blob: e87946a08a21988d6120389fd3e887abf91441db [file] [log] [blame]
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +09001/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09002 * 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>
Geert Uytterhoeven690e4a32012-12-19 22:19:30 +010015#include <linux/prefetch.h>
Jaegeuk Kim351df4b2012-11-02 17:09:16 +090016#include <linux/vmalloc.h>
Chao Yu74de5932013-11-22 09:09:59 +080017#include <linux/swap.h>
Jaegeuk Kim351df4b2012-11-02 17:09:16 +090018
19#include "f2fs.h"
20#include "segment.h"
21#include "node.h"
Namjae Jeon6ec178d2013-04-23 17:51:43 +090022#include <trace/events/f2fs.h>
Jaegeuk Kim351df4b2012-11-02 17:09:16 +090023
Changman Lee9a7f1432013-11-15 10:42:51 +090024#define __reverse_ffz(x) __reverse_ffs(~(x))
25
Jaegeuk Kim7fd9e542013-11-15 13:55:58 +090026static struct kmem_cache *discard_entry_slab;
27
Changman Lee9a7f1432013-11-15 10:42:51 +090028/*
29 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
30 * MSB and LSB are reversed in a byte by f2fs_set_bit.
31 */
32static inline unsigned long __reverse_ffs(unsigned long word)
33{
34 int num = 0;
35
36#if BITS_PER_LONG == 64
37 if ((word & 0xffffffff) == 0) {
38 num += 32;
39 word >>= 32;
40 }
41#endif
42 if ((word & 0xffff) == 0) {
43 num += 16;
44 word >>= 16;
45 }
46 if ((word & 0xff) == 0) {
47 num += 8;
48 word >>= 8;
49 }
50 if ((word & 0xf0) == 0)
51 num += 4;
52 else
53 word >>= 4;
54 if ((word & 0xc) == 0)
55 num += 2;
56 else
57 word >>= 2;
58 if ((word & 0x2) == 0)
59 num += 1;
60 return num;
61}
62
63/*
64 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c becasue
65 * f2fs_set_bit makes MSB and LSB reversed in a byte.
66 * Example:
67 * LSB <--> MSB
68 * f2fs_set_bit(0, bitmap) => 0000 0001
69 * f2fs_set_bit(7, bitmap) => 1000 0000
70 */
71static unsigned long __find_rev_next_bit(const unsigned long *addr,
72 unsigned long size, unsigned long offset)
73{
74 const unsigned long *p = addr + BIT_WORD(offset);
75 unsigned long result = offset & ~(BITS_PER_LONG - 1);
76 unsigned long tmp;
77 unsigned long mask, submask;
78 unsigned long quot, rest;
79
80 if (offset >= size)
81 return size;
82
83 size -= result;
84 offset %= BITS_PER_LONG;
85 if (!offset)
86 goto aligned;
87
88 tmp = *(p++);
89 quot = (offset >> 3) << 3;
90 rest = offset & 0x7;
91 mask = ~0UL << quot;
92 submask = (unsigned char)(0xff << rest) >> rest;
93 submask <<= quot;
94 mask &= submask;
95 tmp &= mask;
96 if (size < BITS_PER_LONG)
97 goto found_first;
98 if (tmp)
99 goto found_middle;
100
101 size -= BITS_PER_LONG;
102 result += BITS_PER_LONG;
103aligned:
104 while (size & ~(BITS_PER_LONG-1)) {
105 tmp = *(p++);
106 if (tmp)
107 goto found_middle;
108 result += BITS_PER_LONG;
109 size -= BITS_PER_LONG;
110 }
111 if (!size)
112 return result;
113 tmp = *p;
114found_first:
115 tmp &= (~0UL >> (BITS_PER_LONG - size));
116 if (tmp == 0UL) /* Are any bits set? */
117 return result + size; /* Nope. */
118found_middle:
119 return result + __reverse_ffs(tmp);
120}
121
122static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
123 unsigned long size, unsigned long offset)
124{
125 const unsigned long *p = addr + BIT_WORD(offset);
126 unsigned long result = offset & ~(BITS_PER_LONG - 1);
127 unsigned long tmp;
128 unsigned long mask, submask;
129 unsigned long quot, rest;
130
131 if (offset >= size)
132 return size;
133
134 size -= result;
135 offset %= BITS_PER_LONG;
136 if (!offset)
137 goto aligned;
138
139 tmp = *(p++);
140 quot = (offset >> 3) << 3;
141 rest = offset & 0x7;
142 mask = ~(~0UL << quot);
143 submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
144 submask <<= quot;
145 mask += submask;
146 tmp |= mask;
147 if (size < BITS_PER_LONG)
148 goto found_first;
149 if (~tmp)
150 goto found_middle;
151
152 size -= BITS_PER_LONG;
153 result += BITS_PER_LONG;
154aligned:
155 while (size & ~(BITS_PER_LONG - 1)) {
156 tmp = *(p++);
157 if (~tmp)
158 goto found_middle;
159 result += BITS_PER_LONG;
160 size -= BITS_PER_LONG;
161 }
162 if (!size)
163 return result;
164 tmp = *p;
165
166found_first:
167 tmp |= ~0UL << size;
168 if (tmp == ~0UL) /* Are any bits zero? */
169 return result + size; /* Nope. */
170found_middle:
171 return result + __reverse_ffz(tmp);
172}
173
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900174/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900175 * This function balances dirty node and dentry pages.
176 * In addition, it controls garbage collection.
177 */
178void f2fs_balance_fs(struct f2fs_sb_info *sbi)
179{
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900180 /*
Jaegeuk Kim029cd282012-12-21 17:20:21 +0900181 * We should do GC or end up with checkpoint, if there are so many dirty
182 * dir/node pages without enough free segments.
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900183 */
Jaegeuk Kim43727522013-02-04 15:11:17 +0900184 if (has_not_enough_free_secs(sbi, 0)) {
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900185 mutex_lock(&sbi->gc_mutex);
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900186 f2fs_gc(sbi);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900187 }
188}
189
Jaegeuk Kim4660f9c2013-10-24 14:19:18 +0900190void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
191{
192 /* check the # of cached NAT entries and prefree segments */
193 if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
194 excess_prefree_segs(sbi))
195 f2fs_sync_fs(sbi->sb, true);
196}
197
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900198static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
199 enum dirty_type dirty_type)
200{
201 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
202
203 /* need not be added */
204 if (IS_CURSEG(sbi, segno))
205 return;
206
207 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
208 dirty_i->nr_dirty[dirty_type]++;
209
210 if (dirty_type == DIRTY) {
211 struct seg_entry *sentry = get_seg_entry(sbi, segno);
Changman Lee4625d6a2013-10-25 17:31:57 +0900212 enum dirty_type t = sentry->type;
Jaegeuk Kimb2f2c392013-04-01 13:52:09 +0900213
Changman Lee4625d6a2013-10-25 17:31:57 +0900214 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
215 dirty_i->nr_dirty[t]++;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900216 }
217}
218
219static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
220 enum dirty_type dirty_type)
221{
222 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
223
224 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
225 dirty_i->nr_dirty[dirty_type]--;
226
227 if (dirty_type == DIRTY) {
Changman Lee4625d6a2013-10-25 17:31:57 +0900228 struct seg_entry *sentry = get_seg_entry(sbi, segno);
229 enum dirty_type t = sentry->type;
Jaegeuk Kimb2f2c392013-04-01 13:52:09 +0900230
Changman Lee4625d6a2013-10-25 17:31:57 +0900231 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
232 dirty_i->nr_dirty[t]--;
Jaegeuk Kimb2f2c392013-04-01 13:52:09 +0900233
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900234 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
235 clear_bit(GET_SECNO(sbi, segno),
236 dirty_i->victim_secmap);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900237 }
238}
239
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900240/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900241 * Should not occur error such as -ENOMEM.
242 * Adding dirty entry into seglist is not critical operation.
243 * If a given segment is one of current working segments, it won't be added.
244 */
Haicheng Li8d8451a2013-06-13 16:59:28 +0800245static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900246{
247 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
248 unsigned short valid_blocks;
249
250 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
251 return;
252
253 mutex_lock(&dirty_i->seglist_lock);
254
255 valid_blocks = get_valid_blocks(sbi, segno, 0);
256
257 if (valid_blocks == 0) {
258 __locate_dirty_segment(sbi, segno, PRE);
259 __remove_dirty_segment(sbi, segno, DIRTY);
260 } else if (valid_blocks < sbi->blocks_per_seg) {
261 __locate_dirty_segment(sbi, segno, DIRTY);
262 } else {
263 /* Recovery routine with SSR needs this */
264 __remove_dirty_segment(sbi, segno, DIRTY);
265 }
266
267 mutex_unlock(&dirty_i->seglist_lock);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900268}
269
Jaegeuk Kim37208872013-11-12 16:55:17 +0900270static void f2fs_issue_discard(struct f2fs_sb_info *sbi,
271 block_t blkstart, block_t blklen)
272{
Jaegeuk Kimf9a4e6d2013-11-28 12:44:05 +0900273 sector_t start = SECTOR_FROM_BLOCK(sbi, blkstart);
274 sector_t len = SECTOR_FROM_BLOCK(sbi, blklen);
Jaegeuk Kim37208872013-11-12 16:55:17 +0900275 blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
Jaegeuk Kim1661d072013-11-12 17:01:00 +0900276 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
Jaegeuk Kim37208872013-11-12 16:55:17 +0900277}
278
Jaegeuk Kimb2955552013-11-12 14:49:56 +0900279static void add_discard_addrs(struct f2fs_sb_info *sbi,
280 unsigned int segno, struct seg_entry *se)
281{
282 struct list_head *head = &SM_I(sbi)->discard_list;
283 struct discard_entry *new;
284 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
285 int max_blocks = sbi->blocks_per_seg;
286 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
287 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
288 unsigned long dmap[entries];
289 unsigned int start = 0, end = -1;
290 int i;
291
292 if (!test_opt(sbi, DISCARD))
293 return;
294
295 /* zero block will be discarded through the prefree list */
296 if (!se->valid_blocks || se->valid_blocks == max_blocks)
297 return;
298
299 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
300 for (i = 0; i < entries; i++)
301 dmap[i] = (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
302
303 while (SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
304 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
305 if (start >= max_blocks)
306 break;
307
308 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
309
310 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
311 INIT_LIST_HEAD(&new->list);
312 new->blkaddr = START_BLOCK(sbi, segno) + start;
313 new->len = end - start;
314
315 list_add_tail(&new->list, head);
316 SM_I(sbi)->nr_discards += end - start;
317 }
318}
319
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900320/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900321 * Should call clear_prefree_segments after checkpoint is done.
322 */
323static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
324{
325 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
Haicheng Li81fb5e82013-05-14 18:20:28 +0800326 unsigned int segno = -1;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900327 unsigned int total_segs = TOTAL_SEGS(sbi);
328
329 mutex_lock(&dirty_i->seglist_lock);
330 while (1) {
331 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
Haicheng Li81fb5e82013-05-14 18:20:28 +0800332 segno + 1);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900333 if (segno >= total_segs)
334 break;
335 __set_test_and_free(sbi, segno);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900336 }
337 mutex_unlock(&dirty_i->seglist_lock);
338}
339
340void clear_prefree_segments(struct f2fs_sb_info *sbi)
341{
Jaegeuk Kimb2955552013-11-12 14:49:56 +0900342 struct list_head *head = &(SM_I(sbi)->discard_list);
343 struct list_head *this, *next;
344 struct discard_entry *entry;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900345 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
Changman Lee29e59c12013-11-11 09:24:37 +0900346 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900347 unsigned int total_segs = TOTAL_SEGS(sbi);
Changman Lee29e59c12013-11-11 09:24:37 +0900348 unsigned int start = 0, end = -1;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900349
350 mutex_lock(&dirty_i->seglist_lock);
Changman Lee29e59c12013-11-11 09:24:37 +0900351
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900352 while (1) {
Changman Lee29e59c12013-11-11 09:24:37 +0900353 int i;
354 start = find_next_bit(prefree_map, total_segs, end + 1);
355 if (start >= total_segs)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900356 break;
Changman Lee29e59c12013-11-11 09:24:37 +0900357 end = find_next_zero_bit(prefree_map, total_segs, start + 1);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900358
Changman Lee29e59c12013-11-11 09:24:37 +0900359 for (i = start; i < end; i++)
360 clear_bit(i, prefree_map);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900361
Changman Lee29e59c12013-11-11 09:24:37 +0900362 dirty_i->nr_dirty[PRE] -= end - start;
363
364 if (!test_opt(sbi, DISCARD))
365 continue;
366
Jaegeuk Kim37208872013-11-12 16:55:17 +0900367 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
368 (end - start) << sbi->log_blocks_per_seg);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900369 }
370 mutex_unlock(&dirty_i->seglist_lock);
Jaegeuk Kimb2955552013-11-12 14:49:56 +0900371
372 /* send small discards */
373 list_for_each_safe(this, next, head) {
374 entry = list_entry(this, struct discard_entry, list);
Jaegeuk Kim37208872013-11-12 16:55:17 +0900375 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
Jaegeuk Kimb2955552013-11-12 14:49:56 +0900376 list_del(&entry->list);
377 SM_I(sbi)->nr_discards -= entry->len;
378 kmem_cache_free(discard_entry_slab, entry);
379 }
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900380}
381
382static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
383{
384 struct sit_info *sit_i = SIT_I(sbi);
385 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
386 sit_i->dirty_sentries++;
387}
388
389static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
390 unsigned int segno, int modified)
391{
392 struct seg_entry *se = get_seg_entry(sbi, segno);
393 se->type = type;
394 if (modified)
395 __mark_sit_entry_dirty(sbi, segno);
396}
397
398static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
399{
400 struct seg_entry *se;
401 unsigned int segno, offset;
402 long int new_vblocks;
403
404 segno = GET_SEGNO(sbi, blkaddr);
405
406 se = get_seg_entry(sbi, segno);
407 new_vblocks = se->valid_blocks + del;
Jaegeuk Kim491c0852014-02-04 13:01:10 +0900408 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900409
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900410 f2fs_bug_on((new_vblocks >> (sizeof(unsigned short) << 3) ||
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900411 (new_vblocks > sbi->blocks_per_seg)));
412
413 se->valid_blocks = new_vblocks;
414 se->mtime = get_mtime(sbi);
415 SIT_I(sbi)->max_mtime = se->mtime;
416
417 /* Update valid block bitmap */
418 if (del > 0) {
419 if (f2fs_set_bit(offset, se->cur_valid_map))
420 BUG();
421 } else {
422 if (!f2fs_clear_bit(offset, se->cur_valid_map))
423 BUG();
424 }
425 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
426 se->ckpt_valid_blocks += del;
427
428 __mark_sit_entry_dirty(sbi, segno);
429
430 /* update total number of valid blocks to be written in ckpt area */
431 SIT_I(sbi)->written_valid_blocks += del;
432
433 if (sbi->segs_per_sec > 1)
434 get_sec_entry(sbi, segno)->valid_blocks += del;
435}
436
Jaegeuk Kim5e443812014-01-28 12:22:14 +0900437void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900438{
Jaegeuk Kim5e443812014-01-28 12:22:14 +0900439 update_sit_entry(sbi, new, 1);
440 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
441 update_sit_entry(sbi, old, -1);
442
443 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
444 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900445}
446
447void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
448{
449 unsigned int segno = GET_SEGNO(sbi, addr);
450 struct sit_info *sit_i = SIT_I(sbi);
451
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900452 f2fs_bug_on(addr == NULL_ADDR);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900453 if (addr == NEW_ADDR)
454 return;
455
456 /* add it into sit main buffer */
457 mutex_lock(&sit_i->sentry_lock);
458
459 update_sit_entry(sbi, addr, -1);
460
461 /* add it into dirty seglist */
462 locate_dirty_segment(sbi, segno);
463
464 mutex_unlock(&sit_i->sentry_lock);
465}
466
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900467/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900468 * This function should be resided under the curseg_mutex lock
469 */
470static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
Haicheng Lie79efe32013-06-13 16:59:27 +0800471 struct f2fs_summary *sum)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900472{
473 struct curseg_info *curseg = CURSEG_I(sbi, type);
474 void *addr = curseg->sum_blk;
Haicheng Lie79efe32013-06-13 16:59:27 +0800475 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900476 memcpy(addr, sum, sizeof(struct f2fs_summary));
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900477}
478
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900479/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900480 * Calculate the number of current summary pages for writing
481 */
482int npages_for_summary_flush(struct f2fs_sb_info *sbi)
483{
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900484 int valid_sum_count = 0;
Fan Li9a479382013-10-29 16:21:47 +0800485 int i, sum_in_page;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900486
487 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
488 if (sbi->ckpt->alloc_type[i] == SSR)
489 valid_sum_count += sbi->blocks_per_seg;
490 else
491 valid_sum_count += curseg_blkoff(sbi, i);
492 }
493
Fan Li9a479382013-10-29 16:21:47 +0800494 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
495 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
496 if (valid_sum_count <= sum_in_page)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900497 return 1;
Fan Li9a479382013-10-29 16:21:47 +0800498 else if ((valid_sum_count - sum_in_page) <=
499 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900500 return 2;
501 return 3;
502}
503
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900504/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900505 * Caller should put this summary page
506 */
507struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
508{
509 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
510}
511
512static void write_sum_page(struct f2fs_sb_info *sbi,
513 struct f2fs_summary_block *sum_blk, block_t blk_addr)
514{
515 struct page *page = grab_meta_page(sbi, blk_addr);
516 void *kaddr = page_address(page);
517 memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
518 set_page_dirty(page);
519 f2fs_put_page(page, 1);
520}
521
Jaegeuk Kim60374682013-03-31 13:58:51 +0900522static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
523{
524 struct curseg_info *curseg = CURSEG_I(sbi, type);
Haicheng Li81fb5e82013-05-14 18:20:28 +0800525 unsigned int segno = curseg->segno + 1;
Jaegeuk Kim60374682013-03-31 13:58:51 +0900526 struct free_segmap_info *free_i = FREE_I(sbi);
527
Haicheng Li81fb5e82013-05-14 18:20:28 +0800528 if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
529 return !test_bit(segno, free_i->free_segmap);
Jaegeuk Kim60374682013-03-31 13:58:51 +0900530 return 0;
531}
532
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900533/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900534 * Find a new segment from the free segments bitmap to right order
535 * This function should be returned with success, otherwise BUG
536 */
537static void get_new_segment(struct f2fs_sb_info *sbi,
538 unsigned int *newseg, bool new_sec, int dir)
539{
540 struct free_segmap_info *free_i = FREE_I(sbi);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900541 unsigned int segno, secno, zoneno;
Jaegeuk Kim53cf9522013-03-31 12:39:49 +0900542 unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900543 unsigned int hint = *newseg / sbi->segs_per_sec;
544 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
545 unsigned int left_start = hint;
546 bool init = true;
547 int go_left = 0;
548 int i;
549
550 write_lock(&free_i->segmap_lock);
551
552 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
553 segno = find_next_zero_bit(free_i->free_segmap,
554 TOTAL_SEGS(sbi), *newseg + 1);
Jaegeuk Kim33afa7f2013-03-31 12:59:53 +0900555 if (segno - *newseg < sbi->segs_per_sec -
556 (*newseg % sbi->segs_per_sec))
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900557 goto got_it;
558 }
559find_other_zone:
Jaegeuk Kim53cf9522013-03-31 12:39:49 +0900560 secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
561 if (secno >= TOTAL_SECS(sbi)) {
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900562 if (dir == ALLOC_RIGHT) {
563 secno = find_next_zero_bit(free_i->free_secmap,
Jaegeuk Kim53cf9522013-03-31 12:39:49 +0900564 TOTAL_SECS(sbi), 0);
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900565 f2fs_bug_on(secno >= TOTAL_SECS(sbi));
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900566 } else {
567 go_left = 1;
568 left_start = hint - 1;
569 }
570 }
571 if (go_left == 0)
572 goto skip_left;
573
574 while (test_bit(left_start, free_i->free_secmap)) {
575 if (left_start > 0) {
576 left_start--;
577 continue;
578 }
579 left_start = find_next_zero_bit(free_i->free_secmap,
Jaegeuk Kim53cf9522013-03-31 12:39:49 +0900580 TOTAL_SECS(sbi), 0);
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900581 f2fs_bug_on(left_start >= TOTAL_SECS(sbi));
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900582 break;
583 }
584 secno = left_start;
585skip_left:
586 hint = secno;
587 segno = secno * sbi->segs_per_sec;
588 zoneno = secno / sbi->secs_per_zone;
589
590 /* give up on finding another zone */
591 if (!init)
592 goto got_it;
593 if (sbi->secs_per_zone == 1)
594 goto got_it;
595 if (zoneno == old_zoneno)
596 goto got_it;
597 if (dir == ALLOC_LEFT) {
598 if (!go_left && zoneno + 1 >= total_zones)
599 goto got_it;
600 if (go_left && zoneno == 0)
601 goto got_it;
602 }
603 for (i = 0; i < NR_CURSEG_TYPE; i++)
604 if (CURSEG_I(sbi, i)->zone == zoneno)
605 break;
606
607 if (i < NR_CURSEG_TYPE) {
608 /* zone is in user, try another */
609 if (go_left)
610 hint = zoneno * sbi->secs_per_zone - 1;
611 else if (zoneno + 1 >= total_zones)
612 hint = 0;
613 else
614 hint = (zoneno + 1) * sbi->secs_per_zone;
615 init = false;
616 goto find_other_zone;
617 }
618got_it:
619 /* set it as dirty segment in free segmap */
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900620 f2fs_bug_on(test_bit(segno, free_i->free_segmap));
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900621 __set_inuse(sbi, segno);
622 *newseg = segno;
623 write_unlock(&free_i->segmap_lock);
624}
625
626static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
627{
628 struct curseg_info *curseg = CURSEG_I(sbi, type);
629 struct summary_footer *sum_footer;
630
631 curseg->segno = curseg->next_segno;
632 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
633 curseg->next_blkoff = 0;
634 curseg->next_segno = NULL_SEGNO;
635
636 sum_footer = &(curseg->sum_blk->footer);
637 memset(sum_footer, 0, sizeof(struct summary_footer));
638 if (IS_DATASEG(type))
639 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
640 if (IS_NODESEG(type))
641 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
642 __set_sit_entry_type(sbi, type, curseg->segno, modified);
643}
644
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900645/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900646 * Allocate a current working segment.
647 * This function always allocates a free segment in LFS manner.
648 */
649static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
650{
651 struct curseg_info *curseg = CURSEG_I(sbi, type);
652 unsigned int segno = curseg->segno;
653 int dir = ALLOC_LEFT;
654
655 write_sum_page(sbi, curseg->sum_blk,
Haicheng Li81fb5e82013-05-14 18:20:28 +0800656 GET_SUM_BLOCK(sbi, segno));
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900657 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
658 dir = ALLOC_RIGHT;
659
660 if (test_opt(sbi, NOHEAP))
661 dir = ALLOC_RIGHT;
662
663 get_new_segment(sbi, &segno, new_sec, dir);
664 curseg->next_segno = segno;
665 reset_curseg(sbi, type, 1);
666 curseg->alloc_type = LFS;
667}
668
669static void __next_free_blkoff(struct f2fs_sb_info *sbi,
670 struct curseg_info *seg, block_t start)
671{
672 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
Changman Leee81c93c2013-11-15 13:21:16 +0900673 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
674 unsigned long target_map[entries];
675 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
676 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
677 int i, pos;
678
679 for (i = 0; i < entries; i++)
680 target_map[i] = ckpt_map[i] | cur_map[i];
681
682 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
683
684 seg->next_blkoff = pos;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900685}
686
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900687/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900688 * If a segment is written by LFS manner, next block offset is just obtained
689 * by increasing the current block offset. However, if a segment is written by
690 * SSR manner, next block offset obtained by calling __next_free_blkoff
691 */
692static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
693 struct curseg_info *seg)
694{
695 if (seg->alloc_type == SSR)
696 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
697 else
698 seg->next_blkoff++;
699}
700
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900701/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900702 * This function always allocates a used segment (from dirty seglist) by SSR
703 * manner, so it should recover the existing segment information of valid blocks
704 */
705static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
706{
707 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
708 struct curseg_info *curseg = CURSEG_I(sbi, type);
709 unsigned int new_segno = curseg->next_segno;
710 struct f2fs_summary_block *sum_node;
711 struct page *sum_page;
712
713 write_sum_page(sbi, curseg->sum_blk,
714 GET_SUM_BLOCK(sbi, curseg->segno));
715 __set_test_and_inuse(sbi, new_segno);
716
717 mutex_lock(&dirty_i->seglist_lock);
718 __remove_dirty_segment(sbi, new_segno, PRE);
719 __remove_dirty_segment(sbi, new_segno, DIRTY);
720 mutex_unlock(&dirty_i->seglist_lock);
721
722 reset_curseg(sbi, type, 1);
723 curseg->alloc_type = SSR;
724 __next_free_blkoff(sbi, curseg, 0);
725
726 if (reuse) {
727 sum_page = get_sum_page(sbi, new_segno);
728 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
729 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
730 f2fs_put_page(sum_page, 1);
731 }
732}
733
Jaegeuk Kim43727522013-02-04 15:11:17 +0900734static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
735{
736 struct curseg_info *curseg = CURSEG_I(sbi, type);
737 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
738
739 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
740 return v_ops->get_victim(sbi,
741 &(curseg)->next_segno, BG_GC, type, SSR);
742
743 /* For data segments, let's do SSR more intensively */
744 for (; type >= CURSEG_HOT_DATA; type--)
745 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
746 BG_GC, type, SSR))
747 return 1;
748 return 0;
749}
750
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900751/*
752 * flush out current segment and replace it with new segment
753 * This function should be returned with success, otherwise BUG
754 */
755static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
756 int type, bool force)
757{
758 struct curseg_info *curseg = CURSEG_I(sbi, type);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900759
Gu Zheng7b405272013-08-19 09:41:15 +0800760 if (force)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900761 new_curseg(sbi, type, true);
Gu Zheng7b405272013-08-19 09:41:15 +0800762 else if (type == CURSEG_WARM_NODE)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900763 new_curseg(sbi, type, false);
Jaegeuk Kim60374682013-03-31 13:58:51 +0900764 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
765 new_curseg(sbi, type, false);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900766 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
767 change_curseg(sbi, type, true);
768 else
769 new_curseg(sbi, type, false);
Jaegeuk Kimdcdfff62013-10-22 20:56:10 +0900770
771 stat_inc_seg_type(sbi, curseg);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900772}
773
774void allocate_new_segments(struct f2fs_sb_info *sbi)
775{
776 struct curseg_info *curseg;
777 unsigned int old_curseg;
778 int i;
779
780 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
781 curseg = CURSEG_I(sbi, i);
782 old_curseg = curseg->segno;
783 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
784 locate_dirty_segment(sbi, old_curseg);
785 }
786}
787
788static const struct segment_allocation default_salloc_ops = {
789 .allocate_segment = allocate_segment_by_default,
790};
791
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900792static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
793{
794 struct curseg_info *curseg = CURSEG_I(sbi, type);
795 if (curseg->next_blkoff < sbi->blocks_per_seg)
796 return true;
797 return false;
798}
799
800static int __get_segment_type_2(struct page *page, enum page_type p_type)
801{
802 if (p_type == DATA)
803 return CURSEG_HOT_DATA;
804 else
805 return CURSEG_HOT_NODE;
806}
807
808static int __get_segment_type_4(struct page *page, enum page_type p_type)
809{
810 if (p_type == DATA) {
811 struct inode *inode = page->mapping->host;
812
813 if (S_ISDIR(inode->i_mode))
814 return CURSEG_HOT_DATA;
815 else
816 return CURSEG_COLD_DATA;
817 } else {
818 if (IS_DNODE(page) && !is_cold_node(page))
819 return CURSEG_HOT_NODE;
820 else
821 return CURSEG_COLD_NODE;
822 }
823}
824
825static int __get_segment_type_6(struct page *page, enum page_type p_type)
826{
827 if (p_type == DATA) {
828 struct inode *inode = page->mapping->host;
829
830 if (S_ISDIR(inode->i_mode))
831 return CURSEG_HOT_DATA;
Jaegeuk Kim354a3392013-06-14 08:52:35 +0900832 else if (is_cold_data(page) || file_is_cold(inode))
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900833 return CURSEG_COLD_DATA;
834 else
835 return CURSEG_WARM_DATA;
836 } else {
837 if (IS_DNODE(page))
838 return is_cold_node(page) ? CURSEG_WARM_NODE :
839 CURSEG_HOT_NODE;
840 else
841 return CURSEG_COLD_NODE;
842 }
843}
844
845static int __get_segment_type(struct page *page, enum page_type p_type)
846{
847 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
848 switch (sbi->active_logs) {
849 case 2:
850 return __get_segment_type_2(page, p_type);
851 case 4:
852 return __get_segment_type_4(page, p_type);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900853 }
Jaegeuk Kim12a67142012-12-21 11:47:05 +0900854 /* NR_CURSEG_TYPE(6) logs by default */
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900855 f2fs_bug_on(sbi->active_logs != NR_CURSEG_TYPE);
Jaegeuk Kim12a67142012-12-21 11:47:05 +0900856 return __get_segment_type_6(page, p_type);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900857}
858
Jaegeuk Kimbfad7c22013-12-16 19:04:05 +0900859void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
860 block_t old_blkaddr, block_t *new_blkaddr,
861 struct f2fs_summary *sum, int type)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900862{
863 struct sit_info *sit_i = SIT_I(sbi);
864 struct curseg_info *curseg;
865 unsigned int old_cursegno;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900866
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900867 curseg = CURSEG_I(sbi, type);
868
869 mutex_lock(&curseg->curseg_mutex);
870
871 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
872 old_cursegno = curseg->segno;
873
874 /*
875 * __add_sum_entry should be resided under the curseg_mutex
876 * because, this function updates a summary entry in the
877 * current summary block.
878 */
Haicheng Lie79efe32013-06-13 16:59:27 +0800879 __add_sum_entry(sbi, type, sum);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900880
881 mutex_lock(&sit_i->sentry_lock);
882 __refresh_next_blkoff(sbi, curseg);
Jaegeuk Kimdcdfff62013-10-22 20:56:10 +0900883
884 stat_inc_block_count(sbi, curseg);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900885
Jaegeuk Kim5e443812014-01-28 12:22:14 +0900886 if (!__has_curseg_space(sbi, type))
887 sit_i->s_ops->allocate_segment(sbi, type, false);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900888 /*
889 * SIT information should be updated before segment allocation,
890 * since SSR needs latest valid block information.
891 */
892 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900893 locate_dirty_segment(sbi, old_cursegno);
Jaegeuk Kim5e443812014-01-28 12:22:14 +0900894
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900895 mutex_unlock(&sit_i->sentry_lock);
896
Jaegeuk Kimbfad7c22013-12-16 19:04:05 +0900897 if (page && IS_NODESEG(type))
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900898 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
899
Jaegeuk Kimbfad7c22013-12-16 19:04:05 +0900900 mutex_unlock(&curseg->curseg_mutex);
901}
902
903static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
904 block_t old_blkaddr, block_t *new_blkaddr,
905 struct f2fs_summary *sum, struct f2fs_io_info *fio)
906{
907 int type = __get_segment_type(page, fio->type);
908
909 allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
910
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900911 /* writeout dirty page into bdev */
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900912 f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900913}
914
Jaegeuk Kim577e3492013-01-24 19:56:11 +0900915void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900916{
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900917 struct f2fs_io_info fio = {
918 .type = META,
Gu Zheng7e8f2302013-12-20 18:17:49 +0800919 .rw = WRITE_SYNC | REQ_META | REQ_PRIO
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900920 };
921
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900922 set_page_writeback(page);
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900923 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900924}
925
926void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
Jaegeuk Kimfb5566d2014-01-08 10:09:51 +0900927 struct f2fs_io_info *fio,
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900928 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
929{
930 struct f2fs_summary sum;
931 set_summary(&sum, nid, 0, 0);
Jaegeuk Kimfb5566d2014-01-08 10:09:51 +0900932 do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900933}
934
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900935void write_data_page(struct page *page, struct dnode_of_data *dn,
936 block_t *new_blkaddr, struct f2fs_io_info *fio)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900937{
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900938 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900939 struct f2fs_summary sum;
940 struct node_info ni;
941
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900942 f2fs_bug_on(dn->data_blkaddr == NULL_ADDR);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900943 get_node_info(sbi, dn->nid, &ni);
944 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
945
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900946 do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900947}
948
Chris Fries6c311ec2014-01-17 14:44:39 -0600949void rewrite_data_page(struct page *page, block_t old_blkaddr,
950 struct f2fs_io_info *fio)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900951{
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900952 struct inode *inode = page->mapping->host;
953 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
954 f2fs_submit_page_mbio(sbi, page, old_blkaddr, fio);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900955}
956
957void recover_data_page(struct f2fs_sb_info *sbi,
958 struct page *page, struct f2fs_summary *sum,
959 block_t old_blkaddr, block_t new_blkaddr)
960{
961 struct sit_info *sit_i = SIT_I(sbi);
962 struct curseg_info *curseg;
963 unsigned int segno, old_cursegno;
964 struct seg_entry *se;
965 int type;
966
967 segno = GET_SEGNO(sbi, new_blkaddr);
968 se = get_seg_entry(sbi, segno);
969 type = se->type;
970
971 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
972 if (old_blkaddr == NULL_ADDR)
973 type = CURSEG_COLD_DATA;
974 else
975 type = CURSEG_WARM_DATA;
976 }
977 curseg = CURSEG_I(sbi, type);
978
979 mutex_lock(&curseg->curseg_mutex);
980 mutex_lock(&sit_i->sentry_lock);
981
982 old_cursegno = curseg->segno;
983
984 /* change the current segment */
985 if (segno != curseg->segno) {
986 curseg->next_segno = segno;
987 change_curseg(sbi, type, true);
988 }
989
Jaegeuk Kim491c0852014-02-04 13:01:10 +0900990 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
Haicheng Lie79efe32013-06-13 16:59:27 +0800991 __add_sum_entry(sbi, type, sum);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900992
993 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900994 locate_dirty_segment(sbi, old_cursegno);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +0900995
996 mutex_unlock(&sit_i->sentry_lock);
997 mutex_unlock(&curseg->curseg_mutex);
998}
999
1000void rewrite_node_page(struct f2fs_sb_info *sbi,
1001 struct page *page, struct f2fs_summary *sum,
1002 block_t old_blkaddr, block_t new_blkaddr)
1003{
1004 struct sit_info *sit_i = SIT_I(sbi);
1005 int type = CURSEG_WARM_NODE;
1006 struct curseg_info *curseg;
1007 unsigned int segno, old_cursegno;
1008 block_t next_blkaddr = next_blkaddr_of_node(page);
1009 unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
Jaegeuk Kim458e6192013-12-11 13:54:01 +09001010 struct f2fs_io_info fio = {
1011 .type = NODE,
1012 .rw = WRITE_SYNC,
Jaegeuk Kim458e6192013-12-11 13:54:01 +09001013 };
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001014
1015 curseg = CURSEG_I(sbi, type);
1016
1017 mutex_lock(&curseg->curseg_mutex);
1018 mutex_lock(&sit_i->sentry_lock);
1019
1020 segno = GET_SEGNO(sbi, new_blkaddr);
1021 old_cursegno = curseg->segno;
1022
1023 /* change the current segment */
1024 if (segno != curseg->segno) {
1025 curseg->next_segno = segno;
1026 change_curseg(sbi, type, true);
1027 }
Jaegeuk Kim491c0852014-02-04 13:01:10 +09001028 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
Haicheng Lie79efe32013-06-13 16:59:27 +08001029 __add_sum_entry(sbi, type, sum);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001030
1031 /* change the current log to the next block addr in advance */
1032 if (next_segno != segno) {
1033 curseg->next_segno = next_segno;
1034 change_curseg(sbi, type, true);
1035 }
Jaegeuk Kim491c0852014-02-04 13:01:10 +09001036 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, next_blkaddr);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001037
1038 /* rewrite node page */
1039 set_page_writeback(page);
Jaegeuk Kim458e6192013-12-11 13:54:01 +09001040 f2fs_submit_page_mbio(sbi, page, new_blkaddr, &fio);
1041 f2fs_submit_merged_bio(sbi, NODE, WRITE);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001042 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001043 locate_dirty_segment(sbi, old_cursegno);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001044
1045 mutex_unlock(&sit_i->sentry_lock);
1046 mutex_unlock(&curseg->curseg_mutex);
1047}
1048
Jaegeuk Kim93dfe2a2013-11-30 12:51:14 +09001049void f2fs_wait_on_page_writeback(struct page *page,
Yuan Zhong5514f0a2014-01-10 07:26:14 +00001050 enum page_type type)
Jaegeuk Kim93dfe2a2013-11-30 12:51:14 +09001051{
1052 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1053 if (PageWriteback(page)) {
Jaegeuk Kim458e6192013-12-11 13:54:01 +09001054 f2fs_submit_merged_bio(sbi, type, WRITE);
Jaegeuk Kim93dfe2a2013-11-30 12:51:14 +09001055 wait_on_page_writeback(page);
1056 }
1057}
1058
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001059static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1060{
1061 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1062 struct curseg_info *seg_i;
1063 unsigned char *kaddr;
1064 struct page *page;
1065 block_t start;
1066 int i, j, offset;
1067
1068 start = start_sum_block(sbi);
1069
1070 page = get_meta_page(sbi, start++);
1071 kaddr = (unsigned char *)page_address(page);
1072
1073 /* Step 1: restore nat cache */
1074 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1075 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1076
1077 /* Step 2: restore sit cache */
1078 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1079 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1080 SUM_JOURNAL_SIZE);
1081 offset = 2 * SUM_JOURNAL_SIZE;
1082
1083 /* Step 3: restore summary entries */
1084 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1085 unsigned short blk_off;
1086 unsigned int segno;
1087
1088 seg_i = CURSEG_I(sbi, i);
1089 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1090 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1091 seg_i->next_segno = segno;
1092 reset_curseg(sbi, i, 0);
1093 seg_i->alloc_type = ckpt->alloc_type[i];
1094 seg_i->next_blkoff = blk_off;
1095
1096 if (seg_i->alloc_type == SSR)
1097 blk_off = sbi->blocks_per_seg;
1098
1099 for (j = 0; j < blk_off; j++) {
1100 struct f2fs_summary *s;
1101 s = (struct f2fs_summary *)(kaddr + offset);
1102 seg_i->sum_blk->entries[j] = *s;
1103 offset += SUMMARY_SIZE;
1104 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1105 SUM_FOOTER_SIZE)
1106 continue;
1107
1108 f2fs_put_page(page, 1);
1109 page = NULL;
1110
1111 page = get_meta_page(sbi, start++);
1112 kaddr = (unsigned char *)page_address(page);
1113 offset = 0;
1114 }
1115 }
1116 f2fs_put_page(page, 1);
1117 return 0;
1118}
1119
1120static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1121{
1122 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1123 struct f2fs_summary_block *sum;
1124 struct curseg_info *curseg;
1125 struct page *new;
1126 unsigned short blk_off;
1127 unsigned int segno = 0;
1128 block_t blk_addr = 0;
1129
1130 /* get segment number and block addr */
1131 if (IS_DATASEG(type)) {
1132 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1133 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1134 CURSEG_HOT_DATA]);
Jaegeuk Kim25ca9232012-11-28 16:12:41 +09001135 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001136 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1137 else
1138 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1139 } else {
1140 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1141 CURSEG_HOT_NODE]);
1142 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1143 CURSEG_HOT_NODE]);
Jaegeuk Kim25ca9232012-11-28 16:12:41 +09001144 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001145 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1146 type - CURSEG_HOT_NODE);
1147 else
1148 blk_addr = GET_SUM_BLOCK(sbi, segno);
1149 }
1150
1151 new = get_meta_page(sbi, blk_addr);
1152 sum = (struct f2fs_summary_block *)page_address(new);
1153
1154 if (IS_NODESEG(type)) {
Jaegeuk Kim25ca9232012-11-28 16:12:41 +09001155 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001156 struct f2fs_summary *ns = &sum->entries[0];
1157 int i;
1158 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1159 ns->version = 0;
1160 ns->ofs_in_node = 0;
1161 }
1162 } else {
1163 if (restore_node_summary(sbi, segno, sum)) {
1164 f2fs_put_page(new, 1);
1165 return -EINVAL;
1166 }
1167 }
1168 }
1169
1170 /* set uncompleted segment to curseg */
1171 curseg = CURSEG_I(sbi, type);
1172 mutex_lock(&curseg->curseg_mutex);
1173 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1174 curseg->next_segno = segno;
1175 reset_curseg(sbi, type, 0);
1176 curseg->alloc_type = ckpt->alloc_type[type];
1177 curseg->next_blkoff = blk_off;
1178 mutex_unlock(&curseg->curseg_mutex);
1179 f2fs_put_page(new, 1);
1180 return 0;
1181}
1182
1183static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1184{
1185 int type = CURSEG_HOT_DATA;
1186
Jaegeuk Kim25ca9232012-11-28 16:12:41 +09001187 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001188 /* restore for compacted data summary */
1189 if (read_compacted_summaries(sbi))
1190 return -EINVAL;
1191 type = CURSEG_HOT_NODE;
1192 }
1193
1194 for (; type <= CURSEG_COLD_NODE; type++)
1195 if (read_normal_summaries(sbi, type))
1196 return -EINVAL;
1197 return 0;
1198}
1199
1200static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1201{
1202 struct page *page;
1203 unsigned char *kaddr;
1204 struct f2fs_summary *summary;
1205 struct curseg_info *seg_i;
1206 int written_size = 0;
1207 int i, j;
1208
1209 page = grab_meta_page(sbi, blkaddr++);
1210 kaddr = (unsigned char *)page_address(page);
1211
1212 /* Step 1: write nat cache */
1213 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1214 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1215 written_size += SUM_JOURNAL_SIZE;
1216
1217 /* Step 2: write sit cache */
1218 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1219 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1220 SUM_JOURNAL_SIZE);
1221 written_size += SUM_JOURNAL_SIZE;
1222
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001223 /* Step 3: write summary entries */
1224 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1225 unsigned short blkoff;
1226 seg_i = CURSEG_I(sbi, i);
1227 if (sbi->ckpt->alloc_type[i] == SSR)
1228 blkoff = sbi->blocks_per_seg;
1229 else
1230 blkoff = curseg_blkoff(sbi, i);
1231
1232 for (j = 0; j < blkoff; j++) {
1233 if (!page) {
1234 page = grab_meta_page(sbi, blkaddr++);
1235 kaddr = (unsigned char *)page_address(page);
1236 written_size = 0;
1237 }
1238 summary = (struct f2fs_summary *)(kaddr + written_size);
1239 *summary = seg_i->sum_blk->entries[j];
1240 written_size += SUMMARY_SIZE;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001241
1242 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1243 SUM_FOOTER_SIZE)
1244 continue;
1245
Chao Yue8d61a72013-10-24 15:08:28 +08001246 set_page_dirty(page);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001247 f2fs_put_page(page, 1);
1248 page = NULL;
1249 }
1250 }
Chao Yue8d61a72013-10-24 15:08:28 +08001251 if (page) {
1252 set_page_dirty(page);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001253 f2fs_put_page(page, 1);
Chao Yue8d61a72013-10-24 15:08:28 +08001254 }
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001255}
1256
1257static void write_normal_summaries(struct f2fs_sb_info *sbi,
1258 block_t blkaddr, int type)
1259{
1260 int i, end;
1261 if (IS_DATASEG(type))
1262 end = type + NR_CURSEG_DATA_TYPE;
1263 else
1264 end = type + NR_CURSEG_NODE_TYPE;
1265
1266 for (i = type; i < end; i++) {
1267 struct curseg_info *sum = CURSEG_I(sbi, i);
1268 mutex_lock(&sum->curseg_mutex);
1269 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1270 mutex_unlock(&sum->curseg_mutex);
1271 }
1272}
1273
1274void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1275{
Jaegeuk Kim25ca9232012-11-28 16:12:41 +09001276 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001277 write_compacted_summaries(sbi, start_blk);
1278 else
1279 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1280}
1281
1282void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1283{
Jaegeuk Kim25ca9232012-11-28 16:12:41 +09001284 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001285 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001286}
1287
1288int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1289 unsigned int val, int alloc)
1290{
1291 int i;
1292
1293 if (type == NAT_JOURNAL) {
1294 for (i = 0; i < nats_in_cursum(sum); i++) {
1295 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1296 return i;
1297 }
1298 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1299 return update_nats_in_cursum(sum, 1);
1300 } else if (type == SIT_JOURNAL) {
1301 for (i = 0; i < sits_in_cursum(sum); i++)
1302 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1303 return i;
1304 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1305 return update_sits_in_cursum(sum, 1);
1306 }
1307 return -1;
1308}
1309
1310static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1311 unsigned int segno)
1312{
1313 struct sit_info *sit_i = SIT_I(sbi);
1314 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1315 block_t blk_addr = sit_i->sit_base_addr + offset;
1316
1317 check_seg_range(sbi, segno);
1318
1319 /* calculate sit block address */
1320 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1321 blk_addr += sit_i->sit_blocks;
1322
1323 return get_meta_page(sbi, blk_addr);
1324}
1325
1326static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1327 unsigned int start)
1328{
1329 struct sit_info *sit_i = SIT_I(sbi);
1330 struct page *src_page, *dst_page;
1331 pgoff_t src_off, dst_off;
1332 void *src_addr, *dst_addr;
1333
1334 src_off = current_sit_addr(sbi, start);
1335 dst_off = next_sit_addr(sbi, src_off);
1336
1337 /* get current sit block page without lock */
1338 src_page = get_meta_page(sbi, src_off);
1339 dst_page = grab_meta_page(sbi, dst_off);
Jaegeuk Kim5d56b672013-10-29 15:14:54 +09001340 f2fs_bug_on(PageDirty(src_page));
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001341
1342 src_addr = page_address(src_page);
1343 dst_addr = page_address(dst_page);
1344 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1345
1346 set_page_dirty(dst_page);
1347 f2fs_put_page(src_page, 1);
1348
1349 set_to_next_sit(sit_i, start);
1350
1351 return dst_page;
1352}
1353
1354static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1355{
1356 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1357 struct f2fs_summary_block *sum = curseg->sum_blk;
1358 int i;
1359
1360 /*
1361 * If the journal area in the current summary is full of sit entries,
1362 * all the sit entries will be flushed. Otherwise the sit entries
1363 * are not able to replace with newly hot sit entries.
1364 */
1365 if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1366 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1367 unsigned int segno;
1368 segno = le32_to_cpu(segno_in_journal(sum, i));
1369 __mark_sit_entry_dirty(sbi, segno);
1370 }
1371 update_sits_in_cursum(sum, -sits_in_cursum(sum));
Haicheng Licffbfa62013-10-18 17:24:07 +08001372 return true;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001373 }
Haicheng Licffbfa62013-10-18 17:24:07 +08001374 return false;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001375}
1376
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +09001377/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001378 * CP calls this function, which flushes SIT entries including sit_journal,
1379 * and moves prefree segs to free segs.
1380 */
1381void flush_sit_entries(struct f2fs_sb_info *sbi)
1382{
1383 struct sit_info *sit_i = SIT_I(sbi);
1384 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1385 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1386 struct f2fs_summary_block *sum = curseg->sum_blk;
1387 unsigned long nsegs = TOTAL_SEGS(sbi);
1388 struct page *page = NULL;
1389 struct f2fs_sit_block *raw_sit = NULL;
1390 unsigned int start = 0, end = 0;
1391 unsigned int segno = -1;
1392 bool flushed;
1393
1394 mutex_lock(&curseg->curseg_mutex);
1395 mutex_lock(&sit_i->sentry_lock);
1396
1397 /*
1398 * "flushed" indicates whether sit entries in journal are flushed
1399 * to the SIT area or not.
1400 */
1401 flushed = flush_sits_in_journal(sbi);
1402
1403 while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1404 struct seg_entry *se = get_seg_entry(sbi, segno);
1405 int sit_offset, offset;
1406
1407 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1408
Jaegeuk Kimb2955552013-11-12 14:49:56 +09001409 /* add discard candidates */
1410 if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards)
1411 add_discard_addrs(sbi, segno, se);
1412
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001413 if (flushed)
1414 goto to_sit_page;
1415
1416 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1417 if (offset >= 0) {
1418 segno_in_journal(sum, offset) = cpu_to_le32(segno);
1419 seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1420 goto flush_done;
1421 }
1422to_sit_page:
1423 if (!page || (start > segno) || (segno > end)) {
1424 if (page) {
1425 f2fs_put_page(page, 1);
1426 page = NULL;
1427 }
1428
1429 start = START_SEGNO(sit_i, segno);
1430 end = start + SIT_ENTRY_PER_BLOCK - 1;
1431
1432 /* read sit block that will be updated */
1433 page = get_next_sit_page(sbi, start);
1434 raw_sit = page_address(page);
1435 }
1436
1437 /* udpate entry in SIT block */
1438 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1439flush_done:
1440 __clear_bit(segno, bitmap);
1441 sit_i->dirty_sentries--;
1442 }
1443 mutex_unlock(&sit_i->sentry_lock);
1444 mutex_unlock(&curseg->curseg_mutex);
1445
1446 /* writeout last modified SIT block */
1447 f2fs_put_page(page, 1);
1448
1449 set_prefree_as_free_segments(sbi);
1450}
1451
1452static int build_sit_info(struct f2fs_sb_info *sbi)
1453{
1454 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1455 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1456 struct sit_info *sit_i;
1457 unsigned int sit_segs, start;
1458 char *src_bitmap, *dst_bitmap;
1459 unsigned int bitmap_size;
1460
1461 /* allocate memory for SIT information */
1462 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1463 if (!sit_i)
1464 return -ENOMEM;
1465
1466 SM_I(sbi)->sit_info = sit_i;
1467
1468 sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1469 if (!sit_i->sentries)
1470 return -ENOMEM;
1471
1472 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1473 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1474 if (!sit_i->dirty_sentries_bitmap)
1475 return -ENOMEM;
1476
1477 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1478 sit_i->sentries[start].cur_valid_map
1479 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1480 sit_i->sentries[start].ckpt_valid_map
1481 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1482 if (!sit_i->sentries[start].cur_valid_map
1483 || !sit_i->sentries[start].ckpt_valid_map)
1484 return -ENOMEM;
1485 }
1486
1487 if (sbi->segs_per_sec > 1) {
Jaegeuk Kim53cf9522013-03-31 12:39:49 +09001488 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001489 sizeof(struct sec_entry));
1490 if (!sit_i->sec_entries)
1491 return -ENOMEM;
1492 }
1493
1494 /* get information related with SIT */
1495 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1496
1497 /* setup SIT bitmap from ckeckpoint pack */
1498 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1499 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1500
Alexandru Gheorghiu79b57932013-03-28 02:24:53 +02001501 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001502 if (!dst_bitmap)
1503 return -ENOMEM;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001504
1505 /* init SIT information */
1506 sit_i->s_ops = &default_salloc_ops;
1507
1508 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1509 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1510 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1511 sit_i->sit_bitmap = dst_bitmap;
1512 sit_i->bitmap_size = bitmap_size;
1513 sit_i->dirty_sentries = 0;
1514 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1515 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1516 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1517 mutex_init(&sit_i->sentry_lock);
1518 return 0;
1519}
1520
1521static int build_free_segmap(struct f2fs_sb_info *sbi)
1522{
1523 struct f2fs_sm_info *sm_info = SM_I(sbi);
1524 struct free_segmap_info *free_i;
1525 unsigned int bitmap_size, sec_bitmap_size;
1526
1527 /* allocate memory for free segmap information */
1528 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1529 if (!free_i)
1530 return -ENOMEM;
1531
1532 SM_I(sbi)->free_info = free_i;
1533
1534 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1535 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1536 if (!free_i->free_segmap)
1537 return -ENOMEM;
1538
Jaegeuk Kim53cf9522013-03-31 12:39:49 +09001539 sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001540 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1541 if (!free_i->free_secmap)
1542 return -ENOMEM;
1543
1544 /* set all segments as dirty temporarily */
1545 memset(free_i->free_segmap, 0xff, bitmap_size);
1546 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1547
1548 /* init free segmap information */
1549 free_i->start_segno =
1550 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1551 free_i->free_segments = 0;
1552 free_i->free_sections = 0;
1553 rwlock_init(&free_i->segmap_lock);
1554 return 0;
1555}
1556
1557static int build_curseg(struct f2fs_sb_info *sbi)
1558{
Namjae Jeon1042d602012-12-01 10:56:13 +09001559 struct curseg_info *array;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001560 int i;
1561
1562 array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1563 if (!array)
1564 return -ENOMEM;
1565
1566 SM_I(sbi)->curseg_array = array;
1567
1568 for (i = 0; i < NR_CURSEG_TYPE; i++) {
1569 mutex_init(&array[i].curseg_mutex);
1570 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1571 if (!array[i].sum_blk)
1572 return -ENOMEM;
1573 array[i].segno = NULL_SEGNO;
1574 array[i].next_blkoff = 0;
1575 }
1576 return restore_curseg_summaries(sbi);
1577}
1578
Chao Yu74de5932013-11-22 09:09:59 +08001579static int ra_sit_pages(struct f2fs_sb_info *sbi, int start, int nrpages)
1580{
Gu Zheng9df27d92014-01-20 18:37:04 +08001581 struct address_space *mapping = META_MAPPING(sbi);
Chao Yu74de5932013-11-22 09:09:59 +08001582 struct page *page;
1583 block_t blk_addr, prev_blk_addr = 0;
1584 int sit_blk_cnt = SIT_BLK_CNT(sbi);
1585 int blkno = start;
Jaegeuk Kim458e6192013-12-11 13:54:01 +09001586 struct f2fs_io_info fio = {
1587 .type = META,
Gu Zheng7e8f2302013-12-20 18:17:49 +08001588 .rw = READ_SYNC | REQ_META | REQ_PRIO
Jaegeuk Kim458e6192013-12-11 13:54:01 +09001589 };
Chao Yu74de5932013-11-22 09:09:59 +08001590
1591 for (; blkno < start + nrpages && blkno < sit_blk_cnt; blkno++) {
1592
1593 blk_addr = current_sit_addr(sbi, blkno * SIT_ENTRY_PER_BLOCK);
1594
1595 if (blkno != start && prev_blk_addr + 1 != blk_addr)
1596 break;
1597 prev_blk_addr = blk_addr;
1598repeat:
1599 page = grab_cache_page(mapping, blk_addr);
1600 if (!page) {
1601 cond_resched();
1602 goto repeat;
1603 }
1604 if (PageUptodate(page)) {
1605 mark_page_accessed(page);
1606 f2fs_put_page(page, 1);
1607 continue;
1608 }
1609
Jaegeuk Kim458e6192013-12-11 13:54:01 +09001610 f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
Chao Yu74de5932013-11-22 09:09:59 +08001611
1612 mark_page_accessed(page);
1613 f2fs_put_page(page, 0);
1614 }
1615
Jaegeuk Kim458e6192013-12-11 13:54:01 +09001616 f2fs_submit_merged_bio(sbi, META, READ);
Chao Yu74de5932013-11-22 09:09:59 +08001617 return blkno - start;
1618}
1619
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001620static void build_sit_entries(struct f2fs_sb_info *sbi)
1621{
1622 struct sit_info *sit_i = SIT_I(sbi);
1623 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1624 struct f2fs_summary_block *sum = curseg->sum_blk;
Chao Yu74de5932013-11-22 09:09:59 +08001625 int sit_blk_cnt = SIT_BLK_CNT(sbi);
1626 unsigned int i, start, end;
1627 unsigned int readed, start_blk = 0;
1628 int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001629
Chao Yu74de5932013-11-22 09:09:59 +08001630 do {
1631 readed = ra_sit_pages(sbi, start_blk, nrpages);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001632
Chao Yu74de5932013-11-22 09:09:59 +08001633 start = start_blk * sit_i->sents_per_block;
1634 end = (start_blk + readed) * sit_i->sents_per_block;
1635
1636 for (; start < end && start < TOTAL_SEGS(sbi); start++) {
1637 struct seg_entry *se = &sit_i->sentries[start];
1638 struct f2fs_sit_block *sit_blk;
1639 struct f2fs_sit_entry sit;
1640 struct page *page;
1641
1642 mutex_lock(&curseg->curseg_mutex);
1643 for (i = 0; i < sits_in_cursum(sum); i++) {
Chris Fries6c311ec2014-01-17 14:44:39 -06001644 if (le32_to_cpu(segno_in_journal(sum, i))
1645 == start) {
Chao Yu74de5932013-11-22 09:09:59 +08001646 sit = sit_in_journal(sum, i);
1647 mutex_unlock(&curseg->curseg_mutex);
1648 goto got_it;
1649 }
1650 }
1651 mutex_unlock(&curseg->curseg_mutex);
1652
1653 page = get_current_sit_page(sbi, start);
1654 sit_blk = (struct f2fs_sit_block *)page_address(page);
1655 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1656 f2fs_put_page(page, 1);
1657got_it:
1658 check_block_count(sbi, start, &sit);
1659 seg_info_from_raw_sit(se, &sit);
1660 if (sbi->segs_per_sec > 1) {
1661 struct sec_entry *e = get_sec_entry(sbi, start);
1662 e->valid_blocks += se->valid_blocks;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001663 }
1664 }
Chao Yu74de5932013-11-22 09:09:59 +08001665 start_blk += readed;
1666 } while (start_blk < sit_blk_cnt);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001667}
1668
1669static void init_free_segmap(struct f2fs_sb_info *sbi)
1670{
1671 unsigned int start;
1672 int type;
1673
1674 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1675 struct seg_entry *sentry = get_seg_entry(sbi, start);
1676 if (!sentry->valid_blocks)
1677 __set_free(sbi, start);
1678 }
1679
1680 /* set use the current segments */
1681 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1682 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1683 __set_test_and_inuse(sbi, curseg_t->segno);
1684 }
1685}
1686
1687static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1688{
1689 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1690 struct free_segmap_info *free_i = FREE_I(sbi);
Namjae Jeon8736fbf2013-06-16 09:49:11 +09001691 unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001692 unsigned short valid_blocks;
1693
Namjae Jeon8736fbf2013-06-16 09:49:11 +09001694 while (1) {
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001695 /* find dirty segment based on free segmap */
Namjae Jeon8736fbf2013-06-16 09:49:11 +09001696 segno = find_next_inuse(free_i, total_segs, offset);
1697 if (segno >= total_segs)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001698 break;
1699 offset = segno + 1;
1700 valid_blocks = get_valid_blocks(sbi, segno, 0);
1701 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1702 continue;
1703 mutex_lock(&dirty_i->seglist_lock);
1704 __locate_dirty_segment(sbi, segno, DIRTY);
1705 mutex_unlock(&dirty_i->seglist_lock);
1706 }
1707}
1708
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +09001709static int init_victim_secmap(struct f2fs_sb_info *sbi)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001710{
1711 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +09001712 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001713
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +09001714 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1715 if (!dirty_i->victim_secmap)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001716 return -ENOMEM;
1717 return 0;
1718}
1719
1720static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1721{
1722 struct dirty_seglist_info *dirty_i;
1723 unsigned int bitmap_size, i;
1724
1725 /* allocate memory for dirty segments list information */
1726 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1727 if (!dirty_i)
1728 return -ENOMEM;
1729
1730 SM_I(sbi)->dirty_info = dirty_i;
1731 mutex_init(&dirty_i->seglist_lock);
1732
1733 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1734
1735 for (i = 0; i < NR_DIRTY_TYPE; i++) {
1736 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001737 if (!dirty_i->dirty_segmap[i])
1738 return -ENOMEM;
1739 }
1740
1741 init_dirty_segmap(sbi);
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +09001742 return init_victim_secmap(sbi);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001743}
1744
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +09001745/*
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001746 * Update min, max modified time for cost-benefit GC algorithm
1747 */
1748static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1749{
1750 struct sit_info *sit_i = SIT_I(sbi);
1751 unsigned int segno;
1752
1753 mutex_lock(&sit_i->sentry_lock);
1754
1755 sit_i->min_mtime = LLONG_MAX;
1756
1757 for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1758 unsigned int i;
1759 unsigned long long mtime = 0;
1760
1761 for (i = 0; i < sbi->segs_per_sec; i++)
1762 mtime += get_seg_entry(sbi, segno + i)->mtime;
1763
1764 mtime = div_u64(mtime, sbi->segs_per_sec);
1765
1766 if (sit_i->min_mtime > mtime)
1767 sit_i->min_mtime = mtime;
1768 }
1769 sit_i->max_mtime = get_mtime(sbi);
1770 mutex_unlock(&sit_i->sentry_lock);
1771}
1772
1773int build_segment_manager(struct f2fs_sb_info *sbi)
1774{
1775 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1776 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
Namjae Jeon1042d602012-12-01 10:56:13 +09001777 struct f2fs_sm_info *sm_info;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001778 int err;
1779
1780 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1781 if (!sm_info)
1782 return -ENOMEM;
1783
1784 /* init sm info */
1785 sbi->sm_info = sm_info;
1786 INIT_LIST_HEAD(&sm_info->wblist_head);
1787 spin_lock_init(&sm_info->wblist_lock);
1788 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1789 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1790 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1791 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1792 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1793 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1794 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +09001795 sm_info->rec_prefree_segments = DEF_RECLAIM_PREFREE_SEGMENTS;
Jaegeuk Kim216fbd62013-11-07 13:13:42 +09001796 sm_info->ipu_policy = F2FS_IPU_DISABLE;
1797 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001798
Jaegeuk Kim7fd9e542013-11-15 13:55:58 +09001799 INIT_LIST_HEAD(&sm_info->discard_list);
1800 sm_info->nr_discards = 0;
1801 sm_info->max_discards = 0;
1802
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001803 err = build_sit_info(sbi);
1804 if (err)
1805 return err;
1806 err = build_free_segmap(sbi);
1807 if (err)
1808 return err;
1809 err = build_curseg(sbi);
1810 if (err)
1811 return err;
1812
1813 /* reinit free segmap based on SIT */
1814 build_sit_entries(sbi);
1815
1816 init_free_segmap(sbi);
1817 err = build_dirty_segmap(sbi);
1818 if (err)
1819 return err;
1820
1821 init_min_max_mtime(sbi);
1822 return 0;
1823}
1824
1825static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1826 enum dirty_type dirty_type)
1827{
1828 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1829
1830 mutex_lock(&dirty_i->seglist_lock);
1831 kfree(dirty_i->dirty_segmap[dirty_type]);
1832 dirty_i->nr_dirty[dirty_type] = 0;
1833 mutex_unlock(&dirty_i->seglist_lock);
1834}
1835
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +09001836static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001837{
1838 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +09001839 kfree(dirty_i->victim_secmap);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001840}
1841
1842static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1843{
1844 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1845 int i;
1846
1847 if (!dirty_i)
1848 return;
1849
1850 /* discard pre-free/dirty segments list */
1851 for (i = 0; i < NR_DIRTY_TYPE; i++)
1852 discard_dirty_segmap(sbi, i);
1853
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +09001854 destroy_victim_secmap(sbi);
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001855 SM_I(sbi)->dirty_info = NULL;
1856 kfree(dirty_i);
1857}
1858
1859static void destroy_curseg(struct f2fs_sb_info *sbi)
1860{
1861 struct curseg_info *array = SM_I(sbi)->curseg_array;
1862 int i;
1863
1864 if (!array)
1865 return;
1866 SM_I(sbi)->curseg_array = NULL;
1867 for (i = 0; i < NR_CURSEG_TYPE; i++)
1868 kfree(array[i].sum_blk);
1869 kfree(array);
1870}
1871
1872static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1873{
1874 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1875 if (!free_i)
1876 return;
1877 SM_I(sbi)->free_info = NULL;
1878 kfree(free_i->free_segmap);
1879 kfree(free_i->free_secmap);
1880 kfree(free_i);
1881}
1882
1883static void destroy_sit_info(struct f2fs_sb_info *sbi)
1884{
1885 struct sit_info *sit_i = SIT_I(sbi);
1886 unsigned int start;
1887
1888 if (!sit_i)
1889 return;
1890
1891 if (sit_i->sentries) {
1892 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1893 kfree(sit_i->sentries[start].cur_valid_map);
1894 kfree(sit_i->sentries[start].ckpt_valid_map);
1895 }
1896 }
1897 vfree(sit_i->sentries);
1898 vfree(sit_i->sec_entries);
1899 kfree(sit_i->dirty_sentries_bitmap);
1900
1901 SM_I(sbi)->sit_info = NULL;
1902 kfree(sit_i->sit_bitmap);
1903 kfree(sit_i);
1904}
1905
1906void destroy_segment_manager(struct f2fs_sb_info *sbi)
1907{
1908 struct f2fs_sm_info *sm_info = SM_I(sbi);
Chao Yu3b03f722013-11-06 09:12:04 +08001909 if (!sm_info)
1910 return;
Jaegeuk Kim351df4b2012-11-02 17:09:16 +09001911 destroy_dirty_segmap(sbi);
1912 destroy_curseg(sbi);
1913 destroy_free_segmap(sbi);
1914 destroy_sit_info(sbi);
1915 sbi->sm_info = NULL;
1916 kfree(sm_info);
1917}
Jaegeuk Kim7fd9e542013-11-15 13:55:58 +09001918
1919int __init create_segment_manager_caches(void)
1920{
1921 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
1922 sizeof(struct discard_entry), NULL);
1923 if (!discard_entry_slab)
1924 return -ENOMEM;
1925 return 0;
1926}
1927
1928void destroy_segment_manager_caches(void)
1929{
1930 kmem_cache_destroy(discard_entry_slab);
1931}