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> |
74de593a | 17 | #include <linux/swap.h> |
60b99b48 | 18 | #include <linux/timer.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 | |
f96999c3 JK |
32 | static unsigned long __reverse_ulong(unsigned char *str) |
33 | { | |
34 | unsigned long tmp = 0; | |
35 | int shift = 24, idx = 0; | |
36 | ||
37 | #if BITS_PER_LONG == 64 | |
38 | shift = 56; | |
39 | #endif | |
40 | while (shift >= 0) { | |
41 | tmp |= (unsigned long)str[idx++] << shift; | |
42 | shift -= BITS_PER_BYTE; | |
43 | } | |
44 | return tmp; | |
45 | } | |
46 | ||
9a7f143a CL |
47 | /* |
48 | * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since | |
49 | * MSB and LSB are reversed in a byte by f2fs_set_bit. | |
50 | */ | |
51 | static inline unsigned long __reverse_ffs(unsigned long word) | |
52 | { | |
53 | int num = 0; | |
54 | ||
55 | #if BITS_PER_LONG == 64 | |
f96999c3 | 56 | if ((word & 0xffffffff00000000UL) == 0) |
9a7f143a | 57 | num += 32; |
f96999c3 | 58 | else |
9a7f143a | 59 | word >>= 32; |
9a7f143a | 60 | #endif |
f96999c3 | 61 | if ((word & 0xffff0000) == 0) |
9a7f143a | 62 | num += 16; |
f96999c3 | 63 | else |
9a7f143a | 64 | word >>= 16; |
f96999c3 JK |
65 | |
66 | if ((word & 0xff00) == 0) | |
9a7f143a | 67 | num += 8; |
f96999c3 | 68 | else |
9a7f143a | 69 | word >>= 8; |
f96999c3 | 70 | |
9a7f143a CL |
71 | if ((word & 0xf0) == 0) |
72 | num += 4; | |
73 | else | |
74 | word >>= 4; | |
f96999c3 | 75 | |
9a7f143a CL |
76 | if ((word & 0xc) == 0) |
77 | num += 2; | |
78 | else | |
79 | word >>= 2; | |
f96999c3 | 80 | |
9a7f143a CL |
81 | if ((word & 0x2) == 0) |
82 | num += 1; | |
83 | return num; | |
84 | } | |
85 | ||
86 | /* | |
e1c42045 | 87 | * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because |
9a7f143a | 88 | * f2fs_set_bit makes MSB and LSB reversed in a byte. |
692223d1 | 89 | * @size must be integral times of unsigned long. |
9a7f143a | 90 | * Example: |
f96999c3 JK |
91 | * MSB <--> LSB |
92 | * f2fs_set_bit(0, bitmap) => 1000 0000 | |
93 | * f2fs_set_bit(7, bitmap) => 0000 0001 | |
9a7f143a CL |
94 | */ |
95 | static unsigned long __find_rev_next_bit(const unsigned long *addr, | |
96 | unsigned long size, unsigned long offset) | |
97 | { | |
98 | const unsigned long *p = addr + BIT_WORD(offset); | |
692223d1 | 99 | unsigned long result = size; |
9a7f143a | 100 | unsigned long tmp; |
9a7f143a CL |
101 | |
102 | if (offset >= size) | |
103 | return size; | |
104 | ||
692223d1 | 105 | size -= (offset & ~(BITS_PER_LONG - 1)); |
9a7f143a | 106 | offset %= BITS_PER_LONG; |
f96999c3 | 107 | |
692223d1 FL |
108 | while (1) { |
109 | if (*p == 0) | |
110 | goto pass; | |
9a7f143a | 111 | |
f96999c3 | 112 | tmp = __reverse_ulong((unsigned char *)p); |
692223d1 FL |
113 | |
114 | tmp &= ~0UL >> offset; | |
115 | if (size < BITS_PER_LONG) | |
116 | tmp &= (~0UL << (BITS_PER_LONG - size)); | |
9a7f143a | 117 | if (tmp) |
692223d1 FL |
118 | goto found; |
119 | pass: | |
120 | if (size <= BITS_PER_LONG) | |
121 | break; | |
9a7f143a | 122 | size -= BITS_PER_LONG; |
692223d1 | 123 | offset = 0; |
f96999c3 | 124 | p++; |
9a7f143a | 125 | } |
692223d1 FL |
126 | return result; |
127 | found: | |
128 | return result - size + __reverse_ffs(tmp); | |
9a7f143a CL |
129 | } |
130 | ||
131 | static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, | |
132 | unsigned long size, unsigned long offset) | |
133 | { | |
134 | const unsigned long *p = addr + BIT_WORD(offset); | |
80609448 | 135 | unsigned long result = size; |
9a7f143a | 136 | unsigned long tmp; |
9a7f143a CL |
137 | |
138 | if (offset >= size) | |
139 | return size; | |
140 | ||
80609448 | 141 | size -= (offset & ~(BITS_PER_LONG - 1)); |
9a7f143a | 142 | offset %= BITS_PER_LONG; |
80609448 JK |
143 | |
144 | while (1) { | |
145 | if (*p == ~0UL) | |
146 | goto pass; | |
147 | ||
f96999c3 | 148 | tmp = __reverse_ulong((unsigned char *)p); |
80609448 JK |
149 | |
150 | if (offset) | |
151 | tmp |= ~0UL << (BITS_PER_LONG - offset); | |
152 | if (size < BITS_PER_LONG) | |
153 | tmp |= ~0UL >> size; | |
f96999c3 | 154 | if (tmp != ~0UL) |
80609448 JK |
155 | goto found; |
156 | pass: | |
157 | if (size <= BITS_PER_LONG) | |
158 | break; | |
9a7f143a | 159 | size -= BITS_PER_LONG; |
80609448 | 160 | offset = 0; |
f96999c3 | 161 | p++; |
9a7f143a | 162 | } |
80609448 JK |
163 | return result; |
164 | found: | |
165 | return result - size + __reverse_ffz(tmp); | |
9a7f143a CL |
166 | } |
167 | ||
88b88a66 JK |
168 | void register_inmem_page(struct inode *inode, struct page *page) |
169 | { | |
170 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
171 | struct inmem_pages *new; | |
9be32d72 | 172 | |
9e4ded3f | 173 | f2fs_trace_pid(page); |
0722b101 | 174 | |
decd36b6 CY |
175 | set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE); |
176 | SetPagePrivate(page); | |
177 | ||
88b88a66 JK |
178 | new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); |
179 | ||
180 | /* add atomic page indices to the list */ | |
181 | new->page = page; | |
182 | INIT_LIST_HEAD(&new->list); | |
decd36b6 | 183 | |
88b88a66 JK |
184 | /* increase reference count with clean state */ |
185 | mutex_lock(&fi->inmem_lock); | |
186 | get_page(page); | |
187 | list_add_tail(&new->list, &fi->inmem_pages); | |
8dcf2ff7 | 188 | inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); |
88b88a66 | 189 | mutex_unlock(&fi->inmem_lock); |
8ce67cb0 JK |
190 | |
191 | trace_f2fs_register_inmem_page(page, INMEM); | |
88b88a66 JK |
192 | } |
193 | ||
28bc106b CY |
194 | static int __revoke_inmem_pages(struct inode *inode, |
195 | struct list_head *head, bool drop, bool recover) | |
29b96b54 | 196 | { |
28bc106b | 197 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
29b96b54 | 198 | struct inmem_pages *cur, *tmp; |
28bc106b | 199 | int err = 0; |
29b96b54 CY |
200 | |
201 | list_for_each_entry_safe(cur, tmp, head, list) { | |
28bc106b CY |
202 | struct page *page = cur->page; |
203 | ||
204 | if (drop) | |
205 | trace_f2fs_commit_inmem_page(page, INMEM_DROP); | |
206 | ||
207 | lock_page(page); | |
29b96b54 | 208 | |
28bc106b CY |
209 | if (recover) { |
210 | struct dnode_of_data dn; | |
211 | struct node_info ni; | |
212 | ||
213 | trace_f2fs_commit_inmem_page(page, INMEM_REVOKE); | |
214 | ||
215 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
216 | if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) { | |
217 | err = -EAGAIN; | |
218 | goto next; | |
219 | } | |
220 | get_node_info(sbi, dn.nid, &ni); | |
221 | f2fs_replace_block(sbi, &dn, dn.data_blkaddr, | |
222 | cur->old_addr, ni.version, true, true); | |
223 | f2fs_put_dnode(&dn); | |
224 | } | |
225 | next: | |
63c52d78 JK |
226 | /* we don't need to invalidate this in the sccessful status */ |
227 | if (drop || recover) | |
228 | ClearPageUptodate(page); | |
28bc106b | 229 | set_page_private(page, 0); |
28bc106b | 230 | f2fs_put_page(page, 1); |
29b96b54 CY |
231 | |
232 | list_del(&cur->list); | |
233 | kmem_cache_free(inmem_entry_slab, cur); | |
234 | dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); | |
235 | } | |
28bc106b | 236 | return err; |
29b96b54 CY |
237 | } |
238 | ||
239 | void drop_inmem_pages(struct inode *inode) | |
240 | { | |
241 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
242 | ||
26dc3d44 JK |
243 | clear_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE); |
244 | ||
29b96b54 | 245 | mutex_lock(&fi->inmem_lock); |
28bc106b | 246 | __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); |
29b96b54 CY |
247 | mutex_unlock(&fi->inmem_lock); |
248 | } | |
249 | ||
28bc106b CY |
250 | static int __commit_inmem_pages(struct inode *inode, |
251 | struct list_head *revoke_list) | |
88b88a66 JK |
252 | { |
253 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
254 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
255 | struct inmem_pages *cur, *tmp; | |
88b88a66 | 256 | struct f2fs_io_info fio = { |
05ca3632 | 257 | .sbi = sbi, |
88b88a66 | 258 | .type = DATA, |
1e84371f | 259 | .rw = WRITE_SYNC | REQ_PRIO, |
4375a336 | 260 | .encrypted_page = NULL, |
88b88a66 | 261 | }; |
29b96b54 | 262 | bool submit_bio = false; |
edb27dee | 263 | int err = 0; |
88b88a66 | 264 | |
88b88a66 | 265 | list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { |
28bc106b CY |
266 | struct page *page = cur->page; |
267 | ||
268 | lock_page(page); | |
269 | if (page->mapping == inode->i_mapping) { | |
270 | trace_f2fs_commit_inmem_page(page, INMEM); | |
271 | ||
272 | set_page_dirty(page); | |
273 | f2fs_wait_on_page_writeback(page, DATA, true); | |
274 | if (clear_page_dirty_for_io(page)) | |
29b96b54 | 275 | inode_dec_dirty_pages(inode); |
28bc106b CY |
276 | |
277 | fio.page = page; | |
29b96b54 CY |
278 | err = do_write_data_page(&fio); |
279 | if (err) { | |
28bc106b | 280 | unlock_page(page); |
29b96b54 | 281 | break; |
70c640b1 | 282 | } |
29b96b54 | 283 | |
28bc106b CY |
284 | /* record old blkaddr for revoking */ |
285 | cur->old_addr = fio.old_blkaddr; | |
decd36b6 | 286 | |
28bc106b CY |
287 | clear_cold_data(page); |
288 | submit_bio = true; | |
289 | } | |
290 | unlock_page(page); | |
291 | list_move_tail(&cur->list, revoke_list); | |
88b88a66 | 292 | } |
29b96b54 CY |
293 | |
294 | if (submit_bio) | |
295 | f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE); | |
28bc106b CY |
296 | |
297 | if (!err) | |
298 | __revoke_inmem_pages(inode, revoke_list, false, false); | |
299 | ||
29b96b54 CY |
300 | return err; |
301 | } | |
302 | ||
303 | int commit_inmem_pages(struct inode *inode) | |
304 | { | |
305 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
306 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
28bc106b CY |
307 | struct list_head revoke_list; |
308 | int err; | |
29b96b54 | 309 | |
28bc106b | 310 | INIT_LIST_HEAD(&revoke_list); |
29b96b54 CY |
311 | f2fs_balance_fs(sbi, true); |
312 | f2fs_lock_op(sbi); | |
313 | ||
314 | mutex_lock(&fi->inmem_lock); | |
28bc106b CY |
315 | err = __commit_inmem_pages(inode, &revoke_list); |
316 | if (err) { | |
317 | int ret; | |
318 | /* | |
319 | * try to revoke all committed pages, but still we could fail | |
320 | * due to no memory or other reason, if that happened, EAGAIN | |
321 | * will be returned, which means in such case, transaction is | |
322 | * already not integrity, caller should use journal to do the | |
323 | * recovery or rewrite & commit last transaction. For other | |
324 | * error number, revoking was done by filesystem itself. | |
325 | */ | |
326 | ret = __revoke_inmem_pages(inode, &revoke_list, false, true); | |
327 | if (ret) | |
328 | err = ret; | |
329 | ||
330 | /* drop all uncommitted pages */ | |
331 | __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); | |
332 | } | |
88b88a66 JK |
333 | mutex_unlock(&fi->inmem_lock); |
334 | ||
29b96b54 | 335 | f2fs_unlock_op(sbi); |
edb27dee | 336 | return err; |
88b88a66 JK |
337 | } |
338 | ||
0a8165d7 | 339 | /* |
351df4b2 JK |
340 | * This function balances dirty node and dentry pages. |
341 | * In addition, it controls garbage collection. | |
342 | */ | |
2c4db1a6 | 343 | void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) |
351df4b2 | 344 | { |
2c4db1a6 JK |
345 | if (!need) |
346 | return; | |
351df4b2 | 347 | /* |
029cd28c JK |
348 | * We should do GC or end up with checkpoint, if there are so many dirty |
349 | * dir/node pages without enough free segments. | |
351df4b2 | 350 | */ |
43727527 | 351 | if (has_not_enough_free_secs(sbi, 0)) { |
351df4b2 | 352 | mutex_lock(&sbi->gc_mutex); |
d530d4d8 | 353 | f2fs_gc(sbi, false); |
351df4b2 JK |
354 | } |
355 | } | |
356 | ||
4660f9c0 JK |
357 | void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) |
358 | { | |
1dcc336b | 359 | /* try to shrink extent cache when there is no enough memory */ |
554df79e JK |
360 | if (!available_free_memory(sbi, EXTENT_CACHE)) |
361 | f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); | |
1dcc336b | 362 | |
1b38dc8e JK |
363 | /* check the # of cached NAT entries */ |
364 | if (!available_free_memory(sbi, NAT_ENTRIES)) | |
365 | try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); | |
366 | ||
31696580 CY |
367 | if (!available_free_memory(sbi, FREE_NIDS)) |
368 | try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES); | |
369 | ||
1b38dc8e JK |
370 | /* checkpoint is the only way to shrink partial cached entries */ |
371 | if (!available_free_memory(sbi, NAT_ENTRIES) || | |
60b99b48 | 372 | !available_free_memory(sbi, INO_ENTRIES) || |
7d768d2c CY |
373 | excess_prefree_segs(sbi) || |
374 | excess_dirty_nats(sbi) || | |
d0239e1b | 375 | (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) { |
e9f5b8b8 CY |
376 | if (test_opt(sbi, DATA_FLUSH)) { |
377 | struct blk_plug plug; | |
378 | ||
379 | blk_start_plug(&plug); | |
36b35a0d | 380 | sync_dirty_inodes(sbi, FILE_INODE); |
e9f5b8b8 CY |
381 | blk_finish_plug(&plug); |
382 | } | |
4660f9c0 | 383 | f2fs_sync_fs(sbi->sb, true); |
42190d2a | 384 | stat_inc_bg_cp_count(sbi->stat_info); |
36b35a0d | 385 | } |
4660f9c0 JK |
386 | } |
387 | ||
2163d198 | 388 | static int issue_flush_thread(void *data) |
6b4afdd7 JK |
389 | { |
390 | struct f2fs_sb_info *sbi = data; | |
a688b9d9 GZ |
391 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
392 | wait_queue_head_t *q = &fcc->flush_wait_queue; | |
6b4afdd7 JK |
393 | repeat: |
394 | if (kthread_should_stop()) | |
395 | return 0; | |
396 | ||
721bd4d5 | 397 | if (!llist_empty(&fcc->issue_list)) { |
740432f8 | 398 | struct bio *bio; |
6b4afdd7 JK |
399 | struct flush_cmd *cmd, *next; |
400 | int ret; | |
401 | ||
740432f8 JK |
402 | bio = f2fs_bio_alloc(0); |
403 | ||
721bd4d5 GZ |
404 | fcc->dispatch_list = llist_del_all(&fcc->issue_list); |
405 | fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); | |
406 | ||
6b4afdd7 JK |
407 | bio->bi_bdev = sbi->sb->s_bdev; |
408 | ret = submit_bio_wait(WRITE_FLUSH, bio); | |
409 | ||
721bd4d5 GZ |
410 | llist_for_each_entry_safe(cmd, next, |
411 | fcc->dispatch_list, llnode) { | |
6b4afdd7 | 412 | cmd->ret = ret; |
6b4afdd7 JK |
413 | complete(&cmd->wait); |
414 | } | |
a4ed23f2 | 415 | bio_put(bio); |
a688b9d9 | 416 | fcc->dispatch_list = NULL; |
6b4afdd7 JK |
417 | } |
418 | ||
a688b9d9 | 419 | wait_event_interruptible(*q, |
721bd4d5 | 420 | kthread_should_stop() || !llist_empty(&fcc->issue_list)); |
6b4afdd7 JK |
421 | goto repeat; |
422 | } | |
423 | ||
424 | int f2fs_issue_flush(struct f2fs_sb_info *sbi) | |
425 | { | |
a688b9d9 | 426 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
adf8d90b | 427 | struct flush_cmd cmd; |
6b4afdd7 | 428 | |
24a9ee0f JK |
429 | trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER), |
430 | test_opt(sbi, FLUSH_MERGE)); | |
431 | ||
0f7b2abd JK |
432 | if (test_opt(sbi, NOBARRIER)) |
433 | return 0; | |
434 | ||
740432f8 JK |
435 | if (!test_opt(sbi, FLUSH_MERGE)) { |
436 | struct bio *bio = f2fs_bio_alloc(0); | |
437 | int ret; | |
438 | ||
439 | bio->bi_bdev = sbi->sb->s_bdev; | |
440 | ret = submit_bio_wait(WRITE_FLUSH, bio); | |
441 | bio_put(bio); | |
442 | return ret; | |
443 | } | |
6b4afdd7 | 444 | |
adf8d90b | 445 | init_completion(&cmd.wait); |
6b4afdd7 | 446 | |
721bd4d5 | 447 | llist_add(&cmd.llnode, &fcc->issue_list); |
6b4afdd7 | 448 | |
a688b9d9 GZ |
449 | if (!fcc->dispatch_list) |
450 | wake_up(&fcc->flush_wait_queue); | |
6b4afdd7 | 451 | |
adf8d90b CY |
452 | wait_for_completion(&cmd.wait); |
453 | ||
454 | return cmd.ret; | |
6b4afdd7 JK |
455 | } |
456 | ||
2163d198 GZ |
457 | int create_flush_cmd_control(struct f2fs_sb_info *sbi) |
458 | { | |
459 | dev_t dev = sbi->sb->s_bdev->bd_dev; | |
460 | struct flush_cmd_control *fcc; | |
461 | int err = 0; | |
462 | ||
463 | fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL); | |
464 | if (!fcc) | |
465 | return -ENOMEM; | |
2163d198 | 466 | init_waitqueue_head(&fcc->flush_wait_queue); |
721bd4d5 | 467 | init_llist_head(&fcc->issue_list); |
6b2920a5 | 468 | SM_I(sbi)->cmd_control_info = fcc; |
2163d198 GZ |
469 | fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, |
470 | "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); | |
471 | if (IS_ERR(fcc->f2fs_issue_flush)) { | |
472 | err = PTR_ERR(fcc->f2fs_issue_flush); | |
473 | kfree(fcc); | |
6b2920a5 | 474 | SM_I(sbi)->cmd_control_info = NULL; |
2163d198 GZ |
475 | return err; |
476 | } | |
2163d198 GZ |
477 | |
478 | return err; | |
479 | } | |
480 | ||
481 | void destroy_flush_cmd_control(struct f2fs_sb_info *sbi) | |
482 | { | |
6b2920a5 | 483 | struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info; |
2163d198 GZ |
484 | |
485 | if (fcc && fcc->f2fs_issue_flush) | |
486 | kthread_stop(fcc->f2fs_issue_flush); | |
487 | kfree(fcc); | |
6b2920a5 | 488 | SM_I(sbi)->cmd_control_info = NULL; |
2163d198 GZ |
489 | } |
490 | ||
351df4b2 JK |
491 | static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, |
492 | enum dirty_type dirty_type) | |
493 | { | |
494 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
495 | ||
496 | /* need not be added */ | |
497 | if (IS_CURSEG(sbi, segno)) | |
498 | return; | |
499 | ||
500 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
501 | dirty_i->nr_dirty[dirty_type]++; | |
502 | ||
503 | if (dirty_type == DIRTY) { | |
504 | struct seg_entry *sentry = get_seg_entry(sbi, segno); | |
4625d6aa | 505 | enum dirty_type t = sentry->type; |
b2f2c390 | 506 | |
ec325b52 JK |
507 | if (unlikely(t >= DIRTY)) { |
508 | f2fs_bug_on(sbi, 1); | |
509 | return; | |
510 | } | |
4625d6aa CL |
511 | if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) |
512 | dirty_i->nr_dirty[t]++; | |
351df4b2 JK |
513 | } |
514 | } | |
515 | ||
516 | static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, | |
517 | enum dirty_type dirty_type) | |
518 | { | |
519 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
520 | ||
521 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) | |
522 | dirty_i->nr_dirty[dirty_type]--; | |
523 | ||
524 | if (dirty_type == DIRTY) { | |
4625d6aa CL |
525 | struct seg_entry *sentry = get_seg_entry(sbi, segno); |
526 | enum dirty_type t = sentry->type; | |
527 | ||
528 | if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) | |
529 | dirty_i->nr_dirty[t]--; | |
b2f2c390 | 530 | |
5ec4e49f JK |
531 | if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0) |
532 | clear_bit(GET_SECNO(sbi, segno), | |
533 | dirty_i->victim_secmap); | |
351df4b2 JK |
534 | } |
535 | } | |
536 | ||
0a8165d7 | 537 | /* |
351df4b2 JK |
538 | * Should not occur error such as -ENOMEM. |
539 | * Adding dirty entry into seglist is not critical operation. | |
540 | * If a given segment is one of current working segments, it won't be added. | |
541 | */ | |
8d8451af | 542 | static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) |
351df4b2 JK |
543 | { |
544 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
545 | unsigned short valid_blocks; | |
546 | ||
547 | if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) | |
548 | return; | |
549 | ||
550 | mutex_lock(&dirty_i->seglist_lock); | |
551 | ||
552 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
553 | ||
554 | if (valid_blocks == 0) { | |
555 | __locate_dirty_segment(sbi, segno, PRE); | |
556 | __remove_dirty_segment(sbi, segno, DIRTY); | |
557 | } else if (valid_blocks < sbi->blocks_per_seg) { | |
558 | __locate_dirty_segment(sbi, segno, DIRTY); | |
559 | } else { | |
560 | /* Recovery routine with SSR needs this */ | |
561 | __remove_dirty_segment(sbi, segno, DIRTY); | |
562 | } | |
563 | ||
564 | mutex_unlock(&dirty_i->seglist_lock); | |
351df4b2 JK |
565 | } |
566 | ||
1e87a78d | 567 | static int f2fs_issue_discard(struct f2fs_sb_info *sbi, |
37208879 JK |
568 | block_t blkstart, block_t blklen) |
569 | { | |
55cf9cb6 CY |
570 | sector_t start = SECTOR_FROM_BLOCK(blkstart); |
571 | sector_t len = SECTOR_FROM_BLOCK(blklen); | |
a66cdd98 JK |
572 | struct seg_entry *se; |
573 | unsigned int offset; | |
574 | block_t i; | |
575 | ||
576 | for (i = blkstart; i < blkstart + blklen; i++) { | |
577 | se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); | |
578 | offset = GET_BLKOFF_FROM_SEG0(sbi, i); | |
579 | ||
580 | if (!f2fs_test_and_set_bit(offset, se->discard_map)) | |
581 | sbi->discard_blks--; | |
582 | } | |
1661d07c | 583 | trace_f2fs_issue_discard(sbi->sb, blkstart, blklen); |
1e87a78d JK |
584 | return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0); |
585 | } | |
586 | ||
e90c2d28 | 587 | bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr) |
1e87a78d | 588 | { |
60b286c4 | 589 | int err = -EOPNOTSUPP; |
40a02be1 JK |
590 | |
591 | if (test_opt(sbi, DISCARD)) { | |
592 | struct seg_entry *se = get_seg_entry(sbi, | |
593 | GET_SEGNO(sbi, blkaddr)); | |
594 | unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); | |
595 | ||
596 | if (f2fs_test_bit(offset, se->discard_map)) | |
e90c2d28 | 597 | return false; |
40a02be1 JK |
598 | |
599 | err = f2fs_issue_discard(sbi, blkaddr, 1); | |
600 | } | |
601 | ||
e90c2d28 | 602 | if (err) { |
381722d2 | 603 | update_meta_page(sbi, NULL, blkaddr); |
e90c2d28 CY |
604 | return true; |
605 | } | |
606 | return false; | |
37208879 JK |
607 | } |
608 | ||
adf4983b | 609 | static void __add_discard_entry(struct f2fs_sb_info *sbi, |
a66cdd98 JK |
610 | struct cp_control *cpc, struct seg_entry *se, |
611 | unsigned int start, unsigned int end) | |
b2955550 JK |
612 | { |
613 | struct list_head *head = &SM_I(sbi)->discard_list; | |
adf4983b JK |
614 | struct discard_entry *new, *last; |
615 | ||
616 | if (!list_empty(head)) { | |
617 | last = list_last_entry(head, struct discard_entry, list); | |
618 | if (START_BLOCK(sbi, cpc->trim_start) + start == | |
619 | last->blkaddr + last->len) { | |
620 | last->len += end - start; | |
621 | goto done; | |
622 | } | |
623 | } | |
624 | ||
625 | new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS); | |
626 | INIT_LIST_HEAD(&new->list); | |
627 | new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start; | |
628 | new->len = end - start; | |
629 | list_add_tail(&new->list, head); | |
630 | done: | |
631 | SM_I(sbi)->nr_discards += end - start; | |
adf4983b JK |
632 | } |
633 | ||
634 | static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc) | |
635 | { | |
b2955550 JK |
636 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); |
637 | int max_blocks = sbi->blocks_per_seg; | |
4b2fecc8 | 638 | struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); |
b2955550 JK |
639 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; |
640 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; | |
a66cdd98 | 641 | unsigned long *discard_map = (unsigned long *)se->discard_map; |
60a3b782 | 642 | unsigned long *dmap = SIT_I(sbi)->tmp_map; |
b2955550 | 643 | unsigned int start = 0, end = -1; |
4b2fecc8 | 644 | bool force = (cpc->reason == CP_DISCARD); |
b2955550 JK |
645 | int i; |
646 | ||
a66cdd98 | 647 | if (se->valid_blocks == max_blocks) |
b2955550 JK |
648 | return; |
649 | ||
a66cdd98 JK |
650 | if (!force) { |
651 | if (!test_opt(sbi, DISCARD) || !se->valid_blocks || | |
912a83b5 DC |
652 | SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards) |
653 | return; | |
4b2fecc8 JK |
654 | } |
655 | ||
b2955550 JK |
656 | /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ |
657 | for (i = 0; i < entries; i++) | |
a66cdd98 | 658 | dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : |
d7bc2484 | 659 | (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; |
b2955550 | 660 | |
4b2fecc8 | 661 | while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) { |
b2955550 JK |
662 | start = __find_rev_next_bit(dmap, max_blocks, end + 1); |
663 | if (start >= max_blocks) | |
664 | break; | |
665 | ||
666 | end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); | |
a66cdd98 | 667 | __add_discard_entry(sbi, cpc, se, start, end); |
b2955550 JK |
668 | } |
669 | } | |
670 | ||
4b2fecc8 JK |
671 | void release_discard_addrs(struct f2fs_sb_info *sbi) |
672 | { | |
673 | struct list_head *head = &(SM_I(sbi)->discard_list); | |
674 | struct discard_entry *entry, *this; | |
675 | ||
676 | /* drop caches */ | |
677 | list_for_each_entry_safe(entry, this, head, list) { | |
678 | list_del(&entry->list); | |
679 | kmem_cache_free(discard_entry_slab, entry); | |
680 | } | |
681 | } | |
682 | ||
0a8165d7 | 683 | /* |
351df4b2 JK |
684 | * Should call clear_prefree_segments after checkpoint is done. |
685 | */ | |
686 | static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) | |
687 | { | |
688 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
b65ee148 | 689 | unsigned int segno; |
351df4b2 JK |
690 | |
691 | mutex_lock(&dirty_i->seglist_lock); | |
7cd8558b | 692 | for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) |
351df4b2 | 693 | __set_test_and_free(sbi, segno); |
351df4b2 JK |
694 | mutex_unlock(&dirty_i->seglist_lock); |
695 | } | |
696 | ||
836b5a63 | 697 | void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc) |
351df4b2 | 698 | { |
b2955550 | 699 | struct list_head *head = &(SM_I(sbi)->discard_list); |
2d7b822a | 700 | struct discard_entry *entry, *this; |
351df4b2 | 701 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); |
29e59c14 | 702 | unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; |
29e59c14 | 703 | unsigned int start = 0, end = -1; |
351df4b2 JK |
704 | |
705 | mutex_lock(&dirty_i->seglist_lock); | |
29e59c14 | 706 | |
351df4b2 | 707 | while (1) { |
29e59c14 | 708 | int i; |
7cd8558b JK |
709 | start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); |
710 | if (start >= MAIN_SEGS(sbi)) | |
351df4b2 | 711 | break; |
7cd8558b JK |
712 | end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), |
713 | start + 1); | |
29e59c14 CL |
714 | |
715 | for (i = start; i < end; i++) | |
716 | clear_bit(i, prefree_map); | |
717 | ||
718 | dirty_i->nr_dirty[PRE] -= end - start; | |
719 | ||
720 | if (!test_opt(sbi, DISCARD)) | |
721 | continue; | |
351df4b2 | 722 | |
37208879 JK |
723 | f2fs_issue_discard(sbi, START_BLOCK(sbi, start), |
724 | (end - start) << sbi->log_blocks_per_seg); | |
351df4b2 JK |
725 | } |
726 | mutex_unlock(&dirty_i->seglist_lock); | |
b2955550 JK |
727 | |
728 | /* send small discards */ | |
2d7b822a | 729 | list_for_each_entry_safe(entry, this, head, list) { |
836b5a63 JK |
730 | if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen) |
731 | goto skip; | |
37208879 | 732 | f2fs_issue_discard(sbi, entry->blkaddr, entry->len); |
f56aa1c5 | 733 | cpc->trimmed += entry->len; |
836b5a63 | 734 | skip: |
b2955550 JK |
735 | list_del(&entry->list); |
736 | SM_I(sbi)->nr_discards -= entry->len; | |
737 | kmem_cache_free(discard_entry_slab, entry); | |
738 | } | |
351df4b2 JK |
739 | } |
740 | ||
184a5cd2 | 741 | static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) |
351df4b2 JK |
742 | { |
743 | struct sit_info *sit_i = SIT_I(sbi); | |
184a5cd2 CY |
744 | |
745 | if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { | |
351df4b2 | 746 | sit_i->dirty_sentries++; |
184a5cd2 CY |
747 | return false; |
748 | } | |
749 | ||
750 | return true; | |
351df4b2 JK |
751 | } |
752 | ||
753 | static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, | |
754 | unsigned int segno, int modified) | |
755 | { | |
756 | struct seg_entry *se = get_seg_entry(sbi, segno); | |
757 | se->type = type; | |
758 | if (modified) | |
759 | __mark_sit_entry_dirty(sbi, segno); | |
760 | } | |
761 | ||
762 | static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) | |
763 | { | |
764 | struct seg_entry *se; | |
765 | unsigned int segno, offset; | |
766 | long int new_vblocks; | |
767 | ||
768 | segno = GET_SEGNO(sbi, blkaddr); | |
769 | ||
770 | se = get_seg_entry(sbi, segno); | |
771 | new_vblocks = se->valid_blocks + del; | |
491c0854 | 772 | offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); |
351df4b2 | 773 | |
9850cf4a | 774 | f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || |
351df4b2 JK |
775 | (new_vblocks > sbi->blocks_per_seg))); |
776 | ||
777 | se->valid_blocks = new_vblocks; | |
778 | se->mtime = get_mtime(sbi); | |
779 | SIT_I(sbi)->max_mtime = se->mtime; | |
780 | ||
781 | /* Update valid block bitmap */ | |
782 | if (del > 0) { | |
52aca074 | 783 | if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) |
05796763 | 784 | f2fs_bug_on(sbi, 1); |
a66cdd98 JK |
785 | if (!f2fs_test_and_set_bit(offset, se->discard_map)) |
786 | sbi->discard_blks--; | |
351df4b2 | 787 | } else { |
52aca074 | 788 | if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) |
05796763 | 789 | f2fs_bug_on(sbi, 1); |
a66cdd98 JK |
790 | if (f2fs_test_and_clear_bit(offset, se->discard_map)) |
791 | sbi->discard_blks++; | |
351df4b2 JK |
792 | } |
793 | if (!f2fs_test_bit(offset, se->ckpt_valid_map)) | |
794 | se->ckpt_valid_blocks += del; | |
795 | ||
796 | __mark_sit_entry_dirty(sbi, segno); | |
797 | ||
798 | /* update total number of valid blocks to be written in ckpt area */ | |
799 | SIT_I(sbi)->written_valid_blocks += del; | |
800 | ||
801 | if (sbi->segs_per_sec > 1) | |
802 | get_sec_entry(sbi, segno)->valid_blocks += del; | |
803 | } | |
804 | ||
5e443818 | 805 | void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) |
351df4b2 | 806 | { |
5e443818 JK |
807 | update_sit_entry(sbi, new, 1); |
808 | if (GET_SEGNO(sbi, old) != NULL_SEGNO) | |
809 | update_sit_entry(sbi, old, -1); | |
810 | ||
811 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); | |
812 | locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); | |
351df4b2 JK |
813 | } |
814 | ||
815 | void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) | |
816 | { | |
817 | unsigned int segno = GET_SEGNO(sbi, addr); | |
818 | struct sit_info *sit_i = SIT_I(sbi); | |
819 | ||
9850cf4a | 820 | f2fs_bug_on(sbi, addr == NULL_ADDR); |
351df4b2 JK |
821 | if (addr == NEW_ADDR) |
822 | return; | |
823 | ||
824 | /* add it into sit main buffer */ | |
825 | mutex_lock(&sit_i->sentry_lock); | |
826 | ||
827 | update_sit_entry(sbi, addr, -1); | |
828 | ||
829 | /* add it into dirty seglist */ | |
830 | locate_dirty_segment(sbi, segno); | |
831 | ||
832 | mutex_unlock(&sit_i->sentry_lock); | |
833 | } | |
834 | ||
6e2c64ad JK |
835 | bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) |
836 | { | |
837 | struct sit_info *sit_i = SIT_I(sbi); | |
838 | unsigned int segno, offset; | |
839 | struct seg_entry *se; | |
840 | bool is_cp = false; | |
841 | ||
842 | if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) | |
843 | return true; | |
844 | ||
845 | mutex_lock(&sit_i->sentry_lock); | |
846 | ||
847 | segno = GET_SEGNO(sbi, blkaddr); | |
848 | se = get_seg_entry(sbi, segno); | |
849 | offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); | |
850 | ||
851 | if (f2fs_test_bit(offset, se->ckpt_valid_map)) | |
852 | is_cp = true; | |
853 | ||
854 | mutex_unlock(&sit_i->sentry_lock); | |
855 | ||
856 | return is_cp; | |
857 | } | |
858 | ||
0a8165d7 | 859 | /* |
351df4b2 JK |
860 | * This function should be resided under the curseg_mutex lock |
861 | */ | |
862 | static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, | |
e79efe3b | 863 | struct f2fs_summary *sum) |
351df4b2 JK |
864 | { |
865 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
866 | void *addr = curseg->sum_blk; | |
e79efe3b | 867 | addr += curseg->next_blkoff * sizeof(struct f2fs_summary); |
351df4b2 | 868 | memcpy(addr, sum, sizeof(struct f2fs_summary)); |
351df4b2 JK |
869 | } |
870 | ||
0a8165d7 | 871 | /* |
351df4b2 JK |
872 | * Calculate the number of current summary pages for writing |
873 | */ | |
3fa06d7b | 874 | int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) |
351df4b2 | 875 | { |
351df4b2 | 876 | int valid_sum_count = 0; |
9a47938b | 877 | int i, sum_in_page; |
351df4b2 JK |
878 | |
879 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
880 | if (sbi->ckpt->alloc_type[i] == SSR) | |
881 | valid_sum_count += sbi->blocks_per_seg; | |
3fa06d7b CY |
882 | else { |
883 | if (for_ra) | |
884 | valid_sum_count += le16_to_cpu( | |
885 | F2FS_CKPT(sbi)->cur_data_blkoff[i]); | |
886 | else | |
887 | valid_sum_count += curseg_blkoff(sbi, i); | |
888 | } | |
351df4b2 JK |
889 | } |
890 | ||
09cbfeaf | 891 | sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - |
9a47938b FL |
892 | SUM_FOOTER_SIZE) / SUMMARY_SIZE; |
893 | if (valid_sum_count <= sum_in_page) | |
351df4b2 | 894 | return 1; |
9a47938b | 895 | else if ((valid_sum_count - sum_in_page) <= |
09cbfeaf | 896 | (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) |
351df4b2 JK |
897 | return 2; |
898 | return 3; | |
899 | } | |
900 | ||
0a8165d7 | 901 | /* |
351df4b2 JK |
902 | * Caller should put this summary page |
903 | */ | |
904 | struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) | |
905 | { | |
906 | return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); | |
907 | } | |
908 | ||
381722d2 | 909 | void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr) |
351df4b2 JK |
910 | { |
911 | struct page *page = grab_meta_page(sbi, blk_addr); | |
381722d2 CY |
912 | void *dst = page_address(page); |
913 | ||
914 | if (src) | |
09cbfeaf | 915 | memcpy(dst, src, PAGE_SIZE); |
381722d2 | 916 | else |
09cbfeaf | 917 | memset(dst, 0, PAGE_SIZE); |
351df4b2 JK |
918 | set_page_dirty(page); |
919 | f2fs_put_page(page, 1); | |
920 | } | |
921 | ||
381722d2 CY |
922 | static void write_sum_page(struct f2fs_sb_info *sbi, |
923 | struct f2fs_summary_block *sum_blk, block_t blk_addr) | |
924 | { | |
925 | update_meta_page(sbi, (void *)sum_blk, blk_addr); | |
926 | } | |
927 | ||
b7ad7512 CY |
928 | static void write_current_sum_page(struct f2fs_sb_info *sbi, |
929 | int type, block_t blk_addr) | |
930 | { | |
931 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
932 | struct page *page = grab_meta_page(sbi, blk_addr); | |
933 | struct f2fs_summary_block *src = curseg->sum_blk; | |
934 | struct f2fs_summary_block *dst; | |
935 | ||
936 | dst = (struct f2fs_summary_block *)page_address(page); | |
937 | ||
938 | mutex_lock(&curseg->curseg_mutex); | |
939 | ||
940 | down_read(&curseg->journal_rwsem); | |
941 | memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); | |
942 | up_read(&curseg->journal_rwsem); | |
943 | ||
944 | memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); | |
945 | memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); | |
946 | ||
947 | mutex_unlock(&curseg->curseg_mutex); | |
948 | ||
949 | set_page_dirty(page); | |
950 | f2fs_put_page(page, 1); | |
951 | } | |
952 | ||
60374688 JK |
953 | static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) |
954 | { | |
955 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
81fb5e87 | 956 | unsigned int segno = curseg->segno + 1; |
60374688 JK |
957 | struct free_segmap_info *free_i = FREE_I(sbi); |
958 | ||
7cd8558b | 959 | if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) |
81fb5e87 | 960 | return !test_bit(segno, free_i->free_segmap); |
60374688 JK |
961 | return 0; |
962 | } | |
963 | ||
0a8165d7 | 964 | /* |
351df4b2 JK |
965 | * Find a new segment from the free segments bitmap to right order |
966 | * This function should be returned with success, otherwise BUG | |
967 | */ | |
968 | static void get_new_segment(struct f2fs_sb_info *sbi, | |
969 | unsigned int *newseg, bool new_sec, int dir) | |
970 | { | |
971 | struct free_segmap_info *free_i = FREE_I(sbi); | |
351df4b2 | 972 | unsigned int segno, secno, zoneno; |
7cd8558b | 973 | unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; |
351df4b2 JK |
974 | unsigned int hint = *newseg / sbi->segs_per_sec; |
975 | unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); | |
976 | unsigned int left_start = hint; | |
977 | bool init = true; | |
978 | int go_left = 0; | |
979 | int i; | |
980 | ||
1a118ccf | 981 | spin_lock(&free_i->segmap_lock); |
351df4b2 JK |
982 | |
983 | if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { | |
984 | segno = find_next_zero_bit(free_i->free_segmap, | |
0ab14356 CY |
985 | (hint + 1) * sbi->segs_per_sec, *newseg + 1); |
986 | if (segno < (hint + 1) * sbi->segs_per_sec) | |
351df4b2 JK |
987 | goto got_it; |
988 | } | |
989 | find_other_zone: | |
7cd8558b JK |
990 | secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); |
991 | if (secno >= MAIN_SECS(sbi)) { | |
351df4b2 JK |
992 | if (dir == ALLOC_RIGHT) { |
993 | secno = find_next_zero_bit(free_i->free_secmap, | |
7cd8558b JK |
994 | MAIN_SECS(sbi), 0); |
995 | f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); | |
351df4b2 JK |
996 | } else { |
997 | go_left = 1; | |
998 | left_start = hint - 1; | |
999 | } | |
1000 | } | |
1001 | if (go_left == 0) | |
1002 | goto skip_left; | |
1003 | ||
1004 | while (test_bit(left_start, free_i->free_secmap)) { | |
1005 | if (left_start > 0) { | |
1006 | left_start--; | |
1007 | continue; | |
1008 | } | |
1009 | left_start = find_next_zero_bit(free_i->free_secmap, | |
7cd8558b JK |
1010 | MAIN_SECS(sbi), 0); |
1011 | f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); | |
351df4b2 JK |
1012 | break; |
1013 | } | |
1014 | secno = left_start; | |
1015 | skip_left: | |
1016 | hint = secno; | |
1017 | segno = secno * sbi->segs_per_sec; | |
1018 | zoneno = secno / sbi->secs_per_zone; | |
1019 | ||
1020 | /* give up on finding another zone */ | |
1021 | if (!init) | |
1022 | goto got_it; | |
1023 | if (sbi->secs_per_zone == 1) | |
1024 | goto got_it; | |
1025 | if (zoneno == old_zoneno) | |
1026 | goto got_it; | |
1027 | if (dir == ALLOC_LEFT) { | |
1028 | if (!go_left && zoneno + 1 >= total_zones) | |
1029 | goto got_it; | |
1030 | if (go_left && zoneno == 0) | |
1031 | goto got_it; | |
1032 | } | |
1033 | for (i = 0; i < NR_CURSEG_TYPE; i++) | |
1034 | if (CURSEG_I(sbi, i)->zone == zoneno) | |
1035 | break; | |
1036 | ||
1037 | if (i < NR_CURSEG_TYPE) { | |
1038 | /* zone is in user, try another */ | |
1039 | if (go_left) | |
1040 | hint = zoneno * sbi->secs_per_zone - 1; | |
1041 | else if (zoneno + 1 >= total_zones) | |
1042 | hint = 0; | |
1043 | else | |
1044 | hint = (zoneno + 1) * sbi->secs_per_zone; | |
1045 | init = false; | |
1046 | goto find_other_zone; | |
1047 | } | |
1048 | got_it: | |
1049 | /* set it as dirty segment in free segmap */ | |
9850cf4a | 1050 | f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); |
351df4b2 JK |
1051 | __set_inuse(sbi, segno); |
1052 | *newseg = segno; | |
1a118ccf | 1053 | spin_unlock(&free_i->segmap_lock); |
351df4b2 JK |
1054 | } |
1055 | ||
1056 | static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) | |
1057 | { | |
1058 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1059 | struct summary_footer *sum_footer; | |
1060 | ||
1061 | curseg->segno = curseg->next_segno; | |
1062 | curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); | |
1063 | curseg->next_blkoff = 0; | |
1064 | curseg->next_segno = NULL_SEGNO; | |
1065 | ||
1066 | sum_footer = &(curseg->sum_blk->footer); | |
1067 | memset(sum_footer, 0, sizeof(struct summary_footer)); | |
1068 | if (IS_DATASEG(type)) | |
1069 | SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); | |
1070 | if (IS_NODESEG(type)) | |
1071 | SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); | |
1072 | __set_sit_entry_type(sbi, type, curseg->segno, modified); | |
1073 | } | |
1074 | ||
0a8165d7 | 1075 | /* |
351df4b2 JK |
1076 | * Allocate a current working segment. |
1077 | * This function always allocates a free segment in LFS manner. | |
1078 | */ | |
1079 | static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) | |
1080 | { | |
1081 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1082 | unsigned int segno = curseg->segno; | |
1083 | int dir = ALLOC_LEFT; | |
1084 | ||
1085 | write_sum_page(sbi, curseg->sum_blk, | |
81fb5e87 | 1086 | GET_SUM_BLOCK(sbi, segno)); |
351df4b2 JK |
1087 | if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) |
1088 | dir = ALLOC_RIGHT; | |
1089 | ||
1090 | if (test_opt(sbi, NOHEAP)) | |
1091 | dir = ALLOC_RIGHT; | |
1092 | ||
1093 | get_new_segment(sbi, &segno, new_sec, dir); | |
1094 | curseg->next_segno = segno; | |
1095 | reset_curseg(sbi, type, 1); | |
1096 | curseg->alloc_type = LFS; | |
1097 | } | |
1098 | ||
1099 | static void __next_free_blkoff(struct f2fs_sb_info *sbi, | |
1100 | struct curseg_info *seg, block_t start) | |
1101 | { | |
1102 | struct seg_entry *se = get_seg_entry(sbi, seg->segno); | |
e81c93cf | 1103 | int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); |
60a3b782 | 1104 | unsigned long *target_map = SIT_I(sbi)->tmp_map; |
e81c93cf CL |
1105 | unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; |
1106 | unsigned long *cur_map = (unsigned long *)se->cur_valid_map; | |
1107 | int i, pos; | |
1108 | ||
1109 | for (i = 0; i < entries; i++) | |
1110 | target_map[i] = ckpt_map[i] | cur_map[i]; | |
1111 | ||
1112 | pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); | |
1113 | ||
1114 | seg->next_blkoff = pos; | |
351df4b2 JK |
1115 | } |
1116 | ||
0a8165d7 | 1117 | /* |
351df4b2 JK |
1118 | * If a segment is written by LFS manner, next block offset is just obtained |
1119 | * by increasing the current block offset. However, if a segment is written by | |
1120 | * SSR manner, next block offset obtained by calling __next_free_blkoff | |
1121 | */ | |
1122 | static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, | |
1123 | struct curseg_info *seg) | |
1124 | { | |
1125 | if (seg->alloc_type == SSR) | |
1126 | __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); | |
1127 | else | |
1128 | seg->next_blkoff++; | |
1129 | } | |
1130 | ||
0a8165d7 | 1131 | /* |
e1c42045 | 1132 | * This function always allocates a used segment(from dirty seglist) by SSR |
351df4b2 JK |
1133 | * manner, so it should recover the existing segment information of valid blocks |
1134 | */ | |
1135 | static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) | |
1136 | { | |
1137 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
1138 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1139 | unsigned int new_segno = curseg->next_segno; | |
1140 | struct f2fs_summary_block *sum_node; | |
1141 | struct page *sum_page; | |
1142 | ||
1143 | write_sum_page(sbi, curseg->sum_blk, | |
1144 | GET_SUM_BLOCK(sbi, curseg->segno)); | |
1145 | __set_test_and_inuse(sbi, new_segno); | |
1146 | ||
1147 | mutex_lock(&dirty_i->seglist_lock); | |
1148 | __remove_dirty_segment(sbi, new_segno, PRE); | |
1149 | __remove_dirty_segment(sbi, new_segno, DIRTY); | |
1150 | mutex_unlock(&dirty_i->seglist_lock); | |
1151 | ||
1152 | reset_curseg(sbi, type, 1); | |
1153 | curseg->alloc_type = SSR; | |
1154 | __next_free_blkoff(sbi, curseg, 0); | |
1155 | ||
1156 | if (reuse) { | |
1157 | sum_page = get_sum_page(sbi, new_segno); | |
1158 | sum_node = (struct f2fs_summary_block *)page_address(sum_page); | |
1159 | memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); | |
1160 | f2fs_put_page(sum_page, 1); | |
1161 | } | |
1162 | } | |
1163 | ||
43727527 JK |
1164 | static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) |
1165 | { | |
1166 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1167 | const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; | |
1168 | ||
1169 | if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0)) | |
1170 | return v_ops->get_victim(sbi, | |
1171 | &(curseg)->next_segno, BG_GC, type, SSR); | |
1172 | ||
1173 | /* For data segments, let's do SSR more intensively */ | |
1174 | for (; type >= CURSEG_HOT_DATA; type--) | |
1175 | if (v_ops->get_victim(sbi, &(curseg)->next_segno, | |
1176 | BG_GC, type, SSR)) | |
1177 | return 1; | |
1178 | return 0; | |
1179 | } | |
1180 | ||
351df4b2 JK |
1181 | /* |
1182 | * flush out current segment and replace it with new segment | |
1183 | * This function should be returned with success, otherwise BUG | |
1184 | */ | |
1185 | static void allocate_segment_by_default(struct f2fs_sb_info *sbi, | |
1186 | int type, bool force) | |
1187 | { | |
1188 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
351df4b2 | 1189 | |
7b405275 | 1190 | if (force) |
351df4b2 | 1191 | new_curseg(sbi, type, true); |
7b405275 | 1192 | else if (type == CURSEG_WARM_NODE) |
351df4b2 | 1193 | new_curseg(sbi, type, false); |
60374688 JK |
1194 | else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) |
1195 | new_curseg(sbi, type, false); | |
351df4b2 JK |
1196 | else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) |
1197 | change_curseg(sbi, type, true); | |
1198 | else | |
1199 | new_curseg(sbi, type, false); | |
dcdfff65 JK |
1200 | |
1201 | stat_inc_seg_type(sbi, curseg); | |
351df4b2 JK |
1202 | } |
1203 | ||
38aa0889 JK |
1204 | static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type) |
1205 | { | |
1206 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1207 | unsigned int old_segno; | |
1208 | ||
1209 | old_segno = curseg->segno; | |
1210 | SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true); | |
1211 | locate_dirty_segment(sbi, old_segno); | |
1212 | } | |
1213 | ||
351df4b2 JK |
1214 | void allocate_new_segments(struct f2fs_sb_info *sbi) |
1215 | { | |
351df4b2 JK |
1216 | int i; |
1217 | ||
38aa0889 JK |
1218 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) |
1219 | __allocate_new_segments(sbi, i); | |
351df4b2 JK |
1220 | } |
1221 | ||
1222 | static const struct segment_allocation default_salloc_ops = { | |
1223 | .allocate_segment = allocate_segment_by_default, | |
1224 | }; | |
1225 | ||
4b2fecc8 JK |
1226 | int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) |
1227 | { | |
f7ef9b83 JK |
1228 | __u64 start = F2FS_BYTES_TO_BLK(range->start); |
1229 | __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; | |
4b2fecc8 JK |
1230 | unsigned int start_segno, end_segno; |
1231 | struct cp_control cpc; | |
c34f42e2 | 1232 | int err = 0; |
4b2fecc8 | 1233 | |
836b5a63 | 1234 | if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) |
4b2fecc8 JK |
1235 | return -EINVAL; |
1236 | ||
9bd27ae4 | 1237 | cpc.trimmed = 0; |
7cd8558b | 1238 | if (end <= MAIN_BLKADDR(sbi)) |
4b2fecc8 JK |
1239 | goto out; |
1240 | ||
1241 | /* start/end segment number in main_area */ | |
7cd8558b JK |
1242 | start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); |
1243 | end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : | |
1244 | GET_SEGNO(sbi, end); | |
4b2fecc8 | 1245 | cpc.reason = CP_DISCARD; |
836b5a63 | 1246 | cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); |
4b2fecc8 JK |
1247 | |
1248 | /* do checkpoint to issue discard commands safely */ | |
bba681cb JK |
1249 | for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) { |
1250 | cpc.trim_start = start_segno; | |
a66cdd98 JK |
1251 | |
1252 | if (sbi->discard_blks == 0) | |
1253 | break; | |
1254 | else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi)) | |
1255 | cpc.trim_end = end_segno; | |
1256 | else | |
1257 | cpc.trim_end = min_t(unsigned int, | |
1258 | rounddown(start_segno + | |
bba681cb JK |
1259 | BATCHED_TRIM_SEGMENTS(sbi), |
1260 | sbi->segs_per_sec) - 1, end_segno); | |
1261 | ||
1262 | mutex_lock(&sbi->gc_mutex); | |
c34f42e2 | 1263 | err = write_checkpoint(sbi, &cpc); |
bba681cb JK |
1264 | mutex_unlock(&sbi->gc_mutex); |
1265 | } | |
4b2fecc8 | 1266 | out: |
f7ef9b83 | 1267 | range->len = F2FS_BLK_TO_BYTES(cpc.trimmed); |
c34f42e2 | 1268 | return err; |
4b2fecc8 JK |
1269 | } |
1270 | ||
351df4b2 JK |
1271 | static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) |
1272 | { | |
1273 | struct curseg_info *curseg = CURSEG_I(sbi, type); | |
1274 | if (curseg->next_blkoff < sbi->blocks_per_seg) | |
1275 | return true; | |
1276 | return false; | |
1277 | } | |
1278 | ||
1279 | static int __get_segment_type_2(struct page *page, enum page_type p_type) | |
1280 | { | |
1281 | if (p_type == DATA) | |
1282 | return CURSEG_HOT_DATA; | |
1283 | else | |
1284 | return CURSEG_HOT_NODE; | |
1285 | } | |
1286 | ||
1287 | static int __get_segment_type_4(struct page *page, enum page_type p_type) | |
1288 | { | |
1289 | if (p_type == DATA) { | |
1290 | struct inode *inode = page->mapping->host; | |
1291 | ||
1292 | if (S_ISDIR(inode->i_mode)) | |
1293 | return CURSEG_HOT_DATA; | |
1294 | else | |
1295 | return CURSEG_COLD_DATA; | |
1296 | } else { | |
a344b9fd JK |
1297 | if (IS_DNODE(page) && is_cold_node(page)) |
1298 | return CURSEG_WARM_NODE; | |
351df4b2 JK |
1299 | else |
1300 | return CURSEG_COLD_NODE; | |
1301 | } | |
1302 | } | |
1303 | ||
1304 | static int __get_segment_type_6(struct page *page, enum page_type p_type) | |
1305 | { | |
1306 | if (p_type == DATA) { | |
1307 | struct inode *inode = page->mapping->host; | |
1308 | ||
1309 | if (S_ISDIR(inode->i_mode)) | |
1310 | return CURSEG_HOT_DATA; | |
354a3399 | 1311 | else if (is_cold_data(page) || file_is_cold(inode)) |
351df4b2 JK |
1312 | return CURSEG_COLD_DATA; |
1313 | else | |
1314 | return CURSEG_WARM_DATA; | |
1315 | } else { | |
1316 | if (IS_DNODE(page)) | |
1317 | return is_cold_node(page) ? CURSEG_WARM_NODE : | |
1318 | CURSEG_HOT_NODE; | |
1319 | else | |
1320 | return CURSEG_COLD_NODE; | |
1321 | } | |
1322 | } | |
1323 | ||
1324 | static int __get_segment_type(struct page *page, enum page_type p_type) | |
1325 | { | |
4081363f | 1326 | switch (F2FS_P_SB(page)->active_logs) { |
351df4b2 JK |
1327 | case 2: |
1328 | return __get_segment_type_2(page, p_type); | |
1329 | case 4: | |
1330 | return __get_segment_type_4(page, p_type); | |
351df4b2 | 1331 | } |
12a67146 | 1332 | /* NR_CURSEG_TYPE(6) logs by default */ |
9850cf4a JK |
1333 | f2fs_bug_on(F2FS_P_SB(page), |
1334 | F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE); | |
12a67146 | 1335 | return __get_segment_type_6(page, p_type); |
351df4b2 JK |
1336 | } |
1337 | ||
bfad7c2d JK |
1338 | void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, |
1339 | block_t old_blkaddr, block_t *new_blkaddr, | |
1340 | struct f2fs_summary *sum, int type) | |
351df4b2 JK |
1341 | { |
1342 | struct sit_info *sit_i = SIT_I(sbi); | |
1343 | struct curseg_info *curseg; | |
38aa0889 JK |
1344 | bool direct_io = (type == CURSEG_DIRECT_IO); |
1345 | ||
1346 | type = direct_io ? CURSEG_WARM_DATA : type; | |
351df4b2 | 1347 | |
351df4b2 JK |
1348 | curseg = CURSEG_I(sbi, type); |
1349 | ||
1350 | mutex_lock(&curseg->curseg_mutex); | |
21cb1d99 | 1351 | mutex_lock(&sit_i->sentry_lock); |
351df4b2 | 1352 | |
38aa0889 | 1353 | /* direct_io'ed data is aligned to the segment for better performance */ |
47e70ca4 JK |
1354 | if (direct_io && curseg->next_blkoff && |
1355 | !has_not_enough_free_secs(sbi, 0)) | |
38aa0889 JK |
1356 | __allocate_new_segments(sbi, type); |
1357 | ||
351df4b2 | 1358 | *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); |
351df4b2 JK |
1359 | |
1360 | /* | |
1361 | * __add_sum_entry should be resided under the curseg_mutex | |
1362 | * because, this function updates a summary entry in the | |
1363 | * current summary block. | |
1364 | */ | |
e79efe3b | 1365 | __add_sum_entry(sbi, type, sum); |
351df4b2 | 1366 | |
351df4b2 | 1367 | __refresh_next_blkoff(sbi, curseg); |
dcdfff65 JK |
1368 | |
1369 | stat_inc_block_count(sbi, curseg); | |
351df4b2 | 1370 | |
5e443818 JK |
1371 | if (!__has_curseg_space(sbi, type)) |
1372 | sit_i->s_ops->allocate_segment(sbi, type, false); | |
351df4b2 JK |
1373 | /* |
1374 | * SIT information should be updated before segment allocation, | |
1375 | * since SSR needs latest valid block information. | |
1376 | */ | |
1377 | refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); | |
5e443818 | 1378 | |
351df4b2 JK |
1379 | mutex_unlock(&sit_i->sentry_lock); |
1380 | ||
bfad7c2d | 1381 | if (page && IS_NODESEG(type)) |
351df4b2 JK |
1382 | fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); |
1383 | ||
bfad7c2d JK |
1384 | mutex_unlock(&curseg->curseg_mutex); |
1385 | } | |
1386 | ||
05ca3632 | 1387 | static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) |
bfad7c2d | 1388 | { |
05ca3632 | 1389 | int type = __get_segment_type(fio->page, fio->type); |
bfad7c2d | 1390 | |
7a9d7548 CY |
1391 | allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, |
1392 | &fio->new_blkaddr, sum, type); | |
bfad7c2d | 1393 | |
351df4b2 | 1394 | /* writeout dirty page into bdev */ |
05ca3632 | 1395 | f2fs_submit_page_mbio(fio); |
351df4b2 JK |
1396 | } |
1397 | ||
577e3495 | 1398 | void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) |
351df4b2 | 1399 | { |
458e6197 | 1400 | struct f2fs_io_info fio = { |
05ca3632 | 1401 | .sbi = sbi, |
458e6197 | 1402 | .type = META, |
cf04e8eb | 1403 | .rw = WRITE_SYNC | REQ_META | REQ_PRIO, |
7a9d7548 CY |
1404 | .old_blkaddr = page->index, |
1405 | .new_blkaddr = page->index, | |
05ca3632 | 1406 | .page = page, |
4375a336 | 1407 | .encrypted_page = NULL, |
458e6197 JK |
1408 | }; |
1409 | ||
2b947003 CY |
1410 | if (unlikely(page->index >= MAIN_BLKADDR(sbi))) |
1411 | fio.rw &= ~REQ_META; | |
1412 | ||
351df4b2 | 1413 | set_page_writeback(page); |
05ca3632 | 1414 | f2fs_submit_page_mbio(&fio); |
351df4b2 JK |
1415 | } |
1416 | ||
05ca3632 | 1417 | void write_node_page(unsigned int nid, struct f2fs_io_info *fio) |
351df4b2 JK |
1418 | { |
1419 | struct f2fs_summary sum; | |
05ca3632 | 1420 | |
351df4b2 | 1421 | set_summary(&sum, nid, 0, 0); |
05ca3632 | 1422 | do_write_page(&sum, fio); |
351df4b2 JK |
1423 | } |
1424 | ||
05ca3632 | 1425 | void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio) |
351df4b2 | 1426 | { |
05ca3632 | 1427 | struct f2fs_sb_info *sbi = fio->sbi; |
351df4b2 JK |
1428 | struct f2fs_summary sum; |
1429 | struct node_info ni; | |
1430 | ||
9850cf4a | 1431 | f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); |
351df4b2 JK |
1432 | get_node_info(sbi, dn->nid, &ni); |
1433 | set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); | |
05ca3632 | 1434 | do_write_page(&sum, fio); |
f28b3434 | 1435 | f2fs_update_data_blkaddr(dn, fio->new_blkaddr); |
351df4b2 JK |
1436 | } |
1437 | ||
05ca3632 | 1438 | void rewrite_data_page(struct f2fs_io_info *fio) |
351df4b2 | 1439 | { |
7a9d7548 | 1440 | fio->new_blkaddr = fio->old_blkaddr; |
05ca3632 JK |
1441 | stat_inc_inplace_blocks(fio->sbi); |
1442 | f2fs_submit_page_mbio(fio); | |
351df4b2 JK |
1443 | } |
1444 | ||
4356e48e | 1445 | void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
19f106bc | 1446 | block_t old_blkaddr, block_t new_blkaddr, |
28bc106b | 1447 | bool recover_curseg, bool recover_newaddr) |
351df4b2 JK |
1448 | { |
1449 | struct sit_info *sit_i = SIT_I(sbi); | |
1450 | struct curseg_info *curseg; | |
1451 | unsigned int segno, old_cursegno; | |
1452 | struct seg_entry *se; | |
1453 | int type; | |
19f106bc | 1454 | unsigned short old_blkoff; |
351df4b2 JK |
1455 | |
1456 | segno = GET_SEGNO(sbi, new_blkaddr); | |
1457 | se = get_seg_entry(sbi, segno); | |
1458 | type = se->type; | |
1459 | ||
19f106bc CY |
1460 | if (!recover_curseg) { |
1461 | /* for recovery flow */ | |
1462 | if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { | |
1463 | if (old_blkaddr == NULL_ADDR) | |
1464 | type = CURSEG_COLD_DATA; | |
1465 | else | |
1466 | type = CURSEG_WARM_DATA; | |
1467 | } | |
1468 | } else { | |
1469 | if (!IS_CURSEG(sbi, segno)) | |
351df4b2 JK |
1470 | type = CURSEG_WARM_DATA; |
1471 | } | |
19f106bc | 1472 | |
351df4b2 JK |
1473 | curseg = CURSEG_I(sbi, type); |
1474 | ||
1475 | mutex_lock(&curseg->curseg_mutex); | |
1476 | mutex_lock(&sit_i->sentry_lock); | |
1477 | ||
1478 | old_cursegno = curseg->segno; | |
19f106bc | 1479 | old_blkoff = curseg->next_blkoff; |
351df4b2 JK |
1480 | |
1481 | /* change the current segment */ | |
1482 | if (segno != curseg->segno) { | |
1483 | curseg->next_segno = segno; | |
1484 | change_curseg(sbi, type, true); | |
1485 | } | |
1486 | ||
491c0854 | 1487 | curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); |
e79efe3b | 1488 | __add_sum_entry(sbi, type, sum); |
351df4b2 | 1489 | |
28bc106b | 1490 | if (!recover_curseg || recover_newaddr) |
6e2c64ad JK |
1491 | update_sit_entry(sbi, new_blkaddr, 1); |
1492 | if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) | |
1493 | update_sit_entry(sbi, old_blkaddr, -1); | |
1494 | ||
1495 | locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); | |
1496 | locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); | |
1497 | ||
351df4b2 | 1498 | locate_dirty_segment(sbi, old_cursegno); |
351df4b2 | 1499 | |
19f106bc CY |
1500 | if (recover_curseg) { |
1501 | if (old_cursegno != curseg->segno) { | |
1502 | curseg->next_segno = old_cursegno; | |
1503 | change_curseg(sbi, type, true); | |
1504 | } | |
1505 | curseg->next_blkoff = old_blkoff; | |
1506 | } | |
1507 | ||
351df4b2 JK |
1508 | mutex_unlock(&sit_i->sentry_lock); |
1509 | mutex_unlock(&curseg->curseg_mutex); | |
1510 | } | |
1511 | ||
528e3459 CY |
1512 | void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, |
1513 | block_t old_addr, block_t new_addr, | |
28bc106b CY |
1514 | unsigned char version, bool recover_curseg, |
1515 | bool recover_newaddr) | |
528e3459 CY |
1516 | { |
1517 | struct f2fs_summary sum; | |
1518 | ||
1519 | set_summary(&sum, dn->nid, dn->ofs_in_node, version); | |
1520 | ||
28bc106b CY |
1521 | __f2fs_replace_block(sbi, &sum, old_addr, new_addr, |
1522 | recover_curseg, recover_newaddr); | |
528e3459 | 1523 | |
f28b3434 | 1524 | f2fs_update_data_blkaddr(dn, new_addr); |
528e3459 CY |
1525 | } |
1526 | ||
93dfe2ac | 1527 | void f2fs_wait_on_page_writeback(struct page *page, |
fec1d657 | 1528 | enum page_type type, bool ordered) |
93dfe2ac | 1529 | { |
93dfe2ac | 1530 | if (PageWriteback(page)) { |
4081363f JK |
1531 | struct f2fs_sb_info *sbi = F2FS_P_SB(page); |
1532 | ||
0c3a5797 | 1533 | f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE); |
fec1d657 JK |
1534 | if (ordered) |
1535 | wait_on_page_writeback(page); | |
1536 | else | |
1537 | wait_for_stable_page(page); | |
93dfe2ac JK |
1538 | } |
1539 | } | |
1540 | ||
08b39fbd CY |
1541 | void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi, |
1542 | block_t blkaddr) | |
1543 | { | |
1544 | struct page *cpage; | |
1545 | ||
1546 | if (blkaddr == NEW_ADDR) | |
1547 | return; | |
1548 | ||
1549 | f2fs_bug_on(sbi, blkaddr == NULL_ADDR); | |
1550 | ||
1551 | cpage = find_lock_page(META_MAPPING(sbi), blkaddr); | |
1552 | if (cpage) { | |
fec1d657 | 1553 | f2fs_wait_on_page_writeback(cpage, DATA, true); |
08b39fbd CY |
1554 | f2fs_put_page(cpage, 1); |
1555 | } | |
1556 | } | |
1557 | ||
351df4b2 JK |
1558 | static int read_compacted_summaries(struct f2fs_sb_info *sbi) |
1559 | { | |
1560 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1561 | struct curseg_info *seg_i; | |
1562 | unsigned char *kaddr; | |
1563 | struct page *page; | |
1564 | block_t start; | |
1565 | int i, j, offset; | |
1566 | ||
1567 | start = start_sum_block(sbi); | |
1568 | ||
1569 | page = get_meta_page(sbi, start++); | |
1570 | kaddr = (unsigned char *)page_address(page); | |
1571 | ||
1572 | /* Step 1: restore nat cache */ | |
1573 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
b7ad7512 | 1574 | memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); |
351df4b2 JK |
1575 | |
1576 | /* Step 2: restore sit cache */ | |
1577 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
b7ad7512 | 1578 | memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); |
351df4b2 JK |
1579 | offset = 2 * SUM_JOURNAL_SIZE; |
1580 | ||
1581 | /* Step 3: restore summary entries */ | |
1582 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1583 | unsigned short blk_off; | |
1584 | unsigned int segno; | |
1585 | ||
1586 | seg_i = CURSEG_I(sbi, i); | |
1587 | segno = le32_to_cpu(ckpt->cur_data_segno[i]); | |
1588 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); | |
1589 | seg_i->next_segno = segno; | |
1590 | reset_curseg(sbi, i, 0); | |
1591 | seg_i->alloc_type = ckpt->alloc_type[i]; | |
1592 | seg_i->next_blkoff = blk_off; | |
1593 | ||
1594 | if (seg_i->alloc_type == SSR) | |
1595 | blk_off = sbi->blocks_per_seg; | |
1596 | ||
1597 | for (j = 0; j < blk_off; j++) { | |
1598 | struct f2fs_summary *s; | |
1599 | s = (struct f2fs_summary *)(kaddr + offset); | |
1600 | seg_i->sum_blk->entries[j] = *s; | |
1601 | offset += SUMMARY_SIZE; | |
09cbfeaf | 1602 | if (offset + SUMMARY_SIZE <= PAGE_SIZE - |
351df4b2 JK |
1603 | SUM_FOOTER_SIZE) |
1604 | continue; | |
1605 | ||
1606 | f2fs_put_page(page, 1); | |
1607 | page = NULL; | |
1608 | ||
1609 | page = get_meta_page(sbi, start++); | |
1610 | kaddr = (unsigned char *)page_address(page); | |
1611 | offset = 0; | |
1612 | } | |
1613 | } | |
1614 | f2fs_put_page(page, 1); | |
1615 | return 0; | |
1616 | } | |
1617 | ||
1618 | static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) | |
1619 | { | |
1620 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1621 | struct f2fs_summary_block *sum; | |
1622 | struct curseg_info *curseg; | |
1623 | struct page *new; | |
1624 | unsigned short blk_off; | |
1625 | unsigned int segno = 0; | |
1626 | block_t blk_addr = 0; | |
1627 | ||
1628 | /* get segment number and block addr */ | |
1629 | if (IS_DATASEG(type)) { | |
1630 | segno = le32_to_cpu(ckpt->cur_data_segno[type]); | |
1631 | blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - | |
1632 | CURSEG_HOT_DATA]); | |
119ee914 | 1633 | if (__exist_node_summaries(sbi)) |
351df4b2 JK |
1634 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); |
1635 | else | |
1636 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); | |
1637 | } else { | |
1638 | segno = le32_to_cpu(ckpt->cur_node_segno[type - | |
1639 | CURSEG_HOT_NODE]); | |
1640 | blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - | |
1641 | CURSEG_HOT_NODE]); | |
119ee914 | 1642 | if (__exist_node_summaries(sbi)) |
351df4b2 JK |
1643 | blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, |
1644 | type - CURSEG_HOT_NODE); | |
1645 | else | |
1646 | blk_addr = GET_SUM_BLOCK(sbi, segno); | |
1647 | } | |
1648 | ||
1649 | new = get_meta_page(sbi, blk_addr); | |
1650 | sum = (struct f2fs_summary_block *)page_address(new); | |
1651 | ||
1652 | if (IS_NODESEG(type)) { | |
119ee914 | 1653 | if (__exist_node_summaries(sbi)) { |
351df4b2 JK |
1654 | struct f2fs_summary *ns = &sum->entries[0]; |
1655 | int i; | |
1656 | for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { | |
1657 | ns->version = 0; | |
1658 | ns->ofs_in_node = 0; | |
1659 | } | |
1660 | } else { | |
d653788a GZ |
1661 | int err; |
1662 | ||
1663 | err = restore_node_summary(sbi, segno, sum); | |
1664 | if (err) { | |
351df4b2 | 1665 | f2fs_put_page(new, 1); |
d653788a | 1666 | return err; |
351df4b2 JK |
1667 | } |
1668 | } | |
1669 | } | |
1670 | ||
1671 | /* set uncompleted segment to curseg */ | |
1672 | curseg = CURSEG_I(sbi, type); | |
1673 | mutex_lock(&curseg->curseg_mutex); | |
b7ad7512 CY |
1674 | |
1675 | /* update journal info */ | |
1676 | down_write(&curseg->journal_rwsem); | |
1677 | memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); | |
1678 | up_write(&curseg->journal_rwsem); | |
1679 | ||
1680 | memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); | |
1681 | memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); | |
351df4b2 JK |
1682 | curseg->next_segno = segno; |
1683 | reset_curseg(sbi, type, 0); | |
1684 | curseg->alloc_type = ckpt->alloc_type[type]; | |
1685 | curseg->next_blkoff = blk_off; | |
1686 | mutex_unlock(&curseg->curseg_mutex); | |
1687 | f2fs_put_page(new, 1); | |
1688 | return 0; | |
1689 | } | |
1690 | ||
1691 | static int restore_curseg_summaries(struct f2fs_sb_info *sbi) | |
1692 | { | |
1693 | int type = CURSEG_HOT_DATA; | |
e4fc5fbf | 1694 | int err; |
351df4b2 | 1695 | |
25ca923b | 1696 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) { |
3fa06d7b CY |
1697 | int npages = npages_for_summary_flush(sbi, true); |
1698 | ||
1699 | if (npages >= 2) | |
1700 | ra_meta_pages(sbi, start_sum_block(sbi), npages, | |
26879fb1 | 1701 | META_CP, true); |
3fa06d7b | 1702 | |
351df4b2 JK |
1703 | /* restore for compacted data summary */ |
1704 | if (read_compacted_summaries(sbi)) | |
1705 | return -EINVAL; | |
1706 | type = CURSEG_HOT_NODE; | |
1707 | } | |
1708 | ||
119ee914 | 1709 | if (__exist_node_summaries(sbi)) |
3fa06d7b | 1710 | ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), |
26879fb1 | 1711 | NR_CURSEG_TYPE - type, META_CP, true); |
3fa06d7b | 1712 | |
e4fc5fbf CY |
1713 | for (; type <= CURSEG_COLD_NODE; type++) { |
1714 | err = read_normal_summaries(sbi, type); | |
1715 | if (err) | |
1716 | return err; | |
1717 | } | |
1718 | ||
351df4b2 JK |
1719 | return 0; |
1720 | } | |
1721 | ||
1722 | static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) | |
1723 | { | |
1724 | struct page *page; | |
1725 | unsigned char *kaddr; | |
1726 | struct f2fs_summary *summary; | |
1727 | struct curseg_info *seg_i; | |
1728 | int written_size = 0; | |
1729 | int i, j; | |
1730 | ||
1731 | page = grab_meta_page(sbi, blkaddr++); | |
1732 | kaddr = (unsigned char *)page_address(page); | |
1733 | ||
1734 | /* Step 1: write nat cache */ | |
1735 | seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); | |
b7ad7512 | 1736 | memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); |
351df4b2 JK |
1737 | written_size += SUM_JOURNAL_SIZE; |
1738 | ||
1739 | /* Step 2: write sit cache */ | |
1740 | seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
b7ad7512 | 1741 | memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); |
351df4b2 JK |
1742 | written_size += SUM_JOURNAL_SIZE; |
1743 | ||
351df4b2 JK |
1744 | /* Step 3: write summary entries */ |
1745 | for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { | |
1746 | unsigned short blkoff; | |
1747 | seg_i = CURSEG_I(sbi, i); | |
1748 | if (sbi->ckpt->alloc_type[i] == SSR) | |
1749 | blkoff = sbi->blocks_per_seg; | |
1750 | else | |
1751 | blkoff = curseg_blkoff(sbi, i); | |
1752 | ||
1753 | for (j = 0; j < blkoff; j++) { | |
1754 | if (!page) { | |
1755 | page = grab_meta_page(sbi, blkaddr++); | |
1756 | kaddr = (unsigned char *)page_address(page); | |
1757 | written_size = 0; | |
1758 | } | |
1759 | summary = (struct f2fs_summary *)(kaddr + written_size); | |
1760 | *summary = seg_i->sum_blk->entries[j]; | |
1761 | written_size += SUMMARY_SIZE; | |
351df4b2 | 1762 | |
09cbfeaf | 1763 | if (written_size + SUMMARY_SIZE <= PAGE_SIZE - |
351df4b2 JK |
1764 | SUM_FOOTER_SIZE) |
1765 | continue; | |
1766 | ||
e8d61a74 | 1767 | set_page_dirty(page); |
351df4b2 JK |
1768 | f2fs_put_page(page, 1); |
1769 | page = NULL; | |
1770 | } | |
1771 | } | |
e8d61a74 CY |
1772 | if (page) { |
1773 | set_page_dirty(page); | |
351df4b2 | 1774 | f2fs_put_page(page, 1); |
e8d61a74 | 1775 | } |
351df4b2 JK |
1776 | } |
1777 | ||
1778 | static void write_normal_summaries(struct f2fs_sb_info *sbi, | |
1779 | block_t blkaddr, int type) | |
1780 | { | |
1781 | int i, end; | |
1782 | if (IS_DATASEG(type)) | |
1783 | end = type + NR_CURSEG_DATA_TYPE; | |
1784 | else | |
1785 | end = type + NR_CURSEG_NODE_TYPE; | |
1786 | ||
b7ad7512 CY |
1787 | for (i = type; i < end; i++) |
1788 | write_current_sum_page(sbi, i, blkaddr + (i - type)); | |
351df4b2 JK |
1789 | } |
1790 | ||
1791 | void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1792 | { | |
25ca923b | 1793 | if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) |
351df4b2 JK |
1794 | write_compacted_summaries(sbi, start_blk); |
1795 | else | |
1796 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); | |
1797 | } | |
1798 | ||
1799 | void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) | |
1800 | { | |
119ee914 | 1801 | write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); |
351df4b2 JK |
1802 | } |
1803 | ||
dfc08a12 | 1804 | int lookup_journal_in_cursum(struct f2fs_journal *journal, int type, |
351df4b2 JK |
1805 | unsigned int val, int alloc) |
1806 | { | |
1807 | int i; | |
1808 | ||
1809 | if (type == NAT_JOURNAL) { | |
dfc08a12 CY |
1810 | for (i = 0; i < nats_in_cursum(journal); i++) { |
1811 | if (le32_to_cpu(nid_in_journal(journal, i)) == val) | |
351df4b2 JK |
1812 | return i; |
1813 | } | |
dfc08a12 CY |
1814 | if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) |
1815 | return update_nats_in_cursum(journal, 1); | |
351df4b2 | 1816 | } else if (type == SIT_JOURNAL) { |
dfc08a12 CY |
1817 | for (i = 0; i < sits_in_cursum(journal); i++) |
1818 | if (le32_to_cpu(segno_in_journal(journal, i)) == val) | |
351df4b2 | 1819 | return i; |
dfc08a12 CY |
1820 | if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) |
1821 | return update_sits_in_cursum(journal, 1); | |
351df4b2 JK |
1822 | } |
1823 | return -1; | |
1824 | } | |
1825 | ||
1826 | static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, | |
1827 | unsigned int segno) | |
1828 | { | |
2cc22186 | 1829 | return get_meta_page(sbi, current_sit_addr(sbi, segno)); |
351df4b2 JK |
1830 | } |
1831 | ||
1832 | static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, | |
1833 | unsigned int start) | |
1834 | { | |
1835 | struct sit_info *sit_i = SIT_I(sbi); | |
1836 | struct page *src_page, *dst_page; | |
1837 | pgoff_t src_off, dst_off; | |
1838 | void *src_addr, *dst_addr; | |
1839 | ||
1840 | src_off = current_sit_addr(sbi, start); | |
1841 | dst_off = next_sit_addr(sbi, src_off); | |
1842 | ||
1843 | /* get current sit block page without lock */ | |
1844 | src_page = get_meta_page(sbi, src_off); | |
1845 | dst_page = grab_meta_page(sbi, dst_off); | |
9850cf4a | 1846 | f2fs_bug_on(sbi, PageDirty(src_page)); |
351df4b2 JK |
1847 | |
1848 | src_addr = page_address(src_page); | |
1849 | dst_addr = page_address(dst_page); | |
09cbfeaf | 1850 | memcpy(dst_addr, src_addr, PAGE_SIZE); |
351df4b2 JK |
1851 | |
1852 | set_page_dirty(dst_page); | |
1853 | f2fs_put_page(src_page, 1); | |
1854 | ||
1855 | set_to_next_sit(sit_i, start); | |
1856 | ||
1857 | return dst_page; | |
1858 | } | |
1859 | ||
184a5cd2 CY |
1860 | static struct sit_entry_set *grab_sit_entry_set(void) |
1861 | { | |
1862 | struct sit_entry_set *ses = | |
80c54505 | 1863 | f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); |
184a5cd2 CY |
1864 | |
1865 | ses->entry_cnt = 0; | |
1866 | INIT_LIST_HEAD(&ses->set_list); | |
1867 | return ses; | |
1868 | } | |
1869 | ||
1870 | static void release_sit_entry_set(struct sit_entry_set *ses) | |
1871 | { | |
1872 | list_del(&ses->set_list); | |
1873 | kmem_cache_free(sit_entry_set_slab, ses); | |
1874 | } | |
1875 | ||
1876 | static void adjust_sit_entry_set(struct sit_entry_set *ses, | |
1877 | struct list_head *head) | |
1878 | { | |
1879 | struct sit_entry_set *next = ses; | |
1880 | ||
1881 | if (list_is_last(&ses->set_list, head)) | |
1882 | return; | |
1883 | ||
1884 | list_for_each_entry_continue(next, head, set_list) | |
1885 | if (ses->entry_cnt <= next->entry_cnt) | |
1886 | break; | |
1887 | ||
1888 | list_move_tail(&ses->set_list, &next->set_list); | |
1889 | } | |
1890 | ||
1891 | static void add_sit_entry(unsigned int segno, struct list_head *head) | |
1892 | { | |
1893 | struct sit_entry_set *ses; | |
1894 | unsigned int start_segno = START_SEGNO(segno); | |
1895 | ||
1896 | list_for_each_entry(ses, head, set_list) { | |
1897 | if (ses->start_segno == start_segno) { | |
1898 | ses->entry_cnt++; | |
1899 | adjust_sit_entry_set(ses, head); | |
1900 | return; | |
1901 | } | |
1902 | } | |
1903 | ||
1904 | ses = grab_sit_entry_set(); | |
1905 | ||
1906 | ses->start_segno = start_segno; | |
1907 | ses->entry_cnt++; | |
1908 | list_add(&ses->set_list, head); | |
1909 | } | |
1910 | ||
1911 | static void add_sits_in_set(struct f2fs_sb_info *sbi) | |
1912 | { | |
1913 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
1914 | struct list_head *set_list = &sm_info->sit_entry_set; | |
1915 | unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; | |
184a5cd2 CY |
1916 | unsigned int segno; |
1917 | ||
7cd8558b | 1918 | for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) |
184a5cd2 CY |
1919 | add_sit_entry(segno, set_list); |
1920 | } | |
1921 | ||
1922 | static void remove_sits_in_journal(struct f2fs_sb_info *sbi) | |
351df4b2 JK |
1923 | { |
1924 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
b7ad7512 | 1925 | struct f2fs_journal *journal = curseg->journal; |
351df4b2 JK |
1926 | int i; |
1927 | ||
b7ad7512 | 1928 | down_write(&curseg->journal_rwsem); |
dfc08a12 | 1929 | for (i = 0; i < sits_in_cursum(journal); i++) { |
184a5cd2 CY |
1930 | unsigned int segno; |
1931 | bool dirtied; | |
1932 | ||
dfc08a12 | 1933 | segno = le32_to_cpu(segno_in_journal(journal, i)); |
184a5cd2 CY |
1934 | dirtied = __mark_sit_entry_dirty(sbi, segno); |
1935 | ||
1936 | if (!dirtied) | |
1937 | add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); | |
351df4b2 | 1938 | } |
dfc08a12 | 1939 | update_sits_in_cursum(journal, -i); |
b7ad7512 | 1940 | up_write(&curseg->journal_rwsem); |
351df4b2 JK |
1941 | } |
1942 | ||
0a8165d7 | 1943 | /* |
351df4b2 JK |
1944 | * CP calls this function, which flushes SIT entries including sit_journal, |
1945 | * and moves prefree segs to free segs. | |
1946 | */ | |
4b2fecc8 | 1947 | void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) |
351df4b2 JK |
1948 | { |
1949 | struct sit_info *sit_i = SIT_I(sbi); | |
1950 | unsigned long *bitmap = sit_i->dirty_sentries_bitmap; | |
1951 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
b7ad7512 | 1952 | struct f2fs_journal *journal = curseg->journal; |
184a5cd2 CY |
1953 | struct sit_entry_set *ses, *tmp; |
1954 | struct list_head *head = &SM_I(sbi)->sit_entry_set; | |
184a5cd2 | 1955 | bool to_journal = true; |
4b2fecc8 | 1956 | struct seg_entry *se; |
351df4b2 | 1957 | |
351df4b2 JK |
1958 | mutex_lock(&sit_i->sentry_lock); |
1959 | ||
2b11a74b WL |
1960 | if (!sit_i->dirty_sentries) |
1961 | goto out; | |
1962 | ||
351df4b2 | 1963 | /* |
184a5cd2 CY |
1964 | * add and account sit entries of dirty bitmap in sit entry |
1965 | * set temporarily | |
351df4b2 | 1966 | */ |
184a5cd2 | 1967 | add_sits_in_set(sbi); |
351df4b2 | 1968 | |
184a5cd2 CY |
1969 | /* |
1970 | * if there are no enough space in journal to store dirty sit | |
1971 | * entries, remove all entries from journal and add and account | |
1972 | * them in sit entry set. | |
1973 | */ | |
dfc08a12 | 1974 | if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL)) |
184a5cd2 | 1975 | remove_sits_in_journal(sbi); |
b2955550 | 1976 | |
184a5cd2 CY |
1977 | /* |
1978 | * there are two steps to flush sit entries: | |
1979 | * #1, flush sit entries to journal in current cold data summary block. | |
1980 | * #2, flush sit entries to sit page. | |
1981 | */ | |
1982 | list_for_each_entry_safe(ses, tmp, head, set_list) { | |
4a257ed6 | 1983 | struct page *page = NULL; |
184a5cd2 CY |
1984 | struct f2fs_sit_block *raw_sit = NULL; |
1985 | unsigned int start_segno = ses->start_segno; | |
1986 | unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, | |
7cd8558b | 1987 | (unsigned long)MAIN_SEGS(sbi)); |
184a5cd2 CY |
1988 | unsigned int segno = start_segno; |
1989 | ||
1990 | if (to_journal && | |
dfc08a12 | 1991 | !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) |
184a5cd2 CY |
1992 | to_journal = false; |
1993 | ||
b7ad7512 CY |
1994 | if (to_journal) { |
1995 | down_write(&curseg->journal_rwsem); | |
1996 | } else { | |
184a5cd2 CY |
1997 | page = get_next_sit_page(sbi, start_segno); |
1998 | raw_sit = page_address(page); | |
351df4b2 | 1999 | } |
351df4b2 | 2000 | |
184a5cd2 CY |
2001 | /* flush dirty sit entries in region of current sit set */ |
2002 | for_each_set_bit_from(segno, bitmap, end) { | |
2003 | int offset, sit_offset; | |
4b2fecc8 JK |
2004 | |
2005 | se = get_seg_entry(sbi, segno); | |
184a5cd2 CY |
2006 | |
2007 | /* add discard candidates */ | |
d7bc2484 | 2008 | if (cpc->reason != CP_DISCARD) { |
4b2fecc8 JK |
2009 | cpc->trim_start = segno; |
2010 | add_discard_addrs(sbi, cpc); | |
2011 | } | |
184a5cd2 CY |
2012 | |
2013 | if (to_journal) { | |
dfc08a12 | 2014 | offset = lookup_journal_in_cursum(journal, |
184a5cd2 CY |
2015 | SIT_JOURNAL, segno, 1); |
2016 | f2fs_bug_on(sbi, offset < 0); | |
dfc08a12 | 2017 | segno_in_journal(journal, offset) = |
184a5cd2 CY |
2018 | cpu_to_le32(segno); |
2019 | seg_info_to_raw_sit(se, | |
dfc08a12 | 2020 | &sit_in_journal(journal, offset)); |
184a5cd2 CY |
2021 | } else { |
2022 | sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); | |
2023 | seg_info_to_raw_sit(se, | |
2024 | &raw_sit->entries[sit_offset]); | |
2025 | } | |
351df4b2 | 2026 | |
184a5cd2 CY |
2027 | __clear_bit(segno, bitmap); |
2028 | sit_i->dirty_sentries--; | |
2029 | ses->entry_cnt--; | |
351df4b2 JK |
2030 | } |
2031 | ||
b7ad7512 CY |
2032 | if (to_journal) |
2033 | up_write(&curseg->journal_rwsem); | |
2034 | else | |
184a5cd2 CY |
2035 | f2fs_put_page(page, 1); |
2036 | ||
2037 | f2fs_bug_on(sbi, ses->entry_cnt); | |
2038 | release_sit_entry_set(ses); | |
351df4b2 | 2039 | } |
184a5cd2 CY |
2040 | |
2041 | f2fs_bug_on(sbi, !list_empty(head)); | |
2042 | f2fs_bug_on(sbi, sit_i->dirty_sentries); | |
184a5cd2 | 2043 | out: |
4b2fecc8 JK |
2044 | if (cpc->reason == CP_DISCARD) { |
2045 | for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) | |
2046 | add_discard_addrs(sbi, cpc); | |
2047 | } | |
351df4b2 | 2048 | mutex_unlock(&sit_i->sentry_lock); |
351df4b2 | 2049 | |
351df4b2 JK |
2050 | set_prefree_as_free_segments(sbi); |
2051 | } | |
2052 | ||
2053 | static int build_sit_info(struct f2fs_sb_info *sbi) | |
2054 | { | |
2055 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
2056 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
2057 | struct sit_info *sit_i; | |
2058 | unsigned int sit_segs, start; | |
2059 | char *src_bitmap, *dst_bitmap; | |
2060 | unsigned int bitmap_size; | |
2061 | ||
2062 | /* allocate memory for SIT information */ | |
2063 | sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); | |
2064 | if (!sit_i) | |
2065 | return -ENOMEM; | |
2066 | ||
2067 | SM_I(sbi)->sit_info = sit_i; | |
2068 | ||
39307a8e JK |
2069 | sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) * |
2070 | sizeof(struct seg_entry), GFP_KERNEL); | |
351df4b2 JK |
2071 | if (!sit_i->sentries) |
2072 | return -ENOMEM; | |
2073 | ||
7cd8558b | 2074 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
39307a8e | 2075 | sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); |
351df4b2 JK |
2076 | if (!sit_i->dirty_sentries_bitmap) |
2077 | return -ENOMEM; | |
2078 | ||
7cd8558b | 2079 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
2080 | sit_i->sentries[start].cur_valid_map |
2081 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
2082 | sit_i->sentries[start].ckpt_valid_map | |
2083 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
a66cdd98 JK |
2084 | sit_i->sentries[start].discard_map |
2085 | = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); | |
2086 | if (!sit_i->sentries[start].cur_valid_map || | |
2087 | !sit_i->sentries[start].ckpt_valid_map || | |
2088 | !sit_i->sentries[start].discard_map) | |
351df4b2 JK |
2089 | return -ENOMEM; |
2090 | } | |
2091 | ||
60a3b782 JK |
2092 | sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); |
2093 | if (!sit_i->tmp_map) | |
2094 | return -ENOMEM; | |
2095 | ||
351df4b2 | 2096 | if (sbi->segs_per_sec > 1) { |
39307a8e JK |
2097 | sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) * |
2098 | sizeof(struct sec_entry), GFP_KERNEL); | |
351df4b2 JK |
2099 | if (!sit_i->sec_entries) |
2100 | return -ENOMEM; | |
2101 | } | |
2102 | ||
2103 | /* get information related with SIT */ | |
2104 | sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; | |
2105 | ||
2106 | /* setup SIT bitmap from ckeckpoint pack */ | |
2107 | bitmap_size = __bitmap_size(sbi, SIT_BITMAP); | |
2108 | src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); | |
2109 | ||
79b5793b | 2110 | dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); |
351df4b2 JK |
2111 | if (!dst_bitmap) |
2112 | return -ENOMEM; | |
351df4b2 JK |
2113 | |
2114 | /* init SIT information */ | |
2115 | sit_i->s_ops = &default_salloc_ops; | |
2116 | ||
2117 | sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); | |
2118 | sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; | |
2119 | sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count); | |
2120 | sit_i->sit_bitmap = dst_bitmap; | |
2121 | sit_i->bitmap_size = bitmap_size; | |
2122 | sit_i->dirty_sentries = 0; | |
2123 | sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; | |
2124 | sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); | |
2125 | sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; | |
2126 | mutex_init(&sit_i->sentry_lock); | |
2127 | return 0; | |
2128 | } | |
2129 | ||
2130 | static int build_free_segmap(struct f2fs_sb_info *sbi) | |
2131 | { | |
351df4b2 JK |
2132 | struct free_segmap_info *free_i; |
2133 | unsigned int bitmap_size, sec_bitmap_size; | |
2134 | ||
2135 | /* allocate memory for free segmap information */ | |
2136 | free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); | |
2137 | if (!free_i) | |
2138 | return -ENOMEM; | |
2139 | ||
2140 | SM_I(sbi)->free_info = free_i; | |
2141 | ||
7cd8558b | 2142 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
39307a8e | 2143 | free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL); |
351df4b2 JK |
2144 | if (!free_i->free_segmap) |
2145 | return -ENOMEM; | |
2146 | ||
7cd8558b | 2147 | sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); |
39307a8e | 2148 | free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL); |
351df4b2 JK |
2149 | if (!free_i->free_secmap) |
2150 | return -ENOMEM; | |
2151 | ||
2152 | /* set all segments as dirty temporarily */ | |
2153 | memset(free_i->free_segmap, 0xff, bitmap_size); | |
2154 | memset(free_i->free_secmap, 0xff, sec_bitmap_size); | |
2155 | ||
2156 | /* init free segmap information */ | |
7cd8558b | 2157 | free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); |
351df4b2 JK |
2158 | free_i->free_segments = 0; |
2159 | free_i->free_sections = 0; | |
1a118ccf | 2160 | spin_lock_init(&free_i->segmap_lock); |
351df4b2 JK |
2161 | return 0; |
2162 | } | |
2163 | ||
2164 | static int build_curseg(struct f2fs_sb_info *sbi) | |
2165 | { | |
1042d60f | 2166 | struct curseg_info *array; |
351df4b2 JK |
2167 | int i; |
2168 | ||
b434babf | 2169 | array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL); |
351df4b2 JK |
2170 | if (!array) |
2171 | return -ENOMEM; | |
2172 | ||
2173 | SM_I(sbi)->curseg_array = array; | |
2174 | ||
2175 | for (i = 0; i < NR_CURSEG_TYPE; i++) { | |
2176 | mutex_init(&array[i].curseg_mutex); | |
09cbfeaf | 2177 | array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL); |
351df4b2 JK |
2178 | if (!array[i].sum_blk) |
2179 | return -ENOMEM; | |
b7ad7512 CY |
2180 | init_rwsem(&array[i].journal_rwsem); |
2181 | array[i].journal = kzalloc(sizeof(struct f2fs_journal), | |
2182 | GFP_KERNEL); | |
2183 | if (!array[i].journal) | |
2184 | return -ENOMEM; | |
351df4b2 JK |
2185 | array[i].segno = NULL_SEGNO; |
2186 | array[i].next_blkoff = 0; | |
2187 | } | |
2188 | return restore_curseg_summaries(sbi); | |
2189 | } | |
2190 | ||
2191 | static void build_sit_entries(struct f2fs_sb_info *sbi) | |
2192 | { | |
2193 | struct sit_info *sit_i = SIT_I(sbi); | |
2194 | struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); | |
b7ad7512 | 2195 | struct f2fs_journal *journal = curseg->journal; |
74de593a CY |
2196 | int sit_blk_cnt = SIT_BLK_CNT(sbi); |
2197 | unsigned int i, start, end; | |
2198 | unsigned int readed, start_blk = 0; | |
e9f5b8b8 | 2199 | int nrpages = MAX_BIO_BLOCKS(sbi) * 8; |
351df4b2 | 2200 | |
74de593a | 2201 | do { |
26879fb1 | 2202 | readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true); |
74de593a CY |
2203 | |
2204 | start = start_blk * sit_i->sents_per_block; | |
2205 | end = (start_blk + readed) * sit_i->sents_per_block; | |
2206 | ||
7cd8558b | 2207 | for (; start < end && start < MAIN_SEGS(sbi); start++) { |
74de593a CY |
2208 | struct seg_entry *se = &sit_i->sentries[start]; |
2209 | struct f2fs_sit_block *sit_blk; | |
2210 | struct f2fs_sit_entry sit; | |
2211 | struct page *page; | |
2212 | ||
b7ad7512 | 2213 | down_read(&curseg->journal_rwsem); |
dfc08a12 CY |
2214 | for (i = 0; i < sits_in_cursum(journal); i++) { |
2215 | if (le32_to_cpu(segno_in_journal(journal, i)) | |
6c311ec6 | 2216 | == start) { |
dfc08a12 | 2217 | sit = sit_in_journal(journal, i); |
b7ad7512 | 2218 | up_read(&curseg->journal_rwsem); |
74de593a CY |
2219 | goto got_it; |
2220 | } | |
351df4b2 | 2221 | } |
b7ad7512 | 2222 | up_read(&curseg->journal_rwsem); |
74de593a CY |
2223 | |
2224 | page = get_current_sit_page(sbi, start); | |
2225 | sit_blk = (struct f2fs_sit_block *)page_address(page); | |
2226 | sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; | |
2227 | f2fs_put_page(page, 1); | |
351df4b2 | 2228 | got_it: |
74de593a CY |
2229 | check_block_count(sbi, start, &sit); |
2230 | seg_info_from_raw_sit(se, &sit); | |
a66cdd98 JK |
2231 | |
2232 | /* build discard map only one time */ | |
2233 | memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE); | |
2234 | sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks; | |
2235 | ||
74de593a CY |
2236 | if (sbi->segs_per_sec > 1) { |
2237 | struct sec_entry *e = get_sec_entry(sbi, start); | |
2238 | e->valid_blocks += se->valid_blocks; | |
2239 | } | |
351df4b2 | 2240 | } |
74de593a CY |
2241 | start_blk += readed; |
2242 | } while (start_blk < sit_blk_cnt); | |
351df4b2 JK |
2243 | } |
2244 | ||
2245 | static void init_free_segmap(struct f2fs_sb_info *sbi) | |
2246 | { | |
2247 | unsigned int start; | |
2248 | int type; | |
2249 | ||
7cd8558b | 2250 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
2251 | struct seg_entry *sentry = get_seg_entry(sbi, start); |
2252 | if (!sentry->valid_blocks) | |
2253 | __set_free(sbi, start); | |
2254 | } | |
2255 | ||
2256 | /* set use the current segments */ | |
2257 | for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { | |
2258 | struct curseg_info *curseg_t = CURSEG_I(sbi, type); | |
2259 | __set_test_and_inuse(sbi, curseg_t->segno); | |
2260 | } | |
2261 | } | |
2262 | ||
2263 | static void init_dirty_segmap(struct f2fs_sb_info *sbi) | |
2264 | { | |
2265 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
2266 | struct free_segmap_info *free_i = FREE_I(sbi); | |
7cd8558b | 2267 | unsigned int segno = 0, offset = 0; |
351df4b2 JK |
2268 | unsigned short valid_blocks; |
2269 | ||
8736fbf0 | 2270 | while (1) { |
351df4b2 | 2271 | /* find dirty segment based on free segmap */ |
7cd8558b JK |
2272 | segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); |
2273 | if (segno >= MAIN_SEGS(sbi)) | |
351df4b2 JK |
2274 | break; |
2275 | offset = segno + 1; | |
2276 | valid_blocks = get_valid_blocks(sbi, segno, 0); | |
ec325b52 | 2277 | if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) |
351df4b2 | 2278 | continue; |
ec325b52 JK |
2279 | if (valid_blocks > sbi->blocks_per_seg) { |
2280 | f2fs_bug_on(sbi, 1); | |
2281 | continue; | |
2282 | } | |
351df4b2 JK |
2283 | mutex_lock(&dirty_i->seglist_lock); |
2284 | __locate_dirty_segment(sbi, segno, DIRTY); | |
2285 | mutex_unlock(&dirty_i->seglist_lock); | |
2286 | } | |
2287 | } | |
2288 | ||
5ec4e49f | 2289 | static int init_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
2290 | { |
2291 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
7cd8558b | 2292 | unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); |
351df4b2 | 2293 | |
39307a8e | 2294 | dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); |
5ec4e49f | 2295 | if (!dirty_i->victim_secmap) |
351df4b2 JK |
2296 | return -ENOMEM; |
2297 | return 0; | |
2298 | } | |
2299 | ||
2300 | static int build_dirty_segmap(struct f2fs_sb_info *sbi) | |
2301 | { | |
2302 | struct dirty_seglist_info *dirty_i; | |
2303 | unsigned int bitmap_size, i; | |
2304 | ||
2305 | /* allocate memory for dirty segments list information */ | |
2306 | dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); | |
2307 | if (!dirty_i) | |
2308 | return -ENOMEM; | |
2309 | ||
2310 | SM_I(sbi)->dirty_info = dirty_i; | |
2311 | mutex_init(&dirty_i->seglist_lock); | |
2312 | ||
7cd8558b | 2313 | bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); |
351df4b2 JK |
2314 | |
2315 | for (i = 0; i < NR_DIRTY_TYPE; i++) { | |
39307a8e | 2316 | dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); |
351df4b2 JK |
2317 | if (!dirty_i->dirty_segmap[i]) |
2318 | return -ENOMEM; | |
2319 | } | |
2320 | ||
2321 | init_dirty_segmap(sbi); | |
5ec4e49f | 2322 | return init_victim_secmap(sbi); |
351df4b2 JK |
2323 | } |
2324 | ||
0a8165d7 | 2325 | /* |
351df4b2 JK |
2326 | * Update min, max modified time for cost-benefit GC algorithm |
2327 | */ | |
2328 | static void init_min_max_mtime(struct f2fs_sb_info *sbi) | |
2329 | { | |
2330 | struct sit_info *sit_i = SIT_I(sbi); | |
2331 | unsigned int segno; | |
2332 | ||
2333 | mutex_lock(&sit_i->sentry_lock); | |
2334 | ||
2335 | sit_i->min_mtime = LLONG_MAX; | |
2336 | ||
7cd8558b | 2337 | for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { |
351df4b2 JK |
2338 | unsigned int i; |
2339 | unsigned long long mtime = 0; | |
2340 | ||
2341 | for (i = 0; i < sbi->segs_per_sec; i++) | |
2342 | mtime += get_seg_entry(sbi, segno + i)->mtime; | |
2343 | ||
2344 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
2345 | ||
2346 | if (sit_i->min_mtime > mtime) | |
2347 | sit_i->min_mtime = mtime; | |
2348 | } | |
2349 | sit_i->max_mtime = get_mtime(sbi); | |
2350 | mutex_unlock(&sit_i->sentry_lock); | |
2351 | } | |
2352 | ||
2353 | int build_segment_manager(struct f2fs_sb_info *sbi) | |
2354 | { | |
2355 | struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); | |
2356 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
1042d60f | 2357 | struct f2fs_sm_info *sm_info; |
351df4b2 JK |
2358 | int err; |
2359 | ||
2360 | sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); | |
2361 | if (!sm_info) | |
2362 | return -ENOMEM; | |
2363 | ||
2364 | /* init sm info */ | |
2365 | sbi->sm_info = sm_info; | |
351df4b2 JK |
2366 | sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); |
2367 | sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); | |
2368 | sm_info->segment_count = le32_to_cpu(raw_super->segment_count); | |
2369 | sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); | |
2370 | sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); | |
2371 | sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); | |
2372 | sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); | |
58c41035 JK |
2373 | sm_info->rec_prefree_segments = sm_info->main_segments * |
2374 | DEF_RECLAIM_PREFREE_SEGMENTS / 100; | |
9b5f136f | 2375 | sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; |
216fbd64 | 2376 | sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; |
c1ce1b02 | 2377 | sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; |
351df4b2 | 2378 | |
7fd9e544 JK |
2379 | INIT_LIST_HEAD(&sm_info->discard_list); |
2380 | sm_info->nr_discards = 0; | |
2381 | sm_info->max_discards = 0; | |
2382 | ||
bba681cb JK |
2383 | sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS; |
2384 | ||
184a5cd2 CY |
2385 | INIT_LIST_HEAD(&sm_info->sit_entry_set); |
2386 | ||
b270ad6f | 2387 | if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) { |
2163d198 GZ |
2388 | err = create_flush_cmd_control(sbi); |
2389 | if (err) | |
a688b9d9 | 2390 | return err; |
6b4afdd7 JK |
2391 | } |
2392 | ||
351df4b2 JK |
2393 | err = build_sit_info(sbi); |
2394 | if (err) | |
2395 | return err; | |
2396 | err = build_free_segmap(sbi); | |
2397 | if (err) | |
2398 | return err; | |
2399 | err = build_curseg(sbi); | |
2400 | if (err) | |
2401 | return err; | |
2402 | ||
2403 | /* reinit free segmap based on SIT */ | |
2404 | build_sit_entries(sbi); | |
2405 | ||
2406 | init_free_segmap(sbi); | |
2407 | err = build_dirty_segmap(sbi); | |
2408 | if (err) | |
2409 | return err; | |
2410 | ||
2411 | init_min_max_mtime(sbi); | |
2412 | return 0; | |
2413 | } | |
2414 | ||
2415 | static void discard_dirty_segmap(struct f2fs_sb_info *sbi, | |
2416 | enum dirty_type dirty_type) | |
2417 | { | |
2418 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
2419 | ||
2420 | mutex_lock(&dirty_i->seglist_lock); | |
39307a8e | 2421 | kvfree(dirty_i->dirty_segmap[dirty_type]); |
351df4b2 JK |
2422 | dirty_i->nr_dirty[dirty_type] = 0; |
2423 | mutex_unlock(&dirty_i->seglist_lock); | |
2424 | } | |
2425 | ||
5ec4e49f | 2426 | static void destroy_victim_secmap(struct f2fs_sb_info *sbi) |
351df4b2 JK |
2427 | { |
2428 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
39307a8e | 2429 | kvfree(dirty_i->victim_secmap); |
351df4b2 JK |
2430 | } |
2431 | ||
2432 | static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) | |
2433 | { | |
2434 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
2435 | int i; | |
2436 | ||
2437 | if (!dirty_i) | |
2438 | return; | |
2439 | ||
2440 | /* discard pre-free/dirty segments list */ | |
2441 | for (i = 0; i < NR_DIRTY_TYPE; i++) | |
2442 | discard_dirty_segmap(sbi, i); | |
2443 | ||
5ec4e49f | 2444 | destroy_victim_secmap(sbi); |
351df4b2 JK |
2445 | SM_I(sbi)->dirty_info = NULL; |
2446 | kfree(dirty_i); | |
2447 | } | |
2448 | ||
2449 | static void destroy_curseg(struct f2fs_sb_info *sbi) | |
2450 | { | |
2451 | struct curseg_info *array = SM_I(sbi)->curseg_array; | |
2452 | int i; | |
2453 | ||
2454 | if (!array) | |
2455 | return; | |
2456 | SM_I(sbi)->curseg_array = NULL; | |
b7ad7512 | 2457 | for (i = 0; i < NR_CURSEG_TYPE; i++) { |
351df4b2 | 2458 | kfree(array[i].sum_blk); |
b7ad7512 CY |
2459 | kfree(array[i].journal); |
2460 | } | |
351df4b2 JK |
2461 | kfree(array); |
2462 | } | |
2463 | ||
2464 | static void destroy_free_segmap(struct f2fs_sb_info *sbi) | |
2465 | { | |
2466 | struct free_segmap_info *free_i = SM_I(sbi)->free_info; | |
2467 | if (!free_i) | |
2468 | return; | |
2469 | SM_I(sbi)->free_info = NULL; | |
39307a8e JK |
2470 | kvfree(free_i->free_segmap); |
2471 | kvfree(free_i->free_secmap); | |
351df4b2 JK |
2472 | kfree(free_i); |
2473 | } | |
2474 | ||
2475 | static void destroy_sit_info(struct f2fs_sb_info *sbi) | |
2476 | { | |
2477 | struct sit_info *sit_i = SIT_I(sbi); | |
2478 | unsigned int start; | |
2479 | ||
2480 | if (!sit_i) | |
2481 | return; | |
2482 | ||
2483 | if (sit_i->sentries) { | |
7cd8558b | 2484 | for (start = 0; start < MAIN_SEGS(sbi); start++) { |
351df4b2 JK |
2485 | kfree(sit_i->sentries[start].cur_valid_map); |
2486 | kfree(sit_i->sentries[start].ckpt_valid_map); | |
a66cdd98 | 2487 | kfree(sit_i->sentries[start].discard_map); |
351df4b2 JK |
2488 | } |
2489 | } | |
60a3b782 JK |
2490 | kfree(sit_i->tmp_map); |
2491 | ||
39307a8e JK |
2492 | kvfree(sit_i->sentries); |
2493 | kvfree(sit_i->sec_entries); | |
2494 | kvfree(sit_i->dirty_sentries_bitmap); | |
351df4b2 JK |
2495 | |
2496 | SM_I(sbi)->sit_info = NULL; | |
2497 | kfree(sit_i->sit_bitmap); | |
2498 | kfree(sit_i); | |
2499 | } | |
2500 | ||
2501 | void destroy_segment_manager(struct f2fs_sb_info *sbi) | |
2502 | { | |
2503 | struct f2fs_sm_info *sm_info = SM_I(sbi); | |
a688b9d9 | 2504 | |
3b03f724 CY |
2505 | if (!sm_info) |
2506 | return; | |
2163d198 | 2507 | destroy_flush_cmd_control(sbi); |
351df4b2 JK |
2508 | destroy_dirty_segmap(sbi); |
2509 | destroy_curseg(sbi); | |
2510 | destroy_free_segmap(sbi); | |
2511 | destroy_sit_info(sbi); | |
2512 | sbi->sm_info = NULL; | |
2513 | kfree(sm_info); | |
2514 | } | |
7fd9e544 JK |
2515 | |
2516 | int __init create_segment_manager_caches(void) | |
2517 | { | |
2518 | discard_entry_slab = f2fs_kmem_cache_create("discard_entry", | |
e8512d2e | 2519 | sizeof(struct discard_entry)); |
7fd9e544 | 2520 | if (!discard_entry_slab) |
184a5cd2 CY |
2521 | goto fail; |
2522 | ||
2523 | sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", | |
c9ee0085 | 2524 | sizeof(struct sit_entry_set)); |
184a5cd2 CY |
2525 | if (!sit_entry_set_slab) |
2526 | goto destory_discard_entry; | |
88b88a66 JK |
2527 | |
2528 | inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", | |
2529 | sizeof(struct inmem_pages)); | |
2530 | if (!inmem_entry_slab) | |
2531 | goto destroy_sit_entry_set; | |
7fd9e544 | 2532 | return 0; |
184a5cd2 | 2533 | |
88b88a66 JK |
2534 | destroy_sit_entry_set: |
2535 | kmem_cache_destroy(sit_entry_set_slab); | |
184a5cd2 CY |
2536 | destory_discard_entry: |
2537 | kmem_cache_destroy(discard_entry_slab); | |
2538 | fail: | |
2539 | return -ENOMEM; | |
7fd9e544 JK |
2540 | } |
2541 | ||
2542 | void destroy_segment_manager_caches(void) | |
2543 | { | |
184a5cd2 | 2544 | kmem_cache_destroy(sit_entry_set_slab); |
7fd9e544 | 2545 | kmem_cache_destroy(discard_entry_slab); |
88b88a66 | 2546 | kmem_cache_destroy(inmem_entry_slab); |
7fd9e544 | 2547 | } |