projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: introduce CONFIG_F2FS_CHECK_FS for BUG_ON control
[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"
6ec178da 21#include <trace/events/f2fs.h>
351df4b2 22
0a8165d7 23/*
351df4b2
JK		 24 * This function balances dirty node and dentry pages.
25 * In addition, it controls garbage collection.
26 */
27void f2fs_balance_fs(struct f2fs_sb_info *sbi)
28{
351df4b2 29 /*
029cd28c
JK		 30 * We should do GC or end up with checkpoint, if there are so many dirty
31 * dir/node pages without enough free segments.
351df4b2 32 */
43727527 33 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 34 mutex_lock(&sbi->gc_mutex);
408e9375 35 f2fs_gc(sbi);
351df4b2
JK		 36 }
37}
38
4660f9c0
JK		 39void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
40{
41 /* check the # of cached NAT entries and prefree segments */
42 if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
43 excess_prefree_segs(sbi))
44 f2fs_sync_fs(sbi->sb, true);
45}
46
351df4b2
JK		 47static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
48 enum dirty_type dirty_type)
49{
50 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
51
52 /* need not be added */
53 if (IS_CURSEG(sbi, segno))
54 return;
55
56 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
57 dirty_i->nr_dirty[dirty_type]++;
58
59 if (dirty_type == DIRTY) {
60 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 61 enum dirty_type t = sentry->type;
b2f2c390 62
4625d6aa
CL		 63 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
64 dirty_i->nr_dirty[t]++;
351df4b2
JK		 65 }
66}
67
68static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
69 enum dirty_type dirty_type)
70{
71 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
72
73 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
74 dirty_i->nr_dirty[dirty_type]--;
75
76 if (dirty_type == DIRTY) {
4625d6aa
CL		 77 struct seg_entry *sentry = get_seg_entry(sbi, segno);
78 enum dirty_type t = sentry->type;
79
80 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
81 dirty_i->nr_dirty[t]--;
b2f2c390 82
5ec4e49f
JK		 83 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
84 clear_bit(GET_SECNO(sbi, segno),
85 dirty_i->victim_secmap);
351df4b2
JK		 86 }
87}
88
0a8165d7 89/*
351df4b2
JK		 90 * Should not occur error such as -ENOMEM.
91 * Adding dirty entry into seglist is not critical operation.
92 * If a given segment is one of current working segments, it won't be added.
93 */
8d8451af 94static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 95{
96 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
97 unsigned short valid_blocks;
98
99 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
100 return;
101
102 mutex_lock(&dirty_i->seglist_lock);
103
104 valid_blocks = get_valid_blocks(sbi, segno, 0);
105
106 if (valid_blocks == 0) {
107 __locate_dirty_segment(sbi, segno, PRE);
108 __remove_dirty_segment(sbi, segno, DIRTY);
109 } else if (valid_blocks < sbi->blocks_per_seg) {
110 __locate_dirty_segment(sbi, segno, DIRTY);
111 } else {
112 /* Recovery routine with SSR needs this */
113 __remove_dirty_segment(sbi, segno, DIRTY);
114 }
115
116 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 117}
118
0a8165d7 119/*
351df4b2
JK		 120 * Should call clear_prefree_segments after checkpoint is done.
121 */
122static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
123{
124 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
81fb5e87 125 unsigned int segno = -1;
351df4b2
JK		 126 unsigned int total_segs = TOTAL_SEGS(sbi);
127
128 mutex_lock(&dirty_i->seglist_lock);
129 while (1) {
130 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
81fb5e87 131 segno + 1);
351df4b2
JK		 132 if (segno >= total_segs)
133 break;
134 __set_test_and_free(sbi, segno);
351df4b2
JK		 135 }
136 mutex_unlock(&dirty_i->seglist_lock);
137}
138
139void clear_prefree_segments(struct f2fs_sb_info *sbi)
140{
141 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
81fb5e87 142 unsigned int segno = -1;
351df4b2
JK		 143 unsigned int total_segs = TOTAL_SEGS(sbi);
144
145 mutex_lock(&dirty_i->seglist_lock);
146 while (1) {
147 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
81fb5e87 148 segno + 1);
351df4b2
JK		 149 if (segno >= total_segs)
150 break;
151
351df4b2
JK		 152 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
153 dirty_i->nr_dirty[PRE]--;
154
155 /* Let's use trim */
156 if (test_opt(sbi, DISCARD))
157 blkdev_issue_discard(sbi->sb->s_bdev,
158 START_BLOCK(sbi, segno) <<
159 sbi->log_sectors_per_block,
160 1 << (sbi->log_sectors_per_block +
161 sbi->log_blocks_per_seg),
162 GFP_NOFS, 0);
163 }
164 mutex_unlock(&dirty_i->seglist_lock);
165}
166
167static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
168{
169 struct sit_info *sit_i = SIT_I(sbi);
170 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
171 sit_i->dirty_sentries++;
172}
173
174static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
175 unsigned int segno, int modified)
176{
177 struct seg_entry *se = get_seg_entry(sbi, segno);
178 se->type = type;
179 if (modified)
180 __mark_sit_entry_dirty(sbi, segno);
181}
182
183static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
184{
185 struct seg_entry *se;
186 unsigned int segno, offset;
187 long int new_vblocks;
188
189 segno = GET_SEGNO(sbi, blkaddr);
190
191 se = get_seg_entry(sbi, segno);
192 new_vblocks = se->valid_blocks + del;
193 offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
194
195 BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) ||
196 (new_vblocks > sbi->blocks_per_seg)));
197
198 se->valid_blocks = new_vblocks;
199 se->mtime = get_mtime(sbi);
200 SIT_I(sbi)->max_mtime = se->mtime;
201
202 /* Update valid block bitmap */
203 if (del > 0) {
204 if (f2fs_set_bit(offset, se->cur_valid_map))
205 BUG();
206 } else {
207 if (!f2fs_clear_bit(offset, se->cur_valid_map))
208 BUG();
209 }
210 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
211 se->ckpt_valid_blocks += del;
212
213 __mark_sit_entry_dirty(sbi, segno);
214
215 /* update total number of valid blocks to be written in ckpt area */
216 SIT_I(sbi)->written_valid_blocks += del;
217
218 if (sbi->segs_per_sec > 1)
219 get_sec_entry(sbi, segno)->valid_blocks += del;
220}
221
222static void refresh_sit_entry(struct f2fs_sb_info *sbi,
223 block_t old_blkaddr, block_t new_blkaddr)
224{
225 update_sit_entry(sbi, new_blkaddr, 1);
226 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
227 update_sit_entry(sbi, old_blkaddr, -1);
228}
229
230void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
231{
232 unsigned int segno = GET_SEGNO(sbi, addr);
233 struct sit_info *sit_i = SIT_I(sbi);
234
235 BUG_ON(addr == NULL_ADDR);
236 if (addr == NEW_ADDR)
237 return;
238
239 /* add it into sit main buffer */
240 mutex_lock(&sit_i->sentry_lock);
241
242 update_sit_entry(sbi, addr, -1);
243
244 /* add it into dirty seglist */
245 locate_dirty_segment(sbi, segno);
246
247 mutex_unlock(&sit_i->sentry_lock);
248}
249
0a8165d7 250/*
351df4b2
JK		 251 * This function should be resided under the curseg_mutex lock
252 */
253static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 254 struct f2fs_summary *sum)
351df4b2
JK		 255{
256 struct curseg_info *curseg = CURSEG_I(sbi, type);
257 void *addr = curseg->sum_blk;
e79efe3b 258 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 259 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 260}
261
0a8165d7 262/*
351df4b2
JK		 263 * Calculate the number of current summary pages for writing
264 */
265int npages_for_summary_flush(struct f2fs_sb_info *sbi)
266{
267 int total_size_bytes = 0;
268 int valid_sum_count = 0;
269 int i, sum_space;
270
271 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
272 if (sbi->ckpt->alloc_type[i] == SSR)
273 valid_sum_count += sbi->blocks_per_seg;
274 else
275 valid_sum_count += curseg_blkoff(sbi, i);
276 }
277
278 total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1)
279 + sizeof(struct nat_journal) + 2
280 + sizeof(struct sit_journal) + 2;
281 sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE;
282 if (total_size_bytes < sum_space)
283 return 1;
284 else if (total_size_bytes < 2 * sum_space)
285 return 2;
286 return 3;
287}
288
0a8165d7 289/*
351df4b2
JK		 290 * Caller should put this summary page
291 */
292struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
293{
294 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
295}
296
297static void write_sum_page(struct f2fs_sb_info *sbi,
298 struct f2fs_summary_block *sum_blk, block_t blk_addr)
299{
300 struct page *page = grab_meta_page(sbi, blk_addr);
301 void *kaddr = page_address(page);
302 memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
303 set_page_dirty(page);
304 f2fs_put_page(page, 1);
305}
306
60374688
JK		 307static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
308{
309 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 310 unsigned int segno = curseg->segno + 1;
60374688
JK		 311 struct free_segmap_info *free_i = FREE_I(sbi);
312
81fb5e87
HL		 313 if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
314 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 315 return 0;
316}
317
0a8165d7 318/*
351df4b2
JK		 319 * Find a new segment from the free segments bitmap to right order
320 * This function should be returned with success, otherwise BUG
321 */
322static void get_new_segment(struct f2fs_sb_info *sbi,
323 unsigned int *newseg, bool new_sec, int dir)
324{
325 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 326 unsigned int segno, secno, zoneno;
53cf9522 327 unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 328 unsigned int hint = *newseg / sbi->segs_per_sec;
329 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
330 unsigned int left_start = hint;
331 bool init = true;
332 int go_left = 0;
333 int i;
334
335 write_lock(&free_i->segmap_lock);
336
337 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
338 segno = find_next_zero_bit(free_i->free_segmap,
339 TOTAL_SEGS(sbi), *newseg + 1);
33afa7fd
JK		 340 if (segno - *newseg < sbi->segs_per_sec -
341 (*newseg % sbi->segs_per_sec))
351df4b2
JK		 342 goto got_it;
343 }
344find_other_zone:
53cf9522
JK		 345 secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
346 if (secno >= TOTAL_SECS(sbi)) {
351df4b2
JK		 347 if (dir == ALLOC_RIGHT) {
348 secno = find_next_zero_bit(free_i->free_secmap,
53cf9522
JK		 349 TOTAL_SECS(sbi), 0);
350 BUG_ON(secno >= TOTAL_SECS(sbi));
351df4b2
JK		 351 } else {
352 go_left = 1;
353 left_start = hint - 1;
354 }
355 }
356 if (go_left == 0)
357 goto skip_left;
358
359 while (test_bit(left_start, free_i->free_secmap)) {
360 if (left_start > 0) {
361 left_start--;
362 continue;
363 }
364 left_start = find_next_zero_bit(free_i->free_secmap,
53cf9522
JK		 365 TOTAL_SECS(sbi), 0);
366 BUG_ON(left_start >= TOTAL_SECS(sbi));
351df4b2
JK		 367 break;
368 }
369 secno = left_start;
370skip_left:
371 hint = secno;
372 segno = secno * sbi->segs_per_sec;
373 zoneno = secno / sbi->secs_per_zone;
374
375 /* give up on finding another zone */
376 if (!init)
377 goto got_it;
378 if (sbi->secs_per_zone == 1)
379 goto got_it;
380 if (zoneno == old_zoneno)
381 goto got_it;
382 if (dir == ALLOC_LEFT) {
383 if (!go_left && zoneno + 1 >= total_zones)
384 goto got_it;
385 if (go_left && zoneno == 0)
386 goto got_it;
387 }
388 for (i = 0; i < NR_CURSEG_TYPE; i++)
389 if (CURSEG_I(sbi, i)->zone == zoneno)
390 break;
391
392 if (i < NR_CURSEG_TYPE) {
393 /* zone is in user, try another */
394 if (go_left)
395 hint = zoneno * sbi->secs_per_zone - 1;
396 else if (zoneno + 1 >= total_zones)
397 hint = 0;
398 else
399 hint = (zoneno + 1) * sbi->secs_per_zone;
400 init = false;
401 goto find_other_zone;
402 }
403got_it:
404 /* set it as dirty segment in free segmap */
405 BUG_ON(test_bit(segno, free_i->free_segmap));
406 __set_inuse(sbi, segno);
407 *newseg = segno;
408 write_unlock(&free_i->segmap_lock);
409}
410
411static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
412{
413 struct curseg_info *curseg = CURSEG_I(sbi, type);
414 struct summary_footer *sum_footer;
415
416 curseg->segno = curseg->next_segno;
417 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
418 curseg->next_blkoff = 0;
419 curseg->next_segno = NULL_SEGNO;
420
421 sum_footer = &(curseg->sum_blk->footer);
422 memset(sum_footer, 0, sizeof(struct summary_footer));
423 if (IS_DATASEG(type))
424 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
425 if (IS_NODESEG(type))
426 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
427 __set_sit_entry_type(sbi, type, curseg->segno, modified);
428}
429
0a8165d7 430/*
351df4b2
JK		 431 * Allocate a current working segment.
432 * This function always allocates a free segment in LFS manner.
433 */
434static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
435{
436 struct curseg_info *curseg = CURSEG_I(sbi, type);
437 unsigned int segno = curseg->segno;
438 int dir = ALLOC_LEFT;
439
440 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 441 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 442 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
443 dir = ALLOC_RIGHT;
444
445 if (test_opt(sbi, NOHEAP))
446 dir = ALLOC_RIGHT;
447
448 get_new_segment(sbi, &segno, new_sec, dir);
449 curseg->next_segno = segno;
450 reset_curseg(sbi, type, 1);
451 curseg->alloc_type = LFS;
452}
453
454static void __next_free_blkoff(struct f2fs_sb_info *sbi,
455 struct curseg_info *seg, block_t start)
456{
457 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
458 block_t ofs;
459 for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
460 if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
461 && !f2fs_test_bit(ofs, se->cur_valid_map))
462 break;
463 }
464 seg->next_blkoff = ofs;
465}
466
0a8165d7 467/*
351df4b2
JK		 468 * If a segment is written by LFS manner, next block offset is just obtained
469 * by increasing the current block offset. However, if a segment is written by
470 * SSR manner, next block offset obtained by calling __next_free_blkoff
471 */
472static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
473 struct curseg_info *seg)
474{
475 if (seg->alloc_type == SSR)
476 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
477 else
478 seg->next_blkoff++;
479}
480
0a8165d7 481/*
351df4b2
JK		 482 * This function always allocates a used segment (from dirty seglist) by SSR
483 * manner, so it should recover the existing segment information of valid blocks
484 */
485static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
486{
487 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
488 struct curseg_info *curseg = CURSEG_I(sbi, type);
489 unsigned int new_segno = curseg->next_segno;
490 struct f2fs_summary_block *sum_node;
491 struct page *sum_page;
492
493 write_sum_page(sbi, curseg->sum_blk,
494 GET_SUM_BLOCK(sbi, curseg->segno));
495 __set_test_and_inuse(sbi, new_segno);
496
497 mutex_lock(&dirty_i->seglist_lock);
498 __remove_dirty_segment(sbi, new_segno, PRE);
499 __remove_dirty_segment(sbi, new_segno, DIRTY);
500 mutex_unlock(&dirty_i->seglist_lock);
501
502 reset_curseg(sbi, type, 1);
503 curseg->alloc_type = SSR;
504 __next_free_blkoff(sbi, curseg, 0);
505
506 if (reuse) {
507 sum_page = get_sum_page(sbi, new_segno);
508 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
509 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
510 f2fs_put_page(sum_page, 1);
511 }
512}
513
43727527
JK		 514static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
515{
516 struct curseg_info *curseg = CURSEG_I(sbi, type);
517 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
518
519 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
520 return v_ops->get_victim(sbi,
521 &(curseg)->next_segno, BG_GC, type, SSR);
522
523 /* For data segments, let's do SSR more intensively */
524 for (; type >= CURSEG_HOT_DATA; type--)
525 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
526 BG_GC, type, SSR))
527 return 1;
528 return 0;
529}
530
351df4b2
JK		 531/*
532 * flush out current segment and replace it with new segment
533 * This function should be returned with success, otherwise BUG
534 */
535static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
536 int type, bool force)
537{
538 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 539
7b405275 540 if (force)
351df4b2 541 new_curseg(sbi, type, true);
7b405275 542 else if (type == CURSEG_WARM_NODE)
351df4b2 543 new_curseg(sbi, type, false);
60374688
JK		 544 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
545 new_curseg(sbi, type, false);
351df4b2
JK		 546 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
547 change_curseg(sbi, type, true);
548 else
549 new_curseg(sbi, type, false);
dcdfff65
JK		 550
551 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 552}
553
554void allocate_new_segments(struct f2fs_sb_info *sbi)
555{
556 struct curseg_info *curseg;
557 unsigned int old_curseg;
558 int i;
559
560 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
561 curseg = CURSEG_I(sbi, i);
562 old_curseg = curseg->segno;
563 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
564 locate_dirty_segment(sbi, old_curseg);
565 }
566}
567
568static const struct segment_allocation default_salloc_ops = {
569 .allocate_segment = allocate_segment_by_default,
570};
571
572static void f2fs_end_io_write(struct bio *bio, int err)
573{
574 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
575 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
576 struct bio_private *p = bio->bi_private;
577
578 do {
579 struct page *page = bvec->bv_page;
580
581 if (--bvec >= bio->bi_io_vec)
582 prefetchw(&bvec->bv_page->flags);
583 if (!uptodate) {
584 SetPageError(page);
585 if (page->mapping)
586 set_bit(AS_EIO, &page->mapping->flags);
25ca923b 587 set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
577e3495 588 p->sbi->sb->s_flags |= MS_RDONLY;
351df4b2
JK		 589 }
590 end_page_writeback(page);
591 dec_page_count(p->sbi, F2FS_WRITEBACK);
592 } while (bvec >= bio->bi_io_vec);
593
594 if (p->is_sync)
595 complete(p->wait);
e2340887
GZ		 596
597 if (!get_pages(p->sbi, F2FS_WRITEBACK) && p->sbi->cp_task)
598 wake_up_process(p->sbi->cp_task);
599
351df4b2
JK		 600 kfree(p);
601 bio_put(bio);
602}
603
3cd8a239 604struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
351df4b2
JK		 605{
606 struct bio *bio;
3cd8a239
JK		 607
608 /* No failure on bio allocation */
609 bio = bio_alloc(GFP_NOIO, npages);
610 bio->bi_bdev = bdev;
d8207f69
GZ		 611 bio->bi_private = NULL;
612
351df4b2
JK		 613 return bio;
614}
615
616static void do_submit_bio(struct f2fs_sb_info *sbi,
617 enum page_type type, bool sync)
618{
619 int rw = sync ? WRITE_SYNC : WRITE;
620 enum page_type btype = type > META ? META : type;
621
622 if (type >= META_FLUSH)
623 rw = WRITE_FLUSH_FUA;
624
8680441c
NJ		 625 if (btype == META)
626 rw |= REQ_META;
627
351df4b2
JK		 628 if (sbi->bio[btype]) {
629 struct bio_private *p = sbi->bio[btype]->bi_private;
630 p->sbi = sbi;
631 sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
6ec178da
NJ		 632
633 trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]);
634
351df4b2
JK		 635 if (type == META_FLUSH) {
636 DECLARE_COMPLETION_ONSTACK(wait);
637 p->is_sync = true;
638 p->wait = &wait;
639 submit_bio(rw, sbi->bio[btype]);
640 wait_for_completion(&wait);
641 } else {
642 p->is_sync = false;
643 submit_bio(rw, sbi->bio[btype]);
644 }
645 sbi->bio[btype] = NULL;
646 }
647}
648
649void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
650{
651 down_write(&sbi->bio_sem);
652 do_submit_bio(sbi, type, sync);
653 up_write(&sbi->bio_sem);
654}
655
656static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
657 block_t blk_addr, enum page_type type)
658{
659 struct block_device *bdev = sbi->sb->s_bdev;
cc7b1bb1 660 int bio_blocks;
351df4b2
JK		 661
662 verify_block_addr(sbi, blk_addr);
663
664 down_write(&sbi->bio_sem);
665
666 inc_page_count(sbi, F2FS_WRITEBACK);
667
668 if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
669 do_submit_bio(sbi, type, false);
670alloc_new:
3cd8a239 671 if (sbi->bio[type] == NULL) {
d8207f69
GZ		 672 struct bio_private *priv;
673retry:
674 priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
675 if (!priv) {
676 cond_resched();
677 goto retry;
678 }
679
cc7b1bb1
CY		 680 bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
681 sbi->bio[type] = f2fs_bio_alloc(bdev, bio_blocks);
3cd8a239 682 sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
d8207f69 683 sbi->bio[type]->bi_private = priv;
3cd8a239
JK		 684 /*
685 * The end_io will be assigned at the sumbission phase.
686 * Until then, let bio_add_page() merge consecutive IOs as much
687 * as possible.
688 */
689 }
351df4b2
JK		 690
691 if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
692 PAGE_CACHE_SIZE) {
693 do_submit_bio(sbi, type, false);
694 goto alloc_new;
695 }
696
697 sbi->last_block_in_bio[type] = blk_addr;
698
699 up_write(&sbi->bio_sem);
6ec178da 700 trace_f2fs_submit_write_page(page, blk_addr, type);
351df4b2
JK		 701}
702
a569469e
JX		 703void f2fs_wait_on_page_writeback(struct page *page,
704 enum page_type type, bool sync)
705{
706 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
707 if (PageWriteback(page)) {
708 f2fs_submit_bio(sbi, type, sync);
709 wait_on_page_writeback(page);
710 }
711}
712
351df4b2
JK		 713static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
714{
715 struct curseg_info *curseg = CURSEG_I(sbi, type);
716 if (curseg->next_blkoff < sbi->blocks_per_seg)
717 return true;
718 return false;
719}
720
721static int __get_segment_type_2(struct page *page, enum page_type p_type)
722{
723 if (p_type == DATA)
724 return CURSEG_HOT_DATA;
725 else
726 return CURSEG_HOT_NODE;
727}
728
729static int __get_segment_type_4(struct page *page, enum page_type p_type)
730{
731 if (p_type == DATA) {
732 struct inode *inode = page->mapping->host;
733
734 if (S_ISDIR(inode->i_mode))
735 return CURSEG_HOT_DATA;
736 else
737 return CURSEG_COLD_DATA;
738 } else {
739 if (IS_DNODE(page) && !is_cold_node(page))
740 return CURSEG_HOT_NODE;
741 else
742 return CURSEG_COLD_NODE;
743 }
744}
745
746static int __get_segment_type_6(struct page *page, enum page_type p_type)
747{
748 if (p_type == DATA) {
749 struct inode *inode = page->mapping->host;
750
751 if (S_ISDIR(inode->i_mode))
752 return CURSEG_HOT_DATA;
354a3399 753 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 754 return CURSEG_COLD_DATA;
755 else
756 return CURSEG_WARM_DATA;
757 } else {
758 if (IS_DNODE(page))
759 return is_cold_node(page) ? CURSEG_WARM_NODE :
760 CURSEG_HOT_NODE;
761 else
762 return CURSEG_COLD_NODE;
763 }
764}
765
766static int __get_segment_type(struct page *page, enum page_type p_type)
767{
768 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
769 switch (sbi->active_logs) {
770 case 2:
771 return __get_segment_type_2(page, p_type);
772 case 4:
773 return __get_segment_type_4(page, p_type);
351df4b2 774 }
12a67146
JK		 775 /* NR_CURSEG_TYPE(6) logs by default */
776 BUG_ON(sbi->active_logs != NR_CURSEG_TYPE);
777 return __get_segment_type_6(page, p_type);
351df4b2
JK		 778}
779
780static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
781 block_t old_blkaddr, block_t *new_blkaddr,
782 struct f2fs_summary *sum, enum page_type p_type)
783{
784 struct sit_info *sit_i = SIT_I(sbi);
785 struct curseg_info *curseg;
786 unsigned int old_cursegno;
787 int type;
788
789 type = __get_segment_type(page, p_type);
790 curseg = CURSEG_I(sbi, type);
791
792 mutex_lock(&curseg->curseg_mutex);
793
794 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
795 old_cursegno = curseg->segno;
796
797 /*
798 * __add_sum_entry should be resided under the curseg_mutex
799 * because, this function updates a summary entry in the
800 * current summary block.
801 */
e79efe3b 802 __add_sum_entry(sbi, type, sum);
351df4b2
JK		 803
804 mutex_lock(&sit_i->sentry_lock);
805 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 806
807 stat_inc_block_count(sbi, curseg);
351df4b2
JK		 808
809 /*
810 * SIT information should be updated before segment allocation,
811 * since SSR needs latest valid block information.
812 */
813 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
814
815 if (!__has_curseg_space(sbi, type))
816 sit_i->s_ops->allocate_segment(sbi, type, false);
817
818 locate_dirty_segment(sbi, old_cursegno);
819 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
820 mutex_unlock(&sit_i->sentry_lock);
821
822 if (p_type == NODE)
823 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
824
825 /* writeout dirty page into bdev */
826 submit_write_page(sbi, page, *new_blkaddr, p_type);
827
828 mutex_unlock(&curseg->curseg_mutex);
829}
830
577e3495 831void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 832{
351df4b2
JK		 833 set_page_writeback(page);
834 submit_write_page(sbi, page, page->index, META);
351df4b2
JK		 835}
836
837void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
838 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
839{
840 struct f2fs_summary sum;
841 set_summary(&sum, nid, 0, 0);
842 do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
843}
844
845void write_data_page(struct inode *inode, struct page *page,
846 struct dnode_of_data *dn, block_t old_blkaddr,
847 block_t *new_blkaddr)
848{
849 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
850 struct f2fs_summary sum;
851 struct node_info ni;
852
853 BUG_ON(old_blkaddr == NULL_ADDR);
854 get_node_info(sbi, dn->nid, &ni);
855 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
856
857 do_write_page(sbi, page, old_blkaddr,
858 new_blkaddr, &sum, DATA);
859}
860
861void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
862 block_t old_blk_addr)
863{
864 submit_write_page(sbi, page, old_blk_addr, DATA);
865}
866
867void recover_data_page(struct f2fs_sb_info *sbi,
868 struct page *page, struct f2fs_summary *sum,
869 block_t old_blkaddr, block_t new_blkaddr)
870{
871 struct sit_info *sit_i = SIT_I(sbi);
872 struct curseg_info *curseg;
873 unsigned int segno, old_cursegno;
874 struct seg_entry *se;
875 int type;
876
877 segno = GET_SEGNO(sbi, new_blkaddr);
878 se = get_seg_entry(sbi, segno);
879 type = se->type;
880
881 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
882 if (old_blkaddr == NULL_ADDR)
883 type = CURSEG_COLD_DATA;
884 else
885 type = CURSEG_WARM_DATA;
886 }
887 curseg = CURSEG_I(sbi, type);
888
889 mutex_lock(&curseg->curseg_mutex);
890 mutex_lock(&sit_i->sentry_lock);
891
892 old_cursegno = curseg->segno;
893
894 /* change the current segment */
895 if (segno != curseg->segno) {
896 curseg->next_segno = segno;
897 change_curseg(sbi, type, true);
898 }
899
900 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
901 (sbi->blocks_per_seg - 1);
e79efe3b 902 __add_sum_entry(sbi, type, sum);
351df4b2
JK		 903
904 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
905
906 locate_dirty_segment(sbi, old_cursegno);
907 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
908
909 mutex_unlock(&sit_i->sentry_lock);
910 mutex_unlock(&curseg->curseg_mutex);
911}
912
913void rewrite_node_page(struct f2fs_sb_info *sbi,
914 struct page *page, struct f2fs_summary *sum,
915 block_t old_blkaddr, block_t new_blkaddr)
916{
917 struct sit_info *sit_i = SIT_I(sbi);
918 int type = CURSEG_WARM_NODE;
919 struct curseg_info *curseg;
920 unsigned int segno, old_cursegno;
921 block_t next_blkaddr = next_blkaddr_of_node(page);
922 unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
923
924 curseg = CURSEG_I(sbi, type);
925
926 mutex_lock(&curseg->curseg_mutex);
927 mutex_lock(&sit_i->sentry_lock);
928
929 segno = GET_SEGNO(sbi, new_blkaddr);
930 old_cursegno = curseg->segno;
931
932 /* change the current segment */
933 if (segno != curseg->segno) {
934 curseg->next_segno = segno;
935 change_curseg(sbi, type, true);
936 }
937 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
938 (sbi->blocks_per_seg - 1);
e79efe3b 939 __add_sum_entry(sbi, type, sum);
351df4b2
JK		 940
941 /* change the current log to the next block addr in advance */
942 if (next_segno != segno) {
943 curseg->next_segno = next_segno;
944 change_curseg(sbi, type, true);
945 }
946 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
947 (sbi->blocks_per_seg - 1);
948
949 /* rewrite node page */
950 set_page_writeback(page);
951 submit_write_page(sbi, page, new_blkaddr, NODE);
952 f2fs_submit_bio(sbi, NODE, true);
953 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
954
955 locate_dirty_segment(sbi, old_cursegno);
956 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
957
958 mutex_unlock(&sit_i->sentry_lock);
959 mutex_unlock(&curseg->curseg_mutex);
960}
961
962static int read_compacted_summaries(struct f2fs_sb_info *sbi)
963{
964 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
965 struct curseg_info *seg_i;
966 unsigned char *kaddr;
967 struct page *page;
968 block_t start;
969 int i, j, offset;
970
971 start = start_sum_block(sbi);
972
973 page = get_meta_page(sbi, start++);
974 kaddr = (unsigned char *)page_address(page);
975
976 /* Step 1: restore nat cache */
977 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
978 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
979
980 /* Step 2: restore sit cache */
981 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
982 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
983 SUM_JOURNAL_SIZE);
984 offset = 2 * SUM_JOURNAL_SIZE;
985
986 /* Step 3: restore summary entries */
987 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
988 unsigned short blk_off;
989 unsigned int segno;
990
991 seg_i = CURSEG_I(sbi, i);
992 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
993 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
994 seg_i->next_segno = segno;
995 reset_curseg(sbi, i, 0);
996 seg_i->alloc_type = ckpt->alloc_type[i];
997 seg_i->next_blkoff = blk_off;
998
999 if (seg_i->alloc_type == SSR)
1000 blk_off = sbi->blocks_per_seg;
1001
1002 for (j = 0; j < blk_off; j++) {
1003 struct f2fs_summary *s;
1004 s = (struct f2fs_summary *)(kaddr + offset);
1005 seg_i->sum_blk->entries[j] = *s;
1006 offset += SUMMARY_SIZE;
1007 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1008 SUM_FOOTER_SIZE)
1009 continue;
1010
1011 f2fs_put_page(page, 1);
1012 page = NULL;
1013
1014 page = get_meta_page(sbi, start++);
1015 kaddr = (unsigned char *)page_address(page);
1016 offset = 0;
1017 }
1018 }
1019 f2fs_put_page(page, 1);
1020 return 0;
1021}
1022
1023static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1024{
1025 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1026 struct f2fs_summary_block *sum;
1027 struct curseg_info *curseg;
1028 struct page *new;
1029 unsigned short blk_off;
1030 unsigned int segno = 0;
1031 block_t blk_addr = 0;
1032
1033 /* get segment number and block addr */
1034 if (IS_DATASEG(type)) {
1035 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1036 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1037 CURSEG_HOT_DATA]);
25ca923b 1038 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
351df4b2
JK		 1039 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1040 else
1041 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1042 } else {
1043 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1044 CURSEG_HOT_NODE]);
1045 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1046 CURSEG_HOT_NODE]);
25ca923b 1047 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
351df4b2
JK		 1048 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1049 type - CURSEG_HOT_NODE);
1050 else
1051 blk_addr = GET_SUM_BLOCK(sbi, segno);
1052 }
1053
1054 new = get_meta_page(sbi, blk_addr);
1055 sum = (struct f2fs_summary_block *)page_address(new);
1056
1057 if (IS_NODESEG(type)) {
25ca923b 1058 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
351df4b2
JK		 1059 struct f2fs_summary *ns = &sum->entries[0];
1060 int i;
1061 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1062 ns->version = 0;
1063 ns->ofs_in_node = 0;
1064 }
1065 } else {
1066 if (restore_node_summary(sbi, segno, sum)) {
1067 f2fs_put_page(new, 1);
1068 return -EINVAL;
1069 }
1070 }
1071 }
1072
1073 /* set uncompleted segment to curseg */
1074 curseg = CURSEG_I(sbi, type);
1075 mutex_lock(&curseg->curseg_mutex);
1076 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1077 curseg->next_segno = segno;
1078 reset_curseg(sbi, type, 0);
1079 curseg->alloc_type = ckpt->alloc_type[type];
1080 curseg->next_blkoff = blk_off;
1081 mutex_unlock(&curseg->curseg_mutex);
1082 f2fs_put_page(new, 1);
1083 return 0;
1084}
1085
1086static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1087{
1088 int type = CURSEG_HOT_DATA;
1089
25ca923b 1090 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
351df4b2
JK		 1091 /* restore for compacted data summary */
1092 if (read_compacted_summaries(sbi))
1093 return -EINVAL;
1094 type = CURSEG_HOT_NODE;
1095 }
1096
1097 for (; type <= CURSEG_COLD_NODE; type++)
1098 if (read_normal_summaries(sbi, type))
1099 return -EINVAL;
1100 return 0;
1101}
1102
1103static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1104{
1105 struct page *page;
1106 unsigned char *kaddr;
1107 struct f2fs_summary *summary;
1108 struct curseg_info *seg_i;
1109 int written_size = 0;
1110 int i, j;
1111
1112 page = grab_meta_page(sbi, blkaddr++);
1113 kaddr = (unsigned char *)page_address(page);
1114
1115 /* Step 1: write nat cache */
1116 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1117 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1118 written_size += SUM_JOURNAL_SIZE;
1119
1120 /* Step 2: write sit cache */
1121 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1122 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1123 SUM_JOURNAL_SIZE);
1124 written_size += SUM_JOURNAL_SIZE;
1125
351df4b2
JK		 1126 /* Step 3: write summary entries */
1127 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1128 unsigned short blkoff;
1129 seg_i = CURSEG_I(sbi, i);
1130 if (sbi->ckpt->alloc_type[i] == SSR)
1131 blkoff = sbi->blocks_per_seg;
1132 else
1133 blkoff = curseg_blkoff(sbi, i);
1134
1135 for (j = 0; j < blkoff; j++) {
1136 if (!page) {
1137 page = grab_meta_page(sbi, blkaddr++);
1138 kaddr = (unsigned char *)page_address(page);
1139 written_size = 0;
1140 }
1141 summary = (struct f2fs_summary *)(kaddr + written_size);
1142 *summary = seg_i->sum_blk->entries[j];
1143 written_size += SUMMARY_SIZE;
351df4b2
JK		 1144
1145 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1146 SUM_FOOTER_SIZE)
1147 continue;
1148
e8d61a74 1149 set_page_dirty(page);
351df4b2
JK		 1150 f2fs_put_page(page, 1);
1151 page = NULL;
1152 }
1153 }
e8d61a74
CY		 1154 if (page) {
1155 set_page_dirty(page);
351df4b2 1156 f2fs_put_page(page, 1);
e8d61a74 1157 }
351df4b2
JK		 1158}
1159
1160static void write_normal_summaries(struct f2fs_sb_info *sbi,
1161 block_t blkaddr, int type)
1162{
1163 int i, end;
1164 if (IS_DATASEG(type))
1165 end = type + NR_CURSEG_DATA_TYPE;
1166 else
1167 end = type + NR_CURSEG_NODE_TYPE;
1168
1169 for (i = type; i < end; i++) {
1170 struct curseg_info *sum = CURSEG_I(sbi, i);
1171 mutex_lock(&sum->curseg_mutex);
1172 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1173 mutex_unlock(&sum->curseg_mutex);
1174 }
1175}
1176
1177void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1178{
25ca923b 1179 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1180 write_compacted_summaries(sbi, start_blk);
1181 else
1182 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1183}
1184
1185void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1186{
25ca923b 1187 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
351df4b2 1188 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1189}
1190
1191int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1192 unsigned int val, int alloc)
1193{
1194 int i;
1195
1196 if (type == NAT_JOURNAL) {
1197 for (i = 0; i < nats_in_cursum(sum); i++) {
1198 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1199 return i;
1200 }
1201 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1202 return update_nats_in_cursum(sum, 1);
1203 } else if (type == SIT_JOURNAL) {
1204 for (i = 0; i < sits_in_cursum(sum); i++)
1205 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1206 return i;
1207 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1208 return update_sits_in_cursum(sum, 1);
1209 }
1210 return -1;
1211}
1212
1213static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1214 unsigned int segno)
1215{
1216 struct sit_info *sit_i = SIT_I(sbi);
1217 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1218 block_t blk_addr = sit_i->sit_base_addr + offset;
1219
1220 check_seg_range(sbi, segno);
1221
1222 /* calculate sit block address */
1223 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1224 blk_addr += sit_i->sit_blocks;
1225
1226 return get_meta_page(sbi, blk_addr);
1227}
1228
1229static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1230 unsigned int start)
1231{
1232 struct sit_info *sit_i = SIT_I(sbi);
1233 struct page *src_page, *dst_page;
1234 pgoff_t src_off, dst_off;
1235 void *src_addr, *dst_addr;
1236
1237 src_off = current_sit_addr(sbi, start);
1238 dst_off = next_sit_addr(sbi, src_off);
1239
1240 /* get current sit block page without lock */
1241 src_page = get_meta_page(sbi, src_off);
1242 dst_page = grab_meta_page(sbi, dst_off);
1243 BUG_ON(PageDirty(src_page));
1244
1245 src_addr = page_address(src_page);
1246 dst_addr = page_address(dst_page);
1247 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1248
1249 set_page_dirty(dst_page);
1250 f2fs_put_page(src_page, 1);
1251
1252 set_to_next_sit(sit_i, start);
1253
1254 return dst_page;
1255}
1256
1257static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1258{
1259 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1260 struct f2fs_summary_block *sum = curseg->sum_blk;
1261 int i;
1262
1263 /*
1264 * If the journal area in the current summary is full of sit entries,
1265 * all the sit entries will be flushed. Otherwise the sit entries
1266 * are not able to replace with newly hot sit entries.
1267 */
1268 if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1269 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1270 unsigned int segno;
1271 segno = le32_to_cpu(segno_in_journal(sum, i));
1272 __mark_sit_entry_dirty(sbi, segno);
1273 }
1274 update_sits_in_cursum(sum, -sits_in_cursum(sum));
cffbfa66 1275 return true;
351df4b2 1276 }
cffbfa66 1277 return false;
351df4b2
JK		 1278}
1279
0a8165d7 1280/*
351df4b2
JK		 1281 * CP calls this function, which flushes SIT entries including sit_journal,
1282 * and moves prefree segs to free segs.
1283 */
1284void flush_sit_entries(struct f2fs_sb_info *sbi)
1285{
1286 struct sit_info *sit_i = SIT_I(sbi);
1287 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1288 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1289 struct f2fs_summary_block *sum = curseg->sum_blk;
1290 unsigned long nsegs = TOTAL_SEGS(sbi);
1291 struct page *page = NULL;
1292 struct f2fs_sit_block *raw_sit = NULL;
1293 unsigned int start = 0, end = 0;
1294 unsigned int segno = -1;
1295 bool flushed;
1296
1297 mutex_lock(&curseg->curseg_mutex);
1298 mutex_lock(&sit_i->sentry_lock);
1299
1300 /*
1301 * "flushed" indicates whether sit entries in journal are flushed
1302 * to the SIT area or not.
1303 */
1304 flushed = flush_sits_in_journal(sbi);
1305
1306 while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1307 struct seg_entry *se = get_seg_entry(sbi, segno);
1308 int sit_offset, offset;
1309
1310 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1311
1312 if (flushed)
1313 goto to_sit_page;
1314
1315 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1316 if (offset >= 0) {
1317 segno_in_journal(sum, offset) = cpu_to_le32(segno);
1318 seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1319 goto flush_done;
1320 }
1321to_sit_page:
1322 if (!page || (start > segno) || (segno > end)) {
1323 if (page) {
1324 f2fs_put_page(page, 1);
1325 page = NULL;
1326 }
1327
1328 start = START_SEGNO(sit_i, segno);
1329 end = start + SIT_ENTRY_PER_BLOCK - 1;
1330
1331 /* read sit block that will be updated */
1332 page = get_next_sit_page(sbi, start);
1333 raw_sit = page_address(page);
1334 }
1335
1336 /* udpate entry in SIT block */
1337 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1338flush_done:
1339 __clear_bit(segno, bitmap);
1340 sit_i->dirty_sentries--;
1341 }
1342 mutex_unlock(&sit_i->sentry_lock);
1343 mutex_unlock(&curseg->curseg_mutex);
1344
1345 /* writeout last modified SIT block */
1346 f2fs_put_page(page, 1);
1347
1348 set_prefree_as_free_segments(sbi);
1349}
1350
1351static int build_sit_info(struct f2fs_sb_info *sbi)
1352{
1353 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1354 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1355 struct sit_info *sit_i;
1356 unsigned int sit_segs, start;
1357 char *src_bitmap, *dst_bitmap;
1358 unsigned int bitmap_size;
1359
1360 /* allocate memory for SIT information */
1361 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1362 if (!sit_i)
1363 return -ENOMEM;
1364
1365 SM_I(sbi)->sit_info = sit_i;
1366
1367 sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1368 if (!sit_i->sentries)
1369 return -ENOMEM;
1370
1371 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1372 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1373 if (!sit_i->dirty_sentries_bitmap)
1374 return -ENOMEM;
1375
1376 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1377 sit_i->sentries[start].cur_valid_map
1378 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1379 sit_i->sentries[start].ckpt_valid_map
1380 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1381 if (!sit_i->sentries[start].cur_valid_map
1382 || !sit_i->sentries[start].ckpt_valid_map)
1383 return -ENOMEM;
1384 }
1385
1386 if (sbi->segs_per_sec > 1) {
53cf9522 1387 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
351df4b2
JK		 1388 sizeof(struct sec_entry));
1389 if (!sit_i->sec_entries)
1390 return -ENOMEM;
1391 }
1392
1393 /* get information related with SIT */
1394 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1395
1396 /* setup SIT bitmap from ckeckpoint pack */
1397 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1398 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1399
79b5793b 1400 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 1401 if (!dst_bitmap)
1402 return -ENOMEM;
351df4b2
JK		 1403
1404 /* init SIT information */
1405 sit_i->s_ops = &default_salloc_ops;
1406
1407 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1408 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1409 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1410 sit_i->sit_bitmap = dst_bitmap;
1411 sit_i->bitmap_size = bitmap_size;
1412 sit_i->dirty_sentries = 0;
1413 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1414 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1415 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1416 mutex_init(&sit_i->sentry_lock);
1417 return 0;
1418}
1419
1420static int build_free_segmap(struct f2fs_sb_info *sbi)
1421{
1422 struct f2fs_sm_info *sm_info = SM_I(sbi);
1423 struct free_segmap_info *free_i;
1424 unsigned int bitmap_size, sec_bitmap_size;
1425
1426 /* allocate memory for free segmap information */
1427 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1428 if (!free_i)
1429 return -ENOMEM;
1430
1431 SM_I(sbi)->free_info = free_i;
1432
1433 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1434 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1435 if (!free_i->free_segmap)
1436 return -ENOMEM;
1437
53cf9522 1438 sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
351df4b2
JK		 1439 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1440 if (!free_i->free_secmap)
1441 return -ENOMEM;
1442
1443 /* set all segments as dirty temporarily */
1444 memset(free_i->free_segmap, 0xff, bitmap_size);
1445 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1446
1447 /* init free segmap information */
1448 free_i->start_segno =
1449 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1450 free_i->free_segments = 0;
1451 free_i->free_sections = 0;
1452 rwlock_init(&free_i->segmap_lock);
1453 return 0;
1454}
1455
1456static int build_curseg(struct f2fs_sb_info *sbi)
1457{
1042d60f 1458 struct curseg_info *array;
351df4b2
JK		 1459 int i;
1460
1461 array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1462 if (!array)
1463 return -ENOMEM;
1464
1465 SM_I(sbi)->curseg_array = array;
1466
1467 for (i = 0; i < NR_CURSEG_TYPE; i++) {
1468 mutex_init(&array[i].curseg_mutex);
1469 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1470 if (!array[i].sum_blk)
1471 return -ENOMEM;
1472 array[i].segno = NULL_SEGNO;
1473 array[i].next_blkoff = 0;
1474 }
1475 return restore_curseg_summaries(sbi);
1476}
1477
1478static void build_sit_entries(struct f2fs_sb_info *sbi)
1479{
1480 struct sit_info *sit_i = SIT_I(sbi);
1481 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1482 struct f2fs_summary_block *sum = curseg->sum_blk;
1483 unsigned int start;
1484
1485 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1486 struct seg_entry *se = &sit_i->sentries[start];
1487 struct f2fs_sit_block *sit_blk;
1488 struct f2fs_sit_entry sit;
1489 struct page *page;
1490 int i;
1491
1492 mutex_lock(&curseg->curseg_mutex);
1493 for (i = 0; i < sits_in_cursum(sum); i++) {
1494 if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1495 sit = sit_in_journal(sum, i);
1496 mutex_unlock(&curseg->curseg_mutex);
1497 goto got_it;
1498 }
1499 }
1500 mutex_unlock(&curseg->curseg_mutex);
1501 page = get_current_sit_page(sbi, start);
1502 sit_blk = (struct f2fs_sit_block *)page_address(page);
1503 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1504 f2fs_put_page(page, 1);
1505got_it:
1506 check_block_count(sbi, start, &sit);
1507 seg_info_from_raw_sit(se, &sit);
1508 if (sbi->segs_per_sec > 1) {
1509 struct sec_entry *e = get_sec_entry(sbi, start);
1510 e->valid_blocks += se->valid_blocks;
1511 }
1512 }
1513}
1514
1515static void init_free_segmap(struct f2fs_sb_info *sbi)
1516{
1517 unsigned int start;
1518 int type;
1519
1520 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1521 struct seg_entry *sentry = get_seg_entry(sbi, start);
1522 if (!sentry->valid_blocks)
1523 __set_free(sbi, start);
1524 }
1525
1526 /* set use the current segments */
1527 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1528 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1529 __set_test_and_inuse(sbi, curseg_t->segno);
1530 }
1531}
1532
1533static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1534{
1535 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1536 struct free_segmap_info *free_i = FREE_I(sbi);
8736fbf0 1537 unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
351df4b2
JK		 1538 unsigned short valid_blocks;
1539
8736fbf0 1540 while (1) {
351df4b2 1541 /* find dirty segment based on free segmap */
8736fbf0
NJ		 1542 segno = find_next_inuse(free_i, total_segs, offset);
1543 if (segno >= total_segs)
351df4b2
JK		 1544 break;
1545 offset = segno + 1;
1546 valid_blocks = get_valid_blocks(sbi, segno, 0);
1547 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1548 continue;
1549 mutex_lock(&dirty_i->seglist_lock);
1550 __locate_dirty_segment(sbi, segno, DIRTY);
1551 mutex_unlock(&dirty_i->seglist_lock);
1552 }
1553}
1554
5ec4e49f 1555static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 1556{
1557 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5ec4e49f 1558 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
351df4b2 1559
5ec4e49f
JK		 1560 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1561 if (!dirty_i->victim_secmap)
351df4b2
JK		 1562 return -ENOMEM;
1563 return 0;
1564}
1565
1566static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1567{
1568 struct dirty_seglist_info *dirty_i;
1569 unsigned int bitmap_size, i;
1570
1571 /* allocate memory for dirty segments list information */
1572 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1573 if (!dirty_i)
1574 return -ENOMEM;
1575
1576 SM_I(sbi)->dirty_info = dirty_i;
1577 mutex_init(&dirty_i->seglist_lock);
1578
1579 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1580
1581 for (i = 0; i < NR_DIRTY_TYPE; i++) {
1582 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 1583 if (!dirty_i->dirty_segmap[i])
1584 return -ENOMEM;
1585 }
1586
1587 init_dirty_segmap(sbi);
5ec4e49f 1588 return init_victim_secmap(sbi);
351df4b2
JK		 1589}
1590
0a8165d7 1591/*
351df4b2
JK		 1592 * Update min, max modified time for cost-benefit GC algorithm
1593 */
1594static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1595{
1596 struct sit_info *sit_i = SIT_I(sbi);
1597 unsigned int segno;
1598
1599 mutex_lock(&sit_i->sentry_lock);
1600
1601 sit_i->min_mtime = LLONG_MAX;
1602
1603 for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1604 unsigned int i;
1605 unsigned long long mtime = 0;
1606
1607 for (i = 0; i < sbi->segs_per_sec; i++)
1608 mtime += get_seg_entry(sbi, segno + i)->mtime;
1609
1610 mtime = div_u64(mtime, sbi->segs_per_sec);
1611
1612 if (sit_i->min_mtime > mtime)
1613 sit_i->min_mtime = mtime;
1614 }
1615 sit_i->max_mtime = get_mtime(sbi);
1616 mutex_unlock(&sit_i->sentry_lock);
1617}
1618
1619int build_segment_manager(struct f2fs_sb_info *sbi)
1620{
1621 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1622 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 1623 struct f2fs_sm_info *sm_info;
351df4b2
JK		 1624 int err;
1625
1626 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1627 if (!sm_info)
1628 return -ENOMEM;
1629
1630 /* init sm info */
1631 sbi->sm_info = sm_info;
1632 INIT_LIST_HEAD(&sm_info->wblist_head);
1633 spin_lock_init(&sm_info->wblist_lock);
1634 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1635 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1636 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1637 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1638 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1639 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1640 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
81eb8d6e 1641 sm_info->rec_prefree_segments = DEF_RECLAIM_PREFREE_SEGMENTS;
351df4b2
JK		 1642
1643 err = build_sit_info(sbi);
1644 if (err)
1645 return err;
1646 err = build_free_segmap(sbi);
1647 if (err)
1648 return err;
1649 err = build_curseg(sbi);
1650 if (err)
1651 return err;
1652
1653 /* reinit free segmap based on SIT */
1654 build_sit_entries(sbi);
1655
1656 init_free_segmap(sbi);
1657 err = build_dirty_segmap(sbi);
1658 if (err)
1659 return err;
1660
1661 init_min_max_mtime(sbi);
1662 return 0;
1663}
1664
1665static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1666 enum dirty_type dirty_type)
1667{
1668 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1669
1670 mutex_lock(&dirty_i->seglist_lock);
1671 kfree(dirty_i->dirty_segmap[dirty_type]);
1672 dirty_i->nr_dirty[dirty_type] = 0;
1673 mutex_unlock(&dirty_i->seglist_lock);
1674}
1675
5ec4e49f 1676static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 1677{
1678 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5ec4e49f 1679 kfree(dirty_i->victim_secmap);
351df4b2
JK		 1680}
1681
1682static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1683{
1684 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1685 int i;
1686
1687 if (!dirty_i)
1688 return;
1689
1690 /* discard pre-free/dirty segments list */
1691 for (i = 0; i < NR_DIRTY_TYPE; i++)
1692 discard_dirty_segmap(sbi, i);
1693
5ec4e49f 1694 destroy_victim_secmap(sbi);
351df4b2
JK		 1695 SM_I(sbi)->dirty_info = NULL;
1696 kfree(dirty_i);
1697}
1698
1699static void destroy_curseg(struct f2fs_sb_info *sbi)
1700{
1701 struct curseg_info *array = SM_I(sbi)->curseg_array;
1702 int i;
1703
1704 if (!array)
1705 return;
1706 SM_I(sbi)->curseg_array = NULL;
1707 for (i = 0; i < NR_CURSEG_TYPE; i++)
1708 kfree(array[i].sum_blk);
1709 kfree(array);
1710}
1711
1712static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1713{
1714 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1715 if (!free_i)
1716 return;
1717 SM_I(sbi)->free_info = NULL;
1718 kfree(free_i->free_segmap);
1719 kfree(free_i->free_secmap);
1720 kfree(free_i);
1721}
1722
1723static void destroy_sit_info(struct f2fs_sb_info *sbi)
1724{
1725 struct sit_info *sit_i = SIT_I(sbi);
1726 unsigned int start;
1727
1728 if (!sit_i)
1729 return;
1730
1731 if (sit_i->sentries) {
1732 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1733 kfree(sit_i->sentries[start].cur_valid_map);
1734 kfree(sit_i->sentries[start].ckpt_valid_map);
1735 }
1736 }
1737 vfree(sit_i->sentries);
1738 vfree(sit_i->sec_entries);
1739 kfree(sit_i->dirty_sentries_bitmap);
1740
1741 SM_I(sbi)->sit_info = NULL;
1742 kfree(sit_i->sit_bitmap);
1743 kfree(sit_i);
1744}
1745
1746void destroy_segment_manager(struct f2fs_sb_info *sbi)
1747{
1748 struct f2fs_sm_info *sm_info = SM_I(sbi);
1749 destroy_dirty_segmap(sbi);
1750 destroy_curseg(sbi);
1751 destroy_free_segmap(sbi);
1752 destroy_sit_info(sbi);
1753 sbi->sm_info = NULL;
1754 kfree(sm_info);
1755}
Linux kernel - Deliverables
This page took 0.123111 seconds and 5 git commands to generate.