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 JK |
16 | #include <linux/vmalloc.h> |
17 | ||
18 | #include "f2fs.h" | |
19 | #include "segment.h" | |
20 | #include "node.h" | |
6ec178da | 21 | #include <trace/events/f2fs.h> |
351df4b2 | 22 | |
0a8165d7 | 23 | /* |
351df4b2 JK |
24 | * This function balances dirty node and dentry pages. |
25 | * In addition, it controls garbage collection. | |
26 | */ | |
27 | void f2fs_balance_fs(struct f2fs_sb_info *sbi) | |
28 | { | |
351df4b2 | 29 | /* |
029cd28c JK |
30 | * We should do GC or end up with checkpoint, if there are so many dirty |
31 | * dir/node pages without enough free segments. | |
351df4b2 | 32 | */ |
43727527 | 33 | if (has_not_enough_free_secs(sbi, 0)) { |
351df4b2 | 34 | mutex_lock(&sbi->gc_mutex); |
408e9375 | 35 | f2fs_gc(sbi); |
351df4b2 JK |
36 | } |
37 | } | |
38 | ||
39 | static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, | |
40 | enum dirty_type dirty_type) | |
41 | { | |
42 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
43 | ||
44 | /* need not be added */ | |
45 | if (IS_CURSEG(sbi, segno)) | |
46 | return; | |
47 | ||
48 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
49 | dirty_i->nr_dirty[dirty_type]++; | |
50 | ||
51 | if (dirty_type == DIRTY) { | |
52 | struct seg_entry *sentry = get_seg_entry(sbi, segno); | |
b2f2c390 JK |
53 | enum dirty_type t = DIRTY_HOT_DATA; |
54 | ||
351df4b2 | 55 | dirty_type = sentry->type; |
b2f2c390 | 56 | |
351df4b2 JK |
57 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) |
58 | dirty_i->nr_dirty[dirty_type]++; | |
b2f2c390 JK |
59 | |
60 | /* Only one bitmap should be set */ | |
61 | for (; t <= DIRTY_COLD_NODE; t++) { | |
62 | if (t == dirty_type) | |
63 | continue; | |
64 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) | |
65 | dirty_i->nr_dirty[t]--; | |
66 | } | |
351df4b2 JK |
67 | } |
68 | } | |
69 | ||
70 | static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, | |
71 | enum dirty_type dirty_type) | |
72 | { | |
73 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
74 | ||
75 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
76 | dirty_i->nr_dirty[dirty_type]--; | |
77 | ||
78 | if (dirty_type == DIRTY) { | |
b2f2c390 JK |
79 | enum dirty_type t = DIRTY_HOT_DATA; |
80 | ||
81 | /* clear all the bitmaps */ | |
82 | for (; t <= DIRTY_COLD_NODE; t++) | |
83 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) | |
84 | dirty_i->nr_dirty[t]--; | |
85 | ||
5ec4e49f JK |
86 | if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0) |
87 | clear_bit(GET_SECNO(sbi, segno), | |
88 | dirty_i->victim_secmap); | |
351df4b2 JK |
89 | } |
90 | } | |
91 | ||
0a8165d7 | 92 | /* |
351df4b2 JK |
93 | * Should not occur error such as -ENOMEM. |
94 | * Adding dirty entry into seglist is not critical operation. | |
95 | * If a given segment is one of current working segments, it won't be added. | |
96 | */ | |
8d8451af | 97 | static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) |
351df4b2 JK |
98 | { |
99 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
100 | unsigned short valid_blocks; | |
101 | ||
102 | if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) | |
103 | return; | |
104 | ||
105 | mutex_lock(&dirty_i->seglist_lock); | |
106 | ||
107 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
108 | ||
109 | if (valid_blocks == 0) { | |
110 | __locate_dirty_segment(sbi, segno, PRE); | |
111 | __remove_dirty_segment(sbi, segno, DIRTY); | |
112 | } else if (valid_blocks < sbi->blocks_per_seg) { | |
113 | __locate_dirty_segment(sbi, segno, DIRTY); | |
114 | } else { | |
115 | /* Recovery routine with SSR needs this */ | |
116 | __remove_dirty_segment(sbi, segno, DIRTY); | |
117 | } | |
118 | ||
119 | mutex_unlock(&dirty_i->seglist_lock); | |
351df4b2 JK |
120 | } |
121 | ||
0a8165d7 | 122 | /* |
351df4b2 JK |
123 | * Should call clear_prefree_segments after checkpoint is done. |
124 | */ | |
125 | static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) | |
126 | { | |
127 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
81fb5e87 | 128 | unsigned int segno = -1; |
351df4b2 JK |
129 | unsigned int total_segs = TOTAL_SEGS(sbi); |
130 | ||
131 | mutex_lock(&dirty_i->seglist_lock); | |
132 | while (1) { | |
133 | segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs, | |
81fb5e87 | 134 | segno + 1); |
351df4b2 JK |
135 | if (segno >= total_segs) |
136 | break; | |
137 | __set_test_and_free(sbi, segno); | |
351df4b2 JK |
138 | } |
139 | mutex_unlock(&dirty_i->seglist_lock); | |
140 | } | |
141 | ||
142 | void clear_prefree_segments(struct f2fs_sb_info *sbi) | |
143 | { | |
144 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
81fb5e87 | 145 | unsigned int segno = -1; |
351df4b2 JK |
146 | unsigned int total_segs = TOTAL_SEGS(sbi); |
147 | ||
148 | mutex_lock(&dirty_i->seglist_lock); | |
149 | while (1) { | |
150 | segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs, | |
81fb5e87 | 151 | segno + 1); |
351df4b2 JK |
152 | if (segno >= total_segs) |
153 | break; | |
154 | ||
351df4b2 JK |
155 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE])) |
156 | dirty_i->nr_dirty[PRE]--; | |
157 | ||
158 | /* Let's use trim */ | |
159 | if (test_opt(sbi, DISCARD)) | |
160 | blkdev_issue_discard(sbi->sb->s_bdev, | |
161 | START_BLOCK(sbi, segno) << | |
162 | sbi->log_sectors_per_block, | |
163 | 1 << (sbi->log_sectors_per_block + | |
164 | sbi->log_blocks_per_seg), | |
165 | GFP_NOFS, 0); | |
166 | } | |
167 | mutex_unlock(&dirty_i->seglist_lock); | |
168 | } | |
169 | ||
170 | static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) | |
171 | { | |
172 | struct sit_info *sit_i = SIT_I(sbi); | |
173 | if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) | |
174 | sit_i->dirty_sentries++; | |
175 | } | |
176 | ||
177 | static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, | |
178 | unsigned int segno, int modified) | |
179 | { | |
180 | struct seg_entry *se = get_seg_entry(sbi, segno); | |
181 | se->type = type; | |
182 | if (modified) | |
183 | __mark_sit_entry_dirty(sbi, segno); | |
184 | } | |
185 | ||
186 | static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) | |
187 | { | |
188 | struct seg_entry *se; | |
189 | unsigned int segno, offset; | |
190 | long int new_vblocks; | |
191 | ||
192 | segno = GET_SEGNO(sbi, blkaddr); | |
193 | ||
194 | se = get_seg_entry(sbi, segno); | |
195 | new_vblocks = se->valid_blocks + del; | |
196 | offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1); | |
197 | ||
198 | BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) || | |
199 | (new_vblocks > sbi->blocks_per_seg))); | |
200 | ||
201 | se->valid_blocks = new_vblocks; | |
202 | se->mtime = get_mtime(sbi); | |
203 | SIT_I(sbi)->max_mtime = se->mtime; | |
204 | ||
205 | /* Update valid block bitmap */ | |
206 | if (del > 0) { | |
207 | if (f2fs_set_bit(offset, se->cur_valid_map)) | |
208 | BUG(); | |
209 | } else { | |
210 | if (!f2fs_clear_bit(offset, se->cur_valid_map)) | |
211 | BUG(); | |
212 | } | |
213 | if (!f2fs_test_bit(offset, se->ckpt_valid_map)) | |
214 | se->ckpt_valid_blocks += del; | |
215 | ||
216 | __mark_sit_entry_dirty(sbi, segno); | |
217 | ||
218 | /* update total number of valid blocks to be written in ckpt area */ | |
219 | SIT_I(sbi)->written_valid_blocks += del; | |
220 | ||
221 | if (sbi->segs_per_sec > 1) | |
222 | get_sec_entry(sbi, segno)->valid_blocks += del; | |
223 | } | |
224 | ||
225 | static void refresh_sit_entry(struct f2fs_sb_info *sbi, | |
226 | block_t old_blkaddr, block_t new_blkaddr) | |
227 | { | |
228 | update_sit_entry(sbi, new_blkaddr, 1); | |
229 | if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) | |
230 | update_sit_entry(sbi, old_blkaddr, -1); | |
231 | } | |
232 | ||
233 | void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) | |
234 | { | |
235 | unsigned int segno = GET_SEGNO(sbi, addr); | |
236 | struct sit_info *sit_i = SIT_I(sbi); | |
237 | ||
238 | BUG_ON(addr == NULL_ADDR); | |
239 | if (addr == NEW_ADDR) | |
240 | return; | |
241 | ||
242 | /* add it into sit main buffer */ | |
243 | mutex_lock(&sit_i->sentry_lock); | |
244 | ||
245 | update_sit_entry(sbi, addr, -1); | |
246 | ||
247 | /* add it into dirty seglist */ | |
248 | locate_dirty_segment(sbi, segno); | |
249 | ||
250 | mutex_unlock(&sit_i->sentry_lock); | |
251 | } | |
252 | ||
0a8165d7 | 253 | /* |
351df4b2 JK |
254 | * This function should be resided under the curseg_mutex lock |
255 | */ | |
256 | static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, | |
e79efe3b | 257 | struct f2fs_summary *sum) |
351df4b2 JK |
258 | { |
259 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
260 | void *addr = curseg->sum_blk; | |
e79efe3b | 261 | addr += curseg->next_blkoff * sizeof(struct f2fs_summary); |
351df4b2 | 262 | memcpy(addr, sum, sizeof(struct f2fs_summary)); |
351df4b2 JK |
263 | } |
264 | ||
0a8165d7 | 265 | /* |
351df4b2 JK |
266 | * Calculate the number of current summary pages for writing |
267 | */ | |
268 | int npages_for_summary_flush(struct f2fs_sb_info *sbi) | |
269 | { | |
270 | int total_size_bytes = 0; | |
271 | int valid_sum_count = 0; | |
272 | int i, sum_space; | |
273 | ||
274 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
275 | if (sbi->ckpt->alloc_type[i] == SSR) | |
276 | valid_sum_count += sbi->blocks_per_seg; | |
277 | else | |
278 | valid_sum_count += curseg_blkoff(sbi, i); | |
279 | } | |
280 | ||
281 | total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1) | |
282 | + sizeof(struct nat_journal) + 2 | |
283 | + sizeof(struct sit_journal) + 2; | |
284 | sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE; | |
285 | if (total_size_bytes < sum_space) | |
286 | return 1; | |
287 | else if (total_size_bytes < 2 * sum_space) | |
288 | return 2; | |
289 | return 3; | |
290 | } | |
291 | ||
0a8165d7 | 292 | /* |
351df4b2 JK |
293 | * Caller should put this summary page |
294 | */ | |
295 | struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) | |
296 | { | |
297 | return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); | |
298 | } | |
299 | ||
300 | static void write_sum_page(struct f2fs_sb_info *sbi, | |
301 | struct f2fs_summary_block *sum_blk, block_t blk_addr) | |
302 | { | |
303 | struct page *page = grab_meta_page(sbi, blk_addr); | |
304 | void *kaddr = page_address(page); | |
305 | memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE); | |
306 | set_page_dirty(page); | |
307 | f2fs_put_page(page, 1); | |
308 | } | |
309 | ||
60374688 JK |
310 | static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) |
311 | { | |
312 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
81fb5e87 | 313 | unsigned int segno = curseg->segno + 1; |
60374688 JK |
314 | struct free_segmap_info *free_i = FREE_I(sbi); |
315 | ||
81fb5e87 HL |
316 | if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec) |
317 | return !test_bit(segno, free_i->free_segmap); | |
60374688 JK |
318 | return 0; |
319 | } | |
320 | ||
0a8165d7 | 321 | /* |
351df4b2 JK |
322 | * Find a new segment from the free segments bitmap to right order |
323 | * This function should be returned with success, otherwise BUG | |
324 | */ | |
325 | static void get_new_segment(struct f2fs_sb_info *sbi, | |
326 | unsigned int *newseg, bool new_sec, int dir) | |
327 | { | |
328 | struct free_segmap_info *free_i = FREE_I(sbi); | |
351df4b2 | 329 | unsigned int segno, secno, zoneno; |
53cf9522 | 330 | unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone; |
351df4b2 JK |
331 | unsigned int hint = *newseg / sbi->segs_per_sec; |
332 | unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); | |
333 | unsigned int left_start = hint; | |
334 | bool init = true; | |
335 | int go_left = 0; | |
336 | int i; | |
337 | ||
338 | write_lock(&free_i->segmap_lock); | |
339 | ||
340 | if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { | |
341 | segno = find_next_zero_bit(free_i->free_segmap, | |
342 | TOTAL_SEGS(sbi), *newseg + 1); | |
33afa7fd JK |
343 | if (segno - *newseg < sbi->segs_per_sec - |
344 | (*newseg % sbi->segs_per_sec)) | |
351df4b2 JK |
345 | goto got_it; |
346 | } | |
347 | find_other_zone: | |
53cf9522 JK |
348 | secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint); |
349 | if (secno >= TOTAL_SECS(sbi)) { | |
351df4b2 JK |
350 | if (dir == ALLOC_RIGHT) { |
351 | secno = find_next_zero_bit(free_i->free_secmap, | |
53cf9522 JK |
352 | TOTAL_SECS(sbi), 0); |
353 | BUG_ON(secno >= TOTAL_SECS(sbi)); | |
351df4b2 JK |
354 | } else { |
355 | go_left = 1; | |
356 | left_start = hint - 1; | |
357 | } | |
358 | } | |
359 | if (go_left == 0) | |
360 | goto skip_left; | |
361 | ||
362 | while (test_bit(left_start, free_i->free_secmap)) { | |
363 | if (left_start > 0) { | |
364 | left_start--; | |
365 | continue; | |
366 | } | |
367 | left_start = find_next_zero_bit(free_i->free_secmap, | |
53cf9522 JK |
368 | TOTAL_SECS(sbi), 0); |
369 | BUG_ON(left_start >= TOTAL_SECS(sbi)); | |
351df4b2 JK |
370 | break; |
371 | } | |
372 | secno = left_start; | |
373 | skip_left: | |
374 | hint = secno; | |
375 | segno = secno * sbi->segs_per_sec; | |
376 | zoneno = secno / sbi->secs_per_zone; | |
377 | ||
378 | /* give up on finding another zone */ | |
379 | if (!init) | |
380 | goto got_it; | |
381 | if (sbi->secs_per_zone == 1) | |
382 | goto got_it; | |
383 | if (zoneno == old_zoneno) | |
384 | goto got_it; | |
385 | if (dir == ALLOC_LEFT) { | |
386 | if (!go_left && zoneno + 1 >= total_zones) | |
387 | goto got_it; | |
388 | if (go_left && zoneno == 0) | |
389 | goto got_it; | |
390 | } | |
391 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
392 | if (CURSEG_I(sbi, i)->zone == zoneno) | |
393 | break; | |
394 | ||
395 | if (i < NR_CURSEG_TYPE) { | |
396 | /* zone is in user, try another */ | |
397 | if (go_left) | |
398 | hint = zoneno * sbi->secs_per_zone - 1; | |
399 | else if (zoneno + 1 >= total_zones) | |
400 | hint = 0; | |
401 | else | |
402 | hint = (zoneno + 1) * sbi->secs_per_zone; | |
403 | init = false; | |
404 | goto find_other_zone; | |
405 | } | |
406 | got_it: | |
407 | /* set it as dirty segment in free segmap */ | |
408 | BUG_ON(test_bit(segno, free_i->free_segmap)); | |
409 | __set_inuse(sbi, segno); | |
410 | *newseg = segno; | |
411 | write_unlock(&free_i->segmap_lock); | |
412 | } | |
413 | ||
414 | static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) | |
415 | { | |
416 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
417 | struct summary_footer *sum_footer; | |
418 | ||
419 | curseg->segno = curseg->next_segno; | |
420 | curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); | |
421 | curseg->next_blkoff = 0; | |
422 | curseg->next_segno = NULL_SEGNO; | |
423 | ||
424 | sum_footer = &(curseg->sum_blk->footer); | |
425 | memset(sum_footer, 0, sizeof(struct summary_footer)); | |
426 | if (IS_DATASEG(type)) | |
427 | SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); | |
428 | if (IS_NODESEG(type)) | |
429 | SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); | |
430 | __set_sit_entry_type(sbi, type, curseg->segno, modified); | |
431 | } | |
432 | ||
0a8165d7 | 433 | /* |
351df4b2 JK |
434 | * Allocate a current working segment. |
435 | * This function always allocates a free segment in LFS manner. | |
436 | */ | |
437 | static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) | |
438 | { | |
439 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
440 | unsigned int segno = curseg->segno; | |
441 | int dir = ALLOC_LEFT; | |
442 | ||
443 | write_sum_page(sbi, curseg->sum_blk, | |
81fb5e87 | 444 | GET_SUM_BLOCK(sbi, segno)); |
351df4b2 JK |
445 | if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) |
446 | dir = ALLOC_RIGHT; | |
447 | ||
448 | if (test_opt(sbi, NOHEAP)) | |
449 | dir = ALLOC_RIGHT; | |
450 | ||
451 | get_new_segment(sbi, &segno, new_sec, dir); | |
452 | curseg->next_segno = segno; | |
453 | reset_curseg(sbi, type, 1); | |
454 | curseg->alloc_type = LFS; | |
455 | } | |
456 | ||
457 | static void __next_free_blkoff(struct f2fs_sb_info *sbi, | |
458 | struct curseg_info *seg, block_t start) | |
459 | { | |
460 | struct seg_entry *se = get_seg_entry(sbi, seg->segno); | |
461 | block_t ofs; | |
462 | for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) { | |
463 | if (!f2fs_test_bit(ofs, se->ckpt_valid_map) | |
464 | && !f2fs_test_bit(ofs, se->cur_valid_map)) | |
465 | break; | |
466 | } | |
467 | seg->next_blkoff = ofs; | |
468 | } | |
469 | ||
0a8165d7 | 470 | /* |
351df4b2 JK |
471 | * If a segment is written by LFS manner, next block offset is just obtained |
472 | * by increasing the current block offset. However, if a segment is written by | |
473 | * SSR manner, next block offset obtained by calling __next_free_blkoff | |
474 | */ | |
475 | static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, | |
476 | struct curseg_info *seg) | |
477 | { | |
478 | if (seg->alloc_type == SSR) | |
479 | __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); | |
480 | else | |
481 | seg->next_blkoff++; | |
482 | } | |
483 | ||
0a8165d7 | 484 | /* |
351df4b2 JK |
485 | * This function always allocates a used segment (from dirty seglist) by SSR |
486 | * manner, so it should recover the existing segment information of valid blocks | |
487 | */ | |
488 | static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) | |
489 | { | |
490 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
491 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
492 | unsigned int new_segno = curseg->next_segno; | |
493 | struct f2fs_summary_block *sum_node; | |
494 | struct page *sum_page; | |
495 | ||
496 | write_sum_page(sbi, curseg->sum_blk, | |
497 | GET_SUM_BLOCK(sbi, curseg->segno)); | |
498 | __set_test_and_inuse(sbi, new_segno); | |
499 | ||
500 | mutex_lock(&dirty_i->seglist_lock); | |
501 | __remove_dirty_segment(sbi, new_segno, PRE); | |
502 | __remove_dirty_segment(sbi, new_segno, DIRTY); | |
503 | mutex_unlock(&dirty_i->seglist_lock); | |
504 | ||
505 | reset_curseg(sbi, type, 1); | |
506 | curseg->alloc_type = SSR; | |
507 | __next_free_blkoff(sbi, curseg, 0); | |
508 | ||
509 | if (reuse) { | |
510 | sum_page = get_sum_page(sbi, new_segno); | |
511 | sum_node = (struct f2fs_summary_block *)page_address(sum_page); | |
512 | memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); | |
513 | f2fs_put_page(sum_page, 1); | |
514 | } | |
515 | } | |
516 | ||
43727527 JK |
517 | static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) |
518 | { | |
519 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
520 | const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; | |
521 | ||
522 | if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0)) | |
523 | return v_ops->get_victim(sbi, | |
524 | &(curseg)->next_segno, BG_GC, type, SSR); | |
525 | ||
526 | /* For data segments, let's do SSR more intensively */ | |
527 | for (; type >= CURSEG_HOT_DATA; type--) | |
528 | if (v_ops->get_victim(sbi, &(curseg)->next_segno, | |
529 | BG_GC, type, SSR)) | |
530 | return 1; | |
531 | return 0; | |
532 | } | |
533 | ||
351df4b2 JK |
534 | /* |
535 | * flush out current segment and replace it with new segment | |
536 | * This function should be returned with success, otherwise BUG | |
537 | */ | |
538 | static void allocate_segment_by_default(struct f2fs_sb_info *sbi, | |
539 | int type, bool force) | |
540 | { | |
541 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
351df4b2 | 542 | |
7b405275 | 543 | if (force) |
351df4b2 | 544 | new_curseg(sbi, type, true); |
7b405275 | 545 | else if (type == CURSEG_WARM_NODE) |
351df4b2 | 546 | new_curseg(sbi, type, false); |
60374688 JK |
547 | else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) |
548 | new_curseg(sbi, type, false); | |
351df4b2 JK |
549 | else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) |
550 | change_curseg(sbi, type, true); | |
551 | else | |
552 | new_curseg(sbi, type, false); | |
35b09d82 | 553 | #ifdef CONFIG_F2FS_STAT_FS |
351df4b2 | 554 | sbi->segment_count[curseg->alloc_type]++; |
35b09d82 | 555 | #endif |
351df4b2 JK |
556 | } |
557 | ||
558 | void allocate_new_segments(struct f2fs_sb_info *sbi) | |
559 | { | |
560 | struct curseg_info *curseg; | |
561 | unsigned int old_curseg; | |
562 | int i; | |
563 | ||
564 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
565 | curseg = CURSEG_I(sbi, i); | |
566 | old_curseg = curseg->segno; | |
567 | SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); | |
568 | locate_dirty_segment(sbi, old_curseg); | |
569 | } | |
570 | } | |
571 | ||
572 | static const struct segment_allocation default_salloc_ops = { | |
573 | .allocate_segment = allocate_segment_by_default, | |
574 | }; | |
575 | ||
576 | static void f2fs_end_io_write(struct bio *bio, int err) | |
577 | { | |
578 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
579 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
580 | struct bio_private *p = bio->bi_private; | |
581 | ||
582 | do { | |
583 | struct page *page = bvec->bv_page; | |
584 | ||
585 | if (--bvec >= bio->bi_io_vec) | |
586 | prefetchw(&bvec->bv_page->flags); | |
587 | if (!uptodate) { | |
588 | SetPageError(page); | |
589 | if (page->mapping) | |
590 | set_bit(AS_EIO, &page->mapping->flags); | |
25ca923b | 591 | set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG); |
577e3495 | 592 | p->sbi->sb->s_flags |= MS_RDONLY; |
351df4b2 JK |
593 | } |
594 | end_page_writeback(page); | |
595 | dec_page_count(p->sbi, F2FS_WRITEBACK); | |
596 | } while (bvec >= bio->bi_io_vec); | |
597 | ||
598 | if (p->is_sync) | |
599 | complete(p->wait); | |
e2340887 GZ |
600 | |
601 | if (!get_pages(p->sbi, F2FS_WRITEBACK) && p->sbi->cp_task) | |
602 | wake_up_process(p->sbi->cp_task); | |
603 | ||
351df4b2 JK |
604 | kfree(p); |
605 | bio_put(bio); | |
606 | } | |
607 | ||
3cd8a239 | 608 | struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages) |
351df4b2 JK |
609 | { |
610 | struct bio *bio; | |
3cd8a239 JK |
611 | |
612 | /* No failure on bio allocation */ | |
613 | bio = bio_alloc(GFP_NOIO, npages); | |
614 | bio->bi_bdev = bdev; | |
d8207f69 GZ |
615 | bio->bi_private = NULL; |
616 | ||
351df4b2 JK |
617 | return bio; |
618 | } | |
619 | ||
620 | static void do_submit_bio(struct f2fs_sb_info *sbi, | |
621 | enum page_type type, bool sync) | |
622 | { | |
623 | int rw = sync ? WRITE_SYNC : WRITE; | |
624 | enum page_type btype = type > META ? META : type; | |
625 | ||
626 | if (type >= META_FLUSH) | |
627 | rw = WRITE_FLUSH_FUA; | |
628 | ||
8680441c NJ |
629 | if (btype == META) |
630 | rw |= REQ_META; | |
631 | ||
351df4b2 JK |
632 | if (sbi->bio[btype]) { |
633 | struct bio_private *p = sbi->bio[btype]->bi_private; | |
634 | p->sbi = sbi; | |
635 | sbi->bio[btype]->bi_end_io = f2fs_end_io_write; | |
6ec178da NJ |
636 | |
637 | trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]); | |
638 | ||
351df4b2 JK |
639 | if (type == META_FLUSH) { |
640 | DECLARE_COMPLETION_ONSTACK(wait); | |
641 | p->is_sync = true; | |
642 | p->wait = &wait; | |
643 | submit_bio(rw, sbi->bio[btype]); | |
644 | wait_for_completion(&wait); | |
645 | } else { | |
646 | p->is_sync = false; | |
647 | submit_bio(rw, sbi->bio[btype]); | |
648 | } | |
649 | sbi->bio[btype] = NULL; | |
650 | } | |
651 | } | |
652 | ||
653 | void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync) | |
654 | { | |
655 | down_write(&sbi->bio_sem); | |
656 | do_submit_bio(sbi, type, sync); | |
657 | up_write(&sbi->bio_sem); | |
658 | } | |
659 | ||
660 | static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page, | |
661 | block_t blk_addr, enum page_type type) | |
662 | { | |
663 | struct block_device *bdev = sbi->sb->s_bdev; | |
cc7b1bb1 | 664 | int bio_blocks; |
351df4b2 JK |
665 | |
666 | verify_block_addr(sbi, blk_addr); | |
667 | ||
668 | down_write(&sbi->bio_sem); | |
669 | ||
670 | inc_page_count(sbi, F2FS_WRITEBACK); | |
671 | ||
672 | if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1) | |
673 | do_submit_bio(sbi, type, false); | |
674 | alloc_new: | |
3cd8a239 | 675 | if (sbi->bio[type] == NULL) { |
d8207f69 GZ |
676 | struct bio_private *priv; |
677 | retry: | |
678 | priv = kmalloc(sizeof(struct bio_private), GFP_NOFS); | |
679 | if (!priv) { | |
680 | cond_resched(); | |
681 | goto retry; | |
682 | } | |
683 | ||
cc7b1bb1 CY |
684 | bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi)); |
685 | sbi->bio[type] = f2fs_bio_alloc(bdev, bio_blocks); | |
3cd8a239 | 686 | sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); |
d8207f69 | 687 | sbi->bio[type]->bi_private = priv; |
3cd8a239 JK |
688 | /* |
689 | * The end_io will be assigned at the sumbission phase. | |
690 | * Until then, let bio_add_page() merge consecutive IOs as much | |
691 | * as possible. | |
692 | */ | |
693 | } | |
351df4b2 JK |
694 | |
695 | if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) < | |
696 | PAGE_CACHE_SIZE) { | |
697 | do_submit_bio(sbi, type, false); | |
698 | goto alloc_new; | |
699 | } | |
700 | ||
701 | sbi->last_block_in_bio[type] = blk_addr; | |
702 | ||
703 | up_write(&sbi->bio_sem); | |
6ec178da | 704 | trace_f2fs_submit_write_page(page, blk_addr, type); |
351df4b2 JK |
705 | } |
706 | ||
a569469e JX |
707 | void f2fs_wait_on_page_writeback(struct page *page, |
708 | enum page_type type, bool sync) | |
709 | { | |
710 | struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb); | |
711 | if (PageWriteback(page)) { | |
712 | f2fs_submit_bio(sbi, type, sync); | |
713 | wait_on_page_writeback(page); | |
714 | } | |
715 | } | |
716 | ||
351df4b2 JK |
717 | static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) |
718 | { | |
719 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
720 | if (curseg->next_blkoff < sbi->blocks_per_seg) | |
721 | return true; | |
722 | return false; | |
723 | } | |
724 | ||
725 | static int __get_segment_type_2(struct page *page, enum page_type p_type) | |
726 | { | |
727 | if (p_type == DATA) | |
728 | return CURSEG_HOT_DATA; | |
729 | else | |
730 | return CURSEG_HOT_NODE; | |
731 | } | |
732 | ||
733 | static int __get_segment_type_4(struct page *page, enum page_type p_type) | |
734 | { | |
735 | if (p_type == DATA) { | |
736 | struct inode *inode = page->mapping->host; | |
737 | ||
738 | if (S_ISDIR(inode->i_mode)) | |
739 | return CURSEG_HOT_DATA; | |
740 | else | |
741 | return CURSEG_COLD_DATA; | |
742 | } else { | |
743 | if (IS_DNODE(page) && !is_cold_node(page)) | |
744 | return CURSEG_HOT_NODE; | |
745 | else | |
746 | return CURSEG_COLD_NODE; | |
747 | } | |
748 | } | |
749 | ||
750 | static int __get_segment_type_6(struct page *page, enum page_type p_type) | |
751 | { | |
752 | if (p_type == DATA) { | |
753 | struct inode *inode = page->mapping->host; | |
754 | ||
755 | if (S_ISDIR(inode->i_mode)) | |
756 | return CURSEG_HOT_DATA; | |
354a3399 | 757 | else if (is_cold_data(page) || file_is_cold(inode)) |
351df4b2 JK |
758 | return CURSEG_COLD_DATA; |
759 | else | |
760 | return CURSEG_WARM_DATA; | |
761 | } else { | |
762 | if (IS_DNODE(page)) | |
763 | return is_cold_node(page) ? CURSEG_WARM_NODE : | |
764 | CURSEG_HOT_NODE; | |
765 | else | |
766 | return CURSEG_COLD_NODE; | |
767 | } | |
768 | } | |
769 | ||
770 | static int __get_segment_type(struct page *page, enum page_type p_type) | |
771 | { | |
772 | struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb); | |
773 | switch (sbi->active_logs) { | |
774 | case 2: | |
775 | return __get_segment_type_2(page, p_type); | |
776 | case 4: | |
777 | return __get_segment_type_4(page, p_type); | |
351df4b2 | 778 | } |
12a67146 JK |
779 | /* NR_CURSEG_TYPE(6) logs by default */ |
780 | BUG_ON(sbi->active_logs != NR_CURSEG_TYPE); | |
781 | return __get_segment_type_6(page, p_type); | |
351df4b2 JK |
782 | } |
783 | ||
784 | static void do_write_page(struct f2fs_sb_info *sbi, struct page *page, | |
785 | block_t old_blkaddr, block_t *new_blkaddr, | |
786 | struct f2fs_summary *sum, enum page_type p_type) | |
787 | { | |
788 | struct sit_info *sit_i = SIT_I(sbi); | |
789 | struct curseg_info *curseg; | |
790 | unsigned int old_cursegno; | |
791 | int type; | |
792 | ||
793 | type = __get_segment_type(page, p_type); | |
794 | curseg = CURSEG_I(sbi, type); | |
795 | ||
796 | mutex_lock(&curseg->curseg_mutex); | |
797 | ||
798 | *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); | |
799 | old_cursegno = curseg->segno; | |
800 | ||
801 | /* | |
802 | * __add_sum_entry should be resided under the curseg_mutex | |
803 | * because, this function updates a summary entry in the | |
804 | * current summary block. | |
805 | */ | |
e79efe3b | 806 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
807 | |
808 | mutex_lock(&sit_i->sentry_lock); | |
809 | __refresh_next_blkoff(sbi, curseg); | |
35b09d82 | 810 | #ifdef CONFIG_F2FS_STAT_FS |
351df4b2 | 811 | sbi->block_count[curseg->alloc_type]++; |
35b09d82 | 812 | #endif |
351df4b2 JK |
813 | |
814 | /* | |
815 | * SIT information should be updated before segment allocation, | |
816 | * since SSR needs latest valid block information. | |
817 | */ | |
818 | refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); | |
819 | ||
820 | if (!__has_curseg_space(sbi, type)) | |
821 | sit_i->s_ops->allocate_segment(sbi, type, false); | |
822 | ||
823 | locate_dirty_segment(sbi, old_cursegno); | |
824 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); | |
825 | mutex_unlock(&sit_i->sentry_lock); | |
826 | ||
827 | if (p_type == NODE) | |
828 | fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); | |
829 | ||
830 | /* writeout dirty page into bdev */ | |
831 | submit_write_page(sbi, page, *new_blkaddr, p_type); | |
832 | ||
833 | mutex_unlock(&curseg->curseg_mutex); | |
834 | } | |
835 | ||
577e3495 | 836 | void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) |
351df4b2 | 837 | { |
351df4b2 JK |
838 | set_page_writeback(page); |
839 | submit_write_page(sbi, page, page->index, META); | |
351df4b2 JK |
840 | } |
841 | ||
842 | void write_node_page(struct f2fs_sb_info *sbi, struct page *page, | |
843 | unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr) | |
844 | { | |
845 | struct f2fs_summary sum; | |
846 | set_summary(&sum, nid, 0, 0); | |
847 | do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE); | |
848 | } | |
849 | ||
850 | void write_data_page(struct inode *inode, struct page *page, | |
851 | struct dnode_of_data *dn, block_t old_blkaddr, | |
852 | block_t *new_blkaddr) | |
853 | { | |
854 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
855 | struct f2fs_summary sum; | |
856 | struct node_info ni; | |
857 | ||
858 | BUG_ON(old_blkaddr == NULL_ADDR); | |
859 | get_node_info(sbi, dn->nid, &ni); | |
860 | set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); | |
861 | ||
862 | do_write_page(sbi, page, old_blkaddr, | |
863 | new_blkaddr, &sum, DATA); | |
864 | } | |
865 | ||
866 | void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page, | |
867 | block_t old_blk_addr) | |
868 | { | |
869 | submit_write_page(sbi, page, old_blk_addr, DATA); | |
870 | } | |
871 | ||
872 | void recover_data_page(struct f2fs_sb_info *sbi, | |
873 | struct page *page, struct f2fs_summary *sum, | |
874 | block_t old_blkaddr, block_t new_blkaddr) | |
875 | { | |
876 | struct sit_info *sit_i = SIT_I(sbi); | |
877 | struct curseg_info *curseg; | |
878 | unsigned int segno, old_cursegno; | |
879 | struct seg_entry *se; | |
880 | int type; | |
881 | ||
882 | segno = GET_SEGNO(sbi, new_blkaddr); | |
883 | se = get_seg_entry(sbi, segno); | |
884 | type = se->type; | |
885 | ||
886 | if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { | |
887 | if (old_blkaddr == NULL_ADDR) | |
888 | type = CURSEG_COLD_DATA; | |
889 | else | |
890 | type = CURSEG_WARM_DATA; | |
891 | } | |
892 | curseg = CURSEG_I(sbi, type); | |
893 | ||
894 | mutex_lock(&curseg->curseg_mutex); | |
895 | mutex_lock(&sit_i->sentry_lock); | |
896 | ||
897 | old_cursegno = curseg->segno; | |
898 | ||
899 | /* change the current segment */ | |
900 | if (segno != curseg->segno) { | |
901 | curseg->next_segno = segno; | |
902 | change_curseg(sbi, type, true); | |
903 | } | |
904 | ||
905 | curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) & | |
906 | (sbi->blocks_per_seg - 1); | |
e79efe3b | 907 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
908 | |
909 | refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); | |
910 | ||
911 | locate_dirty_segment(sbi, old_cursegno); | |
912 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); | |
913 | ||
914 | mutex_unlock(&sit_i->sentry_lock); | |
915 | mutex_unlock(&curseg->curseg_mutex); | |
916 | } | |
917 | ||
918 | void rewrite_node_page(struct f2fs_sb_info *sbi, | |
919 | struct page *page, struct f2fs_summary *sum, | |
920 | block_t old_blkaddr, block_t new_blkaddr) | |
921 | { | |
922 | struct sit_info *sit_i = SIT_I(sbi); | |
923 | int type = CURSEG_WARM_NODE; | |
924 | struct curseg_info *curseg; | |
925 | unsigned int segno, old_cursegno; | |
926 | block_t next_blkaddr = next_blkaddr_of_node(page); | |
927 | unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr); | |
928 | ||
929 | curseg = CURSEG_I(sbi, type); | |
930 | ||
931 | mutex_lock(&curseg->curseg_mutex); | |
932 | mutex_lock(&sit_i->sentry_lock); | |
933 | ||
934 | segno = GET_SEGNO(sbi, new_blkaddr); | |
935 | old_cursegno = curseg->segno; | |
936 | ||
937 | /* change the current segment */ | |
938 | if (segno != curseg->segno) { | |
939 | curseg->next_segno = segno; | |
940 | change_curseg(sbi, type, true); | |
941 | } | |
942 | curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) & | |
943 | (sbi->blocks_per_seg - 1); | |
e79efe3b | 944 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
945 | |
946 | /* change the current log to the next block addr in advance */ | |
947 | if (next_segno != segno) { | |
948 | curseg->next_segno = next_segno; | |
949 | change_curseg(sbi, type, true); | |
950 | } | |
951 | curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) & | |
952 | (sbi->blocks_per_seg - 1); | |
953 | ||
954 | /* rewrite node page */ | |
955 | set_page_writeback(page); | |
956 | submit_write_page(sbi, page, new_blkaddr, NODE); | |
957 | f2fs_submit_bio(sbi, NODE, true); | |
958 | refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); | |
959 | ||
960 | locate_dirty_segment(sbi, old_cursegno); | |
961 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); | |
962 | ||
963 | mutex_unlock(&sit_i->sentry_lock); | |
964 | mutex_unlock(&curseg->curseg_mutex); | |
965 | } | |
966 | ||
967 | static int read_compacted_summaries(struct f2fs_sb_info *sbi) | |
968 | { | |
969 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
970 | struct curseg_info *seg_i; | |
971 | unsigned char *kaddr; | |
972 | struct page *page; | |
973 | block_t start; | |
974 | int i, j, offset; | |
975 | ||
976 | start = start_sum_block(sbi); | |
977 | ||
978 | page = get_meta_page(sbi, start++); | |
979 | kaddr = (unsigned char *)page_address(page); | |
980 | ||
981 | /* Step 1: restore nat cache */ | |
982 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
983 | memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE); | |
984 | ||
985 | /* Step 2: restore sit cache */ | |
986 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
987 | memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE, | |
988 | SUM_JOURNAL_SIZE); | |
989 | offset = 2 * SUM_JOURNAL_SIZE; | |
990 | ||
991 | /* Step 3: restore summary entries */ | |
992 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
993 | unsigned short blk_off; | |
994 | unsigned int segno; | |
995 | ||
996 | seg_i = CURSEG_I(sbi, i); | |
997 | segno = le32_to_cpu(ckpt->cur_data_segno[i]); | |
998 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); | |
999 | seg_i->next_segno = segno; | |
1000 | reset_curseg(sbi, i, 0); | |
1001 | seg_i->alloc_type = ckpt->alloc_type[i]; | |
1002 | seg_i->next_blkoff = blk_off; | |
1003 | ||
1004 | if (seg_i->alloc_type == SSR) | |
1005 | blk_off = sbi->blocks_per_seg; | |
1006 | ||
1007 | for (j = 0; j < blk_off; j++) { | |
1008 | struct f2fs_summary *s; | |
1009 | s = (struct f2fs_summary *)(kaddr + offset); | |
1010 | seg_i->sum_blk->entries[j] = *s; | |
1011 | offset += SUMMARY_SIZE; | |
1012 | if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE - | |
1013 | SUM_FOOTER_SIZE) | |
1014 | continue; | |
1015 | ||
1016 | f2fs_put_page(page, 1); | |
1017 | page = NULL; | |
1018 | ||
1019 | page = get_meta_page(sbi, start++); | |
1020 | kaddr = (unsigned char *)page_address(page); | |
1021 | offset = 0; | |
1022 | } | |
1023 | } | |
1024 | f2fs_put_page(page, 1); | |
1025 | return 0; | |
1026 | } | |
1027 | ||
1028 | static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) | |
1029 | { | |
1030 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1031 | struct f2fs_summary_block *sum; | |
1032 | struct curseg_info *curseg; | |
1033 | struct page *new; | |
1034 | unsigned short blk_off; | |
1035 | unsigned int segno = 0; | |
1036 | block_t blk_addr = 0; | |
1037 | ||
1038 | /* get segment number and block addr */ | |
1039 | if (IS_DATASEG(type)) { | |
1040 | segno = le32_to_cpu(ckpt->cur_data_segno[type]); | |
1041 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - | |
1042 | CURSEG_HOT_DATA]); | |
25ca923b | 1043 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) |
351df4b2 JK |
1044 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); |
1045 | else | |
1046 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); | |
1047 | } else { | |
1048 | segno = le32_to_cpu(ckpt->cur_node_segno[type - | |
1049 | CURSEG_HOT_NODE]); | |
1050 | blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - | |
1051 | CURSEG_HOT_NODE]); | |
25ca923b | 1052 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) |
351df4b2 JK |
1053 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, |
1054 | type - CURSEG_HOT_NODE); | |
1055 | else | |
1056 | blk_addr = GET_SUM_BLOCK(sbi, segno); | |
1057 | } | |
1058 | ||
1059 | new = get_meta_page(sbi, blk_addr); | |
1060 | sum = (struct f2fs_summary_block *)page_address(new); | |
1061 | ||
1062 | if (IS_NODESEG(type)) { | |
25ca923b | 1063 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) { |
351df4b2 JK |
1064 | struct f2fs_summary *ns = &sum->entries[0]; |
1065 | int i; | |
1066 | for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { | |
1067 | ns->version = 0; | |
1068 | ns->ofs_in_node = 0; | |
1069 | } | |
1070 | } else { | |
1071 | if (restore_node_summary(sbi, segno, sum)) { | |
1072 | f2fs_put_page(new, 1); | |
1073 | return -EINVAL; | |
1074 | } | |
1075 | } | |
1076 | } | |
1077 | ||
1078 | /* set uncompleted segment to curseg */ | |
1079 | curseg = CURSEG_I(sbi, type); | |
1080 | mutex_lock(&curseg->curseg_mutex); | |
1081 | memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE); | |
1082 | curseg->next_segno = segno; | |
1083 | reset_curseg(sbi, type, 0); | |
1084 | curseg->alloc_type = ckpt->alloc_type[type]; | |
1085 | curseg->next_blkoff = blk_off; | |
1086 | mutex_unlock(&curseg->curseg_mutex); | |
1087 | f2fs_put_page(new, 1); | |
1088 | return 0; | |
1089 | } | |
1090 | ||
1091 | static int restore_curseg_summaries(struct f2fs_sb_info *sbi) | |
1092 | { | |
1093 | int type = CURSEG_HOT_DATA; | |
1094 | ||
25ca923b | 1095 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) { |
351df4b2 JK |
1096 | /* restore for compacted data summary */ |
1097 | if (read_compacted_summaries(sbi)) | |
1098 | return -EINVAL; | |
1099 | type = CURSEG_HOT_NODE; | |
1100 | } | |
1101 | ||
1102 | for (; type <= CURSEG_COLD_NODE; type++) | |
1103 | if (read_normal_summaries(sbi, type)) | |
1104 | return -EINVAL; | |
1105 | return 0; | |
1106 | } | |
1107 | ||
1108 | static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) | |
1109 | { | |
1110 | struct page *page; | |
1111 | unsigned char *kaddr; | |
1112 | struct f2fs_summary *summary; | |
1113 | struct curseg_info *seg_i; | |
1114 | int written_size = 0; | |
1115 | int i, j; | |
1116 | ||
1117 | page = grab_meta_page(sbi, blkaddr++); | |
1118 | kaddr = (unsigned char *)page_address(page); | |
1119 | ||
1120 | /* Step 1: write nat cache */ | |
1121 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
1122 | memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE); | |
1123 | written_size += SUM_JOURNAL_SIZE; | |
1124 | ||
1125 | /* Step 2: write sit cache */ | |
1126 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1127 | memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits, | |
1128 | SUM_JOURNAL_SIZE); | |
1129 | written_size += SUM_JOURNAL_SIZE; | |
1130 | ||
1131 | set_page_dirty(page); | |
1132 | ||
1133 | /* Step 3: write summary entries */ | |
1134 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1135 | unsigned short blkoff; | |
1136 | seg_i = CURSEG_I(sbi, i); | |
1137 | if (sbi->ckpt->alloc_type[i] == SSR) | |
1138 | blkoff = sbi->blocks_per_seg; | |
1139 | else | |
1140 | blkoff = curseg_blkoff(sbi, i); | |
1141 | ||
1142 | for (j = 0; j < blkoff; j++) { | |
1143 | if (!page) { | |
1144 | page = grab_meta_page(sbi, blkaddr++); | |
1145 | kaddr = (unsigned char *)page_address(page); | |
1146 | written_size = 0; | |
1147 | } | |
1148 | summary = (struct f2fs_summary *)(kaddr + written_size); | |
1149 | *summary = seg_i->sum_blk->entries[j]; | |
1150 | written_size += SUMMARY_SIZE; | |
1151 | set_page_dirty(page); | |
1152 | ||
1153 | if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE - | |
1154 | SUM_FOOTER_SIZE) | |
1155 | continue; | |
1156 | ||
1157 | f2fs_put_page(page, 1); | |
1158 | page = NULL; | |
1159 | } | |
1160 | } | |
1161 | if (page) | |
1162 | f2fs_put_page(page, 1); | |
1163 | } | |
1164 | ||
1165 | static void write_normal_summaries(struct f2fs_sb_info *sbi, | |
1166 | block_t blkaddr, int type) | |
1167 | { | |
1168 | int i, end; | |
1169 | if (IS_DATASEG(type)) | |
1170 | end = type + NR_CURSEG_DATA_TYPE; | |
1171 | else | |
1172 | end = type + NR_CURSEG_NODE_TYPE; | |
1173 | ||
1174 | for (i = type; i < end; i++) { | |
1175 | struct curseg_info *sum = CURSEG_I(sbi, i); | |
1176 | mutex_lock(&sum->curseg_mutex); | |
1177 | write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type)); | |
1178 | mutex_unlock(&sum->curseg_mutex); | |
1179 | } | |
1180 | } | |
1181 | ||
1182 | void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1183 | { | |
25ca923b | 1184 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) |
351df4b2 JK |
1185 | write_compacted_summaries(sbi, start_blk); |
1186 | else | |
1187 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); | |
1188 | } | |
1189 | ||
1190 | void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1191 | { | |
25ca923b | 1192 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG)) |
351df4b2 | 1193 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); |
351df4b2 JK |
1194 | } |
1195 | ||
1196 | int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type, | |
1197 | unsigned int val, int alloc) | |
1198 | { | |
1199 | int i; | |
1200 | ||
1201 | if (type == NAT_JOURNAL) { | |
1202 | for (i = 0; i < nats_in_cursum(sum); i++) { | |
1203 | if (le32_to_cpu(nid_in_journal(sum, i)) == val) | |
1204 | return i; | |
1205 | } | |
1206 | if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) | |
1207 | return update_nats_in_cursum(sum, 1); | |
1208 | } else if (type == SIT_JOURNAL) { | |
1209 | for (i = 0; i < sits_in_cursum(sum); i++) | |
1210 | if (le32_to_cpu(segno_in_journal(sum, i)) == val) | |
1211 | return i; | |
1212 | if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES) | |
1213 | return update_sits_in_cursum(sum, 1); | |
1214 | } | |
1215 | return -1; | |
1216 | } | |
1217 | ||
1218 | static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, | |
1219 | unsigned int segno) | |
1220 | { | |
1221 | struct sit_info *sit_i = SIT_I(sbi); | |
1222 | unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno); | |
1223 | block_t blk_addr = sit_i->sit_base_addr + offset; | |
1224 | ||
1225 | check_seg_range(sbi, segno); | |
1226 | ||
1227 | /* calculate sit block address */ | |
1228 | if (f2fs_test_bit(offset, sit_i->sit_bitmap)) | |
1229 | blk_addr += sit_i->sit_blocks; | |
1230 | ||
1231 | return get_meta_page(sbi, blk_addr); | |
1232 | } | |
1233 | ||
1234 | static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, | |
1235 | unsigned int start) | |
1236 | { | |
1237 | struct sit_info *sit_i = SIT_I(sbi); | |
1238 | struct page *src_page, *dst_page; | |
1239 | pgoff_t src_off, dst_off; | |
1240 | void *src_addr, *dst_addr; | |
1241 | ||
1242 | src_off = current_sit_addr(sbi, start); | |
1243 | dst_off = next_sit_addr(sbi, src_off); | |
1244 | ||
1245 | /* get current sit block page without lock */ | |
1246 | src_page = get_meta_page(sbi, src_off); | |
1247 | dst_page = grab_meta_page(sbi, dst_off); | |
1248 | BUG_ON(PageDirty(src_page)); | |
1249 | ||
1250 | src_addr = page_address(src_page); | |
1251 | dst_addr = page_address(dst_page); | |
1252 | memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); | |
1253 | ||
1254 | set_page_dirty(dst_page); | |
1255 | f2fs_put_page(src_page, 1); | |
1256 | ||
1257 | set_to_next_sit(sit_i, start); | |
1258 | ||
1259 | return dst_page; | |
1260 | } | |
1261 | ||
1262 | static bool flush_sits_in_journal(struct f2fs_sb_info *sbi) | |
1263 | { | |
1264 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1265 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
1266 | int i; | |
1267 | ||
1268 | /* | |
1269 | * If the journal area in the current summary is full of sit entries, | |
1270 | * all the sit entries will be flushed. Otherwise the sit entries | |
1271 | * are not able to replace with newly hot sit entries. | |
1272 | */ | |
1273 | if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) { | |
1274 | for (i = sits_in_cursum(sum) - 1; i >= 0; i--) { | |
1275 | unsigned int segno; | |
1276 | segno = le32_to_cpu(segno_in_journal(sum, i)); | |
1277 | __mark_sit_entry_dirty(sbi, segno); | |
1278 | } | |
1279 | update_sits_in_cursum(sum, -sits_in_cursum(sum)); | |
1280 | return 1; | |
1281 | } | |
1282 | return 0; | |
1283 | } | |
1284 | ||
0a8165d7 | 1285 | /* |
351df4b2 JK |
1286 | * CP calls this function, which flushes SIT entries including sit_journal, |
1287 | * and moves prefree segs to free segs. | |
1288 | */ | |
1289 | void flush_sit_entries(struct f2fs_sb_info *sbi) | |
1290 | { | |
1291 | struct sit_info *sit_i = SIT_I(sbi); | |
1292 | unsigned long *bitmap = sit_i->dirty_sentries_bitmap; | |
1293 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1294 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
1295 | unsigned long nsegs = TOTAL_SEGS(sbi); | |
1296 | struct page *page = NULL; | |
1297 | struct f2fs_sit_block *raw_sit = NULL; | |
1298 | unsigned int start = 0, end = 0; | |
1299 | unsigned int segno = -1; | |
1300 | bool flushed; | |
1301 | ||
1302 | mutex_lock(&curseg->curseg_mutex); | |
1303 | mutex_lock(&sit_i->sentry_lock); | |
1304 | ||
1305 | /* | |
1306 | * "flushed" indicates whether sit entries in journal are flushed | |
1307 | * to the SIT area or not. | |
1308 | */ | |
1309 | flushed = flush_sits_in_journal(sbi); | |
1310 | ||
1311 | while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) { | |
1312 | struct seg_entry *se = get_seg_entry(sbi, segno); | |
1313 | int sit_offset, offset; | |
1314 | ||
1315 | sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); | |
1316 | ||
1317 | if (flushed) | |
1318 | goto to_sit_page; | |
1319 | ||
1320 | offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1); | |
1321 | if (offset >= 0) { | |
1322 | segno_in_journal(sum, offset) = cpu_to_le32(segno); | |
1323 | seg_info_to_raw_sit(se, &sit_in_journal(sum, offset)); | |
1324 | goto flush_done; | |
1325 | } | |
1326 | to_sit_page: | |
1327 | if (!page || (start > segno) || (segno > end)) { | |
1328 | if (page) { | |
1329 | f2fs_put_page(page, 1); | |
1330 | page = NULL; | |
1331 | } | |
1332 | ||
1333 | start = START_SEGNO(sit_i, segno); | |
1334 | end = start + SIT_ENTRY_PER_BLOCK - 1; | |
1335 | ||
1336 | /* read sit block that will be updated */ | |
1337 | page = get_next_sit_page(sbi, start); | |
1338 | raw_sit = page_address(page); | |
1339 | } | |
1340 | ||
1341 | /* udpate entry in SIT block */ | |
1342 | seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]); | |
1343 | flush_done: | |
1344 | __clear_bit(segno, bitmap); | |
1345 | sit_i->dirty_sentries--; | |
1346 | } | |
1347 | mutex_unlock(&sit_i->sentry_lock); | |
1348 | mutex_unlock(&curseg->curseg_mutex); | |
1349 | ||
1350 | /* writeout last modified SIT block */ | |
1351 | f2fs_put_page(page, 1); | |
1352 | ||
1353 | set_prefree_as_free_segments(sbi); | |
1354 | } | |
1355 | ||
1356 | static int build_sit_info(struct f2fs_sb_info *sbi) | |
1357 | { | |
1358 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
1359 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1360 | struct sit_info *sit_i; | |
1361 | unsigned int sit_segs, start; | |
1362 | char *src_bitmap, *dst_bitmap; | |
1363 | unsigned int bitmap_size; | |
1364 | ||
1365 | /* allocate memory for SIT information */ | |
1366 | sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); | |
1367 | if (!sit_i) | |
1368 | return -ENOMEM; | |
1369 | ||
1370 | SM_I(sbi)->sit_info = sit_i; | |
1371 | ||
1372 | sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry)); | |
1373 | if (!sit_i->sentries) | |
1374 | return -ENOMEM; | |
1375 | ||
1376 | bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); | |
1377 | sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL); | |
1378 | if (!sit_i->dirty_sentries_bitmap) | |
1379 | return -ENOMEM; | |
1380 | ||
1381 | for (start = 0; start < TOTAL_SEGS(sbi); start++) { | |
1382 | sit_i->sentries[start].cur_valid_map | |
1383 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
1384 | sit_i->sentries[start].ckpt_valid_map | |
1385 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
1386 | if (!sit_i->sentries[start].cur_valid_map | |
1387 | || !sit_i->sentries[start].ckpt_valid_map) | |
1388 | return -ENOMEM; | |
1389 | } | |
1390 | ||
1391 | if (sbi->segs_per_sec > 1) { | |
53cf9522 | 1392 | sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) * |
351df4b2 JK |
1393 | sizeof(struct sec_entry)); |
1394 | if (!sit_i->sec_entries) | |
1395 | return -ENOMEM; | |
1396 | } | |
1397 | ||
1398 | /* get information related with SIT */ | |
1399 | sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; | |
1400 | ||
1401 | /* setup SIT bitmap from ckeckpoint pack */ | |
1402 | bitmap_size = __bitmap_size(sbi, SIT_BITMAP); | |
1403 | src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); | |
1404 | ||
79b5793b | 1405 | dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); |
351df4b2 JK |
1406 | if (!dst_bitmap) |
1407 | return -ENOMEM; | |
351df4b2 JK |
1408 | |
1409 | /* init SIT information */ | |
1410 | sit_i->s_ops = &default_salloc_ops; | |
1411 | ||
1412 | sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); | |
1413 | sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; | |
1414 | sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count); | |
1415 | sit_i->sit_bitmap = dst_bitmap; | |
1416 | sit_i->bitmap_size = bitmap_size; | |
1417 | sit_i->dirty_sentries = 0; | |
1418 | sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; | |
1419 | sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); | |
1420 | sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; | |
1421 | mutex_init(&sit_i->sentry_lock); | |
1422 | return 0; | |
1423 | } | |
1424 | ||
1425 | static int build_free_segmap(struct f2fs_sb_info *sbi) | |
1426 | { | |
1427 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
1428 | struct free_segmap_info *free_i; | |
1429 | unsigned int bitmap_size, sec_bitmap_size; | |
1430 | ||
1431 | /* allocate memory for free segmap information */ | |
1432 | free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); | |
1433 | if (!free_i) | |
1434 | return -ENOMEM; | |
1435 | ||
1436 | SM_I(sbi)->free_info = free_i; | |
1437 | ||
1438 | bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); | |
1439 | free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL); | |
1440 | if (!free_i->free_segmap) | |
1441 | return -ENOMEM; | |
1442 | ||
53cf9522 | 1443 | sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi)); |
351df4b2 JK |
1444 | free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL); |
1445 | if (!free_i->free_secmap) | |
1446 | return -ENOMEM; | |
1447 | ||
1448 | /* set all segments as dirty temporarily */ | |
1449 | memset(free_i->free_segmap, 0xff, bitmap_size); | |
1450 | memset(free_i->free_secmap, 0xff, sec_bitmap_size); | |
1451 | ||
1452 | /* init free segmap information */ | |
1453 | free_i->start_segno = | |
1454 | (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr); | |
1455 | free_i->free_segments = 0; | |
1456 | free_i->free_sections = 0; | |
1457 | rwlock_init(&free_i->segmap_lock); | |
1458 | return 0; | |
1459 | } | |
1460 | ||
1461 | static int build_curseg(struct f2fs_sb_info *sbi) | |
1462 | { | |
1042d60f | 1463 | struct curseg_info *array; |
351df4b2 JK |
1464 | int i; |
1465 | ||
1466 | array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL); | |
1467 | if (!array) | |
1468 | return -ENOMEM; | |
1469 | ||
1470 | SM_I(sbi)->curseg_array = array; | |
1471 | ||
1472 | for (i = 0; i < NR_CURSEG_TYPE; i++) { | |
1473 | mutex_init(&array[i].curseg_mutex); | |
1474 | array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL); | |
1475 | if (!array[i].sum_blk) | |
1476 | return -ENOMEM; | |
1477 | array[i].segno = NULL_SEGNO; | |
1478 | array[i].next_blkoff = 0; | |
1479 | } | |
1480 | return restore_curseg_summaries(sbi); | |
1481 | } | |
1482 | ||
1483 | static void build_sit_entries(struct f2fs_sb_info *sbi) | |
1484 | { | |
1485 | struct sit_info *sit_i = SIT_I(sbi); | |
1486 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1487 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
1488 | unsigned int start; | |
1489 | ||
1490 | for (start = 0; start < TOTAL_SEGS(sbi); start++) { | |
1491 | struct seg_entry *se = &sit_i->sentries[start]; | |
1492 | struct f2fs_sit_block *sit_blk; | |
1493 | struct f2fs_sit_entry sit; | |
1494 | struct page *page; | |
1495 | int i; | |
1496 | ||
1497 | mutex_lock(&curseg->curseg_mutex); | |
1498 | for (i = 0; i < sits_in_cursum(sum); i++) { | |
1499 | if (le32_to_cpu(segno_in_journal(sum, i)) == start) { | |
1500 | sit = sit_in_journal(sum, i); | |
1501 | mutex_unlock(&curseg->curseg_mutex); | |
1502 | goto got_it; | |
1503 | } | |
1504 | } | |
1505 | mutex_unlock(&curseg->curseg_mutex); | |
1506 | page = get_current_sit_page(sbi, start); | |
1507 | sit_blk = (struct f2fs_sit_block *)page_address(page); | |
1508 | sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; | |
1509 | f2fs_put_page(page, 1); | |
1510 | got_it: | |
1511 | check_block_count(sbi, start, &sit); | |
1512 | seg_info_from_raw_sit(se, &sit); | |
1513 | if (sbi->segs_per_sec > 1) { | |
1514 | struct sec_entry *e = get_sec_entry(sbi, start); | |
1515 | e->valid_blocks += se->valid_blocks; | |
1516 | } | |
1517 | } | |
1518 | } | |
1519 | ||
1520 | static void init_free_segmap(struct f2fs_sb_info *sbi) | |
1521 | { | |
1522 | unsigned int start; | |
1523 | int type; | |
1524 | ||
1525 | for (start = 0; start < TOTAL_SEGS(sbi); start++) { | |
1526 | struct seg_entry *sentry = get_seg_entry(sbi, start); | |
1527 | if (!sentry->valid_blocks) | |
1528 | __set_free(sbi, start); | |
1529 | } | |
1530 | ||
1531 | /* set use the current segments */ | |
1532 | for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { | |
1533 | struct curseg_info *curseg_t = CURSEG_I(sbi, type); | |
1534 | __set_test_and_inuse(sbi, curseg_t->segno); | |
1535 | } | |
1536 | } | |
1537 | ||
1538 | static void init_dirty_segmap(struct f2fs_sb_info *sbi) | |
1539 | { | |
1540 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1541 | struct free_segmap_info *free_i = FREE_I(sbi); | |
8736fbf0 | 1542 | unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi); |
351df4b2 JK |
1543 | unsigned short valid_blocks; |
1544 | ||
8736fbf0 | 1545 | while (1) { |
351df4b2 | 1546 | /* find dirty segment based on free segmap */ |
8736fbf0 NJ |
1547 | segno = find_next_inuse(free_i, total_segs, offset); |
1548 | if (segno >= total_segs) | |
351df4b2 JK |
1549 | break; |
1550 | offset = segno + 1; | |
1551 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
1552 | if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks) | |
1553 | continue; | |
1554 | mutex_lock(&dirty_i->seglist_lock); | |
1555 | __locate_dirty_segment(sbi, segno, DIRTY); | |
1556 | mutex_unlock(&dirty_i->seglist_lock); | |
1557 | } | |
1558 | } | |
1559 | ||
5ec4e49f | 1560 | static int init_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
1561 | { |
1562 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 1563 | unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi)); |
351df4b2 | 1564 | |
5ec4e49f JK |
1565 | dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL); |
1566 | if (!dirty_i->victim_secmap) | |
351df4b2 JK |
1567 | return -ENOMEM; |
1568 | return 0; | |
1569 | } | |
1570 | ||
1571 | static int build_dirty_segmap(struct f2fs_sb_info *sbi) | |
1572 | { | |
1573 | struct dirty_seglist_info *dirty_i; | |
1574 | unsigned int bitmap_size, i; | |
1575 | ||
1576 | /* allocate memory for dirty segments list information */ | |
1577 | dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); | |
1578 | if (!dirty_i) | |
1579 | return -ENOMEM; | |
1580 | ||
1581 | SM_I(sbi)->dirty_info = dirty_i; | |
1582 | mutex_init(&dirty_i->seglist_lock); | |
1583 | ||
1584 | bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); | |
1585 | ||
1586 | for (i = 0; i < NR_DIRTY_TYPE; i++) { | |
1587 | dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL); | |
351df4b2 JK |
1588 | if (!dirty_i->dirty_segmap[i]) |
1589 | return -ENOMEM; | |
1590 | } | |
1591 | ||
1592 | init_dirty_segmap(sbi); | |
5ec4e49f | 1593 | return init_victim_secmap(sbi); |
351df4b2 JK |
1594 | } |
1595 | ||
0a8165d7 | 1596 | /* |
351df4b2 JK |
1597 | * Update min, max modified time for cost-benefit GC algorithm |
1598 | */ | |
1599 | static void init_min_max_mtime(struct f2fs_sb_info *sbi) | |
1600 | { | |
1601 | struct sit_info *sit_i = SIT_I(sbi); | |
1602 | unsigned int segno; | |
1603 | ||
1604 | mutex_lock(&sit_i->sentry_lock); | |
1605 | ||
1606 | sit_i->min_mtime = LLONG_MAX; | |
1607 | ||
1608 | for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) { | |
1609 | unsigned int i; | |
1610 | unsigned long long mtime = 0; | |
1611 | ||
1612 | for (i = 0; i < sbi->segs_per_sec; i++) | |
1613 | mtime += get_seg_entry(sbi, segno + i)->mtime; | |
1614 | ||
1615 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
1616 | ||
1617 | if (sit_i->min_mtime > mtime) | |
1618 | sit_i->min_mtime = mtime; | |
1619 | } | |
1620 | sit_i->max_mtime = get_mtime(sbi); | |
1621 | mutex_unlock(&sit_i->sentry_lock); | |
1622 | } | |
1623 | ||
1624 | int build_segment_manager(struct f2fs_sb_info *sbi) | |
1625 | { | |
1626 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
1627 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1042d60f | 1628 | struct f2fs_sm_info *sm_info; |
351df4b2 JK |
1629 | int err; |
1630 | ||
1631 | sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); | |
1632 | if (!sm_info) | |
1633 | return -ENOMEM; | |
1634 | ||
1635 | /* init sm info */ | |
1636 | sbi->sm_info = sm_info; | |
1637 | INIT_LIST_HEAD(&sm_info->wblist_head); | |
1638 | spin_lock_init(&sm_info->wblist_lock); | |
1639 | sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); | |
1640 | sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); | |
1641 | sm_info->segment_count = le32_to_cpu(raw_super->segment_count); | |
1642 | sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); | |
1643 | sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); | |
1644 | sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); | |
1645 | sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); | |
1646 | ||
1647 | err = build_sit_info(sbi); | |
1648 | if (err) | |
1649 | return err; | |
1650 | err = build_free_segmap(sbi); | |
1651 | if (err) | |
1652 | return err; | |
1653 | err = build_curseg(sbi); | |
1654 | if (err) | |
1655 | return err; | |
1656 | ||
1657 | /* reinit free segmap based on SIT */ | |
1658 | build_sit_entries(sbi); | |
1659 | ||
1660 | init_free_segmap(sbi); | |
1661 | err = build_dirty_segmap(sbi); | |
1662 | if (err) | |
1663 | return err; | |
1664 | ||
1665 | init_min_max_mtime(sbi); | |
1666 | return 0; | |
1667 | } | |
1668 | ||
1669 | static void discard_dirty_segmap(struct f2fs_sb_info *sbi, | |
1670 | enum dirty_type dirty_type) | |
1671 | { | |
1672 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1673 | ||
1674 | mutex_lock(&dirty_i->seglist_lock); | |
1675 | kfree(dirty_i->dirty_segmap[dirty_type]); | |
1676 | dirty_i->nr_dirty[dirty_type] = 0; | |
1677 | mutex_unlock(&dirty_i->seglist_lock); | |
1678 | } | |
1679 | ||
5ec4e49f | 1680 | static void destroy_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
1681 | { |
1682 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 1683 | kfree(dirty_i->victim_secmap); |
351df4b2 JK |
1684 | } |
1685 | ||
1686 | static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) | |
1687 | { | |
1688 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1689 | int i; | |
1690 | ||
1691 | if (!dirty_i) | |
1692 | return; | |
1693 | ||
1694 | /* discard pre-free/dirty segments list */ | |
1695 | for (i = 0; i < NR_DIRTY_TYPE; i++) | |
1696 | discard_dirty_segmap(sbi, i); | |
1697 | ||
5ec4e49f | 1698 | destroy_victim_secmap(sbi); |
351df4b2 JK |
1699 | SM_I(sbi)->dirty_info = NULL; |
1700 | kfree(dirty_i); | |
1701 | } | |
1702 | ||
1703 | static void destroy_curseg(struct f2fs_sb_info *sbi) | |
1704 | { | |
1705 | struct curseg_info *array = SM_I(sbi)->curseg_array; | |
1706 | int i; | |
1707 | ||
1708 | if (!array) | |
1709 | return; | |
1710 | SM_I(sbi)->curseg_array = NULL; | |
1711 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
1712 | kfree(array[i].sum_blk); | |
1713 | kfree(array); | |
1714 | } | |
1715 | ||
1716 | static void destroy_free_segmap(struct f2fs_sb_info *sbi) | |
1717 | { | |
1718 | struct free_segmap_info *free_i = SM_I(sbi)->free_info; | |
1719 | if (!free_i) | |
1720 | return; | |
1721 | SM_I(sbi)->free_info = NULL; | |
1722 | kfree(free_i->free_segmap); | |
1723 | kfree(free_i->free_secmap); | |
1724 | kfree(free_i); | |
1725 | } | |
1726 | ||
1727 | static void destroy_sit_info(struct f2fs_sb_info *sbi) | |
1728 | { | |
1729 | struct sit_info *sit_i = SIT_I(sbi); | |
1730 | unsigned int start; | |
1731 | ||
1732 | if (!sit_i) | |
1733 | return; | |
1734 | ||
1735 | if (sit_i->sentries) { | |
1736 | for (start = 0; start < TOTAL_SEGS(sbi); start++) { | |
1737 | kfree(sit_i->sentries[start].cur_valid_map); | |
1738 | kfree(sit_i->sentries[start].ckpt_valid_map); | |
1739 | } | |
1740 | } | |
1741 | vfree(sit_i->sentries); | |
1742 | vfree(sit_i->sec_entries); | |
1743 | kfree(sit_i->dirty_sentries_bitmap); | |
1744 | ||
1745 | SM_I(sbi)->sit_info = NULL; | |
1746 | kfree(sit_i->sit_bitmap); | |
1747 | kfree(sit_i); | |
1748 | } | |
1749 | ||
1750 | void destroy_segment_manager(struct f2fs_sb_info *sbi) | |
1751 | { | |
1752 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
1753 | destroy_dirty_segmap(sbi); | |
1754 | destroy_curseg(sbi); | |
1755 | destroy_free_segmap(sbi); | |
1756 | destroy_sit_info(sbi); | |
1757 | sbi->sm_info = NULL; | |
1758 | kfree(sm_info); | |
1759 | } |