]>
Commit | Line | Data |
---|---|---|
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 | 16 | #include <linux/vmalloc.h> |
74de593a | 17 | #include <linux/swap.h> |
351df4b2 JK |
18 | |
19 | #include "f2fs.h" | |
20 | #include "segment.h" | |
21 | #include "node.h" | |
6ec178da | 22 | #include <trace/events/f2fs.h> |
351df4b2 | 23 | |
9a7f143a CL |
24 | #define __reverse_ffz(x) __reverse_ffs(~(x)) |
25 | ||
7fd9e544 JK |
26 | static struct kmem_cache *discard_entry_slab; |
27 | ||
9a7f143a CL |
28 | /* |
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 | */ | |
32 | static 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 | */ | |
71 | static 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; | |
103 | aligned: | |
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; | |
114 | found_first: | |
115 | tmp &= (~0UL >> (BITS_PER_LONG - size)); | |
116 | if (tmp == 0UL) /* Are any bits set? */ | |
117 | return result + size; /* Nope. */ | |
118 | found_middle: | |
119 | return result + __reverse_ffs(tmp); | |
120 | } | |
121 | ||
122 | static 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; | |
154 | aligned: | |
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 | ||
166 | found_first: | |
167 | tmp |= ~0UL << size; | |
168 | if (tmp == ~0UL) /* Are any bits zero? */ | |
169 | return result + size; /* Nope. */ | |
170 | found_middle: | |
171 | return result + __reverse_ffz(tmp); | |
172 | } | |
173 | ||
0a8165d7 | 174 | /* |
351df4b2 JK |
175 | * This function balances dirty node and dentry pages. |
176 | * In addition, it controls garbage collection. | |
177 | */ | |
178 | void f2fs_balance_fs(struct f2fs_sb_info *sbi) | |
179 | { | |
351df4b2 | 180 | /* |
029cd28c JK |
181 | * We should do GC or end up with checkpoint, if there are so many dirty |
182 | * dir/node pages without enough free segments. | |
351df4b2 | 183 | */ |
43727527 | 184 | if (has_not_enough_free_secs(sbi, 0)) { |
351df4b2 | 185 | mutex_lock(&sbi->gc_mutex); |
408e9375 | 186 | f2fs_gc(sbi); |
351df4b2 JK |
187 | } |
188 | } | |
189 | ||
4660f9c0 JK |
190 | void 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 | ||
351df4b2 JK |
198 | static 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); | |
4625d6aa | 212 | enum dirty_type t = sentry->type; |
b2f2c390 | 213 | |
4625d6aa CL |
214 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) |
215 | dirty_i->nr_dirty[t]++; | |
351df4b2 JK |
216 | } |
217 | } | |
218 | ||
219 | static 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) { | |
4625d6aa CL |
228 | struct seg_entry *sentry = get_seg_entry(sbi, segno); |
229 | enum dirty_type t = sentry->type; | |
230 | ||
231 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) | |
232 | dirty_i->nr_dirty[t]--; | |
b2f2c390 | 233 | |
5ec4e49f JK |
234 | if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0) |
235 | clear_bit(GET_SECNO(sbi, segno), | |
236 | dirty_i->victim_secmap); | |
351df4b2 JK |
237 | } |
238 | } | |
239 | ||
0a8165d7 | 240 | /* |
351df4b2 JK |
241 | * 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 | */ | |
8d8451af | 245 | static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) |
351df4b2 JK |
246 | { |
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); | |
351df4b2 JK |
268 | } |
269 | ||
37208879 JK |
270 | static void f2fs_issue_discard(struct f2fs_sb_info *sbi, |
271 | block_t blkstart, block_t blklen) | |
272 | { | |
f9a4e6df JK |
273 | sector_t start = SECTOR_FROM_BLOCK(sbi, blkstart); |
274 | sector_t len = SECTOR_FROM_BLOCK(sbi, blklen); | |
37208879 | 275 | blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0); |
1661d07c | 276 | trace_f2fs_issue_discard(sbi->sb, blkstart, blklen); |
37208879 JK |
277 | } |
278 | ||
b2955550 JK |
279 | static 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 | ||
0a8165d7 | 320 | /* |
351df4b2 JK |
321 | * Should call clear_prefree_segments after checkpoint is done. |
322 | */ | |
323 | static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) | |
324 | { | |
325 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
81fb5e87 | 326 | unsigned int segno = -1; |
351df4b2 JK |
327 | 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, | |
81fb5e87 | 332 | segno + 1); |
351df4b2 JK |
333 | if (segno >= total_segs) |
334 | break; | |
335 | __set_test_and_free(sbi, segno); | |
351df4b2 JK |
336 | } |
337 | mutex_unlock(&dirty_i->seglist_lock); | |
338 | } | |
339 | ||
340 | void clear_prefree_segments(struct f2fs_sb_info *sbi) | |
341 | { | |
b2955550 | 342 | struct list_head *head = &(SM_I(sbi)->discard_list); |
2d7b822a | 343 | struct discard_entry *entry, *this; |
351df4b2 | 344 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); |
29e59c14 | 345 | unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; |
351df4b2 | 346 | unsigned int total_segs = TOTAL_SEGS(sbi); |
29e59c14 | 347 | unsigned int start = 0, end = -1; |
351df4b2 JK |
348 | |
349 | mutex_lock(&dirty_i->seglist_lock); | |
29e59c14 | 350 | |
351df4b2 | 351 | while (1) { |
29e59c14 CL |
352 | int i; |
353 | start = find_next_bit(prefree_map, total_segs, end + 1); | |
354 | if (start >= total_segs) | |
351df4b2 | 355 | break; |
29e59c14 CL |
356 | end = find_next_zero_bit(prefree_map, total_segs, start + 1); |
357 | ||
358 | for (i = start; i < end; i++) | |
359 | clear_bit(i, prefree_map); | |
360 | ||
361 | dirty_i->nr_dirty[PRE] -= end - start; | |
362 | ||
363 | if (!test_opt(sbi, DISCARD)) | |
364 | continue; | |
351df4b2 | 365 | |
37208879 JK |
366 | f2fs_issue_discard(sbi, START_BLOCK(sbi, start), |
367 | (end - start) << sbi->log_blocks_per_seg); | |
351df4b2 JK |
368 | } |
369 | mutex_unlock(&dirty_i->seglist_lock); | |
b2955550 JK |
370 | |
371 | /* send small discards */ | |
2d7b822a | 372 | list_for_each_entry_safe(entry, this, head, list) { |
37208879 | 373 | f2fs_issue_discard(sbi, entry->blkaddr, entry->len); |
b2955550 JK |
374 | list_del(&entry->list); |
375 | SM_I(sbi)->nr_discards -= entry->len; | |
376 | kmem_cache_free(discard_entry_slab, entry); | |
377 | } | |
351df4b2 JK |
378 | } |
379 | ||
380 | static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) | |
381 | { | |
382 | struct sit_info *sit_i = SIT_I(sbi); | |
383 | if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) | |
384 | sit_i->dirty_sentries++; | |
385 | } | |
386 | ||
387 | static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, | |
388 | unsigned int segno, int modified) | |
389 | { | |
390 | struct seg_entry *se = get_seg_entry(sbi, segno); | |
391 | se->type = type; | |
392 | if (modified) | |
393 | __mark_sit_entry_dirty(sbi, segno); | |
394 | } | |
395 | ||
396 | static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) | |
397 | { | |
398 | struct seg_entry *se; | |
399 | unsigned int segno, offset; | |
400 | long int new_vblocks; | |
401 | ||
402 | segno = GET_SEGNO(sbi, blkaddr); | |
403 | ||
404 | se = get_seg_entry(sbi, segno); | |
405 | new_vblocks = se->valid_blocks + del; | |
491c0854 | 406 | offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); |
351df4b2 | 407 | |
5d56b671 | 408 | f2fs_bug_on((new_vblocks >> (sizeof(unsigned short) << 3) || |
351df4b2 JK |
409 | (new_vblocks > sbi->blocks_per_seg))); |
410 | ||
411 | se->valid_blocks = new_vblocks; | |
412 | se->mtime = get_mtime(sbi); | |
413 | SIT_I(sbi)->max_mtime = se->mtime; | |
414 | ||
415 | /* Update valid block bitmap */ | |
416 | if (del > 0) { | |
417 | if (f2fs_set_bit(offset, se->cur_valid_map)) | |
418 | BUG(); | |
419 | } else { | |
420 | if (!f2fs_clear_bit(offset, se->cur_valid_map)) | |
421 | BUG(); | |
422 | } | |
423 | if (!f2fs_test_bit(offset, se->ckpt_valid_map)) | |
424 | se->ckpt_valid_blocks += del; | |
425 | ||
426 | __mark_sit_entry_dirty(sbi, segno); | |
427 | ||
428 | /* update total number of valid blocks to be written in ckpt area */ | |
429 | SIT_I(sbi)->written_valid_blocks += del; | |
430 | ||
431 | if (sbi->segs_per_sec > 1) | |
432 | get_sec_entry(sbi, segno)->valid_blocks += del; | |
433 | } | |
434 | ||
5e443818 | 435 | void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) |
351df4b2 | 436 | { |
5e443818 JK |
437 | update_sit_entry(sbi, new, 1); |
438 | if (GET_SEGNO(sbi, old) != NULL_SEGNO) | |
439 | update_sit_entry(sbi, old, -1); | |
440 | ||
441 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); | |
442 | locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); | |
351df4b2 JK |
443 | } |
444 | ||
445 | void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) | |
446 | { | |
447 | unsigned int segno = GET_SEGNO(sbi, addr); | |
448 | struct sit_info *sit_i = SIT_I(sbi); | |
449 | ||
5d56b671 | 450 | f2fs_bug_on(addr == NULL_ADDR); |
351df4b2 JK |
451 | if (addr == NEW_ADDR) |
452 | return; | |
453 | ||
454 | /* add it into sit main buffer */ | |
455 | mutex_lock(&sit_i->sentry_lock); | |
456 | ||
457 | update_sit_entry(sbi, addr, -1); | |
458 | ||
459 | /* add it into dirty seglist */ | |
460 | locate_dirty_segment(sbi, segno); | |
461 | ||
462 | mutex_unlock(&sit_i->sentry_lock); | |
463 | } | |
464 | ||
0a8165d7 | 465 | /* |
351df4b2 JK |
466 | * This function should be resided under the curseg_mutex lock |
467 | */ | |
468 | static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, | |
e79efe3b | 469 | struct f2fs_summary *sum) |
351df4b2 JK |
470 | { |
471 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
472 | void *addr = curseg->sum_blk; | |
e79efe3b | 473 | addr += curseg->next_blkoff * sizeof(struct f2fs_summary); |
351df4b2 | 474 | memcpy(addr, sum, sizeof(struct f2fs_summary)); |
351df4b2 JK |
475 | } |
476 | ||
0a8165d7 | 477 | /* |
351df4b2 JK |
478 | * Calculate the number of current summary pages for writing |
479 | */ | |
480 | int npages_for_summary_flush(struct f2fs_sb_info *sbi) | |
481 | { | |
351df4b2 | 482 | int valid_sum_count = 0; |
9a47938b | 483 | int i, sum_in_page; |
351df4b2 JK |
484 | |
485 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
486 | if (sbi->ckpt->alloc_type[i] == SSR) | |
487 | valid_sum_count += sbi->blocks_per_seg; | |
488 | else | |
489 | valid_sum_count += curseg_blkoff(sbi, i); | |
490 | } | |
491 | ||
9a47938b FL |
492 | sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE - |
493 | SUM_FOOTER_SIZE) / SUMMARY_SIZE; | |
494 | if (valid_sum_count <= sum_in_page) | |
351df4b2 | 495 | return 1; |
9a47938b FL |
496 | else if ((valid_sum_count - sum_in_page) <= |
497 | (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) | |
351df4b2 JK |
498 | return 2; |
499 | return 3; | |
500 | } | |
501 | ||
0a8165d7 | 502 | /* |
351df4b2 JK |
503 | * Caller should put this summary page |
504 | */ | |
505 | struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) | |
506 | { | |
507 | return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); | |
508 | } | |
509 | ||
510 | static void write_sum_page(struct f2fs_sb_info *sbi, | |
511 | struct f2fs_summary_block *sum_blk, block_t blk_addr) | |
512 | { | |
513 | struct page *page = grab_meta_page(sbi, blk_addr); | |
514 | void *kaddr = page_address(page); | |
515 | memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE); | |
516 | set_page_dirty(page); | |
517 | f2fs_put_page(page, 1); | |
518 | } | |
519 | ||
60374688 JK |
520 | static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) |
521 | { | |
522 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
81fb5e87 | 523 | unsigned int segno = curseg->segno + 1; |
60374688 JK |
524 | struct free_segmap_info *free_i = FREE_I(sbi); |
525 | ||
81fb5e87 HL |
526 | if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec) |
527 | return !test_bit(segno, free_i->free_segmap); | |
60374688 JK |
528 | return 0; |
529 | } | |
530 | ||
0a8165d7 | 531 | /* |
351df4b2 JK |
532 | * Find a new segment from the free segments bitmap to right order |
533 | * This function should be returned with success, otherwise BUG | |
534 | */ | |
535 | static void get_new_segment(struct f2fs_sb_info *sbi, | |
536 | unsigned int *newseg, bool new_sec, int dir) | |
537 | { | |
538 | struct free_segmap_info *free_i = FREE_I(sbi); | |
351df4b2 | 539 | unsigned int segno, secno, zoneno; |
53cf9522 | 540 | unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone; |
351df4b2 JK |
541 | unsigned int hint = *newseg / sbi->segs_per_sec; |
542 | unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); | |
543 | unsigned int left_start = hint; | |
544 | bool init = true; | |
545 | int go_left = 0; | |
546 | int i; | |
547 | ||
548 | write_lock(&free_i->segmap_lock); | |
549 | ||
550 | if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { | |
551 | segno = find_next_zero_bit(free_i->free_segmap, | |
552 | TOTAL_SEGS(sbi), *newseg + 1); | |
33afa7fd JK |
553 | if (segno - *newseg < sbi->segs_per_sec - |
554 | (*newseg % sbi->segs_per_sec)) | |
351df4b2 JK |
555 | goto got_it; |
556 | } | |
557 | find_other_zone: | |
53cf9522 JK |
558 | secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint); |
559 | if (secno >= TOTAL_SECS(sbi)) { | |
351df4b2 JK |
560 | if (dir == ALLOC_RIGHT) { |
561 | secno = find_next_zero_bit(free_i->free_secmap, | |
53cf9522 | 562 | TOTAL_SECS(sbi), 0); |
5d56b671 | 563 | f2fs_bug_on(secno >= TOTAL_SECS(sbi)); |
351df4b2 JK |
564 | } else { |
565 | go_left = 1; | |
566 | left_start = hint - 1; | |
567 | } | |
568 | } | |
569 | if (go_left == 0) | |
570 | goto skip_left; | |
571 | ||
572 | while (test_bit(left_start, free_i->free_secmap)) { | |
573 | if (left_start > 0) { | |
574 | left_start--; | |
575 | continue; | |
576 | } | |
577 | left_start = find_next_zero_bit(free_i->free_secmap, | |
53cf9522 | 578 | TOTAL_SECS(sbi), 0); |
5d56b671 | 579 | f2fs_bug_on(left_start >= TOTAL_SECS(sbi)); |
351df4b2 JK |
580 | break; |
581 | } | |
582 | secno = left_start; | |
583 | skip_left: | |
584 | hint = secno; | |
585 | segno = secno * sbi->segs_per_sec; | |
586 | zoneno = secno / sbi->secs_per_zone; | |
587 | ||
588 | /* give up on finding another zone */ | |
589 | if (!init) | |
590 | goto got_it; | |
591 | if (sbi->secs_per_zone == 1) | |
592 | goto got_it; | |
593 | if (zoneno == old_zoneno) | |
594 | goto got_it; | |
595 | if (dir == ALLOC_LEFT) { | |
596 | if (!go_left && zoneno + 1 >= total_zones) | |
597 | goto got_it; | |
598 | if (go_left && zoneno == 0) | |
599 | goto got_it; | |
600 | } | |
601 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
602 | if (CURSEG_I(sbi, i)->zone == zoneno) | |
603 | break; | |
604 | ||
605 | if (i < NR_CURSEG_TYPE) { | |
606 | /* zone is in user, try another */ | |
607 | if (go_left) | |
608 | hint = zoneno * sbi->secs_per_zone - 1; | |
609 | else if (zoneno + 1 >= total_zones) | |
610 | hint = 0; | |
611 | else | |
612 | hint = (zoneno + 1) * sbi->secs_per_zone; | |
613 | init = false; | |
614 | goto find_other_zone; | |
615 | } | |
616 | got_it: | |
617 | /* set it as dirty segment in free segmap */ | |
5d56b671 | 618 | f2fs_bug_on(test_bit(segno, free_i->free_segmap)); |
351df4b2 JK |
619 | __set_inuse(sbi, segno); |
620 | *newseg = segno; | |
621 | write_unlock(&free_i->segmap_lock); | |
622 | } | |
623 | ||
624 | static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) | |
625 | { | |
626 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
627 | struct summary_footer *sum_footer; | |
628 | ||
629 | curseg->segno = curseg->next_segno; | |
630 | curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); | |
631 | curseg->next_blkoff = 0; | |
632 | curseg->next_segno = NULL_SEGNO; | |
633 | ||
634 | sum_footer = &(curseg->sum_blk->footer); | |
635 | memset(sum_footer, 0, sizeof(struct summary_footer)); | |
636 | if (IS_DATASEG(type)) | |
637 | SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); | |
638 | if (IS_NODESEG(type)) | |
639 | SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); | |
640 | __set_sit_entry_type(sbi, type, curseg->segno, modified); | |
641 | } | |
642 | ||
0a8165d7 | 643 | /* |
351df4b2 JK |
644 | * Allocate a current working segment. |
645 | * This function always allocates a free segment in LFS manner. | |
646 | */ | |
647 | static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) | |
648 | { | |
649 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
650 | unsigned int segno = curseg->segno; | |
651 | int dir = ALLOC_LEFT; | |
652 | ||
653 | write_sum_page(sbi, curseg->sum_blk, | |
81fb5e87 | 654 | GET_SUM_BLOCK(sbi, segno)); |
351df4b2 JK |
655 | if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) |
656 | dir = ALLOC_RIGHT; | |
657 | ||
658 | if (test_opt(sbi, NOHEAP)) | |
659 | dir = ALLOC_RIGHT; | |
660 | ||
661 | get_new_segment(sbi, &segno, new_sec, dir); | |
662 | curseg->next_segno = segno; | |
663 | reset_curseg(sbi, type, 1); | |
664 | curseg->alloc_type = LFS; | |
665 | } | |
666 | ||
667 | static void __next_free_blkoff(struct f2fs_sb_info *sbi, | |
668 | struct curseg_info *seg, block_t start) | |
669 | { | |
670 | struct seg_entry *se = get_seg_entry(sbi, seg->segno); | |
e81c93cf CL |
671 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); |
672 | unsigned long target_map[entries]; | |
673 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; | |
674 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; | |
675 | int i, pos; | |
676 | ||
677 | for (i = 0; i < entries; i++) | |
678 | target_map[i] = ckpt_map[i] | cur_map[i]; | |
679 | ||
680 | pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); | |
681 | ||
682 | seg->next_blkoff = pos; | |
351df4b2 JK |
683 | } |
684 | ||
0a8165d7 | 685 | /* |
351df4b2 JK |
686 | * If a segment is written by LFS manner, next block offset is just obtained |
687 | * by increasing the current block offset. However, if a segment is written by | |
688 | * SSR manner, next block offset obtained by calling __next_free_blkoff | |
689 | */ | |
690 | static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, | |
691 | struct curseg_info *seg) | |
692 | { | |
693 | if (seg->alloc_type == SSR) | |
694 | __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); | |
695 | else | |
696 | seg->next_blkoff++; | |
697 | } | |
698 | ||
0a8165d7 | 699 | /* |
351df4b2 JK |
700 | * This function always allocates a used segment (from dirty seglist) by SSR |
701 | * manner, so it should recover the existing segment information of valid blocks | |
702 | */ | |
703 | static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) | |
704 | { | |
705 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
706 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
707 | unsigned int new_segno = curseg->next_segno; | |
708 | struct f2fs_summary_block *sum_node; | |
709 | struct page *sum_page; | |
710 | ||
711 | write_sum_page(sbi, curseg->sum_blk, | |
712 | GET_SUM_BLOCK(sbi, curseg->segno)); | |
713 | __set_test_and_inuse(sbi, new_segno); | |
714 | ||
715 | mutex_lock(&dirty_i->seglist_lock); | |
716 | __remove_dirty_segment(sbi, new_segno, PRE); | |
717 | __remove_dirty_segment(sbi, new_segno, DIRTY); | |
718 | mutex_unlock(&dirty_i->seglist_lock); | |
719 | ||
720 | reset_curseg(sbi, type, 1); | |
721 | curseg->alloc_type = SSR; | |
722 | __next_free_blkoff(sbi, curseg, 0); | |
723 | ||
724 | if (reuse) { | |
725 | sum_page = get_sum_page(sbi, new_segno); | |
726 | sum_node = (struct f2fs_summary_block *)page_address(sum_page); | |
727 | memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); | |
728 | f2fs_put_page(sum_page, 1); | |
729 | } | |
730 | } | |
731 | ||
43727527 JK |
732 | static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) |
733 | { | |
734 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
735 | const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; | |
736 | ||
737 | if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0)) | |
738 | return v_ops->get_victim(sbi, | |
739 | &(curseg)->next_segno, BG_GC, type, SSR); | |
740 | ||
741 | /* For data segments, let's do SSR more intensively */ | |
742 | for (; type >= CURSEG_HOT_DATA; type--) | |
743 | if (v_ops->get_victim(sbi, &(curseg)->next_segno, | |
744 | BG_GC, type, SSR)) | |
745 | return 1; | |
746 | return 0; | |
747 | } | |
748 | ||
351df4b2 JK |
749 | /* |
750 | * flush out current segment and replace it with new segment | |
751 | * This function should be returned with success, otherwise BUG | |
752 | */ | |
753 | static void allocate_segment_by_default(struct f2fs_sb_info *sbi, | |
754 | int type, bool force) | |
755 | { | |
756 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
351df4b2 | 757 | |
7b405275 | 758 | if (force) |
351df4b2 | 759 | new_curseg(sbi, type, true); |
7b405275 | 760 | else if (type == CURSEG_WARM_NODE) |
351df4b2 | 761 | new_curseg(sbi, type, false); |
60374688 JK |
762 | else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) |
763 | new_curseg(sbi, type, false); | |
351df4b2 JK |
764 | else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) |
765 | change_curseg(sbi, type, true); | |
766 | else | |
767 | new_curseg(sbi, type, false); | |
dcdfff65 JK |
768 | |
769 | stat_inc_seg_type(sbi, curseg); | |
351df4b2 JK |
770 | } |
771 | ||
772 | void allocate_new_segments(struct f2fs_sb_info *sbi) | |
773 | { | |
774 | struct curseg_info *curseg; | |
775 | unsigned int old_curseg; | |
776 | int i; | |
777 | ||
778 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
779 | curseg = CURSEG_I(sbi, i); | |
780 | old_curseg = curseg->segno; | |
781 | SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); | |
782 | locate_dirty_segment(sbi, old_curseg); | |
783 | } | |
784 | } | |
785 | ||
786 | static const struct segment_allocation default_salloc_ops = { | |
787 | .allocate_segment = allocate_segment_by_default, | |
788 | }; | |
789 | ||
351df4b2 JK |
790 | static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) |
791 | { | |
792 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
793 | if (curseg->next_blkoff < sbi->blocks_per_seg) | |
794 | return true; | |
795 | return false; | |
796 | } | |
797 | ||
798 | static int __get_segment_type_2(struct page *page, enum page_type p_type) | |
799 | { | |
800 | if (p_type == DATA) | |
801 | return CURSEG_HOT_DATA; | |
802 | else | |
803 | return CURSEG_HOT_NODE; | |
804 | } | |
805 | ||
806 | static int __get_segment_type_4(struct page *page, enum page_type p_type) | |
807 | { | |
808 | if (p_type == DATA) { | |
809 | struct inode *inode = page->mapping->host; | |
810 | ||
811 | if (S_ISDIR(inode->i_mode)) | |
812 | return CURSEG_HOT_DATA; | |
813 | else | |
814 | return CURSEG_COLD_DATA; | |
815 | } else { | |
816 | if (IS_DNODE(page) && !is_cold_node(page)) | |
817 | return CURSEG_HOT_NODE; | |
818 | else | |
819 | return CURSEG_COLD_NODE; | |
820 | } | |
821 | } | |
822 | ||
823 | static int __get_segment_type_6(struct page *page, enum page_type p_type) | |
824 | { | |
825 | if (p_type == DATA) { | |
826 | struct inode *inode = page->mapping->host; | |
827 | ||
828 | if (S_ISDIR(inode->i_mode)) | |
829 | return CURSEG_HOT_DATA; | |
354a3399 | 830 | else if (is_cold_data(page) || file_is_cold(inode)) |
351df4b2 JK |
831 | return CURSEG_COLD_DATA; |
832 | else | |
833 | return CURSEG_WARM_DATA; | |
834 | } else { | |
835 | if (IS_DNODE(page)) | |
836 | return is_cold_node(page) ? CURSEG_WARM_NODE : | |
837 | CURSEG_HOT_NODE; | |
838 | else | |
839 | return CURSEG_COLD_NODE; | |
840 | } | |
841 | } | |
842 | ||
843 | static int __get_segment_type(struct page *page, enum page_type p_type) | |
844 | { | |
845 | struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb); | |
846 | switch (sbi->active_logs) { | |
847 | case 2: | |
848 | return __get_segment_type_2(page, p_type); | |
849 | case 4: | |
850 | return __get_segment_type_4(page, p_type); | |
351df4b2 | 851 | } |
12a67146 | 852 | /* NR_CURSEG_TYPE(6) logs by default */ |
5d56b671 | 853 | f2fs_bug_on(sbi->active_logs != NR_CURSEG_TYPE); |
12a67146 | 854 | return __get_segment_type_6(page, p_type); |
351df4b2 JK |
855 | } |
856 | ||
bfad7c2d JK |
857 | void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, |
858 | block_t old_blkaddr, block_t *new_blkaddr, | |
859 | struct f2fs_summary *sum, int type) | |
351df4b2 JK |
860 | { |
861 | struct sit_info *sit_i = SIT_I(sbi); | |
862 | struct curseg_info *curseg; | |
863 | unsigned int old_cursegno; | |
351df4b2 | 864 | |
351df4b2 JK |
865 | curseg = CURSEG_I(sbi, type); |
866 | ||
867 | mutex_lock(&curseg->curseg_mutex); | |
868 | ||
869 | *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); | |
870 | old_cursegno = curseg->segno; | |
871 | ||
872 | /* | |
873 | * __add_sum_entry should be resided under the curseg_mutex | |
874 | * because, this function updates a summary entry in the | |
875 | * current summary block. | |
876 | */ | |
e79efe3b | 877 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
878 | |
879 | mutex_lock(&sit_i->sentry_lock); | |
880 | __refresh_next_blkoff(sbi, curseg); | |
dcdfff65 JK |
881 | |
882 | stat_inc_block_count(sbi, curseg); | |
351df4b2 | 883 | |
5e443818 JK |
884 | if (!__has_curseg_space(sbi, type)) |
885 | sit_i->s_ops->allocate_segment(sbi, type, false); | |
351df4b2 JK |
886 | /* |
887 | * SIT information should be updated before segment allocation, | |
888 | * since SSR needs latest valid block information. | |
889 | */ | |
890 | refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); | |
351df4b2 | 891 | locate_dirty_segment(sbi, old_cursegno); |
5e443818 | 892 | |
351df4b2 JK |
893 | mutex_unlock(&sit_i->sentry_lock); |
894 | ||
bfad7c2d | 895 | if (page && IS_NODESEG(type)) |
351df4b2 JK |
896 | fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); |
897 | ||
bfad7c2d JK |
898 | mutex_unlock(&curseg->curseg_mutex); |
899 | } | |
900 | ||
901 | static void do_write_page(struct f2fs_sb_info *sbi, struct page *page, | |
902 | block_t old_blkaddr, block_t *new_blkaddr, | |
903 | struct f2fs_summary *sum, struct f2fs_io_info *fio) | |
904 | { | |
905 | int type = __get_segment_type(page, fio->type); | |
906 | ||
907 | allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type); | |
908 | ||
351df4b2 | 909 | /* writeout dirty page into bdev */ |
458e6197 | 910 | f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio); |
351df4b2 JK |
911 | } |
912 | ||
577e3495 | 913 | void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) |
351df4b2 | 914 | { |
458e6197 JK |
915 | struct f2fs_io_info fio = { |
916 | .type = META, | |
7e8f2308 | 917 | .rw = WRITE_SYNC | REQ_META | REQ_PRIO |
458e6197 JK |
918 | }; |
919 | ||
351df4b2 | 920 | set_page_writeback(page); |
458e6197 | 921 | f2fs_submit_page_mbio(sbi, page, page->index, &fio); |
351df4b2 JK |
922 | } |
923 | ||
924 | void write_node_page(struct f2fs_sb_info *sbi, struct page *page, | |
fb5566da | 925 | struct f2fs_io_info *fio, |
351df4b2 JK |
926 | unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr) |
927 | { | |
928 | struct f2fs_summary sum; | |
929 | set_summary(&sum, nid, 0, 0); | |
fb5566da | 930 | do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio); |
351df4b2 JK |
931 | } |
932 | ||
458e6197 JK |
933 | void write_data_page(struct page *page, struct dnode_of_data *dn, |
934 | block_t *new_blkaddr, struct f2fs_io_info *fio) | |
351df4b2 | 935 | { |
458e6197 | 936 | struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); |
351df4b2 JK |
937 | struct f2fs_summary sum; |
938 | struct node_info ni; | |
939 | ||
458e6197 | 940 | f2fs_bug_on(dn->data_blkaddr == NULL_ADDR); |
351df4b2 JK |
941 | get_node_info(sbi, dn->nid, &ni); |
942 | set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); | |
943 | ||
458e6197 | 944 | do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio); |
351df4b2 JK |
945 | } |
946 | ||
6c311ec6 CF |
947 | void rewrite_data_page(struct page *page, block_t old_blkaddr, |
948 | struct f2fs_io_info *fio) | |
351df4b2 | 949 | { |
458e6197 JK |
950 | struct inode *inode = page->mapping->host; |
951 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
952 | f2fs_submit_page_mbio(sbi, page, old_blkaddr, fio); | |
351df4b2 JK |
953 | } |
954 | ||
955 | void recover_data_page(struct f2fs_sb_info *sbi, | |
956 | struct page *page, struct f2fs_summary *sum, | |
957 | block_t old_blkaddr, block_t new_blkaddr) | |
958 | { | |
959 | struct sit_info *sit_i = SIT_I(sbi); | |
960 | struct curseg_info *curseg; | |
961 | unsigned int segno, old_cursegno; | |
962 | struct seg_entry *se; | |
963 | int type; | |
964 | ||
965 | segno = GET_SEGNO(sbi, new_blkaddr); | |
966 | se = get_seg_entry(sbi, segno); | |
967 | type = se->type; | |
968 | ||
969 | if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { | |
970 | if (old_blkaddr == NULL_ADDR) | |
971 | type = CURSEG_COLD_DATA; | |
972 | else | |
973 | type = CURSEG_WARM_DATA; | |
974 | } | |
975 | curseg = CURSEG_I(sbi, type); | |
976 | ||
977 | mutex_lock(&curseg->curseg_mutex); | |
978 | mutex_lock(&sit_i->sentry_lock); | |
979 | ||
980 | old_cursegno = curseg->segno; | |
981 | ||
982 | /* change the current segment */ | |
983 | if (segno != curseg->segno) { | |
984 | curseg->next_segno = segno; | |
985 | change_curseg(sbi, type, true); | |
986 | } | |
987 | ||
491c0854 | 988 | curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); |
e79efe3b | 989 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
990 | |
991 | refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); | |
351df4b2 | 992 | locate_dirty_segment(sbi, old_cursegno); |
351df4b2 JK |
993 | |
994 | mutex_unlock(&sit_i->sentry_lock); | |
995 | mutex_unlock(&curseg->curseg_mutex); | |
996 | } | |
997 | ||
998 | void rewrite_node_page(struct f2fs_sb_info *sbi, | |
999 | struct page *page, struct f2fs_summary *sum, | |
1000 | block_t old_blkaddr, block_t new_blkaddr) | |
1001 | { | |
1002 | struct sit_info *sit_i = SIT_I(sbi); | |
1003 | int type = CURSEG_WARM_NODE; | |
1004 | struct curseg_info *curseg; | |
1005 | unsigned int segno, old_cursegno; | |
1006 | block_t next_blkaddr = next_blkaddr_of_node(page); | |
1007 | unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr); | |
458e6197 JK |
1008 | struct f2fs_io_info fio = { |
1009 | .type = NODE, | |
1010 | .rw = WRITE_SYNC, | |
458e6197 | 1011 | }; |
351df4b2 JK |
1012 | |
1013 | curseg = CURSEG_I(sbi, type); | |
1014 | ||
1015 | mutex_lock(&curseg->curseg_mutex); | |
1016 | mutex_lock(&sit_i->sentry_lock); | |
1017 | ||
1018 | segno = GET_SEGNO(sbi, new_blkaddr); | |
1019 | old_cursegno = curseg->segno; | |
1020 | ||
1021 | /* change the current segment */ | |
1022 | if (segno != curseg->segno) { | |
1023 | curseg->next_segno = segno; | |
1024 | change_curseg(sbi, type, true); | |
1025 | } | |
491c0854 | 1026 | curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); |
e79efe3b | 1027 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
1028 | |
1029 | /* change the current log to the next block addr in advance */ | |
1030 | if (next_segno != segno) { | |
1031 | curseg->next_segno = next_segno; | |
1032 | change_curseg(sbi, type, true); | |
1033 | } | |
491c0854 | 1034 | curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, next_blkaddr); |
351df4b2 JK |
1035 | |
1036 | /* rewrite node page */ | |
1037 | set_page_writeback(page); | |
458e6197 JK |
1038 | f2fs_submit_page_mbio(sbi, page, new_blkaddr, &fio); |
1039 | f2fs_submit_merged_bio(sbi, NODE, WRITE); | |
351df4b2 | 1040 | refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); |
351df4b2 | 1041 | locate_dirty_segment(sbi, old_cursegno); |
351df4b2 JK |
1042 | |
1043 | mutex_unlock(&sit_i->sentry_lock); | |
1044 | mutex_unlock(&curseg->curseg_mutex); | |
1045 | } | |
1046 | ||
df0f8dc0 CY |
1047 | static inline bool is_merged_page(struct f2fs_sb_info *sbi, |
1048 | struct page *page, enum page_type type) | |
1049 | { | |
1050 | enum page_type btype = PAGE_TYPE_OF_BIO(type); | |
1051 | struct f2fs_bio_info *io = &sbi->write_io[btype]; | |
1052 | struct bio *bio = io->bio; | |
1053 | struct bio_vec *bvec; | |
1054 | int i; | |
1055 | ||
1056 | down_read(&io->io_rwsem); | |
1057 | if (!bio) | |
1058 | goto out; | |
1059 | ||
1060 | bio_for_each_segment_all(bvec, bio, i) { | |
1061 | if (page == bvec->bv_page) { | |
1062 | up_read(&io->io_rwsem); | |
1063 | return true; | |
1064 | } | |
1065 | } | |
1066 | ||
1067 | out: | |
1068 | up_read(&io->io_rwsem); | |
1069 | return false; | |
1070 | } | |
1071 | ||
93dfe2ac | 1072 | void f2fs_wait_on_page_writeback(struct page *page, |
5514f0aa | 1073 | enum page_type type) |
93dfe2ac JK |
1074 | { |
1075 | struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb); | |
1076 | if (PageWriteback(page)) { | |
df0f8dc0 CY |
1077 | if (is_merged_page(sbi, page, type)) |
1078 | f2fs_submit_merged_bio(sbi, type, WRITE); | |
93dfe2ac JK |
1079 | wait_on_page_writeback(page); |
1080 | } | |
1081 | } | |
1082 | ||
351df4b2 JK |
1083 | static int read_compacted_summaries(struct f2fs_sb_info *sbi) |
1084 | { | |
1085 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1086 | struct curseg_info *seg_i; | |
1087 | unsigned char *kaddr; | |
1088 | struct page *page; | |
1089 | block_t start; | |
1090 | int i, j, offset; | |
1091 | ||
1092 | start = start_sum_block(sbi); | |
1093 | ||
1094 | page = get_meta_page(sbi, start++); | |
1095 | kaddr = (unsigned char *)page_address(page); | |
1096 | ||
1097 | /* Step 1: restore nat cache */ | |
1098 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
1099 | memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE); | |
1100 | ||
1101 | /* Step 2: restore sit cache */ | |
1102 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1103 | memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE, | |
1104 | SUM_JOURNAL_SIZE); | |
1105 | offset = 2 * SUM_JOURNAL_SIZE; | |
1106 | ||
1107 | /* Step 3: restore summary entries */ | |
1108 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1109 | unsigned short blk_off; | |
1110 | unsigned int segno; | |
1111 | ||
1112 | seg_i = CURSEG_I(sbi, i); | |
1113 | segno = le32_to_cpu(ckpt->cur_data_segno[i]); | |
1114 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); | |
1115 | seg_i->next_segno = segno; | |
1116 | reset_curseg(sbi, i, 0); | |
1117 | seg_i->alloc_type = ckpt->alloc_type[i]; | |
1118 | seg_i->next_blkoff = blk_off; | |
1119 | ||
1120 | if (seg_i->alloc_type == SSR) | |
1121 | blk_off = sbi->blocks_per_seg; | |
1122 | ||
1123 | for (j = 0; j < blk_off; j++) { | |
1124 | struct f2fs_summary *s; | |
1125 | s = (struct f2fs_summary *)(kaddr + offset); | |
1126 | seg_i->sum_blk->entries[j] = *s; | |
1127 | offset += SUMMARY_SIZE; | |
1128 | if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE - | |
1129 | SUM_FOOTER_SIZE) | |
1130 | continue; | |
1131 | ||
1132 | f2fs_put_page(page, 1); | |
1133 | page = NULL; | |
1134 | ||
1135 | page = get_meta_page(sbi, start++); | |
1136 | kaddr = (unsigned char *)page_address(page); | |
1137 | offset = 0; | |
1138 | } | |
1139 | } | |
1140 | f2fs_put_page(page, 1); | |
1141 | return 0; | |
1142 | } | |
1143 | ||
1144 | static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) | |
1145 | { | |
1146 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1147 | struct f2fs_summary_block *sum; | |
1148 | struct curseg_info *curseg; | |
1149 | struct page *new; | |
1150 | unsigned short blk_off; | |
1151 | unsigned int segno = 0; | |
1152 | block_t blk_addr = 0; | |
1153 | ||
1154 | /* get segment number and block addr */ | |
1155 | if (IS_DATASEG(type)) { | |
1156 | segno = le32_to_cpu(ckpt->cur_data_segno[type]); | |
1157 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - | |
1158 | CURSEG_HOT_DATA]); | |
25ca923b | 1159 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) |
351df4b2 JK |
1160 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); |
1161 | else | |
1162 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); | |
1163 | } else { | |
1164 | segno = le32_to_cpu(ckpt->cur_node_segno[type - | |
1165 | CURSEG_HOT_NODE]); | |
1166 | blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - | |
1167 | CURSEG_HOT_NODE]); | |
25ca923b | 1168 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) |
351df4b2 JK |
1169 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, |
1170 | type - CURSEG_HOT_NODE); | |
1171 | else | |
1172 | blk_addr = GET_SUM_BLOCK(sbi, segno); | |
1173 | } | |
1174 | ||
1175 | new = get_meta_page(sbi, blk_addr); | |
1176 | sum = (struct f2fs_summary_block *)page_address(new); | |
1177 | ||
1178 | if (IS_NODESEG(type)) { | |
25ca923b | 1179 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) { |
351df4b2 JK |
1180 | struct f2fs_summary *ns = &sum->entries[0]; |
1181 | int i; | |
1182 | for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { | |
1183 | ns->version = 0; | |
1184 | ns->ofs_in_node = 0; | |
1185 | } | |
1186 | } else { | |
d653788a GZ |
1187 | int err; |
1188 | ||
1189 | err = restore_node_summary(sbi, segno, sum); | |
1190 | if (err) { | |
351df4b2 | 1191 | f2fs_put_page(new, 1); |
d653788a | 1192 | return err; |
351df4b2 JK |
1193 | } |
1194 | } | |
1195 | } | |
1196 | ||
1197 | /* set uncompleted segment to curseg */ | |
1198 | curseg = CURSEG_I(sbi, type); | |
1199 | mutex_lock(&curseg->curseg_mutex); | |
1200 | memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE); | |
1201 | curseg->next_segno = segno; | |
1202 | reset_curseg(sbi, type, 0); | |
1203 | curseg->alloc_type = ckpt->alloc_type[type]; | |
1204 | curseg->next_blkoff = blk_off; | |
1205 | mutex_unlock(&curseg->curseg_mutex); | |
1206 | f2fs_put_page(new, 1); | |
1207 | return 0; | |
1208 | } | |
1209 | ||
1210 | static int restore_curseg_summaries(struct f2fs_sb_info *sbi) | |
1211 | { | |
1212 | int type = CURSEG_HOT_DATA; | |
e4fc5fbf | 1213 | int err; |
351df4b2 | 1214 | |
25ca923b | 1215 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) { |
351df4b2 JK |
1216 | /* restore for compacted data summary */ |
1217 | if (read_compacted_summaries(sbi)) | |
1218 | return -EINVAL; | |
1219 | type = CURSEG_HOT_NODE; | |
1220 | } | |
1221 | ||
e4fc5fbf CY |
1222 | for (; type <= CURSEG_COLD_NODE; type++) { |
1223 | err = read_normal_summaries(sbi, type); | |
1224 | if (err) | |
1225 | return err; | |
1226 | } | |
1227 | ||
351df4b2 JK |
1228 | return 0; |
1229 | } | |
1230 | ||
1231 | static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) | |
1232 | { | |
1233 | struct page *page; | |
1234 | unsigned char *kaddr; | |
1235 | struct f2fs_summary *summary; | |
1236 | struct curseg_info *seg_i; | |
1237 | int written_size = 0; | |
1238 | int i, j; | |
1239 | ||
1240 | page = grab_meta_page(sbi, blkaddr++); | |
1241 | kaddr = (unsigned char *)page_address(page); | |
1242 | ||
1243 | /* Step 1: write nat cache */ | |
1244 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
1245 | memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE); | |
1246 | written_size += SUM_JOURNAL_SIZE; | |
1247 | ||
1248 | /* Step 2: write sit cache */ | |
1249 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1250 | memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits, | |
1251 | SUM_JOURNAL_SIZE); | |
1252 | written_size += SUM_JOURNAL_SIZE; | |
1253 | ||
351df4b2 JK |
1254 | /* Step 3: write summary entries */ |
1255 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1256 | unsigned short blkoff; | |
1257 | seg_i = CURSEG_I(sbi, i); | |
1258 | if (sbi->ckpt->alloc_type[i] == SSR) | |
1259 | blkoff = sbi->blocks_per_seg; | |
1260 | else | |
1261 | blkoff = curseg_blkoff(sbi, i); | |
1262 | ||
1263 | for (j = 0; j < blkoff; j++) { | |
1264 | if (!page) { | |
1265 | page = grab_meta_page(sbi, blkaddr++); | |
1266 | kaddr = (unsigned char *)page_address(page); | |
1267 | written_size = 0; | |
1268 | } | |
1269 | summary = (struct f2fs_summary *)(kaddr + written_size); | |
1270 | *summary = seg_i->sum_blk->entries[j]; | |
1271 | written_size += SUMMARY_SIZE; | |
351df4b2 JK |
1272 | |
1273 | if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE - | |
1274 | SUM_FOOTER_SIZE) | |
1275 | continue; | |
1276 | ||
e8d61a74 | 1277 | set_page_dirty(page); |
351df4b2 JK |
1278 | f2fs_put_page(page, 1); |
1279 | page = NULL; | |
1280 | } | |
1281 | } | |
e8d61a74 CY |
1282 | if (page) { |
1283 | set_page_dirty(page); | |
351df4b2 | 1284 | f2fs_put_page(page, 1); |
e8d61a74 | 1285 | } |
351df4b2 JK |
1286 | } |
1287 | ||
1288 | static void write_normal_summaries(struct f2fs_sb_info *sbi, | |
1289 | block_t blkaddr, int type) | |
1290 | { | |
1291 | int i, end; | |
1292 | if (IS_DATASEG(type)) | |
1293 | end = type + NR_CURSEG_DATA_TYPE; | |
1294 | else | |
1295 | end = type + NR_CURSEG_NODE_TYPE; | |
1296 | ||
1297 | for (i = type; i < end; i++) { | |
1298 | struct curseg_info *sum = CURSEG_I(sbi, i); | |
1299 | mutex_lock(&sum->curseg_mutex); | |
1300 | write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type)); | |
1301 | mutex_unlock(&sum->curseg_mutex); | |
1302 | } | |
1303 | } | |
1304 | ||
1305 | void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1306 | { | |
25ca923b | 1307 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) |
351df4b2 JK |
1308 | write_compacted_summaries(sbi, start_blk); |
1309 | else | |
1310 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); | |
1311 | } | |
1312 | ||
1313 | void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1314 | { | |
25ca923b | 1315 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG)) |
351df4b2 | 1316 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); |
351df4b2 JK |
1317 | } |
1318 | ||
1319 | int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type, | |
1320 | unsigned int val, int alloc) | |
1321 | { | |
1322 | int i; | |
1323 | ||
1324 | if (type == NAT_JOURNAL) { | |
1325 | for (i = 0; i < nats_in_cursum(sum); i++) { | |
1326 | if (le32_to_cpu(nid_in_journal(sum, i)) == val) | |
1327 | return i; | |
1328 | } | |
1329 | if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) | |
1330 | return update_nats_in_cursum(sum, 1); | |
1331 | } else if (type == SIT_JOURNAL) { | |
1332 | for (i = 0; i < sits_in_cursum(sum); i++) | |
1333 | if (le32_to_cpu(segno_in_journal(sum, i)) == val) | |
1334 | return i; | |
1335 | if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES) | |
1336 | return update_sits_in_cursum(sum, 1); | |
1337 | } | |
1338 | return -1; | |
1339 | } | |
1340 | ||
1341 | static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, | |
1342 | unsigned int segno) | |
1343 | { | |
1344 | struct sit_info *sit_i = SIT_I(sbi); | |
1345 | unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno); | |
1346 | block_t blk_addr = sit_i->sit_base_addr + offset; | |
1347 | ||
1348 | check_seg_range(sbi, segno); | |
1349 | ||
1350 | /* calculate sit block address */ | |
1351 | if (f2fs_test_bit(offset, sit_i->sit_bitmap)) | |
1352 | blk_addr += sit_i->sit_blocks; | |
1353 | ||
1354 | return get_meta_page(sbi, blk_addr); | |
1355 | } | |
1356 | ||
1357 | static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, | |
1358 | unsigned int start) | |
1359 | { | |
1360 | struct sit_info *sit_i = SIT_I(sbi); | |
1361 | struct page *src_page, *dst_page; | |
1362 | pgoff_t src_off, dst_off; | |
1363 | void *src_addr, *dst_addr; | |
1364 | ||
1365 | src_off = current_sit_addr(sbi, start); | |
1366 | dst_off = next_sit_addr(sbi, src_off); | |
1367 | ||
1368 | /* get current sit block page without lock */ | |
1369 | src_page = get_meta_page(sbi, src_off); | |
1370 | dst_page = grab_meta_page(sbi, dst_off); | |
5d56b671 | 1371 | f2fs_bug_on(PageDirty(src_page)); |
351df4b2 JK |
1372 | |
1373 | src_addr = page_address(src_page); | |
1374 | dst_addr = page_address(dst_page); | |
1375 | memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); | |
1376 | ||
1377 | set_page_dirty(dst_page); | |
1378 | f2fs_put_page(src_page, 1); | |
1379 | ||
1380 | set_to_next_sit(sit_i, start); | |
1381 | ||
1382 | return dst_page; | |
1383 | } | |
1384 | ||
1385 | static bool flush_sits_in_journal(struct f2fs_sb_info *sbi) | |
1386 | { | |
1387 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1388 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
1389 | int i; | |
1390 | ||
1391 | /* | |
1392 | * If the journal area in the current summary is full of sit entries, | |
1393 | * all the sit entries will be flushed. Otherwise the sit entries | |
1394 | * are not able to replace with newly hot sit entries. | |
1395 | */ | |
1396 | if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) { | |
1397 | for (i = sits_in_cursum(sum) - 1; i >= 0; i--) { | |
1398 | unsigned int segno; | |
1399 | segno = le32_to_cpu(segno_in_journal(sum, i)); | |
1400 | __mark_sit_entry_dirty(sbi, segno); | |
1401 | } | |
1402 | update_sits_in_cursum(sum, -sits_in_cursum(sum)); | |
cffbfa66 | 1403 | return true; |
351df4b2 | 1404 | } |
cffbfa66 | 1405 | return false; |
351df4b2 JK |
1406 | } |
1407 | ||
0a8165d7 | 1408 | /* |
351df4b2 JK |
1409 | * CP calls this function, which flushes SIT entries including sit_journal, |
1410 | * and moves prefree segs to free segs. | |
1411 | */ | |
1412 | void flush_sit_entries(struct f2fs_sb_info *sbi) | |
1413 | { | |
1414 | struct sit_info *sit_i = SIT_I(sbi); | |
1415 | unsigned long *bitmap = sit_i->dirty_sentries_bitmap; | |
1416 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1417 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
1418 | unsigned long nsegs = TOTAL_SEGS(sbi); | |
1419 | struct page *page = NULL; | |
1420 | struct f2fs_sit_block *raw_sit = NULL; | |
1421 | unsigned int start = 0, end = 0; | |
1422 | unsigned int segno = -1; | |
1423 | bool flushed; | |
1424 | ||
1425 | mutex_lock(&curseg->curseg_mutex); | |
1426 | mutex_lock(&sit_i->sentry_lock); | |
1427 | ||
1428 | /* | |
1429 | * "flushed" indicates whether sit entries in journal are flushed | |
1430 | * to the SIT area or not. | |
1431 | */ | |
1432 | flushed = flush_sits_in_journal(sbi); | |
1433 | ||
1434 | while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) { | |
1435 | struct seg_entry *se = get_seg_entry(sbi, segno); | |
1436 | int sit_offset, offset; | |
1437 | ||
1438 | sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); | |
1439 | ||
b2955550 JK |
1440 | /* add discard candidates */ |
1441 | if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards) | |
1442 | add_discard_addrs(sbi, segno, se); | |
1443 | ||
351df4b2 JK |
1444 | if (flushed) |
1445 | goto to_sit_page; | |
1446 | ||
1447 | offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1); | |
1448 | if (offset >= 0) { | |
1449 | segno_in_journal(sum, offset) = cpu_to_le32(segno); | |
1450 | seg_info_to_raw_sit(se, &sit_in_journal(sum, offset)); | |
1451 | goto flush_done; | |
1452 | } | |
1453 | to_sit_page: | |
1454 | if (!page || (start > segno) || (segno > end)) { | |
1455 | if (page) { | |
1456 | f2fs_put_page(page, 1); | |
1457 | page = NULL; | |
1458 | } | |
1459 | ||
1460 | start = START_SEGNO(sit_i, segno); | |
1461 | end = start + SIT_ENTRY_PER_BLOCK - 1; | |
1462 | ||
1463 | /* read sit block that will be updated */ | |
1464 | page = get_next_sit_page(sbi, start); | |
1465 | raw_sit = page_address(page); | |
1466 | } | |
1467 | ||
1468 | /* udpate entry in SIT block */ | |
1469 | seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]); | |
1470 | flush_done: | |
1471 | __clear_bit(segno, bitmap); | |
1472 | sit_i->dirty_sentries--; | |
1473 | } | |
1474 | mutex_unlock(&sit_i->sentry_lock); | |
1475 | mutex_unlock(&curseg->curseg_mutex); | |
1476 | ||
1477 | /* writeout last modified SIT block */ | |
1478 | f2fs_put_page(page, 1); | |
1479 | ||
1480 | set_prefree_as_free_segments(sbi); | |
1481 | } | |
1482 | ||
1483 | static int build_sit_info(struct f2fs_sb_info *sbi) | |
1484 | { | |
1485 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
1486 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1487 | struct sit_info *sit_i; | |
1488 | unsigned int sit_segs, start; | |
1489 | char *src_bitmap, *dst_bitmap; | |
1490 | unsigned int bitmap_size; | |
1491 | ||
1492 | /* allocate memory for SIT information */ | |
1493 | sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); | |
1494 | if (!sit_i) | |
1495 | return -ENOMEM; | |
1496 | ||
1497 | SM_I(sbi)->sit_info = sit_i; | |
1498 | ||
1499 | sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry)); | |
1500 | if (!sit_i->sentries) | |
1501 | return -ENOMEM; | |
1502 | ||
1503 | bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); | |
1504 | sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL); | |
1505 | if (!sit_i->dirty_sentries_bitmap) | |
1506 | return -ENOMEM; | |
1507 | ||
1508 | for (start = 0; start < TOTAL_SEGS(sbi); start++) { | |
1509 | sit_i->sentries[start].cur_valid_map | |
1510 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
1511 | sit_i->sentries[start].ckpt_valid_map | |
1512 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
1513 | if (!sit_i->sentries[start].cur_valid_map | |
1514 | || !sit_i->sentries[start].ckpt_valid_map) | |
1515 | return -ENOMEM; | |
1516 | } | |
1517 | ||
1518 | if (sbi->segs_per_sec > 1) { | |
53cf9522 | 1519 | sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) * |
351df4b2 JK |
1520 | sizeof(struct sec_entry)); |
1521 | if (!sit_i->sec_entries) | |
1522 | return -ENOMEM; | |
1523 | } | |
1524 | ||
1525 | /* get information related with SIT */ | |
1526 | sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; | |
1527 | ||
1528 | /* setup SIT bitmap from ckeckpoint pack */ | |
1529 | bitmap_size = __bitmap_size(sbi, SIT_BITMAP); | |
1530 | src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); | |
1531 | ||
79b5793b | 1532 | dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); |
351df4b2 JK |
1533 | if (!dst_bitmap) |
1534 | return -ENOMEM; | |
351df4b2 JK |
1535 | |
1536 | /* init SIT information */ | |
1537 | sit_i->s_ops = &default_salloc_ops; | |
1538 | ||
1539 | sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); | |
1540 | sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; | |
1541 | sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count); | |
1542 | sit_i->sit_bitmap = dst_bitmap; | |
1543 | sit_i->bitmap_size = bitmap_size; | |
1544 | sit_i->dirty_sentries = 0; | |
1545 | sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; | |
1546 | sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); | |
1547 | sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; | |
1548 | mutex_init(&sit_i->sentry_lock); | |
1549 | return 0; | |
1550 | } | |
1551 | ||
1552 | static int build_free_segmap(struct f2fs_sb_info *sbi) | |
1553 | { | |
1554 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
1555 | struct free_segmap_info *free_i; | |
1556 | unsigned int bitmap_size, sec_bitmap_size; | |
1557 | ||
1558 | /* allocate memory for free segmap information */ | |
1559 | free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); | |
1560 | if (!free_i) | |
1561 | return -ENOMEM; | |
1562 | ||
1563 | SM_I(sbi)->free_info = free_i; | |
1564 | ||
1565 | bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); | |
1566 | free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL); | |
1567 | if (!free_i->free_segmap) | |
1568 | return -ENOMEM; | |
1569 | ||
53cf9522 | 1570 | sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi)); |
351df4b2 JK |
1571 | free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL); |
1572 | if (!free_i->free_secmap) | |
1573 | return -ENOMEM; | |
1574 | ||
1575 | /* set all segments as dirty temporarily */ | |
1576 | memset(free_i->free_segmap, 0xff, bitmap_size); | |
1577 | memset(free_i->free_secmap, 0xff, sec_bitmap_size); | |
1578 | ||
1579 | /* init free segmap information */ | |
1580 | free_i->start_segno = | |
1581 | (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr); | |
1582 | free_i->free_segments = 0; | |
1583 | free_i->free_sections = 0; | |
1584 | rwlock_init(&free_i->segmap_lock); | |
1585 | return 0; | |
1586 | } | |
1587 | ||
1588 | static int build_curseg(struct f2fs_sb_info *sbi) | |
1589 | { | |
1042d60f | 1590 | struct curseg_info *array; |
351df4b2 JK |
1591 | int i; |
1592 | ||
1593 | array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL); | |
1594 | if (!array) | |
1595 | return -ENOMEM; | |
1596 | ||
1597 | SM_I(sbi)->curseg_array = array; | |
1598 | ||
1599 | for (i = 0; i < NR_CURSEG_TYPE; i++) { | |
1600 | mutex_init(&array[i].curseg_mutex); | |
1601 | array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL); | |
1602 | if (!array[i].sum_blk) | |
1603 | return -ENOMEM; | |
1604 | array[i].segno = NULL_SEGNO; | |
1605 | array[i].next_blkoff = 0; | |
1606 | } | |
1607 | return restore_curseg_summaries(sbi); | |
1608 | } | |
1609 | ||
1610 | static void build_sit_entries(struct f2fs_sb_info *sbi) | |
1611 | { | |
1612 | struct sit_info *sit_i = SIT_I(sbi); | |
1613 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1614 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
74de593a CY |
1615 | int sit_blk_cnt = SIT_BLK_CNT(sbi); |
1616 | unsigned int i, start, end; | |
1617 | unsigned int readed, start_blk = 0; | |
1618 | int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi)); | |
351df4b2 | 1619 | |
74de593a | 1620 | do { |
662befda | 1621 | readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT); |
74de593a CY |
1622 | |
1623 | start = start_blk * sit_i->sents_per_block; | |
1624 | end = (start_blk + readed) * sit_i->sents_per_block; | |
1625 | ||
1626 | for (; start < end && start < TOTAL_SEGS(sbi); start++) { | |
1627 | struct seg_entry *se = &sit_i->sentries[start]; | |
1628 | struct f2fs_sit_block *sit_blk; | |
1629 | struct f2fs_sit_entry sit; | |
1630 | struct page *page; | |
1631 | ||
1632 | mutex_lock(&curseg->curseg_mutex); | |
1633 | for (i = 0; i < sits_in_cursum(sum); i++) { | |
6c311ec6 CF |
1634 | if (le32_to_cpu(segno_in_journal(sum, i)) |
1635 | == start) { | |
74de593a CY |
1636 | sit = sit_in_journal(sum, i); |
1637 | mutex_unlock(&curseg->curseg_mutex); | |
1638 | goto got_it; | |
1639 | } | |
351df4b2 | 1640 | } |
74de593a CY |
1641 | mutex_unlock(&curseg->curseg_mutex); |
1642 | ||
1643 | page = get_current_sit_page(sbi, start); | |
1644 | sit_blk = (struct f2fs_sit_block *)page_address(page); | |
1645 | sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; | |
1646 | f2fs_put_page(page, 1); | |
351df4b2 | 1647 | got_it: |
74de593a CY |
1648 | check_block_count(sbi, start, &sit); |
1649 | seg_info_from_raw_sit(se, &sit); | |
1650 | if (sbi->segs_per_sec > 1) { | |
1651 | struct sec_entry *e = get_sec_entry(sbi, start); | |
1652 | e->valid_blocks += se->valid_blocks; | |
1653 | } | |
351df4b2 | 1654 | } |
74de593a CY |
1655 | start_blk += readed; |
1656 | } while (start_blk < sit_blk_cnt); | |
351df4b2 JK |
1657 | } |
1658 | ||
1659 | static void init_free_segmap(struct f2fs_sb_info *sbi) | |
1660 | { | |
1661 | unsigned int start; | |
1662 | int type; | |
1663 | ||
1664 | for (start = 0; start < TOTAL_SEGS(sbi); start++) { | |
1665 | struct seg_entry *sentry = get_seg_entry(sbi, start); | |
1666 | if (!sentry->valid_blocks) | |
1667 | __set_free(sbi, start); | |
1668 | } | |
1669 | ||
1670 | /* set use the current segments */ | |
1671 | for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { | |
1672 | struct curseg_info *curseg_t = CURSEG_I(sbi, type); | |
1673 | __set_test_and_inuse(sbi, curseg_t->segno); | |
1674 | } | |
1675 | } | |
1676 | ||
1677 | static void init_dirty_segmap(struct f2fs_sb_info *sbi) | |
1678 | { | |
1679 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1680 | struct free_segmap_info *free_i = FREE_I(sbi); | |
8736fbf0 | 1681 | unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi); |
351df4b2 JK |
1682 | unsigned short valid_blocks; |
1683 | ||
8736fbf0 | 1684 | while (1) { |
351df4b2 | 1685 | /* find dirty segment based on free segmap */ |
8736fbf0 NJ |
1686 | segno = find_next_inuse(free_i, total_segs, offset); |
1687 | if (segno >= total_segs) | |
351df4b2 JK |
1688 | break; |
1689 | offset = segno + 1; | |
1690 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
1691 | if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks) | |
1692 | continue; | |
1693 | mutex_lock(&dirty_i->seglist_lock); | |
1694 | __locate_dirty_segment(sbi, segno, DIRTY); | |
1695 | mutex_unlock(&dirty_i->seglist_lock); | |
1696 | } | |
1697 | } | |
1698 | ||
5ec4e49f | 1699 | static int init_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
1700 | { |
1701 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 1702 | unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi)); |
351df4b2 | 1703 | |
5ec4e49f JK |
1704 | dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL); |
1705 | if (!dirty_i->victim_secmap) | |
351df4b2 JK |
1706 | return -ENOMEM; |
1707 | return 0; | |
1708 | } | |
1709 | ||
1710 | static int build_dirty_segmap(struct f2fs_sb_info *sbi) | |
1711 | { | |
1712 | struct dirty_seglist_info *dirty_i; | |
1713 | unsigned int bitmap_size, i; | |
1714 | ||
1715 | /* allocate memory for dirty segments list information */ | |
1716 | dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); | |
1717 | if (!dirty_i) | |
1718 | return -ENOMEM; | |
1719 | ||
1720 | SM_I(sbi)->dirty_info = dirty_i; | |
1721 | mutex_init(&dirty_i->seglist_lock); | |
1722 | ||
1723 | bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); | |
1724 | ||
1725 | for (i = 0; i < NR_DIRTY_TYPE; i++) { | |
1726 | dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL); | |
351df4b2 JK |
1727 | if (!dirty_i->dirty_segmap[i]) |
1728 | return -ENOMEM; | |
1729 | } | |
1730 | ||
1731 | init_dirty_segmap(sbi); | |
5ec4e49f | 1732 | return init_victim_secmap(sbi); |
351df4b2 JK |
1733 | } |
1734 | ||
0a8165d7 | 1735 | /* |
351df4b2 JK |
1736 | * Update min, max modified time for cost-benefit GC algorithm |
1737 | */ | |
1738 | static void init_min_max_mtime(struct f2fs_sb_info *sbi) | |
1739 | { | |
1740 | struct sit_info *sit_i = SIT_I(sbi); | |
1741 | unsigned int segno; | |
1742 | ||
1743 | mutex_lock(&sit_i->sentry_lock); | |
1744 | ||
1745 | sit_i->min_mtime = LLONG_MAX; | |
1746 | ||
1747 | for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) { | |
1748 | unsigned int i; | |
1749 | unsigned long long mtime = 0; | |
1750 | ||
1751 | for (i = 0; i < sbi->segs_per_sec; i++) | |
1752 | mtime += get_seg_entry(sbi, segno + i)->mtime; | |
1753 | ||
1754 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
1755 | ||
1756 | if (sit_i->min_mtime > mtime) | |
1757 | sit_i->min_mtime = mtime; | |
1758 | } | |
1759 | sit_i->max_mtime = get_mtime(sbi); | |
1760 | mutex_unlock(&sit_i->sentry_lock); | |
1761 | } | |
1762 | ||
1763 | int build_segment_manager(struct f2fs_sb_info *sbi) | |
1764 | { | |
1765 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
1766 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1042d60f | 1767 | struct f2fs_sm_info *sm_info; |
351df4b2 JK |
1768 | int err; |
1769 | ||
1770 | sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); | |
1771 | if (!sm_info) | |
1772 | return -ENOMEM; | |
1773 | ||
1774 | /* init sm info */ | |
1775 | sbi->sm_info = sm_info; | |
1776 | INIT_LIST_HEAD(&sm_info->wblist_head); | |
1777 | spin_lock_init(&sm_info->wblist_lock); | |
1778 | sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); | |
1779 | sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); | |
1780 | sm_info->segment_count = le32_to_cpu(raw_super->segment_count); | |
1781 | sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); | |
1782 | sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); | |
1783 | sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); | |
1784 | sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); | |
58c41035 JK |
1785 | sm_info->rec_prefree_segments = sm_info->main_segments * |
1786 | DEF_RECLAIM_PREFREE_SEGMENTS / 100; | |
216fbd64 JK |
1787 | sm_info->ipu_policy = F2FS_IPU_DISABLE; |
1788 | sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; | |
351df4b2 | 1789 | |
7fd9e544 JK |
1790 | INIT_LIST_HEAD(&sm_info->discard_list); |
1791 | sm_info->nr_discards = 0; | |
1792 | sm_info->max_discards = 0; | |
1793 | ||
351df4b2 JK |
1794 | err = build_sit_info(sbi); |
1795 | if (err) | |
1796 | return err; | |
1797 | err = build_free_segmap(sbi); | |
1798 | if (err) | |
1799 | return err; | |
1800 | err = build_curseg(sbi); | |
1801 | if (err) | |
1802 | return err; | |
1803 | ||
1804 | /* reinit free segmap based on SIT */ | |
1805 | build_sit_entries(sbi); | |
1806 | ||
1807 | init_free_segmap(sbi); | |
1808 | err = build_dirty_segmap(sbi); | |
1809 | if (err) | |
1810 | return err; | |
1811 | ||
1812 | init_min_max_mtime(sbi); | |
1813 | return 0; | |
1814 | } | |
1815 | ||
1816 | static void discard_dirty_segmap(struct f2fs_sb_info *sbi, | |
1817 | enum dirty_type dirty_type) | |
1818 | { | |
1819 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1820 | ||
1821 | mutex_lock(&dirty_i->seglist_lock); | |
1822 | kfree(dirty_i->dirty_segmap[dirty_type]); | |
1823 | dirty_i->nr_dirty[dirty_type] = 0; | |
1824 | mutex_unlock(&dirty_i->seglist_lock); | |
1825 | } | |
1826 | ||
5ec4e49f | 1827 | static void destroy_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
1828 | { |
1829 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 1830 | kfree(dirty_i->victim_secmap); |
351df4b2 JK |
1831 | } |
1832 | ||
1833 | static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) | |
1834 | { | |
1835 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1836 | int i; | |
1837 | ||
1838 | if (!dirty_i) | |
1839 | return; | |
1840 | ||
1841 | /* discard pre-free/dirty segments list */ | |
1842 | for (i = 0; i < NR_DIRTY_TYPE; i++) | |
1843 | discard_dirty_segmap(sbi, i); | |
1844 | ||
5ec4e49f | 1845 | destroy_victim_secmap(sbi); |
351df4b2 JK |
1846 | SM_I(sbi)->dirty_info = NULL; |
1847 | kfree(dirty_i); | |
1848 | } | |
1849 | ||
1850 | static void destroy_curseg(struct f2fs_sb_info *sbi) | |
1851 | { | |
1852 | struct curseg_info *array = SM_I(sbi)->curseg_array; | |
1853 | int i; | |
1854 | ||
1855 | if (!array) | |
1856 | return; | |
1857 | SM_I(sbi)->curseg_array = NULL; | |
1858 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
1859 | kfree(array[i].sum_blk); | |
1860 | kfree(array); | |
1861 | } | |
1862 | ||
1863 | static void destroy_free_segmap(struct f2fs_sb_info *sbi) | |
1864 | { | |
1865 | struct free_segmap_info *free_i = SM_I(sbi)->free_info; | |
1866 | if (!free_i) | |
1867 | return; | |
1868 | SM_I(sbi)->free_info = NULL; | |
1869 | kfree(free_i->free_segmap); | |
1870 | kfree(free_i->free_secmap); | |
1871 | kfree(free_i); | |
1872 | } | |
1873 | ||
1874 | static void destroy_sit_info(struct f2fs_sb_info *sbi) | |
1875 | { | |
1876 | struct sit_info *sit_i = SIT_I(sbi); | |
1877 | unsigned int start; | |
1878 | ||
1879 | if (!sit_i) | |
1880 | return; | |
1881 | ||
1882 | if (sit_i->sentries) { | |
1883 | for (start = 0; start < TOTAL_SEGS(sbi); start++) { | |
1884 | kfree(sit_i->sentries[start].cur_valid_map); | |
1885 | kfree(sit_i->sentries[start].ckpt_valid_map); | |
1886 | } | |
1887 | } | |
1888 | vfree(sit_i->sentries); | |
1889 | vfree(sit_i->sec_entries); | |
1890 | kfree(sit_i->dirty_sentries_bitmap); | |
1891 | ||
1892 | SM_I(sbi)->sit_info = NULL; | |
1893 | kfree(sit_i->sit_bitmap); | |
1894 | kfree(sit_i); | |
1895 | } | |
1896 | ||
1897 | void destroy_segment_manager(struct f2fs_sb_info *sbi) | |
1898 | { | |
1899 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
3b03f724 CY |
1900 | if (!sm_info) |
1901 | return; | |
351df4b2 JK |
1902 | destroy_dirty_segmap(sbi); |
1903 | destroy_curseg(sbi); | |
1904 | destroy_free_segmap(sbi); | |
1905 | destroy_sit_info(sbi); | |
1906 | sbi->sm_info = NULL; | |
1907 | kfree(sm_info); | |
1908 | } | |
7fd9e544 JK |
1909 | |
1910 | int __init create_segment_manager_caches(void) | |
1911 | { | |
1912 | discard_entry_slab = f2fs_kmem_cache_create("discard_entry", | |
e8512d2e | 1913 | sizeof(struct discard_entry)); |
7fd9e544 JK |
1914 | if (!discard_entry_slab) |
1915 | return -ENOMEM; | |
1916 | return 0; | |
1917 | } | |
1918 | ||
1919 | void destroy_segment_manager_caches(void) | |
1920 | { | |
1921 | kmem_cache_destroy(discard_entry_slab); | |
1922 | } |