]>
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 | 181 | int err; |
9be32d72 | 182 | |
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); | |
9be32d72 | 188 | retry: |
88b88a66 JK |
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); | |
34ba94ba JK |
198 | goto retry; |
199 | } | |
88b88a66 JK |
200 | get_page(page); |
201 | list_add_tail(&new->list, &fi->inmem_pages); | |
202 | mutex_unlock(&fi->inmem_lock); | |
203 | } | |
204 | ||
cbcb2872 JK |
205 | void invalidate_inmem_page(struct inode *inode, struct page *page) |
206 | { | |
207 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
208 | struct inmem_pages *cur; | |
209 | ||
210 | mutex_lock(&fi->inmem_lock); | |
211 | cur = radix_tree_lookup(&fi->inmem_root, page->index); | |
212 | if (cur) { | |
213 | radix_tree_delete(&fi->inmem_root, cur->page->index); | |
214 | f2fs_put_page(cur->page, 0); | |
215 | list_del(&cur->list); | |
216 | kmem_cache_free(inmem_entry_slab, cur); | |
217 | } | |
218 | mutex_unlock(&fi->inmem_lock); | |
219 | } | |
220 | ||
88b88a66 JK |
221 | void commit_inmem_pages(struct inode *inode, bool abort) |
222 | { | |
223 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
224 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
225 | struct inmem_pages *cur, *tmp; | |
226 | bool submit_bio = false; | |
227 | struct f2fs_io_info fio = { | |
228 | .type = DATA, | |
229 | .rw = WRITE_SYNC, | |
230 | }; | |
231 | ||
0341845e JK |
232 | /* |
233 | * The abort is true only when f2fs_evict_inode is called. | |
234 | * Basically, the f2fs_evict_inode doesn't produce any data writes, so | |
235 | * that we don't need to call f2fs_balance_fs. | |
236 | * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this | |
237 | * inode becomes free by iget_locked in f2fs_iget. | |
238 | */ | |
239 | if (!abort) | |
240 | f2fs_balance_fs(sbi); | |
241 | ||
88b88a66 JK |
242 | f2fs_lock_op(sbi); |
243 | ||
244 | mutex_lock(&fi->inmem_lock); | |
245 | list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { | |
246 | lock_page(cur->page); | |
247 | if (!abort && cur->page->mapping == inode->i_mapping) { | |
248 | f2fs_wait_on_page_writeback(cur->page, DATA); | |
249 | if (clear_page_dirty_for_io(cur->page)) | |
250 | inode_dec_dirty_pages(inode); | |
251 | do_write_data_page(cur->page, &fio); | |
252 | submit_bio = true; | |
253 | } | |
34ba94ba | 254 | radix_tree_delete(&fi->inmem_root, cur->page->index); |
88b88a66 JK |
255 | f2fs_put_page(cur->page, 1); |
256 | list_del(&cur->list); | |
257 | kmem_cache_free(inmem_entry_slab, cur); | |
258 | } | |
259 | if (submit_bio) | |
260 | f2fs_submit_merged_bio(sbi, DATA, WRITE); | |
261 | mutex_unlock(&fi->inmem_lock); | |
262 | ||
263 | filemap_fdatawait_range(inode->i_mapping, 0, LLONG_MAX); | |
264 | f2fs_unlock_op(sbi); | |
265 | } | |
266 | ||
0a8165d7 | 267 | /* |
351df4b2 JK |
268 | * This function balances dirty node and dentry pages. |
269 | * In addition, it controls garbage collection. | |
270 | */ | |
271 | void f2fs_balance_fs(struct f2fs_sb_info *sbi) | |
272 | { | |
351df4b2 | 273 | /* |
029cd28c JK |
274 | * We should do GC or end up with checkpoint, if there are so many dirty |
275 | * dir/node pages without enough free segments. | |
351df4b2 | 276 | */ |
43727527 | 277 | if (has_not_enough_free_secs(sbi, 0)) { |
351df4b2 | 278 | mutex_lock(&sbi->gc_mutex); |
408e9375 | 279 | f2fs_gc(sbi); |
351df4b2 JK |
280 | } |
281 | } | |
282 | ||
4660f9c0 JK |
283 | void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) |
284 | { | |
285 | /* check the # of cached NAT entries and prefree segments */ | |
286 | if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) || | |
e5e7ea3c JK |
287 | excess_prefree_segs(sbi) || |
288 | available_free_memory(sbi, INO_ENTRIES)) | |
4660f9c0 JK |
289 | f2fs_sync_fs(sbi->sb, true); |
290 | } | |
291 | ||
2163d198 | 292 | static int issue_flush_thread(void *data) |
6b4afdd7 JK |
293 | { |
294 | struct f2fs_sb_info *sbi = data; | |
a688b9d9 GZ |
295 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
296 | wait_queue_head_t *q = &fcc->flush_wait_queue; | |
6b4afdd7 JK |
297 | repeat: |
298 | if (kthread_should_stop()) | |
299 | return 0; | |
300 | ||
721bd4d5 | 301 | if (!llist_empty(&fcc->issue_list)) { |
6b4afdd7 JK |
302 | struct bio *bio = bio_alloc(GFP_NOIO, 0); |
303 | struct flush_cmd *cmd, *next; | |
304 | int ret; | |
305 | ||
721bd4d5 GZ |
306 | fcc->dispatch_list = llist_del_all(&fcc->issue_list); |
307 | fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); | |
308 | ||
6b4afdd7 JK |
309 | bio->bi_bdev = sbi->sb->s_bdev; |
310 | ret = submit_bio_wait(WRITE_FLUSH, bio); | |
311 | ||
721bd4d5 GZ |
312 | llist_for_each_entry_safe(cmd, next, |
313 | fcc->dispatch_list, llnode) { | |
6b4afdd7 | 314 | cmd->ret = ret; |
6b4afdd7 JK |
315 | complete(&cmd->wait); |
316 | } | |
a4ed23f2 | 317 | bio_put(bio); |
a688b9d9 | 318 | fcc->dispatch_list = NULL; |
6b4afdd7 JK |
319 | } |
320 | ||
a688b9d9 | 321 | wait_event_interruptible(*q, |
721bd4d5 | 322 | kthread_should_stop() || !llist_empty(&fcc->issue_list)); |
6b4afdd7 JK |
323 | goto repeat; |
324 | } | |
325 | ||
326 | int f2fs_issue_flush(struct f2fs_sb_info *sbi) | |
327 | { | |
a688b9d9 | 328 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
adf8d90b | 329 | struct flush_cmd cmd; |
6b4afdd7 | 330 | |
24a9ee0f JK |
331 | trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER), |
332 | test_opt(sbi, FLUSH_MERGE)); | |
333 | ||
0f7b2abd JK |
334 | if (test_opt(sbi, NOBARRIER)) |
335 | return 0; | |
336 | ||
6b4afdd7 JK |
337 | if (!test_opt(sbi, FLUSH_MERGE)) |
338 | return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL); | |
339 | ||
adf8d90b | 340 | init_completion(&cmd.wait); |
6b4afdd7 | 341 | |
721bd4d5 | 342 | llist_add(&cmd.llnode, &fcc->issue_list); |
6b4afdd7 | 343 | |
a688b9d9 GZ |
344 | if (!fcc->dispatch_list) |
345 | wake_up(&fcc->flush_wait_queue); | |
6b4afdd7 | 346 | |
adf8d90b CY |
347 | wait_for_completion(&cmd.wait); |
348 | ||
349 | return cmd.ret; | |
6b4afdd7 JK |
350 | } |
351 | ||
2163d198 GZ |
352 | int create_flush_cmd_control(struct f2fs_sb_info *sbi) |
353 | { | |
354 | dev_t dev = sbi->sb->s_bdev->bd_dev; | |
355 | struct flush_cmd_control *fcc; | |
356 | int err = 0; | |
357 | ||
358 | fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL); | |
359 | if (!fcc) | |
360 | return -ENOMEM; | |
2163d198 | 361 | init_waitqueue_head(&fcc->flush_wait_queue); |
721bd4d5 | 362 | init_llist_head(&fcc->issue_list); |
6b2920a5 | 363 | SM_I(sbi)->cmd_control_info = fcc; |
2163d198 GZ |
364 | fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, |
365 | "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); | |
366 | if (IS_ERR(fcc->f2fs_issue_flush)) { | |
367 | err = PTR_ERR(fcc->f2fs_issue_flush); | |
368 | kfree(fcc); | |
6b2920a5 | 369 | SM_I(sbi)->cmd_control_info = NULL; |
2163d198 GZ |
370 | return err; |
371 | } | |
2163d198 GZ |
372 | |
373 | return err; | |
374 | } | |
375 | ||
376 | void destroy_flush_cmd_control(struct f2fs_sb_info *sbi) | |
377 | { | |
6b2920a5 | 378 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
2163d198 GZ |
379 | |
380 | if (fcc && fcc->f2fs_issue_flush) | |
381 | kthread_stop(fcc->f2fs_issue_flush); | |
382 | kfree(fcc); | |
6b2920a5 | 383 | SM_I(sbi)->cmd_control_info = NULL; |
2163d198 GZ |
384 | } |
385 | ||
351df4b2 JK |
386 | static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, |
387 | enum dirty_type dirty_type) | |
388 | { | |
389 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
390 | ||
391 | /* need not be added */ | |
392 | if (IS_CURSEG(sbi, segno)) | |
393 | return; | |
394 | ||
395 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
396 | dirty_i->nr_dirty[dirty_type]++; | |
397 | ||
398 | if (dirty_type == DIRTY) { | |
399 | struct seg_entry *sentry = get_seg_entry(sbi, segno); | |
4625d6aa | 400 | enum dirty_type t = sentry->type; |
b2f2c390 | 401 | |
ec325b52 JK |
402 | if (unlikely(t >= DIRTY)) { |
403 | f2fs_bug_on(sbi, 1); | |
404 | return; | |
405 | } | |
4625d6aa CL |
406 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) |
407 | dirty_i->nr_dirty[t]++; | |
351df4b2 JK |
408 | } |
409 | } | |
410 | ||
411 | static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, | |
412 | enum dirty_type dirty_type) | |
413 | { | |
414 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
415 | ||
416 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
417 | dirty_i->nr_dirty[dirty_type]--; | |
418 | ||
419 | if (dirty_type == DIRTY) { | |
4625d6aa CL |
420 | struct seg_entry *sentry = get_seg_entry(sbi, segno); |
421 | enum dirty_type t = sentry->type; | |
422 | ||
423 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) | |
424 | dirty_i->nr_dirty[t]--; | |
b2f2c390 | 425 | |
5ec4e49f JK |
426 | if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0) |
427 | clear_bit(GET_SECNO(sbi, segno), | |
428 | dirty_i->victim_secmap); | |
351df4b2 JK |
429 | } |
430 | } | |
431 | ||
0a8165d7 | 432 | /* |
351df4b2 JK |
433 | * Should not occur error such as -ENOMEM. |
434 | * Adding dirty entry into seglist is not critical operation. | |
435 | * If a given segment is one of current working segments, it won't be added. | |
436 | */ | |
8d8451af | 437 | static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) |
351df4b2 JK |
438 | { |
439 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
440 | unsigned short valid_blocks; | |
441 | ||
442 | if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) | |
443 | return; | |
444 | ||
445 | mutex_lock(&dirty_i->seglist_lock); | |
446 | ||
447 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
448 | ||
449 | if (valid_blocks == 0) { | |
450 | __locate_dirty_segment(sbi, segno, PRE); | |
451 | __remove_dirty_segment(sbi, segno, DIRTY); | |
452 | } else if (valid_blocks < sbi->blocks_per_seg) { | |
453 | __locate_dirty_segment(sbi, segno, DIRTY); | |
454 | } else { | |
455 | /* Recovery routine with SSR needs this */ | |
456 | __remove_dirty_segment(sbi, segno, DIRTY); | |
457 | } | |
458 | ||
459 | mutex_unlock(&dirty_i->seglist_lock); | |
351df4b2 JK |
460 | } |
461 | ||
1e87a78d | 462 | static int f2fs_issue_discard(struct f2fs_sb_info *sbi, |
37208879 JK |
463 | block_t blkstart, block_t blklen) |
464 | { | |
55cf9cb6 CY |
465 | sector_t start = SECTOR_FROM_BLOCK(blkstart); |
466 | sector_t len = SECTOR_FROM_BLOCK(blklen); | |
1661d07c | 467 | trace_f2fs_issue_discard(sbi->sb, blkstart, blklen); |
1e87a78d JK |
468 | return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0); |
469 | } | |
470 | ||
cf2271e7 | 471 | void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr) |
1e87a78d | 472 | { |
1e87a78d JK |
473 | if (f2fs_issue_discard(sbi, blkaddr, 1)) { |
474 | struct page *page = grab_meta_page(sbi, blkaddr); | |
475 | /* zero-filled page */ | |
476 | set_page_dirty(page); | |
477 | f2fs_put_page(page, 1); | |
478 | } | |
37208879 JK |
479 | } |
480 | ||
adf4983b JK |
481 | static void __add_discard_entry(struct f2fs_sb_info *sbi, |
482 | struct cp_control *cpc, unsigned int start, unsigned int end) | |
b2955550 JK |
483 | { |
484 | struct list_head *head = &SM_I(sbi)->discard_list; | |
adf4983b JK |
485 | struct discard_entry *new, *last; |
486 | ||
487 | if (!list_empty(head)) { | |
488 | last = list_last_entry(head, struct discard_entry, list); | |
489 | if (START_BLOCK(sbi, cpc->trim_start) + start == | |
490 | last->blkaddr + last->len) { | |
491 | last->len += end - start; | |
492 | goto done; | |
493 | } | |
494 | } | |
495 | ||
496 | new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS); | |
497 | INIT_LIST_HEAD(&new->list); | |
498 | new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start; | |
499 | new->len = end - start; | |
500 | list_add_tail(&new->list, head); | |
501 | done: | |
502 | SM_I(sbi)->nr_discards += end - start; | |
503 | cpc->trimmed += end - start; | |
504 | } | |
505 | ||
506 | static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc) | |
507 | { | |
b2955550 JK |
508 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); |
509 | int max_blocks = sbi->blocks_per_seg; | |
4b2fecc8 | 510 | struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); |
b2955550 JK |
511 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; |
512 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; | |
513 | unsigned long dmap[entries]; | |
514 | unsigned int start = 0, end = -1; | |
4b2fecc8 | 515 | bool force = (cpc->reason == CP_DISCARD); |
b2955550 JK |
516 | int i; |
517 | ||
4b2fecc8 | 518 | if (!force && !test_opt(sbi, DISCARD)) |
b2955550 JK |
519 | return; |
520 | ||
4b2fecc8 JK |
521 | if (force && !se->valid_blocks) { |
522 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
523 | /* | |
524 | * if this segment is registered in the prefree list, then | |
525 | * we should skip adding a discard candidate, and let the | |
526 | * checkpoint do that later. | |
527 | */ | |
528 | mutex_lock(&dirty_i->seglist_lock); | |
529 | if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) { | |
530 | mutex_unlock(&dirty_i->seglist_lock); | |
531 | cpc->trimmed += sbi->blocks_per_seg; | |
532 | return; | |
533 | } | |
534 | mutex_unlock(&dirty_i->seglist_lock); | |
535 | ||
adf4983b | 536 | __add_discard_entry(sbi, cpc, 0, sbi->blocks_per_seg); |
4b2fecc8 JK |
537 | return; |
538 | } | |
539 | ||
b2955550 JK |
540 | /* zero block will be discarded through the prefree list */ |
541 | if (!se->valid_blocks || se->valid_blocks == max_blocks) | |
542 | return; | |
543 | ||
544 | /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ | |
545 | for (i = 0; i < entries; i++) | |
e3fb1b79 | 546 | dmap[i] = ~(cur_map[i] | ckpt_map[i]); |
b2955550 | 547 | |
4b2fecc8 | 548 | while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) { |
b2955550 JK |
549 | start = __find_rev_next_bit(dmap, max_blocks, end + 1); |
550 | if (start >= max_blocks) | |
551 | break; | |
552 | ||
553 | end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); | |
554 | ||
4b2fecc8 JK |
555 | if (end - start < cpc->trim_minlen) |
556 | continue; | |
557 | ||
adf4983b | 558 | __add_discard_entry(sbi, cpc, start, end); |
b2955550 JK |
559 | } |
560 | } | |
561 | ||
4b2fecc8 JK |
562 | void release_discard_addrs(struct f2fs_sb_info *sbi) |
563 | { | |
564 | struct list_head *head = &(SM_I(sbi)->discard_list); | |
565 | struct discard_entry *entry, *this; | |
566 | ||
567 | /* drop caches */ | |
568 | list_for_each_entry_safe(entry, this, head, list) { | |
569 | list_del(&entry->list); | |
570 | kmem_cache_free(discard_entry_slab, entry); | |
571 | } | |
572 | } | |
573 | ||
0a8165d7 | 574 | /* |
351df4b2 JK |
575 | * Should call clear_prefree_segments after checkpoint is done. |
576 | */ | |
577 | static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) | |
578 | { | |
579 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
b65ee148 | 580 | unsigned int segno; |
351df4b2 JK |
581 | |
582 | mutex_lock(&dirty_i->seglist_lock); | |
7cd8558b | 583 | for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) |
351df4b2 | 584 | __set_test_and_free(sbi, segno); |
351df4b2 JK |
585 | mutex_unlock(&dirty_i->seglist_lock); |
586 | } | |
587 | ||
588 | void clear_prefree_segments(struct f2fs_sb_info *sbi) | |
589 | { | |
b2955550 | 590 | struct list_head *head = &(SM_I(sbi)->discard_list); |
2d7b822a | 591 | struct discard_entry *entry, *this; |
351df4b2 | 592 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); |
29e59c14 | 593 | unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; |
29e59c14 | 594 | unsigned int start = 0, end = -1; |
351df4b2 JK |
595 | |
596 | mutex_lock(&dirty_i->seglist_lock); | |
29e59c14 | 597 | |
351df4b2 | 598 | while (1) { |
29e59c14 | 599 | int i; |
7cd8558b JK |
600 | start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); |
601 | if (start >= MAIN_SEGS(sbi)) | |
351df4b2 | 602 | break; |
7cd8558b JK |
603 | end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), |
604 | start + 1); | |
29e59c14 CL |
605 | |
606 | for (i = start; i < end; i++) | |
607 | clear_bit(i, prefree_map); | |
608 | ||
609 | dirty_i->nr_dirty[PRE] -= end - start; | |
610 | ||
611 | if (!test_opt(sbi, DISCARD)) | |
612 | continue; | |
351df4b2 | 613 | |
37208879 JK |
614 | f2fs_issue_discard(sbi, START_BLOCK(sbi, start), |
615 | (end - start) << sbi->log_blocks_per_seg); | |
351df4b2 JK |
616 | } |
617 | mutex_unlock(&dirty_i->seglist_lock); | |
b2955550 JK |
618 | |
619 | /* send small discards */ | |
2d7b822a | 620 | list_for_each_entry_safe(entry, this, head, list) { |
37208879 | 621 | f2fs_issue_discard(sbi, entry->blkaddr, entry->len); |
b2955550 JK |
622 | list_del(&entry->list); |
623 | SM_I(sbi)->nr_discards -= entry->len; | |
624 | kmem_cache_free(discard_entry_slab, entry); | |
625 | } | |
351df4b2 JK |
626 | } |
627 | ||
184a5cd2 | 628 | static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) |
351df4b2 JK |
629 | { |
630 | struct sit_info *sit_i = SIT_I(sbi); | |
184a5cd2 CY |
631 | |
632 | if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { | |
351df4b2 | 633 | sit_i->dirty_sentries++; |
184a5cd2 CY |
634 | return false; |
635 | } | |
636 | ||
637 | return true; | |
351df4b2 JK |
638 | } |
639 | ||
640 | static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, | |
641 | unsigned int segno, int modified) | |
642 | { | |
643 | struct seg_entry *se = get_seg_entry(sbi, segno); | |
644 | se->type = type; | |
645 | if (modified) | |
646 | __mark_sit_entry_dirty(sbi, segno); | |
647 | } | |
648 | ||
649 | static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) | |
650 | { | |
651 | struct seg_entry *se; | |
652 | unsigned int segno, offset; | |
653 | long int new_vblocks; | |
654 | ||
655 | segno = GET_SEGNO(sbi, blkaddr); | |
656 | ||
657 | se = get_seg_entry(sbi, segno); | |
658 | new_vblocks = se->valid_blocks + del; | |
491c0854 | 659 | offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); |
351df4b2 | 660 | |
9850cf4a | 661 | f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || |
351df4b2 JK |
662 | (new_vblocks > sbi->blocks_per_seg))); |
663 | ||
664 | se->valid_blocks = new_vblocks; | |
665 | se->mtime = get_mtime(sbi); | |
666 | SIT_I(sbi)->max_mtime = se->mtime; | |
667 | ||
668 | /* Update valid block bitmap */ | |
669 | if (del > 0) { | |
52aca074 | 670 | if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) |
05796763 | 671 | f2fs_bug_on(sbi, 1); |
351df4b2 | 672 | } else { |
52aca074 | 673 | if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) |
05796763 | 674 | f2fs_bug_on(sbi, 1); |
351df4b2 JK |
675 | } |
676 | if (!f2fs_test_bit(offset, se->ckpt_valid_map)) | |
677 | se->ckpt_valid_blocks += del; | |
678 | ||
679 | __mark_sit_entry_dirty(sbi, segno); | |
680 | ||
681 | /* update total number of valid blocks to be written in ckpt area */ | |
682 | SIT_I(sbi)->written_valid_blocks += del; | |
683 | ||
684 | if (sbi->segs_per_sec > 1) | |
685 | get_sec_entry(sbi, segno)->valid_blocks += del; | |
686 | } | |
687 | ||
5e443818 | 688 | void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) |
351df4b2 | 689 | { |
5e443818 JK |
690 | update_sit_entry(sbi, new, 1); |
691 | if (GET_SEGNO(sbi, old) != NULL_SEGNO) | |
692 | update_sit_entry(sbi, old, -1); | |
693 | ||
694 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); | |
695 | locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); | |
351df4b2 JK |
696 | } |
697 | ||
698 | void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) | |
699 | { | |
700 | unsigned int segno = GET_SEGNO(sbi, addr); | |
701 | struct sit_info *sit_i = SIT_I(sbi); | |
702 | ||
9850cf4a | 703 | f2fs_bug_on(sbi, addr == NULL_ADDR); |
351df4b2 JK |
704 | if (addr == NEW_ADDR) |
705 | return; | |
706 | ||
707 | /* add it into sit main buffer */ | |
708 | mutex_lock(&sit_i->sentry_lock); | |
709 | ||
710 | update_sit_entry(sbi, addr, -1); | |
711 | ||
712 | /* add it into dirty seglist */ | |
713 | locate_dirty_segment(sbi, segno); | |
714 | ||
715 | mutex_unlock(&sit_i->sentry_lock); | |
716 | } | |
717 | ||
0a8165d7 | 718 | /* |
351df4b2 JK |
719 | * This function should be resided under the curseg_mutex lock |
720 | */ | |
721 | static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, | |
e79efe3b | 722 | struct f2fs_summary *sum) |
351df4b2 JK |
723 | { |
724 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
725 | void *addr = curseg->sum_blk; | |
e79efe3b | 726 | addr += curseg->next_blkoff * sizeof(struct f2fs_summary); |
351df4b2 | 727 | memcpy(addr, sum, sizeof(struct f2fs_summary)); |
351df4b2 JK |
728 | } |
729 | ||
0a8165d7 | 730 | /* |
351df4b2 JK |
731 | * Calculate the number of current summary pages for writing |
732 | */ | |
733 | int npages_for_summary_flush(struct f2fs_sb_info *sbi) | |
734 | { | |
351df4b2 | 735 | int valid_sum_count = 0; |
9a47938b | 736 | int i, sum_in_page; |
351df4b2 JK |
737 | |
738 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
739 | if (sbi->ckpt->alloc_type[i] == SSR) | |
740 | valid_sum_count += sbi->blocks_per_seg; | |
741 | else | |
742 | valid_sum_count += curseg_blkoff(sbi, i); | |
743 | } | |
744 | ||
9a47938b FL |
745 | sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE - |
746 | SUM_FOOTER_SIZE) / SUMMARY_SIZE; | |
747 | if (valid_sum_count <= sum_in_page) | |
351df4b2 | 748 | return 1; |
9a47938b FL |
749 | else if ((valid_sum_count - sum_in_page) <= |
750 | (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) | |
351df4b2 JK |
751 | return 2; |
752 | return 3; | |
753 | } | |
754 | ||
0a8165d7 | 755 | /* |
351df4b2 JK |
756 | * Caller should put this summary page |
757 | */ | |
758 | struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) | |
759 | { | |
760 | return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); | |
761 | } | |
762 | ||
763 | static void write_sum_page(struct f2fs_sb_info *sbi, | |
764 | struct f2fs_summary_block *sum_blk, block_t blk_addr) | |
765 | { | |
766 | struct page *page = grab_meta_page(sbi, blk_addr); | |
767 | void *kaddr = page_address(page); | |
768 | memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE); | |
769 | set_page_dirty(page); | |
770 | f2fs_put_page(page, 1); | |
771 | } | |
772 | ||
60374688 JK |
773 | static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) |
774 | { | |
775 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
81fb5e87 | 776 | unsigned int segno = curseg->segno + 1; |
60374688 JK |
777 | struct free_segmap_info *free_i = FREE_I(sbi); |
778 | ||
7cd8558b | 779 | if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) |
81fb5e87 | 780 | return !test_bit(segno, free_i->free_segmap); |
60374688 JK |
781 | return 0; |
782 | } | |
783 | ||
0a8165d7 | 784 | /* |
351df4b2 JK |
785 | * Find a new segment from the free segments bitmap to right order |
786 | * This function should be returned with success, otherwise BUG | |
787 | */ | |
788 | static void get_new_segment(struct f2fs_sb_info *sbi, | |
789 | unsigned int *newseg, bool new_sec, int dir) | |
790 | { | |
791 | struct free_segmap_info *free_i = FREE_I(sbi); | |
351df4b2 | 792 | unsigned int segno, secno, zoneno; |
7cd8558b | 793 | unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; |
351df4b2 JK |
794 | unsigned int hint = *newseg / sbi->segs_per_sec; |
795 | unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); | |
796 | unsigned int left_start = hint; | |
797 | bool init = true; | |
798 | int go_left = 0; | |
799 | int i; | |
800 | ||
801 | write_lock(&free_i->segmap_lock); | |
802 | ||
803 | if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { | |
804 | segno = find_next_zero_bit(free_i->free_segmap, | |
7cd8558b | 805 | MAIN_SEGS(sbi), *newseg + 1); |
33afa7fd JK |
806 | if (segno - *newseg < sbi->segs_per_sec - |
807 | (*newseg % sbi->segs_per_sec)) | |
351df4b2 JK |
808 | goto got_it; |
809 | } | |
810 | find_other_zone: | |
7cd8558b JK |
811 | secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); |
812 | if (secno >= MAIN_SECS(sbi)) { | |
351df4b2 JK |
813 | if (dir == ALLOC_RIGHT) { |
814 | secno = find_next_zero_bit(free_i->free_secmap, | |
7cd8558b JK |
815 | MAIN_SECS(sbi), 0); |
816 | f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); | |
351df4b2 JK |
817 | } else { |
818 | go_left = 1; | |
819 | left_start = hint - 1; | |
820 | } | |
821 | } | |
822 | if (go_left == 0) | |
823 | goto skip_left; | |
824 | ||
825 | while (test_bit(left_start, free_i->free_secmap)) { | |
826 | if (left_start > 0) { | |
827 | left_start--; | |
828 | continue; | |
829 | } | |
830 | left_start = find_next_zero_bit(free_i->free_secmap, | |
7cd8558b JK |
831 | MAIN_SECS(sbi), 0); |
832 | f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); | |
351df4b2 JK |
833 | break; |
834 | } | |
835 | secno = left_start; | |
836 | skip_left: | |
837 | hint = secno; | |
838 | segno = secno * sbi->segs_per_sec; | |
839 | zoneno = secno / sbi->secs_per_zone; | |
840 | ||
841 | /* give up on finding another zone */ | |
842 | if (!init) | |
843 | goto got_it; | |
844 | if (sbi->secs_per_zone == 1) | |
845 | goto got_it; | |
846 | if (zoneno == old_zoneno) | |
847 | goto got_it; | |
848 | if (dir == ALLOC_LEFT) { | |
849 | if (!go_left && zoneno + 1 >= total_zones) | |
850 | goto got_it; | |
851 | if (go_left && zoneno == 0) | |
852 | goto got_it; | |
853 | } | |
854 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
855 | if (CURSEG_I(sbi, i)->zone == zoneno) | |
856 | break; | |
857 | ||
858 | if (i < NR_CURSEG_TYPE) { | |
859 | /* zone is in user, try another */ | |
860 | if (go_left) | |
861 | hint = zoneno * sbi->secs_per_zone - 1; | |
862 | else if (zoneno + 1 >= total_zones) | |
863 | hint = 0; | |
864 | else | |
865 | hint = (zoneno + 1) * sbi->secs_per_zone; | |
866 | init = false; | |
867 | goto find_other_zone; | |
868 | } | |
869 | got_it: | |
870 | /* set it as dirty segment in free segmap */ | |
9850cf4a | 871 | f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); |
351df4b2 JK |
872 | __set_inuse(sbi, segno); |
873 | *newseg = segno; | |
874 | write_unlock(&free_i->segmap_lock); | |
875 | } | |
876 | ||
877 | static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) | |
878 | { | |
879 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
880 | struct summary_footer *sum_footer; | |
881 | ||
882 | curseg->segno = curseg->next_segno; | |
883 | curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); | |
884 | curseg->next_blkoff = 0; | |
885 | curseg->next_segno = NULL_SEGNO; | |
886 | ||
887 | sum_footer = &(curseg->sum_blk->footer); | |
888 | memset(sum_footer, 0, sizeof(struct summary_footer)); | |
889 | if (IS_DATASEG(type)) | |
890 | SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); | |
891 | if (IS_NODESEG(type)) | |
892 | SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); | |
893 | __set_sit_entry_type(sbi, type, curseg->segno, modified); | |
894 | } | |
895 | ||
0a8165d7 | 896 | /* |
351df4b2 JK |
897 | * Allocate a current working segment. |
898 | * This function always allocates a free segment in LFS manner. | |
899 | */ | |
900 | static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) | |
901 | { | |
902 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
903 | unsigned int segno = curseg->segno; | |
904 | int dir = ALLOC_LEFT; | |
905 | ||
906 | write_sum_page(sbi, curseg->sum_blk, | |
81fb5e87 | 907 | GET_SUM_BLOCK(sbi, segno)); |
351df4b2 JK |
908 | if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) |
909 | dir = ALLOC_RIGHT; | |
910 | ||
911 | if (test_opt(sbi, NOHEAP)) | |
912 | dir = ALLOC_RIGHT; | |
913 | ||
914 | get_new_segment(sbi, &segno, new_sec, dir); | |
915 | curseg->next_segno = segno; | |
916 | reset_curseg(sbi, type, 1); | |
917 | curseg->alloc_type = LFS; | |
918 | } | |
919 | ||
920 | static void __next_free_blkoff(struct f2fs_sb_info *sbi, | |
921 | struct curseg_info *seg, block_t start) | |
922 | { | |
923 | struct seg_entry *se = get_seg_entry(sbi, seg->segno); | |
e81c93cf CL |
924 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); |
925 | unsigned long target_map[entries]; | |
926 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; | |
927 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; | |
928 | int i, pos; | |
929 | ||
930 | for (i = 0; i < entries; i++) | |
931 | target_map[i] = ckpt_map[i] | cur_map[i]; | |
932 | ||
933 | pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); | |
934 | ||
935 | seg->next_blkoff = pos; | |
351df4b2 JK |
936 | } |
937 | ||
0a8165d7 | 938 | /* |
351df4b2 JK |
939 | * If a segment is written by LFS manner, next block offset is just obtained |
940 | * by increasing the current block offset. However, if a segment is written by | |
941 | * SSR manner, next block offset obtained by calling __next_free_blkoff | |
942 | */ | |
943 | static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, | |
944 | struct curseg_info *seg) | |
945 | { | |
946 | if (seg->alloc_type == SSR) | |
947 | __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); | |
948 | else | |
949 | seg->next_blkoff++; | |
950 | } | |
951 | ||
0a8165d7 | 952 | /* |
e1c42045 | 953 | * This function always allocates a used segment(from dirty seglist) by SSR |
351df4b2 JK |
954 | * manner, so it should recover the existing segment information of valid blocks |
955 | */ | |
956 | static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) | |
957 | { | |
958 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
959 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
960 | unsigned int new_segno = curseg->next_segno; | |
961 | struct f2fs_summary_block *sum_node; | |
962 | struct page *sum_page; | |
963 | ||
964 | write_sum_page(sbi, curseg->sum_blk, | |
965 | GET_SUM_BLOCK(sbi, curseg->segno)); | |
966 | __set_test_and_inuse(sbi, new_segno); | |
967 | ||
968 | mutex_lock(&dirty_i->seglist_lock); | |
969 | __remove_dirty_segment(sbi, new_segno, PRE); | |
970 | __remove_dirty_segment(sbi, new_segno, DIRTY); | |
971 | mutex_unlock(&dirty_i->seglist_lock); | |
972 | ||
973 | reset_curseg(sbi, type, 1); | |
974 | curseg->alloc_type = SSR; | |
975 | __next_free_blkoff(sbi, curseg, 0); | |
976 | ||
977 | if (reuse) { | |
978 | sum_page = get_sum_page(sbi, new_segno); | |
979 | sum_node = (struct f2fs_summary_block *)page_address(sum_page); | |
980 | memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); | |
981 | f2fs_put_page(sum_page, 1); | |
982 | } | |
983 | } | |
984 | ||
43727527 JK |
985 | static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) |
986 | { | |
987 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
988 | const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; | |
989 | ||
990 | if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0)) | |
991 | return v_ops->get_victim(sbi, | |
992 | &(curseg)->next_segno, BG_GC, type, SSR); | |
993 | ||
994 | /* For data segments, let's do SSR more intensively */ | |
995 | for (; type >= CURSEG_HOT_DATA; type--) | |
996 | if (v_ops->get_victim(sbi, &(curseg)->next_segno, | |
997 | BG_GC, type, SSR)) | |
998 | return 1; | |
999 | return 0; | |
1000 | } | |
1001 | ||
351df4b2 JK |
1002 | /* |
1003 | * flush out current segment and replace it with new segment | |
1004 | * This function should be returned with success, otherwise BUG | |
1005 | */ | |
1006 | static void allocate_segment_by_default(struct f2fs_sb_info *sbi, | |
1007 | int type, bool force) | |
1008 | { | |
1009 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
351df4b2 | 1010 | |
7b405275 | 1011 | if (force) |
351df4b2 | 1012 | new_curseg(sbi, type, true); |
7b405275 | 1013 | else if (type == CURSEG_WARM_NODE) |
351df4b2 | 1014 | new_curseg(sbi, type, false); |
60374688 JK |
1015 | else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) |
1016 | new_curseg(sbi, type, false); | |
351df4b2 JK |
1017 | else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) |
1018 | change_curseg(sbi, type, true); | |
1019 | else | |
1020 | new_curseg(sbi, type, false); | |
dcdfff65 JK |
1021 | |
1022 | stat_inc_seg_type(sbi, curseg); | |
351df4b2 JK |
1023 | } |
1024 | ||
1025 | void allocate_new_segments(struct f2fs_sb_info *sbi) | |
1026 | { | |
1027 | struct curseg_info *curseg; | |
1028 | unsigned int old_curseg; | |
1029 | int i; | |
1030 | ||
1031 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1032 | curseg = CURSEG_I(sbi, i); | |
1033 | old_curseg = curseg->segno; | |
1034 | SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); | |
1035 | locate_dirty_segment(sbi, old_curseg); | |
1036 | } | |
1037 | } | |
1038 | ||
1039 | static const struct segment_allocation default_salloc_ops = { | |
1040 | .allocate_segment = allocate_segment_by_default, | |
1041 | }; | |
1042 | ||
4b2fecc8 JK |
1043 | int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) |
1044 | { | |
4b2fecc8 JK |
1045 | __u64 start = range->start >> sbi->log_blocksize; |
1046 | __u64 end = start + (range->len >> sbi->log_blocksize) - 1; | |
4b2fecc8 JK |
1047 | unsigned int start_segno, end_segno; |
1048 | struct cp_control cpc; | |
1049 | ||
7cd8558b JK |
1050 | if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) || |
1051 | range->len < sbi->blocksize) | |
4b2fecc8 JK |
1052 | return -EINVAL; |
1053 | ||
9bd27ae4 | 1054 | cpc.trimmed = 0; |
7cd8558b | 1055 | if (end <= MAIN_BLKADDR(sbi)) |
4b2fecc8 JK |
1056 | goto out; |
1057 | ||
1058 | /* start/end segment number in main_area */ | |
7cd8558b JK |
1059 | start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); |
1060 | end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : | |
1061 | GET_SEGNO(sbi, end); | |
4b2fecc8 JK |
1062 | cpc.reason = CP_DISCARD; |
1063 | cpc.trim_start = start_segno; | |
1064 | cpc.trim_end = end_segno; | |
1065 | cpc.trim_minlen = range->minlen >> sbi->log_blocksize; | |
4b2fecc8 JK |
1066 | |
1067 | /* do checkpoint to issue discard commands safely */ | |
ca4b02ee | 1068 | mutex_lock(&sbi->gc_mutex); |
4b2fecc8 | 1069 | write_checkpoint(sbi, &cpc); |
ca4b02ee | 1070 | mutex_unlock(&sbi->gc_mutex); |
4b2fecc8 JK |
1071 | out: |
1072 | range->len = cpc.trimmed << sbi->log_blocksize; | |
1073 | return 0; | |
1074 | } | |
1075 | ||
351df4b2 JK |
1076 | static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) |
1077 | { | |
1078 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1079 | if (curseg->next_blkoff < sbi->blocks_per_seg) | |
1080 | return true; | |
1081 | return false; | |
1082 | } | |
1083 | ||
1084 | static int __get_segment_type_2(struct page *page, enum page_type p_type) | |
1085 | { | |
1086 | if (p_type == DATA) | |
1087 | return CURSEG_HOT_DATA; | |
1088 | else | |
1089 | return CURSEG_HOT_NODE; | |
1090 | } | |
1091 | ||
1092 | static int __get_segment_type_4(struct page *page, enum page_type p_type) | |
1093 | { | |
1094 | if (p_type == DATA) { | |
1095 | struct inode *inode = page->mapping->host; | |
1096 | ||
1097 | if (S_ISDIR(inode->i_mode)) | |
1098 | return CURSEG_HOT_DATA; | |
1099 | else | |
1100 | return CURSEG_COLD_DATA; | |
1101 | } else { | |
a344b9fd JK |
1102 | if (IS_DNODE(page) && is_cold_node(page)) |
1103 | return CURSEG_WARM_NODE; | |
351df4b2 JK |
1104 | else |
1105 | return CURSEG_COLD_NODE; | |
1106 | } | |
1107 | } | |
1108 | ||
1109 | static int __get_segment_type_6(struct page *page, enum page_type p_type) | |
1110 | { | |
1111 | if (p_type == DATA) { | |
1112 | struct inode *inode = page->mapping->host; | |
1113 | ||
1114 | if (S_ISDIR(inode->i_mode)) | |
1115 | return CURSEG_HOT_DATA; | |
354a3399 | 1116 | else if (is_cold_data(page) || file_is_cold(inode)) |
351df4b2 JK |
1117 | return CURSEG_COLD_DATA; |
1118 | else | |
1119 | return CURSEG_WARM_DATA; | |
1120 | } else { | |
1121 | if (IS_DNODE(page)) | |
1122 | return is_cold_node(page) ? CURSEG_WARM_NODE : | |
1123 | CURSEG_HOT_NODE; | |
1124 | else | |
1125 | return CURSEG_COLD_NODE; | |
1126 | } | |
1127 | } | |
1128 | ||
1129 | static int __get_segment_type(struct page *page, enum page_type p_type) | |
1130 | { | |
4081363f | 1131 | switch (F2FS_P_SB(page)->active_logs) { |
351df4b2 JK |
1132 | case 2: |
1133 | return __get_segment_type_2(page, p_type); | |
1134 | case 4: | |
1135 | return __get_segment_type_4(page, p_type); | |
351df4b2 | 1136 | } |
12a67146 | 1137 | /* NR_CURSEG_TYPE(6) logs by default */ |
9850cf4a JK |
1138 | f2fs_bug_on(F2FS_P_SB(page), |
1139 | F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE); | |
12a67146 | 1140 | return __get_segment_type_6(page, p_type); |
351df4b2 JK |
1141 | } |
1142 | ||
bfad7c2d JK |
1143 | void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, |
1144 | block_t old_blkaddr, block_t *new_blkaddr, | |
1145 | struct f2fs_summary *sum, int type) | |
351df4b2 JK |
1146 | { |
1147 | struct sit_info *sit_i = SIT_I(sbi); | |
1148 | struct curseg_info *curseg; | |
351df4b2 | 1149 | |
351df4b2 JK |
1150 | curseg = CURSEG_I(sbi, type); |
1151 | ||
1152 | mutex_lock(&curseg->curseg_mutex); | |
1153 | ||
1154 | *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); | |
351df4b2 JK |
1155 | |
1156 | /* | |
1157 | * __add_sum_entry should be resided under the curseg_mutex | |
1158 | * because, this function updates a summary entry in the | |
1159 | * current summary block. | |
1160 | */ | |
e79efe3b | 1161 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
1162 | |
1163 | mutex_lock(&sit_i->sentry_lock); | |
1164 | __refresh_next_blkoff(sbi, curseg); | |
dcdfff65 JK |
1165 | |
1166 | stat_inc_block_count(sbi, curseg); | |
351df4b2 | 1167 | |
5e443818 JK |
1168 | if (!__has_curseg_space(sbi, type)) |
1169 | sit_i->s_ops->allocate_segment(sbi, type, false); | |
351df4b2 JK |
1170 | /* |
1171 | * SIT information should be updated before segment allocation, | |
1172 | * since SSR needs latest valid block information. | |
1173 | */ | |
1174 | refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); | |
5e443818 | 1175 | |
351df4b2 JK |
1176 | mutex_unlock(&sit_i->sentry_lock); |
1177 | ||
bfad7c2d | 1178 | if (page && IS_NODESEG(type)) |
351df4b2 JK |
1179 | fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); |
1180 | ||
bfad7c2d JK |
1181 | mutex_unlock(&curseg->curseg_mutex); |
1182 | } | |
1183 | ||
1184 | static void do_write_page(struct f2fs_sb_info *sbi, struct page *page, | |
1185 | block_t old_blkaddr, block_t *new_blkaddr, | |
1186 | struct f2fs_summary *sum, struct f2fs_io_info *fio) | |
1187 | { | |
1188 | int type = __get_segment_type(page, fio->type); | |
1189 | ||
1190 | allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type); | |
1191 | ||
351df4b2 | 1192 | /* writeout dirty page into bdev */ |
458e6197 | 1193 | f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio); |
351df4b2 JK |
1194 | } |
1195 | ||
577e3495 | 1196 | void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) |
351df4b2 | 1197 | { |
458e6197 JK |
1198 | struct f2fs_io_info fio = { |
1199 | .type = META, | |
7e8f2308 | 1200 | .rw = WRITE_SYNC | REQ_META | REQ_PRIO |
458e6197 JK |
1201 | }; |
1202 | ||
351df4b2 | 1203 | set_page_writeback(page); |
458e6197 | 1204 | f2fs_submit_page_mbio(sbi, page, page->index, &fio); |
351df4b2 JK |
1205 | } |
1206 | ||
1207 | void write_node_page(struct f2fs_sb_info *sbi, struct page *page, | |
fb5566da | 1208 | struct f2fs_io_info *fio, |
351df4b2 JK |
1209 | unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr) |
1210 | { | |
1211 | struct f2fs_summary sum; | |
1212 | set_summary(&sum, nid, 0, 0); | |
fb5566da | 1213 | do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio); |
351df4b2 JK |
1214 | } |
1215 | ||
458e6197 JK |
1216 | void write_data_page(struct page *page, struct dnode_of_data *dn, |
1217 | block_t *new_blkaddr, struct f2fs_io_info *fio) | |
351df4b2 | 1218 | { |
4081363f | 1219 | struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); |
351df4b2 JK |
1220 | struct f2fs_summary sum; |
1221 | struct node_info ni; | |
1222 | ||
9850cf4a | 1223 | f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); |
351df4b2 JK |
1224 | get_node_info(sbi, dn->nid, &ni); |
1225 | set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); | |
1226 | ||
458e6197 | 1227 | do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio); |
351df4b2 JK |
1228 | } |
1229 | ||
6c311ec6 CF |
1230 | void rewrite_data_page(struct page *page, block_t old_blkaddr, |
1231 | struct f2fs_io_info *fio) | |
351df4b2 | 1232 | { |
4081363f | 1233 | f2fs_submit_page_mbio(F2FS_P_SB(page), page, old_blkaddr, fio); |
351df4b2 JK |
1234 | } |
1235 | ||
1236 | void recover_data_page(struct f2fs_sb_info *sbi, | |
1237 | struct page *page, struct f2fs_summary *sum, | |
1238 | block_t old_blkaddr, block_t new_blkaddr) | |
1239 | { | |
1240 | struct sit_info *sit_i = SIT_I(sbi); | |
1241 | struct curseg_info *curseg; | |
1242 | unsigned int segno, old_cursegno; | |
1243 | struct seg_entry *se; | |
1244 | int type; | |
1245 | ||
1246 | segno = GET_SEGNO(sbi, new_blkaddr); | |
1247 | se = get_seg_entry(sbi, segno); | |
1248 | type = se->type; | |
1249 | ||
1250 | if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { | |
1251 | if (old_blkaddr == NULL_ADDR) | |
1252 | type = CURSEG_COLD_DATA; | |
1253 | else | |
1254 | type = CURSEG_WARM_DATA; | |
1255 | } | |
1256 | curseg = CURSEG_I(sbi, type); | |
1257 | ||
1258 | mutex_lock(&curseg->curseg_mutex); | |
1259 | mutex_lock(&sit_i->sentry_lock); | |
1260 | ||
1261 | old_cursegno = curseg->segno; | |
1262 | ||
1263 | /* change the current segment */ | |
1264 | if (segno != curseg->segno) { | |
1265 | curseg->next_segno = segno; | |
1266 | change_curseg(sbi, type, true); | |
1267 | } | |
1268 | ||
491c0854 | 1269 | curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); |
e79efe3b | 1270 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
1271 | |
1272 | refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); | |
351df4b2 | 1273 | locate_dirty_segment(sbi, old_cursegno); |
351df4b2 JK |
1274 | |
1275 | mutex_unlock(&sit_i->sentry_lock); | |
1276 | mutex_unlock(&curseg->curseg_mutex); | |
1277 | } | |
1278 | ||
df0f8dc0 CY |
1279 | static inline bool is_merged_page(struct f2fs_sb_info *sbi, |
1280 | struct page *page, enum page_type type) | |
1281 | { | |
1282 | enum page_type btype = PAGE_TYPE_OF_BIO(type); | |
1283 | struct f2fs_bio_info *io = &sbi->write_io[btype]; | |
df0f8dc0 CY |
1284 | struct bio_vec *bvec; |
1285 | int i; | |
1286 | ||
1287 | down_read(&io->io_rwsem); | |
ce23447f | 1288 | if (!io->bio) |
df0f8dc0 CY |
1289 | goto out; |
1290 | ||
ce23447f | 1291 | bio_for_each_segment_all(bvec, io->bio, i) { |
df0f8dc0 CY |
1292 | if (page == bvec->bv_page) { |
1293 | up_read(&io->io_rwsem); | |
1294 | return true; | |
1295 | } | |
1296 | } | |
1297 | ||
1298 | out: | |
1299 | up_read(&io->io_rwsem); | |
1300 | return false; | |
1301 | } | |
1302 | ||
93dfe2ac | 1303 | void f2fs_wait_on_page_writeback(struct page *page, |
5514f0aa | 1304 | enum page_type type) |
93dfe2ac | 1305 | { |
93dfe2ac | 1306 | if (PageWriteback(page)) { |
4081363f JK |
1307 | struct f2fs_sb_info *sbi = F2FS_P_SB(page); |
1308 | ||
df0f8dc0 CY |
1309 | if (is_merged_page(sbi, page, type)) |
1310 | f2fs_submit_merged_bio(sbi, type, WRITE); | |
93dfe2ac JK |
1311 | wait_on_page_writeback(page); |
1312 | } | |
1313 | } | |
1314 | ||
351df4b2 JK |
1315 | static int read_compacted_summaries(struct f2fs_sb_info *sbi) |
1316 | { | |
1317 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1318 | struct curseg_info *seg_i; | |
1319 | unsigned char *kaddr; | |
1320 | struct page *page; | |
1321 | block_t start; | |
1322 | int i, j, offset; | |
1323 | ||
1324 | start = start_sum_block(sbi); | |
1325 | ||
1326 | page = get_meta_page(sbi, start++); | |
1327 | kaddr = (unsigned char *)page_address(page); | |
1328 | ||
1329 | /* Step 1: restore nat cache */ | |
1330 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
1331 | memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE); | |
1332 | ||
1333 | /* Step 2: restore sit cache */ | |
1334 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1335 | memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE, | |
1336 | SUM_JOURNAL_SIZE); | |
1337 | offset = 2 * SUM_JOURNAL_SIZE; | |
1338 | ||
1339 | /* Step 3: restore summary entries */ | |
1340 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1341 | unsigned short blk_off; | |
1342 | unsigned int segno; | |
1343 | ||
1344 | seg_i = CURSEG_I(sbi, i); | |
1345 | segno = le32_to_cpu(ckpt->cur_data_segno[i]); | |
1346 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); | |
1347 | seg_i->next_segno = segno; | |
1348 | reset_curseg(sbi, i, 0); | |
1349 | seg_i->alloc_type = ckpt->alloc_type[i]; | |
1350 | seg_i->next_blkoff = blk_off; | |
1351 | ||
1352 | if (seg_i->alloc_type == SSR) | |
1353 | blk_off = sbi->blocks_per_seg; | |
1354 | ||
1355 | for (j = 0; j < blk_off; j++) { | |
1356 | struct f2fs_summary *s; | |
1357 | s = (struct f2fs_summary *)(kaddr + offset); | |
1358 | seg_i->sum_blk->entries[j] = *s; | |
1359 | offset += SUMMARY_SIZE; | |
1360 | if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE - | |
1361 | SUM_FOOTER_SIZE) | |
1362 | continue; | |
1363 | ||
1364 | f2fs_put_page(page, 1); | |
1365 | page = NULL; | |
1366 | ||
1367 | page = get_meta_page(sbi, start++); | |
1368 | kaddr = (unsigned char *)page_address(page); | |
1369 | offset = 0; | |
1370 | } | |
1371 | } | |
1372 | f2fs_put_page(page, 1); | |
1373 | return 0; | |
1374 | } | |
1375 | ||
1376 | static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) | |
1377 | { | |
1378 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1379 | struct f2fs_summary_block *sum; | |
1380 | struct curseg_info *curseg; | |
1381 | struct page *new; | |
1382 | unsigned short blk_off; | |
1383 | unsigned int segno = 0; | |
1384 | block_t blk_addr = 0; | |
1385 | ||
1386 | /* get segment number and block addr */ | |
1387 | if (IS_DATASEG(type)) { | |
1388 | segno = le32_to_cpu(ckpt->cur_data_segno[type]); | |
1389 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - | |
1390 | CURSEG_HOT_DATA]); | |
25ca923b | 1391 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) |
351df4b2 JK |
1392 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); |
1393 | else | |
1394 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); | |
1395 | } else { | |
1396 | segno = le32_to_cpu(ckpt->cur_node_segno[type - | |
1397 | CURSEG_HOT_NODE]); | |
1398 | blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - | |
1399 | CURSEG_HOT_NODE]); | |
25ca923b | 1400 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) |
351df4b2 JK |
1401 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, |
1402 | type - CURSEG_HOT_NODE); | |
1403 | else | |
1404 | blk_addr = GET_SUM_BLOCK(sbi, segno); | |
1405 | } | |
1406 | ||
1407 | new = get_meta_page(sbi, blk_addr); | |
1408 | sum = (struct f2fs_summary_block *)page_address(new); | |
1409 | ||
1410 | if (IS_NODESEG(type)) { | |
25ca923b | 1411 | if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) { |
351df4b2 JK |
1412 | struct f2fs_summary *ns = &sum->entries[0]; |
1413 | int i; | |
1414 | for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { | |
1415 | ns->version = 0; | |
1416 | ns->ofs_in_node = 0; | |
1417 | } | |
1418 | } else { | |
d653788a GZ |
1419 | int err; |
1420 | ||
1421 | err = restore_node_summary(sbi, segno, sum); | |
1422 | if (err) { | |
351df4b2 | 1423 | f2fs_put_page(new, 1); |
d653788a | 1424 | return err; |
351df4b2 JK |
1425 | } |
1426 | } | |
1427 | } | |
1428 | ||
1429 | /* set uncompleted segment to curseg */ | |
1430 | curseg = CURSEG_I(sbi, type); | |
1431 | mutex_lock(&curseg->curseg_mutex); | |
1432 | memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE); | |
1433 | curseg->next_segno = segno; | |
1434 | reset_curseg(sbi, type, 0); | |
1435 | curseg->alloc_type = ckpt->alloc_type[type]; | |
1436 | curseg->next_blkoff = blk_off; | |
1437 | mutex_unlock(&curseg->curseg_mutex); | |
1438 | f2fs_put_page(new, 1); | |
1439 | return 0; | |
1440 | } | |
1441 | ||
1442 | static int restore_curseg_summaries(struct f2fs_sb_info *sbi) | |
1443 | { | |
1444 | int type = CURSEG_HOT_DATA; | |
e4fc5fbf | 1445 | int err; |
351df4b2 | 1446 | |
25ca923b | 1447 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) { |
351df4b2 JK |
1448 | /* restore for compacted data summary */ |
1449 | if (read_compacted_summaries(sbi)) | |
1450 | return -EINVAL; | |
1451 | type = CURSEG_HOT_NODE; | |
1452 | } | |
1453 | ||
e4fc5fbf CY |
1454 | for (; type <= CURSEG_COLD_NODE; type++) { |
1455 | err = read_normal_summaries(sbi, type); | |
1456 | if (err) | |
1457 | return err; | |
1458 | } | |
1459 | ||
351df4b2 JK |
1460 | return 0; |
1461 | } | |
1462 | ||
1463 | static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) | |
1464 | { | |
1465 | struct page *page; | |
1466 | unsigned char *kaddr; | |
1467 | struct f2fs_summary *summary; | |
1468 | struct curseg_info *seg_i; | |
1469 | int written_size = 0; | |
1470 | int i, j; | |
1471 | ||
1472 | page = grab_meta_page(sbi, blkaddr++); | |
1473 | kaddr = (unsigned char *)page_address(page); | |
1474 | ||
1475 | /* Step 1: write nat cache */ | |
1476 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
1477 | memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE); | |
1478 | written_size += SUM_JOURNAL_SIZE; | |
1479 | ||
1480 | /* Step 2: write sit cache */ | |
1481 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1482 | memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits, | |
1483 | SUM_JOURNAL_SIZE); | |
1484 | written_size += SUM_JOURNAL_SIZE; | |
1485 | ||
351df4b2 JK |
1486 | /* Step 3: write summary entries */ |
1487 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1488 | unsigned short blkoff; | |
1489 | seg_i = CURSEG_I(sbi, i); | |
1490 | if (sbi->ckpt->alloc_type[i] == SSR) | |
1491 | blkoff = sbi->blocks_per_seg; | |
1492 | else | |
1493 | blkoff = curseg_blkoff(sbi, i); | |
1494 | ||
1495 | for (j = 0; j < blkoff; j++) { | |
1496 | if (!page) { | |
1497 | page = grab_meta_page(sbi, blkaddr++); | |
1498 | kaddr = (unsigned char *)page_address(page); | |
1499 | written_size = 0; | |
1500 | } | |
1501 | summary = (struct f2fs_summary *)(kaddr + written_size); | |
1502 | *summary = seg_i->sum_blk->entries[j]; | |
1503 | written_size += SUMMARY_SIZE; | |
351df4b2 JK |
1504 | |
1505 | if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE - | |
1506 | SUM_FOOTER_SIZE) | |
1507 | continue; | |
1508 | ||
e8d61a74 | 1509 | set_page_dirty(page); |
351df4b2 JK |
1510 | f2fs_put_page(page, 1); |
1511 | page = NULL; | |
1512 | } | |
1513 | } | |
e8d61a74 CY |
1514 | if (page) { |
1515 | set_page_dirty(page); | |
351df4b2 | 1516 | f2fs_put_page(page, 1); |
e8d61a74 | 1517 | } |
351df4b2 JK |
1518 | } |
1519 | ||
1520 | static void write_normal_summaries(struct f2fs_sb_info *sbi, | |
1521 | block_t blkaddr, int type) | |
1522 | { | |
1523 | int i, end; | |
1524 | if (IS_DATASEG(type)) | |
1525 | end = type + NR_CURSEG_DATA_TYPE; | |
1526 | else | |
1527 | end = type + NR_CURSEG_NODE_TYPE; | |
1528 | ||
1529 | for (i = type; i < end; i++) { | |
1530 | struct curseg_info *sum = CURSEG_I(sbi, i); | |
1531 | mutex_lock(&sum->curseg_mutex); | |
1532 | write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type)); | |
1533 | mutex_unlock(&sum->curseg_mutex); | |
1534 | } | |
1535 | } | |
1536 | ||
1537 | void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1538 | { | |
25ca923b | 1539 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) |
351df4b2 JK |
1540 | write_compacted_summaries(sbi, start_blk); |
1541 | else | |
1542 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); | |
1543 | } | |
1544 | ||
1545 | void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1546 | { | |
25ca923b | 1547 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG)) |
351df4b2 | 1548 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); |
351df4b2 JK |
1549 | } |
1550 | ||
1551 | int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type, | |
1552 | unsigned int val, int alloc) | |
1553 | { | |
1554 | int i; | |
1555 | ||
1556 | if (type == NAT_JOURNAL) { | |
1557 | for (i = 0; i < nats_in_cursum(sum); i++) { | |
1558 | if (le32_to_cpu(nid_in_journal(sum, i)) == val) | |
1559 | return i; | |
1560 | } | |
1561 | if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) | |
1562 | return update_nats_in_cursum(sum, 1); | |
1563 | } else if (type == SIT_JOURNAL) { | |
1564 | for (i = 0; i < sits_in_cursum(sum); i++) | |
1565 | if (le32_to_cpu(segno_in_journal(sum, i)) == val) | |
1566 | return i; | |
1567 | if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES) | |
1568 | return update_sits_in_cursum(sum, 1); | |
1569 | } | |
1570 | return -1; | |
1571 | } | |
1572 | ||
1573 | static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, | |
1574 | unsigned int segno) | |
1575 | { | |
2cc22186 | 1576 | return get_meta_page(sbi, current_sit_addr(sbi, segno)); |
351df4b2 JK |
1577 | } |
1578 | ||
1579 | static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, | |
1580 | unsigned int start) | |
1581 | { | |
1582 | struct sit_info *sit_i = SIT_I(sbi); | |
1583 | struct page *src_page, *dst_page; | |
1584 | pgoff_t src_off, dst_off; | |
1585 | void *src_addr, *dst_addr; | |
1586 | ||
1587 | src_off = current_sit_addr(sbi, start); | |
1588 | dst_off = next_sit_addr(sbi, src_off); | |
1589 | ||
1590 | /* get current sit block page without lock */ | |
1591 | src_page = get_meta_page(sbi, src_off); | |
1592 | dst_page = grab_meta_page(sbi, dst_off); | |
9850cf4a | 1593 | f2fs_bug_on(sbi, PageDirty(src_page)); |
351df4b2 JK |
1594 | |
1595 | src_addr = page_address(src_page); | |
1596 | dst_addr = page_address(dst_page); | |
1597 | memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); | |
1598 | ||
1599 | set_page_dirty(dst_page); | |
1600 | f2fs_put_page(src_page, 1); | |
1601 | ||
1602 | set_to_next_sit(sit_i, start); | |
1603 | ||
1604 | return dst_page; | |
1605 | } | |
1606 | ||
184a5cd2 CY |
1607 | static struct sit_entry_set *grab_sit_entry_set(void) |
1608 | { | |
1609 | struct sit_entry_set *ses = | |
1610 | f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC); | |
1611 | ||
1612 | ses->entry_cnt = 0; | |
1613 | INIT_LIST_HEAD(&ses->set_list); | |
1614 | return ses; | |
1615 | } | |
1616 | ||
1617 | static void release_sit_entry_set(struct sit_entry_set *ses) | |
1618 | { | |
1619 | list_del(&ses->set_list); | |
1620 | kmem_cache_free(sit_entry_set_slab, ses); | |
1621 | } | |
1622 | ||
1623 | static void adjust_sit_entry_set(struct sit_entry_set *ses, | |
1624 | struct list_head *head) | |
1625 | { | |
1626 | struct sit_entry_set *next = ses; | |
1627 | ||
1628 | if (list_is_last(&ses->set_list, head)) | |
1629 | return; | |
1630 | ||
1631 | list_for_each_entry_continue(next, head, set_list) | |
1632 | if (ses->entry_cnt <= next->entry_cnt) | |
1633 | break; | |
1634 | ||
1635 | list_move_tail(&ses->set_list, &next->set_list); | |
1636 | } | |
1637 | ||
1638 | static void add_sit_entry(unsigned int segno, struct list_head *head) | |
1639 | { | |
1640 | struct sit_entry_set *ses; | |
1641 | unsigned int start_segno = START_SEGNO(segno); | |
1642 | ||
1643 | list_for_each_entry(ses, head, set_list) { | |
1644 | if (ses->start_segno == start_segno) { | |
1645 | ses->entry_cnt++; | |
1646 | adjust_sit_entry_set(ses, head); | |
1647 | return; | |
1648 | } | |
1649 | } | |
1650 | ||
1651 | ses = grab_sit_entry_set(); | |
1652 | ||
1653 | ses->start_segno = start_segno; | |
1654 | ses->entry_cnt++; | |
1655 | list_add(&ses->set_list, head); | |
1656 | } | |
1657 | ||
1658 | static void add_sits_in_set(struct f2fs_sb_info *sbi) | |
1659 | { | |
1660 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
1661 | struct list_head *set_list = &sm_info->sit_entry_set; | |
1662 | unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; | |
184a5cd2 CY |
1663 | unsigned int segno; |
1664 | ||
7cd8558b | 1665 | for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) |
184a5cd2 CY |
1666 | add_sit_entry(segno, set_list); |
1667 | } | |
1668 | ||
1669 | static void remove_sits_in_journal(struct f2fs_sb_info *sbi) | |
351df4b2 JK |
1670 | { |
1671 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1672 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
1673 | int i; | |
1674 | ||
184a5cd2 CY |
1675 | for (i = sits_in_cursum(sum) - 1; i >= 0; i--) { |
1676 | unsigned int segno; | |
1677 | bool dirtied; | |
1678 | ||
1679 | segno = le32_to_cpu(segno_in_journal(sum, i)); | |
1680 | dirtied = __mark_sit_entry_dirty(sbi, segno); | |
1681 | ||
1682 | if (!dirtied) | |
1683 | add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); | |
351df4b2 | 1684 | } |
184a5cd2 | 1685 | update_sits_in_cursum(sum, -sits_in_cursum(sum)); |
351df4b2 JK |
1686 | } |
1687 | ||
0a8165d7 | 1688 | /* |
351df4b2 JK |
1689 | * CP calls this function, which flushes SIT entries including sit_journal, |
1690 | * and moves prefree segs to free segs. | |
1691 | */ | |
4b2fecc8 | 1692 | void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) |
351df4b2 JK |
1693 | { |
1694 | struct sit_info *sit_i = SIT_I(sbi); | |
1695 | unsigned long *bitmap = sit_i->dirty_sentries_bitmap; | |
1696 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1697 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
184a5cd2 CY |
1698 | struct sit_entry_set *ses, *tmp; |
1699 | struct list_head *head = &SM_I(sbi)->sit_entry_set; | |
184a5cd2 | 1700 | bool to_journal = true; |
4b2fecc8 | 1701 | struct seg_entry *se; |
351df4b2 JK |
1702 | |
1703 | mutex_lock(&curseg->curseg_mutex); | |
1704 | mutex_lock(&sit_i->sentry_lock); | |
1705 | ||
1706 | /* | |
184a5cd2 CY |
1707 | * add and account sit entries of dirty bitmap in sit entry |
1708 | * set temporarily | |
351df4b2 | 1709 | */ |
184a5cd2 | 1710 | add_sits_in_set(sbi); |
351df4b2 | 1711 | |
184a5cd2 CY |
1712 | /* |
1713 | * if there are no enough space in journal to store dirty sit | |
1714 | * entries, remove all entries from journal and add and account | |
1715 | * them in sit entry set. | |
1716 | */ | |
1717 | if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL)) | |
1718 | remove_sits_in_journal(sbi); | |
b2955550 | 1719 | |
184a5cd2 CY |
1720 | if (!sit_i->dirty_sentries) |
1721 | goto out; | |
351df4b2 | 1722 | |
184a5cd2 CY |
1723 | /* |
1724 | * there are two steps to flush sit entries: | |
1725 | * #1, flush sit entries to journal in current cold data summary block. | |
1726 | * #2, flush sit entries to sit page. | |
1727 | */ | |
1728 | list_for_each_entry_safe(ses, tmp, head, set_list) { | |
4a257ed6 | 1729 | struct page *page = NULL; |
184a5cd2 CY |
1730 | struct f2fs_sit_block *raw_sit = NULL; |
1731 | unsigned int start_segno = ses->start_segno; | |
1732 | unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, | |
7cd8558b | 1733 | (unsigned long)MAIN_SEGS(sbi)); |
184a5cd2 CY |
1734 | unsigned int segno = start_segno; |
1735 | ||
1736 | if (to_journal && | |
1737 | !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL)) | |
1738 | to_journal = false; | |
1739 | ||
1740 | if (!to_journal) { | |
1741 | page = get_next_sit_page(sbi, start_segno); | |
1742 | raw_sit = page_address(page); | |
351df4b2 | 1743 | } |
351df4b2 | 1744 | |
184a5cd2 CY |
1745 | /* flush dirty sit entries in region of current sit set */ |
1746 | for_each_set_bit_from(segno, bitmap, end) { | |
1747 | int offset, sit_offset; | |
4b2fecc8 JK |
1748 | |
1749 | se = get_seg_entry(sbi, segno); | |
184a5cd2 CY |
1750 | |
1751 | /* add discard candidates */ | |
4b2fecc8 JK |
1752 | if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards) { |
1753 | cpc->trim_start = segno; | |
1754 | add_discard_addrs(sbi, cpc); | |
1755 | } | |
184a5cd2 CY |
1756 | |
1757 | if (to_journal) { | |
1758 | offset = lookup_journal_in_cursum(sum, | |
1759 | SIT_JOURNAL, segno, 1); | |
1760 | f2fs_bug_on(sbi, offset < 0); | |
1761 | segno_in_journal(sum, offset) = | |
1762 | cpu_to_le32(segno); | |
1763 | seg_info_to_raw_sit(se, | |
1764 | &sit_in_journal(sum, offset)); | |
1765 | } else { | |
1766 | sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); | |
1767 | seg_info_to_raw_sit(se, | |
1768 | &raw_sit->entries[sit_offset]); | |
1769 | } | |
351df4b2 | 1770 | |
184a5cd2 CY |
1771 | __clear_bit(segno, bitmap); |
1772 | sit_i->dirty_sentries--; | |
1773 | ses->entry_cnt--; | |
351df4b2 JK |
1774 | } |
1775 | ||
184a5cd2 CY |
1776 | if (!to_journal) |
1777 | f2fs_put_page(page, 1); | |
1778 | ||
1779 | f2fs_bug_on(sbi, ses->entry_cnt); | |
1780 | release_sit_entry_set(ses); | |
351df4b2 | 1781 | } |
184a5cd2 CY |
1782 | |
1783 | f2fs_bug_on(sbi, !list_empty(head)); | |
1784 | f2fs_bug_on(sbi, sit_i->dirty_sentries); | |
184a5cd2 | 1785 | out: |
4b2fecc8 JK |
1786 | if (cpc->reason == CP_DISCARD) { |
1787 | for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) | |
1788 | add_discard_addrs(sbi, cpc); | |
1789 | } | |
351df4b2 JK |
1790 | mutex_unlock(&sit_i->sentry_lock); |
1791 | mutex_unlock(&curseg->curseg_mutex); | |
1792 | ||
351df4b2 JK |
1793 | set_prefree_as_free_segments(sbi); |
1794 | } | |
1795 | ||
1796 | static int build_sit_info(struct f2fs_sb_info *sbi) | |
1797 | { | |
1798 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
1799 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1800 | struct sit_info *sit_i; | |
1801 | unsigned int sit_segs, start; | |
1802 | char *src_bitmap, *dst_bitmap; | |
1803 | unsigned int bitmap_size; | |
1804 | ||
1805 | /* allocate memory for SIT information */ | |
1806 | sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); | |
1807 | if (!sit_i) | |
1808 | return -ENOMEM; | |
1809 | ||
1810 | SM_I(sbi)->sit_info = sit_i; | |
1811 | ||
7cd8558b | 1812 | sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry)); |
351df4b2 JK |
1813 | if (!sit_i->sentries) |
1814 | return -ENOMEM; | |
1815 | ||
7cd8558b | 1816 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
351df4b2 JK |
1817 | sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL); |
1818 | if (!sit_i->dirty_sentries_bitmap) | |
1819 | return -ENOMEM; | |
1820 | ||
7cd8558b | 1821 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
1822 | sit_i->sentries[start].cur_valid_map |
1823 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
1824 | sit_i->sentries[start].ckpt_valid_map | |
1825 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
1826 | if (!sit_i->sentries[start].cur_valid_map | |
1827 | || !sit_i->sentries[start].ckpt_valid_map) | |
1828 | return -ENOMEM; | |
1829 | } | |
1830 | ||
1831 | if (sbi->segs_per_sec > 1) { | |
7cd8558b | 1832 | sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) * |
351df4b2 JK |
1833 | sizeof(struct sec_entry)); |
1834 | if (!sit_i->sec_entries) | |
1835 | return -ENOMEM; | |
1836 | } | |
1837 | ||
1838 | /* get information related with SIT */ | |
1839 | sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; | |
1840 | ||
1841 | /* setup SIT bitmap from ckeckpoint pack */ | |
1842 | bitmap_size = __bitmap_size(sbi, SIT_BITMAP); | |
1843 | src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); | |
1844 | ||
79b5793b | 1845 | dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); |
351df4b2 JK |
1846 | if (!dst_bitmap) |
1847 | return -ENOMEM; | |
351df4b2 JK |
1848 | |
1849 | /* init SIT information */ | |
1850 | sit_i->s_ops = &default_salloc_ops; | |
1851 | ||
1852 | sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); | |
1853 | sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; | |
1854 | sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count); | |
1855 | sit_i->sit_bitmap = dst_bitmap; | |
1856 | sit_i->bitmap_size = bitmap_size; | |
1857 | sit_i->dirty_sentries = 0; | |
1858 | sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; | |
1859 | sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); | |
1860 | sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; | |
1861 | mutex_init(&sit_i->sentry_lock); | |
1862 | return 0; | |
1863 | } | |
1864 | ||
1865 | static int build_free_segmap(struct f2fs_sb_info *sbi) | |
1866 | { | |
351df4b2 JK |
1867 | struct free_segmap_info *free_i; |
1868 | unsigned int bitmap_size, sec_bitmap_size; | |
1869 | ||
1870 | /* allocate memory for free segmap information */ | |
1871 | free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); | |
1872 | if (!free_i) | |
1873 | return -ENOMEM; | |
1874 | ||
1875 | SM_I(sbi)->free_info = free_i; | |
1876 | ||
7cd8558b | 1877 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
351df4b2 JK |
1878 | free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL); |
1879 | if (!free_i->free_segmap) | |
1880 | return -ENOMEM; | |
1881 | ||
7cd8558b | 1882 | sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); |
351df4b2 JK |
1883 | free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL); |
1884 | if (!free_i->free_secmap) | |
1885 | return -ENOMEM; | |
1886 | ||
1887 | /* set all segments as dirty temporarily */ | |
1888 | memset(free_i->free_segmap, 0xff, bitmap_size); | |
1889 | memset(free_i->free_secmap, 0xff, sec_bitmap_size); | |
1890 | ||
1891 | /* init free segmap information */ | |
7cd8558b | 1892 | free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); |
351df4b2 JK |
1893 | free_i->free_segments = 0; |
1894 | free_i->free_sections = 0; | |
1895 | rwlock_init(&free_i->segmap_lock); | |
1896 | return 0; | |
1897 | } | |
1898 | ||
1899 | static int build_curseg(struct f2fs_sb_info *sbi) | |
1900 | { | |
1042d60f | 1901 | struct curseg_info *array; |
351df4b2 JK |
1902 | int i; |
1903 | ||
b434babf | 1904 | array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL); |
351df4b2 JK |
1905 | if (!array) |
1906 | return -ENOMEM; | |
1907 | ||
1908 | SM_I(sbi)->curseg_array = array; | |
1909 | ||
1910 | for (i = 0; i < NR_CURSEG_TYPE; i++) { | |
1911 | mutex_init(&array[i].curseg_mutex); | |
1912 | array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL); | |
1913 | if (!array[i].sum_blk) | |
1914 | return -ENOMEM; | |
1915 | array[i].segno = NULL_SEGNO; | |
1916 | array[i].next_blkoff = 0; | |
1917 | } | |
1918 | return restore_curseg_summaries(sbi); | |
1919 | } | |
1920 | ||
1921 | static void build_sit_entries(struct f2fs_sb_info *sbi) | |
1922 | { | |
1923 | struct sit_info *sit_i = SIT_I(sbi); | |
1924 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1925 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
74de593a CY |
1926 | int sit_blk_cnt = SIT_BLK_CNT(sbi); |
1927 | unsigned int i, start, end; | |
1928 | unsigned int readed, start_blk = 0; | |
90a893c7 | 1929 | int nrpages = MAX_BIO_BLOCKS(sbi); |
351df4b2 | 1930 | |
74de593a | 1931 | do { |
662befda | 1932 | readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT); |
74de593a CY |
1933 | |
1934 | start = start_blk * sit_i->sents_per_block; | |
1935 | end = (start_blk + readed) * sit_i->sents_per_block; | |
1936 | ||
7cd8558b | 1937 | for (; start < end && start < MAIN_SEGS(sbi); start++) { |
74de593a CY |
1938 | struct seg_entry *se = &sit_i->sentries[start]; |
1939 | struct f2fs_sit_block *sit_blk; | |
1940 | struct f2fs_sit_entry sit; | |
1941 | struct page *page; | |
1942 | ||
1943 | mutex_lock(&curseg->curseg_mutex); | |
1944 | for (i = 0; i < sits_in_cursum(sum); i++) { | |
6c311ec6 CF |
1945 | if (le32_to_cpu(segno_in_journal(sum, i)) |
1946 | == start) { | |
74de593a CY |
1947 | sit = sit_in_journal(sum, i); |
1948 | mutex_unlock(&curseg->curseg_mutex); | |
1949 | goto got_it; | |
1950 | } | |
351df4b2 | 1951 | } |
74de593a CY |
1952 | mutex_unlock(&curseg->curseg_mutex); |
1953 | ||
1954 | page = get_current_sit_page(sbi, start); | |
1955 | sit_blk = (struct f2fs_sit_block *)page_address(page); | |
1956 | sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; | |
1957 | f2fs_put_page(page, 1); | |
351df4b2 | 1958 | got_it: |
74de593a CY |
1959 | check_block_count(sbi, start, &sit); |
1960 | seg_info_from_raw_sit(se, &sit); | |
1961 | if (sbi->segs_per_sec > 1) { | |
1962 | struct sec_entry *e = get_sec_entry(sbi, start); | |
1963 | e->valid_blocks += se->valid_blocks; | |
1964 | } | |
351df4b2 | 1965 | } |
74de593a CY |
1966 | start_blk += readed; |
1967 | } while (start_blk < sit_blk_cnt); | |
351df4b2 JK |
1968 | } |
1969 | ||
1970 | static void init_free_segmap(struct f2fs_sb_info *sbi) | |
1971 | { | |
1972 | unsigned int start; | |
1973 | int type; | |
1974 | ||
7cd8558b | 1975 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
1976 | struct seg_entry *sentry = get_seg_entry(sbi, start); |
1977 | if (!sentry->valid_blocks) | |
1978 | __set_free(sbi, start); | |
1979 | } | |
1980 | ||
1981 | /* set use the current segments */ | |
1982 | for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { | |
1983 | struct curseg_info *curseg_t = CURSEG_I(sbi, type); | |
1984 | __set_test_and_inuse(sbi, curseg_t->segno); | |
1985 | } | |
1986 | } | |
1987 | ||
1988 | static void init_dirty_segmap(struct f2fs_sb_info *sbi) | |
1989 | { | |
1990 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1991 | struct free_segmap_info *free_i = FREE_I(sbi); | |
7cd8558b | 1992 | unsigned int segno = 0, offset = 0; |
351df4b2 JK |
1993 | unsigned short valid_blocks; |
1994 | ||
8736fbf0 | 1995 | while (1) { |
351df4b2 | 1996 | /* find dirty segment based on free segmap */ |
7cd8558b JK |
1997 | segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); |
1998 | if (segno >= MAIN_SEGS(sbi)) | |
351df4b2 JK |
1999 | break; |
2000 | offset = segno + 1; | |
2001 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
ec325b52 | 2002 | if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) |
351df4b2 | 2003 | continue; |
ec325b52 JK |
2004 | if (valid_blocks > sbi->blocks_per_seg) { |
2005 | f2fs_bug_on(sbi, 1); | |
2006 | continue; | |
2007 | } | |
351df4b2 JK |
2008 | mutex_lock(&dirty_i->seglist_lock); |
2009 | __locate_dirty_segment(sbi, segno, DIRTY); | |
2010 | mutex_unlock(&dirty_i->seglist_lock); | |
2011 | } | |
2012 | } | |
2013 | ||
5ec4e49f | 2014 | static int init_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
2015 | { |
2016 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
7cd8558b | 2017 | unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); |
351df4b2 | 2018 | |
5ec4e49f JK |
2019 | dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL); |
2020 | if (!dirty_i->victim_secmap) | |
351df4b2 JK |
2021 | return -ENOMEM; |
2022 | return 0; | |
2023 | } | |
2024 | ||
2025 | static int build_dirty_segmap(struct f2fs_sb_info *sbi) | |
2026 | { | |
2027 | struct dirty_seglist_info *dirty_i; | |
2028 | unsigned int bitmap_size, i; | |
2029 | ||
2030 | /* allocate memory for dirty segments list information */ | |
2031 | dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); | |
2032 | if (!dirty_i) | |
2033 | return -ENOMEM; | |
2034 | ||
2035 | SM_I(sbi)->dirty_info = dirty_i; | |
2036 | mutex_init(&dirty_i->seglist_lock); | |
2037 | ||
7cd8558b | 2038 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
351df4b2 JK |
2039 | |
2040 | for (i = 0; i < NR_DIRTY_TYPE; i++) { | |
2041 | dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL); | |
351df4b2 JK |
2042 | if (!dirty_i->dirty_segmap[i]) |
2043 | return -ENOMEM; | |
2044 | } | |
2045 | ||
2046 | init_dirty_segmap(sbi); | |
5ec4e49f | 2047 | return init_victim_secmap(sbi); |
351df4b2 JK |
2048 | } |
2049 | ||
0a8165d7 | 2050 | /* |
351df4b2 JK |
2051 | * Update min, max modified time for cost-benefit GC algorithm |
2052 | */ | |
2053 | static void init_min_max_mtime(struct f2fs_sb_info *sbi) | |
2054 | { | |
2055 | struct sit_info *sit_i = SIT_I(sbi); | |
2056 | unsigned int segno; | |
2057 | ||
2058 | mutex_lock(&sit_i->sentry_lock); | |
2059 | ||
2060 | sit_i->min_mtime = LLONG_MAX; | |
2061 | ||
7cd8558b | 2062 | for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { |
351df4b2 JK |
2063 | unsigned int i; |
2064 | unsigned long long mtime = 0; | |
2065 | ||
2066 | for (i = 0; i < sbi->segs_per_sec; i++) | |
2067 | mtime += get_seg_entry(sbi, segno + i)->mtime; | |
2068 | ||
2069 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
2070 | ||
2071 | if (sit_i->min_mtime > mtime) | |
2072 | sit_i->min_mtime = mtime; | |
2073 | } | |
2074 | sit_i->max_mtime = get_mtime(sbi); | |
2075 | mutex_unlock(&sit_i->sentry_lock); | |
2076 | } | |
2077 | ||
2078 | int build_segment_manager(struct f2fs_sb_info *sbi) | |
2079 | { | |
2080 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
2081 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1042d60f | 2082 | struct f2fs_sm_info *sm_info; |
351df4b2 JK |
2083 | int err; |
2084 | ||
2085 | sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); | |
2086 | if (!sm_info) | |
2087 | return -ENOMEM; | |
2088 | ||
2089 | /* init sm info */ | |
2090 | sbi->sm_info = sm_info; | |
351df4b2 JK |
2091 | sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); |
2092 | sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); | |
2093 | sm_info->segment_count = le32_to_cpu(raw_super->segment_count); | |
2094 | sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); | |
2095 | sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); | |
2096 | sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); | |
2097 | sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); | |
58c41035 JK |
2098 | sm_info->rec_prefree_segments = sm_info->main_segments * |
2099 | DEF_RECLAIM_PREFREE_SEGMENTS / 100; | |
9b5f136f | 2100 | sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; |
216fbd64 | 2101 | sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; |
c1ce1b02 | 2102 | sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; |
351df4b2 | 2103 | |
7fd9e544 JK |
2104 | INIT_LIST_HEAD(&sm_info->discard_list); |
2105 | sm_info->nr_discards = 0; | |
2106 | sm_info->max_discards = 0; | |
2107 | ||
184a5cd2 CY |
2108 | INIT_LIST_HEAD(&sm_info->sit_entry_set); |
2109 | ||
b270ad6f | 2110 | if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) { |
2163d198 GZ |
2111 | err = create_flush_cmd_control(sbi); |
2112 | if (err) | |
a688b9d9 | 2113 | return err; |
6b4afdd7 JK |
2114 | } |
2115 | ||
351df4b2 JK |
2116 | err = build_sit_info(sbi); |
2117 | if (err) | |
2118 | return err; | |
2119 | err = build_free_segmap(sbi); | |
2120 | if (err) | |
2121 | return err; | |
2122 | err = build_curseg(sbi); | |
2123 | if (err) | |
2124 | return err; | |
2125 | ||
2126 | /* reinit free segmap based on SIT */ | |
2127 | build_sit_entries(sbi); | |
2128 | ||
2129 | init_free_segmap(sbi); | |
2130 | err = build_dirty_segmap(sbi); | |
2131 | if (err) | |
2132 | return err; | |
2133 | ||
2134 | init_min_max_mtime(sbi); | |
2135 | return 0; | |
2136 | } | |
2137 | ||
2138 | static void discard_dirty_segmap(struct f2fs_sb_info *sbi, | |
2139 | enum dirty_type dirty_type) | |
2140 | { | |
2141 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
2142 | ||
2143 | mutex_lock(&dirty_i->seglist_lock); | |
2144 | kfree(dirty_i->dirty_segmap[dirty_type]); | |
2145 | dirty_i->nr_dirty[dirty_type] = 0; | |
2146 | mutex_unlock(&dirty_i->seglist_lock); | |
2147 | } | |
2148 | ||
5ec4e49f | 2149 | static void destroy_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
2150 | { |
2151 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 2152 | kfree(dirty_i->victim_secmap); |
351df4b2 JK |
2153 | } |
2154 | ||
2155 | static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) | |
2156 | { | |
2157 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
2158 | int i; | |
2159 | ||
2160 | if (!dirty_i) | |
2161 | return; | |
2162 | ||
2163 | /* discard pre-free/dirty segments list */ | |
2164 | for (i = 0; i < NR_DIRTY_TYPE; i++) | |
2165 | discard_dirty_segmap(sbi, i); | |
2166 | ||
5ec4e49f | 2167 | destroy_victim_secmap(sbi); |
351df4b2 JK |
2168 | SM_I(sbi)->dirty_info = NULL; |
2169 | kfree(dirty_i); | |
2170 | } | |
2171 | ||
2172 | static void destroy_curseg(struct f2fs_sb_info *sbi) | |
2173 | { | |
2174 | struct curseg_info *array = SM_I(sbi)->curseg_array; | |
2175 | int i; | |
2176 | ||
2177 | if (!array) | |
2178 | return; | |
2179 | SM_I(sbi)->curseg_array = NULL; | |
2180 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
2181 | kfree(array[i].sum_blk); | |
2182 | kfree(array); | |
2183 | } | |
2184 | ||
2185 | static void destroy_free_segmap(struct f2fs_sb_info *sbi) | |
2186 | { | |
2187 | struct free_segmap_info *free_i = SM_I(sbi)->free_info; | |
2188 | if (!free_i) | |
2189 | return; | |
2190 | SM_I(sbi)->free_info = NULL; | |
2191 | kfree(free_i->free_segmap); | |
2192 | kfree(free_i->free_secmap); | |
2193 | kfree(free_i); | |
2194 | } | |
2195 | ||
2196 | static void destroy_sit_info(struct f2fs_sb_info *sbi) | |
2197 | { | |
2198 | struct sit_info *sit_i = SIT_I(sbi); | |
2199 | unsigned int start; | |
2200 | ||
2201 | if (!sit_i) | |
2202 | return; | |
2203 | ||
2204 | if (sit_i->sentries) { | |
7cd8558b | 2205 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
2206 | kfree(sit_i->sentries[start].cur_valid_map); |
2207 | kfree(sit_i->sentries[start].ckpt_valid_map); | |
2208 | } | |
2209 | } | |
2210 | vfree(sit_i->sentries); | |
2211 | vfree(sit_i->sec_entries); | |
2212 | kfree(sit_i->dirty_sentries_bitmap); | |
2213 | ||
2214 | SM_I(sbi)->sit_info = NULL; | |
2215 | kfree(sit_i->sit_bitmap); | |
2216 | kfree(sit_i); | |
2217 | } | |
2218 | ||
2219 | void destroy_segment_manager(struct f2fs_sb_info *sbi) | |
2220 | { | |
2221 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
a688b9d9 | 2222 | |
3b03f724 CY |
2223 | if (!sm_info) |
2224 | return; | |
2163d198 | 2225 | destroy_flush_cmd_control(sbi); |
351df4b2 JK |
2226 | destroy_dirty_segmap(sbi); |
2227 | destroy_curseg(sbi); | |
2228 | destroy_free_segmap(sbi); | |
2229 | destroy_sit_info(sbi); | |
2230 | sbi->sm_info = NULL; | |
2231 | kfree(sm_info); | |
2232 | } | |
7fd9e544 JK |
2233 | |
2234 | int __init create_segment_manager_caches(void) | |
2235 | { | |
2236 | discard_entry_slab = f2fs_kmem_cache_create("discard_entry", | |
e8512d2e | 2237 | sizeof(struct discard_entry)); |
7fd9e544 | 2238 | if (!discard_entry_slab) |
184a5cd2 CY |
2239 | goto fail; |
2240 | ||
2241 | sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", | |
c9ee0085 | 2242 | sizeof(struct sit_entry_set)); |
184a5cd2 CY |
2243 | if (!sit_entry_set_slab) |
2244 | goto destory_discard_entry; | |
88b88a66 JK |
2245 | |
2246 | inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", | |
2247 | sizeof(struct inmem_pages)); | |
2248 | if (!inmem_entry_slab) | |
2249 | goto destroy_sit_entry_set; | |
7fd9e544 | 2250 | return 0; |
184a5cd2 | 2251 | |
88b88a66 JK |
2252 | destroy_sit_entry_set: |
2253 | kmem_cache_destroy(sit_entry_set_slab); | |
184a5cd2 CY |
2254 | destory_discard_entry: |
2255 | kmem_cache_destroy(discard_entry_slab); | |
2256 | fail: | |
2257 | return -ENOMEM; | |
7fd9e544 JK |
2258 | } |
2259 | ||
2260 | void destroy_segment_manager_caches(void) | |
2261 | { | |
184a5cd2 | 2262 | kmem_cache_destroy(sit_entry_set_slab); |
7fd9e544 | 2263 | kmem_cache_destroy(discard_entry_slab); |
88b88a66 | 2264 | kmem_cache_destroy(inmem_entry_slab); |
7fd9e544 | 2265 | } |