]>
Commit | Line | Data |
---|---|---|
0a8165d7 | 1 | /* |
351df4b2 JK |
2 | * fs/f2fs/segment.c |
3 | * | |
4 | * Copyright (c) 2012 Samsung Electronics Co., Ltd. | |
5 | * http://www.samsung.com/ | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | */ | |
11 | #include <linux/fs.h> | |
12 | #include <linux/f2fs_fs.h> | |
13 | #include <linux/bio.h> | |
14 | #include <linux/blkdev.h> | |
690e4a3e | 15 | #include <linux/prefetch.h> |
6b4afdd7 | 16 | #include <linux/kthread.h> |
351df4b2 | 17 | #include <linux/vmalloc.h> |
74de593a | 18 | #include <linux/swap.h> |
351df4b2 JK |
19 | |
20 | #include "f2fs.h" | |
21 | #include "segment.h" | |
22 | #include "node.h" | |
9e4ded3f | 23 | #include "trace.h" |
6ec178da | 24 | #include <trace/events/f2fs.h> |
351df4b2 | 25 | |
9a7f143a CL |
26 | #define __reverse_ffz(x) __reverse_ffs(~(x)) |
27 | ||
7fd9e544 | 28 | static struct kmem_cache *discard_entry_slab; |
184a5cd2 | 29 | static struct kmem_cache *sit_entry_set_slab; |
88b88a66 | 30 | static struct kmem_cache *inmem_entry_slab; |
7fd9e544 | 31 | |
9a7f143a CL |
32 | /* |
33 | * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since | |
34 | * MSB and LSB are reversed in a byte by f2fs_set_bit. | |
35 | */ | |
36 | static inline unsigned long __reverse_ffs(unsigned long word) | |
37 | { | |
38 | int num = 0; | |
39 | ||
40 | #if BITS_PER_LONG == 64 | |
41 | if ((word & 0xffffffff) == 0) { | |
42 | num += 32; | |
43 | word >>= 32; | |
44 | } | |
45 | #endif | |
46 | if ((word & 0xffff) == 0) { | |
47 | num += 16; | |
48 | word >>= 16; | |
49 | } | |
50 | if ((word & 0xff) == 0) { | |
51 | num += 8; | |
52 | word >>= 8; | |
53 | } | |
54 | if ((word & 0xf0) == 0) | |
55 | num += 4; | |
56 | else | |
57 | word >>= 4; | |
58 | if ((word & 0xc) == 0) | |
59 | num += 2; | |
60 | else | |
61 | word >>= 2; | |
62 | if ((word & 0x2) == 0) | |
63 | num += 1; | |
64 | return num; | |
65 | } | |
66 | ||
67 | /* | |
e1c42045 | 68 | * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because |
9a7f143a CL |
69 | * f2fs_set_bit makes MSB and LSB reversed in a byte. |
70 | * Example: | |
71 | * LSB <--> MSB | |
72 | * f2fs_set_bit(0, bitmap) => 0000 0001 | |
73 | * f2fs_set_bit(7, bitmap) => 1000 0000 | |
74 | */ | |
75 | static unsigned long __find_rev_next_bit(const unsigned long *addr, | |
76 | unsigned long size, unsigned long offset) | |
77 | { | |
78 | const unsigned long *p = addr + BIT_WORD(offset); | |
79 | unsigned long result = offset & ~(BITS_PER_LONG - 1); | |
80 | unsigned long tmp; | |
81 | unsigned long mask, submask; | |
82 | unsigned long quot, rest; | |
83 | ||
84 | if (offset >= size) | |
85 | return size; | |
86 | ||
87 | size -= result; | |
88 | offset %= BITS_PER_LONG; | |
89 | if (!offset) | |
90 | goto aligned; | |
91 | ||
92 | tmp = *(p++); | |
93 | quot = (offset >> 3) << 3; | |
94 | rest = offset & 0x7; | |
95 | mask = ~0UL << quot; | |
96 | submask = (unsigned char)(0xff << rest) >> rest; | |
97 | submask <<= quot; | |
98 | mask &= submask; | |
99 | tmp &= mask; | |
100 | if (size < BITS_PER_LONG) | |
101 | goto found_first; | |
102 | if (tmp) | |
103 | goto found_middle; | |
104 | ||
105 | size -= BITS_PER_LONG; | |
106 | result += BITS_PER_LONG; | |
107 | aligned: | |
108 | while (size & ~(BITS_PER_LONG-1)) { | |
109 | tmp = *(p++); | |
110 | if (tmp) | |
111 | goto found_middle; | |
112 | result += BITS_PER_LONG; | |
113 | size -= BITS_PER_LONG; | |
114 | } | |
115 | if (!size) | |
116 | return result; | |
117 | tmp = *p; | |
118 | found_first: | |
119 | tmp &= (~0UL >> (BITS_PER_LONG - size)); | |
120 | if (tmp == 0UL) /* Are any bits set? */ | |
121 | return result + size; /* Nope. */ | |
122 | found_middle: | |
123 | return result + __reverse_ffs(tmp); | |
124 | } | |
125 | ||
126 | static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, | |
127 | unsigned long size, unsigned long offset) | |
128 | { | |
129 | const unsigned long *p = addr + BIT_WORD(offset); | |
130 | unsigned long result = offset & ~(BITS_PER_LONG - 1); | |
131 | unsigned long tmp; | |
132 | unsigned long mask, submask; | |
133 | unsigned long quot, rest; | |
134 | ||
135 | if (offset >= size) | |
136 | return size; | |
137 | ||
138 | size -= result; | |
139 | offset %= BITS_PER_LONG; | |
140 | if (!offset) | |
141 | goto aligned; | |
142 | ||
143 | tmp = *(p++); | |
144 | quot = (offset >> 3) << 3; | |
145 | rest = offset & 0x7; | |
146 | mask = ~(~0UL << quot); | |
147 | submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest); | |
148 | submask <<= quot; | |
149 | mask += submask; | |
150 | tmp |= mask; | |
151 | if (size < BITS_PER_LONG) | |
152 | goto found_first; | |
153 | if (~tmp) | |
154 | goto found_middle; | |
155 | ||
156 | size -= BITS_PER_LONG; | |
157 | result += BITS_PER_LONG; | |
158 | aligned: | |
159 | while (size & ~(BITS_PER_LONG - 1)) { | |
160 | tmp = *(p++); | |
161 | if (~tmp) | |
162 | goto found_middle; | |
163 | result += BITS_PER_LONG; | |
164 | size -= BITS_PER_LONG; | |
165 | } | |
166 | if (!size) | |
167 | return result; | |
168 | tmp = *p; | |
169 | ||
170 | found_first: | |
171 | tmp |= ~0UL << size; | |
172 | if (tmp == ~0UL) /* Are any bits zero? */ | |
173 | return result + size; /* Nope. */ | |
174 | found_middle: | |
175 | return result + __reverse_ffz(tmp); | |
176 | } | |
177 | ||
88b88a66 JK |
178 | void register_inmem_page(struct inode *inode, struct page *page) |
179 | { | |
180 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
181 | struct inmem_pages *new; | |
34ba94ba | 182 | int err; |
9be32d72 | 183 | |
0722b101 | 184 | SetPagePrivate(page); |
9e4ded3f | 185 | f2fs_trace_pid(page); |
0722b101 | 186 | |
88b88a66 JK |
187 | new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); |
188 | ||
189 | /* add atomic page indices to the list */ | |
190 | new->page = page; | |
191 | INIT_LIST_HEAD(&new->list); | |
9be32d72 | 192 | retry: |
88b88a66 JK |
193 | /* increase reference count with clean state */ |
194 | mutex_lock(&fi->inmem_lock); | |
34ba94ba JK |
195 | err = radix_tree_insert(&fi->inmem_root, page->index, new); |
196 | if (err == -EEXIST) { | |
197 | mutex_unlock(&fi->inmem_lock); | |
198 | kmem_cache_free(inmem_entry_slab, new); | |
199 | return; | |
200 | } else if (err) { | |
201 | mutex_unlock(&fi->inmem_lock); | |
34ba94ba JK |
202 | goto retry; |
203 | } | |
88b88a66 JK |
204 | get_page(page); |
205 | list_add_tail(&new->list, &fi->inmem_pages); | |
8dcf2ff7 | 206 | inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); |
88b88a66 | 207 | mutex_unlock(&fi->inmem_lock); |
8ce67cb0 JK |
208 | |
209 | trace_f2fs_register_inmem_page(page, INMEM); | |
88b88a66 JK |
210 | } |
211 | ||
212 | void commit_inmem_pages(struct inode *inode, bool abort) | |
213 | { | |
214 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
215 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
216 | struct inmem_pages *cur, *tmp; | |
217 | bool submit_bio = false; | |
218 | struct f2fs_io_info fio = { | |
05ca3632 | 219 | .sbi = sbi, |
88b88a66 | 220 | .type = DATA, |
1e84371f | 221 | .rw = WRITE_SYNC | REQ_PRIO, |
88b88a66 JK |
222 | }; |
223 | ||
0341845e JK |
224 | /* |
225 | * The abort is true only when f2fs_evict_inode is called. | |
226 | * Basically, the f2fs_evict_inode doesn't produce any data writes, so | |
227 | * that we don't need to call f2fs_balance_fs. | |
228 | * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this | |
229 | * inode becomes free by iget_locked in f2fs_iget. | |
230 | */ | |
70c640b1 | 231 | if (!abort) { |
0341845e | 232 | f2fs_balance_fs(sbi); |
70c640b1 JK |
233 | f2fs_lock_op(sbi); |
234 | } | |
88b88a66 JK |
235 | |
236 | mutex_lock(&fi->inmem_lock); | |
237 | list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { | |
70c640b1 JK |
238 | if (!abort) { |
239 | lock_page(cur->page); | |
240 | if (cur->page->mapping == inode->i_mapping) { | |
241 | f2fs_wait_on_page_writeback(cur->page, DATA); | |
242 | if (clear_page_dirty_for_io(cur->page)) | |
243 | inode_dec_dirty_pages(inode); | |
8ce67cb0 | 244 | trace_f2fs_commit_inmem_page(cur->page, INMEM); |
05ca3632 JK |
245 | fio.page = cur->page; |
246 | do_write_data_page(&fio); | |
70c640b1 JK |
247 | submit_bio = true; |
248 | } | |
249 | f2fs_put_page(cur->page, 1); | |
250 | } else { | |
8ce67cb0 | 251 | trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP); |
70c640b1 | 252 | put_page(cur->page); |
88b88a66 | 253 | } |
34ba94ba | 254 | radix_tree_delete(&fi->inmem_root, cur->page->index); |
88b88a66 JK |
255 | list_del(&cur->list); |
256 | kmem_cache_free(inmem_entry_slab, cur); | |
8dcf2ff7 | 257 | dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); |
88b88a66 | 258 | } |
88b88a66 JK |
259 | mutex_unlock(&fi->inmem_lock); |
260 | ||
70c640b1 JK |
261 | if (!abort) { |
262 | f2fs_unlock_op(sbi); | |
263 | if (submit_bio) | |
264 | f2fs_submit_merged_bio(sbi, DATA, WRITE); | |
265 | } | |
88b88a66 JK |
266 | } |
267 | ||
0a8165d7 | 268 | /* |
351df4b2 JK |
269 | * This function balances dirty node and dentry pages. |
270 | * In addition, it controls garbage collection. | |
271 | */ | |
272 | void f2fs_balance_fs(struct f2fs_sb_info *sbi) | |
273 | { | |
351df4b2 | 274 | /* |
029cd28c JK |
275 | * We should do GC or end up with checkpoint, if there are so many dirty |
276 | * dir/node pages without enough free segments. | |
351df4b2 | 277 | */ |
43727527 | 278 | if (has_not_enough_free_secs(sbi, 0)) { |
351df4b2 | 279 | mutex_lock(&sbi->gc_mutex); |
408e9375 | 280 | f2fs_gc(sbi); |
351df4b2 JK |
281 | } |
282 | } | |
283 | ||
4660f9c0 JK |
284 | void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) |
285 | { | |
1dcc336b CY |
286 | /* try to shrink extent cache when there is no enough memory */ |
287 | f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); | |
288 | ||
4660f9c0 JK |
289 | /* check the # of cached NAT entries and prefree segments */ |
290 | if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) || | |
e5e7ea3c | 291 | excess_prefree_segs(sbi) || |
88a70a69 | 292 | !available_free_memory(sbi, INO_ENTRIES)) |
4660f9c0 JK |
293 | f2fs_sync_fs(sbi->sb, true); |
294 | } | |
295 | ||
2163d198 | 296 | static int issue_flush_thread(void *data) |
6b4afdd7 JK |
297 | { |
298 | struct f2fs_sb_info *sbi = data; | |
a688b9d9 GZ |
299 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
300 | wait_queue_head_t *q = &fcc->flush_wait_queue; | |
6b4afdd7 JK |
301 | repeat: |
302 | if (kthread_should_stop()) | |
303 | return 0; | |
304 | ||
721bd4d5 | 305 | if (!llist_empty(&fcc->issue_list)) { |
6b4afdd7 JK |
306 | struct bio *bio = bio_alloc(GFP_NOIO, 0); |
307 | struct flush_cmd *cmd, *next; | |
308 | int ret; | |
309 | ||
721bd4d5 GZ |
310 | fcc->dispatch_list = llist_del_all(&fcc->issue_list); |
311 | fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); | |
312 | ||
6b4afdd7 JK |
313 | bio->bi_bdev = sbi->sb->s_bdev; |
314 | ret = submit_bio_wait(WRITE_FLUSH, bio); | |
315 | ||
721bd4d5 GZ |
316 | llist_for_each_entry_safe(cmd, next, |
317 | fcc->dispatch_list, llnode) { | |
6b4afdd7 | 318 | cmd->ret = ret; |
6b4afdd7 JK |
319 | complete(&cmd->wait); |
320 | } | |
a4ed23f2 | 321 | bio_put(bio); |
a688b9d9 | 322 | fcc->dispatch_list = NULL; |
6b4afdd7 JK |
323 | } |
324 | ||
a688b9d9 | 325 | wait_event_interruptible(*q, |
721bd4d5 | 326 | kthread_should_stop() || !llist_empty(&fcc->issue_list)); |
6b4afdd7 JK |
327 | goto repeat; |
328 | } | |
329 | ||
330 | int f2fs_issue_flush(struct f2fs_sb_info *sbi) | |
331 | { | |
a688b9d9 | 332 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
adf8d90b | 333 | struct flush_cmd cmd; |
6b4afdd7 | 334 | |
24a9ee0f JK |
335 | trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER), |
336 | test_opt(sbi, FLUSH_MERGE)); | |
337 | ||
0f7b2abd JK |
338 | if (test_opt(sbi, NOBARRIER)) |
339 | return 0; | |
340 | ||
6b4afdd7 JK |
341 | if (!test_opt(sbi, FLUSH_MERGE)) |
342 | return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL); | |
343 | ||
adf8d90b | 344 | init_completion(&cmd.wait); |
6b4afdd7 | 345 | |
721bd4d5 | 346 | llist_add(&cmd.llnode, &fcc->issue_list); |
6b4afdd7 | 347 | |
a688b9d9 GZ |
348 | if (!fcc->dispatch_list) |
349 | wake_up(&fcc->flush_wait_queue); | |
6b4afdd7 | 350 | |
adf8d90b CY |
351 | wait_for_completion(&cmd.wait); |
352 | ||
353 | return cmd.ret; | |
6b4afdd7 JK |
354 | } |
355 | ||
2163d198 GZ |
356 | int create_flush_cmd_control(struct f2fs_sb_info *sbi) |
357 | { | |
358 | dev_t dev = sbi->sb->s_bdev->bd_dev; | |
359 | struct flush_cmd_control *fcc; | |
360 | int err = 0; | |
361 | ||
362 | fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL); | |
363 | if (!fcc) | |
364 | return -ENOMEM; | |
2163d198 | 365 | init_waitqueue_head(&fcc->flush_wait_queue); |
721bd4d5 | 366 | init_llist_head(&fcc->issue_list); |
6b2920a5 | 367 | SM_I(sbi)->cmd_control_info = fcc; |
2163d198 GZ |
368 | fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, |
369 | "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); | |
370 | if (IS_ERR(fcc->f2fs_issue_flush)) { | |
371 | err = PTR_ERR(fcc->f2fs_issue_flush); | |
372 | kfree(fcc); | |
6b2920a5 | 373 | SM_I(sbi)->cmd_control_info = NULL; |
2163d198 GZ |
374 | return err; |
375 | } | |
2163d198 GZ |
376 | |
377 | return err; | |
378 | } | |
379 | ||
380 | void destroy_flush_cmd_control(struct f2fs_sb_info *sbi) | |
381 | { | |
6b2920a5 | 382 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
2163d198 GZ |
383 | |
384 | if (fcc && fcc->f2fs_issue_flush) | |
385 | kthread_stop(fcc->f2fs_issue_flush); | |
386 | kfree(fcc); | |
6b2920a5 | 387 | SM_I(sbi)->cmd_control_info = NULL; |
2163d198 GZ |
388 | } |
389 | ||
351df4b2 JK |
390 | static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, |
391 | enum dirty_type dirty_type) | |
392 | { | |
393 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
394 | ||
395 | /* need not be added */ | |
396 | if (IS_CURSEG(sbi, segno)) | |
397 | return; | |
398 | ||
399 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
400 | dirty_i->nr_dirty[dirty_type]++; | |
401 | ||
402 | if (dirty_type == DIRTY) { | |
403 | struct seg_entry *sentry = get_seg_entry(sbi, segno); | |
4625d6aa | 404 | enum dirty_type t = sentry->type; |
b2f2c390 | 405 | |
ec325b52 JK |
406 | if (unlikely(t >= DIRTY)) { |
407 | f2fs_bug_on(sbi, 1); | |
408 | return; | |
409 | } | |
4625d6aa CL |
410 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) |
411 | dirty_i->nr_dirty[t]++; | |
351df4b2 JK |
412 | } |
413 | } | |
414 | ||
415 | static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, | |
416 | enum dirty_type dirty_type) | |
417 | { | |
418 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
419 | ||
420 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
421 | dirty_i->nr_dirty[dirty_type]--; | |
422 | ||
423 | if (dirty_type == DIRTY) { | |
4625d6aa CL |
424 | struct seg_entry *sentry = get_seg_entry(sbi, segno); |
425 | enum dirty_type t = sentry->type; | |
426 | ||
427 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) | |
428 | dirty_i->nr_dirty[t]--; | |
b2f2c390 | 429 | |
5ec4e49f JK |
430 | if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0) |
431 | clear_bit(GET_SECNO(sbi, segno), | |
432 | dirty_i->victim_secmap); | |
351df4b2 JK |
433 | } |
434 | } | |
435 | ||
0a8165d7 | 436 | /* |
351df4b2 JK |
437 | * Should not occur error such as -ENOMEM. |
438 | * Adding dirty entry into seglist is not critical operation. | |
439 | * If a given segment is one of current working segments, it won't be added. | |
440 | */ | |
8d8451af | 441 | static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) |
351df4b2 JK |
442 | { |
443 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
444 | unsigned short valid_blocks; | |
445 | ||
446 | if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) | |
447 | return; | |
448 | ||
449 | mutex_lock(&dirty_i->seglist_lock); | |
450 | ||
451 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
452 | ||
453 | if (valid_blocks == 0) { | |
454 | __locate_dirty_segment(sbi, segno, PRE); | |
455 | __remove_dirty_segment(sbi, segno, DIRTY); | |
456 | } else if (valid_blocks < sbi->blocks_per_seg) { | |
457 | __locate_dirty_segment(sbi, segno, DIRTY); | |
458 | } else { | |
459 | /* Recovery routine with SSR needs this */ | |
460 | __remove_dirty_segment(sbi, segno, DIRTY); | |
461 | } | |
462 | ||
463 | mutex_unlock(&dirty_i->seglist_lock); | |
351df4b2 JK |
464 | } |
465 | ||
1e87a78d | 466 | static int f2fs_issue_discard(struct f2fs_sb_info *sbi, |
37208879 JK |
467 | block_t blkstart, block_t blklen) |
468 | { | |
55cf9cb6 CY |
469 | sector_t start = SECTOR_FROM_BLOCK(blkstart); |
470 | sector_t len = SECTOR_FROM_BLOCK(blklen); | |
1661d07c | 471 | trace_f2fs_issue_discard(sbi->sb, blkstart, blklen); |
1e87a78d JK |
472 | return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0); |
473 | } | |
474 | ||
cf2271e7 | 475 | void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr) |
1e87a78d | 476 | { |
1e87a78d JK |
477 | if (f2fs_issue_discard(sbi, blkaddr, 1)) { |
478 | struct page *page = grab_meta_page(sbi, blkaddr); | |
479 | /* zero-filled page */ | |
480 | set_page_dirty(page); | |
481 | f2fs_put_page(page, 1); | |
482 | } | |
37208879 JK |
483 | } |
484 | ||
adf4983b JK |
485 | static void __add_discard_entry(struct f2fs_sb_info *sbi, |
486 | struct cp_control *cpc, unsigned int start, unsigned int end) | |
b2955550 JK |
487 | { |
488 | struct list_head *head = &SM_I(sbi)->discard_list; | |
adf4983b JK |
489 | struct discard_entry *new, *last; |
490 | ||
491 | if (!list_empty(head)) { | |
492 | last = list_last_entry(head, struct discard_entry, list); | |
493 | if (START_BLOCK(sbi, cpc->trim_start) + start == | |
494 | last->blkaddr + last->len) { | |
495 | last->len += end - start; | |
496 | goto done; | |
497 | } | |
498 | } | |
499 | ||
500 | new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS); | |
501 | INIT_LIST_HEAD(&new->list); | |
502 | new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start; | |
503 | new->len = end - start; | |
504 | list_add_tail(&new->list, head); | |
505 | done: | |
506 | SM_I(sbi)->nr_discards += end - start; | |
507 | cpc->trimmed += end - start; | |
508 | } | |
509 | ||
510 | static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc) | |
511 | { | |
b2955550 JK |
512 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); |
513 | int max_blocks = sbi->blocks_per_seg; | |
4b2fecc8 | 514 | struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); |
b2955550 JK |
515 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; |
516 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; | |
60a3b782 | 517 | unsigned long *dmap = SIT_I(sbi)->tmp_map; |
b2955550 | 518 | unsigned int start = 0, end = -1; |
4b2fecc8 | 519 | bool force = (cpc->reason == CP_DISCARD); |
b2955550 JK |
520 | int i; |
521 | ||
d7bc2484 JK |
522 | if (!force && (!test_opt(sbi, DISCARD) || |
523 | SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)) | |
b2955550 JK |
524 | return; |
525 | ||
4b2fecc8 JK |
526 | if (force && !se->valid_blocks) { |
527 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
528 | /* | |
529 | * if this segment is registered in the prefree list, then | |
530 | * we should skip adding a discard candidate, and let the | |
531 | * checkpoint do that later. | |
532 | */ | |
533 | mutex_lock(&dirty_i->seglist_lock); | |
534 | if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) { | |
535 | mutex_unlock(&dirty_i->seglist_lock); | |
536 | cpc->trimmed += sbi->blocks_per_seg; | |
537 | return; | |
538 | } | |
539 | mutex_unlock(&dirty_i->seglist_lock); | |
540 | ||
adf4983b | 541 | __add_discard_entry(sbi, cpc, 0, sbi->blocks_per_seg); |
4b2fecc8 JK |
542 | return; |
543 | } | |
544 | ||
b2955550 JK |
545 | /* zero block will be discarded through the prefree list */ |
546 | if (!se->valid_blocks || se->valid_blocks == max_blocks) | |
547 | return; | |
548 | ||
549 | /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ | |
550 | for (i = 0; i < entries; i++) | |
d7bc2484 JK |
551 | dmap[i] = force ? ~ckpt_map[i] : |
552 | (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; | |
b2955550 | 553 | |
4b2fecc8 | 554 | while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) { |
b2955550 JK |
555 | start = __find_rev_next_bit(dmap, max_blocks, end + 1); |
556 | if (start >= max_blocks) | |
557 | break; | |
558 | ||
559 | end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); | |
560 | ||
b28c3f94 | 561 | if (force && end - start < cpc->trim_minlen) |
4b2fecc8 JK |
562 | continue; |
563 | ||
adf4983b | 564 | __add_discard_entry(sbi, cpc, start, end); |
b2955550 JK |
565 | } |
566 | } | |
567 | ||
4b2fecc8 JK |
568 | void release_discard_addrs(struct f2fs_sb_info *sbi) |
569 | { | |
570 | struct list_head *head = &(SM_I(sbi)->discard_list); | |
571 | struct discard_entry *entry, *this; | |
572 | ||
573 | /* drop caches */ | |
574 | list_for_each_entry_safe(entry, this, head, list) { | |
575 | list_del(&entry->list); | |
576 | kmem_cache_free(discard_entry_slab, entry); | |
577 | } | |
578 | } | |
579 | ||
0a8165d7 | 580 | /* |
351df4b2 JK |
581 | * Should call clear_prefree_segments after checkpoint is done. |
582 | */ | |
583 | static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) | |
584 | { | |
585 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
b65ee148 | 586 | unsigned int segno; |
351df4b2 JK |
587 | |
588 | mutex_lock(&dirty_i->seglist_lock); | |
7cd8558b | 589 | for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) |
351df4b2 | 590 | __set_test_and_free(sbi, segno); |
351df4b2 JK |
591 | mutex_unlock(&dirty_i->seglist_lock); |
592 | } | |
593 | ||
594 | void clear_prefree_segments(struct f2fs_sb_info *sbi) | |
595 | { | |
b2955550 | 596 | struct list_head *head = &(SM_I(sbi)->discard_list); |
2d7b822a | 597 | struct discard_entry *entry, *this; |
351df4b2 | 598 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); |
29e59c14 | 599 | unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; |
29e59c14 | 600 | unsigned int start = 0, end = -1; |
351df4b2 JK |
601 | |
602 | mutex_lock(&dirty_i->seglist_lock); | |
29e59c14 | 603 | |
351df4b2 | 604 | while (1) { |
29e59c14 | 605 | int i; |
7cd8558b JK |
606 | start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); |
607 | if (start >= MAIN_SEGS(sbi)) | |
351df4b2 | 608 | break; |
7cd8558b JK |
609 | end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), |
610 | start + 1); | |
29e59c14 CL |
611 | |
612 | for (i = start; i < end; i++) | |
613 | clear_bit(i, prefree_map); | |
614 | ||
615 | dirty_i->nr_dirty[PRE] -= end - start; | |
616 | ||
617 | if (!test_opt(sbi, DISCARD)) | |
618 | continue; | |
351df4b2 | 619 | |
37208879 JK |
620 | f2fs_issue_discard(sbi, START_BLOCK(sbi, start), |
621 | (end - start) << sbi->log_blocks_per_seg); | |
351df4b2 JK |
622 | } |
623 | mutex_unlock(&dirty_i->seglist_lock); | |
b2955550 JK |
624 | |
625 | /* send small discards */ | |
2d7b822a | 626 | list_for_each_entry_safe(entry, this, head, list) { |
37208879 | 627 | f2fs_issue_discard(sbi, entry->blkaddr, entry->len); |
b2955550 JK |
628 | list_del(&entry->list); |
629 | SM_I(sbi)->nr_discards -= entry->len; | |
630 | kmem_cache_free(discard_entry_slab, entry); | |
631 | } | |
351df4b2 JK |
632 | } |
633 | ||
184a5cd2 | 634 | static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) |
351df4b2 JK |
635 | { |
636 | struct sit_info *sit_i = SIT_I(sbi); | |
184a5cd2 CY |
637 | |
638 | if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { | |
351df4b2 | 639 | sit_i->dirty_sentries++; |
184a5cd2 CY |
640 | return false; |
641 | } | |
642 | ||
643 | return true; | |
351df4b2 JK |
644 | } |
645 | ||
646 | static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, | |
647 | unsigned int segno, int modified) | |
648 | { | |
649 | struct seg_entry *se = get_seg_entry(sbi, segno); | |
650 | se->type = type; | |
651 | if (modified) | |
652 | __mark_sit_entry_dirty(sbi, segno); | |
653 | } | |
654 | ||
655 | static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) | |
656 | { | |
657 | struct seg_entry *se; | |
658 | unsigned int segno, offset; | |
659 | long int new_vblocks; | |
660 | ||
661 | segno = GET_SEGNO(sbi, blkaddr); | |
662 | ||
663 | se = get_seg_entry(sbi, segno); | |
664 | new_vblocks = se->valid_blocks + del; | |
491c0854 | 665 | offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); |
351df4b2 | 666 | |
9850cf4a | 667 | f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || |
351df4b2 JK |
668 | (new_vblocks > sbi->blocks_per_seg))); |
669 | ||
670 | se->valid_blocks = new_vblocks; | |
671 | se->mtime = get_mtime(sbi); | |
672 | SIT_I(sbi)->max_mtime = se->mtime; | |
673 | ||
674 | /* Update valid block bitmap */ | |
675 | if (del > 0) { | |
52aca074 | 676 | if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) |
05796763 | 677 | f2fs_bug_on(sbi, 1); |
351df4b2 | 678 | } else { |
52aca074 | 679 | if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) |
05796763 | 680 | f2fs_bug_on(sbi, 1); |
351df4b2 JK |
681 | } |
682 | if (!f2fs_test_bit(offset, se->ckpt_valid_map)) | |
683 | se->ckpt_valid_blocks += del; | |
684 | ||
685 | __mark_sit_entry_dirty(sbi, segno); | |
686 | ||
687 | /* update total number of valid blocks to be written in ckpt area */ | |
688 | SIT_I(sbi)->written_valid_blocks += del; | |
689 | ||
690 | if (sbi->segs_per_sec > 1) | |
691 | get_sec_entry(sbi, segno)->valid_blocks += del; | |
692 | } | |
693 | ||
5e443818 | 694 | void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) |
351df4b2 | 695 | { |
5e443818 JK |
696 | update_sit_entry(sbi, new, 1); |
697 | if (GET_SEGNO(sbi, old) != NULL_SEGNO) | |
698 | update_sit_entry(sbi, old, -1); | |
699 | ||
700 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); | |
701 | locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); | |
351df4b2 JK |
702 | } |
703 | ||
704 | void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) | |
705 | { | |
706 | unsigned int segno = GET_SEGNO(sbi, addr); | |
707 | struct sit_info *sit_i = SIT_I(sbi); | |
708 | ||
9850cf4a | 709 | f2fs_bug_on(sbi, addr == NULL_ADDR); |
351df4b2 JK |
710 | if (addr == NEW_ADDR) |
711 | return; | |
712 | ||
713 | /* add it into sit main buffer */ | |
714 | mutex_lock(&sit_i->sentry_lock); | |
715 | ||
716 | update_sit_entry(sbi, addr, -1); | |
717 | ||
718 | /* add it into dirty seglist */ | |
719 | locate_dirty_segment(sbi, segno); | |
720 | ||
721 | mutex_unlock(&sit_i->sentry_lock); | |
722 | } | |
723 | ||
0a8165d7 | 724 | /* |
351df4b2 JK |
725 | * This function should be resided under the curseg_mutex lock |
726 | */ | |
727 | static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, | |
e79efe3b | 728 | struct f2fs_summary *sum) |
351df4b2 JK |
729 | { |
730 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
731 | void *addr = curseg->sum_blk; | |
e79efe3b | 732 | addr += curseg->next_blkoff * sizeof(struct f2fs_summary); |
351df4b2 | 733 | memcpy(addr, sum, sizeof(struct f2fs_summary)); |
351df4b2 JK |
734 | } |
735 | ||
0a8165d7 | 736 | /* |
351df4b2 JK |
737 | * Calculate the number of current summary pages for writing |
738 | */ | |
3fa06d7b | 739 | int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) |
351df4b2 | 740 | { |
351df4b2 | 741 | int valid_sum_count = 0; |
9a47938b | 742 | int i, sum_in_page; |
351df4b2 JK |
743 | |
744 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
745 | if (sbi->ckpt->alloc_type[i] == SSR) | |
746 | valid_sum_count += sbi->blocks_per_seg; | |
3fa06d7b CY |
747 | else { |
748 | if (for_ra) | |
749 | valid_sum_count += le16_to_cpu( | |
750 | F2FS_CKPT(sbi)->cur_data_blkoff[i]); | |
751 | else | |
752 | valid_sum_count += curseg_blkoff(sbi, i); | |
753 | } | |
351df4b2 JK |
754 | } |
755 | ||
9a47938b FL |
756 | sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE - |
757 | SUM_FOOTER_SIZE) / SUMMARY_SIZE; | |
758 | if (valid_sum_count <= sum_in_page) | |
351df4b2 | 759 | return 1; |
9a47938b FL |
760 | else if ((valid_sum_count - sum_in_page) <= |
761 | (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) | |
351df4b2 JK |
762 | return 2; |
763 | return 3; | |
764 | } | |
765 | ||
0a8165d7 | 766 | /* |
351df4b2 JK |
767 | * Caller should put this summary page |
768 | */ | |
769 | struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) | |
770 | { | |
771 | return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); | |
772 | } | |
773 | ||
774 | static void write_sum_page(struct f2fs_sb_info *sbi, | |
775 | struct f2fs_summary_block *sum_blk, block_t blk_addr) | |
776 | { | |
777 | struct page *page = grab_meta_page(sbi, blk_addr); | |
778 | void *kaddr = page_address(page); | |
779 | memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE); | |
780 | set_page_dirty(page); | |
781 | f2fs_put_page(page, 1); | |
782 | } | |
783 | ||
60374688 JK |
784 | static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) |
785 | { | |
786 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
81fb5e87 | 787 | unsigned int segno = curseg->segno + 1; |
60374688 JK |
788 | struct free_segmap_info *free_i = FREE_I(sbi); |
789 | ||
7cd8558b | 790 | if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) |
81fb5e87 | 791 | return !test_bit(segno, free_i->free_segmap); |
60374688 JK |
792 | return 0; |
793 | } | |
794 | ||
0a8165d7 | 795 | /* |
351df4b2 JK |
796 | * Find a new segment from the free segments bitmap to right order |
797 | * This function should be returned with success, otherwise BUG | |
798 | */ | |
799 | static void get_new_segment(struct f2fs_sb_info *sbi, | |
800 | unsigned int *newseg, bool new_sec, int dir) | |
801 | { | |
802 | struct free_segmap_info *free_i = FREE_I(sbi); | |
351df4b2 | 803 | unsigned int segno, secno, zoneno; |
7cd8558b | 804 | unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; |
351df4b2 JK |
805 | unsigned int hint = *newseg / sbi->segs_per_sec; |
806 | unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); | |
807 | unsigned int left_start = hint; | |
808 | bool init = true; | |
809 | int go_left = 0; | |
810 | int i; | |
811 | ||
1a118ccf | 812 | spin_lock(&free_i->segmap_lock); |
351df4b2 JK |
813 | |
814 | if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { | |
815 | segno = find_next_zero_bit(free_i->free_segmap, | |
7cd8558b | 816 | MAIN_SEGS(sbi), *newseg + 1); |
33afa7fd JK |
817 | if (segno - *newseg < sbi->segs_per_sec - |
818 | (*newseg % sbi->segs_per_sec)) | |
351df4b2 JK |
819 | goto got_it; |
820 | } | |
821 | find_other_zone: | |
7cd8558b JK |
822 | secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); |
823 | if (secno >= MAIN_SECS(sbi)) { | |
351df4b2 JK |
824 | if (dir == ALLOC_RIGHT) { |
825 | secno = find_next_zero_bit(free_i->free_secmap, | |
7cd8558b JK |
826 | MAIN_SECS(sbi), 0); |
827 | f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); | |
351df4b2 JK |
828 | } else { |
829 | go_left = 1; | |
830 | left_start = hint - 1; | |
831 | } | |
832 | } | |
833 | if (go_left == 0) | |
834 | goto skip_left; | |
835 | ||
836 | while (test_bit(left_start, free_i->free_secmap)) { | |
837 | if (left_start > 0) { | |
838 | left_start--; | |
839 | continue; | |
840 | } | |
841 | left_start = find_next_zero_bit(free_i->free_secmap, | |
7cd8558b JK |
842 | MAIN_SECS(sbi), 0); |
843 | f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); | |
351df4b2 JK |
844 | break; |
845 | } | |
846 | secno = left_start; | |
847 | skip_left: | |
848 | hint = secno; | |
849 | segno = secno * sbi->segs_per_sec; | |
850 | zoneno = secno / sbi->secs_per_zone; | |
851 | ||
852 | /* give up on finding another zone */ | |
853 | if (!init) | |
854 | goto got_it; | |
855 | if (sbi->secs_per_zone == 1) | |
856 | goto got_it; | |
857 | if (zoneno == old_zoneno) | |
858 | goto got_it; | |
859 | if (dir == ALLOC_LEFT) { | |
860 | if (!go_left && zoneno + 1 >= total_zones) | |
861 | goto got_it; | |
862 | if (go_left && zoneno == 0) | |
863 | goto got_it; | |
864 | } | |
865 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
866 | if (CURSEG_I(sbi, i)->zone == zoneno) | |
867 | break; | |
868 | ||
869 | if (i < NR_CURSEG_TYPE) { | |
870 | /* zone is in user, try another */ | |
871 | if (go_left) | |
872 | hint = zoneno * sbi->secs_per_zone - 1; | |
873 | else if (zoneno + 1 >= total_zones) | |
874 | hint = 0; | |
875 | else | |
876 | hint = (zoneno + 1) * sbi->secs_per_zone; | |
877 | init = false; | |
878 | goto find_other_zone; | |
879 | } | |
880 | got_it: | |
881 | /* set it as dirty segment in free segmap */ | |
9850cf4a | 882 | f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); |
351df4b2 JK |
883 | __set_inuse(sbi, segno); |
884 | *newseg = segno; | |
1a118ccf | 885 | spin_unlock(&free_i->segmap_lock); |
351df4b2 JK |
886 | } |
887 | ||
888 | static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) | |
889 | { | |
890 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
891 | struct summary_footer *sum_footer; | |
892 | ||
893 | curseg->segno = curseg->next_segno; | |
894 | curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); | |
895 | curseg->next_blkoff = 0; | |
896 | curseg->next_segno = NULL_SEGNO; | |
897 | ||
898 | sum_footer = &(curseg->sum_blk->footer); | |
899 | memset(sum_footer, 0, sizeof(struct summary_footer)); | |
900 | if (IS_DATASEG(type)) | |
901 | SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); | |
902 | if (IS_NODESEG(type)) | |
903 | SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); | |
904 | __set_sit_entry_type(sbi, type, curseg->segno, modified); | |
905 | } | |
906 | ||
0a8165d7 | 907 | /* |
351df4b2 JK |
908 | * Allocate a current working segment. |
909 | * This function always allocates a free segment in LFS manner. | |
910 | */ | |
911 | static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) | |
912 | { | |
913 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
914 | unsigned int segno = curseg->segno; | |
915 | int dir = ALLOC_LEFT; | |
916 | ||
917 | write_sum_page(sbi, curseg->sum_blk, | |
81fb5e87 | 918 | GET_SUM_BLOCK(sbi, segno)); |
351df4b2 JK |
919 | if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) |
920 | dir = ALLOC_RIGHT; | |
921 | ||
922 | if (test_opt(sbi, NOHEAP)) | |
923 | dir = ALLOC_RIGHT; | |
924 | ||
925 | get_new_segment(sbi, &segno, new_sec, dir); | |
926 | curseg->next_segno = segno; | |
927 | reset_curseg(sbi, type, 1); | |
928 | curseg->alloc_type = LFS; | |
929 | } | |
930 | ||
931 | static void __next_free_blkoff(struct f2fs_sb_info *sbi, | |
932 | struct curseg_info *seg, block_t start) | |
933 | { | |
934 | struct seg_entry *se = get_seg_entry(sbi, seg->segno); | |
e81c93cf | 935 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); |
60a3b782 | 936 | unsigned long *target_map = SIT_I(sbi)->tmp_map; |
e81c93cf CL |
937 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; |
938 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; | |
939 | int i, pos; | |
940 | ||
941 | for (i = 0; i < entries; i++) | |
942 | target_map[i] = ckpt_map[i] | cur_map[i]; | |
943 | ||
944 | pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); | |
945 | ||
946 | seg->next_blkoff = pos; | |
351df4b2 JK |
947 | } |
948 | ||
0a8165d7 | 949 | /* |
351df4b2 JK |
950 | * If a segment is written by LFS manner, next block offset is just obtained |
951 | * by increasing the current block offset. However, if a segment is written by | |
952 | * SSR manner, next block offset obtained by calling __next_free_blkoff | |
953 | */ | |
954 | static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, | |
955 | struct curseg_info *seg) | |
956 | { | |
957 | if (seg->alloc_type == SSR) | |
958 | __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); | |
959 | else | |
960 | seg->next_blkoff++; | |
961 | } | |
962 | ||
0a8165d7 | 963 | /* |
e1c42045 | 964 | * This function always allocates a used segment(from dirty seglist) by SSR |
351df4b2 JK |
965 | * manner, so it should recover the existing segment information of valid blocks |
966 | */ | |
967 | static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) | |
968 | { | |
969 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
970 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
971 | unsigned int new_segno = curseg->next_segno; | |
972 | struct f2fs_summary_block *sum_node; | |
973 | struct page *sum_page; | |
974 | ||
975 | write_sum_page(sbi, curseg->sum_blk, | |
976 | GET_SUM_BLOCK(sbi, curseg->segno)); | |
977 | __set_test_and_inuse(sbi, new_segno); | |
978 | ||
979 | mutex_lock(&dirty_i->seglist_lock); | |
980 | __remove_dirty_segment(sbi, new_segno, PRE); | |
981 | __remove_dirty_segment(sbi, new_segno, DIRTY); | |
982 | mutex_unlock(&dirty_i->seglist_lock); | |
983 | ||
984 | reset_curseg(sbi, type, 1); | |
985 | curseg->alloc_type = SSR; | |
986 | __next_free_blkoff(sbi, curseg, 0); | |
987 | ||
988 | if (reuse) { | |
989 | sum_page = get_sum_page(sbi, new_segno); | |
990 | sum_node = (struct f2fs_summary_block *)page_address(sum_page); | |
991 | memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); | |
992 | f2fs_put_page(sum_page, 1); | |
993 | } | |
994 | } | |
995 | ||
43727527 JK |
996 | static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) |
997 | { | |
998 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
999 | const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; | |
1000 | ||
1001 | if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0)) | |
1002 | return v_ops->get_victim(sbi, | |
1003 | &(curseg)->next_segno, BG_GC, type, SSR); | |
1004 | ||
1005 | /* For data segments, let's do SSR more intensively */ | |
1006 | for (; type >= CURSEG_HOT_DATA; type--) | |
1007 | if (v_ops->get_victim(sbi, &(curseg)->next_segno, | |
1008 | BG_GC, type, SSR)) | |
1009 | return 1; | |
1010 | return 0; | |
1011 | } | |
1012 | ||
351df4b2 JK |
1013 | /* |
1014 | * flush out current segment and replace it with new segment | |
1015 | * This function should be returned with success, otherwise BUG | |
1016 | */ | |
1017 | static void allocate_segment_by_default(struct f2fs_sb_info *sbi, | |
1018 | int type, bool force) | |
1019 | { | |
1020 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
351df4b2 | 1021 | |
7b405275 | 1022 | if (force) |
351df4b2 | 1023 | new_curseg(sbi, type, true); |
7b405275 | 1024 | else if (type == CURSEG_WARM_NODE) |
351df4b2 | 1025 | new_curseg(sbi, type, false); |
60374688 JK |
1026 | else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) |
1027 | new_curseg(sbi, type, false); | |
351df4b2 JK |
1028 | else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) |
1029 | change_curseg(sbi, type, true); | |
1030 | else | |
1031 | new_curseg(sbi, type, false); | |
dcdfff65 JK |
1032 | |
1033 | stat_inc_seg_type(sbi, curseg); | |
351df4b2 JK |
1034 | } |
1035 | ||
38aa0889 JK |
1036 | static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type) |
1037 | { | |
1038 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1039 | unsigned int old_segno; | |
1040 | ||
1041 | old_segno = curseg->segno; | |
1042 | SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true); | |
1043 | locate_dirty_segment(sbi, old_segno); | |
1044 | } | |
1045 | ||
351df4b2 JK |
1046 | void allocate_new_segments(struct f2fs_sb_info *sbi) |
1047 | { | |
351df4b2 JK |
1048 | int i; |
1049 | ||
38aa0889 JK |
1050 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) |
1051 | __allocate_new_segments(sbi, i); | |
351df4b2 JK |
1052 | } |
1053 | ||
1054 | static const struct segment_allocation default_salloc_ops = { | |
1055 | .allocate_segment = allocate_segment_by_default, | |
1056 | }; | |
1057 | ||
4b2fecc8 JK |
1058 | int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) |
1059 | { | |
f7ef9b83 JK |
1060 | __u64 start = F2FS_BYTES_TO_BLK(range->start); |
1061 | __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; | |
4b2fecc8 JK |
1062 | unsigned int start_segno, end_segno; |
1063 | struct cp_control cpc; | |
1064 | ||
7cd8558b JK |
1065 | if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) || |
1066 | range->len < sbi->blocksize) | |
4b2fecc8 JK |
1067 | return -EINVAL; |
1068 | ||
9bd27ae4 | 1069 | cpc.trimmed = 0; |
7cd8558b | 1070 | if (end <= MAIN_BLKADDR(sbi)) |
4b2fecc8 JK |
1071 | goto out; |
1072 | ||
1073 | /* start/end segment number in main_area */ | |
7cd8558b JK |
1074 | start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); |
1075 | end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : | |
1076 | GET_SEGNO(sbi, end); | |
4b2fecc8 | 1077 | cpc.reason = CP_DISCARD; |
f7ef9b83 | 1078 | cpc.trim_minlen = F2FS_BYTES_TO_BLK(range->minlen); |
4b2fecc8 JK |
1079 | |
1080 | /* do checkpoint to issue discard commands safely */ | |
bba681cb JK |
1081 | for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) { |
1082 | cpc.trim_start = start_segno; | |
1083 | cpc.trim_end = min_t(unsigned int, rounddown(start_segno + | |
1084 | BATCHED_TRIM_SEGMENTS(sbi), | |
1085 | sbi->segs_per_sec) - 1, end_segno); | |
1086 | ||
1087 | mutex_lock(&sbi->gc_mutex); | |
1088 | write_checkpoint(sbi, &cpc); | |
1089 | mutex_unlock(&sbi->gc_mutex); | |
1090 | } | |
4b2fecc8 | 1091 | out: |
f7ef9b83 | 1092 | range->len = F2FS_BLK_TO_BYTES(cpc.trimmed); |
4b2fecc8 JK |
1093 | return 0; |
1094 | } | |
1095 | ||
351df4b2 JK |
1096 | static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) |
1097 | { | |
1098 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1099 | if (curseg->next_blkoff < sbi->blocks_per_seg) | |
1100 | return true; | |
1101 | return false; | |
1102 | } | |
1103 | ||
1104 | static int __get_segment_type_2(struct page *page, enum page_type p_type) | |
1105 | { | |
1106 | if (p_type == DATA) | |
1107 | return CURSEG_HOT_DATA; | |
1108 | else | |
1109 | return CURSEG_HOT_NODE; | |
1110 | } | |
1111 | ||
1112 | static int __get_segment_type_4(struct page *page, enum page_type p_type) | |
1113 | { | |
1114 | if (p_type == DATA) { | |
1115 | struct inode *inode = page->mapping->host; | |
1116 | ||
1117 | if (S_ISDIR(inode->i_mode)) | |
1118 | return CURSEG_HOT_DATA; | |
1119 | else | |
1120 | return CURSEG_COLD_DATA; | |
1121 | } else { | |
a344b9fd JK |
1122 | if (IS_DNODE(page) && is_cold_node(page)) |
1123 | return CURSEG_WARM_NODE; | |
351df4b2 JK |
1124 | else |
1125 | return CURSEG_COLD_NODE; | |
1126 | } | |
1127 | } | |
1128 | ||
1129 | static int __get_segment_type_6(struct page *page, enum page_type p_type) | |
1130 | { | |
1131 | if (p_type == DATA) { | |
1132 | struct inode *inode = page->mapping->host; | |
1133 | ||
1134 | if (S_ISDIR(inode->i_mode)) | |
1135 | return CURSEG_HOT_DATA; | |
354a3399 | 1136 | else if (is_cold_data(page) || file_is_cold(inode)) |
351df4b2 JK |
1137 | return CURSEG_COLD_DATA; |
1138 | else | |
1139 | return CURSEG_WARM_DATA; | |
1140 | } else { | |
1141 | if (IS_DNODE(page)) | |
1142 | return is_cold_node(page) ? CURSEG_WARM_NODE : | |
1143 | CURSEG_HOT_NODE; | |
1144 | else | |
1145 | return CURSEG_COLD_NODE; | |
1146 | } | |
1147 | } | |
1148 | ||
1149 | static int __get_segment_type(struct page *page, enum page_type p_type) | |
1150 | { | |
4081363f | 1151 | switch (F2FS_P_SB(page)->active_logs) { |
351df4b2 JK |
1152 | case 2: |
1153 | return __get_segment_type_2(page, p_type); | |
1154 | case 4: | |
1155 | return __get_segment_type_4(page, p_type); | |
351df4b2 | 1156 | } |
12a67146 | 1157 | /* NR_CURSEG_TYPE(6) logs by default */ |
9850cf4a JK |
1158 | f2fs_bug_on(F2FS_P_SB(page), |
1159 | F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE); | |
12a67146 | 1160 | return __get_segment_type_6(page, p_type); |
351df4b2 JK |
1161 | } |
1162 | ||
bfad7c2d JK |
1163 | void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, |
1164 | block_t old_blkaddr, block_t *new_blkaddr, | |
1165 | struct f2fs_summary *sum, int type) | |
351df4b2 JK |
1166 | { |
1167 | struct sit_info *sit_i = SIT_I(sbi); | |
1168 | struct curseg_info *curseg; | |
38aa0889 JK |
1169 | bool direct_io = (type == CURSEG_DIRECT_IO); |
1170 | ||
1171 | type = direct_io ? CURSEG_WARM_DATA : type; | |
351df4b2 | 1172 | |
351df4b2 JK |
1173 | curseg = CURSEG_I(sbi, type); |
1174 | ||
1175 | mutex_lock(&curseg->curseg_mutex); | |
21cb1d99 | 1176 | mutex_lock(&sit_i->sentry_lock); |
351df4b2 | 1177 | |
38aa0889 JK |
1178 | /* direct_io'ed data is aligned to the segment for better performance */ |
1179 | if (direct_io && curseg->next_blkoff) | |
1180 | __allocate_new_segments(sbi, type); | |
1181 | ||
351df4b2 | 1182 | *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); |
351df4b2 JK |
1183 | |
1184 | /* | |
1185 | * __add_sum_entry should be resided under the curseg_mutex | |
1186 | * because, this function updates a summary entry in the | |
1187 | * current summary block. | |
1188 | */ | |
e79efe3b | 1189 | __add_sum_entry(sbi, type, sum); |
351df4b2 | 1190 | |
351df4b2 | 1191 | __refresh_next_blkoff(sbi, curseg); |
dcdfff65 JK |
1192 | |
1193 | stat_inc_block_count(sbi, curseg); | |
351df4b2 | 1194 | |
5e443818 JK |
1195 | if (!__has_curseg_space(sbi, type)) |
1196 | sit_i->s_ops->allocate_segment(sbi, type, false); | |
351df4b2 JK |
1197 | /* |
1198 | * SIT information should be updated before segment allocation, | |
1199 | * since SSR needs latest valid block information. | |
1200 | */ | |
1201 | refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); | |
5e443818 | 1202 | |
351df4b2 JK |
1203 | mutex_unlock(&sit_i->sentry_lock); |
1204 | ||
bfad7c2d | 1205 | if (page && IS_NODESEG(type)) |
351df4b2 JK |
1206 | fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); |
1207 | ||
bfad7c2d JK |
1208 | mutex_unlock(&curseg->curseg_mutex); |
1209 | } | |
1210 | ||
05ca3632 | 1211 | static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) |
bfad7c2d | 1212 | { |
05ca3632 | 1213 | int type = __get_segment_type(fio->page, fio->type); |
bfad7c2d | 1214 | |
05ca3632 JK |
1215 | allocate_data_block(fio->sbi, fio->page, fio->blk_addr, |
1216 | &fio->blk_addr, sum, type); | |
bfad7c2d | 1217 | |
351df4b2 | 1218 | /* writeout dirty page into bdev */ |
05ca3632 | 1219 | f2fs_submit_page_mbio(fio); |
351df4b2 JK |
1220 | } |
1221 | ||
577e3495 | 1222 | void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) |
351df4b2 | 1223 | { |
458e6197 | 1224 | struct f2fs_io_info fio = { |
05ca3632 | 1225 | .sbi = sbi, |
458e6197 | 1226 | .type = META, |
cf04e8eb JK |
1227 | .rw = WRITE_SYNC | REQ_META | REQ_PRIO, |
1228 | .blk_addr = page->index, | |
05ca3632 | 1229 | .page = page, |
458e6197 JK |
1230 | }; |
1231 | ||
351df4b2 | 1232 | set_page_writeback(page); |
05ca3632 | 1233 | f2fs_submit_page_mbio(&fio); |
351df4b2 JK |
1234 | } |
1235 | ||
05ca3632 | 1236 | void write_node_page(unsigned int nid, struct f2fs_io_info *fio) |
351df4b2 JK |
1237 | { |
1238 | struct f2fs_summary sum; | |
05ca3632 | 1239 | |
351df4b2 | 1240 | set_summary(&sum, nid, 0, 0); |
05ca3632 | 1241 | do_write_page(&sum, fio); |
351df4b2 JK |
1242 | } |
1243 | ||
05ca3632 | 1244 | void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio) |
351df4b2 | 1245 | { |
05ca3632 | 1246 | struct f2fs_sb_info *sbi = fio->sbi; |
351df4b2 JK |
1247 | struct f2fs_summary sum; |
1248 | struct node_info ni; | |
1249 | ||
9850cf4a | 1250 | f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); |
351df4b2 JK |
1251 | get_node_info(sbi, dn->nid, &ni); |
1252 | set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); | |
05ca3632 | 1253 | do_write_page(&sum, fio); |
e1509cf2 | 1254 | dn->data_blkaddr = fio->blk_addr; |
351df4b2 JK |
1255 | } |
1256 | ||
05ca3632 | 1257 | void rewrite_data_page(struct f2fs_io_info *fio) |
351df4b2 | 1258 | { |
05ca3632 JK |
1259 | stat_inc_inplace_blocks(fio->sbi); |
1260 | f2fs_submit_page_mbio(fio); | |
351df4b2 JK |
1261 | } |
1262 | ||
1263 | void recover_data_page(struct f2fs_sb_info *sbi, | |
1264 | struct page *page, struct f2fs_summary *sum, | |
1265 | block_t old_blkaddr, block_t new_blkaddr) | |
1266 | { | |
1267 | struct sit_info *sit_i = SIT_I(sbi); | |
1268 | struct curseg_info *curseg; | |
1269 | unsigned int segno, old_cursegno; | |
1270 | struct seg_entry *se; | |
1271 | int type; | |
1272 | ||
1273 | segno = GET_SEGNO(sbi, new_blkaddr); | |
1274 | se = get_seg_entry(sbi, segno); | |
1275 | type = se->type; | |
1276 | ||
1277 | if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { | |
1278 | if (old_blkaddr == NULL_ADDR) | |
1279 | type = CURSEG_COLD_DATA; | |
1280 | else | |
1281 | type = CURSEG_WARM_DATA; | |
1282 | } | |
1283 | curseg = CURSEG_I(sbi, type); | |
1284 | ||
1285 | mutex_lock(&curseg->curseg_mutex); | |
1286 | mutex_lock(&sit_i->sentry_lock); | |
1287 | ||
1288 | old_cursegno = curseg->segno; | |
1289 | ||
1290 | /* change the current segment */ | |
1291 | if (segno != curseg->segno) { | |
1292 | curseg->next_segno = segno; | |
1293 | change_curseg(sbi, type, true); | |
1294 | } | |
1295 | ||
491c0854 | 1296 | curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); |
e79efe3b | 1297 | __add_sum_entry(sbi, type, sum); |
351df4b2 JK |
1298 | |
1299 | refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); | |
351df4b2 | 1300 | locate_dirty_segment(sbi, old_cursegno); |
351df4b2 JK |
1301 | |
1302 | mutex_unlock(&sit_i->sentry_lock); | |
1303 | mutex_unlock(&curseg->curseg_mutex); | |
1304 | } | |
1305 | ||
df0f8dc0 CY |
1306 | static inline bool is_merged_page(struct f2fs_sb_info *sbi, |
1307 | struct page *page, enum page_type type) | |
1308 | { | |
1309 | enum page_type btype = PAGE_TYPE_OF_BIO(type); | |
1310 | struct f2fs_bio_info *io = &sbi->write_io[btype]; | |
df0f8dc0 CY |
1311 | struct bio_vec *bvec; |
1312 | int i; | |
1313 | ||
1314 | down_read(&io->io_rwsem); | |
ce23447f | 1315 | if (!io->bio) |
df0f8dc0 CY |
1316 | goto out; |
1317 | ||
ce23447f | 1318 | bio_for_each_segment_all(bvec, io->bio, i) { |
df0f8dc0 CY |
1319 | if (page == bvec->bv_page) { |
1320 | up_read(&io->io_rwsem); | |
1321 | return true; | |
1322 | } | |
1323 | } | |
1324 | ||
1325 | out: | |
1326 | up_read(&io->io_rwsem); | |
1327 | return false; | |
1328 | } | |
1329 | ||
93dfe2ac | 1330 | void f2fs_wait_on_page_writeback(struct page *page, |
5514f0aa | 1331 | enum page_type type) |
93dfe2ac | 1332 | { |
93dfe2ac | 1333 | if (PageWriteback(page)) { |
4081363f JK |
1334 | struct f2fs_sb_info *sbi = F2FS_P_SB(page); |
1335 | ||
df0f8dc0 CY |
1336 | if (is_merged_page(sbi, page, type)) |
1337 | f2fs_submit_merged_bio(sbi, type, WRITE); | |
93dfe2ac JK |
1338 | wait_on_page_writeback(page); |
1339 | } | |
1340 | } | |
1341 | ||
351df4b2 JK |
1342 | static int read_compacted_summaries(struct f2fs_sb_info *sbi) |
1343 | { | |
1344 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1345 | struct curseg_info *seg_i; | |
1346 | unsigned char *kaddr; | |
1347 | struct page *page; | |
1348 | block_t start; | |
1349 | int i, j, offset; | |
1350 | ||
1351 | start = start_sum_block(sbi); | |
1352 | ||
1353 | page = get_meta_page(sbi, start++); | |
1354 | kaddr = (unsigned char *)page_address(page); | |
1355 | ||
1356 | /* Step 1: restore nat cache */ | |
1357 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
1358 | memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE); | |
1359 | ||
1360 | /* Step 2: restore sit cache */ | |
1361 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1362 | memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE, | |
1363 | SUM_JOURNAL_SIZE); | |
1364 | offset = 2 * SUM_JOURNAL_SIZE; | |
1365 | ||
1366 | /* Step 3: restore summary entries */ | |
1367 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1368 | unsigned short blk_off; | |
1369 | unsigned int segno; | |
1370 | ||
1371 | seg_i = CURSEG_I(sbi, i); | |
1372 | segno = le32_to_cpu(ckpt->cur_data_segno[i]); | |
1373 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); | |
1374 | seg_i->next_segno = segno; | |
1375 | reset_curseg(sbi, i, 0); | |
1376 | seg_i->alloc_type = ckpt->alloc_type[i]; | |
1377 | seg_i->next_blkoff = blk_off; | |
1378 | ||
1379 | if (seg_i->alloc_type == SSR) | |
1380 | blk_off = sbi->blocks_per_seg; | |
1381 | ||
1382 | for (j = 0; j < blk_off; j++) { | |
1383 | struct f2fs_summary *s; | |
1384 | s = (struct f2fs_summary *)(kaddr + offset); | |
1385 | seg_i->sum_blk->entries[j] = *s; | |
1386 | offset += SUMMARY_SIZE; | |
1387 | if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE - | |
1388 | SUM_FOOTER_SIZE) | |
1389 | continue; | |
1390 | ||
1391 | f2fs_put_page(page, 1); | |
1392 | page = NULL; | |
1393 | ||
1394 | page = get_meta_page(sbi, start++); | |
1395 | kaddr = (unsigned char *)page_address(page); | |
1396 | offset = 0; | |
1397 | } | |
1398 | } | |
1399 | f2fs_put_page(page, 1); | |
1400 | return 0; | |
1401 | } | |
1402 | ||
1403 | static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) | |
1404 | { | |
1405 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1406 | struct f2fs_summary_block *sum; | |
1407 | struct curseg_info *curseg; | |
1408 | struct page *new; | |
1409 | unsigned short blk_off; | |
1410 | unsigned int segno = 0; | |
1411 | block_t blk_addr = 0; | |
1412 | ||
1413 | /* get segment number and block addr */ | |
1414 | if (IS_DATASEG(type)) { | |
1415 | segno = le32_to_cpu(ckpt->cur_data_segno[type]); | |
1416 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - | |
1417 | CURSEG_HOT_DATA]); | |
119ee914 | 1418 | if (__exist_node_summaries(sbi)) |
351df4b2 JK |
1419 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); |
1420 | else | |
1421 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); | |
1422 | } else { | |
1423 | segno = le32_to_cpu(ckpt->cur_node_segno[type - | |
1424 | CURSEG_HOT_NODE]); | |
1425 | blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - | |
1426 | CURSEG_HOT_NODE]); | |
119ee914 | 1427 | if (__exist_node_summaries(sbi)) |
351df4b2 JK |
1428 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, |
1429 | type - CURSEG_HOT_NODE); | |
1430 | else | |
1431 | blk_addr = GET_SUM_BLOCK(sbi, segno); | |
1432 | } | |
1433 | ||
1434 | new = get_meta_page(sbi, blk_addr); | |
1435 | sum = (struct f2fs_summary_block *)page_address(new); | |
1436 | ||
1437 | if (IS_NODESEG(type)) { | |
119ee914 | 1438 | if (__exist_node_summaries(sbi)) { |
351df4b2 JK |
1439 | struct f2fs_summary *ns = &sum->entries[0]; |
1440 | int i; | |
1441 | for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { | |
1442 | ns->version = 0; | |
1443 | ns->ofs_in_node = 0; | |
1444 | } | |
1445 | } else { | |
d653788a GZ |
1446 | int err; |
1447 | ||
1448 | err = restore_node_summary(sbi, segno, sum); | |
1449 | if (err) { | |
351df4b2 | 1450 | f2fs_put_page(new, 1); |
d653788a | 1451 | return err; |
351df4b2 JK |
1452 | } |
1453 | } | |
1454 | } | |
1455 | ||
1456 | /* set uncompleted segment to curseg */ | |
1457 | curseg = CURSEG_I(sbi, type); | |
1458 | mutex_lock(&curseg->curseg_mutex); | |
1459 | memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE); | |
1460 | curseg->next_segno = segno; | |
1461 | reset_curseg(sbi, type, 0); | |
1462 | curseg->alloc_type = ckpt->alloc_type[type]; | |
1463 | curseg->next_blkoff = blk_off; | |
1464 | mutex_unlock(&curseg->curseg_mutex); | |
1465 | f2fs_put_page(new, 1); | |
1466 | return 0; | |
1467 | } | |
1468 | ||
1469 | static int restore_curseg_summaries(struct f2fs_sb_info *sbi) | |
1470 | { | |
1471 | int type = CURSEG_HOT_DATA; | |
e4fc5fbf | 1472 | int err; |
351df4b2 | 1473 | |
25ca923b | 1474 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) { |
3fa06d7b CY |
1475 | int npages = npages_for_summary_flush(sbi, true); |
1476 | ||
1477 | if (npages >= 2) | |
1478 | ra_meta_pages(sbi, start_sum_block(sbi), npages, | |
1479 | META_CP); | |
1480 | ||
351df4b2 JK |
1481 | /* restore for compacted data summary */ |
1482 | if (read_compacted_summaries(sbi)) | |
1483 | return -EINVAL; | |
1484 | type = CURSEG_HOT_NODE; | |
1485 | } | |
1486 | ||
119ee914 | 1487 | if (__exist_node_summaries(sbi)) |
3fa06d7b CY |
1488 | ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), |
1489 | NR_CURSEG_TYPE - type, META_CP); | |
1490 | ||
e4fc5fbf CY |
1491 | for (; type <= CURSEG_COLD_NODE; type++) { |
1492 | err = read_normal_summaries(sbi, type); | |
1493 | if (err) | |
1494 | return err; | |
1495 | } | |
1496 | ||
351df4b2 JK |
1497 | return 0; |
1498 | } | |
1499 | ||
1500 | static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) | |
1501 | { | |
1502 | struct page *page; | |
1503 | unsigned char *kaddr; | |
1504 | struct f2fs_summary *summary; | |
1505 | struct curseg_info *seg_i; | |
1506 | int written_size = 0; | |
1507 | int i, j; | |
1508 | ||
1509 | page = grab_meta_page(sbi, blkaddr++); | |
1510 | kaddr = (unsigned char *)page_address(page); | |
1511 | ||
1512 | /* Step 1: write nat cache */ | |
1513 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
1514 | memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE); | |
1515 | written_size += SUM_JOURNAL_SIZE; | |
1516 | ||
1517 | /* Step 2: write sit cache */ | |
1518 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1519 | memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits, | |
1520 | SUM_JOURNAL_SIZE); | |
1521 | written_size += SUM_JOURNAL_SIZE; | |
1522 | ||
351df4b2 JK |
1523 | /* Step 3: write summary entries */ |
1524 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1525 | unsigned short blkoff; | |
1526 | seg_i = CURSEG_I(sbi, i); | |
1527 | if (sbi->ckpt->alloc_type[i] == SSR) | |
1528 | blkoff = sbi->blocks_per_seg; | |
1529 | else | |
1530 | blkoff = curseg_blkoff(sbi, i); | |
1531 | ||
1532 | for (j = 0; j < blkoff; j++) { | |
1533 | if (!page) { | |
1534 | page = grab_meta_page(sbi, blkaddr++); | |
1535 | kaddr = (unsigned char *)page_address(page); | |
1536 | written_size = 0; | |
1537 | } | |
1538 | summary = (struct f2fs_summary *)(kaddr + written_size); | |
1539 | *summary = seg_i->sum_blk->entries[j]; | |
1540 | written_size += SUMMARY_SIZE; | |
351df4b2 JK |
1541 | |
1542 | if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE - | |
1543 | SUM_FOOTER_SIZE) | |
1544 | continue; | |
1545 | ||
e8d61a74 | 1546 | set_page_dirty(page); |
351df4b2 JK |
1547 | f2fs_put_page(page, 1); |
1548 | page = NULL; | |
1549 | } | |
1550 | } | |
e8d61a74 CY |
1551 | if (page) { |
1552 | set_page_dirty(page); | |
351df4b2 | 1553 | f2fs_put_page(page, 1); |
e8d61a74 | 1554 | } |
351df4b2 JK |
1555 | } |
1556 | ||
1557 | static void write_normal_summaries(struct f2fs_sb_info *sbi, | |
1558 | block_t blkaddr, int type) | |
1559 | { | |
1560 | int i, end; | |
1561 | if (IS_DATASEG(type)) | |
1562 | end = type + NR_CURSEG_DATA_TYPE; | |
1563 | else | |
1564 | end = type + NR_CURSEG_NODE_TYPE; | |
1565 | ||
1566 | for (i = type; i < end; i++) { | |
1567 | struct curseg_info *sum = CURSEG_I(sbi, i); | |
1568 | mutex_lock(&sum->curseg_mutex); | |
1569 | write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type)); | |
1570 | mutex_unlock(&sum->curseg_mutex); | |
1571 | } | |
1572 | } | |
1573 | ||
1574 | void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1575 | { | |
25ca923b | 1576 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) |
351df4b2 JK |
1577 | write_compacted_summaries(sbi, start_blk); |
1578 | else | |
1579 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); | |
1580 | } | |
1581 | ||
1582 | void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1583 | { | |
119ee914 | 1584 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); |
351df4b2 JK |
1585 | } |
1586 | ||
1587 | int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type, | |
1588 | unsigned int val, int alloc) | |
1589 | { | |
1590 | int i; | |
1591 | ||
1592 | if (type == NAT_JOURNAL) { | |
1593 | for (i = 0; i < nats_in_cursum(sum); i++) { | |
1594 | if (le32_to_cpu(nid_in_journal(sum, i)) == val) | |
1595 | return i; | |
1596 | } | |
1597 | if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) | |
1598 | return update_nats_in_cursum(sum, 1); | |
1599 | } else if (type == SIT_JOURNAL) { | |
1600 | for (i = 0; i < sits_in_cursum(sum); i++) | |
1601 | if (le32_to_cpu(segno_in_journal(sum, i)) == val) | |
1602 | return i; | |
1603 | if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES) | |
1604 | return update_sits_in_cursum(sum, 1); | |
1605 | } | |
1606 | return -1; | |
1607 | } | |
1608 | ||
1609 | static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, | |
1610 | unsigned int segno) | |
1611 | { | |
2cc22186 | 1612 | return get_meta_page(sbi, current_sit_addr(sbi, segno)); |
351df4b2 JK |
1613 | } |
1614 | ||
1615 | static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, | |
1616 | unsigned int start) | |
1617 | { | |
1618 | struct sit_info *sit_i = SIT_I(sbi); | |
1619 | struct page *src_page, *dst_page; | |
1620 | pgoff_t src_off, dst_off; | |
1621 | void *src_addr, *dst_addr; | |
1622 | ||
1623 | src_off = current_sit_addr(sbi, start); | |
1624 | dst_off = next_sit_addr(sbi, src_off); | |
1625 | ||
1626 | /* get current sit block page without lock */ | |
1627 | src_page = get_meta_page(sbi, src_off); | |
1628 | dst_page = grab_meta_page(sbi, dst_off); | |
9850cf4a | 1629 | f2fs_bug_on(sbi, PageDirty(src_page)); |
351df4b2 JK |
1630 | |
1631 | src_addr = page_address(src_page); | |
1632 | dst_addr = page_address(dst_page); | |
1633 | memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); | |
1634 | ||
1635 | set_page_dirty(dst_page); | |
1636 | f2fs_put_page(src_page, 1); | |
1637 | ||
1638 | set_to_next_sit(sit_i, start); | |
1639 | ||
1640 | return dst_page; | |
1641 | } | |
1642 | ||
184a5cd2 CY |
1643 | static struct sit_entry_set *grab_sit_entry_set(void) |
1644 | { | |
1645 | struct sit_entry_set *ses = | |
1646 | f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC); | |
1647 | ||
1648 | ses->entry_cnt = 0; | |
1649 | INIT_LIST_HEAD(&ses->set_list); | |
1650 | return ses; | |
1651 | } | |
1652 | ||
1653 | static void release_sit_entry_set(struct sit_entry_set *ses) | |
1654 | { | |
1655 | list_del(&ses->set_list); | |
1656 | kmem_cache_free(sit_entry_set_slab, ses); | |
1657 | } | |
1658 | ||
1659 | static void adjust_sit_entry_set(struct sit_entry_set *ses, | |
1660 | struct list_head *head) | |
1661 | { | |
1662 | struct sit_entry_set *next = ses; | |
1663 | ||
1664 | if (list_is_last(&ses->set_list, head)) | |
1665 | return; | |
1666 | ||
1667 | list_for_each_entry_continue(next, head, set_list) | |
1668 | if (ses->entry_cnt <= next->entry_cnt) | |
1669 | break; | |
1670 | ||
1671 | list_move_tail(&ses->set_list, &next->set_list); | |
1672 | } | |
1673 | ||
1674 | static void add_sit_entry(unsigned int segno, struct list_head *head) | |
1675 | { | |
1676 | struct sit_entry_set *ses; | |
1677 | unsigned int start_segno = START_SEGNO(segno); | |
1678 | ||
1679 | list_for_each_entry(ses, head, set_list) { | |
1680 | if (ses->start_segno == start_segno) { | |
1681 | ses->entry_cnt++; | |
1682 | adjust_sit_entry_set(ses, head); | |
1683 | return; | |
1684 | } | |
1685 | } | |
1686 | ||
1687 | ses = grab_sit_entry_set(); | |
1688 | ||
1689 | ses->start_segno = start_segno; | |
1690 | ses->entry_cnt++; | |
1691 | list_add(&ses->set_list, head); | |
1692 | } | |
1693 | ||
1694 | static void add_sits_in_set(struct f2fs_sb_info *sbi) | |
1695 | { | |
1696 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
1697 | struct list_head *set_list = &sm_info->sit_entry_set; | |
1698 | unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; | |
184a5cd2 CY |
1699 | unsigned int segno; |
1700 | ||
7cd8558b | 1701 | for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) |
184a5cd2 CY |
1702 | add_sit_entry(segno, set_list); |
1703 | } | |
1704 | ||
1705 | static void remove_sits_in_journal(struct f2fs_sb_info *sbi) | |
351df4b2 JK |
1706 | { |
1707 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1708 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
1709 | int i; | |
1710 | ||
184a5cd2 CY |
1711 | for (i = sits_in_cursum(sum) - 1; i >= 0; i--) { |
1712 | unsigned int segno; | |
1713 | bool dirtied; | |
1714 | ||
1715 | segno = le32_to_cpu(segno_in_journal(sum, i)); | |
1716 | dirtied = __mark_sit_entry_dirty(sbi, segno); | |
1717 | ||
1718 | if (!dirtied) | |
1719 | add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); | |
351df4b2 | 1720 | } |
184a5cd2 | 1721 | update_sits_in_cursum(sum, -sits_in_cursum(sum)); |
351df4b2 JK |
1722 | } |
1723 | ||
0a8165d7 | 1724 | /* |
351df4b2 JK |
1725 | * CP calls this function, which flushes SIT entries including sit_journal, |
1726 | * and moves prefree segs to free segs. | |
1727 | */ | |
4b2fecc8 | 1728 | void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) |
351df4b2 JK |
1729 | { |
1730 | struct sit_info *sit_i = SIT_I(sbi); | |
1731 | unsigned long *bitmap = sit_i->dirty_sentries_bitmap; | |
1732 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1733 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
184a5cd2 CY |
1734 | struct sit_entry_set *ses, *tmp; |
1735 | struct list_head *head = &SM_I(sbi)->sit_entry_set; | |
184a5cd2 | 1736 | bool to_journal = true; |
4b2fecc8 | 1737 | struct seg_entry *se; |
351df4b2 JK |
1738 | |
1739 | mutex_lock(&curseg->curseg_mutex); | |
1740 | mutex_lock(&sit_i->sentry_lock); | |
1741 | ||
2b11a74b WL |
1742 | if (!sit_i->dirty_sentries) |
1743 | goto out; | |
1744 | ||
351df4b2 | 1745 | /* |
184a5cd2 CY |
1746 | * add and account sit entries of dirty bitmap in sit entry |
1747 | * set temporarily | |
351df4b2 | 1748 | */ |
184a5cd2 | 1749 | add_sits_in_set(sbi); |
351df4b2 | 1750 | |
184a5cd2 CY |
1751 | /* |
1752 | * if there are no enough space in journal to store dirty sit | |
1753 | * entries, remove all entries from journal and add and account | |
1754 | * them in sit entry set. | |
1755 | */ | |
1756 | if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL)) | |
1757 | remove_sits_in_journal(sbi); | |
b2955550 | 1758 | |
184a5cd2 CY |
1759 | /* |
1760 | * there are two steps to flush sit entries: | |
1761 | * #1, flush sit entries to journal in current cold data summary block. | |
1762 | * #2, flush sit entries to sit page. | |
1763 | */ | |
1764 | list_for_each_entry_safe(ses, tmp, head, set_list) { | |
4a257ed6 | 1765 | struct page *page = NULL; |
184a5cd2 CY |
1766 | struct f2fs_sit_block *raw_sit = NULL; |
1767 | unsigned int start_segno = ses->start_segno; | |
1768 | unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, | |
7cd8558b | 1769 | (unsigned long)MAIN_SEGS(sbi)); |
184a5cd2 CY |
1770 | unsigned int segno = start_segno; |
1771 | ||
1772 | if (to_journal && | |
1773 | !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL)) | |
1774 | to_journal = false; | |
1775 | ||
1776 | if (!to_journal) { | |
1777 | page = get_next_sit_page(sbi, start_segno); | |
1778 | raw_sit = page_address(page); | |
351df4b2 | 1779 | } |
351df4b2 | 1780 | |
184a5cd2 CY |
1781 | /* flush dirty sit entries in region of current sit set */ |
1782 | for_each_set_bit_from(segno, bitmap, end) { | |
1783 | int offset, sit_offset; | |
4b2fecc8 JK |
1784 | |
1785 | se = get_seg_entry(sbi, segno); | |
184a5cd2 CY |
1786 | |
1787 | /* add discard candidates */ | |
d7bc2484 | 1788 | if (cpc->reason != CP_DISCARD) { |
4b2fecc8 JK |
1789 | cpc->trim_start = segno; |
1790 | add_discard_addrs(sbi, cpc); | |
1791 | } | |
184a5cd2 CY |
1792 | |
1793 | if (to_journal) { | |
1794 | offset = lookup_journal_in_cursum(sum, | |
1795 | SIT_JOURNAL, segno, 1); | |
1796 | f2fs_bug_on(sbi, offset < 0); | |
1797 | segno_in_journal(sum, offset) = | |
1798 | cpu_to_le32(segno); | |
1799 | seg_info_to_raw_sit(se, | |
1800 | &sit_in_journal(sum, offset)); | |
1801 | } else { | |
1802 | sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); | |
1803 | seg_info_to_raw_sit(se, | |
1804 | &raw_sit->entries[sit_offset]); | |
1805 | } | |
351df4b2 | 1806 | |
184a5cd2 CY |
1807 | __clear_bit(segno, bitmap); |
1808 | sit_i->dirty_sentries--; | |
1809 | ses->entry_cnt--; | |
351df4b2 JK |
1810 | } |
1811 | ||
184a5cd2 CY |
1812 | if (!to_journal) |
1813 | f2fs_put_page(page, 1); | |
1814 | ||
1815 | f2fs_bug_on(sbi, ses->entry_cnt); | |
1816 | release_sit_entry_set(ses); | |
351df4b2 | 1817 | } |
184a5cd2 CY |
1818 | |
1819 | f2fs_bug_on(sbi, !list_empty(head)); | |
1820 | f2fs_bug_on(sbi, sit_i->dirty_sentries); | |
184a5cd2 | 1821 | out: |
4b2fecc8 JK |
1822 | if (cpc->reason == CP_DISCARD) { |
1823 | for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) | |
1824 | add_discard_addrs(sbi, cpc); | |
1825 | } | |
351df4b2 JK |
1826 | mutex_unlock(&sit_i->sentry_lock); |
1827 | mutex_unlock(&curseg->curseg_mutex); | |
1828 | ||
351df4b2 JK |
1829 | set_prefree_as_free_segments(sbi); |
1830 | } | |
1831 | ||
1832 | static int build_sit_info(struct f2fs_sb_info *sbi) | |
1833 | { | |
1834 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
1835 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1836 | struct sit_info *sit_i; | |
1837 | unsigned int sit_segs, start; | |
1838 | char *src_bitmap, *dst_bitmap; | |
1839 | unsigned int bitmap_size; | |
1840 | ||
1841 | /* allocate memory for SIT information */ | |
1842 | sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); | |
1843 | if (!sit_i) | |
1844 | return -ENOMEM; | |
1845 | ||
1846 | SM_I(sbi)->sit_info = sit_i; | |
1847 | ||
7cd8558b | 1848 | sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry)); |
351df4b2 JK |
1849 | if (!sit_i->sentries) |
1850 | return -ENOMEM; | |
1851 | ||
7cd8558b | 1852 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
351df4b2 JK |
1853 | sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL); |
1854 | if (!sit_i->dirty_sentries_bitmap) | |
1855 | return -ENOMEM; | |
1856 | ||
7cd8558b | 1857 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
1858 | sit_i->sentries[start].cur_valid_map |
1859 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
1860 | sit_i->sentries[start].ckpt_valid_map | |
1861 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
1862 | if (!sit_i->sentries[start].cur_valid_map | |
1863 | || !sit_i->sentries[start].ckpt_valid_map) | |
1864 | return -ENOMEM; | |
1865 | } | |
1866 | ||
60a3b782 JK |
1867 | sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); |
1868 | if (!sit_i->tmp_map) | |
1869 | return -ENOMEM; | |
1870 | ||
351df4b2 | 1871 | if (sbi->segs_per_sec > 1) { |
7cd8558b | 1872 | sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) * |
351df4b2 JK |
1873 | sizeof(struct sec_entry)); |
1874 | if (!sit_i->sec_entries) | |
1875 | return -ENOMEM; | |
1876 | } | |
1877 | ||
1878 | /* get information related with SIT */ | |
1879 | sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; | |
1880 | ||
1881 | /* setup SIT bitmap from ckeckpoint pack */ | |
1882 | bitmap_size = __bitmap_size(sbi, SIT_BITMAP); | |
1883 | src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); | |
1884 | ||
79b5793b | 1885 | dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); |
351df4b2 JK |
1886 | if (!dst_bitmap) |
1887 | return -ENOMEM; | |
351df4b2 JK |
1888 | |
1889 | /* init SIT information */ | |
1890 | sit_i->s_ops = &default_salloc_ops; | |
1891 | ||
1892 | sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); | |
1893 | sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; | |
1894 | sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count); | |
1895 | sit_i->sit_bitmap = dst_bitmap; | |
1896 | sit_i->bitmap_size = bitmap_size; | |
1897 | sit_i->dirty_sentries = 0; | |
1898 | sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; | |
1899 | sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); | |
1900 | sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; | |
1901 | mutex_init(&sit_i->sentry_lock); | |
1902 | return 0; | |
1903 | } | |
1904 | ||
1905 | static int build_free_segmap(struct f2fs_sb_info *sbi) | |
1906 | { | |
351df4b2 JK |
1907 | struct free_segmap_info *free_i; |
1908 | unsigned int bitmap_size, sec_bitmap_size; | |
1909 | ||
1910 | /* allocate memory for free segmap information */ | |
1911 | free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); | |
1912 | if (!free_i) | |
1913 | return -ENOMEM; | |
1914 | ||
1915 | SM_I(sbi)->free_info = free_i; | |
1916 | ||
7cd8558b | 1917 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
351df4b2 JK |
1918 | free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL); |
1919 | if (!free_i->free_segmap) | |
1920 | return -ENOMEM; | |
1921 | ||
7cd8558b | 1922 | sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); |
351df4b2 JK |
1923 | free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL); |
1924 | if (!free_i->free_secmap) | |
1925 | return -ENOMEM; | |
1926 | ||
1927 | /* set all segments as dirty temporarily */ | |
1928 | memset(free_i->free_segmap, 0xff, bitmap_size); | |
1929 | memset(free_i->free_secmap, 0xff, sec_bitmap_size); | |
1930 | ||
1931 | /* init free segmap information */ | |
7cd8558b | 1932 | free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); |
351df4b2 JK |
1933 | free_i->free_segments = 0; |
1934 | free_i->free_sections = 0; | |
1a118ccf | 1935 | spin_lock_init(&free_i->segmap_lock); |
351df4b2 JK |
1936 | return 0; |
1937 | } | |
1938 | ||
1939 | static int build_curseg(struct f2fs_sb_info *sbi) | |
1940 | { | |
1042d60f | 1941 | struct curseg_info *array; |
351df4b2 JK |
1942 | int i; |
1943 | ||
b434babf | 1944 | array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL); |
351df4b2 JK |
1945 | if (!array) |
1946 | return -ENOMEM; | |
1947 | ||
1948 | SM_I(sbi)->curseg_array = array; | |
1949 | ||
1950 | for (i = 0; i < NR_CURSEG_TYPE; i++) { | |
1951 | mutex_init(&array[i].curseg_mutex); | |
1952 | array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL); | |
1953 | if (!array[i].sum_blk) | |
1954 | return -ENOMEM; | |
1955 | array[i].segno = NULL_SEGNO; | |
1956 | array[i].next_blkoff = 0; | |
1957 | } | |
1958 | return restore_curseg_summaries(sbi); | |
1959 | } | |
1960 | ||
1961 | static void build_sit_entries(struct f2fs_sb_info *sbi) | |
1962 | { | |
1963 | struct sit_info *sit_i = SIT_I(sbi); | |
1964 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
1965 | struct f2fs_summary_block *sum = curseg->sum_blk; | |
74de593a CY |
1966 | int sit_blk_cnt = SIT_BLK_CNT(sbi); |
1967 | unsigned int i, start, end; | |
1968 | unsigned int readed, start_blk = 0; | |
90a893c7 | 1969 | int nrpages = MAX_BIO_BLOCKS(sbi); |
351df4b2 | 1970 | |
74de593a | 1971 | do { |
662befda | 1972 | readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT); |
74de593a CY |
1973 | |
1974 | start = start_blk * sit_i->sents_per_block; | |
1975 | end = (start_blk + readed) * sit_i->sents_per_block; | |
1976 | ||
7cd8558b | 1977 | for (; start < end && start < MAIN_SEGS(sbi); start++) { |
74de593a CY |
1978 | struct seg_entry *se = &sit_i->sentries[start]; |
1979 | struct f2fs_sit_block *sit_blk; | |
1980 | struct f2fs_sit_entry sit; | |
1981 | struct page *page; | |
1982 | ||
1983 | mutex_lock(&curseg->curseg_mutex); | |
1984 | for (i = 0; i < sits_in_cursum(sum); i++) { | |
6c311ec6 CF |
1985 | if (le32_to_cpu(segno_in_journal(sum, i)) |
1986 | == start) { | |
74de593a CY |
1987 | sit = sit_in_journal(sum, i); |
1988 | mutex_unlock(&curseg->curseg_mutex); | |
1989 | goto got_it; | |
1990 | } | |
351df4b2 | 1991 | } |
74de593a CY |
1992 | mutex_unlock(&curseg->curseg_mutex); |
1993 | ||
1994 | page = get_current_sit_page(sbi, start); | |
1995 | sit_blk = (struct f2fs_sit_block *)page_address(page); | |
1996 | sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; | |
1997 | f2fs_put_page(page, 1); | |
351df4b2 | 1998 | got_it: |
74de593a CY |
1999 | check_block_count(sbi, start, &sit); |
2000 | seg_info_from_raw_sit(se, &sit); | |
2001 | if (sbi->segs_per_sec > 1) { | |
2002 | struct sec_entry *e = get_sec_entry(sbi, start); | |
2003 | e->valid_blocks += se->valid_blocks; | |
2004 | } | |
351df4b2 | 2005 | } |
74de593a CY |
2006 | start_blk += readed; |
2007 | } while (start_blk < sit_blk_cnt); | |
351df4b2 JK |
2008 | } |
2009 | ||
2010 | static void init_free_segmap(struct f2fs_sb_info *sbi) | |
2011 | { | |
2012 | unsigned int start; | |
2013 | int type; | |
2014 | ||
7cd8558b | 2015 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
2016 | struct seg_entry *sentry = get_seg_entry(sbi, start); |
2017 | if (!sentry->valid_blocks) | |
2018 | __set_free(sbi, start); | |
2019 | } | |
2020 | ||
2021 | /* set use the current segments */ | |
2022 | for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { | |
2023 | struct curseg_info *curseg_t = CURSEG_I(sbi, type); | |
2024 | __set_test_and_inuse(sbi, curseg_t->segno); | |
2025 | } | |
2026 | } | |
2027 | ||
2028 | static void init_dirty_segmap(struct f2fs_sb_info *sbi) | |
2029 | { | |
2030 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
2031 | struct free_segmap_info *free_i = FREE_I(sbi); | |
7cd8558b | 2032 | unsigned int segno = 0, offset = 0; |
351df4b2 JK |
2033 | unsigned short valid_blocks; |
2034 | ||
8736fbf0 | 2035 | while (1) { |
351df4b2 | 2036 | /* find dirty segment based on free segmap */ |
7cd8558b JK |
2037 | segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); |
2038 | if (segno >= MAIN_SEGS(sbi)) | |
351df4b2 JK |
2039 | break; |
2040 | offset = segno + 1; | |
2041 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
ec325b52 | 2042 | if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) |
351df4b2 | 2043 | continue; |
ec325b52 JK |
2044 | if (valid_blocks > sbi->blocks_per_seg) { |
2045 | f2fs_bug_on(sbi, 1); | |
2046 | continue; | |
2047 | } | |
351df4b2 JK |
2048 | mutex_lock(&dirty_i->seglist_lock); |
2049 | __locate_dirty_segment(sbi, segno, DIRTY); | |
2050 | mutex_unlock(&dirty_i->seglist_lock); | |
2051 | } | |
2052 | } | |
2053 | ||
5ec4e49f | 2054 | static int init_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
2055 | { |
2056 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
7cd8558b | 2057 | unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); |
351df4b2 | 2058 | |
5ec4e49f JK |
2059 | dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL); |
2060 | if (!dirty_i->victim_secmap) | |
351df4b2 JK |
2061 | return -ENOMEM; |
2062 | return 0; | |
2063 | } | |
2064 | ||
2065 | static int build_dirty_segmap(struct f2fs_sb_info *sbi) | |
2066 | { | |
2067 | struct dirty_seglist_info *dirty_i; | |
2068 | unsigned int bitmap_size, i; | |
2069 | ||
2070 | /* allocate memory for dirty segments list information */ | |
2071 | dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); | |
2072 | if (!dirty_i) | |
2073 | return -ENOMEM; | |
2074 | ||
2075 | SM_I(sbi)->dirty_info = dirty_i; | |
2076 | mutex_init(&dirty_i->seglist_lock); | |
2077 | ||
7cd8558b | 2078 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
351df4b2 JK |
2079 | |
2080 | for (i = 0; i < NR_DIRTY_TYPE; i++) { | |
2081 | dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL); | |
351df4b2 JK |
2082 | if (!dirty_i->dirty_segmap[i]) |
2083 | return -ENOMEM; | |
2084 | } | |
2085 | ||
2086 | init_dirty_segmap(sbi); | |
5ec4e49f | 2087 | return init_victim_secmap(sbi); |
351df4b2 JK |
2088 | } |
2089 | ||
0a8165d7 | 2090 | /* |
351df4b2 JK |
2091 | * Update min, max modified time for cost-benefit GC algorithm |
2092 | */ | |
2093 | static void init_min_max_mtime(struct f2fs_sb_info *sbi) | |
2094 | { | |
2095 | struct sit_info *sit_i = SIT_I(sbi); | |
2096 | unsigned int segno; | |
2097 | ||
2098 | mutex_lock(&sit_i->sentry_lock); | |
2099 | ||
2100 | sit_i->min_mtime = LLONG_MAX; | |
2101 | ||
7cd8558b | 2102 | for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { |
351df4b2 JK |
2103 | unsigned int i; |
2104 | unsigned long long mtime = 0; | |
2105 | ||
2106 | for (i = 0; i < sbi->segs_per_sec; i++) | |
2107 | mtime += get_seg_entry(sbi, segno + i)->mtime; | |
2108 | ||
2109 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
2110 | ||
2111 | if (sit_i->min_mtime > mtime) | |
2112 | sit_i->min_mtime = mtime; | |
2113 | } | |
2114 | sit_i->max_mtime = get_mtime(sbi); | |
2115 | mutex_unlock(&sit_i->sentry_lock); | |
2116 | } | |
2117 | ||
2118 | int build_segment_manager(struct f2fs_sb_info *sbi) | |
2119 | { | |
2120 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
2121 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1042d60f | 2122 | struct f2fs_sm_info *sm_info; |
351df4b2 JK |
2123 | int err; |
2124 | ||
2125 | sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); | |
2126 | if (!sm_info) | |
2127 | return -ENOMEM; | |
2128 | ||
2129 | /* init sm info */ | |
2130 | sbi->sm_info = sm_info; | |
351df4b2 JK |
2131 | sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); |
2132 | sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); | |
2133 | sm_info->segment_count = le32_to_cpu(raw_super->segment_count); | |
2134 | sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); | |
2135 | sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); | |
2136 | sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); | |
2137 | sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); | |
58c41035 JK |
2138 | sm_info->rec_prefree_segments = sm_info->main_segments * |
2139 | DEF_RECLAIM_PREFREE_SEGMENTS / 100; | |
9b5f136f | 2140 | sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; |
216fbd64 | 2141 | sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; |
c1ce1b02 | 2142 | sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; |
351df4b2 | 2143 | |
7fd9e544 JK |
2144 | INIT_LIST_HEAD(&sm_info->discard_list); |
2145 | sm_info->nr_discards = 0; | |
2146 | sm_info->max_discards = 0; | |
2147 | ||
bba681cb JK |
2148 | sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS; |
2149 | ||
184a5cd2 CY |
2150 | INIT_LIST_HEAD(&sm_info->sit_entry_set); |
2151 | ||
b270ad6f | 2152 | if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) { |
2163d198 GZ |
2153 | err = create_flush_cmd_control(sbi); |
2154 | if (err) | |
a688b9d9 | 2155 | return err; |
6b4afdd7 JK |
2156 | } |
2157 | ||
351df4b2 JK |
2158 | err = build_sit_info(sbi); |
2159 | if (err) | |
2160 | return err; | |
2161 | err = build_free_segmap(sbi); | |
2162 | if (err) | |
2163 | return err; | |
2164 | err = build_curseg(sbi); | |
2165 | if (err) | |
2166 | return err; | |
2167 | ||
2168 | /* reinit free segmap based on SIT */ | |
2169 | build_sit_entries(sbi); | |
2170 | ||
2171 | init_free_segmap(sbi); | |
2172 | err = build_dirty_segmap(sbi); | |
2173 | if (err) | |
2174 | return err; | |
2175 | ||
2176 | init_min_max_mtime(sbi); | |
2177 | return 0; | |
2178 | } | |
2179 | ||
2180 | static void discard_dirty_segmap(struct f2fs_sb_info *sbi, | |
2181 | enum dirty_type dirty_type) | |
2182 | { | |
2183 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
2184 | ||
2185 | mutex_lock(&dirty_i->seglist_lock); | |
2186 | kfree(dirty_i->dirty_segmap[dirty_type]); | |
2187 | dirty_i->nr_dirty[dirty_type] = 0; | |
2188 | mutex_unlock(&dirty_i->seglist_lock); | |
2189 | } | |
2190 | ||
5ec4e49f | 2191 | static void destroy_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
2192 | { |
2193 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 2194 | kfree(dirty_i->victim_secmap); |
351df4b2 JK |
2195 | } |
2196 | ||
2197 | static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) | |
2198 | { | |
2199 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
2200 | int i; | |
2201 | ||
2202 | if (!dirty_i) | |
2203 | return; | |
2204 | ||
2205 | /* discard pre-free/dirty segments list */ | |
2206 | for (i = 0; i < NR_DIRTY_TYPE; i++) | |
2207 | discard_dirty_segmap(sbi, i); | |
2208 | ||
5ec4e49f | 2209 | destroy_victim_secmap(sbi); |
351df4b2 JK |
2210 | SM_I(sbi)->dirty_info = NULL; |
2211 | kfree(dirty_i); | |
2212 | } | |
2213 | ||
2214 | static void destroy_curseg(struct f2fs_sb_info *sbi) | |
2215 | { | |
2216 | struct curseg_info *array = SM_I(sbi)->curseg_array; | |
2217 | int i; | |
2218 | ||
2219 | if (!array) | |
2220 | return; | |
2221 | SM_I(sbi)->curseg_array = NULL; | |
2222 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
2223 | kfree(array[i].sum_blk); | |
2224 | kfree(array); | |
2225 | } | |
2226 | ||
2227 | static void destroy_free_segmap(struct f2fs_sb_info *sbi) | |
2228 | { | |
2229 | struct free_segmap_info *free_i = SM_I(sbi)->free_info; | |
2230 | if (!free_i) | |
2231 | return; | |
2232 | SM_I(sbi)->free_info = NULL; | |
2233 | kfree(free_i->free_segmap); | |
2234 | kfree(free_i->free_secmap); | |
2235 | kfree(free_i); | |
2236 | } | |
2237 | ||
2238 | static void destroy_sit_info(struct f2fs_sb_info *sbi) | |
2239 | { | |
2240 | struct sit_info *sit_i = SIT_I(sbi); | |
2241 | unsigned int start; | |
2242 | ||
2243 | if (!sit_i) | |
2244 | return; | |
2245 | ||
2246 | if (sit_i->sentries) { | |
7cd8558b | 2247 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
2248 | kfree(sit_i->sentries[start].cur_valid_map); |
2249 | kfree(sit_i->sentries[start].ckpt_valid_map); | |
2250 | } | |
2251 | } | |
60a3b782 JK |
2252 | kfree(sit_i->tmp_map); |
2253 | ||
351df4b2 JK |
2254 | vfree(sit_i->sentries); |
2255 | vfree(sit_i->sec_entries); | |
2256 | kfree(sit_i->dirty_sentries_bitmap); | |
2257 | ||
2258 | SM_I(sbi)->sit_info = NULL; | |
2259 | kfree(sit_i->sit_bitmap); | |
2260 | kfree(sit_i); | |
2261 | } | |
2262 | ||
2263 | void destroy_segment_manager(struct f2fs_sb_info *sbi) | |
2264 | { | |
2265 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
a688b9d9 | 2266 | |
3b03f724 CY |
2267 | if (!sm_info) |
2268 | return; | |
2163d198 | 2269 | destroy_flush_cmd_control(sbi); |
351df4b2 JK |
2270 | destroy_dirty_segmap(sbi); |
2271 | destroy_curseg(sbi); | |
2272 | destroy_free_segmap(sbi); | |
2273 | destroy_sit_info(sbi); | |
2274 | sbi->sm_info = NULL; | |
2275 | kfree(sm_info); | |
2276 | } | |
7fd9e544 JK |
2277 | |
2278 | int __init create_segment_manager_caches(void) | |
2279 | { | |
2280 | discard_entry_slab = f2fs_kmem_cache_create("discard_entry", | |
e8512d2e | 2281 | sizeof(struct discard_entry)); |
7fd9e544 | 2282 | if (!discard_entry_slab) |
184a5cd2 CY |
2283 | goto fail; |
2284 | ||
2285 | sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", | |
c9ee0085 | 2286 | sizeof(struct sit_entry_set)); |
184a5cd2 CY |
2287 | if (!sit_entry_set_slab) |
2288 | goto destory_discard_entry; | |
88b88a66 JK |
2289 | |
2290 | inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", | |
2291 | sizeof(struct inmem_pages)); | |
2292 | if (!inmem_entry_slab) | |
2293 | goto destroy_sit_entry_set; | |
7fd9e544 | 2294 | return 0; |
184a5cd2 | 2295 | |
88b88a66 JK |
2296 | destroy_sit_entry_set: |
2297 | kmem_cache_destroy(sit_entry_set_slab); | |
184a5cd2 CY |
2298 | destory_discard_entry: |
2299 | kmem_cache_destroy(discard_entry_slab); | |
2300 | fail: | |
2301 | return -ENOMEM; | |
7fd9e544 JK |
2302 | } |
2303 | ||
2304 | void destroy_segment_manager_caches(void) | |
2305 | { | |
184a5cd2 | 2306 | kmem_cache_destroy(sit_entry_set_slab); |
7fd9e544 | 2307 | kmem_cache_destroy(discard_entry_slab); |
88b88a66 | 2308 | kmem_cache_destroy(inmem_entry_slab); |
7fd9e544 | 2309 | } |