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> |
6b4afdd7 | 16 | #include <linux/kthread.h> |
351df4b2 | 17 | #include <linux/vmalloc.h> |
74de593a | 18 | #include <linux/swap.h> |
351df4b2 JK |
19 | |
20 | #include "f2fs.h" | |
21 | #include "segment.h" | |
22 | #include "node.h" | |
6ec178da | 23 | #include <trace/events/f2fs.h> |
351df4b2 | 24 | |
9a7f143a CL |
25 | #define __reverse_ffz(x) __reverse_ffs(~(x)) |
26 | ||
7fd9e544 | 27 | static struct kmem_cache *discard_entry_slab; |
184a5cd2 | 28 | static struct kmem_cache *sit_entry_set_slab; |
88b88a66 | 29 | static struct kmem_cache *inmem_entry_slab; |
7fd9e544 | 30 | |
9a7f143a CL |
31 | /* |
32 | * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since | |
33 | * MSB and LSB are reversed in a byte by f2fs_set_bit. | |
34 | */ | |
35 | static inline unsigned long __reverse_ffs(unsigned long word) | |
36 | { | |
37 | int num = 0; | |
38 | ||
39 | #if BITS_PER_LONG == 64 | |
40 | if ((word & 0xffffffff) == 0) { | |
41 | num += 32; | |
42 | word >>= 32; | |
43 | } | |
44 | #endif | |
45 | if ((word & 0xffff) == 0) { | |
46 | num += 16; | |
47 | word >>= 16; | |
48 | } | |
49 | if ((word & 0xff) == 0) { | |
50 | num += 8; | |
51 | word >>= 8; | |
52 | } | |
53 | if ((word & 0xf0) == 0) | |
54 | num += 4; | |
55 | else | |
56 | word >>= 4; | |
57 | if ((word & 0xc) == 0) | |
58 | num += 2; | |
59 | else | |
60 | word >>= 2; | |
61 | if ((word & 0x2) == 0) | |
62 | num += 1; | |
63 | return num; | |
64 | } | |
65 | ||
66 | /* | |
e1c42045 | 67 | * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because |
9a7f143a CL |
68 | * f2fs_set_bit makes MSB and LSB reversed in a byte. |
69 | * Example: | |
70 | * LSB <--> MSB | |
71 | * f2fs_set_bit(0, bitmap) => 0000 0001 | |
72 | * f2fs_set_bit(7, bitmap) => 1000 0000 | |
73 | */ | |
74 | static unsigned long __find_rev_next_bit(const unsigned long *addr, | |
75 | unsigned long size, unsigned long offset) | |
76 | { | |
77 | const unsigned long *p = addr + BIT_WORD(offset); | |
78 | unsigned long result = offset & ~(BITS_PER_LONG - 1); | |
79 | unsigned long tmp; | |
80 | unsigned long mask, submask; | |
81 | unsigned long quot, rest; | |
82 | ||
83 | if (offset >= size) | |
84 | return size; | |
85 | ||
86 | size -= result; | |
87 | offset %= BITS_PER_LONG; | |
88 | if (!offset) | |
89 | goto aligned; | |
90 | ||
91 | tmp = *(p++); | |
92 | quot = (offset >> 3) << 3; | |
93 | rest = offset & 0x7; | |
94 | mask = ~0UL << quot; | |
95 | submask = (unsigned char)(0xff << rest) >> rest; | |
96 | submask <<= quot; | |
97 | mask &= submask; | |
98 | tmp &= mask; | |
99 | if (size < BITS_PER_LONG) | |
100 | goto found_first; | |
101 | if (tmp) | |
102 | goto found_middle; | |
103 | ||
104 | size -= BITS_PER_LONG; | |
105 | result += BITS_PER_LONG; | |
106 | aligned: | |
107 | while (size & ~(BITS_PER_LONG-1)) { | |
108 | tmp = *(p++); | |
109 | if (tmp) | |
110 | goto found_middle; | |
111 | result += BITS_PER_LONG; | |
112 | size -= BITS_PER_LONG; | |
113 | } | |
114 | if (!size) | |
115 | return result; | |
116 | tmp = *p; | |
117 | found_first: | |
118 | tmp &= (~0UL >> (BITS_PER_LONG - size)); | |
119 | if (tmp == 0UL) /* Are any bits set? */ | |
120 | return result + size; /* Nope. */ | |
121 | found_middle: | |
122 | return result + __reverse_ffs(tmp); | |
123 | } | |
124 | ||
125 | static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, | |
126 | unsigned long size, unsigned long offset) | |
127 | { | |
128 | const unsigned long *p = addr + BIT_WORD(offset); | |
129 | unsigned long result = offset & ~(BITS_PER_LONG - 1); | |
130 | unsigned long tmp; | |
131 | unsigned long mask, submask; | |
132 | unsigned long quot, rest; | |
133 | ||
134 | if (offset >= size) | |
135 | return size; | |
136 | ||
137 | size -= result; | |
138 | offset %= BITS_PER_LONG; | |
139 | if (!offset) | |
140 | goto aligned; | |
141 | ||
142 | tmp = *(p++); | |
143 | quot = (offset >> 3) << 3; | |
144 | rest = offset & 0x7; | |
145 | mask = ~(~0UL << quot); | |
146 | submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest); | |
147 | submask <<= quot; | |
148 | mask += submask; | |
149 | tmp |= mask; | |
150 | if (size < BITS_PER_LONG) | |
151 | goto found_first; | |
152 | if (~tmp) | |
153 | goto found_middle; | |
154 | ||
155 | size -= BITS_PER_LONG; | |
156 | result += BITS_PER_LONG; | |
157 | aligned: | |
158 | while (size & ~(BITS_PER_LONG - 1)) { | |
159 | tmp = *(p++); | |
160 | if (~tmp) | |
161 | goto found_middle; | |
162 | result += BITS_PER_LONG; | |
163 | size -= BITS_PER_LONG; | |
164 | } | |
165 | if (!size) | |
166 | return result; | |
167 | tmp = *p; | |
168 | ||
169 | found_first: | |
170 | tmp |= ~0UL << size; | |
171 | if (tmp == ~0UL) /* Are any bits zero? */ | |
172 | return result + size; /* Nope. */ | |
173 | found_middle: | |
174 | return result + __reverse_ffz(tmp); | |
175 | } | |
176 | ||
88b88a66 JK |
177 | void register_inmem_page(struct inode *inode, struct page *page) |
178 | { | |
179 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
180 | struct inmem_pages *new; | |
34ba94ba JK |
181 | int err; |
182 | retry: | |
88b88a66 JK |
183 | new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); |
184 | ||
185 | /* add atomic page indices to the list */ | |
186 | new->page = page; | |
187 | INIT_LIST_HEAD(&new->list); | |
188 | ||
189 | /* increase reference count with clean state */ | |
190 | mutex_lock(&fi->inmem_lock); | |
34ba94ba JK |
191 | err = radix_tree_insert(&fi->inmem_root, page->index, new); |
192 | if (err == -EEXIST) { | |
193 | mutex_unlock(&fi->inmem_lock); | |
194 | kmem_cache_free(inmem_entry_slab, new); | |
195 | return; | |
196 | } else if (err) { | |
197 | mutex_unlock(&fi->inmem_lock); | |
198 | kmem_cache_free(inmem_entry_slab, new); | |
199 | goto retry; | |
200 | } | |
88b88a66 JK |
201 | get_page(page); |
202 | list_add_tail(&new->list, &fi->inmem_pages); | |
203 | mutex_unlock(&fi->inmem_lock); | |
204 | } | |
205 | ||
cbcb2872 JK |
206 | void invalidate_inmem_page(struct inode *inode, struct page *page) |
207 | { | |
208 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
209 | struct inmem_pages *cur; | |
210 | ||
211 | mutex_lock(&fi->inmem_lock); | |
212 | cur = radix_tree_lookup(&fi->inmem_root, page->index); | |
213 | if (cur) { | |
214 | radix_tree_delete(&fi->inmem_root, cur->page->index); | |
215 | f2fs_put_page(cur->page, 0); | |
216 | list_del(&cur->list); | |
217 | kmem_cache_free(inmem_entry_slab, cur); | |
218 | } | |
219 | mutex_unlock(&fi->inmem_lock); | |
220 | } | |
221 | ||
88b88a66 JK |
222 | void commit_inmem_pages(struct inode *inode, bool abort) |
223 | { | |
224 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
225 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
226 | struct inmem_pages *cur, *tmp; | |
227 | bool submit_bio = false; | |
228 | struct f2fs_io_info fio = { | |
229 | .type = DATA, | |
230 | .rw = WRITE_SYNC, | |
231 | }; | |
232 | ||
233 | f2fs_balance_fs(sbi); | |
234 | f2fs_lock_op(sbi); | |
235 | ||
236 | mutex_lock(&fi->inmem_lock); | |
237 | list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { | |
238 | lock_page(cur->page); | |
239 | if (!abort && cur->page->mapping == inode->i_mapping) { | |
240 | f2fs_wait_on_page_writeback(cur->page, DATA); | |
241 | if (clear_page_dirty_for_io(cur->page)) | |
242 | inode_dec_dirty_pages(inode); | |
243 | do_write_data_page(cur->page, &fio); | |
244 | submit_bio = true; | |
245 | } | |
34ba94ba | 246 | radix_tree_delete(&fi->inmem_root, cur->page->index); |
88b88a66 JK |
247 | f2fs_put_page(cur->page, 1); |
248 | list_del(&cur->list); | |
249 | kmem_cache_free(inmem_entry_slab, cur); | |
250 | } | |
251 | if (submit_bio) | |
252 | f2fs_submit_merged_bio(sbi, DATA, WRITE); | |
253 | mutex_unlock(&fi->inmem_lock); | |
254 | ||
255 | filemap_fdatawait_range(inode->i_mapping, 0, LLONG_MAX); | |
256 | f2fs_unlock_op(sbi); | |
257 | } | |
258 | ||
0a8165d7 | 259 | /* |
351df4b2 JK |
260 | * This function balances dirty node and dentry pages. |
261 | * In addition, it controls garbage collection. | |
262 | */ | |
263 | void f2fs_balance_fs(struct f2fs_sb_info *sbi) | |
264 | { | |
351df4b2 | 265 | /* |
029cd28c JK |
266 | * We should do GC or end up with checkpoint, if there are so many dirty |
267 | * dir/node pages without enough free segments. | |
351df4b2 | 268 | */ |
43727527 | 269 | if (has_not_enough_free_secs(sbi, 0)) { |
351df4b2 | 270 | mutex_lock(&sbi->gc_mutex); |
408e9375 | 271 | f2fs_gc(sbi); |
351df4b2 JK |
272 | } |
273 | } | |
274 | ||
4660f9c0 JK |
275 | void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) |
276 | { | |
277 | /* check the # of cached NAT entries and prefree segments */ | |
278 | if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) || | |
279 | excess_prefree_segs(sbi)) | |
280 | f2fs_sync_fs(sbi->sb, true); | |
281 | } | |
282 | ||
2163d198 | 283 | static int issue_flush_thread(void *data) |
6b4afdd7 JK |
284 | { |
285 | struct f2fs_sb_info *sbi = data; | |
a688b9d9 GZ |
286 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
287 | wait_queue_head_t *q = &fcc->flush_wait_queue; | |
6b4afdd7 JK |
288 | repeat: |
289 | if (kthread_should_stop()) | |
290 | return 0; | |
291 | ||
721bd4d5 | 292 | if (!llist_empty(&fcc->issue_list)) { |
6b4afdd7 JK |
293 | struct bio *bio = bio_alloc(GFP_NOIO, 0); |
294 | struct flush_cmd *cmd, *next; | |
295 | int ret; | |
296 | ||
721bd4d5 GZ |
297 | fcc->dispatch_list = llist_del_all(&fcc->issue_list); |
298 | fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); | |
299 | ||
6b4afdd7 JK |
300 | bio->bi_bdev = sbi->sb->s_bdev; |
301 | ret = submit_bio_wait(WRITE_FLUSH, bio); | |
302 | ||
721bd4d5 GZ |
303 | llist_for_each_entry_safe(cmd, next, |
304 | fcc->dispatch_list, llnode) { | |
6b4afdd7 | 305 | cmd->ret = ret; |
6b4afdd7 JK |
306 | complete(&cmd->wait); |
307 | } | |
a4ed23f2 | 308 | bio_put(bio); |
a688b9d9 | 309 | fcc->dispatch_list = NULL; |
6b4afdd7 JK |
310 | } |
311 | ||
a688b9d9 | 312 | wait_event_interruptible(*q, |
721bd4d5 | 313 | kthread_should_stop() || !llist_empty(&fcc->issue_list)); |
6b4afdd7 JK |
314 | goto repeat; |
315 | } | |
316 | ||
317 | int f2fs_issue_flush(struct f2fs_sb_info *sbi) | |
318 | { | |
a688b9d9 | 319 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
adf8d90b | 320 | struct flush_cmd cmd; |
6b4afdd7 | 321 | |
24a9ee0f JK |
322 | trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER), |
323 | test_opt(sbi, FLUSH_MERGE)); | |
324 | ||
0f7b2abd JK |
325 | if (test_opt(sbi, NOBARRIER)) |
326 | return 0; | |
327 | ||
6b4afdd7 JK |
328 | if (!test_opt(sbi, FLUSH_MERGE)) |
329 | return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL); | |
330 | ||
adf8d90b | 331 | init_completion(&cmd.wait); |
6b4afdd7 | 332 | |
721bd4d5 | 333 | llist_add(&cmd.llnode, &fcc->issue_list); |
6b4afdd7 | 334 | |
a688b9d9 GZ |
335 | if (!fcc->dispatch_list) |
336 | wake_up(&fcc->flush_wait_queue); | |
6b4afdd7 | 337 | |
adf8d90b CY |
338 | wait_for_completion(&cmd.wait); |
339 | ||
340 | return cmd.ret; | |
6b4afdd7 JK |
341 | } |
342 | ||
2163d198 GZ |
343 | int create_flush_cmd_control(struct f2fs_sb_info *sbi) |
344 | { | |
345 | dev_t dev = sbi->sb->s_bdev->bd_dev; | |
346 | struct flush_cmd_control *fcc; | |
347 | int err = 0; | |
348 | ||
349 | fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL); | |
350 | if (!fcc) | |
351 | return -ENOMEM; | |
2163d198 | 352 | init_waitqueue_head(&fcc->flush_wait_queue); |
721bd4d5 | 353 | init_llist_head(&fcc->issue_list); |
6b2920a5 | 354 | SM_I(sbi)->cmd_control_info = fcc; |
2163d198 GZ |
355 | fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, |
356 | "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); | |
357 | if (IS_ERR(fcc->f2fs_issue_flush)) { | |
358 | err = PTR_ERR(fcc->f2fs_issue_flush); | |
359 | kfree(fcc); | |
6b2920a5 | 360 | SM_I(sbi)->cmd_control_info = NULL; |
2163d198 GZ |
361 | return err; |
362 | } | |
2163d198 GZ |
363 | |
364 | return err; | |
365 | } | |
366 | ||
367 | void destroy_flush_cmd_control(struct f2fs_sb_info *sbi) | |
368 | { | |
6b2920a5 | 369 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
2163d198 GZ |
370 | |
371 | if (fcc && fcc->f2fs_issue_flush) | |
372 | kthread_stop(fcc->f2fs_issue_flush); | |
373 | kfree(fcc); | |
6b2920a5 | 374 | SM_I(sbi)->cmd_control_info = NULL; |
2163d198 GZ |
375 | } |
376 | ||
351df4b2 JK |
377 | static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, |
378 | enum dirty_type dirty_type) | |
379 | { | |
380 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
381 | ||
382 | /* need not be added */ | |
383 | if (IS_CURSEG(sbi, segno)) | |
384 | return; | |
385 | ||
386 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
387 | dirty_i->nr_dirty[dirty_type]++; | |
388 | ||
389 | if (dirty_type == DIRTY) { | |
390 | struct seg_entry *sentry = get_seg_entry(sbi, segno); | |
4625d6aa | 391 | enum dirty_type t = sentry->type; |
b2f2c390 | 392 | |
ec325b52 JK |
393 | if (unlikely(t >= DIRTY)) { |
394 | f2fs_bug_on(sbi, 1); | |
395 | return; | |
396 | } | |
4625d6aa CL |
397 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) |
398 | dirty_i->nr_dirty[t]++; | |
351df4b2 JK |
399 | } |
400 | } | |
401 | ||
402 | static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, | |
403 | enum dirty_type dirty_type) | |
404 | { | |
405 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
406 | ||
407 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
408 | dirty_i->nr_dirty[dirty_type]--; | |
409 | ||
410 | if (dirty_type == DIRTY) { | |
4625d6aa CL |
411 | struct seg_entry *sentry = get_seg_entry(sbi, segno); |
412 | enum dirty_type t = sentry->type; | |
413 | ||
414 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) | |
415 | dirty_i->nr_dirty[t]--; | |
b2f2c390 | 416 | |
5ec4e49f JK |
417 | if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0) |
418 | clear_bit(GET_SECNO(sbi, segno), | |
419 | dirty_i->victim_secmap); | |
351df4b2 JK |
420 | } |
421 | } | |
422 | ||
0a8165d7 | 423 | /* |
351df4b2 JK |
424 | * Should not occur error such as -ENOMEM. |
425 | * Adding dirty entry into seglist is not critical operation. | |
426 | * If a given segment is one of current working segments, it won't be added. | |
427 | */ | |
8d8451af | 428 | static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) |
351df4b2 JK |
429 | { |
430 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
431 | unsigned short valid_blocks; | |
432 | ||
433 | if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) | |
434 | return; | |
435 | ||
436 | mutex_lock(&dirty_i->seglist_lock); | |
437 | ||
438 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
439 | ||
440 | if (valid_blocks == 0) { | |
441 | __locate_dirty_segment(sbi, segno, PRE); | |
442 | __remove_dirty_segment(sbi, segno, DIRTY); | |
443 | } else if (valid_blocks < sbi->blocks_per_seg) { | |
444 | __locate_dirty_segment(sbi, segno, DIRTY); | |
445 | } else { | |
446 | /* Recovery routine with SSR needs this */ | |
447 | __remove_dirty_segment(sbi, segno, DIRTY); | |
448 | } | |
449 | ||
450 | mutex_unlock(&dirty_i->seglist_lock); | |
351df4b2 JK |
451 | } |
452 | ||
1e87a78d | 453 | static int f2fs_issue_discard(struct f2fs_sb_info *sbi, |
37208879 JK |
454 | block_t blkstart, block_t blklen) |
455 | { | |
55cf9cb6 CY |
456 | sector_t start = SECTOR_FROM_BLOCK(blkstart); |
457 | sector_t len = SECTOR_FROM_BLOCK(blklen); | |
1661d07c | 458 | trace_f2fs_issue_discard(sbi->sb, blkstart, blklen); |
1e87a78d JK |
459 | return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0); |
460 | } | |
461 | ||
cf2271e7 | 462 | void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr) |
1e87a78d | 463 | { |
1e87a78d JK |
464 | if (f2fs_issue_discard(sbi, blkaddr, 1)) { |
465 | struct page *page = grab_meta_page(sbi, blkaddr); | |
466 | /* zero-filled page */ | |
467 | set_page_dirty(page); | |
468 | f2fs_put_page(page, 1); | |
469 | } | |
37208879 JK |
470 | } |
471 | ||
4b2fecc8 | 472 | static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc) |
b2955550 JK |
473 | { |
474 | struct list_head *head = &SM_I(sbi)->discard_list; | |
475 | struct discard_entry *new; | |
476 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); | |
477 | int max_blocks = sbi->blocks_per_seg; | |
4b2fecc8 | 478 | struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); |
b2955550 JK |
479 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; |
480 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; | |
481 | unsigned long dmap[entries]; | |
482 | unsigned int start = 0, end = -1; | |
4b2fecc8 | 483 | bool force = (cpc->reason == CP_DISCARD); |
b2955550 JK |
484 | int i; |
485 | ||
4b2fecc8 | 486 | if (!force && !test_opt(sbi, DISCARD)) |
b2955550 JK |
487 | return; |
488 | ||
4b2fecc8 JK |
489 | if (force && !se->valid_blocks) { |
490 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
491 | /* | |
492 | * if this segment is registered in the prefree list, then | |
493 | * we should skip adding a discard candidate, and let the | |
494 | * checkpoint do that later. | |
495 | */ | |
496 | mutex_lock(&dirty_i->seglist_lock); | |
497 | if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) { | |
498 | mutex_unlock(&dirty_i->seglist_lock); | |
499 | cpc->trimmed += sbi->blocks_per_seg; | |
500 | return; | |
501 | } | |
502 | mutex_unlock(&dirty_i->seglist_lock); | |
503 | ||
504 | new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS); | |
505 | INIT_LIST_HEAD(&new->list); | |
506 | new->blkaddr = START_BLOCK(sbi, cpc->trim_start); | |
507 | new->len = sbi->blocks_per_seg; | |
508 | list_add_tail(&new->list, head); | |
509 | SM_I(sbi)->nr_discards += sbi->blocks_per_seg; | |
510 | cpc->trimmed += sbi->blocks_per_seg; | |
511 | return; | |
512 | } | |
513 | ||
b2955550 JK |
514 | /* zero block will be discarded through the prefree list */ |
515 | if (!se->valid_blocks || se->valid_blocks == max_blocks) | |
516 | return; | |
517 | ||
518 | /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ | |
519 | for (i = 0; i < entries; i++) | |
520 | dmap[i] = (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; | |
521 | ||
4b2fecc8 | 522 | while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) { |
b2955550 JK |
523 | start = __find_rev_next_bit(dmap, max_blocks, end + 1); |
524 | if (start >= max_blocks) | |
525 | break; | |
526 | ||
527 | end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); | |
528 | ||
4b2fecc8 JK |
529 | if (end - start < cpc->trim_minlen) |
530 | continue; | |
531 | ||
b2955550 JK |
532 | new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS); |
533 | INIT_LIST_HEAD(&new->list); | |
4b2fecc8 | 534 | new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start; |
b2955550 | 535 | new->len = end - start; |
4b2fecc8 | 536 | cpc->trimmed += end - start; |
b2955550 JK |
537 | |
538 | list_add_tail(&new->list, head); | |
539 | SM_I(sbi)->nr_discards += end - start; | |
540 | } | |
541 | } | |
542 | ||
4b2fecc8 JK |
543 | void release_discard_addrs(struct f2fs_sb_info *sbi) |
544 | { | |
545 | struct list_head *head = &(SM_I(sbi)->discard_list); | |
546 | struct discard_entry *entry, *this; | |
547 | ||
548 | /* drop caches */ | |
549 | list_for_each_entry_safe(entry, this, head, list) { | |
550 | list_del(&entry->list); | |
551 | kmem_cache_free(discard_entry_slab, entry); | |
552 | } | |
553 | } | |
554 | ||
0a8165d7 | 555 | /* |
351df4b2 JK |
556 | * Should call clear_prefree_segments after checkpoint is done. |
557 | */ | |
558 | static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) | |
559 | { | |
560 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
b65ee148 | 561 | unsigned int segno; |
351df4b2 JK |
562 | |
563 | mutex_lock(&dirty_i->seglist_lock); | |
7cd8558b | 564 | for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) |
351df4b2 | 565 | __set_test_and_free(sbi, segno); |
351df4b2 JK |
566 | mutex_unlock(&dirty_i->seglist_lock); |
567 | } | |
568 | ||
569 | void clear_prefree_segments(struct f2fs_sb_info *sbi) | |
570 | { | |
b2955550 | 571 | struct list_head *head = &(SM_I(sbi)->discard_list); |
2d7b822a | 572 | struct discard_entry *entry, *this; |
351df4b2 | 573 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); |
29e59c14 | 574 | unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; |
29e59c14 | 575 | unsigned int start = 0, end = -1; |
351df4b2 JK |
576 | |
577 | mutex_lock(&dirty_i->seglist_lock); | |
29e59c14 | 578 | |
351df4b2 | 579 | while (1) { |
29e59c14 | 580 | int i; |
7cd8558b JK |
581 | start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); |
582 | if (start >= MAIN_SEGS(sbi)) | |
351df4b2 | 583 | break; |
7cd8558b JK |
584 | end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), |
585 | start + 1); | |
29e59c14 CL |
586 | |
587 | for (i = start; i < end; i++) | |
588 | clear_bit(i, prefree_map); | |
589 | ||
590 | dirty_i->nr_dirty[PRE] -= end - start; | |
591 | ||
592 | if (!test_opt(sbi, DISCARD)) | |
593 | continue; | |
351df4b2 | 594 | |
37208879 JK |
595 | f2fs_issue_discard(sbi, START_BLOCK(sbi, start), |
596 | (end - start) << sbi->log_blocks_per_seg); | |
351df4b2 JK |
597 | } |
598 | mutex_unlock(&dirty_i->seglist_lock); | |
b2955550 JK |
599 | |
600 | /* send small discards */ | |
2d7b822a | 601 | list_for_each_entry_safe(entry, this, head, list) { |
37208879 | 602 | f2fs_issue_discard(sbi, entry->blkaddr, entry->len); |
b2955550 JK |
603 | list_del(&entry->list); |
604 | SM_I(sbi)->nr_discards -= entry->len; | |
605 | kmem_cache_free(discard_entry_slab, entry); | |
606 | } | |
351df4b2 JK |
607 | } |
608 | ||
184a5cd2 | 609 | static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) |
351df4b2 JK |
610 | { |
611 | struct sit_info *sit_i = SIT_I(sbi); | |
184a5cd2 CY |
612 | |
613 | if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { | |
351df4b2 | 614 | sit_i->dirty_sentries++; |
184a5cd2 CY |
615 | return false; |
616 | } | |
617 | ||
618 | return true; | |
351df4b2 JK |
619 | } |
620 | ||
621 | static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, | |
622 | unsigned int segno, int modified) | |
623 | { | |
624 | struct seg_entry *se = get_seg_entry(sbi, segno); | |
625 | se->type = type; | |
626 | if (modified) | |
627 | __mark_sit_entry_dirty(sbi, segno); | |
628 | } | |
629 | ||
630 | static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) | |
631 | { | |
632 | struct seg_entry *se; | |
633 | unsigned int segno, offset; | |
634 | long int new_vblocks; | |
635 | ||
636 | segno = GET_SEGNO(sbi, blkaddr); | |
637 | ||
638 | se = get_seg_entry(sbi, segno); | |
639 | new_vblocks = se->valid_blocks + del; | |
491c0854 | 640 | offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); |
351df4b2 | 641 | |
9850cf4a | 642 | f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || |
351df4b2 JK |
643 | (new_vblocks > sbi->blocks_per_seg))); |
644 | ||
645 | se->valid_blocks = new_vblocks; | |
646 | se->mtime = get_mtime(sbi); | |
647 | SIT_I(sbi)->max_mtime = se->mtime; | |
648 | ||
649 | /* Update valid block bitmap */ | |
650 | if (del > 0) { | |
52aca074 | 651 | if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) |
05796763 | 652 | f2fs_bug_on(sbi, 1); |
351df4b2 | 653 | } else { |
52aca074 | 654 | if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) |
05796763 | 655 | f2fs_bug_on(sbi, 1); |
351df4b2 JK |
656 | } |
657 | if (!f2fs_test_bit(offset, se->ckpt_valid_map)) | |
658 | se->ckpt_valid_blocks += del; | |
659 | ||
660 | __mark_sit_entry_dirty(sbi, segno); | |
661 | ||
662 | /* update total number of valid blocks to be written in ckpt area */ | |
663 | SIT_I(sbi)->written_valid_blocks += del; | |
664 | ||
665 | if (sbi->segs_per_sec > 1) | |
666 | get_sec_entry(sbi, segno)->valid_blocks += del; | |
667 | } | |
668 | ||
5e443818 | 669 | void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) |
351df4b2 | 670 | { |
5e443818 JK |
671 | update_sit_entry(sbi, new, 1); |
672 | if (GET_SEGNO(sbi, old) != NULL_SEGNO) | |
673 | update_sit_entry(sbi, old, -1); | |
674 | ||
675 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); | |
676 | locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); | |
351df4b2 JK |
677 | } |
678 | ||
679 | void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) | |
680 | { | |
681 | unsigned int segno = GET_SEGNO(sbi, addr); | |
682 | struct sit_info *sit_i = SIT_I(sbi); | |
683 | ||
9850cf4a | 684 | f2fs_bug_on(sbi, addr == NULL_ADDR); |
351df4b2 JK |
685 | if (addr == NEW_ADDR) |
686 | return; | |
687 | ||
688 | /* add it into sit main buffer */ | |
689 | mutex_lock(&sit_i->sentry_lock); | |
690 | ||
691 | update_sit_entry(sbi, addr, -1); | |
692 | ||
693 | /* add it into dirty seglist */ | |
694 | locate_dirty_segment(sbi, segno); | |
695 | ||
696 | mutex_unlock(&sit_i->sentry_lock); | |
697 | } | |
698 | ||
0a8165d7 | 699 | /* |
351df4b2 JK |
700 | * This function should be resided under the curseg_mutex lock |
701 | */ | |
702 | static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, | |
e79efe3b | 703 | struct f2fs_summary *sum) |
351df4b2 JK |
704 | { |
705 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
706 | void *addr = curseg->sum_blk; | |
e79efe3b | 707 | addr += curseg->next_blkoff * sizeof(struct f2fs_summary); |
351df4b2 | 708 | memcpy(addr, sum, sizeof(struct f2fs_summary)); |
351df4b2 JK |
709 | } |
710 | ||
0a8165d7 | 711 | /* |
351df4b2 JK |
712 | * Calculate the number of current summary pages for writing |
713 | */ | |
714 | int npages_for_summary_flush(struct f2fs_sb_info *sbi) | |
715 | { | |
351df4b2 | 716 | int valid_sum_count = 0; |
9a47938b | 717 | int i, sum_in_page; |
351df4b2 JK |
718 | |
719 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
720 | if (sbi->ckpt->alloc_type[i] == SSR) | |
721 | valid_sum_count += sbi->blocks_per_seg; | |
722 | else | |
723 | valid_sum_count += curseg_blkoff(sbi, i); | |
724 | } | |
725 | ||
9a47938b FL |
726 | sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE - |
727 | SUM_FOOTER_SIZE) / SUMMARY_SIZE; | |
728 | if (valid_sum_count <= sum_in_page) | |
351df4b2 | 729 | return 1; |
9a47938b FL |
730 | else if ((valid_sum_count - sum_in_page) <= |
731 | (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) | |
351df4b2 JK |
732 | return 2; |
733 | return 3; | |
734 | } | |
735 | ||
0a8165d7 | 736 | /* |
351df4b2 JK |
737 | * Caller should put this summary page |
738 | */ | |
739 | struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) | |
740 | { | |
741 | return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); | |
742 | } | |
743 | ||
744 | static void write_sum_page(struct f2fs_sb_info *sbi, | |
745 | struct f2fs_summary_block *sum_blk, block_t blk_addr) | |
746 | { | |
747 | struct page *page = grab_meta_page(sbi, blk_addr); | |
748 | void *kaddr = page_address(page); | |
749 | memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE); | |
750 | set_page_dirty(page); | |
751 | f2fs_put_page(page, 1); | |
752 | } | |
753 | ||
60374688 JK |
754 | static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) |
755 | { | |
756 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
81fb5e87 | 757 | unsigned int segno = curseg->segno + 1; |
60374688 JK |
758 | struct free_segmap_info *free_i = FREE_I(sbi); |
759 | ||
7cd8558b | 760 | if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) |
81fb5e87 | 761 | return !test_bit(segno, free_i->free_segmap); |
60374688 JK |
762 | return 0; |
763 | } | |
764 | ||
0a8165d7 | 765 | /* |
351df4b2 JK |
766 | * Find a new segment from the free segments bitmap to right order |
767 | * This function should be returned with success, otherwise BUG | |
768 | */ | |
769 | static void get_new_segment(struct f2fs_sb_info *sbi, | |
770 | unsigned int *newseg, bool new_sec, int dir) | |
771 | { | |
772 | struct free_segmap_info *free_i = FREE_I(sbi); | |
351df4b2 | 773 | unsigned int segno, secno, zoneno; |
7cd8558b | 774 | unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; |
351df4b2 JK |
775 | unsigned int hint = *newseg / sbi->segs_per_sec; |
776 | unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); | |
777 | unsigned int left_start = hint; | |
778 | bool init = true; | |
779 | int go_left = 0; | |
780 | int i; | |
781 | ||
782 | write_lock(&free_i->segmap_lock); | |
783 | ||
784 | if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { | |
785 | segno = find_next_zero_bit(free_i->free_segmap, | |
7cd8558b | 786 | MAIN_SEGS(sbi), *newseg + 1); |
33afa7fd JK |
787 | if (segno - *newseg < sbi->segs_per_sec - |
788 | (*newseg % sbi->segs_per_sec)) | |
351df4b2 JK |
789 | goto got_it; |
790 | } | |
791 | find_other_zone: | |
7cd8558b JK |
792 | secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); |
793 | if (secno >= MAIN_SECS(sbi)) { | |
351df4b2 JK |
794 | if (dir == ALLOC_RIGHT) { |
795 | secno = find_next_zero_bit(free_i->free_secmap, | |
7cd8558b JK |
796 | MAIN_SECS(sbi), 0); |
797 | f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); | |
351df4b2 JK |
798 | } else { |
799 | go_left = 1; | |
800 | left_start = hint - 1; | |
801 | } | |
802 | } | |
803 | if (go_left == 0) | |
804 | goto skip_left; | |
805 | ||
806 | while (test_bit(left_start, free_i->free_secmap)) { | |
807 | if (left_start > 0) { | |
808 | left_start--; | |
809 | continue; | |
810 | } | |
811 | left_start = find_next_zero_bit(free_i->free_secmap, | |
7cd8558b JK |
812 | MAIN_SECS(sbi), 0); |
813 | f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); | |
351df4b2 JK |
814 | break; |
815 | } | |
816 | secno = left_start; | |
817 | skip_left: | |
818 | hint = secno; | |
819 | segno = secno * sbi->segs_per_sec; | |
820 | zoneno = secno / sbi->secs_per_zone; | |
821 | ||
822 | /* give up on finding another zone */ | |
823 | if (!init) | |
824 | goto got_it; | |
825 | if (sbi->secs_per_zone == 1) | |
826 | goto got_it; | |
827 | if (zoneno == old_zoneno) | |
828 | goto got_it; | |
829 | if (dir == ALLOC_LEFT) { | |
830 | if (!go_left && zoneno + 1 >= total_zones) | |
831 | goto got_it; | |
832 | if (go_left && zoneno == 0) | |
833 | goto got_it; | |
834 | } | |
835 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
836 | if (CURSEG_I(sbi, i)->zone == zoneno) | |
837 | break; | |
838 | ||
839 | if (i < NR_CURSEG_TYPE) { | |
840 | /* zone is in user, try another */ | |
841 | if (go_left) | |
842 | hint = zoneno * sbi->secs_per_zone - 1; | |
843 | else if (zoneno + 1 >= total_zones) | |
844 | hint = 0; | |
845 | else | |
846 | hint = (zoneno + 1) * sbi->secs_per_zone; | |
847 | init = false; | |
848 | goto find_other_zone; | |
849 | } | |
850 | got_it: | |
851 | /* set it as dirty segment in free segmap */ | |
9850cf4a | 852 | f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); |
351df4b2 JK |
853 | __set_inuse(sbi, segno); |
854 | *newseg = segno; | |
855 | write_unlock(&free_i->segmap_lock); | |
856 | } | |
857 | ||
858 | static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) | |
859 | { | |
860 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
861 | struct summary_footer *sum_footer; | |
862 | ||
863 | curseg->segno = curseg->next_segno; | |
864 | curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); | |
865 | curseg->next_blkoff = 0; | |
866 | curseg->next_segno = NULL_SEGNO; | |
867 | ||
868 | sum_footer = &(curseg->sum_blk->footer); | |
869 | memset(sum_footer, 0, sizeof(struct summary_footer)); | |
870 | if (IS_DATASEG(type)) | |
871 | SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); | |
872 | if (IS_NODESEG(type)) | |
873 | SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); | |
874 | __set_sit_entry_type(sbi, type, curseg->segno, modified); | |
875 | } | |
876 | ||
0a8165d7 | 877 | /* |
351df4b2 JK |
878 | * Allocate a current working segment. |
879 | * This function always allocates a free segment in LFS manner. | |
880 | */ | |
881 | static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) | |
882 | { | |
883 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
884 | unsigned int segno = curseg->segno; | |
885 | int dir = ALLOC_LEFT; | |
886 | ||
887 | write_sum_page(sbi, curseg->sum_blk, | |
81fb5e87 | 888 | GET_SUM_BLOCK(sbi, segno)); |
351df4b2 JK |
889 | if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) |
890 | dir = ALLOC_RIGHT; | |
891 | ||
892 | if (test_opt(sbi, NOHEAP)) | |
893 | dir = ALLOC_RIGHT; | |
894 | ||
895 | get_new_segment(sbi, &segno, new_sec, dir); | |
896 | curseg->next_segno = segno; | |
897 | reset_curseg(sbi, type, 1); | |
898 | curseg->alloc_type = LFS; | |
899 | } | |
900 | ||
901 | static void __next_free_blkoff(struct f2fs_sb_info *sbi, | |
902 | struct curseg_info *seg, block_t start) | |
903 | { | |
904 | struct seg_entry *se = get_seg_entry(sbi, seg->segno); | |
e81c93cf CL |
905 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); |
906 | unsigned long target_map[entries]; | |
907 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; | |
908 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; | |
909 | int i, pos; | |
910 | ||
911 | for (i = 0; i < entries; i++) | |
912 | target_map[i] = ckpt_map[i] | cur_map[i]; | |
913 | ||
914 | pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); | |
915 | ||
916 | seg->next_blkoff = pos; | |
351df4b2 JK |
917 | } |
918 | ||
0a8165d7 | 919 | /* |
351df4b2 JK |
920 | * If a segment is written by LFS manner, next block offset is just obtained |
921 | * by increasing the current block offset. However, if a segment is written by | |
922 | * SSR manner, next block offset obtained by calling __next_free_blkoff | |
923 | */ | |
924 | static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, | |
925 | struct curseg_info *seg) | |
926 | { | |
927 | if (seg->alloc_type == SSR) | |
928 | __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); | |
929 | else | |
930 | seg->next_blkoff++; | |
931 | } | |
932 | ||
0a8165d7 | 933 | /* |
e1c42045 | 934 | * This function always allocates a used segment(from dirty seglist) by SSR |
351df4b2 JK |
935 | * manner, so it should recover the existing segment information of valid blocks |
936 | */ | |
937 | static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) | |
938 | { | |
939 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
940 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
941 | unsigned int new_segno = curseg->next_segno; | |
942 | struct f2fs_summary_block *sum_node; | |
943 | struct page *sum_page; | |
944 | ||
945 | write_sum_page(sbi, curseg->sum_blk, | |
946 | GET_SUM_BLOCK(sbi, curseg->segno)); | |
947 | __set_test_and_inuse(sbi, new_segno); | |
948 | ||
949 | mutex_lock(&dirty_i->seglist_lock); | |
950 | __remove_dirty_segment(sbi, new_segno, PRE); | |
951 | __remove_dirty_segment(sbi, new_segno, DIRTY); | |
952 | mutex_unlock(&dirty_i->seglist_lock); | |
953 | ||
954 | reset_curseg(sbi, type, 1); | |
955 | curseg->alloc_type = SSR; | |
956 | __next_free_blkoff(sbi, curseg, 0); | |
957 | ||
958 | if (reuse) { | |
959 | sum_page = get_sum_page(sbi, new_segno); | |
960 | sum_node = (struct f2fs_summary_block *)page_address(sum_page); | |
961 | memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); | |
962 | f2fs_put_page(sum_page, 1); | |
963 | } | |
964 | } | |
965 | ||
43727527 JK |
966 | static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) |
967 | { | |
968 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
969 | const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; | |
970 | ||
971 | if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0)) | |
972 | return v_ops->get_victim(sbi, | |
973 | &(curseg)->next_segno, BG_GC, type, SSR); | |
974 | ||
975 | /* For data segments, let's do SSR more intensively */ | |
976 | for (; type >= CURSEG_HOT_DATA; type--) | |
977 | if (v_ops->get_victim(sbi, &(curseg)->next_segno, | |
978 | BG_GC, type, SSR)) | |
979 | return 1; | |
980 | return 0; | |
981 | } | |
982 | ||
351df4b2 JK |
983 | /* |
984 | * flush out current segment and replace it with new segment | |
985 | * This function should be returned with success, otherwise BUG | |
986 | */ | |
987 | static void allocate_segment_by_default(struct f2fs_sb_info *sbi, | |
988 | int type, bool force) | |
989 | { | |
990 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
351df4b2 | 991 | |
7b405275 | 992 | if (force) |
351df4b2 | 993 | new_curseg(sbi, type, true); |
7b405275 | 994 | else if (type == CURSEG_WARM_NODE) |
351df4b2 | 995 | new_curseg(sbi, type, false); |
60374688 JK |
996 | else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) |
997 | new_curseg(sbi, type, false); | |
351df4b2 JK |
998 | else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) |
999 | change_curseg(sbi, type, true); | |
1000 | else | |
1001 | new_curseg(sbi, type, false); | |
dcdfff65 JK |
1002 | |
1003 | stat_inc_seg_type(sbi, curseg); | |
351df4b2 JK |
1004 | } |
1005 | ||
1006 | void allocate_new_segments(struct f2fs_sb_info *sbi) | |
1007 | { | |
1008 | struct curseg_info *curseg; | |
1009 | unsigned int old_curseg; | |
1010 | int i; | |
1011 | ||
1012 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1013 | curseg = CURSEG_I(sbi, i); | |
1014 | old_curseg = curseg->segno; | |
1015 | SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); | |
1016 | locate_dirty_segment(sbi, old_curseg); | |
1017 | } | |
1018 | } | |
1019 | ||
1020 | static const struct segment_allocation default_salloc_ops = { | |
1021 | .allocate_segment = allocate_segment_by_default, | |
1022 | }; | |
1023 | ||
4b2fecc8 JK |
1024 | int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) |
1025 | { | |
4b2fecc8 JK |
1026 | __u64 start = range->start >> sbi->log_blocksize; |
1027 | __u64 end = start + (range->len >> sbi->log_blocksize) - 1; | |
4b2fecc8 JK |
1028 | unsigned int start_segno, end_segno; |
1029 | struct cp_control cpc; | |
1030 | ||
7cd8558b JK |
1031 | if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) || |
1032 | range->len < sbi->blocksize) | |
4b2fecc8 JK |
1033 | return -EINVAL; |
1034 | ||
9bd27ae4 | 1035 | cpc.trimmed = 0; |
7cd8558b | 1036 | if (end <= MAIN_BLKADDR(sbi)) |
4b2fecc8 JK |
1037 | goto out; |
1038 | ||
1039 | /* start/end segment number in main_area */ | |
7cd8558b JK |
1040 | start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); |
1041 | end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : | |
1042 | GET_SEGNO(sbi, end); | |
4b2fecc8 JK |
1043 | cpc.reason = CP_DISCARD; |
1044 | cpc.trim_start = start_segno; | |
1045 | cpc.trim_end = end_segno; | |
1046 | cpc.trim_minlen = range->minlen >> sbi->log_blocksize; | |
4b2fecc8 JK |
1047 | |
1048 | /* do checkpoint to issue discard commands safely */ | |
1049 | write_checkpoint(sbi, &cpc); | |
1050 | out: | |
1051 | range->len = cpc.trimmed << sbi->log_blocksize; | |
1052 | return 0; | |
1053 | } | |
1054 | ||
351df4b2 JK |
1055 | static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) |
1056 | { | |
1057 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1058 | if (curseg->next_blkoff < sbi->blocks_per_seg) | |
1059 | return true; | |
1060 | return false; | |
1061 | } | |
1062 | ||
1063 | static int __get_segment_type_2(struct page *page, enum page_type p_type) | |
1064 | { | |
1065 | if (p_type == DATA) | |
1066 | return CURSEG_HOT_DATA; | |
1067 | else | |
1068 | return CURSEG_HOT_NODE; | |
1069 | } | |
1070 | ||
1071 | static int __get_segment_type_4(struct page *page, enum page_type p_type) | |
1072 | { | |
1073 | if (p_type == DATA) { | |
1074 | struct inode *inode = page->mapping->host; | |
1075 | ||
1076 | if (S_ISDIR(inode->i_mode)) | |
1077 | return CURSEG_HOT_DATA; | |
1078 | else | |
1079 | return CURSEG_COLD_DATA; | |
1080 | } else { | |
1081 | if (IS_DNODE(page) && !is_cold_node(page)) | |
1082 | return CURSEG_HOT_NODE; | |
1083 | else | |
1084 | return CURSEG_COLD_NODE; | |
1085 | } | |
1086 | } | |
1087 | ||
1088 | static int __get_segment_type_6(struct page *page, enum page_type p_type) | |
1089 | { | |
1090 | if (p_type == DATA) { | |
1091 | struct inode *inode = page->mapping->host; | |
1092 | ||
1093 | if (S_ISDIR(inode->i_mode)) | |
1094 | return CURSEG_HOT_DATA; | |
354a3399 | 1095 | else if (is_cold_data(page) || file_is_cold(inode)) |
351df4b2 JK |
1096 | return CURSEG_COLD_DATA; |
1097 | else | |
1098 | return CURSEG_WARM_DATA; | |
1099 | } else { | |
1100 | if (IS_DNODE(page)) | |
1101 | return is_cold_node(page) ? CURSEG_WARM_NODE : | |
1102 | CURSEG_HOT_NODE; | |
1103 | else | |
1104 | return CURSEG_COLD_NODE; | |
1105 | } | |
1106 | } | |
1107 | ||
1108 | static int __get_segment_type(struct page *page, enum page_type p_type) | |
1109 | { | |
4081363f | 1110 | switch (F2FS_P_SB(page)->active_logs) { |
351df4b2 JK |
1111 | case 2: |
1112 | return __get_segment_type_2(page, p_type); | |
1113 | case 4: | |
1114 | return __get_segment_type_4(page, p_type); | |
351df4b2 | 1115 | } |
12a67146 | 1116 | /* NR_CURSEG_TYPE(6) logs by default */ |
9850cf4a JK |
1117 | f2fs_bug_on(F2FS_P_SB(page), |
1118 | F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE); | |
12a67146 | 1119 | return __get_segment_type_6(page, p_type); |
351df4b2 JK |
1120 | } |
1121 | ||
bfad7c2d JK |
1122 | void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, |
1123 | block_t old_blkaddr, block_t *new_blkaddr, | |
1124 | struct f2fs_summary *sum, int type) | |
351df4b2 JK |
1125 | { |
1126 | struct sit_info *sit_i = SIT_I(sbi); | |
1127 | struct curseg_info *curseg; | |
351df4b2 | 1128 | |
351df4b2 JK |
1129 | curseg = CURSEG_I(sbi, type); |
1130 | ||
1131 | mutex_lock(&curseg->curseg_mutex); | |
1132 | ||
1133 | *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); | |
351df4b2 JK |
1134 | |
1135 | /* | |
1136 | * __add_sum_entry should be resided under the curseg_mutex | |
1137 | * because, this function updates a summary entry in the | |
1138 | * current summary block. | |
1139 | */ | |
e79efe3b | 1140 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
1141 | |
1142 | mutex_lock(&sit_i->sentry_lock); | |
1143 | __refresh_next_blkoff(sbi, curseg); | |
dcdfff65 JK |
1144 | |
1145 | stat_inc_block_count(sbi, curseg); | |
351df4b2 | 1146 | |
5e443818 JK |
1147 | if (!__has_curseg_space(sbi, type)) |
1148 | sit_i->s_ops->allocate_segment(sbi, type, false); | |
351df4b2 JK |
1149 | /* |
1150 | * SIT information should be updated before segment allocation, | |
1151 | * since SSR needs latest valid block information. | |
1152 | */ | |
1153 | refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); | |
5e443818 | 1154 | |
351df4b2 JK |
1155 | mutex_unlock(&sit_i->sentry_lock); |
1156 | ||
bfad7c2d | 1157 | if (page && IS_NODESEG(type)) |
351df4b2 JK |
1158 | fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); |
1159 | ||
bfad7c2d JK |
1160 | mutex_unlock(&curseg->curseg_mutex); |
1161 | } | |
1162 | ||
1163 | static void do_write_page(struct f2fs_sb_info *sbi, struct page *page, | |
1164 | block_t old_blkaddr, block_t *new_blkaddr, | |
1165 | struct f2fs_summary *sum, struct f2fs_io_info *fio) | |
1166 | { | |
1167 | int type = __get_segment_type(page, fio->type); | |
1168 | ||
1169 | allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type); | |
1170 | ||
351df4b2 | 1171 | /* writeout dirty page into bdev */ |
458e6197 | 1172 | f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio); |
351df4b2 JK |
1173 | } |
1174 | ||
577e3495 | 1175 | void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) |
351df4b2 | 1176 | { |
458e6197 JK |
1177 | struct f2fs_io_info fio = { |
1178 | .type = META, | |
7e8f2308 | 1179 | .rw = WRITE_SYNC | REQ_META | REQ_PRIO |
458e6197 JK |
1180 | }; |
1181 | ||
351df4b2 | 1182 | set_page_writeback(page); |
458e6197 | 1183 | f2fs_submit_page_mbio(sbi, page, page->index, &fio); |
351df4b2 JK |
1184 | } |
1185 | ||
1186 | void write_node_page(struct f2fs_sb_info *sbi, struct page *page, | |
fb5566da | 1187 | struct f2fs_io_info *fio, |
351df4b2 JK |
1188 | unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr) |
1189 | { | |
1190 | struct f2fs_summary sum; | |
1191 | set_summary(&sum, nid, 0, 0); | |
fb5566da | 1192 | do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio); |
351df4b2 JK |
1193 | } |
1194 | ||
458e6197 JK |
1195 | void write_data_page(struct page *page, struct dnode_of_data *dn, |
1196 | block_t *new_blkaddr, struct f2fs_io_info *fio) | |
351df4b2 | 1197 | { |
4081363f | 1198 | struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); |
351df4b2 JK |
1199 | struct f2fs_summary sum; |
1200 | struct node_info ni; | |
1201 | ||
9850cf4a | 1202 | f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); |
351df4b2 JK |
1203 | get_node_info(sbi, dn->nid, &ni); |
1204 | set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); | |
1205 | ||
458e6197 | 1206 | do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio); |
351df4b2 JK |
1207 | } |
1208 | ||
6c311ec6 CF |
1209 | void rewrite_data_page(struct page *page, block_t old_blkaddr, |
1210 | struct f2fs_io_info *fio) | |
351df4b2 | 1211 | { |
4081363f | 1212 | f2fs_submit_page_mbio(F2FS_P_SB(page), page, old_blkaddr, fio); |
351df4b2 JK |
1213 | } |
1214 | ||
1215 | void recover_data_page(struct f2fs_sb_info *sbi, | |
1216 | struct page *page, struct f2fs_summary *sum, | |
1217 | block_t old_blkaddr, block_t new_blkaddr) | |
1218 | { | |
1219 | struct sit_info *sit_i = SIT_I(sbi); | |
1220 | struct curseg_info *curseg; | |
1221 | unsigned int segno, old_cursegno; | |
1222 | struct seg_entry *se; | |
1223 | int type; | |
1224 | ||
1225 | segno = GET_SEGNO(sbi, new_blkaddr); | |
1226 | se = get_seg_entry(sbi, segno); | |
1227 | type = se->type; | |
1228 | ||
1229 | if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { | |
1230 | if (old_blkaddr == NULL_ADDR) | |
1231 | type = CURSEG_COLD_DATA; | |
1232 | else | |
1233 | type = CURSEG_WARM_DATA; | |
1234 | } | |
1235 | curseg = CURSEG_I(sbi, type); | |
1236 | ||
1237 | mutex_lock(&curseg->curseg_mutex); | |
1238 | mutex_lock(&sit_i->sentry_lock); | |
1239 | ||
1240 | old_cursegno = curseg->segno; | |
1241 | ||
1242 | /* change the current segment */ | |
1243 | if (segno != curseg->segno) { | |
1244 | curseg->next_segno = segno; | |
1245 | change_curseg(sbi, type, true); | |
1246 | } | |
1247 | ||
491c0854 | 1248 | curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); |
e79efe3b | 1249 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
1250 | |
1251 | refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); | |
351df4b2 | 1252 | locate_dirty_segment(sbi, old_cursegno); |
351df4b2 JK |
1253 | |
1254 | mutex_unlock(&sit_i->sentry_lock); | |
1255 | mutex_unlock(&curseg->curseg_mutex); | |
1256 | } | |
1257 | ||
df0f8dc0 CY |
1258 | static inline bool is_merged_page(struct f2fs_sb_info *sbi, |
1259 | struct page *page, enum page_type type) | |
1260 | { | |
1261 | enum page_type btype = PAGE_TYPE_OF_BIO(type); | |
1262 | struct f2fs_bio_info *io = &sbi->write_io[btype]; | |
df0f8dc0 CY |
1263 | struct bio_vec *bvec; |
1264 | int i; | |
1265 | ||
1266 | down_read(&io->io_rwsem); | |
ce23447f | 1267 | if (!io->bio) |
df0f8dc0 CY |
1268 | goto out; |
1269 | ||
ce23447f | 1270 | bio_for_each_segment_all(bvec, io->bio, i) { |
df0f8dc0 CY |
1271 | if (page == bvec->bv_page) { |
1272 | up_read(&io->io_rwsem); | |
1273 | return true; | |
1274 | } | |
1275 | } | |
1276 | ||
1277 | out: | |
1278 | up_read(&io->io_rwsem); | |
1279 | return false; | |
1280 | } | |
1281 | ||
93dfe2ac | 1282 | void f2fs_wait_on_page_writeback(struct page *page, |
5514f0aa | 1283 | enum page_type type) |
93dfe2ac | 1284 | { |
93dfe2ac | 1285 | if (PageWriteback(page)) { |
4081363f JK |
1286 | struct f2fs_sb_info *sbi = F2FS_P_SB(page); |
1287 | ||
df0f8dc0 CY |
1288 | if (is_merged_page(sbi, page, type)) |
1289 | f2fs_submit_merged_bio(sbi, type, WRITE); | |
93dfe2ac JK |
1290 | wait_on_page_writeback(page); |
1291 | } | |
1292 | } | |
1293 | ||
351df4b2 JK |
1294 | static int read_compacted_summaries(struct f2fs_sb_info *sbi) |
1295 | { | |
1296 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1297 | struct curseg_info *seg_i; | |
1298 | unsigned char *kaddr; | |
1299 | struct page *page; | |
1300 | block_t start; | |
1301 | int i, j, offset; | |
1302 | ||
1303 | start = start_sum_block(sbi); | |
1304 | ||
1305 | page = get_meta_page(sbi, start++); | |
1306 | kaddr = (unsigned char *)page_address(page); | |
1307 | ||
1308 | /* Step 1: restore nat cache */ | |
1309 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
1310 | memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE); | |
1311 | ||
1312 | /* Step 2: restore sit cache */ | |
1313 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1314 | memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE, | |
1315 | SUM_JOURNAL_SIZE); | |
1316 | offset = 2 * SUM_JOURNAL_SIZE; | |
1317 | ||
1318 | /* Step 3: restore summary entries */ | |
1319 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1320 | unsigned short blk_off; | |
1321 | unsigned int segno; | |
1322 | ||
1323 | seg_i = CURSEG_I(sbi, i); | |
1324 | segno = le32_to_cpu(ckpt->cur_data_segno[i]); | |
1325 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); | |
1326 | seg_i->next_segno = segno; | |
1327 | reset_curseg(sbi, i, 0); | |
1328 | seg_i->alloc_type = ckpt->alloc_type[i]; | |
1329 | seg_i->next_blkoff = blk_off; | |
1330 | ||
1331 | if (seg_i->alloc_type == SSR) | |
1332 | blk_off = sbi->blocks_per_seg; | |
1333 | ||
1334 | for (j = 0; j < blk_off; j++) { | |
1335 | struct f2fs_summary *s; | |
1336 | s = (struct f2fs_summary *)(kaddr + offset); | |
1337 | seg_i->sum_blk->entries[j] = *s; | |
1338 | offset += SUMMARY_SIZE; | |
1339 | if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE - | |
1340 | SUM_FOOTER_SIZE) | |
1341 | continue; | |
1342 | ||
1343 | f2fs_put_page(page, 1); | |
1344 | page = NULL; | |
1345 | ||
1346 | page = get_meta_page(sbi, start++); | |
1347 | kaddr = (unsigned char *)page_address(page); | |
1348 | offset = 0; | |
1349 | } | |
1350 | } | |
1351 | f2fs_put_page(page, 1); | |
1352 | return 0; | |
1353 | } | |
1354 | ||
1355 | static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) | |
1356 | { | |
1357 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1358 | struct f2fs_summary_block *sum; | |
1359 | struct curseg_info *curseg; | |
1360 | struct page *new; | |
1361 | unsigned short blk_off; | |
1362 | unsigned int segno = 0; | |
1363 | block_t blk_addr = 0; | |
1364 | ||
1365 | /* get segment number and block addr */ | |
1366 | if (IS_DATASEG(type)) { | |
1367 | segno = le32_to_cpu(ckpt->cur_data_segno[type]); | |
1368 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - | |
1369 | CURSEG_HOT_DATA]); | |
25ca923b | 1370 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) |
351df4b2 JK |
1371 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); |
1372 | else | |
1373 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); | |
1374 | } else { | |
1375 | segno = le32_to_cpu(ckpt->cur_node_segno[type - | |
1376 | CURSEG_HOT_NODE]); | |
1377 | blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - | |
1378 | CURSEG_HOT_NODE]); | |
25ca923b | 1379 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) |
351df4b2 JK |
1380 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, |
1381 | type - CURSEG_HOT_NODE); | |
1382 | else | |
1383 | blk_addr = GET_SUM_BLOCK(sbi, segno); | |
1384 | } | |
1385 | ||
1386 | new = get_meta_page(sbi, blk_addr); | |
1387 | sum = (struct f2fs_summary_block *)page_address(new); | |
1388 | ||
1389 | if (IS_NODESEG(type)) { | |
25ca923b | 1390 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) { |
351df4b2 JK |
1391 | struct f2fs_summary *ns = &sum->entries[0]; |
1392 | int i; | |
1393 | for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { | |
1394 | ns->version = 0; | |
1395 | ns->ofs_in_node = 0; | |
1396 | } | |
1397 | } else { | |
d653788a GZ |
1398 | int err; |
1399 | ||
1400 | err = restore_node_summary(sbi, segno, sum); | |
1401 | if (err) { | |
351df4b2 | 1402 | f2fs_put_page(new, 1); |
d653788a | 1403 | return err; |
351df4b2 JK |
1404 | } |
1405 | } | |
1406 | } | |
1407 | ||
1408 | /* set uncompleted segment to curseg */ | |
1409 | curseg = CURSEG_I(sbi, type); | |
1410 | mutex_lock(&curseg->curseg_mutex); | |
1411 | memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE); | |
1412 | curseg->next_segno = segno; | |
1413 | reset_curseg(sbi, type, 0); | |
1414 | curseg->alloc_type = ckpt->alloc_type[type]; | |
1415 | curseg->next_blkoff = blk_off; | |
1416 | mutex_unlock(&curseg->curseg_mutex); | |
1417 | f2fs_put_page(new, 1); | |
1418 | return 0; | |
1419 | } | |
1420 | ||
1421 | static int restore_curseg_summaries(struct f2fs_sb_info *sbi) | |
1422 | { | |
1423 | int type = CURSEG_HOT_DATA; | |
e4fc5fbf | 1424 | int err; |
351df4b2 | 1425 | |
25ca923b | 1426 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) { |
351df4b2 JK |
1427 | /* restore for compacted data summary */ |
1428 | if (read_compacted_summaries(sbi)) | |
1429 | return -EINVAL; | |
1430 | type = CURSEG_HOT_NODE; | |
1431 | } | |
1432 | ||
e4fc5fbf CY |
1433 | for (; type <= CURSEG_COLD_NODE; type++) { |
1434 | err = read_normal_summaries(sbi, type); | |
1435 | if (err) | |
1436 | return err; | |
1437 | } | |
1438 | ||
351df4b2 JK |
1439 | return 0; |
1440 | } | |
1441 | ||
1442 | static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) | |
1443 | { | |
1444 | struct page *page; | |
1445 | unsigned char *kaddr; | |
1446 | struct f2fs_summary *summary; | |
1447 | struct curseg_info *seg_i; | |
1448 | int written_size = 0; | |
1449 | int i, j; | |
1450 | ||
1451 | page = grab_meta_page(sbi, blkaddr++); | |
1452 | kaddr = (unsigned char *)page_address(page); | |
1453 | ||
1454 | /* Step 1: write nat cache */ | |
1455 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
1456 | memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE); | |
1457 | written_size += SUM_JOURNAL_SIZE; | |
1458 | ||
1459 | /* Step 2: write sit cache */ | |
1460 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1461 | memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits, | |
1462 | SUM_JOURNAL_SIZE); | |
1463 | written_size += SUM_JOURNAL_SIZE; | |
1464 | ||
351df4b2 JK |
1465 | /* Step 3: write summary entries */ |
1466 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1467 | unsigned short blkoff; | |
1468 | seg_i = CURSEG_I(sbi, i); | |
1469 | if (sbi->ckpt->alloc_type[i] == SSR) | |
1470 | blkoff = sbi->blocks_per_seg; | |
1471 | else | |
1472 | blkoff = curseg_blkoff(sbi, i); | |
1473 | ||
1474 | for (j = 0; j < blkoff; j++) { | |
1475 | if (!page) { | |
1476 | page = grab_meta_page(sbi, blkaddr++); | |
1477 | kaddr = (unsigned char *)page_address(page); | |
1478 | written_size = 0; | |
1479 | } | |
1480 | summary = (struct f2fs_summary *)(kaddr + written_size); | |
1481 | *summary = seg_i->sum_blk->entries[j]; | |
1482 | written_size += SUMMARY_SIZE; | |
351df4b2 JK |
1483 | |
1484 | if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE - | |
1485 | SUM_FOOTER_SIZE) | |
1486 | continue; | |
1487 | ||
e8d61a74 | 1488 | set_page_dirty(page); |
351df4b2 JK |
1489 | f2fs_put_page(page, 1); |
1490 | page = NULL; | |
1491 | } | |
1492 | } | |
e8d61a74 CY |
1493 | if (page) { |
1494 | set_page_dirty(page); | |
351df4b2 | 1495 | f2fs_put_page(page, 1); |
e8d61a74 | 1496 | } |
351df4b2 JK |
1497 | } |
1498 | ||
1499 | static void write_normal_summaries(struct f2fs_sb_info *sbi, | |
1500 | block_t blkaddr, int type) | |
1501 | { | |
1502 | int i, end; | |
1503 | if (IS_DATASEG(type)) | |
1504 | end = type + NR_CURSEG_DATA_TYPE; | |
1505 | else | |
1506 | end = type + NR_CURSEG_NODE_TYPE; | |
1507 | ||
1508 | for (i = type; i < end; i++) { | |
1509 | struct curseg_info *sum = CURSEG_I(sbi, i); | |
1510 | mutex_lock(&sum->curseg_mutex); | |
1511 | write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type)); | |
1512 | mutex_unlock(&sum->curseg_mutex); | |
1513 | } | |
1514 | } | |
1515 | ||
1516 | void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1517 | { | |
25ca923b | 1518 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) |
351df4b2 JK |
1519 | write_compacted_summaries(sbi, start_blk); |
1520 | else | |
1521 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); | |
1522 | } | |
1523 | ||
1524 | void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1525 | { | |
25ca923b | 1526 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG)) |
351df4b2 | 1527 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); |
351df4b2 JK |
1528 | } |
1529 | ||
1530 | int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type, | |
1531 | unsigned int val, int alloc) | |
1532 | { | |
1533 | int i; | |
1534 | ||
1535 | if (type == NAT_JOURNAL) { | |
1536 | for (i = 0; i < nats_in_cursum(sum); i++) { | |
1537 | if (le32_to_cpu(nid_in_journal(sum, i)) == val) | |
1538 | return i; | |
1539 | } | |
1540 | if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) | |
1541 | return update_nats_in_cursum(sum, 1); | |
1542 | } else if (type == SIT_JOURNAL) { | |
1543 | for (i = 0; i < sits_in_cursum(sum); i++) | |
1544 | if (le32_to_cpu(segno_in_journal(sum, i)) == val) | |
1545 | return i; | |
1546 | if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES) | |
1547 | return update_sits_in_cursum(sum, 1); | |
1548 | } | |
1549 | return -1; | |
1550 | } | |
1551 | ||
1552 | static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, | |
1553 | unsigned int segno) | |
1554 | { | |
1555 | struct sit_info *sit_i = SIT_I(sbi); | |
d3a14afd | 1556 | unsigned int offset = SIT_BLOCK_OFFSET(segno); |
351df4b2 JK |
1557 | block_t blk_addr = sit_i->sit_base_addr + offset; |
1558 | ||
1559 | check_seg_range(sbi, segno); | |
1560 | ||
1561 | /* calculate sit block address */ | |
1562 | if (f2fs_test_bit(offset, sit_i->sit_bitmap)) | |
1563 | blk_addr += sit_i->sit_blocks; | |
1564 | ||
1565 | return get_meta_page(sbi, blk_addr); | |
1566 | } | |
1567 | ||
1568 | static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, | |
1569 | unsigned int start) | |
1570 | { | |
1571 | struct sit_info *sit_i = SIT_I(sbi); | |
1572 | struct page *src_page, *dst_page; | |
1573 | pgoff_t src_off, dst_off; | |
1574 | void *src_addr, *dst_addr; | |
1575 | ||
1576 | src_off = current_sit_addr(sbi, start); | |
1577 | dst_off = next_sit_addr(sbi, src_off); | |
1578 | ||
1579 | /* get current sit block page without lock */ | |
1580 | src_page = get_meta_page(sbi, src_off); | |
1581 | dst_page = grab_meta_page(sbi, dst_off); | |
9850cf4a | 1582 | f2fs_bug_on(sbi, PageDirty(src_page)); |
351df4b2 JK |
1583 | |
1584 | src_addr = page_address(src_page); | |
1585 | dst_addr = page_address(dst_page); | |
1586 | memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); | |
1587 | ||
1588 | set_page_dirty(dst_page); | |
1589 | f2fs_put_page(src_page, 1); | |
1590 | ||
1591 | set_to_next_sit(sit_i, start); | |
1592 | ||
1593 | return dst_page; | |
1594 | } | |
1595 | ||
184a5cd2 CY |
1596 | static struct sit_entry_set *grab_sit_entry_set(void) |
1597 | { | |
1598 | struct sit_entry_set *ses = | |
1599 | f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC); | |
1600 | ||
1601 | ses->entry_cnt = 0; | |
1602 | INIT_LIST_HEAD(&ses->set_list); | |
1603 | return ses; | |
1604 | } | |
1605 | ||
1606 | static void release_sit_entry_set(struct sit_entry_set *ses) | |
1607 | { | |
1608 | list_del(&ses->set_list); | |
1609 | kmem_cache_free(sit_entry_set_slab, ses); | |
1610 | } | |
1611 | ||
1612 | static void adjust_sit_entry_set(struct sit_entry_set *ses, | |
1613 | struct list_head *head) | |
1614 | { | |
1615 | struct sit_entry_set *next = ses; | |
1616 | ||
1617 | if (list_is_last(&ses->set_list, head)) | |
1618 | return; | |
1619 | ||
1620 | list_for_each_entry_continue(next, head, set_list) | |
1621 | if (ses->entry_cnt <= next->entry_cnt) | |
1622 | break; | |
1623 | ||
1624 | list_move_tail(&ses->set_list, &next->set_list); | |
1625 | } | |
1626 | ||
1627 | static void add_sit_entry(unsigned int segno, struct list_head *head) | |
1628 | { | |
1629 | struct sit_entry_set *ses; | |
1630 | unsigned int start_segno = START_SEGNO(segno); | |
1631 | ||
1632 | list_for_each_entry(ses, head, set_list) { | |
1633 | if (ses->start_segno == start_segno) { | |
1634 | ses->entry_cnt++; | |
1635 | adjust_sit_entry_set(ses, head); | |
1636 | return; | |
1637 | } | |
1638 | } | |
1639 | ||
1640 | ses = grab_sit_entry_set(); | |
1641 | ||
1642 | ses->start_segno = start_segno; | |
1643 | ses->entry_cnt++; | |
1644 | list_add(&ses->set_list, head); | |
1645 | } | |
1646 | ||
1647 | static void add_sits_in_set(struct f2fs_sb_info *sbi) | |
1648 | { | |
1649 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
1650 | struct list_head *set_list = &sm_info->sit_entry_set; | |
1651 | unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; | |
184a5cd2 CY |
1652 | unsigned int segno; |
1653 | ||
7cd8558b | 1654 | for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) |
184a5cd2 CY |
1655 | add_sit_entry(segno, set_list); |
1656 | } | |
1657 | ||
1658 | static void remove_sits_in_journal(struct f2fs_sb_info *sbi) | |
351df4b2 JK |
1659 | { |
1660 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1661 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
1662 | int i; | |
1663 | ||
184a5cd2 CY |
1664 | for (i = sits_in_cursum(sum) - 1; i >= 0; i--) { |
1665 | unsigned int segno; | |
1666 | bool dirtied; | |
1667 | ||
1668 | segno = le32_to_cpu(segno_in_journal(sum, i)); | |
1669 | dirtied = __mark_sit_entry_dirty(sbi, segno); | |
1670 | ||
1671 | if (!dirtied) | |
1672 | add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); | |
351df4b2 | 1673 | } |
184a5cd2 | 1674 | update_sits_in_cursum(sum, -sits_in_cursum(sum)); |
351df4b2 JK |
1675 | } |
1676 | ||
0a8165d7 | 1677 | /* |
351df4b2 JK |
1678 | * CP calls this function, which flushes SIT entries including sit_journal, |
1679 | * and moves prefree segs to free segs. | |
1680 | */ | |
4b2fecc8 | 1681 | void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) |
351df4b2 JK |
1682 | { |
1683 | struct sit_info *sit_i = SIT_I(sbi); | |
1684 | unsigned long *bitmap = sit_i->dirty_sentries_bitmap; | |
1685 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1686 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
184a5cd2 CY |
1687 | struct sit_entry_set *ses, *tmp; |
1688 | struct list_head *head = &SM_I(sbi)->sit_entry_set; | |
184a5cd2 | 1689 | bool to_journal = true; |
4b2fecc8 | 1690 | struct seg_entry *se; |
351df4b2 JK |
1691 | |
1692 | mutex_lock(&curseg->curseg_mutex); | |
1693 | mutex_lock(&sit_i->sentry_lock); | |
1694 | ||
1695 | /* | |
184a5cd2 CY |
1696 | * add and account sit entries of dirty bitmap in sit entry |
1697 | * set temporarily | |
351df4b2 | 1698 | */ |
184a5cd2 | 1699 | add_sits_in_set(sbi); |
351df4b2 | 1700 | |
184a5cd2 CY |
1701 | /* |
1702 | * if there are no enough space in journal to store dirty sit | |
1703 | * entries, remove all entries from journal and add and account | |
1704 | * them in sit entry set. | |
1705 | */ | |
1706 | if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL)) | |
1707 | remove_sits_in_journal(sbi); | |
b2955550 | 1708 | |
184a5cd2 CY |
1709 | if (!sit_i->dirty_sentries) |
1710 | goto out; | |
351df4b2 | 1711 | |
184a5cd2 CY |
1712 | /* |
1713 | * there are two steps to flush sit entries: | |
1714 | * #1, flush sit entries to journal in current cold data summary block. | |
1715 | * #2, flush sit entries to sit page. | |
1716 | */ | |
1717 | list_for_each_entry_safe(ses, tmp, head, set_list) { | |
4a257ed6 | 1718 | struct page *page = NULL; |
184a5cd2 CY |
1719 | struct f2fs_sit_block *raw_sit = NULL; |
1720 | unsigned int start_segno = ses->start_segno; | |
1721 | unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, | |
7cd8558b | 1722 | (unsigned long)MAIN_SEGS(sbi)); |
184a5cd2 CY |
1723 | unsigned int segno = start_segno; |
1724 | ||
1725 | if (to_journal && | |
1726 | !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL)) | |
1727 | to_journal = false; | |
1728 | ||
1729 | if (!to_journal) { | |
1730 | page = get_next_sit_page(sbi, start_segno); | |
1731 | raw_sit = page_address(page); | |
351df4b2 | 1732 | } |
351df4b2 | 1733 | |
184a5cd2 CY |
1734 | /* flush dirty sit entries in region of current sit set */ |
1735 | for_each_set_bit_from(segno, bitmap, end) { | |
1736 | int offset, sit_offset; | |
4b2fecc8 JK |
1737 | |
1738 | se = get_seg_entry(sbi, segno); | |
184a5cd2 CY |
1739 | |
1740 | /* add discard candidates */ | |
4b2fecc8 JK |
1741 | if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards) { |
1742 | cpc->trim_start = segno; | |
1743 | add_discard_addrs(sbi, cpc); | |
1744 | } | |
184a5cd2 CY |
1745 | |
1746 | if (to_journal) { | |
1747 | offset = lookup_journal_in_cursum(sum, | |
1748 | SIT_JOURNAL, segno, 1); | |
1749 | f2fs_bug_on(sbi, offset < 0); | |
1750 | segno_in_journal(sum, offset) = | |
1751 | cpu_to_le32(segno); | |
1752 | seg_info_to_raw_sit(se, | |
1753 | &sit_in_journal(sum, offset)); | |
1754 | } else { | |
1755 | sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); | |
1756 | seg_info_to_raw_sit(se, | |
1757 | &raw_sit->entries[sit_offset]); | |
1758 | } | |
351df4b2 | 1759 | |
184a5cd2 CY |
1760 | __clear_bit(segno, bitmap); |
1761 | sit_i->dirty_sentries--; | |
1762 | ses->entry_cnt--; | |
351df4b2 JK |
1763 | } |
1764 | ||
184a5cd2 CY |
1765 | if (!to_journal) |
1766 | f2fs_put_page(page, 1); | |
1767 | ||
1768 | f2fs_bug_on(sbi, ses->entry_cnt); | |
1769 | release_sit_entry_set(ses); | |
351df4b2 | 1770 | } |
184a5cd2 CY |
1771 | |
1772 | f2fs_bug_on(sbi, !list_empty(head)); | |
1773 | f2fs_bug_on(sbi, sit_i->dirty_sentries); | |
184a5cd2 | 1774 | out: |
4b2fecc8 JK |
1775 | if (cpc->reason == CP_DISCARD) { |
1776 | for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) | |
1777 | add_discard_addrs(sbi, cpc); | |
1778 | } | |
351df4b2 JK |
1779 | mutex_unlock(&sit_i->sentry_lock); |
1780 | mutex_unlock(&curseg->curseg_mutex); | |
1781 | ||
351df4b2 JK |
1782 | set_prefree_as_free_segments(sbi); |
1783 | } | |
1784 | ||
1785 | static int build_sit_info(struct f2fs_sb_info *sbi) | |
1786 | { | |
1787 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
1788 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1789 | struct sit_info *sit_i; | |
1790 | unsigned int sit_segs, start; | |
1791 | char *src_bitmap, *dst_bitmap; | |
1792 | unsigned int bitmap_size; | |
1793 | ||
1794 | /* allocate memory for SIT information */ | |
1795 | sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); | |
1796 | if (!sit_i) | |
1797 | return -ENOMEM; | |
1798 | ||
1799 | SM_I(sbi)->sit_info = sit_i; | |
1800 | ||
7cd8558b | 1801 | sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry)); |
351df4b2 JK |
1802 | if (!sit_i->sentries) |
1803 | return -ENOMEM; | |
1804 | ||
7cd8558b | 1805 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
351df4b2 JK |
1806 | sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL); |
1807 | if (!sit_i->dirty_sentries_bitmap) | |
1808 | return -ENOMEM; | |
1809 | ||
7cd8558b | 1810 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
1811 | sit_i->sentries[start].cur_valid_map |
1812 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
1813 | sit_i->sentries[start].ckpt_valid_map | |
1814 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
1815 | if (!sit_i->sentries[start].cur_valid_map | |
1816 | || !sit_i->sentries[start].ckpt_valid_map) | |
1817 | return -ENOMEM; | |
1818 | } | |
1819 | ||
1820 | if (sbi->segs_per_sec > 1) { | |
7cd8558b | 1821 | sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) * |
351df4b2 JK |
1822 | sizeof(struct sec_entry)); |
1823 | if (!sit_i->sec_entries) | |
1824 | return -ENOMEM; | |
1825 | } | |
1826 | ||
1827 | /* get information related with SIT */ | |
1828 | sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; | |
1829 | ||
1830 | /* setup SIT bitmap from ckeckpoint pack */ | |
1831 | bitmap_size = __bitmap_size(sbi, SIT_BITMAP); | |
1832 | src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); | |
1833 | ||
79b5793b | 1834 | dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); |
351df4b2 JK |
1835 | if (!dst_bitmap) |
1836 | return -ENOMEM; | |
351df4b2 JK |
1837 | |
1838 | /* init SIT information */ | |
1839 | sit_i->s_ops = &default_salloc_ops; | |
1840 | ||
1841 | sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); | |
1842 | sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; | |
1843 | sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count); | |
1844 | sit_i->sit_bitmap = dst_bitmap; | |
1845 | sit_i->bitmap_size = bitmap_size; | |
1846 | sit_i->dirty_sentries = 0; | |
1847 | sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; | |
1848 | sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); | |
1849 | sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; | |
1850 | mutex_init(&sit_i->sentry_lock); | |
1851 | return 0; | |
1852 | } | |
1853 | ||
1854 | static int build_free_segmap(struct f2fs_sb_info *sbi) | |
1855 | { | |
351df4b2 JK |
1856 | struct free_segmap_info *free_i; |
1857 | unsigned int bitmap_size, sec_bitmap_size; | |
1858 | ||
1859 | /* allocate memory for free segmap information */ | |
1860 | free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); | |
1861 | if (!free_i) | |
1862 | return -ENOMEM; | |
1863 | ||
1864 | SM_I(sbi)->free_info = free_i; | |
1865 | ||
7cd8558b | 1866 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
351df4b2 JK |
1867 | free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL); |
1868 | if (!free_i->free_segmap) | |
1869 | return -ENOMEM; | |
1870 | ||
7cd8558b | 1871 | sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); |
351df4b2 JK |
1872 | free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL); |
1873 | if (!free_i->free_secmap) | |
1874 | return -ENOMEM; | |
1875 | ||
1876 | /* set all segments as dirty temporarily */ | |
1877 | memset(free_i->free_segmap, 0xff, bitmap_size); | |
1878 | memset(free_i->free_secmap, 0xff, sec_bitmap_size); | |
1879 | ||
1880 | /* init free segmap information */ | |
7cd8558b | 1881 | free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); |
351df4b2 JK |
1882 | free_i->free_segments = 0; |
1883 | free_i->free_sections = 0; | |
1884 | rwlock_init(&free_i->segmap_lock); | |
1885 | return 0; | |
1886 | } | |
1887 | ||
1888 | static int build_curseg(struct f2fs_sb_info *sbi) | |
1889 | { | |
1042d60f | 1890 | struct curseg_info *array; |
351df4b2 JK |
1891 | int i; |
1892 | ||
b434babf | 1893 | array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL); |
351df4b2 JK |
1894 | if (!array) |
1895 | return -ENOMEM; | |
1896 | ||
1897 | SM_I(sbi)->curseg_array = array; | |
1898 | ||
1899 | for (i = 0; i < NR_CURSEG_TYPE; i++) { | |
1900 | mutex_init(&array[i].curseg_mutex); | |
1901 | array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL); | |
1902 | if (!array[i].sum_blk) | |
1903 | return -ENOMEM; | |
1904 | array[i].segno = NULL_SEGNO; | |
1905 | array[i].next_blkoff = 0; | |
1906 | } | |
1907 | return restore_curseg_summaries(sbi); | |
1908 | } | |
1909 | ||
1910 | static void build_sit_entries(struct f2fs_sb_info *sbi) | |
1911 | { | |
1912 | struct sit_info *sit_i = SIT_I(sbi); | |
1913 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1914 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
74de593a CY |
1915 | int sit_blk_cnt = SIT_BLK_CNT(sbi); |
1916 | unsigned int i, start, end; | |
1917 | unsigned int readed, start_blk = 0; | |
90a893c7 | 1918 | int nrpages = MAX_BIO_BLOCKS(sbi); |
351df4b2 | 1919 | |
74de593a | 1920 | do { |
662befda | 1921 | readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT); |
74de593a CY |
1922 | |
1923 | start = start_blk * sit_i->sents_per_block; | |
1924 | end = (start_blk + readed) * sit_i->sents_per_block; | |
1925 | ||
7cd8558b | 1926 | for (; start < end && start < MAIN_SEGS(sbi); start++) { |
74de593a CY |
1927 | struct seg_entry *se = &sit_i->sentries[start]; |
1928 | struct f2fs_sit_block *sit_blk; | |
1929 | struct f2fs_sit_entry sit; | |
1930 | struct page *page; | |
1931 | ||
1932 | mutex_lock(&curseg->curseg_mutex); | |
1933 | for (i = 0; i < sits_in_cursum(sum); i++) { | |
6c311ec6 CF |
1934 | if (le32_to_cpu(segno_in_journal(sum, i)) |
1935 | == start) { | |
74de593a CY |
1936 | sit = sit_in_journal(sum, i); |
1937 | mutex_unlock(&curseg->curseg_mutex); | |
1938 | goto got_it; | |
1939 | } | |
351df4b2 | 1940 | } |
74de593a CY |
1941 | mutex_unlock(&curseg->curseg_mutex); |
1942 | ||
1943 | page = get_current_sit_page(sbi, start); | |
1944 | sit_blk = (struct f2fs_sit_block *)page_address(page); | |
1945 | sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; | |
1946 | f2fs_put_page(page, 1); | |
351df4b2 | 1947 | got_it: |
74de593a CY |
1948 | check_block_count(sbi, start, &sit); |
1949 | seg_info_from_raw_sit(se, &sit); | |
1950 | if (sbi->segs_per_sec > 1) { | |
1951 | struct sec_entry *e = get_sec_entry(sbi, start); | |
1952 | e->valid_blocks += se->valid_blocks; | |
1953 | } | |
351df4b2 | 1954 | } |
74de593a CY |
1955 | start_blk += readed; |
1956 | } while (start_blk < sit_blk_cnt); | |
351df4b2 JK |
1957 | } |
1958 | ||
1959 | static void init_free_segmap(struct f2fs_sb_info *sbi) | |
1960 | { | |
1961 | unsigned int start; | |
1962 | int type; | |
1963 | ||
7cd8558b | 1964 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
1965 | struct seg_entry *sentry = get_seg_entry(sbi, start); |
1966 | if (!sentry->valid_blocks) | |
1967 | __set_free(sbi, start); | |
1968 | } | |
1969 | ||
1970 | /* set use the current segments */ | |
1971 | for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { | |
1972 | struct curseg_info *curseg_t = CURSEG_I(sbi, type); | |
1973 | __set_test_and_inuse(sbi, curseg_t->segno); | |
1974 | } | |
1975 | } | |
1976 | ||
1977 | static void init_dirty_segmap(struct f2fs_sb_info *sbi) | |
1978 | { | |
1979 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1980 | struct free_segmap_info *free_i = FREE_I(sbi); | |
7cd8558b | 1981 | unsigned int segno = 0, offset = 0; |
351df4b2 JK |
1982 | unsigned short valid_blocks; |
1983 | ||
8736fbf0 | 1984 | while (1) { |
351df4b2 | 1985 | /* find dirty segment based on free segmap */ |
7cd8558b JK |
1986 | segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); |
1987 | if (segno >= MAIN_SEGS(sbi)) | |
351df4b2 JK |
1988 | break; |
1989 | offset = segno + 1; | |
1990 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
ec325b52 | 1991 | if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) |
351df4b2 | 1992 | continue; |
ec325b52 JK |
1993 | if (valid_blocks > sbi->blocks_per_seg) { |
1994 | f2fs_bug_on(sbi, 1); | |
1995 | continue; | |
1996 | } | |
351df4b2 JK |
1997 | mutex_lock(&dirty_i->seglist_lock); |
1998 | __locate_dirty_segment(sbi, segno, DIRTY); | |
1999 | mutex_unlock(&dirty_i->seglist_lock); | |
2000 | } | |
2001 | } | |
2002 | ||
5ec4e49f | 2003 | static int init_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
2004 | { |
2005 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
7cd8558b | 2006 | unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); |
351df4b2 | 2007 | |
5ec4e49f JK |
2008 | dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL); |
2009 | if (!dirty_i->victim_secmap) | |
351df4b2 JK |
2010 | return -ENOMEM; |
2011 | return 0; | |
2012 | } | |
2013 | ||
2014 | static int build_dirty_segmap(struct f2fs_sb_info *sbi) | |
2015 | { | |
2016 | struct dirty_seglist_info *dirty_i; | |
2017 | unsigned int bitmap_size, i; | |
2018 | ||
2019 | /* allocate memory for dirty segments list information */ | |
2020 | dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); | |
2021 | if (!dirty_i) | |
2022 | return -ENOMEM; | |
2023 | ||
2024 | SM_I(sbi)->dirty_info = dirty_i; | |
2025 | mutex_init(&dirty_i->seglist_lock); | |
2026 | ||
7cd8558b | 2027 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
351df4b2 JK |
2028 | |
2029 | for (i = 0; i < NR_DIRTY_TYPE; i++) { | |
2030 | dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL); | |
351df4b2 JK |
2031 | if (!dirty_i->dirty_segmap[i]) |
2032 | return -ENOMEM; | |
2033 | } | |
2034 | ||
2035 | init_dirty_segmap(sbi); | |
5ec4e49f | 2036 | return init_victim_secmap(sbi); |
351df4b2 JK |
2037 | } |
2038 | ||
0a8165d7 | 2039 | /* |
351df4b2 JK |
2040 | * Update min, max modified time for cost-benefit GC algorithm |
2041 | */ | |
2042 | static void init_min_max_mtime(struct f2fs_sb_info *sbi) | |
2043 | { | |
2044 | struct sit_info *sit_i = SIT_I(sbi); | |
2045 | unsigned int segno; | |
2046 | ||
2047 | mutex_lock(&sit_i->sentry_lock); | |
2048 | ||
2049 | sit_i->min_mtime = LLONG_MAX; | |
2050 | ||
7cd8558b | 2051 | for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { |
351df4b2 JK |
2052 | unsigned int i; |
2053 | unsigned long long mtime = 0; | |
2054 | ||
2055 | for (i = 0; i < sbi->segs_per_sec; i++) | |
2056 | mtime += get_seg_entry(sbi, segno + i)->mtime; | |
2057 | ||
2058 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
2059 | ||
2060 | if (sit_i->min_mtime > mtime) | |
2061 | sit_i->min_mtime = mtime; | |
2062 | } | |
2063 | sit_i->max_mtime = get_mtime(sbi); | |
2064 | mutex_unlock(&sit_i->sentry_lock); | |
2065 | } | |
2066 | ||
2067 | int build_segment_manager(struct f2fs_sb_info *sbi) | |
2068 | { | |
2069 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
2070 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1042d60f | 2071 | struct f2fs_sm_info *sm_info; |
351df4b2 JK |
2072 | int err; |
2073 | ||
2074 | sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); | |
2075 | if (!sm_info) | |
2076 | return -ENOMEM; | |
2077 | ||
2078 | /* init sm info */ | |
2079 | sbi->sm_info = sm_info; | |
351df4b2 JK |
2080 | sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); |
2081 | sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); | |
2082 | sm_info->segment_count = le32_to_cpu(raw_super->segment_count); | |
2083 | sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); | |
2084 | sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); | |
2085 | sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); | |
2086 | sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); | |
58c41035 JK |
2087 | sm_info->rec_prefree_segments = sm_info->main_segments * |
2088 | DEF_RECLAIM_PREFREE_SEGMENTS / 100; | |
9b5f136f | 2089 | sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; |
216fbd64 | 2090 | sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; |
c1ce1b02 | 2091 | sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; |
351df4b2 | 2092 | |
7fd9e544 JK |
2093 | INIT_LIST_HEAD(&sm_info->discard_list); |
2094 | sm_info->nr_discards = 0; | |
2095 | sm_info->max_discards = 0; | |
2096 | ||
184a5cd2 CY |
2097 | INIT_LIST_HEAD(&sm_info->sit_entry_set); |
2098 | ||
b270ad6f | 2099 | if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) { |
2163d198 GZ |
2100 | err = create_flush_cmd_control(sbi); |
2101 | if (err) | |
a688b9d9 | 2102 | return err; |
6b4afdd7 JK |
2103 | } |
2104 | ||
351df4b2 JK |
2105 | err = build_sit_info(sbi); |
2106 | if (err) | |
2107 | return err; | |
2108 | err = build_free_segmap(sbi); | |
2109 | if (err) | |
2110 | return err; | |
2111 | err = build_curseg(sbi); | |
2112 | if (err) | |
2113 | return err; | |
2114 | ||
2115 | /* reinit free segmap based on SIT */ | |
2116 | build_sit_entries(sbi); | |
2117 | ||
2118 | init_free_segmap(sbi); | |
2119 | err = build_dirty_segmap(sbi); | |
2120 | if (err) | |
2121 | return err; | |
2122 | ||
2123 | init_min_max_mtime(sbi); | |
2124 | return 0; | |
2125 | } | |
2126 | ||
2127 | static void discard_dirty_segmap(struct f2fs_sb_info *sbi, | |
2128 | enum dirty_type dirty_type) | |
2129 | { | |
2130 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
2131 | ||
2132 | mutex_lock(&dirty_i->seglist_lock); | |
2133 | kfree(dirty_i->dirty_segmap[dirty_type]); | |
2134 | dirty_i->nr_dirty[dirty_type] = 0; | |
2135 | mutex_unlock(&dirty_i->seglist_lock); | |
2136 | } | |
2137 | ||
5ec4e49f | 2138 | static void destroy_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
2139 | { |
2140 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 2141 | kfree(dirty_i->victim_secmap); |
351df4b2 JK |
2142 | } |
2143 | ||
2144 | static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) | |
2145 | { | |
2146 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
2147 | int i; | |
2148 | ||
2149 | if (!dirty_i) | |
2150 | return; | |
2151 | ||
2152 | /* discard pre-free/dirty segments list */ | |
2153 | for (i = 0; i < NR_DIRTY_TYPE; i++) | |
2154 | discard_dirty_segmap(sbi, i); | |
2155 | ||
5ec4e49f | 2156 | destroy_victim_secmap(sbi); |
351df4b2 JK |
2157 | SM_I(sbi)->dirty_info = NULL; |
2158 | kfree(dirty_i); | |
2159 | } | |
2160 | ||
2161 | static void destroy_curseg(struct f2fs_sb_info *sbi) | |
2162 | { | |
2163 | struct curseg_info *array = SM_I(sbi)->curseg_array; | |
2164 | int i; | |
2165 | ||
2166 | if (!array) | |
2167 | return; | |
2168 | SM_I(sbi)->curseg_array = NULL; | |
2169 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
2170 | kfree(array[i].sum_blk); | |
2171 | kfree(array); | |
2172 | } | |
2173 | ||
2174 | static void destroy_free_segmap(struct f2fs_sb_info *sbi) | |
2175 | { | |
2176 | struct free_segmap_info *free_i = SM_I(sbi)->free_info; | |
2177 | if (!free_i) | |
2178 | return; | |
2179 | SM_I(sbi)->free_info = NULL; | |
2180 | kfree(free_i->free_segmap); | |
2181 | kfree(free_i->free_secmap); | |
2182 | kfree(free_i); | |
2183 | } | |
2184 | ||
2185 | static void destroy_sit_info(struct f2fs_sb_info *sbi) | |
2186 | { | |
2187 | struct sit_info *sit_i = SIT_I(sbi); | |
2188 | unsigned int start; | |
2189 | ||
2190 | if (!sit_i) | |
2191 | return; | |
2192 | ||
2193 | if (sit_i->sentries) { | |
7cd8558b | 2194 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
2195 | kfree(sit_i->sentries[start].cur_valid_map); |
2196 | kfree(sit_i->sentries[start].ckpt_valid_map); | |
2197 | } | |
2198 | } | |
2199 | vfree(sit_i->sentries); | |
2200 | vfree(sit_i->sec_entries); | |
2201 | kfree(sit_i->dirty_sentries_bitmap); | |
2202 | ||
2203 | SM_I(sbi)->sit_info = NULL; | |
2204 | kfree(sit_i->sit_bitmap); | |
2205 | kfree(sit_i); | |
2206 | } | |
2207 | ||
2208 | void destroy_segment_manager(struct f2fs_sb_info *sbi) | |
2209 | { | |
2210 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
a688b9d9 | 2211 | |
3b03f724 CY |
2212 | if (!sm_info) |
2213 | return; | |
2163d198 | 2214 | destroy_flush_cmd_control(sbi); |
351df4b2 JK |
2215 | destroy_dirty_segmap(sbi); |
2216 | destroy_curseg(sbi); | |
2217 | destroy_free_segmap(sbi); | |
2218 | destroy_sit_info(sbi); | |
2219 | sbi->sm_info = NULL; | |
2220 | kfree(sm_info); | |
2221 | } | |
7fd9e544 JK |
2222 | |
2223 | int __init create_segment_manager_caches(void) | |
2224 | { | |
2225 | discard_entry_slab = f2fs_kmem_cache_create("discard_entry", | |
e8512d2e | 2226 | sizeof(struct discard_entry)); |
7fd9e544 | 2227 | if (!discard_entry_slab) |
184a5cd2 CY |
2228 | goto fail; |
2229 | ||
2230 | sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", | |
2231 | sizeof(struct nat_entry_set)); | |
2232 | if (!sit_entry_set_slab) | |
2233 | goto destory_discard_entry; | |
88b88a66 JK |
2234 | |
2235 | inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", | |
2236 | sizeof(struct inmem_pages)); | |
2237 | if (!inmem_entry_slab) | |
2238 | goto destroy_sit_entry_set; | |
7fd9e544 | 2239 | return 0; |
184a5cd2 | 2240 | |
88b88a66 JK |
2241 | destroy_sit_entry_set: |
2242 | kmem_cache_destroy(sit_entry_set_slab); | |
184a5cd2 CY |
2243 | destory_discard_entry: |
2244 | kmem_cache_destroy(discard_entry_slab); | |
2245 | fail: | |
2246 | return -ENOMEM; | |
7fd9e544 JK |
2247 | } |
2248 | ||
2249 | void destroy_segment_manager_caches(void) | |
2250 | { | |
184a5cd2 | 2251 | kmem_cache_destroy(sit_entry_set_slab); |
7fd9e544 | 2252 | kmem_cache_destroy(discard_entry_slab); |
88b88a66 | 2253 | kmem_cache_destroy(inmem_entry_slab); |
7fd9e544 | 2254 | } |