projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: add MAINTAINERS entry
[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>
351df4b2
JK		 16#include <linux/vmalloc.h>
17
18#include "f2fs.h"
19#include "segment.h"
20#include "node.h"
21
22static int need_to_flush(struct f2fs_sb_info *sbi)
23{
24 unsigned int pages_per_sec = (1 << sbi->log_blocks_per_seg) *
25 sbi->segs_per_sec;
26 int node_secs = ((get_pages(sbi, F2FS_DIRTY_NODES) + pages_per_sec - 1)
27 >> sbi->log_blocks_per_seg) / sbi->segs_per_sec;
28 int dent_secs = ((get_pages(sbi, F2FS_DIRTY_DENTS) + pages_per_sec - 1)
29 >> sbi->log_blocks_per_seg) / sbi->segs_per_sec;
30
31 if (sbi->por_doing)
32 return 0;
33
34 if (free_sections(sbi) <= (node_secs + 2 * dent_secs +
35 reserved_sections(sbi)))
36 return 1;
37 return 0;
38}
39
0a8165d7 40/*
351df4b2
JK		 41 * This function balances dirty node and dentry pages.
42 * In addition, it controls garbage collection.
43 */
44void f2fs_balance_fs(struct f2fs_sb_info *sbi)
45{
46 struct writeback_control wbc = {
47 .sync_mode = WB_SYNC_ALL,
48 .nr_to_write = LONG_MAX,
49 .for_reclaim = 0,
50 };
51
52 if (sbi->por_doing)
53 return;
54
55 /*
56 * We should do checkpoint when there are so many dirty node pages
57 * with enough free segments. After then, we should do GC.
58 */
59 if (need_to_flush(sbi)) {
60 sync_dirty_dir_inodes(sbi);
61 sync_node_pages(sbi, 0, &wbc);
62 }
63
64 if (has_not_enough_free_secs(sbi)) {
65 mutex_lock(&sbi->gc_mutex);
66 f2fs_gc(sbi, 1);
67 }
68}
69
70static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
71 enum dirty_type dirty_type)
72{
73 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
74
75 /* need not be added */
76 if (IS_CURSEG(sbi, segno))
77 return;
78
79 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
80 dirty_i->nr_dirty[dirty_type]++;
81
82 if (dirty_type == DIRTY) {
83 struct seg_entry *sentry = get_seg_entry(sbi, segno);
84 dirty_type = sentry->type;
85 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
86 dirty_i->nr_dirty[dirty_type]++;
87 }
88}
89
90static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
91 enum dirty_type dirty_type)
92{
93 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
94
95 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
96 dirty_i->nr_dirty[dirty_type]--;
97
98 if (dirty_type == DIRTY) {
99 struct seg_entry *sentry = get_seg_entry(sbi, segno);
100 dirty_type = sentry->type;
101 if (test_and_clear_bit(segno,
102 dirty_i->dirty_segmap[dirty_type]))
103 dirty_i->nr_dirty[dirty_type]--;
104 clear_bit(segno, dirty_i->victim_segmap[FG_GC]);
105 clear_bit(segno, dirty_i->victim_segmap[BG_GC]);
106 }
107}
108
0a8165d7 109/*
351df4b2
JK		 110 * Should not occur error such as -ENOMEM.
111 * Adding dirty entry into seglist is not critical operation.
112 * If a given segment is one of current working segments, it won't be added.
113 */
114void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
115{
116 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
117 unsigned short valid_blocks;
118
119 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
120 return;
121
122 mutex_lock(&dirty_i->seglist_lock);
123
124 valid_blocks = get_valid_blocks(sbi, segno, 0);
125
126 if (valid_blocks == 0) {
127 __locate_dirty_segment(sbi, segno, PRE);
128 __remove_dirty_segment(sbi, segno, DIRTY);
129 } else if (valid_blocks < sbi->blocks_per_seg) {
130 __locate_dirty_segment(sbi, segno, DIRTY);
131 } else {
132 /* Recovery routine with SSR needs this */
133 __remove_dirty_segment(sbi, segno, DIRTY);
134 }
135
136 mutex_unlock(&dirty_i->seglist_lock);
137 return;
138}
139
0a8165d7 140/*
351df4b2
JK		 141 * Should call clear_prefree_segments after checkpoint is done.
142 */
143static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
144{
145 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
146 unsigned int segno, offset = 0;
147 unsigned int total_segs = TOTAL_SEGS(sbi);
148
149 mutex_lock(&dirty_i->seglist_lock);
150 while (1) {
151 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
152 offset);
153 if (segno >= total_segs)
154 break;
155 __set_test_and_free(sbi, segno);
156 offset = segno + 1;
157 }
158 mutex_unlock(&dirty_i->seglist_lock);
159}
160
161void clear_prefree_segments(struct f2fs_sb_info *sbi)
162{
163 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
164 unsigned int segno, offset = 0;
165 unsigned int total_segs = TOTAL_SEGS(sbi);
166
167 mutex_lock(&dirty_i->seglist_lock);
168 while (1) {
169 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
170 offset);
171 if (segno >= total_segs)
172 break;
173
174 offset = segno + 1;
175 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
176 dirty_i->nr_dirty[PRE]--;
177
178 /* Let's use trim */
179 if (test_opt(sbi, DISCARD))
180 blkdev_issue_discard(sbi->sb->s_bdev,
181 START_BLOCK(sbi, segno) <<
182 sbi->log_sectors_per_block,
183 1 << (sbi->log_sectors_per_block +
184 sbi->log_blocks_per_seg),
185 GFP_NOFS, 0);
186 }
187 mutex_unlock(&dirty_i->seglist_lock);
188}
189
190static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
191{
192 struct sit_info *sit_i = SIT_I(sbi);
193 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
194 sit_i->dirty_sentries++;
195}
196
197static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
198 unsigned int segno, int modified)
199{
200 struct seg_entry *se = get_seg_entry(sbi, segno);
201 se->type = type;
202 if (modified)
203 __mark_sit_entry_dirty(sbi, segno);
204}
205
206static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
207{
208 struct seg_entry *se;
209 unsigned int segno, offset;
210 long int new_vblocks;
211
212 segno = GET_SEGNO(sbi, blkaddr);
213
214 se = get_seg_entry(sbi, segno);
215 new_vblocks = se->valid_blocks + del;
216 offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
217
218 BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) ||
219 (new_vblocks > sbi->blocks_per_seg)));
220
221 se->valid_blocks = new_vblocks;
222 se->mtime = get_mtime(sbi);
223 SIT_I(sbi)->max_mtime = se->mtime;
224
225 /* Update valid block bitmap */
226 if (del > 0) {
227 if (f2fs_set_bit(offset, se->cur_valid_map))
228 BUG();
229 } else {
230 if (!f2fs_clear_bit(offset, se->cur_valid_map))
231 BUG();
232 }
233 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
234 se->ckpt_valid_blocks += del;
235
236 __mark_sit_entry_dirty(sbi, segno);
237
238 /* update total number of valid blocks to be written in ckpt area */
239 SIT_I(sbi)->written_valid_blocks += del;
240
241 if (sbi->segs_per_sec > 1)
242 get_sec_entry(sbi, segno)->valid_blocks += del;
243}
244
245static void refresh_sit_entry(struct f2fs_sb_info *sbi,
246 block_t old_blkaddr, block_t new_blkaddr)
247{
248 update_sit_entry(sbi, new_blkaddr, 1);
249 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
250 update_sit_entry(sbi, old_blkaddr, -1);
251}
252
253void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
254{
255 unsigned int segno = GET_SEGNO(sbi, addr);
256 struct sit_info *sit_i = SIT_I(sbi);
257
258 BUG_ON(addr == NULL_ADDR);
259 if (addr == NEW_ADDR)
260 return;
261
262 /* add it into sit main buffer */
263 mutex_lock(&sit_i->sentry_lock);
264
265 update_sit_entry(sbi, addr, -1);
266
267 /* add it into dirty seglist */
268 locate_dirty_segment(sbi, segno);
269
270 mutex_unlock(&sit_i->sentry_lock);
271}
272
0a8165d7 273/*
351df4b2
JK		 274 * This function should be resided under the curseg_mutex lock
275 */
276static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
277 struct f2fs_summary *sum, unsigned short offset)
278{
279 struct curseg_info *curseg = CURSEG_I(sbi, type);
280 void *addr = curseg->sum_blk;
281 addr += offset * sizeof(struct f2fs_summary);
282 memcpy(addr, sum, sizeof(struct f2fs_summary));
283 return;
284}
285
0a8165d7 286/*
351df4b2
JK		 287 * Calculate the number of current summary pages for writing
288 */
289int npages_for_summary_flush(struct f2fs_sb_info *sbi)
290{
291 int total_size_bytes = 0;
292 int valid_sum_count = 0;
293 int i, sum_space;
294
295 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
296 if (sbi->ckpt->alloc_type[i] == SSR)
297 valid_sum_count += sbi->blocks_per_seg;
298 else
299 valid_sum_count += curseg_blkoff(sbi, i);
300 }
301
302 total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1)
303 + sizeof(struct nat_journal) + 2
304 + sizeof(struct sit_journal) + 2;
305 sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE;
306 if (total_size_bytes < sum_space)
307 return 1;
308 else if (total_size_bytes < 2 * sum_space)
309 return 2;
310 return 3;
311}
312
0a8165d7 313/*
351df4b2
JK		 314 * Caller should put this summary page
315 */
316struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
317{
318 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
319}
320
321static void write_sum_page(struct f2fs_sb_info *sbi,
322 struct f2fs_summary_block *sum_blk, block_t blk_addr)
323{
324 struct page *page = grab_meta_page(sbi, blk_addr);
325 void *kaddr = page_address(page);
326 memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
327 set_page_dirty(page);
328 f2fs_put_page(page, 1);
329}
330
331static unsigned int check_prefree_segments(struct f2fs_sb_info *sbi,
332 int ofs_unit, int type)
333{
334 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
335 unsigned long *prefree_segmap = dirty_i->dirty_segmap[PRE];
336 unsigned int segno, next_segno, i;
337 int ofs = 0;
338
339 /*
340 * If there is not enough reserved sections,
341 * we should not reuse prefree segments.
342 */
343 if (has_not_enough_free_secs(sbi))
344 return NULL_SEGNO;
345
346 /*
347 * NODE page should not reuse prefree segment,
348 * since those information is used for SPOR.
349 */
350 if (IS_NODESEG(type))
351 return NULL_SEGNO;
352next:
353 segno = find_next_bit(prefree_segmap, TOTAL_SEGS(sbi), ofs++);
354 ofs = ((segno / ofs_unit) * ofs_unit) + ofs_unit;
355 if (segno < TOTAL_SEGS(sbi)) {
356 /* skip intermediate segments in a section */
357 if (segno % ofs_unit)
358 goto next;
359
360 /* skip if whole section is not prefree */
361 next_segno = find_next_zero_bit(prefree_segmap,
362 TOTAL_SEGS(sbi), segno + 1);
363 if (next_segno - segno < ofs_unit)
364 goto next;
365
366 /* skip if whole section was not free at the last checkpoint */
367 for (i = 0; i < ofs_unit; i++)
368 if (get_seg_entry(sbi, segno)->ckpt_valid_blocks)
369 goto next;
370 return segno;
371 }
372 return NULL_SEGNO;
373}
374
0a8165d7 375/*
351df4b2
JK		 376 * Find a new segment from the free segments bitmap to right order
377 * This function should be returned with success, otherwise BUG
378 */
379static void get_new_segment(struct f2fs_sb_info *sbi,
380 unsigned int *newseg, bool new_sec, int dir)
381{
382 struct free_segmap_info *free_i = FREE_I(sbi);
383 unsigned int total_secs = sbi->total_sections;
384 unsigned int segno, secno, zoneno;
385 unsigned int total_zones = sbi->total_sections / sbi->secs_per_zone;
386 unsigned int hint = *newseg / sbi->segs_per_sec;
387 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
388 unsigned int left_start = hint;
389 bool init = true;
390 int go_left = 0;
391 int i;
392
393 write_lock(&free_i->segmap_lock);
394
395 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
396 segno = find_next_zero_bit(free_i->free_segmap,
397 TOTAL_SEGS(sbi), *newseg + 1);
398 if (segno < TOTAL_SEGS(sbi))
399 goto got_it;
400 }
401find_other_zone:
402 secno = find_next_zero_bit(free_i->free_secmap, total_secs, hint);
403 if (secno >= total_secs) {
404 if (dir == ALLOC_RIGHT) {
405 secno = find_next_zero_bit(free_i->free_secmap,
406 total_secs, 0);
407 BUG_ON(secno >= total_secs);
408 } else {
409 go_left = 1;
410 left_start = hint - 1;
411 }
412 }
413 if (go_left == 0)
414 goto skip_left;
415
416 while (test_bit(left_start, free_i->free_secmap)) {
417 if (left_start > 0) {
418 left_start--;
419 continue;
420 }
421 left_start = find_next_zero_bit(free_i->free_secmap,
422 total_secs, 0);
423 BUG_ON(left_start >= total_secs);
424 break;
425 }
426 secno = left_start;
427skip_left:
428 hint = secno;
429 segno = secno * sbi->segs_per_sec;
430 zoneno = secno / sbi->secs_per_zone;
431
432 /* give up on finding another zone */
433 if (!init)
434 goto got_it;
435 if (sbi->secs_per_zone == 1)
436 goto got_it;
437 if (zoneno == old_zoneno)
438 goto got_it;
439 if (dir == ALLOC_LEFT) {
440 if (!go_left && zoneno + 1 >= total_zones)
441 goto got_it;
442 if (go_left && zoneno == 0)
443 goto got_it;
444 }
445 for (i = 0; i < NR_CURSEG_TYPE; i++)
446 if (CURSEG_I(sbi, i)->zone == zoneno)
447 break;
448
449 if (i < NR_CURSEG_TYPE) {
450 /* zone is in user, try another */
451 if (go_left)
452 hint = zoneno * sbi->secs_per_zone - 1;
453 else if (zoneno + 1 >= total_zones)
454 hint = 0;
455 else
456 hint = (zoneno + 1) * sbi->secs_per_zone;
457 init = false;
458 goto find_other_zone;
459 }
460got_it:
461 /* set it as dirty segment in free segmap */
462 BUG_ON(test_bit(segno, free_i->free_segmap));
463 __set_inuse(sbi, segno);
464 *newseg = segno;
465 write_unlock(&free_i->segmap_lock);
466}
467
468static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
469{
470 struct curseg_info *curseg = CURSEG_I(sbi, type);
471 struct summary_footer *sum_footer;
472
473 curseg->segno = curseg->next_segno;
474 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
475 curseg->next_blkoff = 0;
476 curseg->next_segno = NULL_SEGNO;
477
478 sum_footer = &(curseg->sum_blk->footer);
479 memset(sum_footer, 0, sizeof(struct summary_footer));
480 if (IS_DATASEG(type))
481 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
482 if (IS_NODESEG(type))
483 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
484 __set_sit_entry_type(sbi, type, curseg->segno, modified);
485}
486
0a8165d7 487/*
351df4b2
JK		 488 * Allocate a current working segment.
489 * This function always allocates a free segment in LFS manner.
490 */
491static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
492{
493 struct curseg_info *curseg = CURSEG_I(sbi, type);
494 unsigned int segno = curseg->segno;
495 int dir = ALLOC_LEFT;
496
497 write_sum_page(sbi, curseg->sum_blk,
498 GET_SUM_BLOCK(sbi, curseg->segno));
499 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
500 dir = ALLOC_RIGHT;
501
502 if (test_opt(sbi, NOHEAP))
503 dir = ALLOC_RIGHT;
504
505 get_new_segment(sbi, &segno, new_sec, dir);
506 curseg->next_segno = segno;
507 reset_curseg(sbi, type, 1);
508 curseg->alloc_type = LFS;
509}
510
511static void __next_free_blkoff(struct f2fs_sb_info *sbi,
512 struct curseg_info *seg, block_t start)
513{
514 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
515 block_t ofs;
516 for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
517 if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
518 && !f2fs_test_bit(ofs, se->cur_valid_map))
519 break;
520 }
521 seg->next_blkoff = ofs;
522}
523
0a8165d7 524/*
351df4b2
JK		 525 * If a segment is written by LFS manner, next block offset is just obtained
526 * by increasing the current block offset. However, if a segment is written by
527 * SSR manner, next block offset obtained by calling __next_free_blkoff
528 */
529static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
530 struct curseg_info *seg)
531{
532 if (seg->alloc_type == SSR)
533 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
534 else
535 seg->next_blkoff++;
536}
537
0a8165d7 538/*
351df4b2
JK		 539 * This function always allocates a used segment (from dirty seglist) by SSR
540 * manner, so it should recover the existing segment information of valid blocks
541 */
542static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
543{
544 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
545 struct curseg_info *curseg = CURSEG_I(sbi, type);
546 unsigned int new_segno = curseg->next_segno;
547 struct f2fs_summary_block *sum_node;
548 struct page *sum_page;
549
550 write_sum_page(sbi, curseg->sum_blk,
551 GET_SUM_BLOCK(sbi, curseg->segno));
552 __set_test_and_inuse(sbi, new_segno);
553
554 mutex_lock(&dirty_i->seglist_lock);
555 __remove_dirty_segment(sbi, new_segno, PRE);
556 __remove_dirty_segment(sbi, new_segno, DIRTY);
557 mutex_unlock(&dirty_i->seglist_lock);
558
559 reset_curseg(sbi, type, 1);
560 curseg->alloc_type = SSR;
561 __next_free_blkoff(sbi, curseg, 0);
562
563 if (reuse) {
564 sum_page = get_sum_page(sbi, new_segno);
565 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
566 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
567 f2fs_put_page(sum_page, 1);
568 }
569}
570
571/*
572 * flush out current segment and replace it with new segment
573 * This function should be returned with success, otherwise BUG
574 */
575static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
576 int type, bool force)
577{
578 struct curseg_info *curseg = CURSEG_I(sbi, type);
579 unsigned int ofs_unit;
580
581 if (force) {
582 new_curseg(sbi, type, true);
583 goto out;
584 }
585
586 ofs_unit = need_SSR(sbi) ? 1 : sbi->segs_per_sec;
587 curseg->next_segno = check_prefree_segments(sbi, ofs_unit, type);
588
589 if (curseg->next_segno != NULL_SEGNO)
590 change_curseg(sbi, type, false);
591 else if (type == CURSEG_WARM_NODE)
592 new_curseg(sbi, type, false);
593 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
594 change_curseg(sbi, type, true);
595 else
596 new_curseg(sbi, type, false);
597out:
598 sbi->segment_count[curseg->alloc_type]++;
599}
600
601void allocate_new_segments(struct f2fs_sb_info *sbi)
602{
603 struct curseg_info *curseg;
604 unsigned int old_curseg;
605 int i;
606
607 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
608 curseg = CURSEG_I(sbi, i);
609 old_curseg = curseg->segno;
610 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
611 locate_dirty_segment(sbi, old_curseg);
612 }
613}
614
615static const struct segment_allocation default_salloc_ops = {
616 .allocate_segment = allocate_segment_by_default,
617};
618
619static void f2fs_end_io_write(struct bio *bio, int err)
620{
621 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
622 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
623 struct bio_private *p = bio->bi_private;
624
625 do {
626 struct page *page = bvec->bv_page;
627
628 if (--bvec >= bio->bi_io_vec)
629 prefetchw(&bvec->bv_page->flags);
630 if (!uptodate) {
631 SetPageError(page);
632 if (page->mapping)
633 set_bit(AS_EIO, &page->mapping->flags);
25ca923b 634 set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
351df4b2
JK		 635 }
636 end_page_writeback(page);
637 dec_page_count(p->sbi, F2FS_WRITEBACK);
638 } while (bvec >= bio->bi_io_vec);
639
640 if (p->is_sync)
641 complete(p->wait);
642 kfree(p);
643 bio_put(bio);
644}
645
3cd8a239 646struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
351df4b2
JK		 647{
648 struct bio *bio;
3cd8a239 649 struct bio_private *priv;
351df4b2 650retry:
3cd8a239
JK		 651 priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
652 if (!priv) {
351df4b2 653 cond_resched();
c212991a 654 goto retry;
351df4b2 655 }
3cd8a239
JK		 656
657 /* No failure on bio allocation */
658 bio = bio_alloc(GFP_NOIO, npages);
659 bio->bi_bdev = bdev;
660 bio->bi_private = priv;
351df4b2
JK		 661 return bio;
662}
663
664static void do_submit_bio(struct f2fs_sb_info *sbi,
665 enum page_type type, bool sync)
666{
667 int rw = sync ? WRITE_SYNC : WRITE;
668 enum page_type btype = type > META ? META : type;
669
670 if (type >= META_FLUSH)
671 rw = WRITE_FLUSH_FUA;
672
673 if (sbi->bio[btype]) {
674 struct bio_private *p = sbi->bio[btype]->bi_private;
675 p->sbi = sbi;
676 sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
677 if (type == META_FLUSH) {
678 DECLARE_COMPLETION_ONSTACK(wait);
679 p->is_sync = true;
680 p->wait = &wait;
681 submit_bio(rw, sbi->bio[btype]);
682 wait_for_completion(&wait);
683 } else {
684 p->is_sync = false;
685 submit_bio(rw, sbi->bio[btype]);
686 }
687 sbi->bio[btype] = NULL;
688 }
689}
690
691void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
692{
693 down_write(&sbi->bio_sem);
694 do_submit_bio(sbi, type, sync);
695 up_write(&sbi->bio_sem);
696}
697
698static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
699 block_t blk_addr, enum page_type type)
700{
701 struct block_device *bdev = sbi->sb->s_bdev;
702
703 verify_block_addr(sbi, blk_addr);
704
705 down_write(&sbi->bio_sem);
706
707 inc_page_count(sbi, F2FS_WRITEBACK);
708
709 if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
710 do_submit_bio(sbi, type, false);
711alloc_new:
3cd8a239
JK		 712 if (sbi->bio[type] == NULL) {
713 sbi->bio[type] = f2fs_bio_alloc(bdev, bio_get_nr_vecs(bdev));
714 sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
715 /*
716 * The end_io will be assigned at the sumbission phase.
717 * Until then, let bio_add_page() merge consecutive IOs as much
718 * as possible.
719 */
720 }
351df4b2
JK		 721
722 if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
723 PAGE_CACHE_SIZE) {
724 do_submit_bio(sbi, type, false);
725 goto alloc_new;
726 }
727
728 sbi->last_block_in_bio[type] = blk_addr;
729
730 up_write(&sbi->bio_sem);
731}
732
733static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
734{
735 struct curseg_info *curseg = CURSEG_I(sbi, type);
736 if (curseg->next_blkoff < sbi->blocks_per_seg)
737 return true;
738 return false;
739}
740
741static int __get_segment_type_2(struct page *page, enum page_type p_type)
742{
743 if (p_type == DATA)
744 return CURSEG_HOT_DATA;
745 else
746 return CURSEG_HOT_NODE;
747}
748
749static int __get_segment_type_4(struct page *page, enum page_type p_type)
750{
751 if (p_type == DATA) {
752 struct inode *inode = page->mapping->host;
753
754 if (S_ISDIR(inode->i_mode))
755 return CURSEG_HOT_DATA;
756 else
757 return CURSEG_COLD_DATA;
758 } else {
759 if (IS_DNODE(page) && !is_cold_node(page))
760 return CURSEG_HOT_NODE;
761 else
762 return CURSEG_COLD_NODE;
763 }
764}
765
766static int __get_segment_type_6(struct page *page, enum page_type p_type)
767{
768 if (p_type == DATA) {
769 struct inode *inode = page->mapping->host;
770
771 if (S_ISDIR(inode->i_mode))
772 return CURSEG_HOT_DATA;
773 else if (is_cold_data(page) || is_cold_file(inode))
774 return CURSEG_COLD_DATA;
775 else
776 return CURSEG_WARM_DATA;
777 } else {
778 if (IS_DNODE(page))
779 return is_cold_node(page) ? CURSEG_WARM_NODE :
780 CURSEG_HOT_NODE;
781 else
782 return CURSEG_COLD_NODE;
783 }
784}
785
786static int __get_segment_type(struct page *page, enum page_type p_type)
787{
788 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
789 switch (sbi->active_logs) {
790 case 2:
791 return __get_segment_type_2(page, p_type);
792 case 4:
793 return __get_segment_type_4(page, p_type);
351df4b2 794 }
12a67146
JK		 795 /* NR_CURSEG_TYPE(6) logs by default */
796 BUG_ON(sbi->active_logs != NR_CURSEG_TYPE);
797 return __get_segment_type_6(page, p_type);
351df4b2
JK		 798}
799
800static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
801 block_t old_blkaddr, block_t *new_blkaddr,
802 struct f2fs_summary *sum, enum page_type p_type)
803{
804 struct sit_info *sit_i = SIT_I(sbi);
805 struct curseg_info *curseg;
806 unsigned int old_cursegno;
807 int type;
808
809 type = __get_segment_type(page, p_type);
810 curseg = CURSEG_I(sbi, type);
811
812 mutex_lock(&curseg->curseg_mutex);
813
814 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
815 old_cursegno = curseg->segno;
816
817 /*
818 * __add_sum_entry should be resided under the curseg_mutex
819 * because, this function updates a summary entry in the
820 * current summary block.
821 */
822 __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
823
824 mutex_lock(&sit_i->sentry_lock);
825 __refresh_next_blkoff(sbi, curseg);
826 sbi->block_count[curseg->alloc_type]++;
827
828 /*
829 * SIT information should be updated before segment allocation,
830 * since SSR needs latest valid block information.
831 */
832 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
833
834 if (!__has_curseg_space(sbi, type))
835 sit_i->s_ops->allocate_segment(sbi, type, false);
836
837 locate_dirty_segment(sbi, old_cursegno);
838 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
839 mutex_unlock(&sit_i->sentry_lock);
840
841 if (p_type == NODE)
842 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
843
844 /* writeout dirty page into bdev */
845 submit_write_page(sbi, page, *new_blkaddr, p_type);
846
847 mutex_unlock(&curseg->curseg_mutex);
848}
849
850int write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
851 struct writeback_control *wbc)
852{
853 if (wbc->for_reclaim)
854 return AOP_WRITEPAGE_ACTIVATE;
855
856 set_page_writeback(page);
857 submit_write_page(sbi, page, page->index, META);
858 return 0;
859}
860
861void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
862 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
863{
864 struct f2fs_summary sum;
865 set_summary(&sum, nid, 0, 0);
866 do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
867}
868
869void write_data_page(struct inode *inode, struct page *page,
870 struct dnode_of_data *dn, block_t old_blkaddr,
871 block_t *new_blkaddr)
872{
873 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
874 struct f2fs_summary sum;
875 struct node_info ni;
876
877 BUG_ON(old_blkaddr == NULL_ADDR);
878 get_node_info(sbi, dn->nid, &ni);
879 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
880
881 do_write_page(sbi, page, old_blkaddr,
882 new_blkaddr, &sum, DATA);
883}
884
885void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
886 block_t old_blk_addr)
887{
888 submit_write_page(sbi, page, old_blk_addr, DATA);
889}
890
891void recover_data_page(struct f2fs_sb_info *sbi,
892 struct page *page, struct f2fs_summary *sum,
893 block_t old_blkaddr, block_t new_blkaddr)
894{
895 struct sit_info *sit_i = SIT_I(sbi);
896 struct curseg_info *curseg;
897 unsigned int segno, old_cursegno;
898 struct seg_entry *se;
899 int type;
900
901 segno = GET_SEGNO(sbi, new_blkaddr);
902 se = get_seg_entry(sbi, segno);
903 type = se->type;
904
905 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
906 if (old_blkaddr == NULL_ADDR)
907 type = CURSEG_COLD_DATA;
908 else
909 type = CURSEG_WARM_DATA;
910 }
911 curseg = CURSEG_I(sbi, type);
912
913 mutex_lock(&curseg->curseg_mutex);
914 mutex_lock(&sit_i->sentry_lock);
915
916 old_cursegno = curseg->segno;
917
918 /* change the current segment */
919 if (segno != curseg->segno) {
920 curseg->next_segno = segno;
921 change_curseg(sbi, type, true);
922 }
923
924 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
925 (sbi->blocks_per_seg - 1);
926 __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
927
928 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
929
930 locate_dirty_segment(sbi, old_cursegno);
931 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
932
933 mutex_unlock(&sit_i->sentry_lock);
934 mutex_unlock(&curseg->curseg_mutex);
935}
936
937void rewrite_node_page(struct f2fs_sb_info *sbi,
938 struct page *page, struct f2fs_summary *sum,
939 block_t old_blkaddr, block_t new_blkaddr)
940{
941 struct sit_info *sit_i = SIT_I(sbi);
942 int type = CURSEG_WARM_NODE;
943 struct curseg_info *curseg;
944 unsigned int segno, old_cursegno;
945 block_t next_blkaddr = next_blkaddr_of_node(page);
946 unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
947
948 curseg = CURSEG_I(sbi, type);
949
950 mutex_lock(&curseg->curseg_mutex);
951 mutex_lock(&sit_i->sentry_lock);
952
953 segno = GET_SEGNO(sbi, new_blkaddr);
954 old_cursegno = curseg->segno;
955
956 /* change the current segment */
957 if (segno != curseg->segno) {
958 curseg->next_segno = segno;
959 change_curseg(sbi, type, true);
960 }
961 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
962 (sbi->blocks_per_seg - 1);
963 __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
964
965 /* change the current log to the next block addr in advance */
966 if (next_segno != segno) {
967 curseg->next_segno = next_segno;
968 change_curseg(sbi, type, true);
969 }
970 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
971 (sbi->blocks_per_seg - 1);
972
973 /* rewrite node page */
974 set_page_writeback(page);
975 submit_write_page(sbi, page, new_blkaddr, NODE);
976 f2fs_submit_bio(sbi, NODE, true);
977 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
978
979 locate_dirty_segment(sbi, old_cursegno);
980 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
981
982 mutex_unlock(&sit_i->sentry_lock);
983 mutex_unlock(&curseg->curseg_mutex);
984}
985
986static int read_compacted_summaries(struct f2fs_sb_info *sbi)
987{
988 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
989 struct curseg_info *seg_i;
990 unsigned char *kaddr;
991 struct page *page;
992 block_t start;
993 int i, j, offset;
994
995 start = start_sum_block(sbi);
996
997 page = get_meta_page(sbi, start++);
998 kaddr = (unsigned char *)page_address(page);
999
1000 /* Step 1: restore nat cache */
1001 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1002 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1003
1004 /* Step 2: restore sit cache */
1005 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1006 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1007 SUM_JOURNAL_SIZE);
1008 offset = 2 * SUM_JOURNAL_SIZE;
1009
1010 /* Step 3: restore summary entries */
1011 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1012 unsigned short blk_off;
1013 unsigned int segno;
1014
1015 seg_i = CURSEG_I(sbi, i);
1016 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1017 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1018 seg_i->next_segno = segno;
1019 reset_curseg(sbi, i, 0);
1020 seg_i->alloc_type = ckpt->alloc_type[i];
1021 seg_i->next_blkoff = blk_off;
1022
1023 if (seg_i->alloc_type == SSR)
1024 blk_off = sbi->blocks_per_seg;
1025
1026 for (j = 0; j < blk_off; j++) {
1027 struct f2fs_summary *s;
1028 s = (struct f2fs_summary *)(kaddr + offset);
1029 seg_i->sum_blk->entries[j] = *s;
1030 offset += SUMMARY_SIZE;
1031 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1032 SUM_FOOTER_SIZE)
1033 continue;
1034
1035 f2fs_put_page(page, 1);
1036 page = NULL;
1037
1038 page = get_meta_page(sbi, start++);
1039 kaddr = (unsigned char *)page_address(page);
1040 offset = 0;
1041 }
1042 }
1043 f2fs_put_page(page, 1);
1044 return 0;
1045}
1046
1047static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1048{
1049 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1050 struct f2fs_summary_block *sum;
1051 struct curseg_info *curseg;
1052 struct page *new;
1053 unsigned short blk_off;
1054 unsigned int segno = 0;
1055 block_t blk_addr = 0;
1056
1057 /* get segment number and block addr */
1058 if (IS_DATASEG(type)) {
1059 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1060 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1061 CURSEG_HOT_DATA]);
25ca923b 1062 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
351df4b2
JK		 1063 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1064 else
1065 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1066 } else {
1067 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1068 CURSEG_HOT_NODE]);
1069 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1070 CURSEG_HOT_NODE]);
25ca923b 1071 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
351df4b2
JK		 1072 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1073 type - CURSEG_HOT_NODE);
1074 else
1075 blk_addr = GET_SUM_BLOCK(sbi, segno);
1076 }
1077
1078 new = get_meta_page(sbi, blk_addr);
1079 sum = (struct f2fs_summary_block *)page_address(new);
1080
1081 if (IS_NODESEG(type)) {
25ca923b 1082 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
351df4b2
JK		 1083 struct f2fs_summary *ns = &sum->entries[0];
1084 int i;
1085 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1086 ns->version = 0;
1087 ns->ofs_in_node = 0;
1088 }
1089 } else {
1090 if (restore_node_summary(sbi, segno, sum)) {
1091 f2fs_put_page(new, 1);
1092 return -EINVAL;
1093 }
1094 }
1095 }
1096
1097 /* set uncompleted segment to curseg */
1098 curseg = CURSEG_I(sbi, type);
1099 mutex_lock(&curseg->curseg_mutex);
1100 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1101 curseg->next_segno = segno;
1102 reset_curseg(sbi, type, 0);
1103 curseg->alloc_type = ckpt->alloc_type[type];
1104 curseg->next_blkoff = blk_off;
1105 mutex_unlock(&curseg->curseg_mutex);
1106 f2fs_put_page(new, 1);
1107 return 0;
1108}
1109
1110static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1111{
1112 int type = CURSEG_HOT_DATA;
1113
25ca923b 1114 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
351df4b2
JK		 1115 /* restore for compacted data summary */
1116 if (read_compacted_summaries(sbi))
1117 return -EINVAL;
1118 type = CURSEG_HOT_NODE;
1119 }
1120
1121 for (; type <= CURSEG_COLD_NODE; type++)
1122 if (read_normal_summaries(sbi, type))
1123 return -EINVAL;
1124 return 0;
1125}
1126
1127static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1128{
1129 struct page *page;
1130 unsigned char *kaddr;
1131 struct f2fs_summary *summary;
1132 struct curseg_info *seg_i;
1133 int written_size = 0;
1134 int i, j;
1135
1136 page = grab_meta_page(sbi, blkaddr++);
1137 kaddr = (unsigned char *)page_address(page);
1138
1139 /* Step 1: write nat cache */
1140 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1141 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1142 written_size += SUM_JOURNAL_SIZE;
1143
1144 /* Step 2: write sit cache */
1145 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1146 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1147 SUM_JOURNAL_SIZE);
1148 written_size += SUM_JOURNAL_SIZE;
1149
1150 set_page_dirty(page);
1151
1152 /* Step 3: write summary entries */
1153 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1154 unsigned short blkoff;
1155 seg_i = CURSEG_I(sbi, i);
1156 if (sbi->ckpt->alloc_type[i] == SSR)
1157 blkoff = sbi->blocks_per_seg;
1158 else
1159 blkoff = curseg_blkoff(sbi, i);
1160
1161 for (j = 0; j < blkoff; j++) {
1162 if (!page) {
1163 page = grab_meta_page(sbi, blkaddr++);
1164 kaddr = (unsigned char *)page_address(page);
1165 written_size = 0;
1166 }
1167 summary = (struct f2fs_summary *)(kaddr + written_size);
1168 *summary = seg_i->sum_blk->entries[j];
1169 written_size += SUMMARY_SIZE;
1170 set_page_dirty(page);
1171
1172 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1173 SUM_FOOTER_SIZE)
1174 continue;
1175
1176 f2fs_put_page(page, 1);
1177 page = NULL;
1178 }
1179 }
1180 if (page)
1181 f2fs_put_page(page, 1);
1182}
1183
1184static void write_normal_summaries(struct f2fs_sb_info *sbi,
1185 block_t blkaddr, int type)
1186{
1187 int i, end;
1188 if (IS_DATASEG(type))
1189 end = type + NR_CURSEG_DATA_TYPE;
1190 else
1191 end = type + NR_CURSEG_NODE_TYPE;
1192
1193 for (i = type; i < end; i++) {
1194 struct curseg_info *sum = CURSEG_I(sbi, i);
1195 mutex_lock(&sum->curseg_mutex);
1196 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1197 mutex_unlock(&sum->curseg_mutex);
1198 }
1199}
1200
1201void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1202{
25ca923b 1203 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1204 write_compacted_summaries(sbi, start_blk);
1205 else
1206 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1207}
1208
1209void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1210{
25ca923b 1211 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
351df4b2
JK		 1212 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1213 return;
1214}
1215
1216int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1217 unsigned int val, int alloc)
1218{
1219 int i;
1220
1221 if (type == NAT_JOURNAL) {
1222 for (i = 0; i < nats_in_cursum(sum); i++) {
1223 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1224 return i;
1225 }
1226 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1227 return update_nats_in_cursum(sum, 1);
1228 } else if (type == SIT_JOURNAL) {
1229 for (i = 0; i < sits_in_cursum(sum); i++)
1230 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1231 return i;
1232 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1233 return update_sits_in_cursum(sum, 1);
1234 }
1235 return -1;
1236}
1237
1238static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1239 unsigned int segno)
1240{
1241 struct sit_info *sit_i = SIT_I(sbi);
1242 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1243 block_t blk_addr = sit_i->sit_base_addr + offset;
1244
1245 check_seg_range(sbi, segno);
1246
1247 /* calculate sit block address */
1248 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1249 blk_addr += sit_i->sit_blocks;
1250
1251 return get_meta_page(sbi, blk_addr);
1252}
1253
1254static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1255 unsigned int start)
1256{
1257 struct sit_info *sit_i = SIT_I(sbi);
1258 struct page *src_page, *dst_page;
1259 pgoff_t src_off, dst_off;
1260 void *src_addr, *dst_addr;
1261
1262 src_off = current_sit_addr(sbi, start);
1263 dst_off = next_sit_addr(sbi, src_off);
1264
1265 /* get current sit block page without lock */
1266 src_page = get_meta_page(sbi, src_off);
1267 dst_page = grab_meta_page(sbi, dst_off);
1268 BUG_ON(PageDirty(src_page));
1269
1270 src_addr = page_address(src_page);
1271 dst_addr = page_address(dst_page);
1272 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1273
1274 set_page_dirty(dst_page);
1275 f2fs_put_page(src_page, 1);
1276
1277 set_to_next_sit(sit_i, start);
1278
1279 return dst_page;
1280}
1281
1282static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1283{
1284 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1285 struct f2fs_summary_block *sum = curseg->sum_blk;
1286 int i;
1287
1288 /*
1289 * If the journal area in the current summary is full of sit entries,
1290 * all the sit entries will be flushed. Otherwise the sit entries
1291 * are not able to replace with newly hot sit entries.
1292 */
1293 if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1294 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1295 unsigned int segno;
1296 segno = le32_to_cpu(segno_in_journal(sum, i));
1297 __mark_sit_entry_dirty(sbi, segno);
1298 }
1299 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1300 return 1;
1301 }
1302 return 0;
1303}
1304
0a8165d7 1305/*
351df4b2
JK		 1306 * CP calls this function, which flushes SIT entries including sit_journal,
1307 * and moves prefree segs to free segs.
1308 */
1309void flush_sit_entries(struct f2fs_sb_info *sbi)
1310{
1311 struct sit_info *sit_i = SIT_I(sbi);
1312 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1313 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1314 struct f2fs_summary_block *sum = curseg->sum_blk;
1315 unsigned long nsegs = TOTAL_SEGS(sbi);
1316 struct page *page = NULL;
1317 struct f2fs_sit_block *raw_sit = NULL;
1318 unsigned int start = 0, end = 0;
1319 unsigned int segno = -1;
1320 bool flushed;
1321
1322 mutex_lock(&curseg->curseg_mutex);
1323 mutex_lock(&sit_i->sentry_lock);
1324
1325 /*
1326 * "flushed" indicates whether sit entries in journal are flushed
1327 * to the SIT area or not.
1328 */
1329 flushed = flush_sits_in_journal(sbi);
1330
1331 while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1332 struct seg_entry *se = get_seg_entry(sbi, segno);
1333 int sit_offset, offset;
1334
1335 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1336
1337 if (flushed)
1338 goto to_sit_page;
1339
1340 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1341 if (offset >= 0) {
1342 segno_in_journal(sum, offset) = cpu_to_le32(segno);
1343 seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1344 goto flush_done;
1345 }
1346to_sit_page:
1347 if (!page || (start > segno) || (segno > end)) {
1348 if (page) {
1349 f2fs_put_page(page, 1);
1350 page = NULL;
1351 }
1352
1353 start = START_SEGNO(sit_i, segno);
1354 end = start + SIT_ENTRY_PER_BLOCK - 1;
1355
1356 /* read sit block that will be updated */
1357 page = get_next_sit_page(sbi, start);
1358 raw_sit = page_address(page);
1359 }
1360
1361 /* udpate entry in SIT block */
1362 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1363flush_done:
1364 __clear_bit(segno, bitmap);
1365 sit_i->dirty_sentries--;
1366 }
1367 mutex_unlock(&sit_i->sentry_lock);
1368 mutex_unlock(&curseg->curseg_mutex);
1369
1370 /* writeout last modified SIT block */
1371 f2fs_put_page(page, 1);
1372
1373 set_prefree_as_free_segments(sbi);
1374}
1375
1376static int build_sit_info(struct f2fs_sb_info *sbi)
1377{
1378 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1379 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1380 struct sit_info *sit_i;
1381 unsigned int sit_segs, start;
1382 char *src_bitmap, *dst_bitmap;
1383 unsigned int bitmap_size;
1384
1385 /* allocate memory for SIT information */
1386 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1387 if (!sit_i)
1388 return -ENOMEM;
1389
1390 SM_I(sbi)->sit_info = sit_i;
1391
1392 sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1393 if (!sit_i->sentries)
1394 return -ENOMEM;
1395
1396 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1397 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1398 if (!sit_i->dirty_sentries_bitmap)
1399 return -ENOMEM;
1400
1401 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1402 sit_i->sentries[start].cur_valid_map
1403 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1404 sit_i->sentries[start].ckpt_valid_map
1405 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1406 if (!sit_i->sentries[start].cur_valid_map
1407 || !sit_i->sentries[start].ckpt_valid_map)
1408 return -ENOMEM;
1409 }
1410
1411 if (sbi->segs_per_sec > 1) {
1412 sit_i->sec_entries = vzalloc(sbi->total_sections *
1413 sizeof(struct sec_entry));
1414 if (!sit_i->sec_entries)
1415 return -ENOMEM;
1416 }
1417
1418 /* get information related with SIT */
1419 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1420
1421 /* setup SIT bitmap from ckeckpoint pack */
1422 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1423 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1424
1425 dst_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1426 if (!dst_bitmap)
1427 return -ENOMEM;
1428 memcpy(dst_bitmap, src_bitmap, bitmap_size);
1429
1430 /* init SIT information */
1431 sit_i->s_ops = &default_salloc_ops;
1432
1433 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1434 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1435 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1436 sit_i->sit_bitmap = dst_bitmap;
1437 sit_i->bitmap_size = bitmap_size;
1438 sit_i->dirty_sentries = 0;
1439 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1440 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1441 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1442 mutex_init(&sit_i->sentry_lock);
1443 return 0;
1444}
1445
1446static int build_free_segmap(struct f2fs_sb_info *sbi)
1447{
1448 struct f2fs_sm_info *sm_info = SM_I(sbi);
1449 struct free_segmap_info *free_i;
1450 unsigned int bitmap_size, sec_bitmap_size;
1451
1452 /* allocate memory for free segmap information */
1453 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1454 if (!free_i)
1455 return -ENOMEM;
1456
1457 SM_I(sbi)->free_info = free_i;
1458
1459 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1460 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1461 if (!free_i->free_segmap)
1462 return -ENOMEM;
1463
1464 sec_bitmap_size = f2fs_bitmap_size(sbi->total_sections);
1465 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1466 if (!free_i->free_secmap)
1467 return -ENOMEM;
1468
1469 /* set all segments as dirty temporarily */
1470 memset(free_i->free_segmap, 0xff, bitmap_size);
1471 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1472
1473 /* init free segmap information */
1474 free_i->start_segno =
1475 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1476 free_i->free_segments = 0;
1477 free_i->free_sections = 0;
1478 rwlock_init(&free_i->segmap_lock);
1479 return 0;
1480}
1481
1482static int build_curseg(struct f2fs_sb_info *sbi)
1483{
1042d60f 1484 struct curseg_info *array;
351df4b2
JK		 1485 int i;
1486
1487 array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1488 if (!array)
1489 return -ENOMEM;
1490
1491 SM_I(sbi)->curseg_array = array;
1492
1493 for (i = 0; i < NR_CURSEG_TYPE; i++) {
1494 mutex_init(&array[i].curseg_mutex);
1495 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1496 if (!array[i].sum_blk)
1497 return -ENOMEM;
1498 array[i].segno = NULL_SEGNO;
1499 array[i].next_blkoff = 0;
1500 }
1501 return restore_curseg_summaries(sbi);
1502}
1503
1504static void build_sit_entries(struct f2fs_sb_info *sbi)
1505{
1506 struct sit_info *sit_i = SIT_I(sbi);
1507 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1508 struct f2fs_summary_block *sum = curseg->sum_blk;
1509 unsigned int start;
1510
1511 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1512 struct seg_entry *se = &sit_i->sentries[start];
1513 struct f2fs_sit_block *sit_blk;
1514 struct f2fs_sit_entry sit;
1515 struct page *page;
1516 int i;
1517
1518 mutex_lock(&curseg->curseg_mutex);
1519 for (i = 0; i < sits_in_cursum(sum); i++) {
1520 if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1521 sit = sit_in_journal(sum, i);
1522 mutex_unlock(&curseg->curseg_mutex);
1523 goto got_it;
1524 }
1525 }
1526 mutex_unlock(&curseg->curseg_mutex);
1527 page = get_current_sit_page(sbi, start);
1528 sit_blk = (struct f2fs_sit_block *)page_address(page);
1529 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1530 f2fs_put_page(page, 1);
1531got_it:
1532 check_block_count(sbi, start, &sit);
1533 seg_info_from_raw_sit(se, &sit);
1534 if (sbi->segs_per_sec > 1) {
1535 struct sec_entry *e = get_sec_entry(sbi, start);
1536 e->valid_blocks += se->valid_blocks;
1537 }
1538 }
1539}
1540
1541static void init_free_segmap(struct f2fs_sb_info *sbi)
1542{
1543 unsigned int start;
1544 int type;
1545
1546 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1547 struct seg_entry *sentry = get_seg_entry(sbi, start);
1548 if (!sentry->valid_blocks)
1549 __set_free(sbi, start);
1550 }
1551
1552 /* set use the current segments */
1553 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1554 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1555 __set_test_and_inuse(sbi, curseg_t->segno);
1556 }
1557}
1558
1559static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1560{
1561 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1562 struct free_segmap_info *free_i = FREE_I(sbi);
1563 unsigned int segno = 0, offset = 0;
1564 unsigned short valid_blocks;
1565
1566 while (segno < TOTAL_SEGS(sbi)) {
1567 /* find dirty segment based on free segmap */
1568 segno = find_next_inuse(free_i, TOTAL_SEGS(sbi), offset);
1569 if (segno >= TOTAL_SEGS(sbi))
1570 break;
1571 offset = segno + 1;
1572 valid_blocks = get_valid_blocks(sbi, segno, 0);
1573 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1574 continue;
1575 mutex_lock(&dirty_i->seglist_lock);
1576 __locate_dirty_segment(sbi, segno, DIRTY);
1577 mutex_unlock(&dirty_i->seglist_lock);
1578 }
1579}
1580
1581static int init_victim_segmap(struct f2fs_sb_info *sbi)
1582{
1583 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1584 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1585
1586 dirty_i->victim_segmap[FG_GC] = kzalloc(bitmap_size, GFP_KERNEL);
1587 dirty_i->victim_segmap[BG_GC] = kzalloc(bitmap_size, GFP_KERNEL);
1588 if (!dirty_i->victim_segmap[FG_GC] || !dirty_i->victim_segmap[BG_GC])
1589 return -ENOMEM;
1590 return 0;
1591}
1592
1593static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1594{
1595 struct dirty_seglist_info *dirty_i;
1596 unsigned int bitmap_size, i;
1597
1598 /* allocate memory for dirty segments list information */
1599 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1600 if (!dirty_i)
1601 return -ENOMEM;
1602
1603 SM_I(sbi)->dirty_info = dirty_i;
1604 mutex_init(&dirty_i->seglist_lock);
1605
1606 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1607
1608 for (i = 0; i < NR_DIRTY_TYPE; i++) {
1609 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1610 dirty_i->nr_dirty[i] = 0;
1611 if (!dirty_i->dirty_segmap[i])
1612 return -ENOMEM;
1613 }
1614
1615 init_dirty_segmap(sbi);
1616 return init_victim_segmap(sbi);
1617}
1618
0a8165d7 1619/*
351df4b2
JK		 1620 * Update min, max modified time for cost-benefit GC algorithm
1621 */
1622static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1623{
1624 struct sit_info *sit_i = SIT_I(sbi);
1625 unsigned int segno;
1626
1627 mutex_lock(&sit_i->sentry_lock);
1628
1629 sit_i->min_mtime = LLONG_MAX;
1630
1631 for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1632 unsigned int i;
1633 unsigned long long mtime = 0;
1634
1635 for (i = 0; i < sbi->segs_per_sec; i++)
1636 mtime += get_seg_entry(sbi, segno + i)->mtime;
1637
1638 mtime = div_u64(mtime, sbi->segs_per_sec);
1639
1640 if (sit_i->min_mtime > mtime)
1641 sit_i->min_mtime = mtime;
1642 }
1643 sit_i->max_mtime = get_mtime(sbi);
1644 mutex_unlock(&sit_i->sentry_lock);
1645}
1646
1647int build_segment_manager(struct f2fs_sb_info *sbi)
1648{
1649 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1650 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 1651 struct f2fs_sm_info *sm_info;
351df4b2
JK		 1652 int err;
1653
1654 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1655 if (!sm_info)
1656 return -ENOMEM;
1657
1658 /* init sm info */
1659 sbi->sm_info = sm_info;
1660 INIT_LIST_HEAD(&sm_info->wblist_head);
1661 spin_lock_init(&sm_info->wblist_lock);
1662 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1663 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1664 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1665 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1666 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1667 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1668 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1669
1670 err = build_sit_info(sbi);
1671 if (err)
1672 return err;
1673 err = build_free_segmap(sbi);
1674 if (err)
1675 return err;
1676 err = build_curseg(sbi);
1677 if (err)
1678 return err;
1679
1680 /* reinit free segmap based on SIT */
1681 build_sit_entries(sbi);
1682
1683 init_free_segmap(sbi);
1684 err = build_dirty_segmap(sbi);
1685 if (err)
1686 return err;
1687
1688 init_min_max_mtime(sbi);
1689 return 0;
1690}
1691
1692static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1693 enum dirty_type dirty_type)
1694{
1695 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1696
1697 mutex_lock(&dirty_i->seglist_lock);
1698 kfree(dirty_i->dirty_segmap[dirty_type]);
1699 dirty_i->nr_dirty[dirty_type] = 0;
1700 mutex_unlock(&dirty_i->seglist_lock);
1701}
1702
1703void reset_victim_segmap(struct f2fs_sb_info *sbi)
1704{
1705 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1706 memset(DIRTY_I(sbi)->victim_segmap[FG_GC], 0, bitmap_size);
1707}
1708
1709static void destroy_victim_segmap(struct f2fs_sb_info *sbi)
1710{
1711 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1712
1713 kfree(dirty_i->victim_segmap[FG_GC]);
1714 kfree(dirty_i->victim_segmap[BG_GC]);
1715}
1716
1717static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1718{
1719 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1720 int i;
1721
1722 if (!dirty_i)
1723 return;
1724
1725 /* discard pre-free/dirty segments list */
1726 for (i = 0; i < NR_DIRTY_TYPE; i++)
1727 discard_dirty_segmap(sbi, i);
1728
1729 destroy_victim_segmap(sbi);
1730 SM_I(sbi)->dirty_info = NULL;
1731 kfree(dirty_i);
1732}
1733
1734static void destroy_curseg(struct f2fs_sb_info *sbi)
1735{
1736 struct curseg_info *array = SM_I(sbi)->curseg_array;
1737 int i;
1738
1739 if (!array)
1740 return;
1741 SM_I(sbi)->curseg_array = NULL;
1742 for (i = 0; i < NR_CURSEG_TYPE; i++)
1743 kfree(array[i].sum_blk);
1744 kfree(array);
1745}
1746
1747static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1748{
1749 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1750 if (!free_i)
1751 return;
1752 SM_I(sbi)->free_info = NULL;
1753 kfree(free_i->free_segmap);
1754 kfree(free_i->free_secmap);
1755 kfree(free_i);
1756}
1757
1758static void destroy_sit_info(struct f2fs_sb_info *sbi)
1759{
1760 struct sit_info *sit_i = SIT_I(sbi);
1761 unsigned int start;
1762
1763 if (!sit_i)
1764 return;
1765
1766 if (sit_i->sentries) {
1767 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1768 kfree(sit_i->sentries[start].cur_valid_map);
1769 kfree(sit_i->sentries[start].ckpt_valid_map);
1770 }
1771 }
1772 vfree(sit_i->sentries);
1773 vfree(sit_i->sec_entries);
1774 kfree(sit_i->dirty_sentries_bitmap);
1775
1776 SM_I(sbi)->sit_info = NULL;
1777 kfree(sit_i->sit_bitmap);
1778 kfree(sit_i);
1779}
1780
1781void destroy_segment_manager(struct f2fs_sb_info *sbi)
1782{
1783 struct f2fs_sm_info *sm_info = SM_I(sbi);
1784 destroy_dirty_segmap(sbi);
1785 destroy_curseg(sbi);
1786 destroy_free_segmap(sbi);
1787 destroy_sit_info(sbi);
1788 sbi->sm_info = NULL;
1789 kfree(sm_info);
1790}
Linux kernel - Deliverables
This page took 0.090265 seconds and 5 git commands to generate.