projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: move bio_private allocation out of f2fs_bio_alloc()
[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
39static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
40 enum dirty_type dirty_type)
41{
42 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
43
44 /* need not be added */
45 if (IS_CURSEG(sbi, segno))
46 return;
47
48 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
49 dirty_i->nr_dirty[dirty_type]++;
50
51 if (dirty_type == DIRTY) {
52 struct seg_entry *sentry = get_seg_entry(sbi, segno);
b2f2c390
JK		 53 enum dirty_type t = DIRTY_HOT_DATA;
54
351df4b2 55 dirty_type = sentry->type;
b2f2c390 56
351df4b2
JK		 57 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
58 dirty_i->nr_dirty[dirty_type]++;
b2f2c390
JK		 59
60 /* Only one bitmap should be set */
61 for (; t <= DIRTY_COLD_NODE; t++) {
62 if (t == dirty_type)
63 continue;
64 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
65 dirty_i->nr_dirty[t]--;
66 }
351df4b2
JK		 67 }
68}
69
70static void __remove_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 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
76 dirty_i->nr_dirty[dirty_type]--;
77
78 if (dirty_type == DIRTY) {
b2f2c390
JK		 79 enum dirty_type t = DIRTY_HOT_DATA;
80
81 /* clear all the bitmaps */
82 for (; t <= DIRTY_COLD_NODE; t++)
83 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
84 dirty_i->nr_dirty[t]--;
85
5ec4e49f
JK		 86 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
87 clear_bit(GET_SECNO(sbi, segno),
88 dirty_i->victim_secmap);
351df4b2
JK		 89 }
90}
91
0a8165d7 92/*
351df4b2
JK		 93 * Should not occur error such as -ENOMEM.
94 * Adding dirty entry into seglist is not critical operation.
95 * If a given segment is one of current working segments, it won't be added.
96 */
8d8451af 97static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 98{
99 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
100 unsigned short valid_blocks;
101
102 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
103 return;
104
105 mutex_lock(&dirty_i->seglist_lock);
106
107 valid_blocks = get_valid_blocks(sbi, segno, 0);
108
109 if (valid_blocks == 0) {
110 __locate_dirty_segment(sbi, segno, PRE);
111 __remove_dirty_segment(sbi, segno, DIRTY);
112 } else if (valid_blocks < sbi->blocks_per_seg) {
113 __locate_dirty_segment(sbi, segno, DIRTY);
114 } else {
115 /* Recovery routine with SSR needs this */
116 __remove_dirty_segment(sbi, segno, DIRTY);
117 }
118
119 mutex_unlock(&dirty_i->seglist_lock);
120 return;
121}
122
0a8165d7 123/*
351df4b2
JK		 124 * Should call clear_prefree_segments after checkpoint is done.
125 */
126static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
127{
128 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
81fb5e87 129 unsigned int segno = -1;
351df4b2
JK		 130 unsigned int total_segs = TOTAL_SEGS(sbi);
131
132 mutex_lock(&dirty_i->seglist_lock);
133 while (1) {
134 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
81fb5e87 135 segno + 1);
351df4b2
JK		 136 if (segno >= total_segs)
137 break;
138 __set_test_and_free(sbi, segno);
351df4b2
JK		 139 }
140 mutex_unlock(&dirty_i->seglist_lock);
141}
142
143void clear_prefree_segments(struct f2fs_sb_info *sbi)
144{
145 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
81fb5e87 146 unsigned int segno = -1;
351df4b2
JK		 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,
81fb5e87 152 segno + 1);
351df4b2
JK		 153 if (segno >= total_segs)
154 break;
155
351df4b2
JK		 156 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
157 dirty_i->nr_dirty[PRE]--;
158
159 /* Let's use trim */
160 if (test_opt(sbi, DISCARD))
161 blkdev_issue_discard(sbi->sb->s_bdev,
162 START_BLOCK(sbi, segno) <<
163 sbi->log_sectors_per_block,
164 1 << (sbi->log_sectors_per_block +
165 sbi->log_blocks_per_seg),
166 GFP_NOFS, 0);
167 }
168 mutex_unlock(&dirty_i->seglist_lock);
169}
170
171static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
172{
173 struct sit_info *sit_i = SIT_I(sbi);
174 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
175 sit_i->dirty_sentries++;
176}
177
178static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
179 unsigned int segno, int modified)
180{
181 struct seg_entry *se = get_seg_entry(sbi, segno);
182 se->type = type;
183 if (modified)
184 __mark_sit_entry_dirty(sbi, segno);
185}
186
187static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
188{
189 struct seg_entry *se;
190 unsigned int segno, offset;
191 long int new_vblocks;
192
193 segno = GET_SEGNO(sbi, blkaddr);
194
195 se = get_seg_entry(sbi, segno);
196 new_vblocks = se->valid_blocks + del;
197 offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
198
199 BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) ||
200 (new_vblocks > sbi->blocks_per_seg)));
201
202 se->valid_blocks = new_vblocks;
203 se->mtime = get_mtime(sbi);
204 SIT_I(sbi)->max_mtime = se->mtime;
205
206 /* Update valid block bitmap */
207 if (del > 0) {
208 if (f2fs_set_bit(offset, se->cur_valid_map))
209 BUG();
210 } else {
211 if (!f2fs_clear_bit(offset, se->cur_valid_map))
212 BUG();
213 }
214 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
215 se->ckpt_valid_blocks += del;
216
217 __mark_sit_entry_dirty(sbi, segno);
218
219 /* update total number of valid blocks to be written in ckpt area */
220 SIT_I(sbi)->written_valid_blocks += del;
221
222 if (sbi->segs_per_sec > 1)
223 get_sec_entry(sbi, segno)->valid_blocks += del;
224}
225
226static void refresh_sit_entry(struct f2fs_sb_info *sbi,
227 block_t old_blkaddr, block_t new_blkaddr)
228{
229 update_sit_entry(sbi, new_blkaddr, 1);
230 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
231 update_sit_entry(sbi, old_blkaddr, -1);
232}
233
234void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
235{
236 unsigned int segno = GET_SEGNO(sbi, addr);
237 struct sit_info *sit_i = SIT_I(sbi);
238
239 BUG_ON(addr == NULL_ADDR);
240 if (addr == NEW_ADDR)
241 return;
242
243 /* add it into sit main buffer */
244 mutex_lock(&sit_i->sentry_lock);
245
246 update_sit_entry(sbi, addr, -1);
247
248 /* add it into dirty seglist */
249 locate_dirty_segment(sbi, segno);
250
251 mutex_unlock(&sit_i->sentry_lock);
252}
253
0a8165d7 254/*
351df4b2
JK		 255 * This function should be resided under the curseg_mutex lock
256 */
257static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 258 struct f2fs_summary *sum)
351df4b2
JK		 259{
260 struct curseg_info *curseg = CURSEG_I(sbi, type);
261 void *addr = curseg->sum_blk;
e79efe3b 262 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2
JK		 263 memcpy(addr, sum, sizeof(struct f2fs_summary));
264 return;
265}
266
0a8165d7 267/*
351df4b2
JK		 268 * Calculate the number of current summary pages for writing
269 */
270int npages_for_summary_flush(struct f2fs_sb_info *sbi)
271{
272 int total_size_bytes = 0;
273 int valid_sum_count = 0;
274 int i, sum_space;
275
276 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
277 if (sbi->ckpt->alloc_type[i] == SSR)
278 valid_sum_count += sbi->blocks_per_seg;
279 else
280 valid_sum_count += curseg_blkoff(sbi, i);
281 }
282
283 total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1)
284 + sizeof(struct nat_journal) + 2
285 + sizeof(struct sit_journal) + 2;
286 sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE;
287 if (total_size_bytes < sum_space)
288 return 1;
289 else if (total_size_bytes < 2 * sum_space)
290 return 2;
291 return 3;
292}
293
0a8165d7 294/*
351df4b2
JK		 295 * Caller should put this summary page
296 */
297struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
298{
299 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
300}
301
302static void write_sum_page(struct f2fs_sb_info *sbi,
303 struct f2fs_summary_block *sum_blk, block_t blk_addr)
304{
305 struct page *page = grab_meta_page(sbi, blk_addr);
306 void *kaddr = page_address(page);
307 memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
308 set_page_dirty(page);
309 f2fs_put_page(page, 1);
310}
311
60374688
JK		 312static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
313{
314 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 315 unsigned int segno = curseg->segno + 1;
60374688
JK		 316 struct free_segmap_info *free_i = FREE_I(sbi);
317
81fb5e87
HL		 318 if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
319 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 320 return 0;
321}
322
0a8165d7 323/*
351df4b2
JK		 324 * Find a new segment from the free segments bitmap to right order
325 * This function should be returned with success, otherwise BUG
326 */
327static void get_new_segment(struct f2fs_sb_info *sbi,
328 unsigned int *newseg, bool new_sec, int dir)
329{
330 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 331 unsigned int segno, secno, zoneno;
53cf9522 332 unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 333 unsigned int hint = *newseg / sbi->segs_per_sec;
334 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
335 unsigned int left_start = hint;
336 bool init = true;
337 int go_left = 0;
338 int i;
339
340 write_lock(&free_i->segmap_lock);
341
342 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
343 segno = find_next_zero_bit(free_i->free_segmap,
344 TOTAL_SEGS(sbi), *newseg + 1);
33afa7fd
JK		 345 if (segno - *newseg < sbi->segs_per_sec -
346 (*newseg % sbi->segs_per_sec))
351df4b2
JK		 347 goto got_it;
348 }
349find_other_zone:
53cf9522
JK		 350 secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
351 if (secno >= TOTAL_SECS(sbi)) {
351df4b2
JK		 352 if (dir == ALLOC_RIGHT) {
353 secno = find_next_zero_bit(free_i->free_secmap,
53cf9522
JK		 354 TOTAL_SECS(sbi), 0);
355 BUG_ON(secno >= TOTAL_SECS(sbi));
351df4b2
JK		 356 } else {
357 go_left = 1;
358 left_start = hint - 1;
359 }
360 }
361 if (go_left == 0)
362 goto skip_left;
363
364 while (test_bit(left_start, free_i->free_secmap)) {
365 if (left_start > 0) {
366 left_start--;
367 continue;
368 }
369 left_start = find_next_zero_bit(free_i->free_secmap,
53cf9522
JK		 370 TOTAL_SECS(sbi), 0);
371 BUG_ON(left_start >= TOTAL_SECS(sbi));
351df4b2
JK		 372 break;
373 }
374 secno = left_start;
375skip_left:
376 hint = secno;
377 segno = secno * sbi->segs_per_sec;
378 zoneno = secno / sbi->secs_per_zone;
379
380 /* give up on finding another zone */
381 if (!init)
382 goto got_it;
383 if (sbi->secs_per_zone == 1)
384 goto got_it;
385 if (zoneno == old_zoneno)
386 goto got_it;
387 if (dir == ALLOC_LEFT) {
388 if (!go_left && zoneno + 1 >= total_zones)
389 goto got_it;
390 if (go_left && zoneno == 0)
391 goto got_it;
392 }
393 for (i = 0; i < NR_CURSEG_TYPE; i++)
394 if (CURSEG_I(sbi, i)->zone == zoneno)
395 break;
396
397 if (i < NR_CURSEG_TYPE) {
398 /* zone is in user, try another */
399 if (go_left)
400 hint = zoneno * sbi->secs_per_zone - 1;
401 else if (zoneno + 1 >= total_zones)
402 hint = 0;
403 else
404 hint = (zoneno + 1) * sbi->secs_per_zone;
405 init = false;
406 goto find_other_zone;
407 }
408got_it:
409 /* set it as dirty segment in free segmap */
410 BUG_ON(test_bit(segno, free_i->free_segmap));
411 __set_inuse(sbi, segno);
412 *newseg = segno;
413 write_unlock(&free_i->segmap_lock);
414}
415
416static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
417{
418 struct curseg_info *curseg = CURSEG_I(sbi, type);
419 struct summary_footer *sum_footer;
420
421 curseg->segno = curseg->next_segno;
422 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
423 curseg->next_blkoff = 0;
424 curseg->next_segno = NULL_SEGNO;
425
426 sum_footer = &(curseg->sum_blk->footer);
427 memset(sum_footer, 0, sizeof(struct summary_footer));
428 if (IS_DATASEG(type))
429 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
430 if (IS_NODESEG(type))
431 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
432 __set_sit_entry_type(sbi, type, curseg->segno, modified);
433}
434
0a8165d7 435/*
351df4b2
JK		 436 * Allocate a current working segment.
437 * This function always allocates a free segment in LFS manner.
438 */
439static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
440{
441 struct curseg_info *curseg = CURSEG_I(sbi, type);
442 unsigned int segno = curseg->segno;
443 int dir = ALLOC_LEFT;
444
445 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 446 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 447 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
448 dir = ALLOC_RIGHT;
449
450 if (test_opt(sbi, NOHEAP))
451 dir = ALLOC_RIGHT;
452
453 get_new_segment(sbi, &segno, new_sec, dir);
454 curseg->next_segno = segno;
455 reset_curseg(sbi, type, 1);
456 curseg->alloc_type = LFS;
457}
458
459static void __next_free_blkoff(struct f2fs_sb_info *sbi,
460 struct curseg_info *seg, block_t start)
461{
462 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
463 block_t ofs;
464 for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
465 if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
466 && !f2fs_test_bit(ofs, se->cur_valid_map))
467 break;
468 }
469 seg->next_blkoff = ofs;
470}
471
0a8165d7 472/*
351df4b2
JK		 473 * If a segment is written by LFS manner, next block offset is just obtained
474 * by increasing the current block offset. However, if a segment is written by
475 * SSR manner, next block offset obtained by calling __next_free_blkoff
476 */
477static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
478 struct curseg_info *seg)
479{
480 if (seg->alloc_type == SSR)
481 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
482 else
483 seg->next_blkoff++;
484}
485
0a8165d7 486/*
351df4b2
JK		 487 * This function always allocates a used segment (from dirty seglist) by SSR
488 * manner, so it should recover the existing segment information of valid blocks
489 */
490static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
491{
492 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
493 struct curseg_info *curseg = CURSEG_I(sbi, type);
494 unsigned int new_segno = curseg->next_segno;
495 struct f2fs_summary_block *sum_node;
496 struct page *sum_page;
497
498 write_sum_page(sbi, curseg->sum_blk,
499 GET_SUM_BLOCK(sbi, curseg->segno));
500 __set_test_and_inuse(sbi, new_segno);
501
502 mutex_lock(&dirty_i->seglist_lock);
503 __remove_dirty_segment(sbi, new_segno, PRE);
504 __remove_dirty_segment(sbi, new_segno, DIRTY);
505 mutex_unlock(&dirty_i->seglist_lock);
506
507 reset_curseg(sbi, type, 1);
508 curseg->alloc_type = SSR;
509 __next_free_blkoff(sbi, curseg, 0);
510
511 if (reuse) {
512 sum_page = get_sum_page(sbi, new_segno);
513 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
514 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
515 f2fs_put_page(sum_page, 1);
516 }
517}
518
43727527
JK		 519static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
520{
521 struct curseg_info *curseg = CURSEG_I(sbi, type);
522 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
523
524 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
525 return v_ops->get_victim(sbi,
526 &(curseg)->next_segno, BG_GC, type, SSR);
527
528 /* For data segments, let's do SSR more intensively */
529 for (; type >= CURSEG_HOT_DATA; type--)
530 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
531 BG_GC, type, SSR))
532 return 1;
533 return 0;
534}
535
351df4b2
JK		 536/*
537 * flush out current segment and replace it with new segment
538 * This function should be returned with success, otherwise BUG
539 */
540static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
541 int type, bool force)
542{
543 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 544
545 if (force) {
546 new_curseg(sbi, type, true);
547 goto out;
548 }
549
763bfe1b 550 if (type == CURSEG_WARM_NODE)
351df4b2 551 new_curseg(sbi, type, false);
60374688
JK		 552 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
553 new_curseg(sbi, type, false);
351df4b2
JK		 554 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
555 change_curseg(sbi, type, true);
556 else
557 new_curseg(sbi, type, false);
558out:
35b09d82 559#ifdef CONFIG_F2FS_STAT_FS
351df4b2 560 sbi->segment_count[curseg->alloc_type]++;
35b09d82
NJ		 561#endif
562 return;
351df4b2
JK		 563}
564
565void allocate_new_segments(struct f2fs_sb_info *sbi)
566{
567 struct curseg_info *curseg;
568 unsigned int old_curseg;
569 int i;
570
571 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
572 curseg = CURSEG_I(sbi, i);
573 old_curseg = curseg->segno;
574 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
575 locate_dirty_segment(sbi, old_curseg);
576 }
577}
578
579static const struct segment_allocation default_salloc_ops = {
580 .allocate_segment = allocate_segment_by_default,
581};
582
583static void f2fs_end_io_write(struct bio *bio, int err)
584{
585 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
586 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
587 struct bio_private *p = bio->bi_private;
588
589 do {
590 struct page *page = bvec->bv_page;
591
592 if (--bvec >= bio->bi_io_vec)
593 prefetchw(&bvec->bv_page->flags);
594 if (!uptodate) {
595 SetPageError(page);
596 if (page->mapping)
597 set_bit(AS_EIO, &page->mapping->flags);
25ca923b 598 set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
577e3495 599 p->sbi->sb->s_flags |= MS_RDONLY;
351df4b2
JK		 600 }
601 end_page_writeback(page);
602 dec_page_count(p->sbi, F2FS_WRITEBACK);
603 } while (bvec >= bio->bi_io_vec);
604
605 if (p->is_sync)
606 complete(p->wait);
607 kfree(p);
608 bio_put(bio);
609}
610
3cd8a239 611struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
351df4b2
JK		 612{
613 struct bio *bio;
3cd8a239
JK		 614
615 /* No failure on bio allocation */
616 bio = bio_alloc(GFP_NOIO, npages);
617 bio->bi_bdev = bdev;
d8207f69
GZ		 618 bio->bi_private = NULL;
619
351df4b2
JK		 620 return bio;
621}
622
623static void do_submit_bio(struct f2fs_sb_info *sbi,
624 enum page_type type, bool sync)
625{
626 int rw = sync ? WRITE_SYNC : WRITE;
627 enum page_type btype = type > META ? META : type;
628
629 if (type >= META_FLUSH)
630 rw = WRITE_FLUSH_FUA;
631
8680441c
NJ		 632 if (btype == META)
633 rw |= REQ_META;
634
351df4b2
JK		 635 if (sbi->bio[btype]) {
636 struct bio_private *p = sbi->bio[btype]->bi_private;
637 p->sbi = sbi;
638 sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
6ec178da
NJ		 639
640 trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]);
641
351df4b2
JK		 642 if (type == META_FLUSH) {
643 DECLARE_COMPLETION_ONSTACK(wait);
644 p->is_sync = true;
645 p->wait = &wait;
646 submit_bio(rw, sbi->bio[btype]);
647 wait_for_completion(&wait);
648 } else {
649 p->is_sync = false;
650 submit_bio(rw, sbi->bio[btype]);
651 }
652 sbi->bio[btype] = NULL;
653 }
654}
655
656void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
657{
658 down_write(&sbi->bio_sem);
659 do_submit_bio(sbi, type, sync);
660 up_write(&sbi->bio_sem);
661}
662
663static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
664 block_t blk_addr, enum page_type type)
665{
666 struct block_device *bdev = sbi->sb->s_bdev;
667
668 verify_block_addr(sbi, blk_addr);
669
670 down_write(&sbi->bio_sem);
671
672 inc_page_count(sbi, F2FS_WRITEBACK);
673
674 if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
675 do_submit_bio(sbi, type, false);
676alloc_new:
3cd8a239 677 if (sbi->bio[type] == NULL) {
d8207f69
GZ		 678 struct bio_private *priv;
679retry:
680 priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
681 if (!priv) {
682 cond_resched();
683 goto retry;
684 }
685
ac5d156c 686 sbi->bio[type] = f2fs_bio_alloc(bdev, max_hw_blocks(sbi));
3cd8a239 687 sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
d8207f69 688 sbi->bio[type]->bi_private = priv;
3cd8a239
JK		 689 /*
690 * The end_io will be assigned at the sumbission phase.
691 * Until then, let bio_add_page() merge consecutive IOs as much
692 * as possible.
693 */
694 }
351df4b2
JK		 695
696 if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
697 PAGE_CACHE_SIZE) {
698 do_submit_bio(sbi, type, false);
699 goto alloc_new;
700 }
701
702 sbi->last_block_in_bio[type] = blk_addr;
703
704 up_write(&sbi->bio_sem);
6ec178da 705 trace_f2fs_submit_write_page(page, blk_addr, type);
351df4b2
JK		 706}
707
708static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
709{
710 struct curseg_info *curseg = CURSEG_I(sbi, type);
711 if (curseg->next_blkoff < sbi->blocks_per_seg)
712 return true;
713 return false;
714}
715
716static int __get_segment_type_2(struct page *page, enum page_type p_type)
717{
718 if (p_type == DATA)
719 return CURSEG_HOT_DATA;
720 else
721 return CURSEG_HOT_NODE;
722}
723
724static int __get_segment_type_4(struct page *page, enum page_type p_type)
725{
726 if (p_type == DATA) {
727 struct inode *inode = page->mapping->host;
728
729 if (S_ISDIR(inode->i_mode))
730 return CURSEG_HOT_DATA;
731 else
732 return CURSEG_COLD_DATA;
733 } else {
734 if (IS_DNODE(page) && !is_cold_node(page))
735 return CURSEG_HOT_NODE;
736 else
737 return CURSEG_COLD_NODE;
738 }
739}
740
741static int __get_segment_type_6(struct page *page, enum page_type p_type)
742{
743 if (p_type == DATA) {
744 struct inode *inode = page->mapping->host;
745
746 if (S_ISDIR(inode->i_mode))
747 return CURSEG_HOT_DATA;
354a3399 748 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 749 return CURSEG_COLD_DATA;
750 else
751 return CURSEG_WARM_DATA;
752 } else {
753 if (IS_DNODE(page))
754 return is_cold_node(page) ? CURSEG_WARM_NODE :
755 CURSEG_HOT_NODE;
756 else
757 return CURSEG_COLD_NODE;
758 }
759}
760
761static int __get_segment_type(struct page *page, enum page_type p_type)
762{
763 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
764 switch (sbi->active_logs) {
765 case 2:
766 return __get_segment_type_2(page, p_type);
767 case 4:
768 return __get_segment_type_4(page, p_type);
351df4b2 769 }
12a67146
JK		 770 /* NR_CURSEG_TYPE(6) logs by default */
771 BUG_ON(sbi->active_logs != NR_CURSEG_TYPE);
772 return __get_segment_type_6(page, p_type);
351df4b2
JK		 773}
774
775static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
776 block_t old_blkaddr, block_t *new_blkaddr,
777 struct f2fs_summary *sum, enum page_type p_type)
778{
779 struct sit_info *sit_i = SIT_I(sbi);
780 struct curseg_info *curseg;
781 unsigned int old_cursegno;
782 int type;
783
784 type = __get_segment_type(page, p_type);
785 curseg = CURSEG_I(sbi, type);
786
787 mutex_lock(&curseg->curseg_mutex);
788
789 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
790 old_cursegno = curseg->segno;
791
792 /*
793 * __add_sum_entry should be resided under the curseg_mutex
794 * because, this function updates a summary entry in the
795 * current summary block.
796 */
e79efe3b 797 __add_sum_entry(sbi, type, sum);
351df4b2
JK		 798
799 mutex_lock(&sit_i->sentry_lock);
800 __refresh_next_blkoff(sbi, curseg);
35b09d82 801#ifdef CONFIG_F2FS_STAT_FS
351df4b2 802 sbi->block_count[curseg->alloc_type]++;
35b09d82 803#endif
351df4b2
JK		 804
805 /*
806 * SIT information should be updated before segment allocation,
807 * since SSR needs latest valid block information.
808 */
809 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
810
811 if (!__has_curseg_space(sbi, type))
812 sit_i->s_ops->allocate_segment(sbi, type, false);
813
814 locate_dirty_segment(sbi, old_cursegno);
815 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
816 mutex_unlock(&sit_i->sentry_lock);
817
818 if (p_type == NODE)
819 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
820
821 /* writeout dirty page into bdev */
822 submit_write_page(sbi, page, *new_blkaddr, p_type);
823
824 mutex_unlock(&curseg->curseg_mutex);
825}
826
577e3495 827void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 828{
351df4b2
JK		 829 set_page_writeback(page);
830 submit_write_page(sbi, page, page->index, META);
351df4b2
JK		 831}
832
833void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
834 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
835{
836 struct f2fs_summary sum;
837 set_summary(&sum, nid, 0, 0);
838 do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
839}
840
841void write_data_page(struct inode *inode, struct page *page,
842 struct dnode_of_data *dn, block_t old_blkaddr,
843 block_t *new_blkaddr)
844{
845 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
846 struct f2fs_summary sum;
847 struct node_info ni;
848
849 BUG_ON(old_blkaddr == NULL_ADDR);
850 get_node_info(sbi, dn->nid, &ni);
851 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
852
853 do_write_page(sbi, page, old_blkaddr,
854 new_blkaddr, &sum, DATA);
855}
856
857void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
858 block_t old_blk_addr)
859{
860 submit_write_page(sbi, page, old_blk_addr, DATA);
861}
862
863void recover_data_page(struct f2fs_sb_info *sbi,
864 struct page *page, struct f2fs_summary *sum,
865 block_t old_blkaddr, block_t new_blkaddr)
866{
867 struct sit_info *sit_i = SIT_I(sbi);
868 struct curseg_info *curseg;
869 unsigned int segno, old_cursegno;
870 struct seg_entry *se;
871 int type;
872
873 segno = GET_SEGNO(sbi, new_blkaddr);
874 se = get_seg_entry(sbi, segno);
875 type = se->type;
876
877 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
878 if (old_blkaddr == NULL_ADDR)
879 type = CURSEG_COLD_DATA;
880 else
881 type = CURSEG_WARM_DATA;
882 }
883 curseg = CURSEG_I(sbi, type);
884
885 mutex_lock(&curseg->curseg_mutex);
886 mutex_lock(&sit_i->sentry_lock);
887
888 old_cursegno = curseg->segno;
889
890 /* change the current segment */
891 if (segno != curseg->segno) {
892 curseg->next_segno = segno;
893 change_curseg(sbi, type, true);
894 }
895
896 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
897 (sbi->blocks_per_seg - 1);
e79efe3b 898 __add_sum_entry(sbi, type, sum);
351df4b2
JK		 899
900 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
901
902 locate_dirty_segment(sbi, old_cursegno);
903 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
904
905 mutex_unlock(&sit_i->sentry_lock);
906 mutex_unlock(&curseg->curseg_mutex);
907}
908
909void rewrite_node_page(struct f2fs_sb_info *sbi,
910 struct page *page, struct f2fs_summary *sum,
911 block_t old_blkaddr, block_t new_blkaddr)
912{
913 struct sit_info *sit_i = SIT_I(sbi);
914 int type = CURSEG_WARM_NODE;
915 struct curseg_info *curseg;
916 unsigned int segno, old_cursegno;
917 block_t next_blkaddr = next_blkaddr_of_node(page);
918 unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
919
920 curseg = CURSEG_I(sbi, type);
921
922 mutex_lock(&curseg->curseg_mutex);
923 mutex_lock(&sit_i->sentry_lock);
924
925 segno = GET_SEGNO(sbi, new_blkaddr);
926 old_cursegno = curseg->segno;
927
928 /* change the current segment */
929 if (segno != curseg->segno) {
930 curseg->next_segno = segno;
931 change_curseg(sbi, type, true);
932 }
933 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
934 (sbi->blocks_per_seg - 1);
e79efe3b 935 __add_sum_entry(sbi, type, sum);
351df4b2
JK		 936
937 /* change the current log to the next block addr in advance */
938 if (next_segno != segno) {
939 curseg->next_segno = next_segno;
940 change_curseg(sbi, type, true);
941 }
942 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
943 (sbi->blocks_per_seg - 1);
944
945 /* rewrite node page */
946 set_page_writeback(page);
947 submit_write_page(sbi, page, new_blkaddr, NODE);
948 f2fs_submit_bio(sbi, NODE, true);
949 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
950
951 locate_dirty_segment(sbi, old_cursegno);
952 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
953
954 mutex_unlock(&sit_i->sentry_lock);
955 mutex_unlock(&curseg->curseg_mutex);
956}
957
958static int read_compacted_summaries(struct f2fs_sb_info *sbi)
959{
960 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
961 struct curseg_info *seg_i;
962 unsigned char *kaddr;
963 struct page *page;
964 block_t start;
965 int i, j, offset;
966
967 start = start_sum_block(sbi);
968
969 page = get_meta_page(sbi, start++);
970 kaddr = (unsigned char *)page_address(page);
971
972 /* Step 1: restore nat cache */
973 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
974 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
975
976 /* Step 2: restore sit cache */
977 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
978 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
979 SUM_JOURNAL_SIZE);
980 offset = 2 * SUM_JOURNAL_SIZE;
981
982 /* Step 3: restore summary entries */
983 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
984 unsigned short blk_off;
985 unsigned int segno;
986
987 seg_i = CURSEG_I(sbi, i);
988 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
989 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
990 seg_i->next_segno = segno;
991 reset_curseg(sbi, i, 0);
992 seg_i->alloc_type = ckpt->alloc_type[i];
993 seg_i->next_blkoff = blk_off;
994
995 if (seg_i->alloc_type == SSR)
996 blk_off = sbi->blocks_per_seg;
997
998 for (j = 0; j < blk_off; j++) {
999 struct f2fs_summary *s;
1000 s = (struct f2fs_summary *)(kaddr + offset);
1001 seg_i->sum_blk->entries[j] = *s;
1002 offset += SUMMARY_SIZE;
1003 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1004 SUM_FOOTER_SIZE)
1005 continue;
1006
1007 f2fs_put_page(page, 1);
1008 page = NULL;
1009
1010 page = get_meta_page(sbi, start++);
1011 kaddr = (unsigned char *)page_address(page);
1012 offset = 0;
1013 }
1014 }
1015 f2fs_put_page(page, 1);
1016 return 0;
1017}
1018
1019static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1020{
1021 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1022 struct f2fs_summary_block *sum;
1023 struct curseg_info *curseg;
1024 struct page *new;
1025 unsigned short blk_off;
1026 unsigned int segno = 0;
1027 block_t blk_addr = 0;
1028
1029 /* get segment number and block addr */
1030 if (IS_DATASEG(type)) {
1031 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1032 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1033 CURSEG_HOT_DATA]);
25ca923b 1034 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
351df4b2
JK		 1035 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1036 else
1037 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1038 } else {
1039 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1040 CURSEG_HOT_NODE]);
1041 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1042 CURSEG_HOT_NODE]);
25ca923b 1043 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
351df4b2
JK		 1044 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1045 type - CURSEG_HOT_NODE);
1046 else
1047 blk_addr = GET_SUM_BLOCK(sbi, segno);
1048 }
1049
1050 new = get_meta_page(sbi, blk_addr);
1051 sum = (struct f2fs_summary_block *)page_address(new);
1052
1053 if (IS_NODESEG(type)) {
25ca923b 1054 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
351df4b2
JK		 1055 struct f2fs_summary *ns = &sum->entries[0];
1056 int i;
1057 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1058 ns->version = 0;
1059 ns->ofs_in_node = 0;
1060 }
1061 } else {
1062 if (restore_node_summary(sbi, segno, sum)) {
1063 f2fs_put_page(new, 1);
1064 return -EINVAL;
1065 }
1066 }
1067 }
1068
1069 /* set uncompleted segment to curseg */
1070 curseg = CURSEG_I(sbi, type);
1071 mutex_lock(&curseg->curseg_mutex);
1072 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1073 curseg->next_segno = segno;
1074 reset_curseg(sbi, type, 0);
1075 curseg->alloc_type = ckpt->alloc_type[type];
1076 curseg->next_blkoff = blk_off;
1077 mutex_unlock(&curseg->curseg_mutex);
1078 f2fs_put_page(new, 1);
1079 return 0;
1080}
1081
1082static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1083{
1084 int type = CURSEG_HOT_DATA;
1085
25ca923b 1086 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
351df4b2
JK		 1087 /* restore for compacted data summary */
1088 if (read_compacted_summaries(sbi))
1089 return -EINVAL;
1090 type = CURSEG_HOT_NODE;
1091 }
1092
1093 for (; type <= CURSEG_COLD_NODE; type++)
1094 if (read_normal_summaries(sbi, type))
1095 return -EINVAL;
1096 return 0;
1097}
1098
1099static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1100{
1101 struct page *page;
1102 unsigned char *kaddr;
1103 struct f2fs_summary *summary;
1104 struct curseg_info *seg_i;
1105 int written_size = 0;
1106 int i, j;
1107
1108 page = grab_meta_page(sbi, blkaddr++);
1109 kaddr = (unsigned char *)page_address(page);
1110
1111 /* Step 1: write nat cache */
1112 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1113 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1114 written_size += SUM_JOURNAL_SIZE;
1115
1116 /* Step 2: write sit cache */
1117 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1118 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1119 SUM_JOURNAL_SIZE);
1120 written_size += SUM_JOURNAL_SIZE;
1121
1122 set_page_dirty(page);
1123
1124 /* Step 3: write summary entries */
1125 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1126 unsigned short blkoff;
1127 seg_i = CURSEG_I(sbi, i);
1128 if (sbi->ckpt->alloc_type[i] == SSR)
1129 blkoff = sbi->blocks_per_seg;
1130 else
1131 blkoff = curseg_blkoff(sbi, i);
1132
1133 for (j = 0; j < blkoff; j++) {
1134 if (!page) {
1135 page = grab_meta_page(sbi, blkaddr++);
1136 kaddr = (unsigned char *)page_address(page);
1137 written_size = 0;
1138 }
1139 summary = (struct f2fs_summary *)(kaddr + written_size);
1140 *summary = seg_i->sum_blk->entries[j];
1141 written_size += SUMMARY_SIZE;
1142 set_page_dirty(page);
1143
1144 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1145 SUM_FOOTER_SIZE)
1146 continue;
1147
1148 f2fs_put_page(page, 1);
1149 page = NULL;
1150 }
1151 }
1152 if (page)
1153 f2fs_put_page(page, 1);
1154}
1155
1156static void write_normal_summaries(struct f2fs_sb_info *sbi,
1157 block_t blkaddr, int type)
1158{
1159 int i, end;
1160 if (IS_DATASEG(type))
1161 end = type + NR_CURSEG_DATA_TYPE;
1162 else
1163 end = type + NR_CURSEG_NODE_TYPE;
1164
1165 for (i = type; i < end; i++) {
1166 struct curseg_info *sum = CURSEG_I(sbi, i);
1167 mutex_lock(&sum->curseg_mutex);
1168 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1169 mutex_unlock(&sum->curseg_mutex);
1170 }
1171}
1172
1173void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1174{
25ca923b 1175 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1176 write_compacted_summaries(sbi, start_blk);
1177 else
1178 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1179}
1180
1181void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1182{
25ca923b 1183 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
351df4b2
JK		 1184 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1185 return;
1186}
1187
1188int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1189 unsigned int val, int alloc)
1190{
1191 int i;
1192
1193 if (type == NAT_JOURNAL) {
1194 for (i = 0; i < nats_in_cursum(sum); i++) {
1195 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1196 return i;
1197 }
1198 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1199 return update_nats_in_cursum(sum, 1);
1200 } else if (type == SIT_JOURNAL) {
1201 for (i = 0; i < sits_in_cursum(sum); i++)
1202 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1203 return i;
1204 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1205 return update_sits_in_cursum(sum, 1);
1206 }
1207 return -1;
1208}
1209
1210static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1211 unsigned int segno)
1212{
1213 struct sit_info *sit_i = SIT_I(sbi);
1214 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1215 block_t blk_addr = sit_i->sit_base_addr + offset;
1216
1217 check_seg_range(sbi, segno);
1218
1219 /* calculate sit block address */
1220 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1221 blk_addr += sit_i->sit_blocks;
1222
1223 return get_meta_page(sbi, blk_addr);
1224}
1225
1226static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1227 unsigned int start)
1228{
1229 struct sit_info *sit_i = SIT_I(sbi);
1230 struct page *src_page, *dst_page;
1231 pgoff_t src_off, dst_off;
1232 void *src_addr, *dst_addr;
1233
1234 src_off = current_sit_addr(sbi, start);
1235 dst_off = next_sit_addr(sbi, src_off);
1236
1237 /* get current sit block page without lock */
1238 src_page = get_meta_page(sbi, src_off);
1239 dst_page = grab_meta_page(sbi, dst_off);
1240 BUG_ON(PageDirty(src_page));
1241
1242 src_addr = page_address(src_page);
1243 dst_addr = page_address(dst_page);
1244 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1245
1246 set_page_dirty(dst_page);
1247 f2fs_put_page(src_page, 1);
1248
1249 set_to_next_sit(sit_i, start);
1250
1251 return dst_page;
1252}
1253
1254static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1255{
1256 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1257 struct f2fs_summary_block *sum = curseg->sum_blk;
1258 int i;
1259
1260 /*
1261 * If the journal area in the current summary is full of sit entries,
1262 * all the sit entries will be flushed. Otherwise the sit entries
1263 * are not able to replace with newly hot sit entries.
1264 */
1265 if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1266 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1267 unsigned int segno;
1268 segno = le32_to_cpu(segno_in_journal(sum, i));
1269 __mark_sit_entry_dirty(sbi, segno);
1270 }
1271 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1272 return 1;
1273 }
1274 return 0;
1275}
1276
0a8165d7 1277/*
351df4b2
JK		 1278 * CP calls this function, which flushes SIT entries including sit_journal,
1279 * and moves prefree segs to free segs.
1280 */
1281void flush_sit_entries(struct f2fs_sb_info *sbi)
1282{
1283 struct sit_info *sit_i = SIT_I(sbi);
1284 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1285 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1286 struct f2fs_summary_block *sum = curseg->sum_blk;
1287 unsigned long nsegs = TOTAL_SEGS(sbi);
1288 struct page *page = NULL;
1289 struct f2fs_sit_block *raw_sit = NULL;
1290 unsigned int start = 0, end = 0;
1291 unsigned int segno = -1;
1292 bool flushed;
1293
1294 mutex_lock(&curseg->curseg_mutex);
1295 mutex_lock(&sit_i->sentry_lock);
1296
1297 /*
1298 * "flushed" indicates whether sit entries in journal are flushed
1299 * to the SIT area or not.
1300 */
1301 flushed = flush_sits_in_journal(sbi);
1302
1303 while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1304 struct seg_entry *se = get_seg_entry(sbi, segno);
1305 int sit_offset, offset;
1306
1307 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1308
1309 if (flushed)
1310 goto to_sit_page;
1311
1312 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1313 if (offset >= 0) {
1314 segno_in_journal(sum, offset) = cpu_to_le32(segno);
1315 seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1316 goto flush_done;
1317 }
1318to_sit_page:
1319 if (!page || (start > segno) || (segno > end)) {
1320 if (page) {
1321 f2fs_put_page(page, 1);
1322 page = NULL;
1323 }
1324
1325 start = START_SEGNO(sit_i, segno);
1326 end = start + SIT_ENTRY_PER_BLOCK - 1;
1327
1328 /* read sit block that will be updated */
1329 page = get_next_sit_page(sbi, start);
1330 raw_sit = page_address(page);
1331 }
1332
1333 /* udpate entry in SIT block */
1334 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1335flush_done:
1336 __clear_bit(segno, bitmap);
1337 sit_i->dirty_sentries--;
1338 }
1339 mutex_unlock(&sit_i->sentry_lock);
1340 mutex_unlock(&curseg->curseg_mutex);
1341
1342 /* writeout last modified SIT block */
1343 f2fs_put_page(page, 1);
1344
1345 set_prefree_as_free_segments(sbi);
1346}
1347
1348static int build_sit_info(struct f2fs_sb_info *sbi)
1349{
1350 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1351 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1352 struct sit_info *sit_i;
1353 unsigned int sit_segs, start;
1354 char *src_bitmap, *dst_bitmap;
1355 unsigned int bitmap_size;
1356
1357 /* allocate memory for SIT information */
1358 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1359 if (!sit_i)
1360 return -ENOMEM;
1361
1362 SM_I(sbi)->sit_info = sit_i;
1363
1364 sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1365 if (!sit_i->sentries)
1366 return -ENOMEM;
1367
1368 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1369 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1370 if (!sit_i->dirty_sentries_bitmap)
1371 return -ENOMEM;
1372
1373 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1374 sit_i->sentries[start].cur_valid_map
1375 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1376 sit_i->sentries[start].ckpt_valid_map
1377 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1378 if (!sit_i->sentries[start].cur_valid_map
1379 || !sit_i->sentries[start].ckpt_valid_map)
1380 return -ENOMEM;
1381 }
1382
1383 if (sbi->segs_per_sec > 1) {
53cf9522 1384 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
351df4b2
JK		 1385 sizeof(struct sec_entry));
1386 if (!sit_i->sec_entries)
1387 return -ENOMEM;
1388 }
1389
1390 /* get information related with SIT */
1391 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1392
1393 /* setup SIT bitmap from ckeckpoint pack */
1394 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1395 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1396
79b5793b 1397 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 1398 if (!dst_bitmap)
1399 return -ENOMEM;
351df4b2
JK		 1400
1401 /* init SIT information */
1402 sit_i->s_ops = &default_salloc_ops;
1403
1404 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1405 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1406 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1407 sit_i->sit_bitmap = dst_bitmap;
1408 sit_i->bitmap_size = bitmap_size;
1409 sit_i->dirty_sentries = 0;
1410 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1411 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1412 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1413 mutex_init(&sit_i->sentry_lock);
1414 return 0;
1415}
1416
1417static int build_free_segmap(struct f2fs_sb_info *sbi)
1418{
1419 struct f2fs_sm_info *sm_info = SM_I(sbi);
1420 struct free_segmap_info *free_i;
1421 unsigned int bitmap_size, sec_bitmap_size;
1422
1423 /* allocate memory for free segmap information */
1424 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1425 if (!free_i)
1426 return -ENOMEM;
1427
1428 SM_I(sbi)->free_info = free_i;
1429
1430 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1431 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1432 if (!free_i->free_segmap)
1433 return -ENOMEM;
1434
53cf9522 1435 sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
351df4b2
JK		 1436 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1437 if (!free_i->free_secmap)
1438 return -ENOMEM;
1439
1440 /* set all segments as dirty temporarily */
1441 memset(free_i->free_segmap, 0xff, bitmap_size);
1442 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1443
1444 /* init free segmap information */
1445 free_i->start_segno =
1446 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1447 free_i->free_segments = 0;
1448 free_i->free_sections = 0;
1449 rwlock_init(&free_i->segmap_lock);
1450 return 0;
1451}
1452
1453static int build_curseg(struct f2fs_sb_info *sbi)
1454{
1042d60f 1455 struct curseg_info *array;
351df4b2
JK		 1456 int i;
1457
1458 array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1459 if (!array)
1460 return -ENOMEM;
1461
1462 SM_I(sbi)->curseg_array = array;
1463
1464 for (i = 0; i < NR_CURSEG_TYPE; i++) {
1465 mutex_init(&array[i].curseg_mutex);
1466 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1467 if (!array[i].sum_blk)
1468 return -ENOMEM;
1469 array[i].segno = NULL_SEGNO;
1470 array[i].next_blkoff = 0;
1471 }
1472 return restore_curseg_summaries(sbi);
1473}
1474
1475static void build_sit_entries(struct f2fs_sb_info *sbi)
1476{
1477 struct sit_info *sit_i = SIT_I(sbi);
1478 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1479 struct f2fs_summary_block *sum = curseg->sum_blk;
1480 unsigned int start;
1481
1482 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1483 struct seg_entry *se = &sit_i->sentries[start];
1484 struct f2fs_sit_block *sit_blk;
1485 struct f2fs_sit_entry sit;
1486 struct page *page;
1487 int i;
1488
1489 mutex_lock(&curseg->curseg_mutex);
1490 for (i = 0; i < sits_in_cursum(sum); i++) {
1491 if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1492 sit = sit_in_journal(sum, i);
1493 mutex_unlock(&curseg->curseg_mutex);
1494 goto got_it;
1495 }
1496 }
1497 mutex_unlock(&curseg->curseg_mutex);
1498 page = get_current_sit_page(sbi, start);
1499 sit_blk = (struct f2fs_sit_block *)page_address(page);
1500 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1501 f2fs_put_page(page, 1);
1502got_it:
1503 check_block_count(sbi, start, &sit);
1504 seg_info_from_raw_sit(se, &sit);
1505 if (sbi->segs_per_sec > 1) {
1506 struct sec_entry *e = get_sec_entry(sbi, start);
1507 e->valid_blocks += se->valid_blocks;
1508 }
1509 }
1510}
1511
1512static void init_free_segmap(struct f2fs_sb_info *sbi)
1513{
1514 unsigned int start;
1515 int type;
1516
1517 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1518 struct seg_entry *sentry = get_seg_entry(sbi, start);
1519 if (!sentry->valid_blocks)
1520 __set_free(sbi, start);
1521 }
1522
1523 /* set use the current segments */
1524 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1525 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1526 __set_test_and_inuse(sbi, curseg_t->segno);
1527 }
1528}
1529
1530static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1531{
1532 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1533 struct free_segmap_info *free_i = FREE_I(sbi);
8736fbf0 1534 unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
351df4b2
JK		 1535 unsigned short valid_blocks;
1536
8736fbf0 1537 while (1) {
351df4b2 1538 /* find dirty segment based on free segmap */
8736fbf0
NJ		 1539 segno = find_next_inuse(free_i, total_segs, offset);
1540 if (segno >= total_segs)
351df4b2
JK		 1541 break;
1542 offset = segno + 1;
1543 valid_blocks = get_valid_blocks(sbi, segno, 0);
1544 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1545 continue;
1546 mutex_lock(&dirty_i->seglist_lock);
1547 __locate_dirty_segment(sbi, segno, DIRTY);
1548 mutex_unlock(&dirty_i->seglist_lock);
1549 }
1550}
1551
5ec4e49f 1552static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 1553{
1554 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5ec4e49f 1555 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
351df4b2 1556
5ec4e49f
JK		 1557 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1558 if (!dirty_i->victim_secmap)
351df4b2
JK		 1559 return -ENOMEM;
1560 return 0;
1561}
1562
1563static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1564{
1565 struct dirty_seglist_info *dirty_i;
1566 unsigned int bitmap_size, i;
1567
1568 /* allocate memory for dirty segments list information */
1569 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1570 if (!dirty_i)
1571 return -ENOMEM;
1572
1573 SM_I(sbi)->dirty_info = dirty_i;
1574 mutex_init(&dirty_i->seglist_lock);
1575
1576 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1577
1578 for (i = 0; i < NR_DIRTY_TYPE; i++) {
1579 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 1580 if (!dirty_i->dirty_segmap[i])
1581 return -ENOMEM;
1582 }
1583
1584 init_dirty_segmap(sbi);
5ec4e49f 1585 return init_victim_secmap(sbi);
351df4b2
JK		 1586}
1587
0a8165d7 1588/*
351df4b2
JK		 1589 * Update min, max modified time for cost-benefit GC algorithm
1590 */
1591static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1592{
1593 struct sit_info *sit_i = SIT_I(sbi);
1594 unsigned int segno;
1595
1596 mutex_lock(&sit_i->sentry_lock);
1597
1598 sit_i->min_mtime = LLONG_MAX;
1599
1600 for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1601 unsigned int i;
1602 unsigned long long mtime = 0;
1603
1604 for (i = 0; i < sbi->segs_per_sec; i++)
1605 mtime += get_seg_entry(sbi, segno + i)->mtime;
1606
1607 mtime = div_u64(mtime, sbi->segs_per_sec);
1608
1609 if (sit_i->min_mtime > mtime)
1610 sit_i->min_mtime = mtime;
1611 }
1612 sit_i->max_mtime = get_mtime(sbi);
1613 mutex_unlock(&sit_i->sentry_lock);
1614}
1615
1616int build_segment_manager(struct f2fs_sb_info *sbi)
1617{
1618 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1619 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 1620 struct f2fs_sm_info *sm_info;
351df4b2
JK		 1621 int err;
1622
1623 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1624 if (!sm_info)
1625 return -ENOMEM;
1626
1627 /* init sm info */
1628 sbi->sm_info = sm_info;
1629 INIT_LIST_HEAD(&sm_info->wblist_head);
1630 spin_lock_init(&sm_info->wblist_lock);
1631 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1632 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1633 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1634 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1635 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1636 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1637 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1638
1639 err = build_sit_info(sbi);
1640 if (err)
1641 return err;
1642 err = build_free_segmap(sbi);
1643 if (err)
1644 return err;
1645 err = build_curseg(sbi);
1646 if (err)
1647 return err;
1648
1649 /* reinit free segmap based on SIT */
1650 build_sit_entries(sbi);
1651
1652 init_free_segmap(sbi);
1653 err = build_dirty_segmap(sbi);
1654 if (err)
1655 return err;
1656
1657 init_min_max_mtime(sbi);
1658 return 0;
1659}
1660
1661static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1662 enum dirty_type dirty_type)
1663{
1664 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1665
1666 mutex_lock(&dirty_i->seglist_lock);
1667 kfree(dirty_i->dirty_segmap[dirty_type]);
1668 dirty_i->nr_dirty[dirty_type] = 0;
1669 mutex_unlock(&dirty_i->seglist_lock);
1670}
1671
5ec4e49f 1672static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 1673{
1674 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5ec4e49f 1675 kfree(dirty_i->victim_secmap);
351df4b2
JK		 1676}
1677
1678static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1679{
1680 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1681 int i;
1682
1683 if (!dirty_i)
1684 return;
1685
1686 /* discard pre-free/dirty segments list */
1687 for (i = 0; i < NR_DIRTY_TYPE; i++)
1688 discard_dirty_segmap(sbi, i);
1689
5ec4e49f 1690 destroy_victim_secmap(sbi);
351df4b2
JK		 1691 SM_I(sbi)->dirty_info = NULL;
1692 kfree(dirty_i);
1693}
1694
1695static void destroy_curseg(struct f2fs_sb_info *sbi)
1696{
1697 struct curseg_info *array = SM_I(sbi)->curseg_array;
1698 int i;
1699
1700 if (!array)
1701 return;
1702 SM_I(sbi)->curseg_array = NULL;
1703 for (i = 0; i < NR_CURSEG_TYPE; i++)
1704 kfree(array[i].sum_blk);
1705 kfree(array);
1706}
1707
1708static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1709{
1710 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1711 if (!free_i)
1712 return;
1713 SM_I(sbi)->free_info = NULL;
1714 kfree(free_i->free_segmap);
1715 kfree(free_i->free_secmap);
1716 kfree(free_i);
1717}
1718
1719static void destroy_sit_info(struct f2fs_sb_info *sbi)
1720{
1721 struct sit_info *sit_i = SIT_I(sbi);
1722 unsigned int start;
1723
1724 if (!sit_i)
1725 return;
1726
1727 if (sit_i->sentries) {
1728 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1729 kfree(sit_i->sentries[start].cur_valid_map);
1730 kfree(sit_i->sentries[start].ckpt_valid_map);
1731 }
1732 }
1733 vfree(sit_i->sentries);
1734 vfree(sit_i->sec_entries);
1735 kfree(sit_i->dirty_sentries_bitmap);
1736
1737 SM_I(sbi)->sit_info = NULL;
1738 kfree(sit_i->sit_bitmap);
1739 kfree(sit_i);
1740}
1741
1742void destroy_segment_manager(struct f2fs_sb_info *sbi)
1743{
1744 struct f2fs_sm_info *sm_info = SM_I(sbi);
1745 destroy_dirty_segmap(sbi);
1746 destroy_curseg(sbi);
1747 destroy_free_segmap(sbi);
1748 destroy_sit_info(sbi);
1749 sbi->sm_info = NULL;
1750 kfree(sm_info);
1751}
Linux kernel - Deliverables
This page took 0.138704 seconds and 5 git commands to generate.