]>
Commit | Line | Data |
---|---|---|
1 | /* | |
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> | |
15 | #include <linux/prefetch.h> | |
16 | #include <linux/kthread.h> | |
17 | #include <linux/swap.h> | |
18 | #include <linux/timer.h> | |
19 | ||
20 | #include "f2fs.h" | |
21 | #include "segment.h" | |
22 | #include "node.h" | |
23 | #include "trace.h" | |
24 | #include <trace/events/f2fs.h> | |
25 | ||
26 | #define __reverse_ffz(x) __reverse_ffs(~(x)) | |
27 | ||
28 | static struct kmem_cache *discard_entry_slab; | |
29 | static struct kmem_cache *discard_cmd_slab; | |
30 | static struct kmem_cache *sit_entry_set_slab; | |
31 | static struct kmem_cache *inmem_entry_slab; | |
32 | ||
33 | static unsigned long __reverse_ulong(unsigned char *str) | |
34 | { | |
35 | unsigned long tmp = 0; | |
36 | int shift = 24, idx = 0; | |
37 | ||
38 | #if BITS_PER_LONG == 64 | |
39 | shift = 56; | |
40 | #endif | |
41 | while (shift >= 0) { | |
42 | tmp |= (unsigned long)str[idx++] << shift; | |
43 | shift -= BITS_PER_BYTE; | |
44 | } | |
45 | return tmp; | |
46 | } | |
47 | ||
48 | /* | |
49 | * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since | |
50 | * MSB and LSB are reversed in a byte by f2fs_set_bit. | |
51 | */ | |
52 | static inline unsigned long __reverse_ffs(unsigned long word) | |
53 | { | |
54 | int num = 0; | |
55 | ||
56 | #if BITS_PER_LONG == 64 | |
57 | if ((word & 0xffffffff00000000UL) == 0) | |
58 | num += 32; | |
59 | else | |
60 | word >>= 32; | |
61 | #endif | |
62 | if ((word & 0xffff0000) == 0) | |
63 | num += 16; | |
64 | else | |
65 | word >>= 16; | |
66 | ||
67 | if ((word & 0xff00) == 0) | |
68 | num += 8; | |
69 | else | |
70 | word >>= 8; | |
71 | ||
72 | if ((word & 0xf0) == 0) | |
73 | num += 4; | |
74 | else | |
75 | word >>= 4; | |
76 | ||
77 | if ((word & 0xc) == 0) | |
78 | num += 2; | |
79 | else | |
80 | word >>= 2; | |
81 | ||
82 | if ((word & 0x2) == 0) | |
83 | num += 1; | |
84 | return num; | |
85 | } | |
86 | ||
87 | /* | |
88 | * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because | |
89 | * f2fs_set_bit makes MSB and LSB reversed in a byte. | |
90 | * @size must be integral times of unsigned long. | |
91 | * Example: | |
92 | * MSB <--> LSB | |
93 | * f2fs_set_bit(0, bitmap) => 1000 0000 | |
94 | * f2fs_set_bit(7, bitmap) => 0000 0001 | |
95 | */ | |
96 | static unsigned long __find_rev_next_bit(const unsigned long *addr, | |
97 | unsigned long size, unsigned long offset) | |
98 | { | |
99 | const unsigned long *p = addr + BIT_WORD(offset); | |
100 | unsigned long result = size; | |
101 | unsigned long tmp; | |
102 | ||
103 | if (offset >= size) | |
104 | return size; | |
105 | ||
106 | size -= (offset & ~(BITS_PER_LONG - 1)); | |
107 | offset %= BITS_PER_LONG; | |
108 | ||
109 | while (1) { | |
110 | if (*p == 0) | |
111 | goto pass; | |
112 | ||
113 | tmp = __reverse_ulong((unsigned char *)p); | |
114 | ||
115 | tmp &= ~0UL >> offset; | |
116 | if (size < BITS_PER_LONG) | |
117 | tmp &= (~0UL << (BITS_PER_LONG - size)); | |
118 | if (tmp) | |
119 | goto found; | |
120 | pass: | |
121 | if (size <= BITS_PER_LONG) | |
122 | break; | |
123 | size -= BITS_PER_LONG; | |
124 | offset = 0; | |
125 | p++; | |
126 | } | |
127 | return result; | |
128 | found: | |
129 | return result - size + __reverse_ffs(tmp); | |
130 | } | |
131 | ||
132 | static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, | |
133 | unsigned long size, unsigned long offset) | |
134 | { | |
135 | const unsigned long *p = addr + BIT_WORD(offset); | |
136 | unsigned long result = size; | |
137 | unsigned long tmp; | |
138 | ||
139 | if (offset >= size) | |
140 | return size; | |
141 | ||
142 | size -= (offset & ~(BITS_PER_LONG - 1)); | |
143 | offset %= BITS_PER_LONG; | |
144 | ||
145 | while (1) { | |
146 | if (*p == ~0UL) | |
147 | goto pass; | |
148 | ||
149 | tmp = __reverse_ulong((unsigned char *)p); | |
150 | ||
151 | if (offset) | |
152 | tmp |= ~0UL << (BITS_PER_LONG - offset); | |
153 | if (size < BITS_PER_LONG) | |
154 | tmp |= ~0UL >> size; | |
155 | if (tmp != ~0UL) | |
156 | goto found; | |
157 | pass: | |
158 | if (size <= BITS_PER_LONG) | |
159 | break; | |
160 | size -= BITS_PER_LONG; | |
161 | offset = 0; | |
162 | p++; | |
163 | } | |
164 | return result; | |
165 | found: | |
166 | return result - size + __reverse_ffz(tmp); | |
167 | } | |
168 | ||
169 | void register_inmem_page(struct inode *inode, struct page *page) | |
170 | { | |
171 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
172 | struct inmem_pages *new; | |
173 | ||
174 | f2fs_trace_pid(page); | |
175 | ||
176 | set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE); | |
177 | SetPagePrivate(page); | |
178 | ||
179 | new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); | |
180 | ||
181 | /* add atomic page indices to the list */ | |
182 | new->page = page; | |
183 | INIT_LIST_HEAD(&new->list); | |
184 | ||
185 | /* increase reference count with clean state */ | |
186 | mutex_lock(&fi->inmem_lock); | |
187 | get_page(page); | |
188 | list_add_tail(&new->list, &fi->inmem_pages); | |
189 | inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); | |
190 | mutex_unlock(&fi->inmem_lock); | |
191 | ||
192 | trace_f2fs_register_inmem_page(page, INMEM); | |
193 | } | |
194 | ||
195 | static int __revoke_inmem_pages(struct inode *inode, | |
196 | struct list_head *head, bool drop, bool recover) | |
197 | { | |
198 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
199 | struct inmem_pages *cur, *tmp; | |
200 | int err = 0; | |
201 | ||
202 | list_for_each_entry_safe(cur, tmp, head, list) { | |
203 | struct page *page = cur->page; | |
204 | ||
205 | if (drop) | |
206 | trace_f2fs_commit_inmem_page(page, INMEM_DROP); | |
207 | ||
208 | lock_page(page); | |
209 | ||
210 | if (recover) { | |
211 | struct dnode_of_data dn; | |
212 | struct node_info ni; | |
213 | ||
214 | trace_f2fs_commit_inmem_page(page, INMEM_REVOKE); | |
215 | ||
216 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
217 | if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) { | |
218 | err = -EAGAIN; | |
219 | goto next; | |
220 | } | |
221 | get_node_info(sbi, dn.nid, &ni); | |
222 | f2fs_replace_block(sbi, &dn, dn.data_blkaddr, | |
223 | cur->old_addr, ni.version, true, true); | |
224 | f2fs_put_dnode(&dn); | |
225 | } | |
226 | next: | |
227 | /* we don't need to invalidate this in the sccessful status */ | |
228 | if (drop || recover) | |
229 | ClearPageUptodate(page); | |
230 | set_page_private(page, 0); | |
231 | ClearPagePrivate(page); | |
232 | f2fs_put_page(page, 1); | |
233 | ||
234 | list_del(&cur->list); | |
235 | kmem_cache_free(inmem_entry_slab, cur); | |
236 | dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); | |
237 | } | |
238 | return err; | |
239 | } | |
240 | ||
241 | void drop_inmem_pages(struct inode *inode) | |
242 | { | |
243 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
244 | ||
245 | mutex_lock(&fi->inmem_lock); | |
246 | __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); | |
247 | mutex_unlock(&fi->inmem_lock); | |
248 | ||
249 | clear_inode_flag(inode, FI_ATOMIC_FILE); | |
250 | stat_dec_atomic_write(inode); | |
251 | } | |
252 | ||
253 | void drop_inmem_page(struct inode *inode, struct page *page) | |
254 | { | |
255 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
256 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
257 | struct list_head *head = &fi->inmem_pages; | |
258 | struct inmem_pages *cur = NULL; | |
259 | ||
260 | f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page)); | |
261 | ||
262 | mutex_lock(&fi->inmem_lock); | |
263 | list_for_each_entry(cur, head, list) { | |
264 | if (cur->page == page) | |
265 | break; | |
266 | } | |
267 | ||
268 | f2fs_bug_on(sbi, !cur || cur->page != page); | |
269 | list_del(&cur->list); | |
270 | mutex_unlock(&fi->inmem_lock); | |
271 | ||
272 | dec_page_count(sbi, F2FS_INMEM_PAGES); | |
273 | kmem_cache_free(inmem_entry_slab, cur); | |
274 | ||
275 | ClearPageUptodate(page); | |
276 | set_page_private(page, 0); | |
277 | ClearPagePrivate(page); | |
278 | f2fs_put_page(page, 0); | |
279 | ||
280 | trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE); | |
281 | } | |
282 | ||
283 | static int __commit_inmem_pages(struct inode *inode, | |
284 | struct list_head *revoke_list) | |
285 | { | |
286 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
287 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
288 | struct inmem_pages *cur, *tmp; | |
289 | struct f2fs_io_info fio = { | |
290 | .sbi = sbi, | |
291 | .type = DATA, | |
292 | .op = REQ_OP_WRITE, | |
293 | .op_flags = REQ_SYNC | REQ_PRIO, | |
294 | .encrypted_page = NULL, | |
295 | }; | |
296 | pgoff_t last_idx = ULONG_MAX; | |
297 | int err = 0; | |
298 | ||
299 | list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { | |
300 | struct page *page = cur->page; | |
301 | ||
302 | lock_page(page); | |
303 | if (page->mapping == inode->i_mapping) { | |
304 | trace_f2fs_commit_inmem_page(page, INMEM); | |
305 | ||
306 | set_page_dirty(page); | |
307 | f2fs_wait_on_page_writeback(page, DATA, true); | |
308 | if (clear_page_dirty_for_io(page)) { | |
309 | inode_dec_dirty_pages(inode); | |
310 | remove_dirty_inode(inode); | |
311 | } | |
312 | ||
313 | fio.page = page; | |
314 | err = do_write_data_page(&fio); | |
315 | if (err) { | |
316 | unlock_page(page); | |
317 | break; | |
318 | } | |
319 | ||
320 | /* record old blkaddr for revoking */ | |
321 | cur->old_addr = fio.old_blkaddr; | |
322 | last_idx = page->index; | |
323 | } | |
324 | unlock_page(page); | |
325 | list_move_tail(&cur->list, revoke_list); | |
326 | } | |
327 | ||
328 | if (last_idx != ULONG_MAX) | |
329 | f2fs_submit_merged_bio_cond(sbi, inode, 0, last_idx, | |
330 | DATA, WRITE); | |
331 | ||
332 | if (!err) | |
333 | __revoke_inmem_pages(inode, revoke_list, false, false); | |
334 | ||
335 | return err; | |
336 | } | |
337 | ||
338 | int commit_inmem_pages(struct inode *inode) | |
339 | { | |
340 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
341 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
342 | struct list_head revoke_list; | |
343 | int err; | |
344 | ||
345 | INIT_LIST_HEAD(&revoke_list); | |
346 | f2fs_balance_fs(sbi, true); | |
347 | f2fs_lock_op(sbi); | |
348 | ||
349 | set_inode_flag(inode, FI_ATOMIC_COMMIT); | |
350 | ||
351 | mutex_lock(&fi->inmem_lock); | |
352 | err = __commit_inmem_pages(inode, &revoke_list); | |
353 | if (err) { | |
354 | int ret; | |
355 | /* | |
356 | * try to revoke all committed pages, but still we could fail | |
357 | * due to no memory or other reason, if that happened, EAGAIN | |
358 | * will be returned, which means in such case, transaction is | |
359 | * already not integrity, caller should use journal to do the | |
360 | * recovery or rewrite & commit last transaction. For other | |
361 | * error number, revoking was done by filesystem itself. | |
362 | */ | |
363 | ret = __revoke_inmem_pages(inode, &revoke_list, false, true); | |
364 | if (ret) | |
365 | err = ret; | |
366 | ||
367 | /* drop all uncommitted pages */ | |
368 | __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); | |
369 | } | |
370 | mutex_unlock(&fi->inmem_lock); | |
371 | ||
372 | clear_inode_flag(inode, FI_ATOMIC_COMMIT); | |
373 | ||
374 | f2fs_unlock_op(sbi); | |
375 | return err; | |
376 | } | |
377 | ||
378 | /* | |
379 | * This function balances dirty node and dentry pages. | |
380 | * In addition, it controls garbage collection. | |
381 | */ | |
382 | void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) | |
383 | { | |
384 | #ifdef CONFIG_F2FS_FAULT_INJECTION | |
385 | if (time_to_inject(sbi, FAULT_CHECKPOINT)) { | |
386 | f2fs_show_injection_info(FAULT_CHECKPOINT); | |
387 | f2fs_stop_checkpoint(sbi, false); | |
388 | } | |
389 | #endif | |
390 | ||
391 | /* balance_fs_bg is able to be pending */ | |
392 | if (need && excess_cached_nats(sbi)) | |
393 | f2fs_balance_fs_bg(sbi); | |
394 | ||
395 | /* | |
396 | * We should do GC or end up with checkpoint, if there are so many dirty | |
397 | * dir/node pages without enough free segments. | |
398 | */ | |
399 | if (has_not_enough_free_secs(sbi, 0, 0)) { | |
400 | mutex_lock(&sbi->gc_mutex); | |
401 | f2fs_gc(sbi, false, false, NULL_SEGNO); | |
402 | } | |
403 | } | |
404 | ||
405 | void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) | |
406 | { | |
407 | /* try to shrink extent cache when there is no enough memory */ | |
408 | if (!available_free_memory(sbi, EXTENT_CACHE)) | |
409 | f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); | |
410 | ||
411 | /* check the # of cached NAT entries */ | |
412 | if (!available_free_memory(sbi, NAT_ENTRIES)) | |
413 | try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); | |
414 | ||
415 | if (!available_free_memory(sbi, FREE_NIDS)) | |
416 | try_to_free_nids(sbi, MAX_FREE_NIDS); | |
417 | else | |
418 | build_free_nids(sbi, false, false); | |
419 | ||
420 | if (!is_idle(sbi)) | |
421 | return; | |
422 | ||
423 | /* checkpoint is the only way to shrink partial cached entries */ | |
424 | if (!available_free_memory(sbi, NAT_ENTRIES) || | |
425 | !available_free_memory(sbi, INO_ENTRIES) || | |
426 | excess_prefree_segs(sbi) || | |
427 | excess_dirty_nats(sbi) || | |
428 | f2fs_time_over(sbi, CP_TIME)) { | |
429 | if (test_opt(sbi, DATA_FLUSH)) { | |
430 | struct blk_plug plug; | |
431 | ||
432 | blk_start_plug(&plug); | |
433 | sync_dirty_inodes(sbi, FILE_INODE); | |
434 | blk_finish_plug(&plug); | |
435 | } | |
436 | f2fs_sync_fs(sbi->sb, true); | |
437 | stat_inc_bg_cp_count(sbi->stat_info); | |
438 | } | |
439 | } | |
440 | ||
441 | static int __submit_flush_wait(struct f2fs_sb_info *sbi, | |
442 | struct block_device *bdev) | |
443 | { | |
444 | struct bio *bio = f2fs_bio_alloc(0); | |
445 | int ret; | |
446 | ||
447 | bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; | |
448 | bio->bi_bdev = bdev; | |
449 | ret = submit_bio_wait(bio); | |
450 | bio_put(bio); | |
451 | ||
452 | trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), | |
453 | test_opt(sbi, FLUSH_MERGE), ret); | |
454 | return ret; | |
455 | } | |
456 | ||
457 | static int submit_flush_wait(struct f2fs_sb_info *sbi) | |
458 | { | |
459 | int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev); | |
460 | int i; | |
461 | ||
462 | if (!sbi->s_ndevs || ret) | |
463 | return ret; | |
464 | ||
465 | for (i = 1; i < sbi->s_ndevs; i++) { | |
466 | ret = __submit_flush_wait(sbi, FDEV(i).bdev); | |
467 | if (ret) | |
468 | break; | |
469 | } | |
470 | return ret; | |
471 | } | |
472 | ||
473 | static int issue_flush_thread(void *data) | |
474 | { | |
475 | struct f2fs_sb_info *sbi = data; | |
476 | struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; | |
477 | wait_queue_head_t *q = &fcc->flush_wait_queue; | |
478 | repeat: | |
479 | if (kthread_should_stop()) | |
480 | return 0; | |
481 | ||
482 | if (!llist_empty(&fcc->issue_list)) { | |
483 | struct flush_cmd *cmd, *next; | |
484 | int ret; | |
485 | ||
486 | fcc->dispatch_list = llist_del_all(&fcc->issue_list); | |
487 | fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); | |
488 | ||
489 | ret = submit_flush_wait(sbi); | |
490 | atomic_inc(&fcc->issued_flush); | |
491 | ||
492 | llist_for_each_entry_safe(cmd, next, | |
493 | fcc->dispatch_list, llnode) { | |
494 | cmd->ret = ret; | |
495 | complete(&cmd->wait); | |
496 | } | |
497 | fcc->dispatch_list = NULL; | |
498 | } | |
499 | ||
500 | wait_event_interruptible(*q, | |
501 | kthread_should_stop() || !llist_empty(&fcc->issue_list)); | |
502 | goto repeat; | |
503 | } | |
504 | ||
505 | int f2fs_issue_flush(struct f2fs_sb_info *sbi) | |
506 | { | |
507 | struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; | |
508 | struct flush_cmd cmd; | |
509 | int ret; | |
510 | ||
511 | if (test_opt(sbi, NOBARRIER)) | |
512 | return 0; | |
513 | ||
514 | if (!test_opt(sbi, FLUSH_MERGE)) { | |
515 | ret = submit_flush_wait(sbi); | |
516 | atomic_inc(&fcc->issued_flush); | |
517 | return ret; | |
518 | } | |
519 | ||
520 | if (!atomic_read(&fcc->issing_flush)) { | |
521 | atomic_inc(&fcc->issing_flush); | |
522 | ret = submit_flush_wait(sbi); | |
523 | atomic_dec(&fcc->issing_flush); | |
524 | ||
525 | atomic_inc(&fcc->issued_flush); | |
526 | return ret; | |
527 | } | |
528 | ||
529 | init_completion(&cmd.wait); | |
530 | ||
531 | atomic_inc(&fcc->issing_flush); | |
532 | llist_add(&cmd.llnode, &fcc->issue_list); | |
533 | ||
534 | if (!fcc->dispatch_list) | |
535 | wake_up(&fcc->flush_wait_queue); | |
536 | ||
537 | if (fcc->f2fs_issue_flush) { | |
538 | wait_for_completion(&cmd.wait); | |
539 | atomic_dec(&fcc->issing_flush); | |
540 | } else { | |
541 | llist_del_all(&fcc->issue_list); | |
542 | atomic_set(&fcc->issing_flush, 0); | |
543 | } | |
544 | ||
545 | return cmd.ret; | |
546 | } | |
547 | ||
548 | int create_flush_cmd_control(struct f2fs_sb_info *sbi) | |
549 | { | |
550 | dev_t dev = sbi->sb->s_bdev->bd_dev; | |
551 | struct flush_cmd_control *fcc; | |
552 | int err = 0; | |
553 | ||
554 | if (SM_I(sbi)->fcc_info) { | |
555 | fcc = SM_I(sbi)->fcc_info; | |
556 | goto init_thread; | |
557 | } | |
558 | ||
559 | fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL); | |
560 | if (!fcc) | |
561 | return -ENOMEM; | |
562 | atomic_set(&fcc->issued_flush, 0); | |
563 | atomic_set(&fcc->issing_flush, 0); | |
564 | init_waitqueue_head(&fcc->flush_wait_queue); | |
565 | init_llist_head(&fcc->issue_list); | |
566 | SM_I(sbi)->fcc_info = fcc; | |
567 | init_thread: | |
568 | fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, | |
569 | "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); | |
570 | if (IS_ERR(fcc->f2fs_issue_flush)) { | |
571 | err = PTR_ERR(fcc->f2fs_issue_flush); | |
572 | kfree(fcc); | |
573 | SM_I(sbi)->fcc_info = NULL; | |
574 | return err; | |
575 | } | |
576 | ||
577 | return err; | |
578 | } | |
579 | ||
580 | void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) | |
581 | { | |
582 | struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; | |
583 | ||
584 | if (fcc && fcc->f2fs_issue_flush) { | |
585 | struct task_struct *flush_thread = fcc->f2fs_issue_flush; | |
586 | ||
587 | fcc->f2fs_issue_flush = NULL; | |
588 | kthread_stop(flush_thread); | |
589 | } | |
590 | if (free) { | |
591 | kfree(fcc); | |
592 | SM_I(sbi)->fcc_info = NULL; | |
593 | } | |
594 | } | |
595 | ||
596 | static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, | |
597 | enum dirty_type dirty_type) | |
598 | { | |
599 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
600 | ||
601 | /* need not be added */ | |
602 | if (IS_CURSEG(sbi, segno)) | |
603 | return; | |
604 | ||
605 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
606 | dirty_i->nr_dirty[dirty_type]++; | |
607 | ||
608 | if (dirty_type == DIRTY) { | |
609 | struct seg_entry *sentry = get_seg_entry(sbi, segno); | |
610 | enum dirty_type t = sentry->type; | |
611 | ||
612 | if (unlikely(t >= DIRTY)) { | |
613 | f2fs_bug_on(sbi, 1); | |
614 | return; | |
615 | } | |
616 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) | |
617 | dirty_i->nr_dirty[t]++; | |
618 | } | |
619 | } | |
620 | ||
621 | static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, | |
622 | enum dirty_type dirty_type) | |
623 | { | |
624 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
625 | ||
626 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
627 | dirty_i->nr_dirty[dirty_type]--; | |
628 | ||
629 | if (dirty_type == DIRTY) { | |
630 | struct seg_entry *sentry = get_seg_entry(sbi, segno); | |
631 | enum dirty_type t = sentry->type; | |
632 | ||
633 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) | |
634 | dirty_i->nr_dirty[t]--; | |
635 | ||
636 | if (get_valid_blocks(sbi, segno, true) == 0) | |
637 | clear_bit(GET_SEC_FROM_SEG(sbi, segno), | |
638 | dirty_i->victim_secmap); | |
639 | } | |
640 | } | |
641 | ||
642 | /* | |
643 | * Should not occur error such as -ENOMEM. | |
644 | * Adding dirty entry into seglist is not critical operation. | |
645 | * If a given segment is one of current working segments, it won't be added. | |
646 | */ | |
647 | static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) | |
648 | { | |
649 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
650 | unsigned short valid_blocks; | |
651 | ||
652 | if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) | |
653 | return; | |
654 | ||
655 | mutex_lock(&dirty_i->seglist_lock); | |
656 | ||
657 | valid_blocks = get_valid_blocks(sbi, segno, false); | |
658 | ||
659 | if (valid_blocks == 0) { | |
660 | __locate_dirty_segment(sbi, segno, PRE); | |
661 | __remove_dirty_segment(sbi, segno, DIRTY); | |
662 | } else if (valid_blocks < sbi->blocks_per_seg) { | |
663 | __locate_dirty_segment(sbi, segno, DIRTY); | |
664 | } else { | |
665 | /* Recovery routine with SSR needs this */ | |
666 | __remove_dirty_segment(sbi, segno, DIRTY); | |
667 | } | |
668 | ||
669 | mutex_unlock(&dirty_i->seglist_lock); | |
670 | } | |
671 | ||
672 | static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, | |
673 | struct block_device *bdev, block_t lstart, | |
674 | block_t start, block_t len) | |
675 | { | |
676 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
677 | struct list_head *pend_list; | |
678 | struct discard_cmd *dc; | |
679 | ||
680 | f2fs_bug_on(sbi, !len); | |
681 | ||
682 | pend_list = &dcc->pend_list[plist_idx(len)]; | |
683 | ||
684 | dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS); | |
685 | INIT_LIST_HEAD(&dc->list); | |
686 | dc->bdev = bdev; | |
687 | dc->lstart = lstart; | |
688 | dc->start = start; | |
689 | dc->len = len; | |
690 | dc->state = D_PREP; | |
691 | dc->error = 0; | |
692 | init_completion(&dc->wait); | |
693 | list_add_tail(&dc->list, pend_list); | |
694 | atomic_inc(&dcc->discard_cmd_cnt); | |
695 | dcc->undiscard_blks += len; | |
696 | ||
697 | return dc; | |
698 | } | |
699 | ||
700 | static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi, | |
701 | struct block_device *bdev, block_t lstart, | |
702 | block_t start, block_t len, | |
703 | struct rb_node *parent, struct rb_node **p) | |
704 | { | |
705 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
706 | struct discard_cmd *dc; | |
707 | ||
708 | dc = __create_discard_cmd(sbi, bdev, lstart, start, len); | |
709 | ||
710 | rb_link_node(&dc->rb_node, parent, p); | |
711 | rb_insert_color(&dc->rb_node, &dcc->root); | |
712 | ||
713 | return dc; | |
714 | } | |
715 | ||
716 | static void __detach_discard_cmd(struct discard_cmd_control *dcc, | |
717 | struct discard_cmd *dc) | |
718 | { | |
719 | if (dc->state == D_DONE) | |
720 | atomic_dec(&dcc->issing_discard); | |
721 | ||
722 | list_del(&dc->list); | |
723 | rb_erase(&dc->rb_node, &dcc->root); | |
724 | dcc->undiscard_blks -= dc->len; | |
725 | ||
726 | kmem_cache_free(discard_cmd_slab, dc); | |
727 | ||
728 | atomic_dec(&dcc->discard_cmd_cnt); | |
729 | } | |
730 | ||
731 | static void __remove_discard_cmd(struct f2fs_sb_info *sbi, | |
732 | struct discard_cmd *dc) | |
733 | { | |
734 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
735 | ||
736 | if (dc->error == -EOPNOTSUPP) | |
737 | dc->error = 0; | |
738 | ||
739 | if (dc->error) | |
740 | f2fs_msg(sbi->sb, KERN_INFO, | |
741 | "Issue discard failed, ret: %d", dc->error); | |
742 | __detach_discard_cmd(dcc, dc); | |
743 | } | |
744 | ||
745 | static void f2fs_submit_discard_endio(struct bio *bio) | |
746 | { | |
747 | struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; | |
748 | ||
749 | dc->error = bio->bi_error; | |
750 | dc->state = D_DONE; | |
751 | complete(&dc->wait); | |
752 | bio_put(bio); | |
753 | } | |
754 | ||
755 | /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ | |
756 | static void __submit_discard_cmd(struct f2fs_sb_info *sbi, | |
757 | struct discard_cmd *dc) | |
758 | { | |
759 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
760 | struct bio *bio = NULL; | |
761 | ||
762 | if (dc->state != D_PREP) | |
763 | return; | |
764 | ||
765 | trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len); | |
766 | ||
767 | dc->error = __blkdev_issue_discard(dc->bdev, | |
768 | SECTOR_FROM_BLOCK(dc->start), | |
769 | SECTOR_FROM_BLOCK(dc->len), | |
770 | GFP_NOFS, 0, &bio); | |
771 | if (!dc->error) { | |
772 | /* should keep before submission to avoid D_DONE right away */ | |
773 | dc->state = D_SUBMIT; | |
774 | atomic_inc(&dcc->issued_discard); | |
775 | atomic_inc(&dcc->issing_discard); | |
776 | if (bio) { | |
777 | bio->bi_private = dc; | |
778 | bio->bi_end_io = f2fs_submit_discard_endio; | |
779 | bio->bi_opf |= REQ_SYNC; | |
780 | submit_bio(bio); | |
781 | list_move_tail(&dc->list, &dcc->wait_list); | |
782 | } | |
783 | } else { | |
784 | __remove_discard_cmd(sbi, dc); | |
785 | } | |
786 | } | |
787 | ||
788 | static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi, | |
789 | struct block_device *bdev, block_t lstart, | |
790 | block_t start, block_t len, | |
791 | struct rb_node **insert_p, | |
792 | struct rb_node *insert_parent) | |
793 | { | |
794 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
795 | struct rb_node **p = &dcc->root.rb_node; | |
796 | struct rb_node *parent = NULL; | |
797 | struct discard_cmd *dc = NULL; | |
798 | ||
799 | if (insert_p && insert_parent) { | |
800 | parent = insert_parent; | |
801 | p = insert_p; | |
802 | goto do_insert; | |
803 | } | |
804 | ||
805 | p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart); | |
806 | do_insert: | |
807 | dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p); | |
808 | if (!dc) | |
809 | return NULL; | |
810 | ||
811 | return dc; | |
812 | } | |
813 | ||
814 | static void __relocate_discard_cmd(struct discard_cmd_control *dcc, | |
815 | struct discard_cmd *dc) | |
816 | { | |
817 | list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]); | |
818 | } | |
819 | ||
820 | static void __punch_discard_cmd(struct f2fs_sb_info *sbi, | |
821 | struct discard_cmd *dc, block_t blkaddr) | |
822 | { | |
823 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
824 | struct discard_info di = dc->di; | |
825 | bool modified = false; | |
826 | ||
827 | if (dc->state == D_DONE || dc->len == 1) { | |
828 | __remove_discard_cmd(sbi, dc); | |
829 | return; | |
830 | } | |
831 | ||
832 | dcc->undiscard_blks -= di.len; | |
833 | ||
834 | if (blkaddr > di.lstart) { | |
835 | dc->len = blkaddr - dc->lstart; | |
836 | dcc->undiscard_blks += dc->len; | |
837 | __relocate_discard_cmd(dcc, dc); | |
838 | f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root)); | |
839 | modified = true; | |
840 | } | |
841 | ||
842 | if (blkaddr < di.lstart + di.len - 1) { | |
843 | if (modified) { | |
844 | __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, | |
845 | di.start + blkaddr + 1 - di.lstart, | |
846 | di.lstart + di.len - 1 - blkaddr, | |
847 | NULL, NULL); | |
848 | f2fs_bug_on(sbi, | |
849 | !__check_rb_tree_consistence(sbi, &dcc->root)); | |
850 | } else { | |
851 | dc->lstart++; | |
852 | dc->len--; | |
853 | dc->start++; | |
854 | dcc->undiscard_blks += dc->len; | |
855 | __relocate_discard_cmd(dcc, dc); | |
856 | f2fs_bug_on(sbi, | |
857 | !__check_rb_tree_consistence(sbi, &dcc->root)); | |
858 | } | |
859 | } | |
860 | } | |
861 | ||
862 | static void __update_discard_tree_range(struct f2fs_sb_info *sbi, | |
863 | struct block_device *bdev, block_t lstart, | |
864 | block_t start, block_t len) | |
865 | { | |
866 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
867 | struct discard_cmd *prev_dc = NULL, *next_dc = NULL; | |
868 | struct discard_cmd *dc; | |
869 | struct discard_info di = {0}; | |
870 | struct rb_node **insert_p = NULL, *insert_parent = NULL; | |
871 | block_t end = lstart + len; | |
872 | ||
873 | mutex_lock(&dcc->cmd_lock); | |
874 | ||
875 | dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root, | |
876 | NULL, lstart, | |
877 | (struct rb_entry **)&prev_dc, | |
878 | (struct rb_entry **)&next_dc, | |
879 | &insert_p, &insert_parent, true); | |
880 | if (dc) | |
881 | prev_dc = dc; | |
882 | ||
883 | if (!prev_dc) { | |
884 | di.lstart = lstart; | |
885 | di.len = next_dc ? next_dc->lstart - lstart : len; | |
886 | di.len = min(di.len, len); | |
887 | di.start = start; | |
888 | } | |
889 | ||
890 | while (1) { | |
891 | struct rb_node *node; | |
892 | bool merged = false; | |
893 | struct discard_cmd *tdc = NULL; | |
894 | ||
895 | if (prev_dc) { | |
896 | di.lstart = prev_dc->lstart + prev_dc->len; | |
897 | if (di.lstart < lstart) | |
898 | di.lstart = lstart; | |
899 | if (di.lstart >= end) | |
900 | break; | |
901 | ||
902 | if (!next_dc || next_dc->lstart > end) | |
903 | di.len = end - di.lstart; | |
904 | else | |
905 | di.len = next_dc->lstart - di.lstart; | |
906 | di.start = start + di.lstart - lstart; | |
907 | } | |
908 | ||
909 | if (!di.len) | |
910 | goto next; | |
911 | ||
912 | if (prev_dc && prev_dc->state == D_PREP && | |
913 | prev_dc->bdev == bdev && | |
914 | __is_discard_back_mergeable(&di, &prev_dc->di)) { | |
915 | prev_dc->di.len += di.len; | |
916 | dcc->undiscard_blks += di.len; | |
917 | __relocate_discard_cmd(dcc, prev_dc); | |
918 | f2fs_bug_on(sbi, | |
919 | !__check_rb_tree_consistence(sbi, &dcc->root)); | |
920 | di = prev_dc->di; | |
921 | tdc = prev_dc; | |
922 | merged = true; | |
923 | } | |
924 | ||
925 | if (next_dc && next_dc->state == D_PREP && | |
926 | next_dc->bdev == bdev && | |
927 | __is_discard_front_mergeable(&di, &next_dc->di)) { | |
928 | next_dc->di.lstart = di.lstart; | |
929 | next_dc->di.len += di.len; | |
930 | next_dc->di.start = di.start; | |
931 | dcc->undiscard_blks += di.len; | |
932 | __relocate_discard_cmd(dcc, next_dc); | |
933 | if (tdc) | |
934 | __remove_discard_cmd(sbi, tdc); | |
935 | f2fs_bug_on(sbi, | |
936 | !__check_rb_tree_consistence(sbi, &dcc->root)); | |
937 | merged = true; | |
938 | } | |
939 | ||
940 | if (!merged) { | |
941 | __insert_discard_tree(sbi, bdev, di.lstart, di.start, | |
942 | di.len, NULL, NULL); | |
943 | f2fs_bug_on(sbi, | |
944 | !__check_rb_tree_consistence(sbi, &dcc->root)); | |
945 | } | |
946 | next: | |
947 | prev_dc = next_dc; | |
948 | if (!prev_dc) | |
949 | break; | |
950 | ||
951 | node = rb_next(&prev_dc->rb_node); | |
952 | next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); | |
953 | } | |
954 | ||
955 | mutex_unlock(&dcc->cmd_lock); | |
956 | } | |
957 | ||
958 | static int __queue_discard_cmd(struct f2fs_sb_info *sbi, | |
959 | struct block_device *bdev, block_t blkstart, block_t blklen) | |
960 | { | |
961 | block_t lblkstart = blkstart; | |
962 | ||
963 | trace_f2fs_queue_discard(bdev, blkstart, blklen); | |
964 | ||
965 | if (sbi->s_ndevs) { | |
966 | int devi = f2fs_target_device_index(sbi, blkstart); | |
967 | ||
968 | blkstart -= FDEV(devi).start_blk; | |
969 | } | |
970 | __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); | |
971 | return 0; | |
972 | } | |
973 | ||
974 | /* This should be covered by global mutex, &sit_i->sentry_lock */ | |
975 | void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) | |
976 | { | |
977 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
978 | struct discard_cmd *dc; | |
979 | ||
980 | mutex_lock(&dcc->cmd_lock); | |
981 | ||
982 | dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr); | |
983 | if (dc) { | |
984 | if (dc->state != D_PREP) | |
985 | wait_for_completion_io(&dc->wait); | |
986 | __punch_discard_cmd(sbi, dc, blkaddr); | |
987 | } | |
988 | ||
989 | mutex_unlock(&dcc->cmd_lock); | |
990 | } | |
991 | ||
992 | /* This comes from f2fs_put_super */ | |
993 | void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi) | |
994 | { | |
995 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
996 | struct list_head *pend_list; | |
997 | struct list_head *wait_list = &(dcc->wait_list); | |
998 | struct discard_cmd *dc, *tmp; | |
999 | struct blk_plug plug; | |
1000 | int i; | |
1001 | ||
1002 | mutex_lock(&dcc->cmd_lock); | |
1003 | ||
1004 | blk_start_plug(&plug); | |
1005 | for (i = 0; i < MAX_PLIST_NUM; i++) { | |
1006 | pend_list = &dcc->pend_list[i]; | |
1007 | list_for_each_entry_safe(dc, tmp, pend_list, list) | |
1008 | __submit_discard_cmd(sbi, dc); | |
1009 | } | |
1010 | blk_finish_plug(&plug); | |
1011 | ||
1012 | list_for_each_entry_safe(dc, tmp, wait_list, list) { | |
1013 | wait_for_completion_io(&dc->wait); | |
1014 | __remove_discard_cmd(sbi, dc); | |
1015 | } | |
1016 | ||
1017 | mutex_unlock(&dcc->cmd_lock); | |
1018 | } | |
1019 | ||
1020 | static int issue_discard_thread(void *data) | |
1021 | { | |
1022 | struct f2fs_sb_info *sbi = data; | |
1023 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
1024 | wait_queue_head_t *q = &dcc->discard_wait_queue; | |
1025 | struct list_head *pend_list; | |
1026 | struct list_head *wait_list = &dcc->wait_list; | |
1027 | struct discard_cmd *dc, *tmp; | |
1028 | struct blk_plug plug; | |
1029 | int iter = 0, i; | |
1030 | repeat: | |
1031 | if (kthread_should_stop()) | |
1032 | return 0; | |
1033 | ||
1034 | mutex_lock(&dcc->cmd_lock); | |
1035 | blk_start_plug(&plug); | |
1036 | for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { | |
1037 | pend_list = &dcc->pend_list[i]; | |
1038 | list_for_each_entry_safe(dc, tmp, pend_list, list) { | |
1039 | f2fs_bug_on(sbi, dc->state != D_PREP); | |
1040 | ||
1041 | if (is_idle(sbi)) | |
1042 | __submit_discard_cmd(sbi, dc); | |
1043 | ||
1044 | if (iter++ > DISCARD_ISSUE_RATE) | |
1045 | goto next_step; | |
1046 | } | |
1047 | } | |
1048 | next_step: | |
1049 | blk_finish_plug(&plug); | |
1050 | ||
1051 | list_for_each_entry_safe(dc, tmp, wait_list, list) { | |
1052 | if (dc->state == D_DONE) { | |
1053 | wait_for_completion_io(&dc->wait); | |
1054 | __remove_discard_cmd(sbi, dc); | |
1055 | } | |
1056 | } | |
1057 | mutex_unlock(&dcc->cmd_lock); | |
1058 | ||
1059 | iter = 0; | |
1060 | congestion_wait(BLK_RW_SYNC, HZ/50); | |
1061 | ||
1062 | wait_event_interruptible(*q, kthread_should_stop() || | |
1063 | atomic_read(&dcc->discard_cmd_cnt)); | |
1064 | goto repeat; | |
1065 | } | |
1066 | ||
1067 | #ifdef CONFIG_BLK_DEV_ZONED | |
1068 | static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, | |
1069 | struct block_device *bdev, block_t blkstart, block_t blklen) | |
1070 | { | |
1071 | sector_t sector, nr_sects; | |
1072 | block_t lblkstart = blkstart; | |
1073 | int devi = 0; | |
1074 | ||
1075 | if (sbi->s_ndevs) { | |
1076 | devi = f2fs_target_device_index(sbi, blkstart); | |
1077 | blkstart -= FDEV(devi).start_blk; | |
1078 | } | |
1079 | ||
1080 | /* | |
1081 | * We need to know the type of the zone: for conventional zones, | |
1082 | * use regular discard if the drive supports it. For sequential | |
1083 | * zones, reset the zone write pointer. | |
1084 | */ | |
1085 | switch (get_blkz_type(sbi, bdev, blkstart)) { | |
1086 | ||
1087 | case BLK_ZONE_TYPE_CONVENTIONAL: | |
1088 | if (!blk_queue_discard(bdev_get_queue(bdev))) | |
1089 | return 0; | |
1090 | return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); | |
1091 | case BLK_ZONE_TYPE_SEQWRITE_REQ: | |
1092 | case BLK_ZONE_TYPE_SEQWRITE_PREF: | |
1093 | sector = SECTOR_FROM_BLOCK(blkstart); | |
1094 | nr_sects = SECTOR_FROM_BLOCK(blklen); | |
1095 | ||
1096 | if (sector & (bdev_zone_sectors(bdev) - 1) || | |
1097 | nr_sects != bdev_zone_sectors(bdev)) { | |
1098 | f2fs_msg(sbi->sb, KERN_INFO, | |
1099 | "(%d) %s: Unaligned discard attempted (block %x + %x)", | |
1100 | devi, sbi->s_ndevs ? FDEV(devi).path: "", | |
1101 | blkstart, blklen); | |
1102 | return -EIO; | |
1103 | } | |
1104 | trace_f2fs_issue_reset_zone(bdev, blkstart); | |
1105 | return blkdev_reset_zones(bdev, sector, | |
1106 | nr_sects, GFP_NOFS); | |
1107 | default: | |
1108 | /* Unknown zone type: broken device ? */ | |
1109 | return -EIO; | |
1110 | } | |
1111 | } | |
1112 | #endif | |
1113 | ||
1114 | static int __issue_discard_async(struct f2fs_sb_info *sbi, | |
1115 | struct block_device *bdev, block_t blkstart, block_t blklen) | |
1116 | { | |
1117 | #ifdef CONFIG_BLK_DEV_ZONED | |
1118 | if (f2fs_sb_mounted_blkzoned(sbi->sb) && | |
1119 | bdev_zoned_model(bdev) != BLK_ZONED_NONE) | |
1120 | return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); | |
1121 | #endif | |
1122 | return __queue_discard_cmd(sbi, bdev, blkstart, blklen); | |
1123 | } | |
1124 | ||
1125 | static int f2fs_issue_discard(struct f2fs_sb_info *sbi, | |
1126 | block_t blkstart, block_t blklen) | |
1127 | { | |
1128 | sector_t start = blkstart, len = 0; | |
1129 | struct block_device *bdev; | |
1130 | struct seg_entry *se; | |
1131 | unsigned int offset; | |
1132 | block_t i; | |
1133 | int err = 0; | |
1134 | ||
1135 | bdev = f2fs_target_device(sbi, blkstart, NULL); | |
1136 | ||
1137 | for (i = blkstart; i < blkstart + blklen; i++, len++) { | |
1138 | if (i != start) { | |
1139 | struct block_device *bdev2 = | |
1140 | f2fs_target_device(sbi, i, NULL); | |
1141 | ||
1142 | if (bdev2 != bdev) { | |
1143 | err = __issue_discard_async(sbi, bdev, | |
1144 | start, len); | |
1145 | if (err) | |
1146 | return err; | |
1147 | bdev = bdev2; | |
1148 | start = i; | |
1149 | len = 0; | |
1150 | } | |
1151 | } | |
1152 | ||
1153 | se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); | |
1154 | offset = GET_BLKOFF_FROM_SEG0(sbi, i); | |
1155 | ||
1156 | if (!f2fs_test_and_set_bit(offset, se->discard_map)) | |
1157 | sbi->discard_blks--; | |
1158 | } | |
1159 | ||
1160 | if (len) | |
1161 | err = __issue_discard_async(sbi, bdev, start, len); | |
1162 | return err; | |
1163 | } | |
1164 | ||
1165 | static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, | |
1166 | bool check_only) | |
1167 | { | |
1168 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); | |
1169 | int max_blocks = sbi->blocks_per_seg; | |
1170 | struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); | |
1171 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; | |
1172 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; | |
1173 | unsigned long *discard_map = (unsigned long *)se->discard_map; | |
1174 | unsigned long *dmap = SIT_I(sbi)->tmp_map; | |
1175 | unsigned int start = 0, end = -1; | |
1176 | bool force = (cpc->reason == CP_DISCARD); | |
1177 | struct discard_entry *de = NULL; | |
1178 | struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; | |
1179 | int i; | |
1180 | ||
1181 | if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi)) | |
1182 | return false; | |
1183 | ||
1184 | if (!force) { | |
1185 | if (!test_opt(sbi, DISCARD) || !se->valid_blocks || | |
1186 | SM_I(sbi)->dcc_info->nr_discards >= | |
1187 | SM_I(sbi)->dcc_info->max_discards) | |
1188 | return false; | |
1189 | } | |
1190 | ||
1191 | /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ | |
1192 | for (i = 0; i < entries; i++) | |
1193 | dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : | |
1194 | (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; | |
1195 | ||
1196 | while (force || SM_I(sbi)->dcc_info->nr_discards <= | |
1197 | SM_I(sbi)->dcc_info->max_discards) { | |
1198 | start = __find_rev_next_bit(dmap, max_blocks, end + 1); | |
1199 | if (start >= max_blocks) | |
1200 | break; | |
1201 | ||
1202 | end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); | |
1203 | if (force && start && end != max_blocks | |
1204 | && (end - start) < cpc->trim_minlen) | |
1205 | continue; | |
1206 | ||
1207 | if (check_only) | |
1208 | return true; | |
1209 | ||
1210 | if (!de) { | |
1211 | de = f2fs_kmem_cache_alloc(discard_entry_slab, | |
1212 | GFP_F2FS_ZERO); | |
1213 | de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); | |
1214 | list_add_tail(&de->list, head); | |
1215 | } | |
1216 | ||
1217 | for (i = start; i < end; i++) | |
1218 | __set_bit_le(i, (void *)de->discard_map); | |
1219 | ||
1220 | SM_I(sbi)->dcc_info->nr_discards += end - start; | |
1221 | } | |
1222 | return false; | |
1223 | } | |
1224 | ||
1225 | void release_discard_addrs(struct f2fs_sb_info *sbi) | |
1226 | { | |
1227 | struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); | |
1228 | struct discard_entry *entry, *this; | |
1229 | ||
1230 | /* drop caches */ | |
1231 | list_for_each_entry_safe(entry, this, head, list) { | |
1232 | list_del(&entry->list); | |
1233 | kmem_cache_free(discard_entry_slab, entry); | |
1234 | } | |
1235 | } | |
1236 | ||
1237 | /* | |
1238 | * Should call clear_prefree_segments after checkpoint is done. | |
1239 | */ | |
1240 | static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) | |
1241 | { | |
1242 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1243 | unsigned int segno; | |
1244 | ||
1245 | mutex_lock(&dirty_i->seglist_lock); | |
1246 | for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) | |
1247 | __set_test_and_free(sbi, segno); | |
1248 | mutex_unlock(&dirty_i->seglist_lock); | |
1249 | } | |
1250 | ||
1251 | void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc) | |
1252 | { | |
1253 | struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); | |
1254 | struct discard_entry *entry, *this; | |
1255 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1256 | unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; | |
1257 | unsigned int start = 0, end = -1; | |
1258 | unsigned int secno, start_segno; | |
1259 | bool force = (cpc->reason == CP_DISCARD); | |
1260 | ||
1261 | mutex_lock(&dirty_i->seglist_lock); | |
1262 | ||
1263 | while (1) { | |
1264 | int i; | |
1265 | start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); | |
1266 | if (start >= MAIN_SEGS(sbi)) | |
1267 | break; | |
1268 | end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), | |
1269 | start + 1); | |
1270 | ||
1271 | for (i = start; i < end; i++) | |
1272 | clear_bit(i, prefree_map); | |
1273 | ||
1274 | dirty_i->nr_dirty[PRE] -= end - start; | |
1275 | ||
1276 | if (!test_opt(sbi, DISCARD)) | |
1277 | continue; | |
1278 | ||
1279 | if (force && start >= cpc->trim_start && | |
1280 | (end - 1) <= cpc->trim_end) | |
1281 | continue; | |
1282 | ||
1283 | if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) { | |
1284 | f2fs_issue_discard(sbi, START_BLOCK(sbi, start), | |
1285 | (end - start) << sbi->log_blocks_per_seg); | |
1286 | continue; | |
1287 | } | |
1288 | next: | |
1289 | secno = GET_SEC_FROM_SEG(sbi, start); | |
1290 | start_segno = GET_SEG_FROM_SEC(sbi, secno); | |
1291 | if (!IS_CURSEC(sbi, secno) && | |
1292 | !get_valid_blocks(sbi, start, true)) | |
1293 | f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), | |
1294 | sbi->segs_per_sec << sbi->log_blocks_per_seg); | |
1295 | ||
1296 | start = start_segno + sbi->segs_per_sec; | |
1297 | if (start < end) | |
1298 | goto next; | |
1299 | else | |
1300 | end = start - 1; | |
1301 | } | |
1302 | mutex_unlock(&dirty_i->seglist_lock); | |
1303 | ||
1304 | /* send small discards */ | |
1305 | list_for_each_entry_safe(entry, this, head, list) { | |
1306 | unsigned int cur_pos = 0, next_pos, len, total_len = 0; | |
1307 | bool is_valid = test_bit_le(0, entry->discard_map); | |
1308 | ||
1309 | find_next: | |
1310 | if (is_valid) { | |
1311 | next_pos = find_next_zero_bit_le(entry->discard_map, | |
1312 | sbi->blocks_per_seg, cur_pos); | |
1313 | len = next_pos - cur_pos; | |
1314 | ||
1315 | if (force && len < cpc->trim_minlen) | |
1316 | goto skip; | |
1317 | ||
1318 | f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, | |
1319 | len); | |
1320 | cpc->trimmed += len; | |
1321 | total_len += len; | |
1322 | } else { | |
1323 | next_pos = find_next_bit_le(entry->discard_map, | |
1324 | sbi->blocks_per_seg, cur_pos); | |
1325 | } | |
1326 | skip: | |
1327 | cur_pos = next_pos; | |
1328 | is_valid = !is_valid; | |
1329 | ||
1330 | if (cur_pos < sbi->blocks_per_seg) | |
1331 | goto find_next; | |
1332 | ||
1333 | list_del(&entry->list); | |
1334 | SM_I(sbi)->dcc_info->nr_discards -= total_len; | |
1335 | kmem_cache_free(discard_entry_slab, entry); | |
1336 | } | |
1337 | ||
1338 | wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue); | |
1339 | } | |
1340 | ||
1341 | static int create_discard_cmd_control(struct f2fs_sb_info *sbi) | |
1342 | { | |
1343 | dev_t dev = sbi->sb->s_bdev->bd_dev; | |
1344 | struct discard_cmd_control *dcc; | |
1345 | int err = 0, i; | |
1346 | ||
1347 | if (SM_I(sbi)->dcc_info) { | |
1348 | dcc = SM_I(sbi)->dcc_info; | |
1349 | goto init_thread; | |
1350 | } | |
1351 | ||
1352 | dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL); | |
1353 | if (!dcc) | |
1354 | return -ENOMEM; | |
1355 | ||
1356 | INIT_LIST_HEAD(&dcc->entry_list); | |
1357 | for (i = 0; i < MAX_PLIST_NUM; i++) | |
1358 | INIT_LIST_HEAD(&dcc->pend_list[i]); | |
1359 | INIT_LIST_HEAD(&dcc->wait_list); | |
1360 | mutex_init(&dcc->cmd_lock); | |
1361 | atomic_set(&dcc->issued_discard, 0); | |
1362 | atomic_set(&dcc->issing_discard, 0); | |
1363 | atomic_set(&dcc->discard_cmd_cnt, 0); | |
1364 | dcc->nr_discards = 0; | |
1365 | dcc->max_discards = 0; | |
1366 | dcc->undiscard_blks = 0; | |
1367 | dcc->root = RB_ROOT; | |
1368 | ||
1369 | init_waitqueue_head(&dcc->discard_wait_queue); | |
1370 | SM_I(sbi)->dcc_info = dcc; | |
1371 | init_thread: | |
1372 | dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, | |
1373 | "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); | |
1374 | if (IS_ERR(dcc->f2fs_issue_discard)) { | |
1375 | err = PTR_ERR(dcc->f2fs_issue_discard); | |
1376 | kfree(dcc); | |
1377 | SM_I(sbi)->dcc_info = NULL; | |
1378 | return err; | |
1379 | } | |
1380 | ||
1381 | return err; | |
1382 | } | |
1383 | ||
1384 | static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) | |
1385 | { | |
1386 | struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; | |
1387 | ||
1388 | if (!dcc) | |
1389 | return; | |
1390 | ||
1391 | if (dcc->f2fs_issue_discard) { | |
1392 | struct task_struct *discard_thread = dcc->f2fs_issue_discard; | |
1393 | ||
1394 | dcc->f2fs_issue_discard = NULL; | |
1395 | kthread_stop(discard_thread); | |
1396 | } | |
1397 | ||
1398 | kfree(dcc); | |
1399 | SM_I(sbi)->dcc_info = NULL; | |
1400 | } | |
1401 | ||
1402 | static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) | |
1403 | { | |
1404 | struct sit_info *sit_i = SIT_I(sbi); | |
1405 | ||
1406 | if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { | |
1407 | sit_i->dirty_sentries++; | |
1408 | return false; | |
1409 | } | |
1410 | ||
1411 | return true; | |
1412 | } | |
1413 | ||
1414 | static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, | |
1415 | unsigned int segno, int modified) | |
1416 | { | |
1417 | struct seg_entry *se = get_seg_entry(sbi, segno); | |
1418 | se->type = type; | |
1419 | if (modified) | |
1420 | __mark_sit_entry_dirty(sbi, segno); | |
1421 | } | |
1422 | ||
1423 | static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) | |
1424 | { | |
1425 | struct seg_entry *se; | |
1426 | unsigned int segno, offset; | |
1427 | long int new_vblocks; | |
1428 | ||
1429 | segno = GET_SEGNO(sbi, blkaddr); | |
1430 | ||
1431 | se = get_seg_entry(sbi, segno); | |
1432 | new_vblocks = se->valid_blocks + del; | |
1433 | offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); | |
1434 | ||
1435 | f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || | |
1436 | (new_vblocks > sbi->blocks_per_seg))); | |
1437 | ||
1438 | se->valid_blocks = new_vblocks; | |
1439 | se->mtime = get_mtime(sbi); | |
1440 | SIT_I(sbi)->max_mtime = se->mtime; | |
1441 | ||
1442 | /* Update valid block bitmap */ | |
1443 | if (del > 0) { | |
1444 | if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) { | |
1445 | #ifdef CONFIG_F2FS_CHECK_FS | |
1446 | if (f2fs_test_and_set_bit(offset, | |
1447 | se->cur_valid_map_mir)) | |
1448 | f2fs_bug_on(sbi, 1); | |
1449 | else | |
1450 | WARN_ON(1); | |
1451 | #else | |
1452 | f2fs_bug_on(sbi, 1); | |
1453 | #endif | |
1454 | } | |
1455 | if (f2fs_discard_en(sbi) && | |
1456 | !f2fs_test_and_set_bit(offset, se->discard_map)) | |
1457 | sbi->discard_blks--; | |
1458 | ||
1459 | /* don't overwrite by SSR to keep node chain */ | |
1460 | if (se->type == CURSEG_WARM_NODE) { | |
1461 | if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) | |
1462 | se->ckpt_valid_blocks++; | |
1463 | } | |
1464 | } else { | |
1465 | if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) { | |
1466 | #ifdef CONFIG_F2FS_CHECK_FS | |
1467 | if (!f2fs_test_and_clear_bit(offset, | |
1468 | se->cur_valid_map_mir)) | |
1469 | f2fs_bug_on(sbi, 1); | |
1470 | else | |
1471 | WARN_ON(1); | |
1472 | #else | |
1473 | f2fs_bug_on(sbi, 1); | |
1474 | #endif | |
1475 | } | |
1476 | if (f2fs_discard_en(sbi) && | |
1477 | f2fs_test_and_clear_bit(offset, se->discard_map)) | |
1478 | sbi->discard_blks++; | |
1479 | } | |
1480 | if (!f2fs_test_bit(offset, se->ckpt_valid_map)) | |
1481 | se->ckpt_valid_blocks += del; | |
1482 | ||
1483 | __mark_sit_entry_dirty(sbi, segno); | |
1484 | ||
1485 | /* update total number of valid blocks to be written in ckpt area */ | |
1486 | SIT_I(sbi)->written_valid_blocks += del; | |
1487 | ||
1488 | if (sbi->segs_per_sec > 1) | |
1489 | get_sec_entry(sbi, segno)->valid_blocks += del; | |
1490 | } | |
1491 | ||
1492 | void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) | |
1493 | { | |
1494 | update_sit_entry(sbi, new, 1); | |
1495 | if (GET_SEGNO(sbi, old) != NULL_SEGNO) | |
1496 | update_sit_entry(sbi, old, -1); | |
1497 | ||
1498 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); | |
1499 | locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); | |
1500 | } | |
1501 | ||
1502 | void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) | |
1503 | { | |
1504 | unsigned int segno = GET_SEGNO(sbi, addr); | |
1505 | struct sit_info *sit_i = SIT_I(sbi); | |
1506 | ||
1507 | f2fs_bug_on(sbi, addr == NULL_ADDR); | |
1508 | if (addr == NEW_ADDR) | |
1509 | return; | |
1510 | ||
1511 | /* add it into sit main buffer */ | |
1512 | mutex_lock(&sit_i->sentry_lock); | |
1513 | ||
1514 | update_sit_entry(sbi, addr, -1); | |
1515 | ||
1516 | /* add it into dirty seglist */ | |
1517 | locate_dirty_segment(sbi, segno); | |
1518 | ||
1519 | mutex_unlock(&sit_i->sentry_lock); | |
1520 | } | |
1521 | ||
1522 | bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) | |
1523 | { | |
1524 | struct sit_info *sit_i = SIT_I(sbi); | |
1525 | unsigned int segno, offset; | |
1526 | struct seg_entry *se; | |
1527 | bool is_cp = false; | |
1528 | ||
1529 | if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) | |
1530 | return true; | |
1531 | ||
1532 | mutex_lock(&sit_i->sentry_lock); | |
1533 | ||
1534 | segno = GET_SEGNO(sbi, blkaddr); | |
1535 | se = get_seg_entry(sbi, segno); | |
1536 | offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); | |
1537 | ||
1538 | if (f2fs_test_bit(offset, se->ckpt_valid_map)) | |
1539 | is_cp = true; | |
1540 | ||
1541 | mutex_unlock(&sit_i->sentry_lock); | |
1542 | ||
1543 | return is_cp; | |
1544 | } | |
1545 | ||
1546 | /* | |
1547 | * This function should be resided under the curseg_mutex lock | |
1548 | */ | |
1549 | static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, | |
1550 | struct f2fs_summary *sum) | |
1551 | { | |
1552 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1553 | void *addr = curseg->sum_blk; | |
1554 | addr += curseg->next_blkoff * sizeof(struct f2fs_summary); | |
1555 | memcpy(addr, sum, sizeof(struct f2fs_summary)); | |
1556 | } | |
1557 | ||
1558 | /* | |
1559 | * Calculate the number of current summary pages for writing | |
1560 | */ | |
1561 | int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) | |
1562 | { | |
1563 | int valid_sum_count = 0; | |
1564 | int i, sum_in_page; | |
1565 | ||
1566 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1567 | if (sbi->ckpt->alloc_type[i] == SSR) | |
1568 | valid_sum_count += sbi->blocks_per_seg; | |
1569 | else { | |
1570 | if (for_ra) | |
1571 | valid_sum_count += le16_to_cpu( | |
1572 | F2FS_CKPT(sbi)->cur_data_blkoff[i]); | |
1573 | else | |
1574 | valid_sum_count += curseg_blkoff(sbi, i); | |
1575 | } | |
1576 | } | |
1577 | ||
1578 | sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - | |
1579 | SUM_FOOTER_SIZE) / SUMMARY_SIZE; | |
1580 | if (valid_sum_count <= sum_in_page) | |
1581 | return 1; | |
1582 | else if ((valid_sum_count - sum_in_page) <= | |
1583 | (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) | |
1584 | return 2; | |
1585 | return 3; | |
1586 | } | |
1587 | ||
1588 | /* | |
1589 | * Caller should put this summary page | |
1590 | */ | |
1591 | struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) | |
1592 | { | |
1593 | return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); | |
1594 | } | |
1595 | ||
1596 | void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr) | |
1597 | { | |
1598 | struct page *page = grab_meta_page(sbi, blk_addr); | |
1599 | void *dst = page_address(page); | |
1600 | ||
1601 | if (src) | |
1602 | memcpy(dst, src, PAGE_SIZE); | |
1603 | else | |
1604 | memset(dst, 0, PAGE_SIZE); | |
1605 | set_page_dirty(page); | |
1606 | f2fs_put_page(page, 1); | |
1607 | } | |
1608 | ||
1609 | static void write_sum_page(struct f2fs_sb_info *sbi, | |
1610 | struct f2fs_summary_block *sum_blk, block_t blk_addr) | |
1611 | { | |
1612 | update_meta_page(sbi, (void *)sum_blk, blk_addr); | |
1613 | } | |
1614 | ||
1615 | static void write_current_sum_page(struct f2fs_sb_info *sbi, | |
1616 | int type, block_t blk_addr) | |
1617 | { | |
1618 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1619 | struct page *page = grab_meta_page(sbi, blk_addr); | |
1620 | struct f2fs_summary_block *src = curseg->sum_blk; | |
1621 | struct f2fs_summary_block *dst; | |
1622 | ||
1623 | dst = (struct f2fs_summary_block *)page_address(page); | |
1624 | ||
1625 | mutex_lock(&curseg->curseg_mutex); | |
1626 | ||
1627 | down_read(&curseg->journal_rwsem); | |
1628 | memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); | |
1629 | up_read(&curseg->journal_rwsem); | |
1630 | ||
1631 | memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); | |
1632 | memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); | |
1633 | ||
1634 | mutex_unlock(&curseg->curseg_mutex); | |
1635 | ||
1636 | set_page_dirty(page); | |
1637 | f2fs_put_page(page, 1); | |
1638 | } | |
1639 | ||
1640 | static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) | |
1641 | { | |
1642 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1643 | unsigned int segno = curseg->segno + 1; | |
1644 | struct free_segmap_info *free_i = FREE_I(sbi); | |
1645 | ||
1646 | if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) | |
1647 | return !test_bit(segno, free_i->free_segmap); | |
1648 | return 0; | |
1649 | } | |
1650 | ||
1651 | /* | |
1652 | * Find a new segment from the free segments bitmap to right order | |
1653 | * This function should be returned with success, otherwise BUG | |
1654 | */ | |
1655 | static void get_new_segment(struct f2fs_sb_info *sbi, | |
1656 | unsigned int *newseg, bool new_sec, int dir) | |
1657 | { | |
1658 | struct free_segmap_info *free_i = FREE_I(sbi); | |
1659 | unsigned int segno, secno, zoneno; | |
1660 | unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; | |
1661 | unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); | |
1662 | unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); | |
1663 | unsigned int left_start = hint; | |
1664 | bool init = true; | |
1665 | int go_left = 0; | |
1666 | int i; | |
1667 | ||
1668 | spin_lock(&free_i->segmap_lock); | |
1669 | ||
1670 | if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { | |
1671 | segno = find_next_zero_bit(free_i->free_segmap, | |
1672 | GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); | |
1673 | if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) | |
1674 | goto got_it; | |
1675 | } | |
1676 | find_other_zone: | |
1677 | secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); | |
1678 | if (secno >= MAIN_SECS(sbi)) { | |
1679 | if (dir == ALLOC_RIGHT) { | |
1680 | secno = find_next_zero_bit(free_i->free_secmap, | |
1681 | MAIN_SECS(sbi), 0); | |
1682 | f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); | |
1683 | } else { | |
1684 | go_left = 1; | |
1685 | left_start = hint - 1; | |
1686 | } | |
1687 | } | |
1688 | if (go_left == 0) | |
1689 | goto skip_left; | |
1690 | ||
1691 | while (test_bit(left_start, free_i->free_secmap)) { | |
1692 | if (left_start > 0) { | |
1693 | left_start--; | |
1694 | continue; | |
1695 | } | |
1696 | left_start = find_next_zero_bit(free_i->free_secmap, | |
1697 | MAIN_SECS(sbi), 0); | |
1698 | f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); | |
1699 | break; | |
1700 | } | |
1701 | secno = left_start; | |
1702 | skip_left: | |
1703 | hint = secno; | |
1704 | segno = GET_SEG_FROM_SEC(sbi, secno); | |
1705 | zoneno = GET_ZONE_FROM_SEC(sbi, secno); | |
1706 | ||
1707 | /* give up on finding another zone */ | |
1708 | if (!init) | |
1709 | goto got_it; | |
1710 | if (sbi->secs_per_zone == 1) | |
1711 | goto got_it; | |
1712 | if (zoneno == old_zoneno) | |
1713 | goto got_it; | |
1714 | if (dir == ALLOC_LEFT) { | |
1715 | if (!go_left && zoneno + 1 >= total_zones) | |
1716 | goto got_it; | |
1717 | if (go_left && zoneno == 0) | |
1718 | goto got_it; | |
1719 | } | |
1720 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
1721 | if (CURSEG_I(sbi, i)->zone == zoneno) | |
1722 | break; | |
1723 | ||
1724 | if (i < NR_CURSEG_TYPE) { | |
1725 | /* zone is in user, try another */ | |
1726 | if (go_left) | |
1727 | hint = zoneno * sbi->secs_per_zone - 1; | |
1728 | else if (zoneno + 1 >= total_zones) | |
1729 | hint = 0; | |
1730 | else | |
1731 | hint = (zoneno + 1) * sbi->secs_per_zone; | |
1732 | init = false; | |
1733 | goto find_other_zone; | |
1734 | } | |
1735 | got_it: | |
1736 | /* set it as dirty segment in free segmap */ | |
1737 | f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); | |
1738 | __set_inuse(sbi, segno); | |
1739 | *newseg = segno; | |
1740 | spin_unlock(&free_i->segmap_lock); | |
1741 | } | |
1742 | ||
1743 | static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) | |
1744 | { | |
1745 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1746 | struct summary_footer *sum_footer; | |
1747 | ||
1748 | curseg->segno = curseg->next_segno; | |
1749 | curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); | |
1750 | curseg->next_blkoff = 0; | |
1751 | curseg->next_segno = NULL_SEGNO; | |
1752 | ||
1753 | sum_footer = &(curseg->sum_blk->footer); | |
1754 | memset(sum_footer, 0, sizeof(struct summary_footer)); | |
1755 | if (IS_DATASEG(type)) | |
1756 | SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); | |
1757 | if (IS_NODESEG(type)) | |
1758 | SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); | |
1759 | __set_sit_entry_type(sbi, type, curseg->segno, modified); | |
1760 | } | |
1761 | ||
1762 | static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) | |
1763 | { | |
1764 | /* if segs_per_sec is large than 1, we need to keep original policy. */ | |
1765 | if (sbi->segs_per_sec != 1) | |
1766 | return CURSEG_I(sbi, type)->segno; | |
1767 | ||
1768 | if (type == CURSEG_HOT_DATA || IS_NODESEG(type)) | |
1769 | return 0; | |
1770 | ||
1771 | if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) | |
1772 | return SIT_I(sbi)->last_victim[ALLOC_NEXT]; | |
1773 | return CURSEG_I(sbi, type)->segno; | |
1774 | } | |
1775 | ||
1776 | /* | |
1777 | * Allocate a current working segment. | |
1778 | * This function always allocates a free segment in LFS manner. | |
1779 | */ | |
1780 | static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) | |
1781 | { | |
1782 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1783 | unsigned int segno = curseg->segno; | |
1784 | int dir = ALLOC_LEFT; | |
1785 | ||
1786 | write_sum_page(sbi, curseg->sum_blk, | |
1787 | GET_SUM_BLOCK(sbi, segno)); | |
1788 | if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) | |
1789 | dir = ALLOC_RIGHT; | |
1790 | ||
1791 | if (test_opt(sbi, NOHEAP)) | |
1792 | dir = ALLOC_RIGHT; | |
1793 | ||
1794 | segno = __get_next_segno(sbi, type); | |
1795 | get_new_segment(sbi, &segno, new_sec, dir); | |
1796 | curseg->next_segno = segno; | |
1797 | reset_curseg(sbi, type, 1); | |
1798 | curseg->alloc_type = LFS; | |
1799 | } | |
1800 | ||
1801 | static void __next_free_blkoff(struct f2fs_sb_info *sbi, | |
1802 | struct curseg_info *seg, block_t start) | |
1803 | { | |
1804 | struct seg_entry *se = get_seg_entry(sbi, seg->segno); | |
1805 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); | |
1806 | unsigned long *target_map = SIT_I(sbi)->tmp_map; | |
1807 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; | |
1808 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; | |
1809 | int i, pos; | |
1810 | ||
1811 | for (i = 0; i < entries; i++) | |
1812 | target_map[i] = ckpt_map[i] | cur_map[i]; | |
1813 | ||
1814 | pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); | |
1815 | ||
1816 | seg->next_blkoff = pos; | |
1817 | } | |
1818 | ||
1819 | /* | |
1820 | * If a segment is written by LFS manner, next block offset is just obtained | |
1821 | * by increasing the current block offset. However, if a segment is written by | |
1822 | * SSR manner, next block offset obtained by calling __next_free_blkoff | |
1823 | */ | |
1824 | static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, | |
1825 | struct curseg_info *seg) | |
1826 | { | |
1827 | if (seg->alloc_type == SSR) | |
1828 | __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); | |
1829 | else | |
1830 | seg->next_blkoff++; | |
1831 | } | |
1832 | ||
1833 | /* | |
1834 | * This function always allocates a used segment(from dirty seglist) by SSR | |
1835 | * manner, so it should recover the existing segment information of valid blocks | |
1836 | */ | |
1837 | static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) | |
1838 | { | |
1839 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1840 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1841 | unsigned int new_segno = curseg->next_segno; | |
1842 | struct f2fs_summary_block *sum_node; | |
1843 | struct page *sum_page; | |
1844 | ||
1845 | write_sum_page(sbi, curseg->sum_blk, | |
1846 | GET_SUM_BLOCK(sbi, curseg->segno)); | |
1847 | __set_test_and_inuse(sbi, new_segno); | |
1848 | ||
1849 | mutex_lock(&dirty_i->seglist_lock); | |
1850 | __remove_dirty_segment(sbi, new_segno, PRE); | |
1851 | __remove_dirty_segment(sbi, new_segno, DIRTY); | |
1852 | mutex_unlock(&dirty_i->seglist_lock); | |
1853 | ||
1854 | reset_curseg(sbi, type, 1); | |
1855 | curseg->alloc_type = SSR; | |
1856 | __next_free_blkoff(sbi, curseg, 0); | |
1857 | ||
1858 | if (reuse) { | |
1859 | sum_page = get_sum_page(sbi, new_segno); | |
1860 | sum_node = (struct f2fs_summary_block *)page_address(sum_page); | |
1861 | memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); | |
1862 | f2fs_put_page(sum_page, 1); | |
1863 | } | |
1864 | } | |
1865 | ||
1866 | static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) | |
1867 | { | |
1868 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1869 | const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; | |
1870 | unsigned segno = NULL_SEGNO; | |
1871 | int i, cnt; | |
1872 | bool reversed = false; | |
1873 | ||
1874 | /* need_SSR() already forces to do this */ | |
1875 | if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) { | |
1876 | curseg->next_segno = segno; | |
1877 | return 1; | |
1878 | } | |
1879 | ||
1880 | /* For node segments, let's do SSR more intensively */ | |
1881 | if (IS_NODESEG(type)) { | |
1882 | if (type >= CURSEG_WARM_NODE) { | |
1883 | reversed = true; | |
1884 | i = CURSEG_COLD_NODE; | |
1885 | } else { | |
1886 | i = CURSEG_HOT_NODE; | |
1887 | } | |
1888 | cnt = NR_CURSEG_NODE_TYPE; | |
1889 | } else { | |
1890 | if (type >= CURSEG_WARM_DATA) { | |
1891 | reversed = true; | |
1892 | i = CURSEG_COLD_DATA; | |
1893 | } else { | |
1894 | i = CURSEG_HOT_DATA; | |
1895 | } | |
1896 | cnt = NR_CURSEG_DATA_TYPE; | |
1897 | } | |
1898 | ||
1899 | for (; cnt-- > 0; reversed ? i-- : i++) { | |
1900 | if (i == type) | |
1901 | continue; | |
1902 | if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) { | |
1903 | curseg->next_segno = segno; | |
1904 | return 1; | |
1905 | } | |
1906 | } | |
1907 | return 0; | |
1908 | } | |
1909 | ||
1910 | /* | |
1911 | * flush out current segment and replace it with new segment | |
1912 | * This function should be returned with success, otherwise BUG | |
1913 | */ | |
1914 | static void allocate_segment_by_default(struct f2fs_sb_info *sbi, | |
1915 | int type, bool force) | |
1916 | { | |
1917 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1918 | ||
1919 | if (force) | |
1920 | new_curseg(sbi, type, true); | |
1921 | else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && | |
1922 | type == CURSEG_WARM_NODE) | |
1923 | new_curseg(sbi, type, false); | |
1924 | else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) | |
1925 | new_curseg(sbi, type, false); | |
1926 | else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) | |
1927 | change_curseg(sbi, type, true); | |
1928 | else | |
1929 | new_curseg(sbi, type, false); | |
1930 | ||
1931 | stat_inc_seg_type(sbi, curseg); | |
1932 | } | |
1933 | ||
1934 | void allocate_new_segments(struct f2fs_sb_info *sbi) | |
1935 | { | |
1936 | struct curseg_info *curseg; | |
1937 | unsigned int old_segno; | |
1938 | int i; | |
1939 | ||
1940 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1941 | curseg = CURSEG_I(sbi, i); | |
1942 | old_segno = curseg->segno; | |
1943 | SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); | |
1944 | locate_dirty_segment(sbi, old_segno); | |
1945 | } | |
1946 | } | |
1947 | ||
1948 | static const struct segment_allocation default_salloc_ops = { | |
1949 | .allocate_segment = allocate_segment_by_default, | |
1950 | }; | |
1951 | ||
1952 | bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc) | |
1953 | { | |
1954 | __u64 trim_start = cpc->trim_start; | |
1955 | bool has_candidate = false; | |
1956 | ||
1957 | mutex_lock(&SIT_I(sbi)->sentry_lock); | |
1958 | for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { | |
1959 | if (add_discard_addrs(sbi, cpc, true)) { | |
1960 | has_candidate = true; | |
1961 | break; | |
1962 | } | |
1963 | } | |
1964 | mutex_unlock(&SIT_I(sbi)->sentry_lock); | |
1965 | ||
1966 | cpc->trim_start = trim_start; | |
1967 | return has_candidate; | |
1968 | } | |
1969 | ||
1970 | int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) | |
1971 | { | |
1972 | __u64 start = F2FS_BYTES_TO_BLK(range->start); | |
1973 | __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; | |
1974 | unsigned int start_segno, end_segno; | |
1975 | struct cp_control cpc; | |
1976 | int err = 0; | |
1977 | ||
1978 | if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) | |
1979 | return -EINVAL; | |
1980 | ||
1981 | cpc.trimmed = 0; | |
1982 | if (end <= MAIN_BLKADDR(sbi)) | |
1983 | goto out; | |
1984 | ||
1985 | if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { | |
1986 | f2fs_msg(sbi->sb, KERN_WARNING, | |
1987 | "Found FS corruption, run fsck to fix."); | |
1988 | goto out; | |
1989 | } | |
1990 | ||
1991 | /* start/end segment number in main_area */ | |
1992 | start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); | |
1993 | end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : | |
1994 | GET_SEGNO(sbi, end); | |
1995 | cpc.reason = CP_DISCARD; | |
1996 | cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); | |
1997 | ||
1998 | /* do checkpoint to issue discard commands safely */ | |
1999 | for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) { | |
2000 | cpc.trim_start = start_segno; | |
2001 | ||
2002 | if (sbi->discard_blks == 0) | |
2003 | break; | |
2004 | else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi)) | |
2005 | cpc.trim_end = end_segno; | |
2006 | else | |
2007 | cpc.trim_end = min_t(unsigned int, | |
2008 | rounddown(start_segno + | |
2009 | BATCHED_TRIM_SEGMENTS(sbi), | |
2010 | sbi->segs_per_sec) - 1, end_segno); | |
2011 | ||
2012 | mutex_lock(&sbi->gc_mutex); | |
2013 | err = write_checkpoint(sbi, &cpc); | |
2014 | mutex_unlock(&sbi->gc_mutex); | |
2015 | if (err) | |
2016 | break; | |
2017 | ||
2018 | schedule(); | |
2019 | } | |
2020 | out: | |
2021 | range->len = F2FS_BLK_TO_BYTES(cpc.trimmed); | |
2022 | return err; | |
2023 | } | |
2024 | ||
2025 | static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) | |
2026 | { | |
2027 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
2028 | if (curseg->next_blkoff < sbi->blocks_per_seg) | |
2029 | return true; | |
2030 | return false; | |
2031 | } | |
2032 | ||
2033 | static int __get_segment_type_2(struct page *page, enum page_type p_type) | |
2034 | { | |
2035 | if (p_type == DATA) | |
2036 | return CURSEG_HOT_DATA; | |
2037 | else | |
2038 | return CURSEG_HOT_NODE; | |
2039 | } | |
2040 | ||
2041 | static int __get_segment_type_4(struct page *page, enum page_type p_type) | |
2042 | { | |
2043 | if (p_type == DATA) { | |
2044 | struct inode *inode = page->mapping->host; | |
2045 | ||
2046 | if (S_ISDIR(inode->i_mode)) | |
2047 | return CURSEG_HOT_DATA; | |
2048 | else | |
2049 | return CURSEG_COLD_DATA; | |
2050 | } else { | |
2051 | if (IS_DNODE(page) && is_cold_node(page)) | |
2052 | return CURSEG_WARM_NODE; | |
2053 | else | |
2054 | return CURSEG_COLD_NODE; | |
2055 | } | |
2056 | } | |
2057 | ||
2058 | static int __get_segment_type_6(struct page *page, enum page_type p_type) | |
2059 | { | |
2060 | if (p_type == DATA) { | |
2061 | struct inode *inode = page->mapping->host; | |
2062 | ||
2063 | if (is_cold_data(page) || file_is_cold(inode)) | |
2064 | return CURSEG_COLD_DATA; | |
2065 | if (is_inode_flag_set(inode, FI_HOT_DATA)) | |
2066 | return CURSEG_HOT_DATA; | |
2067 | return CURSEG_WARM_DATA; | |
2068 | } else { | |
2069 | if (IS_DNODE(page)) | |
2070 | return is_cold_node(page) ? CURSEG_WARM_NODE : | |
2071 | CURSEG_HOT_NODE; | |
2072 | return CURSEG_COLD_NODE; | |
2073 | } | |
2074 | } | |
2075 | ||
2076 | static int __get_segment_type(struct page *page, enum page_type p_type) | |
2077 | { | |
2078 | switch (F2FS_P_SB(page)->active_logs) { | |
2079 | case 2: | |
2080 | return __get_segment_type_2(page, p_type); | |
2081 | case 4: | |
2082 | return __get_segment_type_4(page, p_type); | |
2083 | } | |
2084 | /* NR_CURSEG_TYPE(6) logs by default */ | |
2085 | f2fs_bug_on(F2FS_P_SB(page), | |
2086 | F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE); | |
2087 | return __get_segment_type_6(page, p_type); | |
2088 | } | |
2089 | ||
2090 | void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, | |
2091 | block_t old_blkaddr, block_t *new_blkaddr, | |
2092 | struct f2fs_summary *sum, int type) | |
2093 | { | |
2094 | struct sit_info *sit_i = SIT_I(sbi); | |
2095 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
2096 | ||
2097 | mutex_lock(&curseg->curseg_mutex); | |
2098 | mutex_lock(&sit_i->sentry_lock); | |
2099 | ||
2100 | *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); | |
2101 | ||
2102 | f2fs_wait_discard_bio(sbi, *new_blkaddr); | |
2103 | ||
2104 | /* | |
2105 | * __add_sum_entry should be resided under the curseg_mutex | |
2106 | * because, this function updates a summary entry in the | |
2107 | * current summary block. | |
2108 | */ | |
2109 | __add_sum_entry(sbi, type, sum); | |
2110 | ||
2111 | __refresh_next_blkoff(sbi, curseg); | |
2112 | ||
2113 | stat_inc_block_count(sbi, curseg); | |
2114 | ||
2115 | if (!__has_curseg_space(sbi, type)) | |
2116 | sit_i->s_ops->allocate_segment(sbi, type, false); | |
2117 | /* | |
2118 | * SIT information should be updated after segment allocation, | |
2119 | * since we need to keep dirty segments precisely under SSR. | |
2120 | */ | |
2121 | refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); | |
2122 | ||
2123 | mutex_unlock(&sit_i->sentry_lock); | |
2124 | ||
2125 | if (page && IS_NODESEG(type)) | |
2126 | fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); | |
2127 | ||
2128 | mutex_unlock(&curseg->curseg_mutex); | |
2129 | } | |
2130 | ||
2131 | static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) | |
2132 | { | |
2133 | int type = __get_segment_type(fio->page, fio->type); | |
2134 | int err; | |
2135 | ||
2136 | if (fio->type == NODE || fio->type == DATA) | |
2137 | mutex_lock(&fio->sbi->wio_mutex[fio->type]); | |
2138 | reallocate: | |
2139 | allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, | |
2140 | &fio->new_blkaddr, sum, type); | |
2141 | ||
2142 | /* writeout dirty page into bdev */ | |
2143 | err = f2fs_submit_page_mbio(fio); | |
2144 | if (err == -EAGAIN) { | |
2145 | fio->old_blkaddr = fio->new_blkaddr; | |
2146 | goto reallocate; | |
2147 | } | |
2148 | ||
2149 | if (fio->type == NODE || fio->type == DATA) | |
2150 | mutex_unlock(&fio->sbi->wio_mutex[fio->type]); | |
2151 | } | |
2152 | ||
2153 | void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) | |
2154 | { | |
2155 | struct f2fs_io_info fio = { | |
2156 | .sbi = sbi, | |
2157 | .type = META, | |
2158 | .op = REQ_OP_WRITE, | |
2159 | .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, | |
2160 | .old_blkaddr = page->index, | |
2161 | .new_blkaddr = page->index, | |
2162 | .page = page, | |
2163 | .encrypted_page = NULL, | |
2164 | }; | |
2165 | ||
2166 | if (unlikely(page->index >= MAIN_BLKADDR(sbi))) | |
2167 | fio.op_flags &= ~REQ_META; | |
2168 | ||
2169 | set_page_writeback(page); | |
2170 | f2fs_submit_page_mbio(&fio); | |
2171 | } | |
2172 | ||
2173 | void write_node_page(unsigned int nid, struct f2fs_io_info *fio) | |
2174 | { | |
2175 | struct f2fs_summary sum; | |
2176 | ||
2177 | set_summary(&sum, nid, 0, 0); | |
2178 | do_write_page(&sum, fio); | |
2179 | } | |
2180 | ||
2181 | void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio) | |
2182 | { | |
2183 | struct f2fs_sb_info *sbi = fio->sbi; | |
2184 | struct f2fs_summary sum; | |
2185 | struct node_info ni; | |
2186 | ||
2187 | f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); | |
2188 | get_node_info(sbi, dn->nid, &ni); | |
2189 | set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); | |
2190 | do_write_page(&sum, fio); | |
2191 | f2fs_update_data_blkaddr(dn, fio->new_blkaddr); | |
2192 | } | |
2193 | ||
2194 | int rewrite_data_page(struct f2fs_io_info *fio) | |
2195 | { | |
2196 | fio->new_blkaddr = fio->old_blkaddr; | |
2197 | stat_inc_inplace_blocks(fio->sbi); | |
2198 | return f2fs_submit_page_bio(fio); | |
2199 | } | |
2200 | ||
2201 | void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, | |
2202 | block_t old_blkaddr, block_t new_blkaddr, | |
2203 | bool recover_curseg, bool recover_newaddr) | |
2204 | { | |
2205 | struct sit_info *sit_i = SIT_I(sbi); | |
2206 | struct curseg_info *curseg; | |
2207 | unsigned int segno, old_cursegno; | |
2208 | struct seg_entry *se; | |
2209 | int type; | |
2210 | unsigned short old_blkoff; | |
2211 | ||
2212 | segno = GET_SEGNO(sbi, new_blkaddr); | |
2213 | se = get_seg_entry(sbi, segno); | |
2214 | type = se->type; | |
2215 | ||
2216 | if (!recover_curseg) { | |
2217 | /* for recovery flow */ | |
2218 | if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { | |
2219 | if (old_blkaddr == NULL_ADDR) | |
2220 | type = CURSEG_COLD_DATA; | |
2221 | else | |
2222 | type = CURSEG_WARM_DATA; | |
2223 | } | |
2224 | } else { | |
2225 | if (!IS_CURSEG(sbi, segno)) | |
2226 | type = CURSEG_WARM_DATA; | |
2227 | } | |
2228 | ||
2229 | curseg = CURSEG_I(sbi, type); | |
2230 | ||
2231 | mutex_lock(&curseg->curseg_mutex); | |
2232 | mutex_lock(&sit_i->sentry_lock); | |
2233 | ||
2234 | old_cursegno = curseg->segno; | |
2235 | old_blkoff = curseg->next_blkoff; | |
2236 | ||
2237 | /* change the current segment */ | |
2238 | if (segno != curseg->segno) { | |
2239 | curseg->next_segno = segno; | |
2240 | change_curseg(sbi, type, true); | |
2241 | } | |
2242 | ||
2243 | curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); | |
2244 | __add_sum_entry(sbi, type, sum); | |
2245 | ||
2246 | if (!recover_curseg || recover_newaddr) | |
2247 | update_sit_entry(sbi, new_blkaddr, 1); | |
2248 | if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) | |
2249 | update_sit_entry(sbi, old_blkaddr, -1); | |
2250 | ||
2251 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); | |
2252 | locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); | |
2253 | ||
2254 | locate_dirty_segment(sbi, old_cursegno); | |
2255 | ||
2256 | if (recover_curseg) { | |
2257 | if (old_cursegno != curseg->segno) { | |
2258 | curseg->next_segno = old_cursegno; | |
2259 | change_curseg(sbi, type, true); | |
2260 | } | |
2261 | curseg->next_blkoff = old_blkoff; | |
2262 | } | |
2263 | ||
2264 | mutex_unlock(&sit_i->sentry_lock); | |
2265 | mutex_unlock(&curseg->curseg_mutex); | |
2266 | } | |
2267 | ||
2268 | void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, | |
2269 | block_t old_addr, block_t new_addr, | |
2270 | unsigned char version, bool recover_curseg, | |
2271 | bool recover_newaddr) | |
2272 | { | |
2273 | struct f2fs_summary sum; | |
2274 | ||
2275 | set_summary(&sum, dn->nid, dn->ofs_in_node, version); | |
2276 | ||
2277 | __f2fs_replace_block(sbi, &sum, old_addr, new_addr, | |
2278 | recover_curseg, recover_newaddr); | |
2279 | ||
2280 | f2fs_update_data_blkaddr(dn, new_addr); | |
2281 | } | |
2282 | ||
2283 | void f2fs_wait_on_page_writeback(struct page *page, | |
2284 | enum page_type type, bool ordered) | |
2285 | { | |
2286 | if (PageWriteback(page)) { | |
2287 | struct f2fs_sb_info *sbi = F2FS_P_SB(page); | |
2288 | ||
2289 | f2fs_submit_merged_bio_cond(sbi, page->mapping->host, | |
2290 | 0, page->index, type, WRITE); | |
2291 | if (ordered) | |
2292 | wait_on_page_writeback(page); | |
2293 | else | |
2294 | wait_for_stable_page(page); | |
2295 | } | |
2296 | } | |
2297 | ||
2298 | void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi, | |
2299 | block_t blkaddr) | |
2300 | { | |
2301 | struct page *cpage; | |
2302 | ||
2303 | if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) | |
2304 | return; | |
2305 | ||
2306 | cpage = find_lock_page(META_MAPPING(sbi), blkaddr); | |
2307 | if (cpage) { | |
2308 | f2fs_wait_on_page_writeback(cpage, DATA, true); | |
2309 | f2fs_put_page(cpage, 1); | |
2310 | } | |
2311 | } | |
2312 | ||
2313 | static int read_compacted_summaries(struct f2fs_sb_info *sbi) | |
2314 | { | |
2315 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
2316 | struct curseg_info *seg_i; | |
2317 | unsigned char *kaddr; | |
2318 | struct page *page; | |
2319 | block_t start; | |
2320 | int i, j, offset; | |
2321 | ||
2322 | start = start_sum_block(sbi); | |
2323 | ||
2324 | page = get_meta_page(sbi, start++); | |
2325 | kaddr = (unsigned char *)page_address(page); | |
2326 | ||
2327 | /* Step 1: restore nat cache */ | |
2328 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
2329 | memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); | |
2330 | ||
2331 | /* Step 2: restore sit cache */ | |
2332 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
2333 | memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); | |
2334 | offset = 2 * SUM_JOURNAL_SIZE; | |
2335 | ||
2336 | /* Step 3: restore summary entries */ | |
2337 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
2338 | unsigned short blk_off; | |
2339 | unsigned int segno; | |
2340 | ||
2341 | seg_i = CURSEG_I(sbi, i); | |
2342 | segno = le32_to_cpu(ckpt->cur_data_segno[i]); | |
2343 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); | |
2344 | seg_i->next_segno = segno; | |
2345 | reset_curseg(sbi, i, 0); | |
2346 | seg_i->alloc_type = ckpt->alloc_type[i]; | |
2347 | seg_i->next_blkoff = blk_off; | |
2348 | ||
2349 | if (seg_i->alloc_type == SSR) | |
2350 | blk_off = sbi->blocks_per_seg; | |
2351 | ||
2352 | for (j = 0; j < blk_off; j++) { | |
2353 | struct f2fs_summary *s; | |
2354 | s = (struct f2fs_summary *)(kaddr + offset); | |
2355 | seg_i->sum_blk->entries[j] = *s; | |
2356 | offset += SUMMARY_SIZE; | |
2357 | if (offset + SUMMARY_SIZE <= PAGE_SIZE - | |
2358 | SUM_FOOTER_SIZE) | |
2359 | continue; | |
2360 | ||
2361 | f2fs_put_page(page, 1); | |
2362 | page = NULL; | |
2363 | ||
2364 | page = get_meta_page(sbi, start++); | |
2365 | kaddr = (unsigned char *)page_address(page); | |
2366 | offset = 0; | |
2367 | } | |
2368 | } | |
2369 | f2fs_put_page(page, 1); | |
2370 | return 0; | |
2371 | } | |
2372 | ||
2373 | static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) | |
2374 | { | |
2375 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
2376 | struct f2fs_summary_block *sum; | |
2377 | struct curseg_info *curseg; | |
2378 | struct page *new; | |
2379 | unsigned short blk_off; | |
2380 | unsigned int segno = 0; | |
2381 | block_t blk_addr = 0; | |
2382 | ||
2383 | /* get segment number and block addr */ | |
2384 | if (IS_DATASEG(type)) { | |
2385 | segno = le32_to_cpu(ckpt->cur_data_segno[type]); | |
2386 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - | |
2387 | CURSEG_HOT_DATA]); | |
2388 | if (__exist_node_summaries(sbi)) | |
2389 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); | |
2390 | else | |
2391 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); | |
2392 | } else { | |
2393 | segno = le32_to_cpu(ckpt->cur_node_segno[type - | |
2394 | CURSEG_HOT_NODE]); | |
2395 | blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - | |
2396 | CURSEG_HOT_NODE]); | |
2397 | if (__exist_node_summaries(sbi)) | |
2398 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, | |
2399 | type - CURSEG_HOT_NODE); | |
2400 | else | |
2401 | blk_addr = GET_SUM_BLOCK(sbi, segno); | |
2402 | } | |
2403 | ||
2404 | new = get_meta_page(sbi, blk_addr); | |
2405 | sum = (struct f2fs_summary_block *)page_address(new); | |
2406 | ||
2407 | if (IS_NODESEG(type)) { | |
2408 | if (__exist_node_summaries(sbi)) { | |
2409 | struct f2fs_summary *ns = &sum->entries[0]; | |
2410 | int i; | |
2411 | for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { | |
2412 | ns->version = 0; | |
2413 | ns->ofs_in_node = 0; | |
2414 | } | |
2415 | } else { | |
2416 | int err; | |
2417 | ||
2418 | err = restore_node_summary(sbi, segno, sum); | |
2419 | if (err) { | |
2420 | f2fs_put_page(new, 1); | |
2421 | return err; | |
2422 | } | |
2423 | } | |
2424 | } | |
2425 | ||
2426 | /* set uncompleted segment to curseg */ | |
2427 | curseg = CURSEG_I(sbi, type); | |
2428 | mutex_lock(&curseg->curseg_mutex); | |
2429 | ||
2430 | /* update journal info */ | |
2431 | down_write(&curseg->journal_rwsem); | |
2432 | memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); | |
2433 | up_write(&curseg->journal_rwsem); | |
2434 | ||
2435 | memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); | |
2436 | memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); | |
2437 | curseg->next_segno = segno; | |
2438 | reset_curseg(sbi, type, 0); | |
2439 | curseg->alloc_type = ckpt->alloc_type[type]; | |
2440 | curseg->next_blkoff = blk_off; | |
2441 | mutex_unlock(&curseg->curseg_mutex); | |
2442 | f2fs_put_page(new, 1); | |
2443 | return 0; | |
2444 | } | |
2445 | ||
2446 | static int restore_curseg_summaries(struct f2fs_sb_info *sbi) | |
2447 | { | |
2448 | int type = CURSEG_HOT_DATA; | |
2449 | int err; | |
2450 | ||
2451 | if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { | |
2452 | int npages = npages_for_summary_flush(sbi, true); | |
2453 | ||
2454 | if (npages >= 2) | |
2455 | ra_meta_pages(sbi, start_sum_block(sbi), npages, | |
2456 | META_CP, true); | |
2457 | ||
2458 | /* restore for compacted data summary */ | |
2459 | if (read_compacted_summaries(sbi)) | |
2460 | return -EINVAL; | |
2461 | type = CURSEG_HOT_NODE; | |
2462 | } | |
2463 | ||
2464 | if (__exist_node_summaries(sbi)) | |
2465 | ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), | |
2466 | NR_CURSEG_TYPE - type, META_CP, true); | |
2467 | ||
2468 | for (; type <= CURSEG_COLD_NODE; type++) { | |
2469 | err = read_normal_summaries(sbi, type); | |
2470 | if (err) | |
2471 | return err; | |
2472 | } | |
2473 | ||
2474 | return 0; | |
2475 | } | |
2476 | ||
2477 | static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) | |
2478 | { | |
2479 | struct page *page; | |
2480 | unsigned char *kaddr; | |
2481 | struct f2fs_summary *summary; | |
2482 | struct curseg_info *seg_i; | |
2483 | int written_size = 0; | |
2484 | int i, j; | |
2485 | ||
2486 | page = grab_meta_page(sbi, blkaddr++); | |
2487 | kaddr = (unsigned char *)page_address(page); | |
2488 | ||
2489 | /* Step 1: write nat cache */ | |
2490 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
2491 | memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); | |
2492 | written_size += SUM_JOURNAL_SIZE; | |
2493 | ||
2494 | /* Step 2: write sit cache */ | |
2495 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
2496 | memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); | |
2497 | written_size += SUM_JOURNAL_SIZE; | |
2498 | ||
2499 | /* Step 3: write summary entries */ | |
2500 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
2501 | unsigned short blkoff; | |
2502 | seg_i = CURSEG_I(sbi, i); | |
2503 | if (sbi->ckpt->alloc_type[i] == SSR) | |
2504 | blkoff = sbi->blocks_per_seg; | |
2505 | else | |
2506 | blkoff = curseg_blkoff(sbi, i); | |
2507 | ||
2508 | for (j = 0; j < blkoff; j++) { | |
2509 | if (!page) { | |
2510 | page = grab_meta_page(sbi, blkaddr++); | |
2511 | kaddr = (unsigned char *)page_address(page); | |
2512 | written_size = 0; | |
2513 | } | |
2514 | summary = (struct f2fs_summary *)(kaddr + written_size); | |
2515 | *summary = seg_i->sum_blk->entries[j]; | |
2516 | written_size += SUMMARY_SIZE; | |
2517 | ||
2518 | if (written_size + SUMMARY_SIZE <= PAGE_SIZE - | |
2519 | SUM_FOOTER_SIZE) | |
2520 | continue; | |
2521 | ||
2522 | set_page_dirty(page); | |
2523 | f2fs_put_page(page, 1); | |
2524 | page = NULL; | |
2525 | } | |
2526 | } | |
2527 | if (page) { | |
2528 | set_page_dirty(page); | |
2529 | f2fs_put_page(page, 1); | |
2530 | } | |
2531 | } | |
2532 | ||
2533 | static void write_normal_summaries(struct f2fs_sb_info *sbi, | |
2534 | block_t blkaddr, int type) | |
2535 | { | |
2536 | int i, end; | |
2537 | if (IS_DATASEG(type)) | |
2538 | end = type + NR_CURSEG_DATA_TYPE; | |
2539 | else | |
2540 | end = type + NR_CURSEG_NODE_TYPE; | |
2541 | ||
2542 | for (i = type; i < end; i++) | |
2543 | write_current_sum_page(sbi, i, blkaddr + (i - type)); | |
2544 | } | |
2545 | ||
2546 | void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
2547 | { | |
2548 | if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) | |
2549 | write_compacted_summaries(sbi, start_blk); | |
2550 | else | |
2551 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); | |
2552 | } | |
2553 | ||
2554 | void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
2555 | { | |
2556 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); | |
2557 | } | |
2558 | ||
2559 | int lookup_journal_in_cursum(struct f2fs_journal *journal, int type, | |
2560 | unsigned int val, int alloc) | |
2561 | { | |
2562 | int i; | |
2563 | ||
2564 | if (type == NAT_JOURNAL) { | |
2565 | for (i = 0; i < nats_in_cursum(journal); i++) { | |
2566 | if (le32_to_cpu(nid_in_journal(journal, i)) == val) | |
2567 | return i; | |
2568 | } | |
2569 | if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) | |
2570 | return update_nats_in_cursum(journal, 1); | |
2571 | } else if (type == SIT_JOURNAL) { | |
2572 | for (i = 0; i < sits_in_cursum(journal); i++) | |
2573 | if (le32_to_cpu(segno_in_journal(journal, i)) == val) | |
2574 | return i; | |
2575 | if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) | |
2576 | return update_sits_in_cursum(journal, 1); | |
2577 | } | |
2578 | return -1; | |
2579 | } | |
2580 | ||
2581 | static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, | |
2582 | unsigned int segno) | |
2583 | { | |
2584 | return get_meta_page(sbi, current_sit_addr(sbi, segno)); | |
2585 | } | |
2586 | ||
2587 | static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, | |
2588 | unsigned int start) | |
2589 | { | |
2590 | struct sit_info *sit_i = SIT_I(sbi); | |
2591 | struct page *src_page, *dst_page; | |
2592 | pgoff_t src_off, dst_off; | |
2593 | void *src_addr, *dst_addr; | |
2594 | ||
2595 | src_off = current_sit_addr(sbi, start); | |
2596 | dst_off = next_sit_addr(sbi, src_off); | |
2597 | ||
2598 | /* get current sit block page without lock */ | |
2599 | src_page = get_meta_page(sbi, src_off); | |
2600 | dst_page = grab_meta_page(sbi, dst_off); | |
2601 | f2fs_bug_on(sbi, PageDirty(src_page)); | |
2602 | ||
2603 | src_addr = page_address(src_page); | |
2604 | dst_addr = page_address(dst_page); | |
2605 | memcpy(dst_addr, src_addr, PAGE_SIZE); | |
2606 | ||
2607 | set_page_dirty(dst_page); | |
2608 | f2fs_put_page(src_page, 1); | |
2609 | ||
2610 | set_to_next_sit(sit_i, start); | |
2611 | ||
2612 | return dst_page; | |
2613 | } | |
2614 | ||
2615 | static struct sit_entry_set *grab_sit_entry_set(void) | |
2616 | { | |
2617 | struct sit_entry_set *ses = | |
2618 | f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); | |
2619 | ||
2620 | ses->entry_cnt = 0; | |
2621 | INIT_LIST_HEAD(&ses->set_list); | |
2622 | return ses; | |
2623 | } | |
2624 | ||
2625 | static void release_sit_entry_set(struct sit_entry_set *ses) | |
2626 | { | |
2627 | list_del(&ses->set_list); | |
2628 | kmem_cache_free(sit_entry_set_slab, ses); | |
2629 | } | |
2630 | ||
2631 | static void adjust_sit_entry_set(struct sit_entry_set *ses, | |
2632 | struct list_head *head) | |
2633 | { | |
2634 | struct sit_entry_set *next = ses; | |
2635 | ||
2636 | if (list_is_last(&ses->set_list, head)) | |
2637 | return; | |
2638 | ||
2639 | list_for_each_entry_continue(next, head, set_list) | |
2640 | if (ses->entry_cnt <= next->entry_cnt) | |
2641 | break; | |
2642 | ||
2643 | list_move_tail(&ses->set_list, &next->set_list); | |
2644 | } | |
2645 | ||
2646 | static void add_sit_entry(unsigned int segno, struct list_head *head) | |
2647 | { | |
2648 | struct sit_entry_set *ses; | |
2649 | unsigned int start_segno = START_SEGNO(segno); | |
2650 | ||
2651 | list_for_each_entry(ses, head, set_list) { | |
2652 | if (ses->start_segno == start_segno) { | |
2653 | ses->entry_cnt++; | |
2654 | adjust_sit_entry_set(ses, head); | |
2655 | return; | |
2656 | } | |
2657 | } | |
2658 | ||
2659 | ses = grab_sit_entry_set(); | |
2660 | ||
2661 | ses->start_segno = start_segno; | |
2662 | ses->entry_cnt++; | |
2663 | list_add(&ses->set_list, head); | |
2664 | } | |
2665 | ||
2666 | static void add_sits_in_set(struct f2fs_sb_info *sbi) | |
2667 | { | |
2668 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
2669 | struct list_head *set_list = &sm_info->sit_entry_set; | |
2670 | unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; | |
2671 | unsigned int segno; | |
2672 | ||
2673 | for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) | |
2674 | add_sit_entry(segno, set_list); | |
2675 | } | |
2676 | ||
2677 | static void remove_sits_in_journal(struct f2fs_sb_info *sbi) | |
2678 | { | |
2679 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
2680 | struct f2fs_journal *journal = curseg->journal; | |
2681 | int i; | |
2682 | ||
2683 | down_write(&curseg->journal_rwsem); | |
2684 | for (i = 0; i < sits_in_cursum(journal); i++) { | |
2685 | unsigned int segno; | |
2686 | bool dirtied; | |
2687 | ||
2688 | segno = le32_to_cpu(segno_in_journal(journal, i)); | |
2689 | dirtied = __mark_sit_entry_dirty(sbi, segno); | |
2690 | ||
2691 | if (!dirtied) | |
2692 | add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); | |
2693 | } | |
2694 | update_sits_in_cursum(journal, -i); | |
2695 | up_write(&curseg->journal_rwsem); | |
2696 | } | |
2697 | ||
2698 | /* | |
2699 | * CP calls this function, which flushes SIT entries including sit_journal, | |
2700 | * and moves prefree segs to free segs. | |
2701 | */ | |
2702 | void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) | |
2703 | { | |
2704 | struct sit_info *sit_i = SIT_I(sbi); | |
2705 | unsigned long *bitmap = sit_i->dirty_sentries_bitmap; | |
2706 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
2707 | struct f2fs_journal *journal = curseg->journal; | |
2708 | struct sit_entry_set *ses, *tmp; | |
2709 | struct list_head *head = &SM_I(sbi)->sit_entry_set; | |
2710 | bool to_journal = true; | |
2711 | struct seg_entry *se; | |
2712 | ||
2713 | mutex_lock(&sit_i->sentry_lock); | |
2714 | ||
2715 | if (!sit_i->dirty_sentries) | |
2716 | goto out; | |
2717 | ||
2718 | /* | |
2719 | * add and account sit entries of dirty bitmap in sit entry | |
2720 | * set temporarily | |
2721 | */ | |
2722 | add_sits_in_set(sbi); | |
2723 | ||
2724 | /* | |
2725 | * if there are no enough space in journal to store dirty sit | |
2726 | * entries, remove all entries from journal and add and account | |
2727 | * them in sit entry set. | |
2728 | */ | |
2729 | if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL)) | |
2730 | remove_sits_in_journal(sbi); | |
2731 | ||
2732 | /* | |
2733 | * there are two steps to flush sit entries: | |
2734 | * #1, flush sit entries to journal in current cold data summary block. | |
2735 | * #2, flush sit entries to sit page. | |
2736 | */ | |
2737 | list_for_each_entry_safe(ses, tmp, head, set_list) { | |
2738 | struct page *page = NULL; | |
2739 | struct f2fs_sit_block *raw_sit = NULL; | |
2740 | unsigned int start_segno = ses->start_segno; | |
2741 | unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, | |
2742 | (unsigned long)MAIN_SEGS(sbi)); | |
2743 | unsigned int segno = start_segno; | |
2744 | ||
2745 | if (to_journal && | |
2746 | !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) | |
2747 | to_journal = false; | |
2748 | ||
2749 | if (to_journal) { | |
2750 | down_write(&curseg->journal_rwsem); | |
2751 | } else { | |
2752 | page = get_next_sit_page(sbi, start_segno); | |
2753 | raw_sit = page_address(page); | |
2754 | } | |
2755 | ||
2756 | /* flush dirty sit entries in region of current sit set */ | |
2757 | for_each_set_bit_from(segno, bitmap, end) { | |
2758 | int offset, sit_offset; | |
2759 | ||
2760 | se = get_seg_entry(sbi, segno); | |
2761 | ||
2762 | /* add discard candidates */ | |
2763 | if (cpc->reason != CP_DISCARD) { | |
2764 | cpc->trim_start = segno; | |
2765 | add_discard_addrs(sbi, cpc, false); | |
2766 | } | |
2767 | ||
2768 | if (to_journal) { | |
2769 | offset = lookup_journal_in_cursum(journal, | |
2770 | SIT_JOURNAL, segno, 1); | |
2771 | f2fs_bug_on(sbi, offset < 0); | |
2772 | segno_in_journal(journal, offset) = | |
2773 | cpu_to_le32(segno); | |
2774 | seg_info_to_raw_sit(se, | |
2775 | &sit_in_journal(journal, offset)); | |
2776 | } else { | |
2777 | sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); | |
2778 | seg_info_to_raw_sit(se, | |
2779 | &raw_sit->entries[sit_offset]); | |
2780 | } | |
2781 | ||
2782 | __clear_bit(segno, bitmap); | |
2783 | sit_i->dirty_sentries--; | |
2784 | ses->entry_cnt--; | |
2785 | } | |
2786 | ||
2787 | if (to_journal) | |
2788 | up_write(&curseg->journal_rwsem); | |
2789 | else | |
2790 | f2fs_put_page(page, 1); | |
2791 | ||
2792 | f2fs_bug_on(sbi, ses->entry_cnt); | |
2793 | release_sit_entry_set(ses); | |
2794 | } | |
2795 | ||
2796 | f2fs_bug_on(sbi, !list_empty(head)); | |
2797 | f2fs_bug_on(sbi, sit_i->dirty_sentries); | |
2798 | out: | |
2799 | if (cpc->reason == CP_DISCARD) { | |
2800 | __u64 trim_start = cpc->trim_start; | |
2801 | ||
2802 | for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) | |
2803 | add_discard_addrs(sbi, cpc, false); | |
2804 | ||
2805 | cpc->trim_start = trim_start; | |
2806 | } | |
2807 | mutex_unlock(&sit_i->sentry_lock); | |
2808 | ||
2809 | set_prefree_as_free_segments(sbi); | |
2810 | } | |
2811 | ||
2812 | static int build_sit_info(struct f2fs_sb_info *sbi) | |
2813 | { | |
2814 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
2815 | struct sit_info *sit_i; | |
2816 | unsigned int sit_segs, start; | |
2817 | char *src_bitmap; | |
2818 | unsigned int bitmap_size; | |
2819 | ||
2820 | /* allocate memory for SIT information */ | |
2821 | sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); | |
2822 | if (!sit_i) | |
2823 | return -ENOMEM; | |
2824 | ||
2825 | SM_I(sbi)->sit_info = sit_i; | |
2826 | ||
2827 | sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) * | |
2828 | sizeof(struct seg_entry), GFP_KERNEL); | |
2829 | if (!sit_i->sentries) | |
2830 | return -ENOMEM; | |
2831 | ||
2832 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); | |
2833 | sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); | |
2834 | if (!sit_i->dirty_sentries_bitmap) | |
2835 | return -ENOMEM; | |
2836 | ||
2837 | for (start = 0; start < MAIN_SEGS(sbi); start++) { | |
2838 | sit_i->sentries[start].cur_valid_map | |
2839 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
2840 | sit_i->sentries[start].ckpt_valid_map | |
2841 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
2842 | if (!sit_i->sentries[start].cur_valid_map || | |
2843 | !sit_i->sentries[start].ckpt_valid_map) | |
2844 | return -ENOMEM; | |
2845 | ||
2846 | #ifdef CONFIG_F2FS_CHECK_FS | |
2847 | sit_i->sentries[start].cur_valid_map_mir | |
2848 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
2849 | if (!sit_i->sentries[start].cur_valid_map_mir) | |
2850 | return -ENOMEM; | |
2851 | #endif | |
2852 | ||
2853 | if (f2fs_discard_en(sbi)) { | |
2854 | sit_i->sentries[start].discard_map | |
2855 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
2856 | if (!sit_i->sentries[start].discard_map) | |
2857 | return -ENOMEM; | |
2858 | } | |
2859 | } | |
2860 | ||
2861 | sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
2862 | if (!sit_i->tmp_map) | |
2863 | return -ENOMEM; | |
2864 | ||
2865 | if (sbi->segs_per_sec > 1) { | |
2866 | sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) * | |
2867 | sizeof(struct sec_entry), GFP_KERNEL); | |
2868 | if (!sit_i->sec_entries) | |
2869 | return -ENOMEM; | |
2870 | } | |
2871 | ||
2872 | /* get information related with SIT */ | |
2873 | sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; | |
2874 | ||
2875 | /* setup SIT bitmap from ckeckpoint pack */ | |
2876 | bitmap_size = __bitmap_size(sbi, SIT_BITMAP); | |
2877 | src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); | |
2878 | ||
2879 | sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); | |
2880 | if (!sit_i->sit_bitmap) | |
2881 | return -ENOMEM; | |
2882 | ||
2883 | #ifdef CONFIG_F2FS_CHECK_FS | |
2884 | sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); | |
2885 | if (!sit_i->sit_bitmap_mir) | |
2886 | return -ENOMEM; | |
2887 | #endif | |
2888 | ||
2889 | /* init SIT information */ | |
2890 | sit_i->s_ops = &default_salloc_ops; | |
2891 | ||
2892 | sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); | |
2893 | sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; | |
2894 | sit_i->written_valid_blocks = 0; | |
2895 | sit_i->bitmap_size = bitmap_size; | |
2896 | sit_i->dirty_sentries = 0; | |
2897 | sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; | |
2898 | sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); | |
2899 | sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; | |
2900 | mutex_init(&sit_i->sentry_lock); | |
2901 | return 0; | |
2902 | } | |
2903 | ||
2904 | static int build_free_segmap(struct f2fs_sb_info *sbi) | |
2905 | { | |
2906 | struct free_segmap_info *free_i; | |
2907 | unsigned int bitmap_size, sec_bitmap_size; | |
2908 | ||
2909 | /* allocate memory for free segmap information */ | |
2910 | free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); | |
2911 | if (!free_i) | |
2912 | return -ENOMEM; | |
2913 | ||
2914 | SM_I(sbi)->free_info = free_i; | |
2915 | ||
2916 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); | |
2917 | free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL); | |
2918 | if (!free_i->free_segmap) | |
2919 | return -ENOMEM; | |
2920 | ||
2921 | sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); | |
2922 | free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL); | |
2923 | if (!free_i->free_secmap) | |
2924 | return -ENOMEM; | |
2925 | ||
2926 | /* set all segments as dirty temporarily */ | |
2927 | memset(free_i->free_segmap, 0xff, bitmap_size); | |
2928 | memset(free_i->free_secmap, 0xff, sec_bitmap_size); | |
2929 | ||
2930 | /* init free segmap information */ | |
2931 | free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); | |
2932 | free_i->free_segments = 0; | |
2933 | free_i->free_sections = 0; | |
2934 | spin_lock_init(&free_i->segmap_lock); | |
2935 | return 0; | |
2936 | } | |
2937 | ||
2938 | static int build_curseg(struct f2fs_sb_info *sbi) | |
2939 | { | |
2940 | struct curseg_info *array; | |
2941 | int i; | |
2942 | ||
2943 | array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL); | |
2944 | if (!array) | |
2945 | return -ENOMEM; | |
2946 | ||
2947 | SM_I(sbi)->curseg_array = array; | |
2948 | ||
2949 | for (i = 0; i < NR_CURSEG_TYPE; i++) { | |
2950 | mutex_init(&array[i].curseg_mutex); | |
2951 | array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
2952 | if (!array[i].sum_blk) | |
2953 | return -ENOMEM; | |
2954 | init_rwsem(&array[i].journal_rwsem); | |
2955 | array[i].journal = kzalloc(sizeof(struct f2fs_journal), | |
2956 | GFP_KERNEL); | |
2957 | if (!array[i].journal) | |
2958 | return -ENOMEM; | |
2959 | array[i].segno = NULL_SEGNO; | |
2960 | array[i].next_blkoff = 0; | |
2961 | } | |
2962 | return restore_curseg_summaries(sbi); | |
2963 | } | |
2964 | ||
2965 | static void build_sit_entries(struct f2fs_sb_info *sbi) | |
2966 | { | |
2967 | struct sit_info *sit_i = SIT_I(sbi); | |
2968 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
2969 | struct f2fs_journal *journal = curseg->journal; | |
2970 | struct seg_entry *se; | |
2971 | struct f2fs_sit_entry sit; | |
2972 | int sit_blk_cnt = SIT_BLK_CNT(sbi); | |
2973 | unsigned int i, start, end; | |
2974 | unsigned int readed, start_blk = 0; | |
2975 | ||
2976 | do { | |
2977 | readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES, | |
2978 | META_SIT, true); | |
2979 | ||
2980 | start = start_blk * sit_i->sents_per_block; | |
2981 | end = (start_blk + readed) * sit_i->sents_per_block; | |
2982 | ||
2983 | for (; start < end && start < MAIN_SEGS(sbi); start++) { | |
2984 | struct f2fs_sit_block *sit_blk; | |
2985 | struct page *page; | |
2986 | ||
2987 | se = &sit_i->sentries[start]; | |
2988 | page = get_current_sit_page(sbi, start); | |
2989 | sit_blk = (struct f2fs_sit_block *)page_address(page); | |
2990 | sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; | |
2991 | f2fs_put_page(page, 1); | |
2992 | ||
2993 | check_block_count(sbi, start, &sit); | |
2994 | seg_info_from_raw_sit(se, &sit); | |
2995 | ||
2996 | /* build discard map only one time */ | |
2997 | if (f2fs_discard_en(sbi)) { | |
2998 | memcpy(se->discard_map, se->cur_valid_map, | |
2999 | SIT_VBLOCK_MAP_SIZE); | |
3000 | sbi->discard_blks += sbi->blocks_per_seg - | |
3001 | se->valid_blocks; | |
3002 | } | |
3003 | ||
3004 | if (sbi->segs_per_sec > 1) | |
3005 | get_sec_entry(sbi, start)->valid_blocks += | |
3006 | se->valid_blocks; | |
3007 | } | |
3008 | start_blk += readed; | |
3009 | } while (start_blk < sit_blk_cnt); | |
3010 | ||
3011 | down_read(&curseg->journal_rwsem); | |
3012 | for (i = 0; i < sits_in_cursum(journal); i++) { | |
3013 | unsigned int old_valid_blocks; | |
3014 | ||
3015 | start = le32_to_cpu(segno_in_journal(journal, i)); | |
3016 | se = &sit_i->sentries[start]; | |
3017 | sit = sit_in_journal(journal, i); | |
3018 | ||
3019 | old_valid_blocks = se->valid_blocks; | |
3020 | ||
3021 | check_block_count(sbi, start, &sit); | |
3022 | seg_info_from_raw_sit(se, &sit); | |
3023 | ||
3024 | if (f2fs_discard_en(sbi)) { | |
3025 | memcpy(se->discard_map, se->cur_valid_map, | |
3026 | SIT_VBLOCK_MAP_SIZE); | |
3027 | sbi->discard_blks += old_valid_blocks - | |
3028 | se->valid_blocks; | |
3029 | } | |
3030 | ||
3031 | if (sbi->segs_per_sec > 1) | |
3032 | get_sec_entry(sbi, start)->valid_blocks += | |
3033 | se->valid_blocks - old_valid_blocks; | |
3034 | } | |
3035 | up_read(&curseg->journal_rwsem); | |
3036 | } | |
3037 | ||
3038 | static void init_free_segmap(struct f2fs_sb_info *sbi) | |
3039 | { | |
3040 | unsigned int start; | |
3041 | int type; | |
3042 | ||
3043 | for (start = 0; start < MAIN_SEGS(sbi); start++) { | |
3044 | struct seg_entry *sentry = get_seg_entry(sbi, start); | |
3045 | if (!sentry->valid_blocks) | |
3046 | __set_free(sbi, start); | |
3047 | else | |
3048 | SIT_I(sbi)->written_valid_blocks += | |
3049 | sentry->valid_blocks; | |
3050 | } | |
3051 | ||
3052 | /* set use the current segments */ | |
3053 | for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { | |
3054 | struct curseg_info *curseg_t = CURSEG_I(sbi, type); | |
3055 | __set_test_and_inuse(sbi, curseg_t->segno); | |
3056 | } | |
3057 | } | |
3058 | ||
3059 | static void init_dirty_segmap(struct f2fs_sb_info *sbi) | |
3060 | { | |
3061 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
3062 | struct free_segmap_info *free_i = FREE_I(sbi); | |
3063 | unsigned int segno = 0, offset = 0; | |
3064 | unsigned short valid_blocks; | |
3065 | ||
3066 | while (1) { | |
3067 | /* find dirty segment based on free segmap */ | |
3068 | segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); | |
3069 | if (segno >= MAIN_SEGS(sbi)) | |
3070 | break; | |
3071 | offset = segno + 1; | |
3072 | valid_blocks = get_valid_blocks(sbi, segno, false); | |
3073 | if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) | |
3074 | continue; | |
3075 | if (valid_blocks > sbi->blocks_per_seg) { | |
3076 | f2fs_bug_on(sbi, 1); | |
3077 | continue; | |
3078 | } | |
3079 | mutex_lock(&dirty_i->seglist_lock); | |
3080 | __locate_dirty_segment(sbi, segno, DIRTY); | |
3081 | mutex_unlock(&dirty_i->seglist_lock); | |
3082 | } | |
3083 | } | |
3084 | ||
3085 | static int init_victim_secmap(struct f2fs_sb_info *sbi) | |
3086 | { | |
3087 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
3088 | unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); | |
3089 | ||
3090 | dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); | |
3091 | if (!dirty_i->victim_secmap) | |
3092 | return -ENOMEM; | |
3093 | return 0; | |
3094 | } | |
3095 | ||
3096 | static int build_dirty_segmap(struct f2fs_sb_info *sbi) | |
3097 | { | |
3098 | struct dirty_seglist_info *dirty_i; | |
3099 | unsigned int bitmap_size, i; | |
3100 | ||
3101 | /* allocate memory for dirty segments list information */ | |
3102 | dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); | |
3103 | if (!dirty_i) | |
3104 | return -ENOMEM; | |
3105 | ||
3106 | SM_I(sbi)->dirty_info = dirty_i; | |
3107 | mutex_init(&dirty_i->seglist_lock); | |
3108 | ||
3109 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); | |
3110 | ||
3111 | for (i = 0; i < NR_DIRTY_TYPE; i++) { | |
3112 | dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); | |
3113 | if (!dirty_i->dirty_segmap[i]) | |
3114 | return -ENOMEM; | |
3115 | } | |
3116 | ||
3117 | init_dirty_segmap(sbi); | |
3118 | return init_victim_secmap(sbi); | |
3119 | } | |
3120 | ||
3121 | /* | |
3122 | * Update min, max modified time for cost-benefit GC algorithm | |
3123 | */ | |
3124 | static void init_min_max_mtime(struct f2fs_sb_info *sbi) | |
3125 | { | |
3126 | struct sit_info *sit_i = SIT_I(sbi); | |
3127 | unsigned int segno; | |
3128 | ||
3129 | mutex_lock(&sit_i->sentry_lock); | |
3130 | ||
3131 | sit_i->min_mtime = LLONG_MAX; | |
3132 | ||
3133 | for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { | |
3134 | unsigned int i; | |
3135 | unsigned long long mtime = 0; | |
3136 | ||
3137 | for (i = 0; i < sbi->segs_per_sec; i++) | |
3138 | mtime += get_seg_entry(sbi, segno + i)->mtime; | |
3139 | ||
3140 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
3141 | ||
3142 | if (sit_i->min_mtime > mtime) | |
3143 | sit_i->min_mtime = mtime; | |
3144 | } | |
3145 | sit_i->max_mtime = get_mtime(sbi); | |
3146 | mutex_unlock(&sit_i->sentry_lock); | |
3147 | } | |
3148 | ||
3149 | int build_segment_manager(struct f2fs_sb_info *sbi) | |
3150 | { | |
3151 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
3152 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
3153 | struct f2fs_sm_info *sm_info; | |
3154 | int err; | |
3155 | ||
3156 | sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); | |
3157 | if (!sm_info) | |
3158 | return -ENOMEM; | |
3159 | ||
3160 | /* init sm info */ | |
3161 | sbi->sm_info = sm_info; | |
3162 | sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); | |
3163 | sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); | |
3164 | sm_info->segment_count = le32_to_cpu(raw_super->segment_count); | |
3165 | sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); | |
3166 | sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); | |
3167 | sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); | |
3168 | sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); | |
3169 | sm_info->rec_prefree_segments = sm_info->main_segments * | |
3170 | DEF_RECLAIM_PREFREE_SEGMENTS / 100; | |
3171 | if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) | |
3172 | sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; | |
3173 | ||
3174 | if (!test_opt(sbi, LFS)) | |
3175 | sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; | |
3176 | sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; | |
3177 | sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; | |
3178 | sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; | |
3179 | ||
3180 | sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS; | |
3181 | ||
3182 | INIT_LIST_HEAD(&sm_info->sit_entry_set); | |
3183 | ||
3184 | if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) { | |
3185 | err = create_flush_cmd_control(sbi); | |
3186 | if (err) | |
3187 | return err; | |
3188 | } | |
3189 | ||
3190 | err = create_discard_cmd_control(sbi); | |
3191 | if (err) | |
3192 | return err; | |
3193 | ||
3194 | err = build_sit_info(sbi); | |
3195 | if (err) | |
3196 | return err; | |
3197 | err = build_free_segmap(sbi); | |
3198 | if (err) | |
3199 | return err; | |
3200 | err = build_curseg(sbi); | |
3201 | if (err) | |
3202 | return err; | |
3203 | ||
3204 | /* reinit free segmap based on SIT */ | |
3205 | build_sit_entries(sbi); | |
3206 | ||
3207 | init_free_segmap(sbi); | |
3208 | err = build_dirty_segmap(sbi); | |
3209 | if (err) | |
3210 | return err; | |
3211 | ||
3212 | init_min_max_mtime(sbi); | |
3213 | return 0; | |
3214 | } | |
3215 | ||
3216 | static void discard_dirty_segmap(struct f2fs_sb_info *sbi, | |
3217 | enum dirty_type dirty_type) | |
3218 | { | |
3219 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
3220 | ||
3221 | mutex_lock(&dirty_i->seglist_lock); | |
3222 | kvfree(dirty_i->dirty_segmap[dirty_type]); | |
3223 | dirty_i->nr_dirty[dirty_type] = 0; | |
3224 | mutex_unlock(&dirty_i->seglist_lock); | |
3225 | } | |
3226 | ||
3227 | static void destroy_victim_secmap(struct f2fs_sb_info *sbi) | |
3228 | { | |
3229 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
3230 | kvfree(dirty_i->victim_secmap); | |
3231 | } | |
3232 | ||
3233 | static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) | |
3234 | { | |
3235 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
3236 | int i; | |
3237 | ||
3238 | if (!dirty_i) | |
3239 | return; | |
3240 | ||
3241 | /* discard pre-free/dirty segments list */ | |
3242 | for (i = 0; i < NR_DIRTY_TYPE; i++) | |
3243 | discard_dirty_segmap(sbi, i); | |
3244 | ||
3245 | destroy_victim_secmap(sbi); | |
3246 | SM_I(sbi)->dirty_info = NULL; | |
3247 | kfree(dirty_i); | |
3248 | } | |
3249 | ||
3250 | static void destroy_curseg(struct f2fs_sb_info *sbi) | |
3251 | { | |
3252 | struct curseg_info *array = SM_I(sbi)->curseg_array; | |
3253 | int i; | |
3254 | ||
3255 | if (!array) | |
3256 | return; | |
3257 | SM_I(sbi)->curseg_array = NULL; | |
3258 | for (i = 0; i < NR_CURSEG_TYPE; i++) { | |
3259 | kfree(array[i].sum_blk); | |
3260 | kfree(array[i].journal); | |
3261 | } | |
3262 | kfree(array); | |
3263 | } | |
3264 | ||
3265 | static void destroy_free_segmap(struct f2fs_sb_info *sbi) | |
3266 | { | |
3267 | struct free_segmap_info *free_i = SM_I(sbi)->free_info; | |
3268 | if (!free_i) | |
3269 | return; | |
3270 | SM_I(sbi)->free_info = NULL; | |
3271 | kvfree(free_i->free_segmap); | |
3272 | kvfree(free_i->free_secmap); | |
3273 | kfree(free_i); | |
3274 | } | |
3275 | ||
3276 | static void destroy_sit_info(struct f2fs_sb_info *sbi) | |
3277 | { | |
3278 | struct sit_info *sit_i = SIT_I(sbi); | |
3279 | unsigned int start; | |
3280 | ||
3281 | if (!sit_i) | |
3282 | return; | |
3283 | ||
3284 | if (sit_i->sentries) { | |
3285 | for (start = 0; start < MAIN_SEGS(sbi); start++) { | |
3286 | kfree(sit_i->sentries[start].cur_valid_map); | |
3287 | #ifdef CONFIG_F2FS_CHECK_FS | |
3288 | kfree(sit_i->sentries[start].cur_valid_map_mir); | |
3289 | #endif | |
3290 | kfree(sit_i->sentries[start].ckpt_valid_map); | |
3291 | kfree(sit_i->sentries[start].discard_map); | |
3292 | } | |
3293 | } | |
3294 | kfree(sit_i->tmp_map); | |
3295 | ||
3296 | kvfree(sit_i->sentries); | |
3297 | kvfree(sit_i->sec_entries); | |
3298 | kvfree(sit_i->dirty_sentries_bitmap); | |
3299 | ||
3300 | SM_I(sbi)->sit_info = NULL; | |
3301 | kfree(sit_i->sit_bitmap); | |
3302 | #ifdef CONFIG_F2FS_CHECK_FS | |
3303 | kfree(sit_i->sit_bitmap_mir); | |
3304 | #endif | |
3305 | kfree(sit_i); | |
3306 | } | |
3307 | ||
3308 | void destroy_segment_manager(struct f2fs_sb_info *sbi) | |
3309 | { | |
3310 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
3311 | ||
3312 | if (!sm_info) | |
3313 | return; | |
3314 | destroy_flush_cmd_control(sbi, true); | |
3315 | destroy_discard_cmd_control(sbi); | |
3316 | destroy_dirty_segmap(sbi); | |
3317 | destroy_curseg(sbi); | |
3318 | destroy_free_segmap(sbi); | |
3319 | destroy_sit_info(sbi); | |
3320 | sbi->sm_info = NULL; | |
3321 | kfree(sm_info); | |
3322 | } | |
3323 | ||
3324 | int __init create_segment_manager_caches(void) | |
3325 | { | |
3326 | discard_entry_slab = f2fs_kmem_cache_create("discard_entry", | |
3327 | sizeof(struct discard_entry)); | |
3328 | if (!discard_entry_slab) | |
3329 | goto fail; | |
3330 | ||
3331 | discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd", | |
3332 | sizeof(struct discard_cmd)); | |
3333 | if (!discard_cmd_slab) | |
3334 | goto destroy_discard_entry; | |
3335 | ||
3336 | sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", | |
3337 | sizeof(struct sit_entry_set)); | |
3338 | if (!sit_entry_set_slab) | |
3339 | goto destroy_discard_cmd; | |
3340 | ||
3341 | inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", | |
3342 | sizeof(struct inmem_pages)); | |
3343 | if (!inmem_entry_slab) | |
3344 | goto destroy_sit_entry_set; | |
3345 | return 0; | |
3346 | ||
3347 | destroy_sit_entry_set: | |
3348 | kmem_cache_destroy(sit_entry_set_slab); | |
3349 | destroy_discard_cmd: | |
3350 | kmem_cache_destroy(discard_cmd_slab); | |
3351 | destroy_discard_entry: | |
3352 | kmem_cache_destroy(discard_entry_slab); | |
3353 | fail: | |
3354 | return -ENOMEM; | |
3355 | } | |
3356 | ||
3357 | void destroy_segment_manager_caches(void) | |
3358 | { | |
3359 | kmem_cache_destroy(sit_entry_set_slab); | |
3360 | kmem_cache_destroy(discard_cmd_slab); | |
3361 | kmem_cache_destroy(discard_entry_slab); | |
3362 | kmem_cache_destroy(inmem_entry_slab); | |
3363 | } |