = NULL;
6b4afdd7
JK
310 }
311
a688b9d9 312 wait_event_interruptible(*q,
721bd4d5 313 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK
314 goto repeat;
315}
316
317int f2fs_issue_flush(struct f2fs_sb_info *sbi)
318{
a688b9d9 319 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 320 struct flush_cmd cmd;
6b4afdd7 321
24a9ee0f
JK
322 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
323 test_opt(sbi, FLUSH_MERGE));
324
0f7b2abd
JK
325 if (test_opt(sbi, NOBARRIER))
326 return 0;
327
6b4afdd7
JK
328 if (!test_opt(sbi, FLUSH_MERGE))
329 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
330
adf8d90b 331 init_completion(&cmd.wait);
6b4afdd7 332
721bd4d5 333 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 334
a688b9d9
GZ
335 if (!fcc->dispatch_list)
336 wake_up(&fcc->flush_wait_queue);
6b4afdd7 337
adf8d90b
CY
338 wait_for_completion(&cmd.wait);
339
340 return cmd.ret;
6b4afdd7
JK
341}
342
2163d198
GZ
343int create_flush_cmd_control(struct f2fs_sb_info *sbi)
344{
345 dev_t dev = sbi->sb->s_bdev->bd_dev;
346 struct flush_cmd_control *fcc;
347 int err = 0;
348
349 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
350 if (!fcc)
351 return -ENOMEM;
2163d198 352 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 353 init_llist_head(&fcc->issue_list);
6b2920a5 354 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ
355 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
356 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
357 if (IS_ERR(fcc->f2fs_issue_flush)) {
358 err = PTR_ERR(fcc->f2fs_issue_flush);
359 kfree(fcc);
6b2920a5 360 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ
361 return err;
362 }
2163d198
GZ
363
364 return err;
365}
366
367void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
368{
6b2920a5 369 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ
370
371 if (fcc && fcc->f2fs_issue_flush)
372 kthread_stop(fcc->f2fs_issue_flush);
373 kfree(fcc);
6b2920a5 374 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ
375}
376
351df4b2
JK
377static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
378 enum dirty_type dirty_type)
379{
380 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
381
382 /* need not be added */
383 if (IS_CURSEG(sbi, segno))
384 return;
385
386 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
387 dirty_i->nr_dirty[dirty_type]++;
388
389 if (dirty_type == DIRTY) {
390 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 391 enum dirty_type t = sentry->type;
b2f2c390 392
ec325b52
JK
393 if (unlikely(t >= DIRTY)) {
394 f2fs_bug_on(sbi, 1);
395 return;
396 }
4625d6aa
CL
397 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
398 dirty_i->nr_dirty[t]++;
351df4b2
JK
399 }
400}
401
402static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
403 enum dirty_type dirty_type)
404{
405 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
406
407 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
408 dirty_i->nr_dirty[dirty_type]--;
409
410 if (dirty_type == DIRTY) {
4625d6aa
CL
411 struct seg_entry *sentry = get_seg_entry(sbi, segno);
412 enum dirty_type t = sentry->type;
413
414 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
415 dirty_i->nr_dirty[t]--;
b2f2c390 416
5ec4e49f
JK
417 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
418 clear_bit(GET_SECNO(sbi, segno),
419 dirty_i->victim_secmap);
351df4b2
JK
420 }
421}
422
0a8165d7 423/*
351df4b2
JK
424 * Should not occur error such as -ENOMEM.
425 * Adding dirty entry into seglist is not critical operation.
426 * If a given segment is one of current working segments, it won't be added.
427 */
8d8451af 428static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK
429{
430 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
431 unsigned short valid_blocks;
432
433 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
434 return;
435
436 mutex_lock(&dirty_i->seglist_lock);
437
438 valid_blocks = get_valid_blocks(sbi, segno, 0);
439
440 if (valid_blocks == 0) {
441 __locate_dirty_segment(sbi, segno, PRE);
442 __remove_dirty_segment(sbi, segno, DIRTY);
443 } else if (valid_blocks < sbi->blocks_per_seg) {
444 __locate_dirty_segment(sbi, segno, DIRTY);
445 } else {
446 /* Recovery routine with SSR needs this */
447 __remove_dirty_segment(sbi, segno, DIRTY);
448 }
449
450 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK
451}
452
1e87a78d 453static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK
454 block_t blkstart, block_t blklen)
455{
55cf9cb6
CY
456 sector_t start = SECTOR_FROM_BLOCK(blkstart);
457 sector_t len = SECTOR_FROM_BLOCK(blklen);
1661d07c 458 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK
459 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
460}
461
cf2271e7 462void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 463{
1e87a78d
JK
464 if (f2fs_issue_discard(sbi, blkaddr, 1)) {
465 struct page *page = grab_meta_page(sbi, blkaddr);
466 /* zero-filled page */
467 set_page_dirty(page);
468 f2fs_put_page(page, 1);
469 }
37208879
JK
470}
471
adf4983b
JK
472static void __add_discard_entry(struct f2fs_sb_info *sbi,
473 struct cp_control *cpc, unsigned int start, unsigned int end)
b2955550
JK
474{
475 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK
476 struct discard_entry *new, *last;
477
478 if (!list_empty(head)) {
479 last = list_last_entry(head, struct discard_entry, list);
480 if (START_BLOCK(sbi, cpc->trim_start) + start ==
481 last->blkaddr + last->len) {
482 last->len += end - start;
483 goto done;
484 }
485 }
486
487 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
488 INIT_LIST_HEAD(&new->list);
489 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
490 new->len = end - start;
491 list_add_tail(&new->list, head);
492done:
493 SM_I(sbi)->nr_discards += end - start;
494 cpc->trimmed += end - start;
495}
496
497static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
498{
b2955550
JK
499 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
500 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 501 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK
502 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
503 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
504 unsigned long dmap[entries];
505 unsigned int start = 0, end = -1;
4b2fecc8 506 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK
507 int i;
508
4b2fecc8 509 if (!force && !test_opt(sbi, DISCARD))
b2955550
JK
510 return;
511
4b2fecc8
JK
512 if (force && !se->valid_blocks) {
513 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
514 /*
515 * if this segment is registered in the prefree list, then
516 * we should skip adding a discard candidate, and let the
517 * checkpoint do that later.
518 */
519 mutex_lock(&dirty_i->seglist_lock);
520 if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) {
521 mutex_unlock(&dirty_i->seglist_lock);
522 cpc->trimmed += sbi->blocks_per_seg;
523 return;
524 }
525 mutex_unlock(&dirty_i->seglist_lock);
526
adf4983b 527 __add_discard_entry(sbi, cpc, 0, sbi->blocks_per_seg);
4b2fecc8
JK
528 return;
529 }
530
b2955550
JK
531 /* zero block will be discarded through the prefree list */
532 if (!se->valid_blocks || se->valid_blocks == max_blocks)
533 return;
534
535 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
536 for (i = 0; i < entries; i++)
e3fb1b79 537 dmap[i] = ~(cur_map[i] | ckpt_map[i]);
b2955550 538
4b2fecc8 539 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK
540 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
541 if (start >= max_blocks)
542 break;
543
544 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
545
4b2fecc8
JK
546 if (end - start < cpc->trim_minlen)
547 continue;
548
adf4983b 549 __add_discard_entry(sbi, cpc, start, end);
b2955550
JK
550 }
551}
552
4b2fecc8
JK
553void release_discard_addrs(struct f2fs_sb_info *sbi)
554{
555 struct list_head *head = &(SM_I(sbi)->discard_list);
556 struct discard_entry *entry, *this;
557
558 /* drop caches */
559 list_for_each_entry_safe(entry, this, head, list) {
560 list_del(&entry->list);
561 kmem_cache_free(discard_entry_slab, entry);
562 }
563}
564
0a8165d7 565/*
351df4b2
JK
566 * Should call clear_prefree_segments after checkpoint is done.
567 */
568static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
569{
570 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 571 unsigned int segno;
351df4b2
JK
572
573 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 574 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 575 __set_test_and_free(sbi, segno);
351df4b2
JK
576 mutex_unlock(&dirty_i->seglist_lock);
577}
578
579void clear_prefree_segments(struct f2fs_sb_info *sbi)
580{
b2955550 581 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 582 struct discard_entry *entry, *this;
351df4b2 583 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 584 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 585 unsigned int start = 0, end = -1;
351df4b2
JK
586
587 mutex_lock(&dirty_i->seglist_lock);
29e59c14 588
351df4b2 589 while (1) {
29e59c14 590 int i;
7cd8558b
JK
591 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
592 if (start >= MAIN_SEGS(sbi))
351df4b2 593 break;
7cd8558b
JK
594 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
595 start + 1);
29e59c14
CL
596
597 for (i = start; i < end; i++)
598 clear_bit(i, prefree_map);
599
600 dirty_i->nr_dirty[PRE] -= end - start;
601
602 if (!test_opt(sbi, DISCARD))
603 continue;
351df4b2 604
37208879
JK
605 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
606 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK
607 }
608 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK
609
610 /* send small discards */
2d7b822a 611 list_for_each_entry_safe(entry, this, head, list) {
37208879 612 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
b2955550
JK
613 list_del(&entry->list);
614 SM_I(sbi)->nr_discards -= entry->len;
615 kmem_cache_free(discard_entry_slab, entry);
616 }
351df4b2
JK
617}
618
184a5cd2 619static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK
620{
621 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY
622
623 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 624 sit_i->dirty_sentries++;
184a5cd2
CY
625 return false;
626 }
627
628 return true;
351df4b2
JK
629}
630
631static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
632 unsigned int segno, int modified)
633{
634 struct seg_entry *se = get_seg_entry(sbi, segno);
635 se->type = type;
636 if (modified)
637 __mark_sit_entry_dirty(sbi, segno);
638}
639
640static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
641{
642 struct seg_entry *se;
643 unsigned int segno, offset;
644 long int new_vblocks;
645
646 segno = GET_SEGNO(sbi, blkaddr);
647
648 se = get_seg_entry(sbi, segno);
649 new_vblocks = se->valid_blocks + del;
491c0854 650 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 651
9850cf4a 652 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK
653 (new_vblocks > sbi->blocks_per_seg)));
654
655 se->valid_blocks = new_vblocks;
656 se->mtime = get_mtime(sbi);
657 SIT_I(sbi)->max_mtime = se->mtime;
658
659 /* Update valid block bitmap */
660 if (del > 0) {
52aca074 661 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 662 f2fs_bug_on(sbi, 1);
351df4b2 663 } else {
52aca074 664 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 665 f2fs_bug_on(sbi, 1);
351df4b2
JK
666 }
667 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
668 se->ckpt_valid_blocks += del;
669
670 __mark_sit_entry_dirty(sbi, segno);
671
672 /* update total number of valid blocks to be written in ckpt area */
673 SIT_I(sbi)->written_valid_blocks += del;
674
675 if (sbi->segs_per_sec > 1)
676 get_sec_entry(sbi, segno)->valid_blocks += del;
677}
678
5e443818 679void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 680{
5e443818
JK
681 update_sit_entry(sbi, new, 1);
682 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
683 update_sit_entry(sbi, old, -1);
684
685 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
686 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK
687}
688
689void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
690{
691 unsigned int segno = GET_SEGNO(sbi, addr);
692 struct sit_info *sit_i = SIT_I(sbi);
693
9850cf4a 694 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK
695 if (addr == NEW_ADDR)
696 return;
697
698 /* add it into sit main buffer */
699 mutex_lock(&sit_i->sentry_lock);
700
701 update_sit_entry(sbi, addr, -1);
702
703 /* add it into dirty seglist */
704 locate_dirty_segment(sbi, segno);
705
706 mutex_unlock(&sit_i->sentry_lock);
707}
708
0a8165d7 709/*
351df4b2
JK
710 * This function should be resided under the curseg_mutex lock
711 */
712static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 713 struct f2fs_summary *sum)
351df4b2
JK
714{
715 struct curseg_info *curseg = CURSEG_I(sbi, type);
716 void *addr = curseg->sum_blk;
e79efe3b 717 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 718 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK
719}
720
0a8165d7 721/*
351df4b2
JK
722 * Calculate the number of current summary pages for writing
723 */
724int npages_for_summary_flush(struct f2fs_sb_info *sbi)
725{
351df4b2 726 int valid_sum_count = 0;
9a47938b 727 int i, sum_in_page;
351df4b2
JK
728
729 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
730 if (sbi->ckpt->alloc_type[i] == SSR)
731 valid_sum_count += sbi->blocks_per_seg;
732 else
733 valid_sum_count += curseg_blkoff(sbi, i);
734 }
735
9a47938b
FL
736 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
737 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
738 if (valid_sum_count <= sum_in_page)
351df4b2 739 return 1;
9a47938b
FL
740 else if ((valid_sum_count - sum_in_page) <=
741 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK
742 return 2;
743 return 3;
744}
745
0a8165d7 746/*
351df4b2
JK
747 * Caller should put this summary page
748 */
749struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
750{
751 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
752}
753
754static void write_sum_page(struct f2fs_sb_info *sbi,
755 struct f2fs_summary_block *sum_blk, block_t blk_addr)
756{
757 struct page *page = grab_meta_page(sbi, blk_addr);
758 void *kaddr = page_address(page);
759 memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
760 set_page_dirty(page);
761 f2fs_put_page(page, 1);
762}
763
60374688
JK
764static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
765{
766 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 767 unsigned int segno = curseg->segno + 1;
60374688
JK
768 struct free_segmap_info *free_i = FREE_I(sbi);
769
7cd8558b 770 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 771 return !test_bit(segno, free_i->free_segmap);
60374688
JK
772 return 0;
773}
774
0a8165d7 775/*
351df4b2
JK
776 * Find a new segment from the free segments bitmap to right order
777 * This function should be returned with success, otherwise BUG
778 */
779static void get_new_segment(struct f2fs_sb_info *sbi,
780 unsigned int *newseg, bool new_sec, int dir)
781{
782 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 783 unsigned int segno, secno, zoneno;
7cd8558b 784 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK
785 unsigned int hint = *newseg / sbi->segs_per_sec;
786 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
787 unsigned int left_start = hint;
788 bool init = true;
789 int go_left = 0;
790 int i;
791
792 write_lock(&free_i->segmap_lock);
793
794 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
795 segno = find_next_zero_bit(free_i->free_segmap,
7cd8558b 796 MAIN_SEGS(sbi), *newseg + 1);
33afa7fd
JK
797 if (segno - *newseg < sbi->segs_per_sec -
798 (*newseg % sbi->segs_per_sec))
351df4b2
JK
799 goto got_it;
800 }
801find_other_zone:
7cd8558b
JK
802 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
803 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK
804 if (dir == ALLOC_RIGHT) {
805 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK
806 MAIN_SECS(sbi), 0);
807 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK
808 } else {
809 go_left = 1;
810 left_start = hint - 1;
811 }
812 }
813 if (go_left == 0)
814 goto skip_left;
815
816 while (test_bit(left_start, free_i->free_secmap)) {
817 if (left_start > 0) {
818 left_start--;
819 continue;
820 }
821 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK
822 MAIN_SECS(sbi), 0);
823 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK
824 break;
825 }
826 secno = left_start;
827skip_left:
828 hint = secno;
829 segno = secno * sbi->segs_per_sec;
830 zoneno = secno / sbi->secs_per_zone;
831
832 /* give up on finding another zone */
833 if (!init)
834 goto got_it;
835 if (sbi->secs_per_zone == 1)
836 goto got_it;
837 if (zoneno == old_zoneno)
838 goto got_it;
839 if (dir == ALLOC_LEFT) {
840 if (!go_left && zoneno + 1 >= total_zones)
841 goto got_it;
842 if (go_left && zoneno == 0)
843 goto got_it;
844 }
845 for (i = 0; i < NR_CURSEG_TYPE; i++)
846 if (CURSEG_I(sbi, i)->zone == zoneno)
847 break;
848
849 if (i < NR_CURSEG_TYPE) {
850 /* zone is in user, try another */
851 if (go_left)
852 hint = zoneno * sbi->secs_per_zone - 1;
853 else if (zoneno + 1 >= total_zones)
854 hint = 0;
855 else
856 hint = (zoneno + 1) * sbi->secs_per_zone;
857 init = false;
858 goto find_other_zone;
859 }
860got_it:
861 /* set it as dirty segment in free segmap */
9850cf4a 862 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK
863 __set_inuse(sbi, segno);
864 *newseg = segno;
865 write_unlock(&free_i->segmap_lock);
866}
867
868static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
869{
870 struct curseg_info *curseg = CURSEG_I(sbi, type);
871 struct summary_footer *sum_footer;
872
873 curseg->segno = curseg->next_segno;
874 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
875 curseg->next_blkoff = 0;
876 curseg->next_segno = NULL_SEGNO;
877
878 sum_footer = &(curseg->sum_blk->footer);
879 memset(sum_footer, 0, sizeof(struct summary_footer));
880 if (IS_DATASEG(type))
881 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
882 if (IS_NODESEG(type))
883 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
884 __set_sit_entry_type(sbi, type, curseg->segno, modified);
885}
886
0a8165d7 887/*
351df4b2
JK
888 * Allocate a current working segment.
889 * This function always allocates a free segment in LFS manner.
890 */
891static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
892{
893 struct curseg_info *curseg = CURSEG_I(sbi, type);
894 unsigned int segno = curseg->segno;
895 int dir = ALLOC_LEFT;
896
897 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 898 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK
899 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
900 dir = ALLOC_RIGHT;
901
902 if (test_opt(sbi, NOHEAP))
903 dir = ALLOC_RIGHT;
904
905 get_new_segment(sbi, &segno, new_sec, dir);
906 curseg->next_segno = segno;
907 reset_curseg(sbi, type, 1);
908 curseg->alloc_type = LFS;
909}
910
911static void __next_free_blkoff(struct f2fs_sb_info *sbi,
912 struct curseg_info *seg, block_t start)
913{
914 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf
CL
915 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
916 unsigned long target_map[entries];
917 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
918 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
919 int i, pos;
920
921 for (i = 0; i < entries; i++)
922 target_map[i] = ckpt_map[i] | cur_map[i];
923
924 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
925
926 seg->next_blkoff = pos;
351df4b2
JK
927}
928
0a8165d7 929/*
351df4b2
JK
930 * If a segment is written by LFS manner, next block offset is just obtained
931 * by increasing the current block offset. However, if a segment is written by
932 * SSR manner, next block offset obtained by calling __next_free_blkoff
933 */
934static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
935 struct curseg_info *seg)
936{
937 if (seg->alloc_type == SSR)
938 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
939 else
940 seg->next_blkoff++;
941}
942
0a8165d7 943/*
e1c42045 944 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK
945 * manner, so it should recover the existing segment information of valid blocks
946 */
947static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
948{
949 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
950 struct curseg_info *curseg = CURSEG_I(sbi, type);
951 unsigned int new_segno = curseg->next_segno;
952 struct f2fs_summary_block *sum_node;
953 struct page *sum_page;
954
955 write_sum_page(sbi, curseg->sum_blk,
956 GET_SUM_BLOCK(sbi, curseg->segno));
957 __set_test_and_inuse(sbi, new_segno);
958
959 mutex_lock(&dirty_i->seglist_lock);
960 __remove_dirty_segment(sbi, new_segno, PRE);
961 __remove_dirty_segment(sbi, new_segno, DIRTY);
962 mutex_unlock(&dirty_i->seglist_lock);
963
964 reset_curseg(sbi, type, 1);
965 curseg->alloc_type = SSR;
966 __next_free_blkoff(sbi, curseg, 0);
967
968 if (reuse) {
969 sum_page = get_sum_page(sbi, new_segno);
970 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
971 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
972 f2fs_put_page(sum_page, 1);
973 }
974}
975
43727527
JK
976static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
977{
978 struct curseg_info *curseg = CURSEG_I(sbi, type);
979 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
980
981 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
982 return v_ops->get_victim(sbi,
983 &(curseg)->next_segno, BG_GC, type, SSR);
984
985 /* For data segments, let's do SSR more intensively */
986 for (; type >= CURSEG_HOT_DATA; type--)
987 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
988 BG_GC, type, SSR))
989 return 1;
990 return 0;
991}
992
351df4b2
JK
993/*
994 * flush out current segment and replace it with new segment
995 * This function should be returned with success, otherwise BUG
996 */
997static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
998 int type, bool force)
999{
1000 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1001
7b405275 1002 if (force)
351df4b2 1003 new_curseg(sbi, type, true);
7b405275 1004 else if (type == CURSEG_WARM_NODE)
351df4b2 1005 new_curseg(sbi, type, false);
60374688
JK
1006 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1007 new_curseg(sbi, type, false);
351df4b2
JK
1008 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1009 change_curseg(sbi, type, true);
1010 else
1011 new_curseg(sbi, type, false);
dcdfff65
JK
1012
1013 stat_inc_seg_type(sbi, curseg);
351df4b2
JK
1014}
1015
1016void allocate_new_segments(struct f2fs_sb_info *sbi)
1017{
1018 struct curseg_info *curseg;
1019 unsigned int old_curseg;
1020 int i;
1021
1022 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1023 curseg = CURSEG_I(sbi, i);
1024 old_curseg = curseg->segno;
1025 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1026 locate_dirty_segment(sbi, old_curseg);
1027 }
1028}
1029
1030static const struct segment_allocation default_salloc_ops = {
1031 .allocate_segment = allocate_segment_by_default,
1032};
1033
4b2fecc8
JK
1034int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1035{
4b2fecc8
JK
1036 __u64 start = range->start >> sbi->log_blocksize;
1037 __u64 end = start + (range->len >> sbi->log_blocksize) - 1;
4b2fecc8
JK
1038 unsigned int start_segno, end_segno;
1039 struct cp_control cpc;
1040
7cd8558b
JK
1041 if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) ||
1042 range->len < sbi->blocksize)
4b2fecc8
JK
1043 return -EINVAL;
1044
9bd27ae4 1045 cpc.trimmed = 0;
7cd8558b 1046 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK
1047 goto out;
1048
1049 /* start/end segment number in main_area */
7cd8558b
JK
1050 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1051 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1052 GET_SEGNO(sbi, end);
4b2fecc8
JK
1053 cpc.reason = CP_DISCARD;
1054 cpc.trim_start = start_segno;
1055 cpc.trim_end = end_segno;
1056 cpc.trim_minlen = range->minlen >> sbi->log_blocksize;
4b2fecc8
JK
1057
1058 /* do checkpoint to issue discard commands safely */
ca4b02ee 1059 mutex_lock(&sbi->gc_mutex);
4b2fecc8 1060 write_checkpoint(sbi, &cpc);
ca4b02ee 1061 mutex_unlock(&sbi->gc_mutex);
4b2fecc8
JK
1062out:
1063 range->len = cpc.trimmed << sbi->log_blocksize;
1064 return 0;
1065}
1066
351df4b2
JK
1067static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1068{
1069 struct curseg_info *curseg = CURSEG_I(sbi, type);
1070 if (curseg->next_blkoff < sbi->blocks_per_seg)
1071 return true;
1072 return false;
1073}
1074
1075static int __get_segment_type_2(struct page *page, enum page_type p_type)
1076{
1077 if (p_type == DATA)
1078 return CURSEG_HOT_DATA;
1079 else
1080 return CURSEG_HOT_NODE;
1081}
1082
1083static int __get_segment_type_4(struct page *page, enum page_type p_type)
1084{
1085 if (p_type == DATA) {
1086 struct inode *inode = page->mapping->host;
1087
1088 if (S_ISDIR(inode->i_mode))
1089 return CURSEG_HOT_DATA;
1090 else
1091 return CURSEG_COLD_DATA;
1092 } else {
a344b9fd
JK
1093 if (IS_DNODE(page) && is_cold_node(page))
1094 return CURSEG_WARM_NODE;
351df4b2
JK
1095 else
1096 return CURSEG_COLD_NODE;
1097 }
1098}
1099
1100static int __get_segment_type_6(struct page *page, enum page_type p_type)
1101{
1102 if (p_type == DATA) {
1103 struct inode *inode = page->mapping->host;
1104
1105 if (S_ISDIR(inode->i_mode))
1106 return CURSEG_HOT_DATA;
354a3399 1107 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK
1108 return CURSEG_COLD_DATA;
1109 else
1110 return CURSEG_WARM_DATA;
1111 } else {
1112 if (IS_DNODE(page))
1113 return is_cold_node(page) ? CURSEG_WARM_NODE :
1114 CURSEG_HOT_NODE;
1115 else
1116 return CURSEG_COLD_NODE;
1117 }
1118}
1119
1120static int __get_segment_type(struct page *page, enum page_type p_type)
1121{
4081363f 1122 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK
1123 case 2:
1124 return __get_segment_type_2(page, p_type);
1125 case 4:
1126 return __get_segment_type_4(page, p_type);
351df4b2 1127 }
12a67146 1128 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK
1129 f2fs_bug_on(F2FS_P_SB(page),
1130 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1131 return __get_segment_type_6(page, p_type);
351df4b2
JK
1132}
1133
bfad7c2d
JK
1134void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1135 block_t old_blkaddr, block_t *new_blkaddr,
1136 struct f2fs_summary *sum, int type)
351df4b2
JK
1137{
1138 struct sit_info *sit_i = SIT_I(sbi);
1139 struct curseg_info *curseg;
351df4b2 1140
351df4b2
JK
1141 curseg = CURSEG_I(sbi, type);
1142
1143 mutex_lock(&curseg->curseg_mutex);
1144
1145 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK
1146
1147 /*
1148 * __add_sum_entry should be resided under the curseg_mutex
1149 * because, this function updates a summary entry in the
1150 * current summary block.
1151 */
e79efe3b 1152 __add_sum_entry(sbi, type, sum);
351df4b2
JK
1153
1154 mutex_lock(&sit_i->sentry_lock);
1155 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK
1156
1157 stat_inc_block_count(sbi, curseg);
351df4b2 1158
5e443818
JK
1159 if (!__has_curseg_space(sbi, type))
1160 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK
1161 /*
1162 * SIT information should be updated before segment allocation,
1163 * since SSR needs latest valid block information.
1164 */
1165 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1166
351df4b2
JK
1167 mutex_unlock(&sit_i->sentry_lock);
1168
bfad7c2d 1169 if (page && IS_NODESEG(type))
351df4b2
JK
1170 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1171
bfad7c2d
JK
1172 mutex_unlock(&curseg->curseg_mutex);
1173}
1174
1175static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1176 block_t old_blkaddr, block_t *new_blkaddr,
1177 struct f2fs_summary *sum, struct f2fs_io_info *fio)
1178{
1179 int type = __get_segment_type(page, fio->type);
1180
1181 allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
1182
351df4b2 1183 /* writeout dirty page into bdev */
458e6197 1184 f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
351df4b2
JK
1185}
1186
577e3495 1187void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1188{
458e6197
JK
1189 struct f2fs_io_info fio = {
1190 .type = META,
7e8f2308 1191 .rw = WRITE_SYNC | REQ_META | REQ_PRIO
458e6197
JK
1192 };
1193
351df4b2 1194 set_page_writeback(page);
458e6197 1195 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
351df4b2
JK
1196}
1197
1198void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
fb5566da 1199 struct f2fs_io_info *fio,
351df4b2
JK
1200 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
1201{
1202 struct f2fs_summary sum;
1203 set_summary(&sum, nid, 0, 0);
fb5566da 1204 do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
351df4b2
JK
1205}
1206
458e6197
JK
1207void write_data_page(struct page *page, struct dnode_of_data *dn,
1208 block_t *new_blkaddr, struct f2fs_io_info *fio)
351df4b2 1209{
4081363f 1210 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
351df4b2
JK
1211 struct f2fs_summary sum;
1212 struct node_info ni;
1213
9850cf4a 1214 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK
1215 get_node_info(sbi, dn->nid, &ni);
1216 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1217
458e6197 1218 do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
351df4b2
JK
1219}
1220
6c311ec6
CF
1221void rewrite_data_page(struct page *page, block_t old_blkaddr,
1222 struct f2fs_io_info *fio)
351df4b2 1223{
4081363f 1224 f2fs_submit_page_mbio(F2FS_P_SB(page), page, old_blkaddr, fio);
351df4b2
JK
1225}
1226
1227void recover_data_page(struct f2fs_sb_info *sbi,
1228 struct page *page, struct f2fs_summary *sum,
1229 block_t old_blkaddr, block_t new_blkaddr)
1230{
1231 struct sit_info *sit_i = SIT_I(sbi);
1232 struct curseg_info *curseg;
1233 unsigned int segno, old_cursegno;
1234 struct seg_entry *se;
1235 int type;
1236
1237 segno = GET_SEGNO(sbi, new_blkaddr);
1238 se = get_seg_entry(sbi, segno);
1239 type = se->type;
1240
1241 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1242 if (old_blkaddr == NULL_ADDR)
1243 type = CURSEG_COLD_DATA;
1244 else
1245 type = CURSEG_WARM_DATA;
1246 }
1247 curseg = CURSEG_I(sbi, type);
1248
1249 mutex_lock(&curseg->curseg_mutex);
1250 mutex_lock(&sit_i->sentry_lock);
1251
1252 old_cursegno = curseg->segno;
1253
1254 /* change the current segment */
1255 if (segno != curseg->segno) {
1256 curseg->next_segno = segno;
1257 change_curseg(sbi, type, true);
1258 }
1259
491c0854 1260 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1261 __add_sum_entry(sbi, type, sum);
351df4b2
JK
1262
1263 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
351df4b2 1264 locate_dirty_segment(sbi, old_cursegno);
351df4b2
JK
1265
1266 mutex_unlock(&sit_i->sentry_lock);
1267 mutex_unlock(&curseg->curseg_mutex);
1268}
1269
df0f8dc0
CY
1270static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1271 struct page *page, enum page_type type)
1272{
1273 enum page_type btype = PAGE_TYPE_OF_BIO(type);
1274 struct f2fs_bio_info *io = &sbi->write_io[btype];
df0f8dc0
CY
1275 struct bio_vec *bvec;
1276 int i;
1277
1278 down_read(&io->io_rwsem);
ce23447f 1279 if (!io->bio)
df0f8dc0
CY
1280 goto out;
1281
ce23447f 1282 bio_for_each_segment_all(bvec, io->bio, i) {
df0f8dc0
CY
1283 if (page == bvec->bv_page) {
1284 up_read(&io->io_rwsem);
1285 return true;
1286 }
1287 }
1288
1289out:
1290 up_read(&io->io_rwsem);
1291 return false;
1292}
1293
93dfe2ac 1294void f2fs_wait_on_page_writeback(struct page *page,
5514f0aa 1295 enum page_type type)
93dfe2ac 1296{
93dfe2ac 1297 if (PageWriteback(page)) {
4081363f
JK
1298 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1299
df0f8dc0
CY
1300 if (is_merged_page(sbi, page, type))
1301 f2fs_submit_merged_bio(sbi, type, WRITE);
93dfe2ac
JK
1302 wait_on_page_writeback(page);
1303 }
1304}
1305
351df4b2
JK
1306static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1307{
1308 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1309 struct curseg_info *seg_i;
1310 unsigned char *kaddr;
1311 struct page *page;
1312 block_t start;
1313 int i, j, offset;
1314
1315 start = start_sum_block(sbi);
1316
1317 page = get_meta_page(sbi, start++);
1318 kaddr = (unsigned char *)page_address(page);
1319
1320 /* Step 1: restore nat cache */
1321 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1322 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1323
1324 /* Step 2: restore sit cache */
1325 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1326 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1327 SUM_JOURNAL_SIZE);
1328 offset = 2 * SUM_JOURNAL_SIZE;
1329
1330 /* Step 3: restore summary entries */
1331 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1332 unsigned short blk_off;
1333 unsigned int segno;
1334
1335 seg_i = CURSEG_I(sbi, i);
1336 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1337 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1338 seg_i->next_segno = segno;
1339 reset_curseg(sbi, i, 0);
1340 seg_i->alloc_type = ckpt->alloc_type[i];
1341 seg_i->next_blkoff = blk_off;
1342
1343 if (seg_i->alloc_type == SSR)
1344 blk_off = sbi->blocks_per_seg;
1345
1346 for (j = 0; j < blk_off; j++) {
1347 struct f2fs_summary *s;
1348 s = (struct f2fs_summary *)(kaddr + offset);
1349 seg_i->sum_blk->entries[j] = *s;
1350 offset += SUMMARY_SIZE;
1351 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1352 SUM_FOOTER_SIZE)
1353 continue;
1354
1355 f2fs_put_page(page, 1);
1356 page = NULL;
1357
1358 page = get_meta_page(sbi, start++);
1359 kaddr = (unsigned char *)page_address(page);
1360 offset = 0;
1361 }
1362 }
1363 f2fs_put_page(page, 1);
1364 return 0;
1365}
1366
1367static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1368{
1369 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1370 struct f2fs_summary_block *sum;
1371 struct curseg_info *curseg;
1372 struct page *new;
1373 unsigned short blk_off;
1374 unsigned int segno = 0;
1375 block_t blk_addr = 0;
1376
1377 /* get segment number and block addr */
1378 if (IS_DATASEG(type)) {
1379 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1380 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1381 CURSEG_HOT_DATA]);
25ca923b 1382 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
351df4b2
JK
1383 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1384 else
1385 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1386 } else {
1387 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1388 CURSEG_HOT_NODE]);
1389 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1390 CURSEG_HOT_NODE]);
25ca923b 1391 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
351df4b2
JK
1392 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1393 type - CURSEG_HOT_NODE);
1394 else
1395 blk_addr = GET_SUM_BLOCK(sbi, segno);
1396 }
1397
1398 new = get_meta_page(sbi, blk_addr);
1399 sum = (struct f2fs_summary_block *)page_address(new);
1400
1401 if (IS_NODESEG(type)) {
25ca923b 1402 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
351df4b2
JK
1403 struct f2fs_summary *ns = &sum->entries[0];
1404 int i;
1405 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1406 ns->version = 0;
1407 ns->ofs_in_node = 0;
1408 }
1409 } else {
d653788a
GZ
1410 int err;
1411
1412 err = restore_node_summary(sbi, segno, sum);
1413 if (err) {
351df4b2 1414 f2fs_put_page(new, 1);
d653788a 1415 return err;
351df4b2
JK
1416 }
1417 }
1418 }
1419
1420 /* set uncompleted segment to curseg */
1421 curseg = CURSEG_I(sbi, type);
1422 mutex_lock(&curseg->curseg_mutex);
1423 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1424 curseg->next_segno = segno;
1425 reset_curseg(sbi, type, 0);
1426 curseg->alloc_type = ckpt->alloc_type[type];
1427 curseg->next_blkoff = blk_off;
1428 mutex_unlock(&curseg->curseg_mutex);
1429 f2fs_put_page(new, 1);
1430 return 0;
1431}
1432
1433static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1434{
1435 int type = CURSEG_HOT_DATA;
e4fc5fbf 1436 int err;
351df4b2 1437
25ca923b 1438 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
351df4b2
JK
1439 /* restore for compacted data summary */
1440 if (read_compacted_summaries(sbi))
1441 return -EINVAL;
1442 type = CURSEG_HOT_NODE;
1443 }
1444
e4fc5fbf
CY
1445 for (; type <= CURSEG_COLD_NODE; type++) {
1446 err = read_normal_summaries(sbi, type);
1447 if (err)
1448 return err;
1449 }
1450
351df4b2
JK
1451 return 0;
1452}
1453
1454static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1455{
1456 struct page *page;
1457 unsigned char *kaddr;
1458 struct f2fs_summary *summary;
1459 struct curseg_info *seg_i;
1460 int written_size = 0;
1461 int i, j;
1462
1463 page = grab_meta_page(sbi, blkaddr++);
1464 kaddr = (unsigned char *)page_address(page);
1465
1466 /* Step 1: write nat cache */
1467 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1468 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1469 written_size += SUM_JOURNAL_SIZE;
1470
1471 /* Step 2: write sit cache */
1472 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1473 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1474 SUM_JOURNAL_SIZE);
1475 written_size += SUM_JOURNAL_SIZE;
1476
351df4b2
JK
1477 /* Step 3: write summary entries */
1478 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1479 unsigned short blkoff;
1480 seg_i = CURSEG_I(sbi, i);
1481 if (sbi->ckpt->alloc_type[i] == SSR)
1482 blkoff = sbi->blocks_per_seg;
1483 else
1484 blkoff = curseg_blkoff(sbi, i);
1485
1486 for (j = 0; j < blkoff; j++) {
1487 if (!page) {
1488 page = grab_meta_page(sbi, blkaddr++);
1489 kaddr = (unsigned char *)page_address(page);
1490 written_size = 0;
1491 }
1492 summary = (struct f2fs_summary *)(kaddr + written_size);
1493 *summary = seg_i->sum_blk->entries[j];
1494 written_size += SUMMARY_SIZE;
351df4b2
JK
1495
1496 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1497 SUM_FOOTER_SIZE)
1498 continue;
1499
e8d61a74 1500 set_page_dirty(page);
351df4b2
JK
1501 f2fs_put_page(page, 1);
1502 page = NULL;
1503 }
1504 }
e8d61a74
CY
1505 if (page) {
1506 set_page_dirty(page);
351df4b2 1507 f2fs_put_page(page, 1);
e8d61a74 1508 }
351df4b2
JK
1509}
1510
1511static void write_normal_summaries(struct f2fs_sb_info *sbi,
1512 block_t blkaddr, int type)
1513{
1514 int i, end;
1515 if (IS_DATASEG(type))
1516 end = type + NR_CURSEG_DATA_TYPE;
1517 else
1518 end = type + NR_CURSEG_NODE_TYPE;
1519
1520 for (i = type; i < end; i++) {
1521 struct curseg_info *sum = CURSEG_I(sbi, i);
1522 mutex_lock(&sum->curseg_mutex);
1523 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1524 mutex_unlock(&sum->curseg_mutex);
1525 }
1526}
1527
1528void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1529{
25ca923b 1530 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK
1531 write_compacted_summaries(sbi, start_blk);
1532 else
1533 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1534}
1535
1536void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1537{
25ca923b 1538 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
351df4b2 1539 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK
1540}
1541
1542int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1543 unsigned int val, int alloc)
1544{
1545 int i;
1546
1547 if (type == NAT_JOURNAL) {
1548 for (i = 0; i < nats_in_cursum(sum); i++) {
1549 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1550 return i;
1551 }
1552 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1553 return update_nats_in_cursum(sum, 1);
1554 } else if (type == SIT_JOURNAL) {
1555 for (i = 0; i < sits_in_cursum(sum); i++)
1556 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1557 return i;
1558 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1559 return update_sits_in_cursum(sum, 1);
1560 }
1561 return -1;
1562}
1563
1564static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1565 unsigned int segno)
1566{
2cc22186 1567 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK
1568}
1569
1570static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1571 unsigned int start)
1572{
1573 struct sit_info *sit_i = SIT_I(sbi);
1574 struct page *src_page, *dst_page;
1575 pgoff_t src_off, dst_off;
1576 void *src_addr, *dst_addr;
1577
1578 src_off = current_sit_addr(sbi, start);
1579 dst_off = next_sit_addr(sbi, src_off);
1580
1581 /* get current sit block page without lock */
1582 src_page = get_meta_page(sbi, src_off);
1583 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1584 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK
1585
1586 src_addr = page_address(src_page);
1587 dst_addr = page_address(dst_page);
1588 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1589
1590 set_page_dirty(dst_page);
1591 f2fs_put_page(src_page, 1);
1592
1593 set_to_next_sit(sit_i, start);
1594
1595 return dst_page;
1596}
1597
184a5cd2
CY
1598static struct sit_entry_set *grab_sit_entry_set(void)
1599{
1600 struct sit_entry_set *ses =
1601 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1602
1603 ses->entry_cnt = 0;
1604 INIT_LIST_HEAD(&ses->set_list);
1605 return ses;
1606}
1607
1608static void release_sit_entry_set(struct sit_entry_set *ses)
1609{
1610 list_del(&ses->set_list);
1611 kmem_cache_free(sit_entry_set_slab, ses);
1612}
1613
1614static void adjust_sit_entry_set(struct sit_entry_set *ses,
1615 struct list_head *head)
1616{
1617 struct sit_entry_set *next = ses;
1618
1619 if (list_is_last(&ses->set_list, head))
1620 return;
1621
1622 list_for_each_entry_continue(next, head, set_list)
1623 if (ses->entry_cnt <= next->entry_cnt)
1624 break;
1625
1626 list_move_tail(&ses->set_list, &next->set_list);
1627}
1628
1629static void add_sit_entry(unsigned int segno, struct list_head *head)
1630{
1631 struct sit_entry_set *ses;
1632 unsigned int start_segno = START_SEGNO(segno);
1633
1634 list_for_each_entry(ses, head, set_list) {
1635 if (ses->start_segno == start_segno) {
1636 ses->entry_cnt++;
1637 adjust_sit_entry_set(ses, head);
1638 return;
1639 }
1640 }
1641
1642 ses = grab_sit_entry_set();
1643
1644 ses->start_segno = start_segno;
1645 ses->entry_cnt++;
1646 list_add(&ses->set_list, head);
1647}
1648
1649static void add_sits_in_set(struct f2fs_sb_info *sbi)
1650{
1651 struct f2fs_sm_info *sm_info = SM_I(sbi);
1652 struct list_head *set_list = &sm_info->sit_entry_set;
1653 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY
1654 unsigned int segno;
1655
7cd8558b 1656 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY
1657 add_sit_entry(segno, set_list);
1658}
1659
1660static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK
1661{
1662 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1663 struct f2fs_summary_block *sum = curseg->sum_blk;
1664 int i;
1665
184a5cd2
CY
1666 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1667 unsigned int segno;
1668 bool dirtied;
1669
1670 segno = le32_to_cpu(segno_in_journal(sum, i));
1671 dirtied = __mark_sit_entry_dirty(sbi, segno);
1672
1673 if (!dirtied)
1674 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1675 }
184a5cd2 1676 update_sits_in_cursum(sum, -sits_in_cursum(sum));
351df4b2
JK
1677}
1678
0a8165d7 1679/*
351df4b2
JK
1680 * CP calls this function, which flushes SIT entries including sit_journal,
1681 * and moves prefree segs to free segs.
1682 */
4b2fecc8 1683void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK
1684{
1685 struct sit_info *sit_i = SIT_I(sbi);
1686 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1687 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1688 struct f2fs_summary_block *sum = curseg->sum_blk;
184a5cd2
CY
1689 struct sit_entry_set *ses, *tmp;
1690 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1691 bool to_journal = true;
4b2fecc8 1692 struct seg_entry *se;
351df4b2
JK
1693
1694 mutex_lock(&curseg->curseg_mutex);
1695 mutex_lock(&sit_i->sentry_lock);
1696
1697 /*
184a5cd2
CY
1698 * add and account sit entries of dirty bitmap in sit entry
1699 * set temporarily
351df4b2 1700 */
184a5cd2 1701 add_sits_in_set(sbi);
351df4b2 1702
184a5cd2
CY
1703 /*
1704 * if there are no enough space in journal to store dirty sit
1705 * entries, remove all entries from journal and add and account
1706 * them in sit entry set.
1707 */
1708 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1709 remove_sits_in_journal(sbi);
b2955550 1710
184a5cd2
CY
1711 if (!sit_i->dirty_sentries)
1712 goto out;
351df4b2 1713
184a5cd2
CY
1714 /*
1715 * there are two steps to flush sit entries:
1716 * #1, flush sit entries to journal in current cold data summary block.
1717 * #2, flush sit entries to sit page.
1718 */
1719 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1720 struct page *page = NULL;
184a5cd2
CY
1721 struct f2fs_sit_block *raw_sit = NULL;
1722 unsigned int start_segno = ses->start_segno;
1723 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1724 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY
1725 unsigned int segno = start_segno;
1726
1727 if (to_journal &&
1728 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1729 to_journal = false;
1730
1731 if (!to_journal) {
1732 page = get_next_sit_page(sbi, start_segno);
1733 raw_sit = page_address(page);
351df4b2 1734 }
351df4b2 1735
184a5cd2
CY
1736 /* flush dirty sit entries in region of current sit set */
1737 for_each_set_bit_from(segno, bitmap, end) {
1738 int offset, sit_offset;
4b2fecc8
JK
1739
1740 se = get_seg_entry(sbi, segno);
184a5cd2
CY
1741
1742 /* add discard candidates */
4b2fecc8
JK
1743 if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards) {
1744 cpc->trim_start = segno;
1745 add_discard_addrs(sbi, cpc);
1746 }
184a5cd2
CY
1747
1748 if (to_journal) {
1749 offset = lookup_journal_in_cursum(sum,
1750 SIT_JOURNAL, segno, 1);
1751 f2fs_bug_on(sbi, offset < 0);
1752 segno_in_journal(sum, offset) =
1753 cpu_to_le32(segno);
1754 seg_info_to_raw_sit(se,
1755 &sit_in_journal(sum, offset));
1756 } else {
1757 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1758 seg_info_to_raw_sit(se,
1759 &raw_sit->entries[sit_offset]);
1760 }
351df4b2 1761
184a5cd2
CY
1762 __clear_bit(segno, bitmap);
1763 sit_i->dirty_sentries--;
1764 ses->entry_cnt--;
351df4b2
JK
1765 }
1766
184a5cd2
CY
1767 if (!to_journal)
1768 f2fs_put_page(page, 1);
1769
1770 f2fs_bug_on(sbi, ses->entry_cnt);
1771 release_sit_entry_set(ses);
351df4b2 1772 }
184a5cd2
CY
1773
1774 f2fs_bug_on(sbi, !list_empty(head));
1775 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 1776out:
4b2fecc8
JK
1777 if (cpc->reason == CP_DISCARD) {
1778 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1779 add_discard_addrs(sbi, cpc);
1780 }
351df4b2
JK
1781 mutex_unlock(&sit_i->sentry_lock);
1782 mutex_unlock(&curseg->curseg_mutex);
1783
351df4b2
JK
1784 set_prefree_as_free_segments(sbi);
1785}
1786
1787static int build_sit_info(struct f2fs_sb_info *sbi)
1788{
1789 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1790 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1791 struct sit_info *sit_i;
1792 unsigned int sit_segs, start;
1793 char *src_bitmap, *dst_bitmap;
1794 unsigned int bitmap_size;
1795
1796 /* allocate memory for SIT information */
1797 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1798 if (!sit_i)
1799 return -ENOMEM;
1800
1801 SM_I(sbi)->sit_info = sit_i;
1802
7cd8558b 1803 sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
351df4b2
JK
1804 if (!sit_i->sentries)
1805 return -ENOMEM;
1806
7cd8558b 1807 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK
1808 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1809 if (!sit_i->dirty_sentries_bitmap)
1810 return -ENOMEM;
1811
7cd8558b 1812 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK
1813 sit_i->sentries[start].cur_valid_map
1814 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1815 sit_i->sentries[start].ckpt_valid_map
1816 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1817 if (!sit_i->sentries[start].cur_valid_map
1818 || !sit_i->sentries[start].ckpt_valid_map)
1819 return -ENOMEM;
1820 }
1821
1822 if (sbi->segs_per_sec > 1) {
7cd8558b 1823 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
351df4b2
JK
1824 sizeof(struct sec_entry));
1825 if (!sit_i->sec_entries)
1826 return -ENOMEM;
1827 }
1828
1829 /* get information related with SIT */
1830 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1831
1832 /* setup SIT bitmap from ckeckpoint pack */
1833 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1834 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1835
79b5793b 1836 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK
1837 if (!dst_bitmap)
1838 return -ENOMEM;
351df4b2
JK
1839
1840 /* init SIT information */
1841 sit_i->s_ops = &default_salloc_ops;
1842
1843 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1844 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1845 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1846 sit_i->sit_bitmap = dst_bitmap;
1847 sit_i->bitmap_size = bitmap_size;
1848 sit_i->dirty_sentries = 0;
1849 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1850 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1851 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1852 mutex_init(&sit_i->sentry_lock);
1853 return 0;
1854}
1855
1856static int build_free_segmap(struct f2fs_sb_info *sbi)
1857{
351df4b2
JK
1858 struct free_segmap_info *free_i;
1859 unsigned int bitmap_size, sec_bitmap_size;
1860
1861 /* allocate memory for free segmap information */
1862 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1863 if (!free_i)
1864 return -ENOMEM;
1865
1866 SM_I(sbi)->free_info = free_i;
1867
7cd8558b 1868 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK
1869 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1870 if (!free_i->free_segmap)
1871 return -ENOMEM;
1872
7cd8558b 1873 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2
JK
1874 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1875 if (!free_i->free_secmap)
1876 return -ENOMEM;
1877
1878 /* set all segments as dirty temporarily */
1879 memset(free_i->free_segmap, 0xff, bitmap_size);
1880 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1881
1882 /* init free segmap information */
7cd8558b 1883 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK
1884 free_i->free_segments = 0;
1885 free_i->free_sections = 0;
1886 rwlock_init(&free_i->segmap_lock);
1887 return 0;
1888}
1889
1890static int build_curseg(struct f2fs_sb_info *sbi)
1891{
1042d60f 1892 struct curseg_info *array;
351df4b2
JK
1893 int i;
1894
b434babf 1895 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK
1896 if (!array)
1897 return -ENOMEM;
1898
1899 SM_I(sbi)->curseg_array = array;
1900
1901 for (i = 0; i < NR_CURSEG_TYPE; i++) {
1902 mutex_init(&array[i].curseg_mutex);
1903 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1904 if (!array[i].sum_blk)
1905 return -ENOMEM;
1906 array[i].segno = NULL_SEGNO;
1907 array[i].next_blkoff = 0;
1908 }
1909 return restore_curseg_summaries(sbi);
1910}
1911
1912static void build_sit_entries(struct f2fs_sb_info *sbi)
1913{
1914 struct sit_info *sit_i = SIT_I(sbi);
1915 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1916 struct f2fs_summary_block *sum = curseg->sum_blk;
74de593a
CY
1917 int sit_blk_cnt = SIT_BLK_CNT(sbi);
1918 unsigned int i, start, end;
1919 unsigned int readed, start_blk = 0;
90a893c7 1920 int nrpages = MAX_BIO_BLOCKS(sbi);
351df4b2 1921
74de593a 1922 do {
662befda 1923 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
74de593a
CY
1924
1925 start = start_blk * sit_i->sents_per_block;
1926 end = (start_blk + readed) * sit_i->sents_per_block;
1927
7cd8558b 1928 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY
1929 struct seg_entry *se = &sit_i->sentries[start];
1930 struct f2fs_sit_block *sit_blk;
1931 struct f2fs_sit_entry sit;
1932 struct page *page;
1933
1934 mutex_lock(&curseg->curseg_mutex);
1935 for (i = 0; i < sits_in_cursum(sum); i++) {
6c311ec6
CF
1936 if (le32_to_cpu(segno_in_journal(sum, i))
1937 == start) {
74de593a
CY
1938 sit = sit_in_journal(sum, i);
1939 mutex_unlock(&curseg->curseg_mutex);
1940 goto got_it;
1941 }
351df4b2 1942 }
74de593a
CY
1943 mutex_unlock(&curseg->curseg_mutex);
1944
1945 page = get_current_sit_page(sbi, start);
1946 sit_blk = (struct f2fs_sit_block *)page_address(page);
1947 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1948 f2fs_put_page(page, 1);
351df4b2 1949got_it:
74de593a
CY
1950 check_block_count(sbi, start, &sit);
1951 seg_info_from_raw_sit(se, &sit);
1952 if (sbi->segs_per_sec > 1) {
1953 struct sec_entry *e = get_sec_entry(sbi, start);
1954 e->valid_blocks += se->valid_blocks;
1955 }
351df4b2 1956 }
74de593a
CY
1957 start_blk += readed;
1958 } while (start_blk < sit_blk_cnt);
351df4b2
JK
1959}
1960
1961static void init_free_segmap(struct f2fs_sb_info *sbi)
1962{
1963 unsigned int start;
1964 int type;
1965
7cd8558b 1966 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK
1967 struct seg_entry *sentry = get_seg_entry(sbi, start);
1968 if (!sentry->valid_blocks)
1969 __set_free(sbi, start);
1970 }
1971
1972 /* set use the current segments */
1973 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1974 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1975 __set_test_and_inuse(sbi, curseg_t->segno);
1976 }
1977}
1978
1979static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1980{
1981 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1982 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 1983 unsigned int segno = 0, offset = 0;
351df4b2
JK
1984 unsigned short valid_blocks;
1985
8736fbf0 1986 while (1) {
351df4b2 1987 /* find dirty segment based on free segmap */
7cd8558b
JK
1988 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
1989 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK
1990 break;
1991 offset = segno + 1;
1992 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 1993 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 1994 continue;
ec325b52
JK
1995 if (valid_blocks > sbi->blocks_per_seg) {
1996 f2fs_bug_on(sbi, 1);
1997 continue;
1998 }
351df4b2
JK
1999 mutex_lock(&dirty_i->seglist_lock);
2000 __locate_dirty_segment(sbi, segno, DIRTY);
2001 mutex_unlock(&dirty_i->seglist_lock);
2002 }
2003}
2004
5ec4e49f 2005static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK
2006{
2007 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2008 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2009
5ec4e49f
JK
2010 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2011 if (!dirty_i->victim_secmap)
351df4b2
JK
2012 return -ENOMEM;
2013 return 0;
2014}
2015
2016static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2017{
2018 struct dirty_seglist_info *dirty_i;
2019 unsigned int bitmap_size, i;
2020
2021 /* allocate memory for dirty segments list information */
2022 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2023 if (!dirty_i)
2024 return -ENOMEM;
2025
2026 SM_I(sbi)->dirty_info = dirty_i;
2027 mutex_init(&dirty_i->seglist_lock);
2028
7cd8558b 2029 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK
2030
2031 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2032 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK
2033 if (!dirty_i->dirty_segmap[i])
2034 return -ENOMEM;
2035 }
2036
2037 init_dirty_segmap(sbi);
5ec4e49f 2038 return init_victim_secmap(sbi);
351df4b2
JK
2039}
2040
0a8165d7 2041/*
351df4b2
JK
2042 * Update min, max modified time for cost-benefit GC algorithm
2043 */
2044static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2045{
2046 struct sit_info *sit_i = SIT_I(sbi);
2047 unsigned int segno;
2048
2049 mutex_lock(&sit_i->sentry_lock);
2050
2051 sit_i->min_mtime = LLONG_MAX;
2052
7cd8558b 2053 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK
2054 unsigned int i;
2055 unsigned long long mtime = 0;
2056
2057 for (i = 0; i < sbi->segs_per_sec; i++)
2058 mtime += get_seg_entry(sbi, segno + i)->mtime;
2059
2060 mtime = div_u64(mtime, sbi->segs_per_sec);
2061
2062 if (sit_i->min_mtime > mtime)
2063 sit_i->min_mtime = mtime;
2064 }
2065 sit_i->max_mtime = get_mtime(sbi);
2066 mutex_unlock(&sit_i->sentry_lock);
2067}
2068
2069int build_segment_manager(struct f2fs_sb_info *sbi)
2070{
2071 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2072 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2073 struct f2fs_sm_info *sm_info;
351df4b2
JK
2074 int err;
2075
2076 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2077 if (!sm_info)
2078 return -ENOMEM;
2079
2080 /* init sm info */
2081 sbi->sm_info = sm_info;
351df4b2
JK
2082 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2083 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2084 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2085 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2086 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2087 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2088 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK
2089 sm_info->rec_prefree_segments = sm_info->main_segments *
2090 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2091 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2092 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2093 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2094
7fd9e544
JK
2095 INIT_LIST_HEAD(&sm_info->discard_list);
2096 sm_info->nr_discards = 0;
2097 sm_info->max_discards = 0;
2098
184a5cd2
CY
2099 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2100
b270ad6f 2101 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ
2102 err = create_flush_cmd_control(sbi);
2103 if (err)
a688b9d9 2104 return err;
6b4afdd7
JK
2105 }
2106
351df4b2
JK
2107 err = build_sit_info(sbi);
2108 if (err)
2109 return err;
2110 err = build_free_segmap(sbi);
2111 if (err)
2112 return err;
2113 err = build_curseg(sbi);
2114 if (err)
2115 return err;
2116
2117 /* reinit free segmap based on SIT */
2118 build_sit_entries(sbi);
2119
2120 init_free_segmap(sbi);
2121 err = build_dirty_segmap(sbi);
2122 if (err)
2123 return err;
2124
2125 init_min_max_mtime(sbi);
2126 return 0;
2127}
2128
2129static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2130 enum dirty_type dirty_type)
2131{
2132 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2133
2134 mutex_lock(&dirty_i->seglist_lock);
2135 kfree(dirty_i->dirty_segmap[dirty_type]);
2136 dirty_i->nr_dirty[dirty_type] = 0;
2137 mutex_unlock(&dirty_i->seglist_lock);
2138}
2139
5ec4e49f 2140static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK
2141{
2142 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5ec4e49f 2143 kfree(dirty_i->victim_secmap);
351df4b2
JK
2144}
2145
2146static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2147{
2148 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2149 int i;
2150
2151 if (!dirty_i)
2152 return;
2153
2154 /* discard pre-free/dirty segments list */
2155 for (i = 0; i < NR_DIRTY_TYPE; i++)
2156 discard_dirty_segmap(sbi, i);
2157
5ec4e49f 2158 destroy_victim_secmap(sbi);
351df4b2
JK
2159 SM_I(sbi)->dirty_info = NULL;
2160 kfree(dirty_i);
2161}
2162
2163static void destroy_curseg(struct f2fs_sb_info *sbi)
2164{
2165 struct curseg_info *array = SM_I(sbi)->curseg_array;
2166 int i;
2167
2168 if (!array)
2169 return;
2170 SM_I(sbi)->curseg_array = NULL;
2171 for (i = 0; i < NR_CURSEG_TYPE; i++)
2172 kfree(array[i].sum_blk);
2173 kfree(array);
2174}
2175
2176static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2177{
2178 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2179 if (!free_i)
2180 return;
2181 SM_I(sbi)->free_info = NULL;
2182 kfree(free_i->free_segmap);
2183 kfree(free_i->free_secmap);
2184 kfree(free_i);
2185}
2186
2187static void destroy_sit_info(struct f2fs_sb_info *sbi)
2188{
2189 struct sit_info *sit_i = SIT_I(sbi);
2190 unsigned int start;
2191
2192 if (!sit_i)
2193 return;
2194
2195 if (sit_i->sentries) {
7cd8558b 2196 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK
2197 kfree(sit_i->sentries[start].cur_valid_map);
2198 kfree(sit_i->sentries[start].ckpt_valid_map);
2199 }
2200 }
2201 vfree(sit_i->sentries);
2202 vfree(sit_i->sec_entries);
2203 kfree(sit_i->dirty_sentries_bitmap);
2204
2205 SM_I(sbi)->sit_info = NULL;
2206 kfree(sit_i->sit_bitmap);
2207 kfree(sit_i);
2208}
2209
2210void destroy_segment_manager(struct f2fs_sb_info *sbi)
2211{
2212 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2213
3b03f724
CY
2214 if (!sm_info)
2215 return;
2163d198 2216 destroy_flush_cmd_control(sbi);
351df4b2
JK
2217 destroy_dirty_segmap(sbi);
2218 destroy_curseg(sbi);
2219 destroy_free_segmap(sbi);
2220 destroy_sit_info(sbi);
2221 sbi->sm_info = NULL;
2222 kfree(sm_info);
2223}
7fd9e544
JK
2224
2225int __init create_segment_manager_caches(void)
2226{
2227 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2228 sizeof(struct discard_entry));
7fd9e544 2229 if (!discard_entry_slab)
184a5cd2
CY
2230 goto fail;
2231
2232 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2233 sizeof(struct nat_entry_set));
2234 if (!sit_entry_set_slab)
2235 goto destory_discard_entry;
88b88a66
JK
2236
2237 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2238 sizeof(struct inmem_pages));
2239 if (!inmem_entry_slab)
2240 goto destroy_sit_entry_set;
7fd9e544 2241 return 0;
184a5cd2 2242
88b88a66
JK
2243destroy_sit_entry_set:
2244 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY
2245destory_discard_entry:
2246 kmem_cache_destroy(discard_entry_slab);
2247fail:
2248 return -ENOMEM;
7fd9e544
JK
2249}
2250
2251void destroy_segment_manager_caches(void)
2252{
184a5cd2 2253 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2254 kmem_cache_destroy(discard_entry_slab);
88b88a66 2255 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2256}