]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: shrink nat_cache entries
[mirror_ubuntu-artful-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
351df4b2 17#include <linux/vmalloc.h>
74de593a 18#include <linux/swap.h>
351df4b2
JK		 19
20#include "f2fs.h"
21#include "segment.h"
22#include "node.h"
9e4ded3f 23#include "trace.h"
6ec178da 24#include <trace/events/f2fs.h>
351df4b2 25
9a7f143a
CL		 26#define __reverse_ffz(x) __reverse_ffs(~(x))
27
7fd9e544 28static struct kmem_cache *discard_entry_slab;
184a5cd2 29static struct kmem_cache *sit_entry_set_slab;
88b88a66 30static struct kmem_cache *inmem_entry_slab;
7fd9e544 31
9a7f143a
CL		 32/*
33 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
34 * MSB and LSB are reversed in a byte by f2fs_set_bit.
35 */
36static inline unsigned long __reverse_ffs(unsigned long word)
37{
38 int num = 0;
39
40#if BITS_PER_LONG == 64
41 if ((word & 0xffffffff) == 0) {
42 num += 32;
43 word >>= 32;
44 }
45#endif
46 if ((word & 0xffff) == 0) {
47 num += 16;
48 word >>= 16;
49 }
50 if ((word & 0xff) == 0) {
51 num += 8;
52 word >>= 8;
53 }
54 if ((word & 0xf0) == 0)
55 num += 4;
56 else
57 word >>= 4;
58 if ((word & 0xc) == 0)
59 num += 2;
60 else
61 word >>= 2;
62 if ((word & 0x2) == 0)
63 num += 1;
64 return num;
65}
66
67/*
e1c42045 68 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a
CL		 69 * f2fs_set_bit makes MSB and LSB reversed in a byte.
70 * Example:
71 * LSB <--> MSB
72 * f2fs_set_bit(0, bitmap) => 0000 0001
73 * f2fs_set_bit(7, bitmap) => 1000 0000
74 */
75static unsigned long __find_rev_next_bit(const unsigned long *addr,
76 unsigned long size, unsigned long offset)
77{
e19ef527
JK		 78 while (!f2fs_test_bit(offset, (unsigned char *)addr))
79 offset++;
80
81 if (offset > size)
82 offset = size;
83
84 return offset;
85#if 0
9a7f143a
CL		 86 const unsigned long *p = addr + BIT_WORD(offset);
87 unsigned long result = offset & ~(BITS_PER_LONG - 1);
88 unsigned long tmp;
89 unsigned long mask, submask;
90 unsigned long quot, rest;
91
92 if (offset >= size)
93 return size;
94
95 size -= result;
96 offset %= BITS_PER_LONG;
97 if (!offset)
98 goto aligned;
99
100 tmp = *(p++);
101 quot = (offset >> 3) << 3;
102 rest = offset & 0x7;
103 mask = ~0UL << quot;
104 submask = (unsigned char)(0xff << rest) >> rest;
105 submask <<= quot;
106 mask &= submask;
107 tmp &= mask;
108 if (size < BITS_PER_LONG)
109 goto found_first;
110 if (tmp)
111 goto found_middle;
112
113 size -= BITS_PER_LONG;
114 result += BITS_PER_LONG;
115aligned:
116 while (size & ~(BITS_PER_LONG-1)) {
117 tmp = *(p++);
118 if (tmp)
119 goto found_middle;
120 result += BITS_PER_LONG;
121 size -= BITS_PER_LONG;
122 }
123 if (!size)
124 return result;
125 tmp = *p;
126found_first:
127 tmp &= (~0UL >> (BITS_PER_LONG - size));
128 if (tmp == 0UL) /* Are any bits set? */
129 return result + size; /* Nope. */
130found_middle:
131 return result + __reverse_ffs(tmp);
e19ef527 132#endif
9a7f143a
CL		 133}
134
135static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137{
e19ef527
JK		 138 while (f2fs_test_bit(offset, (unsigned char *)addr))
139 offset++;
140
141 if (offset > size)
142 offset = size;
143
144 return offset;
145#if 0
9a7f143a
CL		 146 const unsigned long *p = addr + BIT_WORD(offset);
147 unsigned long result = offset & ~(BITS_PER_LONG - 1);
148 unsigned long tmp;
149 unsigned long mask, submask;
150 unsigned long quot, rest;
151
152 if (offset >= size)
153 return size;
154
155 size -= result;
156 offset %= BITS_PER_LONG;
157 if (!offset)
158 goto aligned;
159
160 tmp = *(p++);
161 quot = (offset >> 3) << 3;
162 rest = offset & 0x7;
163 mask = ~(~0UL << quot);
164 submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
165 submask <<= quot;
166 mask += submask;
167 tmp |= mask;
168 if (size < BITS_PER_LONG)
169 goto found_first;
170 if (~tmp)
171 goto found_middle;
172
173 size -= BITS_PER_LONG;
174 result += BITS_PER_LONG;
175aligned:
176 while (size & ~(BITS_PER_LONG - 1)) {
177 tmp = *(p++);
178 if (~tmp)
179 goto found_middle;
180 result += BITS_PER_LONG;
181 size -= BITS_PER_LONG;
182 }
183 if (!size)
184 return result;
185 tmp = *p;
186
187found_first:
188 tmp |= ~0UL << size;
189 if (tmp == ~0UL) /* Are any bits zero? */
190 return result + size; /* Nope. */
191found_middle:
192 return result + __reverse_ffz(tmp);
e19ef527 193#endif
9a7f143a
CL		 194}
195
88b88a66
JK		 196void register_inmem_page(struct inode *inode, struct page *page)
197{
198 struct f2fs_inode_info *fi = F2FS_I(inode);
199 struct inmem_pages *new;
34ba94ba 200 int err;
9be32d72 201
0722b101 202 SetPagePrivate(page);
9e4ded3f 203 f2fs_trace_pid(page);
0722b101 204
88b88a66
JK		 205 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
206
207 /* add atomic page indices to the list */
208 new->page = page;
209 INIT_LIST_HEAD(&new->list);
9be32d72 210retry:
88b88a66
JK		 211 /* increase reference count with clean state */
212 mutex_lock(&fi->inmem_lock);
34ba94ba
JK		 213 err = radix_tree_insert(&fi->inmem_root, page->index, new);
214 if (err == -EEXIST) {
215 mutex_unlock(&fi->inmem_lock);
216 kmem_cache_free(inmem_entry_slab, new);
217 return;
218 } else if (err) {
219 mutex_unlock(&fi->inmem_lock);
34ba94ba
JK		 220 goto retry;
221 }
88b88a66
JK		 222 get_page(page);
223 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 224 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 225 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 226
227 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 228}
229
230void commit_inmem_pages(struct inode *inode, bool abort)
231{
232 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
233 struct f2fs_inode_info *fi = F2FS_I(inode);
234 struct inmem_pages *cur, *tmp;
235 bool submit_bio = false;
236 struct f2fs_io_info fio = {
05ca3632 237 .sbi = sbi,
88b88a66 238 .type = DATA,
1e84371f 239 .rw = WRITE_SYNC | REQ_PRIO,
4375a336 240 .encrypted_page = NULL,
88b88a66
JK		 241 };
242
0341845e
JK		 243 /*
244 * The abort is true only when f2fs_evict_inode is called.
245 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
246 * that we don't need to call f2fs_balance_fs.
247 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
248 * inode becomes free by iget_locked in f2fs_iget.
249 */
70c640b1 250 if (!abort) {
0341845e 251 f2fs_balance_fs(sbi);
70c640b1
JK		 252 f2fs_lock_op(sbi);
253 }
88b88a66
JK		 254
255 mutex_lock(&fi->inmem_lock);
256 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
70c640b1
JK		 257 if (!abort) {
258 lock_page(cur->page);
259 if (cur->page->mapping == inode->i_mapping) {
6282adbf 260 set_page_dirty(cur->page);
70c640b1
JK		 261 f2fs_wait_on_page_writeback(cur->page, DATA);
262 if (clear_page_dirty_for_io(cur->page))
263 inode_dec_dirty_pages(inode);
8ce67cb0 264 trace_f2fs_commit_inmem_page(cur->page, INMEM);
05ca3632
JK		 265 fio.page = cur->page;
266 do_write_data_page(&fio);
70c640b1
JK		 267 submit_bio = true;
268 }
269 f2fs_put_page(cur->page, 1);
270 } else {
8ce67cb0 271 trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
70c640b1 272 put_page(cur->page);
88b88a66 273 }
34ba94ba 274 radix_tree_delete(&fi->inmem_root, cur->page->index);
88b88a66
JK		 275 list_del(&cur->list);
276 kmem_cache_free(inmem_entry_slab, cur);
8dcf2ff7 277 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 278 }
88b88a66
JK		 279 mutex_unlock(&fi->inmem_lock);
280
70c640b1
JK		 281 if (!abort) {
282 f2fs_unlock_op(sbi);
283 if (submit_bio)
284 f2fs_submit_merged_bio(sbi, DATA, WRITE);
285 }
88b88a66
JK		 286}
287
0a8165d7 288/*
351df4b2
JK		 289 * This function balances dirty node and dentry pages.
290 * In addition, it controls garbage collection.
291 */
292void f2fs_balance_fs(struct f2fs_sb_info *sbi)
293{
351df4b2 294 /*
029cd28c
JK		 295 * We should do GC or end up with checkpoint, if there are so many dirty
296 * dir/node pages without enough free segments.
351df4b2 297 */
43727527 298 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 299 mutex_lock(&sbi->gc_mutex);
408e9375 300 f2fs_gc(sbi);
351df4b2
JK		 301 }
302}
303
4660f9c0
JK		 304void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
305{
1dcc336b
CY		 306 /* try to shrink extent cache when there is no enough memory */
307 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
308
1b38dc8e
JK		 309 /* check the # of cached NAT entries */
310 if (!available_free_memory(sbi, NAT_ENTRIES))
311 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
312
313 /* checkpoint is the only way to shrink partial cached entries */
314 if (!available_free_memory(sbi, NAT_ENTRIES) ||
e5e7ea3c 315 excess_prefree_segs(sbi) ||
88a70a69 316 !available_free_memory(sbi, INO_ENTRIES))
4660f9c0
JK		 317 f2fs_sync_fs(sbi->sb, true);
318}
319
2163d198 320static int issue_flush_thread(void *data)
6b4afdd7
JK		 321{
322 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 323 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
324 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 325repeat:
326 if (kthread_should_stop())
327 return 0;
328
721bd4d5 329 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 330 struct bio *bio = bio_alloc(GFP_NOIO, 0);
331 struct flush_cmd *cmd, *next;
332 int ret;
333
721bd4d5
GZ		 334 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
335 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
336
6b4afdd7
JK		 337 bio->bi_bdev = sbi->sb->s_bdev;
338 ret = submit_bio_wait(WRITE_FLUSH, bio);
339
721bd4d5
GZ		 340 llist_for_each_entry_safe(cmd, next,
341 fcc->dispatch_list, llnode) {
6b4afdd7 342 cmd->ret = ret;
6b4afdd7
JK		 343 complete(&cmd->wait);
344 }
a4ed23f2 345 bio_put(bio);
a688b9d9 346 fcc->dispatch_list = NULL;
6b4afdd7
JK		 347 }
348
a688b9d9 349 wait_event_interruptible(*q,
721bd4d5 350 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 351 goto repeat;
352}
353
354int f2fs_issue_flush(struct f2fs_sb_info *sbi)
355{
a688b9d9 356 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 357 struct flush_cmd cmd;
6b4afdd7 358
24a9ee0f
JK		 359 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
360 test_opt(sbi, FLUSH_MERGE));
361
0f7b2abd
JK		 362 if (test_opt(sbi, NOBARRIER))
363 return 0;
364
6b4afdd7
JK		 365 if (!test_opt(sbi, FLUSH_MERGE))
366 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
367
adf8d90b 368 init_completion(&cmd.wait);
6b4afdd7 369
721bd4d5 370 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 371
a688b9d9
GZ		 372 if (!fcc->dispatch_list)
373 wake_up(&fcc->flush_wait_queue);
6b4afdd7 374
adf8d90b
CY		 375 wait_for_completion(&cmd.wait);
376
377 return cmd.ret;
6b4afdd7
JK		 378}
379
2163d198
GZ		 380int create_flush_cmd_control(struct f2fs_sb_info *sbi)
381{
382 dev_t dev = sbi->sb->s_bdev->bd_dev;
383 struct flush_cmd_control *fcc;
384 int err = 0;
385
386 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
387 if (!fcc)
388 return -ENOMEM;
2163d198 389 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 390 init_llist_head(&fcc->issue_list);
6b2920a5 391 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 392 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
393 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
394 if (IS_ERR(fcc->f2fs_issue_flush)) {
395 err = PTR_ERR(fcc->f2fs_issue_flush);
396 kfree(fcc);
6b2920a5 397 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 398 return err;
399 }
2163d198
GZ		 400
401 return err;
402}
403
404void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
405{
6b2920a5 406 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 407
408 if (fcc && fcc->f2fs_issue_flush)
409 kthread_stop(fcc->f2fs_issue_flush);
410 kfree(fcc);
6b2920a5 411 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 412}
413
351df4b2
JK		 414static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
415 enum dirty_type dirty_type)
416{
417 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
418
419 /* need not be added */
420 if (IS_CURSEG(sbi, segno))
421 return;
422
423 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
424 dirty_i->nr_dirty[dirty_type]++;
425
426 if (dirty_type == DIRTY) {
427 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 428 enum dirty_type t = sentry->type;
b2f2c390 429
ec325b52
JK		 430 if (unlikely(t >= DIRTY)) {
431 f2fs_bug_on(sbi, 1);
432 return;
433 }
4625d6aa
CL		 434 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
435 dirty_i->nr_dirty[t]++;
351df4b2
JK		 436 }
437}
438
439static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
440 enum dirty_type dirty_type)
441{
442 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
443
444 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
445 dirty_i->nr_dirty[dirty_type]--;
446
447 if (dirty_type == DIRTY) {
4625d6aa
CL		 448 struct seg_entry *sentry = get_seg_entry(sbi, segno);
449 enum dirty_type t = sentry->type;
450
451 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
452 dirty_i->nr_dirty[t]--;
b2f2c390 453
5ec4e49f
JK		 454 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
455 clear_bit(GET_SECNO(sbi, segno),
456 dirty_i->victim_secmap);
351df4b2
JK		 457 }
458}
459
0a8165d7 460/*
351df4b2
JK		 461 * Should not occur error such as -ENOMEM.
462 * Adding dirty entry into seglist is not critical operation.
463 * If a given segment is one of current working segments, it won't be added.
464 */
8d8451af 465static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 466{
467 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
468 unsigned short valid_blocks;
469
470 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
471 return;
472
473 mutex_lock(&dirty_i->seglist_lock);
474
475 valid_blocks = get_valid_blocks(sbi, segno, 0);
476
477 if (valid_blocks == 0) {
478 __locate_dirty_segment(sbi, segno, PRE);
479 __remove_dirty_segment(sbi, segno, DIRTY);
480 } else if (valid_blocks < sbi->blocks_per_seg) {
481 __locate_dirty_segment(sbi, segno, DIRTY);
482 } else {
483 /* Recovery routine with SSR needs this */
484 __remove_dirty_segment(sbi, segno, DIRTY);
485 }
486
487 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 488}
489
1e87a78d 490static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 491 block_t blkstart, block_t blklen)
492{
55cf9cb6
CY		 493 sector_t start = SECTOR_FROM_BLOCK(blkstart);
494 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 495 struct seg_entry *se;
496 unsigned int offset;
497 block_t i;
498
499 for (i = blkstart; i < blkstart + blklen; i++) {
500 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
501 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
502
503 if (!f2fs_test_and_set_bit(offset, se->discard_map))
504 sbi->discard_blks--;
505 }
1661d07c 506 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 507 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
508}
509
cf2271e7 510void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 511{
40a02be1
JK		 512 int err = -ENOTSUPP;
513
514 if (test_opt(sbi, DISCARD)) {
515 struct seg_entry *se = get_seg_entry(sbi,
516 GET_SEGNO(sbi, blkaddr));
517 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
518
519 if (f2fs_test_bit(offset, se->discard_map))
520 return;
521
522 err = f2fs_issue_discard(sbi, blkaddr, 1);
523 }
524
381722d2
CY		 525 if (err)
526 update_meta_page(sbi, NULL, blkaddr);
37208879
JK		 527}
528
adf4983b 529static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 530 struct cp_control *cpc, struct seg_entry *se,
531 unsigned int start, unsigned int end)
b2955550
JK		 532{
533 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 534 struct discard_entry *new, *last;
535
536 if (!list_empty(head)) {
537 last = list_last_entry(head, struct discard_entry, list);
538 if (START_BLOCK(sbi, cpc->trim_start) + start ==
539 last->blkaddr + last->len) {
540 last->len += end - start;
541 goto done;
542 }
543 }
544
545 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
546 INIT_LIST_HEAD(&new->list);
547 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
548 new->len = end - start;
549 list_add_tail(&new->list, head);
550done:
551 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 552}
553
554static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
555{
b2955550
JK		 556 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
557 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 558 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 559 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
560 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 561 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 562 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 563 unsigned int start = 0, end = -1;
4b2fecc8 564 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 565 int i;
566
a66cdd98 567 if (se->valid_blocks == max_blocks)
b2955550
JK		 568 return;
569
a66cdd98
JK		 570 if (!force) {
571 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 572 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
573 return;
4b2fecc8
JK		 574 }
575
b2955550
JK		 576 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
577 for (i = 0; i < entries; i++)
a66cdd98 578 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 579 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 580
4b2fecc8 581 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 582 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
583 if (start >= max_blocks)
584 break;
585
586 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 587 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 588 }
589}
590
4b2fecc8
JK		 591void release_discard_addrs(struct f2fs_sb_info *sbi)
592{
593 struct list_head *head = &(SM_I(sbi)->discard_list);
594 struct discard_entry *entry, *this;
595
596 /* drop caches */
597 list_for_each_entry_safe(entry, this, head, list) {
598 list_del(&entry->list);
599 kmem_cache_free(discard_entry_slab, entry);
600 }
601}
602
0a8165d7 603/*
351df4b2
JK		 604 * Should call clear_prefree_segments after checkpoint is done.
605 */
606static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
607{
608 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 609 unsigned int segno;
351df4b2
JK		 610
611 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 612 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 613 __set_test_and_free(sbi, segno);
351df4b2
JK		 614 mutex_unlock(&dirty_i->seglist_lock);
615}
616
836b5a63 617void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 618{
b2955550 619 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 620 struct discard_entry *entry, *this;
351df4b2 621 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 622 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 623 unsigned int start = 0, end = -1;
351df4b2
JK		 624
625 mutex_lock(&dirty_i->seglist_lock);
29e59c14 626
351df4b2 627 while (1) {
29e59c14 628 int i;
7cd8558b
JK		 629 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
630 if (start >= MAIN_SEGS(sbi))
351df4b2 631 break;
7cd8558b
JK		 632 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
633 start + 1);
29e59c14
CL		 634
635 for (i = start; i < end; i++)
636 clear_bit(i, prefree_map);
637
638 dirty_i->nr_dirty[PRE] -= end - start;
639
640 if (!test_opt(sbi, DISCARD))
641 continue;
351df4b2 642
37208879
JK		 643 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
644 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 645 }
646 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 647
648 /* send small discards */
2d7b822a 649 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 650 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
651 goto skip;
37208879 652 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 653 cpc->trimmed += entry->len;
836b5a63 654skip:
b2955550
JK		 655 list_del(&entry->list);
656 SM_I(sbi)->nr_discards -= entry->len;
657 kmem_cache_free(discard_entry_slab, entry);
658 }
351df4b2
JK		 659}
660
184a5cd2 661static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 662{
663 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 664
665 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 666 sit_i->dirty_sentries++;
184a5cd2
CY		 667 return false;
668 }
669
670 return true;
351df4b2
JK		 671}
672
673static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
674 unsigned int segno, int modified)
675{
676 struct seg_entry *se = get_seg_entry(sbi, segno);
677 se->type = type;
678 if (modified)
679 __mark_sit_entry_dirty(sbi, segno);
680}
681
682static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
683{
684 struct seg_entry *se;
685 unsigned int segno, offset;
686 long int new_vblocks;
687
688 segno = GET_SEGNO(sbi, blkaddr);
689
690 se = get_seg_entry(sbi, segno);
691 new_vblocks = se->valid_blocks + del;
491c0854 692 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 693
9850cf4a 694 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 695 (new_vblocks > sbi->blocks_per_seg)));
696
697 se->valid_blocks = new_vblocks;
698 se->mtime = get_mtime(sbi);
699 SIT_I(sbi)->max_mtime = se->mtime;
700
701 /* Update valid block bitmap */
702 if (del > 0) {
52aca074 703 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 704 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 705 if (!f2fs_test_and_set_bit(offset, se->discard_map))
706 sbi->discard_blks--;
351df4b2 707 } else {
52aca074 708 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 709 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 710 if (f2fs_test_and_clear_bit(offset, se->discard_map))
711 sbi->discard_blks++;
351df4b2
JK		 712 }
713 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
714 se->ckpt_valid_blocks += del;
715
716 __mark_sit_entry_dirty(sbi, segno);
717
718 /* update total number of valid blocks to be written in ckpt area */
719 SIT_I(sbi)->written_valid_blocks += del;
720
721 if (sbi->segs_per_sec > 1)
722 get_sec_entry(sbi, segno)->valid_blocks += del;
723}
724
5e443818 725void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 726{
5e443818
JK		 727 update_sit_entry(sbi, new, 1);
728 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
729 update_sit_entry(sbi, old, -1);
730
731 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
732 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 733}
734
735void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
736{
737 unsigned int segno = GET_SEGNO(sbi, addr);
738 struct sit_info *sit_i = SIT_I(sbi);
739
9850cf4a 740 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 741 if (addr == NEW_ADDR)
742 return;
743
744 /* add it into sit main buffer */
745 mutex_lock(&sit_i->sentry_lock);
746
747 update_sit_entry(sbi, addr, -1);
748
749 /* add it into dirty seglist */
750 locate_dirty_segment(sbi, segno);
751
752 mutex_unlock(&sit_i->sentry_lock);
753}
754
0a8165d7 755/*
351df4b2
JK		 756 * This function should be resided under the curseg_mutex lock
757 */
758static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 759 struct f2fs_summary *sum)
351df4b2
JK		 760{
761 struct curseg_info *curseg = CURSEG_I(sbi, type);
762 void *addr = curseg->sum_blk;
e79efe3b 763 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 764 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 765}
766
0a8165d7 767/*
351df4b2
JK		 768 * Calculate the number of current summary pages for writing
769 */
3fa06d7b 770int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 771{
351df4b2 772 int valid_sum_count = 0;
9a47938b 773 int i, sum_in_page;
351df4b2
JK		 774
775 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
776 if (sbi->ckpt->alloc_type[i] == SSR)
777 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 778 else {
779 if (for_ra)
780 valid_sum_count += le16_to_cpu(
781 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
782 else
783 valid_sum_count += curseg_blkoff(sbi, i);
784 }
351df4b2
JK		 785 }
786
9a47938b
FL		 787 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
788 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
789 if (valid_sum_count <= sum_in_page)
351df4b2 790 return 1;
9a47938b
FL		 791 else if ((valid_sum_count - sum_in_page) <=
792 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 793 return 2;
794 return 3;
795}
796
0a8165d7 797/*
351df4b2
JK		 798 * Caller should put this summary page
799 */
800struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
801{
802 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
803}
804
381722d2 805void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 806{
807 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 808 void *dst = page_address(page);
809
810 if (src)
811 memcpy(dst, src, PAGE_CACHE_SIZE);
812 else
813 memset(dst, 0, PAGE_CACHE_SIZE);
351df4b2
JK		 814 set_page_dirty(page);
815 f2fs_put_page(page, 1);
816}
817
381722d2
CY		 818static void write_sum_page(struct f2fs_sb_info *sbi,
819 struct f2fs_summary_block *sum_blk, block_t blk_addr)
820{
821 update_meta_page(sbi, (void *)sum_blk, blk_addr);
822}
823
60374688
JK		 824static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
825{
826 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 827 unsigned int segno = curseg->segno + 1;
60374688
JK		 828 struct free_segmap_info *free_i = FREE_I(sbi);
829
7cd8558b 830 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 831 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 832 return 0;
833}
834
0a8165d7 835/*
351df4b2
JK		 836 * Find a new segment from the free segments bitmap to right order
837 * This function should be returned with success, otherwise BUG
838 */
839static void get_new_segment(struct f2fs_sb_info *sbi,
840 unsigned int *newseg, bool new_sec, int dir)
841{
842 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 843 unsigned int segno, secno, zoneno;
7cd8558b 844 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 845 unsigned int hint = *newseg / sbi->segs_per_sec;
846 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
847 unsigned int left_start = hint;
848 bool init = true;
849 int go_left = 0;
850 int i;
851
1a118ccf 852 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 853
854 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
855 segno = find_next_zero_bit(free_i->free_segmap,
7cd8558b 856 MAIN_SEGS(sbi), *newseg + 1);
33afa7fd
JK		 857 if (segno - *newseg < sbi->segs_per_sec -
858 (*newseg % sbi->segs_per_sec))
351df4b2
JK		 859 goto got_it;
860 }
861find_other_zone:
7cd8558b
JK		 862 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
863 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 864 if (dir == ALLOC_RIGHT) {
865 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 866 MAIN_SECS(sbi), 0);
867 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 868 } else {
869 go_left = 1;
870 left_start = hint - 1;
871 }
872 }
873 if (go_left == 0)
874 goto skip_left;
875
876 while (test_bit(left_start, free_i->free_secmap)) {
877 if (left_start > 0) {
878 left_start--;
879 continue;
880 }
881 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 882 MAIN_SECS(sbi), 0);
883 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 884 break;
885 }
886 secno = left_start;
887skip_left:
888 hint = secno;
889 segno = secno * sbi->segs_per_sec;
890 zoneno = secno / sbi->secs_per_zone;
891
892 /* give up on finding another zone */
893 if (!init)
894 goto got_it;
895 if (sbi->secs_per_zone == 1)
896 goto got_it;
897 if (zoneno == old_zoneno)
898 goto got_it;
899 if (dir == ALLOC_LEFT) {
900 if (!go_left && zoneno + 1 >= total_zones)
901 goto got_it;
902 if (go_left && zoneno == 0)
903 goto got_it;
904 }
905 for (i = 0; i < NR_CURSEG_TYPE; i++)
906 if (CURSEG_I(sbi, i)->zone == zoneno)
907 break;
908
909 if (i < NR_CURSEG_TYPE) {
910 /* zone is in user, try another */
911 if (go_left)
912 hint = zoneno * sbi->secs_per_zone - 1;
913 else if (zoneno + 1 >= total_zones)
914 hint = 0;
915 else
916 hint = (zoneno + 1) * sbi->secs_per_zone;
917 init = false;
918 goto find_other_zone;
919 }
920got_it:
921 /* set it as dirty segment in free segmap */
9850cf4a 922 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 923 __set_inuse(sbi, segno);
924 *newseg = segno;
1a118ccf 925 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 926}
927
928static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
929{
930 struct curseg_info *curseg = CURSEG_I(sbi, type);
931 struct summary_footer *sum_footer;
932
933 curseg->segno = curseg->next_segno;
934 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
935 curseg->next_blkoff = 0;
936 curseg->next_segno = NULL_SEGNO;
937
938 sum_footer = &(curseg->sum_blk->footer);
939 memset(sum_footer, 0, sizeof(struct summary_footer));
940 if (IS_DATASEG(type))
941 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
942 if (IS_NODESEG(type))
943 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
944 __set_sit_entry_type(sbi, type, curseg->segno, modified);
945}
946
0a8165d7 947/*
351df4b2
JK		 948 * Allocate a current working segment.
949 * This function always allocates a free segment in LFS manner.
950 */
951static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
952{
953 struct curseg_info *curseg = CURSEG_I(sbi, type);
954 unsigned int segno = curseg->segno;
955 int dir = ALLOC_LEFT;
956
957 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 958 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 959 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
960 dir = ALLOC_RIGHT;
961
962 if (test_opt(sbi, NOHEAP))
963 dir = ALLOC_RIGHT;
964
965 get_new_segment(sbi, &segno, new_sec, dir);
966 curseg->next_segno = segno;
967 reset_curseg(sbi, type, 1);
968 curseg->alloc_type = LFS;
969}
970
971static void __next_free_blkoff(struct f2fs_sb_info *sbi,
972 struct curseg_info *seg, block_t start)
973{
974 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 975 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 976 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 977 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
978 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
979 int i, pos;
980
981 for (i = 0; i < entries; i++)
982 target_map[i] = ckpt_map[i] | cur_map[i];
983
984 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
985
986 seg->next_blkoff = pos;
351df4b2
JK		 987}
988
0a8165d7 989/*
351df4b2
JK		 990 * If a segment is written by LFS manner, next block offset is just obtained
991 * by increasing the current block offset. However, if a segment is written by
992 * SSR manner, next block offset obtained by calling __next_free_blkoff
993 */
994static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
995 struct curseg_info *seg)
996{
997 if (seg->alloc_type == SSR)
998 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
999 else
1000 seg->next_blkoff++;
1001}
1002
0a8165d7 1003/*
e1c42045 1004 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1005 * manner, so it should recover the existing segment information of valid blocks
1006 */
1007static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1008{
1009 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1010 struct curseg_info *curseg = CURSEG_I(sbi, type);
1011 unsigned int new_segno = curseg->next_segno;
1012 struct f2fs_summary_block *sum_node;
1013 struct page *sum_page;
1014
1015 write_sum_page(sbi, curseg->sum_blk,
1016 GET_SUM_BLOCK(sbi, curseg->segno));
1017 __set_test_and_inuse(sbi, new_segno);
1018
1019 mutex_lock(&dirty_i->seglist_lock);
1020 __remove_dirty_segment(sbi, new_segno, PRE);
1021 __remove_dirty_segment(sbi, new_segno, DIRTY);
1022 mutex_unlock(&dirty_i->seglist_lock);
1023
1024 reset_curseg(sbi, type, 1);
1025 curseg->alloc_type = SSR;
1026 __next_free_blkoff(sbi, curseg, 0);
1027
1028 if (reuse) {
1029 sum_page = get_sum_page(sbi, new_segno);
1030 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1031 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1032 f2fs_put_page(sum_page, 1);
1033 }
1034}
1035
43727527
JK		 1036static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1037{
1038 struct curseg_info *curseg = CURSEG_I(sbi, type);
1039 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1040
1041 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1042 return v_ops->get_victim(sbi,
1043 &(curseg)->next_segno, BG_GC, type, SSR);
1044
1045 /* For data segments, let's do SSR more intensively */
1046 for (; type >= CURSEG_HOT_DATA; type--)
1047 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1048 BG_GC, type, SSR))
1049 return 1;
1050 return 0;
1051}
1052
351df4b2
JK		 1053/*
1054 * flush out current segment and replace it with new segment
1055 * This function should be returned with success, otherwise BUG
1056 */
1057static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1058 int type, bool force)
1059{
1060 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1061
7b405275 1062 if (force)
351df4b2 1063 new_curseg(sbi, type, true);
7b405275 1064 else if (type == CURSEG_WARM_NODE)
351df4b2 1065 new_curseg(sbi, type, false);
60374688
JK		 1066 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1067 new_curseg(sbi, type, false);
351df4b2
JK		 1068 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1069 change_curseg(sbi, type, true);
1070 else
1071 new_curseg(sbi, type, false);
dcdfff65
JK		 1072
1073 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1074}
1075
38aa0889
JK		 1076static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1077{
1078 struct curseg_info *curseg = CURSEG_I(sbi, type);
1079 unsigned int old_segno;
1080
1081 old_segno = curseg->segno;
1082 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1083 locate_dirty_segment(sbi, old_segno);
1084}
1085
351df4b2
JK		 1086void allocate_new_segments(struct f2fs_sb_info *sbi)
1087{
351df4b2
JK		 1088 int i;
1089
38aa0889
JK		 1090 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1091 __allocate_new_segments(sbi, i);
351df4b2
JK		 1092}
1093
1094static const struct segment_allocation default_salloc_ops = {
1095 .allocate_segment = allocate_segment_by_default,
1096};
1097
4b2fecc8
JK		 1098int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1099{
f7ef9b83
JK		 1100 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1101 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1102 unsigned int start_segno, end_segno;
1103 struct cp_control cpc;
1104
836b5a63 1105 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1106 return -EINVAL;
1107
9bd27ae4 1108 cpc.trimmed = 0;
7cd8558b 1109 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1110 goto out;
1111
1112 /* start/end segment number in main_area */
7cd8558b
JK		 1113 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1114 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1115 GET_SEGNO(sbi, end);
4b2fecc8 1116 cpc.reason = CP_DISCARD;
836b5a63 1117 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1118
1119 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1120 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1121 cpc.trim_start = start_segno;
a66cdd98
JK		 1122
1123 if (sbi->discard_blks == 0)
1124 break;
1125 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1126 cpc.trim_end = end_segno;
1127 else
1128 cpc.trim_end = min_t(unsigned int,
1129 rounddown(start_segno +
bba681cb
JK		 1130 BATCHED_TRIM_SEGMENTS(sbi),
1131 sbi->segs_per_sec) - 1, end_segno);
1132
1133 mutex_lock(&sbi->gc_mutex);
1134 write_checkpoint(sbi, &cpc);
1135 mutex_unlock(&sbi->gc_mutex);
1136 }
4b2fecc8 1137out:
f7ef9b83 1138 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
4b2fecc8
JK		 1139 return 0;
1140}
1141
351df4b2
JK		 1142static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1143{
1144 struct curseg_info *curseg = CURSEG_I(sbi, type);
1145 if (curseg->next_blkoff < sbi->blocks_per_seg)
1146 return true;
1147 return false;
1148}
1149
1150static int __get_segment_type_2(struct page *page, enum page_type p_type)
1151{
1152 if (p_type == DATA)
1153 return CURSEG_HOT_DATA;
1154 else
1155 return CURSEG_HOT_NODE;
1156}
1157
1158static int __get_segment_type_4(struct page *page, enum page_type p_type)
1159{
1160 if (p_type == DATA) {
1161 struct inode *inode = page->mapping->host;
1162
1163 if (S_ISDIR(inode->i_mode))
1164 return CURSEG_HOT_DATA;
1165 else
1166 return CURSEG_COLD_DATA;
1167 } else {
a344b9fd
JK		 1168 if (IS_DNODE(page) && is_cold_node(page))
1169 return CURSEG_WARM_NODE;
351df4b2
JK		 1170 else
1171 return CURSEG_COLD_NODE;
1172 }
1173}
1174
1175static int __get_segment_type_6(struct page *page, enum page_type p_type)
1176{
1177 if (p_type == DATA) {
1178 struct inode *inode = page->mapping->host;
1179
1180 if (S_ISDIR(inode->i_mode))
1181 return CURSEG_HOT_DATA;
354a3399 1182 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1183 return CURSEG_COLD_DATA;
1184 else
1185 return CURSEG_WARM_DATA;
1186 } else {
1187 if (IS_DNODE(page))
1188 return is_cold_node(page) ? CURSEG_WARM_NODE :
1189 CURSEG_HOT_NODE;
1190 else
1191 return CURSEG_COLD_NODE;
1192 }
1193}
1194
1195static int __get_segment_type(struct page *page, enum page_type p_type)
1196{
4081363f 1197 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1198 case 2:
1199 return __get_segment_type_2(page, p_type);
1200 case 4:
1201 return __get_segment_type_4(page, p_type);
351df4b2 1202 }
12a67146 1203 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1204 f2fs_bug_on(F2FS_P_SB(page),
1205 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1206 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1207}
1208
bfad7c2d
JK		 1209void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1210 block_t old_blkaddr, block_t *new_blkaddr,
1211 struct f2fs_summary *sum, int type)
351df4b2
JK		 1212{
1213 struct sit_info *sit_i = SIT_I(sbi);
1214 struct curseg_info *curseg;
38aa0889
JK		 1215 bool direct_io = (type == CURSEG_DIRECT_IO);
1216
1217 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1218
351df4b2
JK		 1219 curseg = CURSEG_I(sbi, type);
1220
1221 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1222 mutex_lock(&sit_i->sentry_lock);
351df4b2 1223
38aa0889
JK		 1224 /* direct_io'ed data is aligned to the segment for better performance */
1225 if (direct_io && curseg->next_blkoff)
1226 __allocate_new_segments(sbi, type);
1227
351df4b2 1228 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1229
1230 /*
1231 * __add_sum_entry should be resided under the curseg_mutex
1232 * because, this function updates a summary entry in the
1233 * current summary block.
1234 */
e79efe3b 1235 __add_sum_entry(sbi, type, sum);
351df4b2 1236
351df4b2 1237 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1238
1239 stat_inc_block_count(sbi, curseg);
351df4b2 1240
5e443818
JK		 1241 if (!__has_curseg_space(sbi, type))
1242 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1243 /*
1244 * SIT information should be updated before segment allocation,
1245 * since SSR needs latest valid block information.
1246 */
1247 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1248
351df4b2
JK		 1249 mutex_unlock(&sit_i->sentry_lock);
1250
bfad7c2d 1251 if (page && IS_NODESEG(type))
351df4b2
JK		 1252 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1253
bfad7c2d
JK		 1254 mutex_unlock(&curseg->curseg_mutex);
1255}
1256
05ca3632 1257static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1258{
05ca3632 1259 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1260
05ca3632
JK		 1261 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1262 &fio->blk_addr, sum, type);
bfad7c2d 1263
351df4b2 1264 /* writeout dirty page into bdev */
05ca3632 1265 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1266}
1267
577e3495 1268void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1269{
458e6197 1270 struct f2fs_io_info fio = {
05ca3632 1271 .sbi = sbi,
458e6197 1272 .type = META,
cf04e8eb
JK		 1273 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1274 .blk_addr = page->index,
05ca3632 1275 .page = page,
4375a336 1276 .encrypted_page = NULL,
458e6197
JK		 1277 };
1278
351df4b2 1279 set_page_writeback(page);
05ca3632 1280 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1281}
1282
05ca3632 1283void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1284{
1285 struct f2fs_summary sum;
05ca3632 1286
351df4b2 1287 set_summary(&sum, nid, 0, 0);
05ca3632 1288 do_write_page(&sum, fio);
351df4b2
JK		 1289}
1290
05ca3632 1291void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1292{
05ca3632 1293 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1294 struct f2fs_summary sum;
1295 struct node_info ni;
1296
9850cf4a 1297 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1298 get_node_info(sbi, dn->nid, &ni);
1299 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1300 do_write_page(&sum, fio);
e1509cf2 1301 dn->data_blkaddr = fio->blk_addr;
351df4b2
JK		 1302}
1303
05ca3632 1304void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1305{
05ca3632
JK		 1306 stat_inc_inplace_blocks(fio->sbi);
1307 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1308}
1309
528e3459
CY		 1310static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1311 struct f2fs_summary *sum,
19f106bc
CY		 1312 block_t old_blkaddr, block_t new_blkaddr,
1313 bool recover_curseg)
351df4b2
JK		 1314{
1315 struct sit_info *sit_i = SIT_I(sbi);
1316 struct curseg_info *curseg;
1317 unsigned int segno, old_cursegno;
1318 struct seg_entry *se;
1319 int type;
19f106bc 1320 unsigned short old_blkoff;
351df4b2
JK		 1321
1322 segno = GET_SEGNO(sbi, new_blkaddr);
1323 se = get_seg_entry(sbi, segno);
1324 type = se->type;
1325
19f106bc
CY		 1326 if (!recover_curseg) {
1327 /* for recovery flow */
1328 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1329 if (old_blkaddr == NULL_ADDR)
1330 type = CURSEG_COLD_DATA;
1331 else
1332 type = CURSEG_WARM_DATA;
1333 }
1334 } else {
1335 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1336 type = CURSEG_WARM_DATA;
1337 }
19f106bc 1338
351df4b2
JK		 1339 curseg = CURSEG_I(sbi, type);
1340
1341 mutex_lock(&curseg->curseg_mutex);
1342 mutex_lock(&sit_i->sentry_lock);
1343
1344 old_cursegno = curseg->segno;
19f106bc 1345 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1346
1347 /* change the current segment */
1348 if (segno != curseg->segno) {
1349 curseg->next_segno = segno;
1350 change_curseg(sbi, type, true);
1351 }
1352
491c0854 1353 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1354 __add_sum_entry(sbi, type, sum);
351df4b2
JK		 1355
1356 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
351df4b2 1357 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1358
19f106bc
CY		 1359 if (recover_curseg) {
1360 if (old_cursegno != curseg->segno) {
1361 curseg->next_segno = old_cursegno;
1362 change_curseg(sbi, type, true);
1363 }
1364 curseg->next_blkoff = old_blkoff;
1365 }
1366
351df4b2
JK		 1367 mutex_unlock(&sit_i->sentry_lock);
1368 mutex_unlock(&curseg->curseg_mutex);
1369}
1370
528e3459
CY		 1371void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1372 block_t old_addr, block_t new_addr,
1373 unsigned char version, bool recover_curseg)
1374{
1375 struct f2fs_summary sum;
1376
1377 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1378
1379 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1380
1381 dn->data_blkaddr = new_addr;
1382 set_data_blkaddr(dn);
1383 f2fs_update_extent_cache(dn);
1384}
1385
df0f8dc0
CY		 1386static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1387 struct page *page, enum page_type type)
1388{
1389 enum page_type btype = PAGE_TYPE_OF_BIO(type);
1390 struct f2fs_bio_info *io = &sbi->write_io[btype];
df0f8dc0 1391 struct bio_vec *bvec;
4375a336 1392 struct page *target;
df0f8dc0
CY		 1393 int i;
1394
1395 down_read(&io->io_rwsem);
4375a336
JK		 1396 if (!io->bio) {
1397 up_read(&io->io_rwsem);
1398 return false;
1399 }
df0f8dc0 1400
ce23447f 1401 bio_for_each_segment_all(bvec, io->bio, i) {
4375a336
JK		 1402
1403 if (bvec->bv_page->mapping) {
1404 target = bvec->bv_page;
1405 } else {
1406 struct f2fs_crypto_ctx *ctx;
1407
1408 /* encrypted page */
1409 ctx = (struct f2fs_crypto_ctx *)page_private(
1410 bvec->bv_page);
ca40b030 1411 target = ctx->w.control_page;
4375a336
JK		 1412 }
1413
1414 if (page == target) {
df0f8dc0
CY		 1415 up_read(&io->io_rwsem);
1416 return true;
1417 }
1418 }
1419
df0f8dc0
CY		 1420 up_read(&io->io_rwsem);
1421 return false;
1422}
1423
93dfe2ac 1424void f2fs_wait_on_page_writeback(struct page *page,
5514f0aa 1425 enum page_type type)
93dfe2ac 1426{
93dfe2ac 1427 if (PageWriteback(page)) {
4081363f
JK		 1428 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1429
df0f8dc0
CY		 1430 if (is_merged_page(sbi, page, type))
1431 f2fs_submit_merged_bio(sbi, type, WRITE);
93dfe2ac
JK		 1432 wait_on_page_writeback(page);
1433 }
1434}
1435
351df4b2
JK		 1436static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1437{
1438 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1439 struct curseg_info *seg_i;
1440 unsigned char *kaddr;
1441 struct page *page;
1442 block_t start;
1443 int i, j, offset;
1444
1445 start = start_sum_block(sbi);
1446
1447 page = get_meta_page(sbi, start++);
1448 kaddr = (unsigned char *)page_address(page);
1449
1450 /* Step 1: restore nat cache */
1451 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1452 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1453
1454 /* Step 2: restore sit cache */
1455 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1456 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1457 SUM_JOURNAL_SIZE);
1458 offset = 2 * SUM_JOURNAL_SIZE;
1459
1460 /* Step 3: restore summary entries */
1461 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1462 unsigned short blk_off;
1463 unsigned int segno;
1464
1465 seg_i = CURSEG_I(sbi, i);
1466 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1467 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1468 seg_i->next_segno = segno;
1469 reset_curseg(sbi, i, 0);
1470 seg_i->alloc_type = ckpt->alloc_type[i];
1471 seg_i->next_blkoff = blk_off;
1472
1473 if (seg_i->alloc_type == SSR)
1474 blk_off = sbi->blocks_per_seg;
1475
1476 for (j = 0; j < blk_off; j++) {
1477 struct f2fs_summary *s;
1478 s = (struct f2fs_summary *)(kaddr + offset);
1479 seg_i->sum_blk->entries[j] = *s;
1480 offset += SUMMARY_SIZE;
1481 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1482 SUM_FOOTER_SIZE)
1483 continue;
1484
1485 f2fs_put_page(page, 1);
1486 page = NULL;
1487
1488 page = get_meta_page(sbi, start++);
1489 kaddr = (unsigned char *)page_address(page);
1490 offset = 0;
1491 }
1492 }
1493 f2fs_put_page(page, 1);
1494 return 0;
1495}
1496
1497static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1498{
1499 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1500 struct f2fs_summary_block *sum;
1501 struct curseg_info *curseg;
1502 struct page *new;
1503 unsigned short blk_off;
1504 unsigned int segno = 0;
1505 block_t blk_addr = 0;
1506
1507 /* get segment number and block addr */
1508 if (IS_DATASEG(type)) {
1509 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1510 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1511 CURSEG_HOT_DATA]);
119ee914 1512 if (__exist_node_summaries(sbi))
351df4b2
JK		 1513 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1514 else
1515 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1516 } else {
1517 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1518 CURSEG_HOT_NODE]);
1519 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1520 CURSEG_HOT_NODE]);
119ee914 1521 if (__exist_node_summaries(sbi))
351df4b2
JK		 1522 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1523 type - CURSEG_HOT_NODE);
1524 else
1525 blk_addr = GET_SUM_BLOCK(sbi, segno);
1526 }
1527
1528 new = get_meta_page(sbi, blk_addr);
1529 sum = (struct f2fs_summary_block *)page_address(new);
1530
1531 if (IS_NODESEG(type)) {
119ee914 1532 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1533 struct f2fs_summary *ns = &sum->entries[0];
1534 int i;
1535 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1536 ns->version = 0;
1537 ns->ofs_in_node = 0;
1538 }
1539 } else {
d653788a
GZ		 1540 int err;
1541
1542 err = restore_node_summary(sbi, segno, sum);
1543 if (err) {
351df4b2 1544 f2fs_put_page(new, 1);
d653788a 1545 return err;
351df4b2
JK		 1546 }
1547 }
1548 }
1549
1550 /* set uncompleted segment to curseg */
1551 curseg = CURSEG_I(sbi, type);
1552 mutex_lock(&curseg->curseg_mutex);
1553 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1554 curseg->next_segno = segno;
1555 reset_curseg(sbi, type, 0);
1556 curseg->alloc_type = ckpt->alloc_type[type];
1557 curseg->next_blkoff = blk_off;
1558 mutex_unlock(&curseg->curseg_mutex);
1559 f2fs_put_page(new, 1);
1560 return 0;
1561}
1562
1563static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1564{
1565 int type = CURSEG_HOT_DATA;
e4fc5fbf 1566 int err;
351df4b2 1567
25ca923b 1568 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1569 int npages = npages_for_summary_flush(sbi, true);
1570
1571 if (npages >= 2)
1572 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1573 META_CP);
1574
351df4b2
JK		 1575 /* restore for compacted data summary */
1576 if (read_compacted_summaries(sbi))
1577 return -EINVAL;
1578 type = CURSEG_HOT_NODE;
1579 }
1580
119ee914 1581 if (__exist_node_summaries(sbi))
3fa06d7b
CY		 1582 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1583 NR_CURSEG_TYPE - type, META_CP);
1584
e4fc5fbf
CY		 1585 for (; type <= CURSEG_COLD_NODE; type++) {
1586 err = read_normal_summaries(sbi, type);
1587 if (err)
1588 return err;
1589 }
1590
351df4b2
JK		 1591 return 0;
1592}
1593
1594static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1595{
1596 struct page *page;
1597 unsigned char *kaddr;
1598 struct f2fs_summary *summary;
1599 struct curseg_info *seg_i;
1600 int written_size = 0;
1601 int i, j;
1602
1603 page = grab_meta_page(sbi, blkaddr++);
1604 kaddr = (unsigned char *)page_address(page);
1605
1606 /* Step 1: write nat cache */
1607 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1608 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1609 written_size += SUM_JOURNAL_SIZE;
1610
1611 /* Step 2: write sit cache */
1612 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1613 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1614 SUM_JOURNAL_SIZE);
1615 written_size += SUM_JOURNAL_SIZE;
1616
351df4b2
JK		 1617 /* Step 3: write summary entries */
1618 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1619 unsigned short blkoff;
1620 seg_i = CURSEG_I(sbi, i);
1621 if (sbi->ckpt->alloc_type[i] == SSR)
1622 blkoff = sbi->blocks_per_seg;
1623 else
1624 blkoff = curseg_blkoff(sbi, i);
1625
1626 for (j = 0; j < blkoff; j++) {
1627 if (!page) {
1628 page = grab_meta_page(sbi, blkaddr++);
1629 kaddr = (unsigned char *)page_address(page);
1630 written_size = 0;
1631 }
1632 summary = (struct f2fs_summary *)(kaddr + written_size);
1633 *summary = seg_i->sum_blk->entries[j];
1634 written_size += SUMMARY_SIZE;
351df4b2
JK		 1635
1636 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1637 SUM_FOOTER_SIZE)
1638 continue;
1639
e8d61a74 1640 set_page_dirty(page);
351df4b2
JK		 1641 f2fs_put_page(page, 1);
1642 page = NULL;
1643 }
1644 }
e8d61a74
CY		 1645 if (page) {
1646 set_page_dirty(page);
351df4b2 1647 f2fs_put_page(page, 1);
e8d61a74 1648 }
351df4b2
JK		 1649}
1650
1651static void write_normal_summaries(struct f2fs_sb_info *sbi,
1652 block_t blkaddr, int type)
1653{
1654 int i, end;
1655 if (IS_DATASEG(type))
1656 end = type + NR_CURSEG_DATA_TYPE;
1657 else
1658 end = type + NR_CURSEG_NODE_TYPE;
1659
1660 for (i = type; i < end; i++) {
1661 struct curseg_info *sum = CURSEG_I(sbi, i);
1662 mutex_lock(&sum->curseg_mutex);
1663 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1664 mutex_unlock(&sum->curseg_mutex);
1665 }
1666}
1667
1668void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1669{
25ca923b 1670 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1671 write_compacted_summaries(sbi, start_blk);
1672 else
1673 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1674}
1675
1676void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1677{
119ee914 1678 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1679}
1680
1681int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1682 unsigned int val, int alloc)
1683{
1684 int i;
1685
1686 if (type == NAT_JOURNAL) {
1687 for (i = 0; i < nats_in_cursum(sum); i++) {
1688 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1689 return i;
1690 }
1691 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1692 return update_nats_in_cursum(sum, 1);
1693 } else if (type == SIT_JOURNAL) {
1694 for (i = 0; i < sits_in_cursum(sum); i++)
1695 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1696 return i;
1697 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1698 return update_sits_in_cursum(sum, 1);
1699 }
1700 return -1;
1701}
1702
1703static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1704 unsigned int segno)
1705{
2cc22186 1706 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1707}
1708
1709static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1710 unsigned int start)
1711{
1712 struct sit_info *sit_i = SIT_I(sbi);
1713 struct page *src_page, *dst_page;
1714 pgoff_t src_off, dst_off;
1715 void *src_addr, *dst_addr;
1716
1717 src_off = current_sit_addr(sbi, start);
1718 dst_off = next_sit_addr(sbi, src_off);
1719
1720 /* get current sit block page without lock */
1721 src_page = get_meta_page(sbi, src_off);
1722 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1723 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1724
1725 src_addr = page_address(src_page);
1726 dst_addr = page_address(dst_page);
1727 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1728
1729 set_page_dirty(dst_page);
1730 f2fs_put_page(src_page, 1);
1731
1732 set_to_next_sit(sit_i, start);
1733
1734 return dst_page;
1735}
1736
184a5cd2
CY		 1737static struct sit_entry_set *grab_sit_entry_set(void)
1738{
1739 struct sit_entry_set *ses =
1740 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1741
1742 ses->entry_cnt = 0;
1743 INIT_LIST_HEAD(&ses->set_list);
1744 return ses;
1745}
1746
1747static void release_sit_entry_set(struct sit_entry_set *ses)
1748{
1749 list_del(&ses->set_list);
1750 kmem_cache_free(sit_entry_set_slab, ses);
1751}
1752
1753static void adjust_sit_entry_set(struct sit_entry_set *ses,
1754 struct list_head *head)
1755{
1756 struct sit_entry_set *next = ses;
1757
1758 if (list_is_last(&ses->set_list, head))
1759 return;
1760
1761 list_for_each_entry_continue(next, head, set_list)
1762 if (ses->entry_cnt <= next->entry_cnt)
1763 break;
1764
1765 list_move_tail(&ses->set_list, &next->set_list);
1766}
1767
1768static void add_sit_entry(unsigned int segno, struct list_head *head)
1769{
1770 struct sit_entry_set *ses;
1771 unsigned int start_segno = START_SEGNO(segno);
1772
1773 list_for_each_entry(ses, head, set_list) {
1774 if (ses->start_segno == start_segno) {
1775 ses->entry_cnt++;
1776 adjust_sit_entry_set(ses, head);
1777 return;
1778 }
1779 }
1780
1781 ses = grab_sit_entry_set();
1782
1783 ses->start_segno = start_segno;
1784 ses->entry_cnt++;
1785 list_add(&ses->set_list, head);
1786}
1787
1788static void add_sits_in_set(struct f2fs_sb_info *sbi)
1789{
1790 struct f2fs_sm_info *sm_info = SM_I(sbi);
1791 struct list_head *set_list = &sm_info->sit_entry_set;
1792 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1793 unsigned int segno;
1794
7cd8558b 1795 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1796 add_sit_entry(segno, set_list);
1797}
1798
1799static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1800{
1801 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1802 struct f2fs_summary_block *sum = curseg->sum_blk;
1803 int i;
1804
184a5cd2
CY		 1805 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1806 unsigned int segno;
1807 bool dirtied;
1808
1809 segno = le32_to_cpu(segno_in_journal(sum, i));
1810 dirtied = __mark_sit_entry_dirty(sbi, segno);
1811
1812 if (!dirtied)
1813 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1814 }
184a5cd2 1815 update_sits_in_cursum(sum, -sits_in_cursum(sum));
351df4b2
JK		 1816}
1817
0a8165d7 1818/*
351df4b2
JK		 1819 * CP calls this function, which flushes SIT entries including sit_journal,
1820 * and moves prefree segs to free segs.
1821 */
4b2fecc8 1822void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1823{
1824 struct sit_info *sit_i = SIT_I(sbi);
1825 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1826 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1827 struct f2fs_summary_block *sum = curseg->sum_blk;
184a5cd2
CY		 1828 struct sit_entry_set *ses, *tmp;
1829 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1830 bool to_journal = true;
4b2fecc8 1831 struct seg_entry *se;
351df4b2
JK		 1832
1833 mutex_lock(&curseg->curseg_mutex);
1834 mutex_lock(&sit_i->sentry_lock);
1835
2b11a74b
WL		 1836 if (!sit_i->dirty_sentries)
1837 goto out;
1838
351df4b2 1839 /*
184a5cd2
CY		 1840 * add and account sit entries of dirty bitmap in sit entry
1841 * set temporarily
351df4b2 1842 */
184a5cd2 1843 add_sits_in_set(sbi);
351df4b2 1844
184a5cd2
CY		 1845 /*
1846 * if there are no enough space in journal to store dirty sit
1847 * entries, remove all entries from journal and add and account
1848 * them in sit entry set.
1849 */
1850 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1851 remove_sits_in_journal(sbi);
b2955550 1852
184a5cd2
CY		 1853 /*
1854 * there are two steps to flush sit entries:
1855 * #1, flush sit entries to journal in current cold data summary block.
1856 * #2, flush sit entries to sit page.
1857 */
1858 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1859 struct page *page = NULL;
184a5cd2
CY		 1860 struct f2fs_sit_block *raw_sit = NULL;
1861 unsigned int start_segno = ses->start_segno;
1862 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1863 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1864 unsigned int segno = start_segno;
1865
1866 if (to_journal &&
1867 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1868 to_journal = false;
1869
1870 if (!to_journal) {
1871 page = get_next_sit_page(sbi, start_segno);
1872 raw_sit = page_address(page);
351df4b2 1873 }
351df4b2 1874
184a5cd2
CY		 1875 /* flush dirty sit entries in region of current sit set */
1876 for_each_set_bit_from(segno, bitmap, end) {
1877 int offset, sit_offset;
4b2fecc8
JK		 1878
1879 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 1880
1881 /* add discard candidates */
d7bc2484 1882 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 1883 cpc->trim_start = segno;
1884 add_discard_addrs(sbi, cpc);
1885 }
184a5cd2
CY		 1886
1887 if (to_journal) {
1888 offset = lookup_journal_in_cursum(sum,
1889 SIT_JOURNAL, segno, 1);
1890 f2fs_bug_on(sbi, offset < 0);
1891 segno_in_journal(sum, offset) =
1892 cpu_to_le32(segno);
1893 seg_info_to_raw_sit(se,
1894 &sit_in_journal(sum, offset));
1895 } else {
1896 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1897 seg_info_to_raw_sit(se,
1898 &raw_sit->entries[sit_offset]);
1899 }
351df4b2 1900
184a5cd2
CY		 1901 __clear_bit(segno, bitmap);
1902 sit_i->dirty_sentries--;
1903 ses->entry_cnt--;
351df4b2
JK		 1904 }
1905
184a5cd2
CY		 1906 if (!to_journal)
1907 f2fs_put_page(page, 1);
1908
1909 f2fs_bug_on(sbi, ses->entry_cnt);
1910 release_sit_entry_set(ses);
351df4b2 1911 }
184a5cd2
CY		 1912
1913 f2fs_bug_on(sbi, !list_empty(head));
1914 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 1915out:
4b2fecc8
JK		 1916 if (cpc->reason == CP_DISCARD) {
1917 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1918 add_discard_addrs(sbi, cpc);
1919 }
351df4b2
JK		 1920 mutex_unlock(&sit_i->sentry_lock);
1921 mutex_unlock(&curseg->curseg_mutex);
1922
351df4b2
JK		 1923 set_prefree_as_free_segments(sbi);
1924}
1925
1926static int build_sit_info(struct f2fs_sb_info *sbi)
1927{
1928 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1929 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1930 struct sit_info *sit_i;
1931 unsigned int sit_segs, start;
1932 char *src_bitmap, *dst_bitmap;
1933 unsigned int bitmap_size;
1934
1935 /* allocate memory for SIT information */
1936 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1937 if (!sit_i)
1938 return -ENOMEM;
1939
1940 SM_I(sbi)->sit_info = sit_i;
1941
7cd8558b 1942 sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
351df4b2
JK		 1943 if (!sit_i->sentries)
1944 return -ENOMEM;
1945
7cd8558b 1946 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 1947 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1948 if (!sit_i->dirty_sentries_bitmap)
1949 return -ENOMEM;
1950
7cd8558b 1951 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 1952 sit_i->sentries[start].cur_valid_map
1953 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1954 sit_i->sentries[start].ckpt_valid_map
1955 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 1956 sit_i->sentries[start].discard_map
1957 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1958 if (!sit_i->sentries[start].cur_valid_map ||
1959 !sit_i->sentries[start].ckpt_valid_map ||
1960 !sit_i->sentries[start].discard_map)
351df4b2
JK		 1961 return -ENOMEM;
1962 }
1963
60a3b782
JK		 1964 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1965 if (!sit_i->tmp_map)
1966 return -ENOMEM;
1967
351df4b2 1968 if (sbi->segs_per_sec > 1) {
7cd8558b 1969 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
351df4b2
JK		 1970 sizeof(struct sec_entry));
1971 if (!sit_i->sec_entries)
1972 return -ENOMEM;
1973 }
1974
1975 /* get information related with SIT */
1976 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1977
1978 /* setup SIT bitmap from ckeckpoint pack */
1979 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1980 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1981
79b5793b 1982 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 1983 if (!dst_bitmap)
1984 return -ENOMEM;
351df4b2
JK		 1985
1986 /* init SIT information */
1987 sit_i->s_ops = &default_salloc_ops;
1988
1989 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1990 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1991 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1992 sit_i->sit_bitmap = dst_bitmap;
1993 sit_i->bitmap_size = bitmap_size;
1994 sit_i->dirty_sentries = 0;
1995 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1996 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1997 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1998 mutex_init(&sit_i->sentry_lock);
1999 return 0;
2000}
2001
2002static int build_free_segmap(struct f2fs_sb_info *sbi)
2003{
351df4b2
JK		 2004 struct free_segmap_info *free_i;
2005 unsigned int bitmap_size, sec_bitmap_size;
2006
2007 /* allocate memory for free segmap information */
2008 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2009 if (!free_i)
2010 return -ENOMEM;
2011
2012 SM_I(sbi)->free_info = free_i;
2013
7cd8558b 2014 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2015 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
2016 if (!free_i->free_segmap)
2017 return -ENOMEM;
2018
7cd8558b 2019 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2
JK		 2020 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
2021 if (!free_i->free_secmap)
2022 return -ENOMEM;
2023
2024 /* set all segments as dirty temporarily */
2025 memset(free_i->free_segmap, 0xff, bitmap_size);
2026 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2027
2028 /* init free segmap information */
7cd8558b 2029 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2030 free_i->free_segments = 0;
2031 free_i->free_sections = 0;
1a118ccf 2032 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2033 return 0;
2034}
2035
2036static int build_curseg(struct f2fs_sb_info *sbi)
2037{
1042d60f 2038 struct curseg_info *array;
351df4b2
JK		 2039 int i;
2040
b434babf 2041 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2042 if (!array)
2043 return -ENOMEM;
2044
2045 SM_I(sbi)->curseg_array = array;
2046
2047 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2048 mutex_init(&array[i].curseg_mutex);
2049 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2050 if (!array[i].sum_blk)
2051 return -ENOMEM;
2052 array[i].segno = NULL_SEGNO;
2053 array[i].next_blkoff = 0;
2054 }
2055 return restore_curseg_summaries(sbi);
2056}
2057
2058static void build_sit_entries(struct f2fs_sb_info *sbi)
2059{
2060 struct sit_info *sit_i = SIT_I(sbi);
2061 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2062 struct f2fs_summary_block *sum = curseg->sum_blk;
74de593a
CY		 2063 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2064 unsigned int i, start, end;
2065 unsigned int readed, start_blk = 0;
90a893c7 2066 int nrpages = MAX_BIO_BLOCKS(sbi);
351df4b2 2067
74de593a 2068 do {
662befda 2069 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
74de593a
CY		 2070
2071 start = start_blk * sit_i->sents_per_block;
2072 end = (start_blk + readed) * sit_i->sents_per_block;
2073
7cd8558b 2074 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2075 struct seg_entry *se = &sit_i->sentries[start];
2076 struct f2fs_sit_block *sit_blk;
2077 struct f2fs_sit_entry sit;
2078 struct page *page;
2079
2080 mutex_lock(&curseg->curseg_mutex);
2081 for (i = 0; i < sits_in_cursum(sum); i++) {
6c311ec6
CF		 2082 if (le32_to_cpu(segno_in_journal(sum, i))
2083 == start) {
74de593a
CY		 2084 sit = sit_in_journal(sum, i);
2085 mutex_unlock(&curseg->curseg_mutex);
2086 goto got_it;
2087 }
351df4b2 2088 }
74de593a
CY		 2089 mutex_unlock(&curseg->curseg_mutex);
2090
2091 page = get_current_sit_page(sbi, start);
2092 sit_blk = (struct f2fs_sit_block *)page_address(page);
2093 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2094 f2fs_put_page(page, 1);
351df4b2 2095got_it:
74de593a
CY		 2096 check_block_count(sbi, start, &sit);
2097 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2098
2099 /* build discard map only one time */
2100 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2101 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2102
74de593a
CY		 2103 if (sbi->segs_per_sec > 1) {
2104 struct sec_entry *e = get_sec_entry(sbi, start);
2105 e->valid_blocks += se->valid_blocks;
2106 }
351df4b2 2107 }
74de593a
CY		 2108 start_blk += readed;
2109 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2110}
2111
2112static void init_free_segmap(struct f2fs_sb_info *sbi)
2113{
2114 unsigned int start;
2115 int type;
2116
7cd8558b 2117 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2118 struct seg_entry *sentry = get_seg_entry(sbi, start);
2119 if (!sentry->valid_blocks)
2120 __set_free(sbi, start);
2121 }
2122
2123 /* set use the current segments */
2124 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2125 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2126 __set_test_and_inuse(sbi, curseg_t->segno);
2127 }
2128}
2129
2130static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2131{
2132 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2133 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2134 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2135 unsigned short valid_blocks;
2136
8736fbf0 2137 while (1) {
351df4b2 2138 /* find dirty segment based on free segmap */
7cd8558b
JK		 2139 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2140 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2141 break;
2142 offset = segno + 1;
2143 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2144 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2145 continue;
ec325b52
JK		 2146 if (valid_blocks > sbi->blocks_per_seg) {
2147 f2fs_bug_on(sbi, 1);
2148 continue;
2149 }
351df4b2
JK		 2150 mutex_lock(&dirty_i->seglist_lock);
2151 __locate_dirty_segment(sbi, segno, DIRTY);
2152 mutex_unlock(&dirty_i->seglist_lock);
2153 }
2154}
2155
5ec4e49f 2156static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2157{
2158 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2159 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2160
5ec4e49f
JK		 2161 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2162 if (!dirty_i->victim_secmap)
351df4b2
JK		 2163 return -ENOMEM;
2164 return 0;
2165}
2166
2167static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2168{
2169 struct dirty_seglist_info *dirty_i;
2170 unsigned int bitmap_size, i;
2171
2172 /* allocate memory for dirty segments list information */
2173 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2174 if (!dirty_i)
2175 return -ENOMEM;
2176
2177 SM_I(sbi)->dirty_info = dirty_i;
2178 mutex_init(&dirty_i->seglist_lock);
2179
7cd8558b 2180 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2181
2182 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2183 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2184 if (!dirty_i->dirty_segmap[i])
2185 return -ENOMEM;
2186 }
2187
2188 init_dirty_segmap(sbi);
5ec4e49f 2189 return init_victim_secmap(sbi);
351df4b2
JK		 2190}
2191
0a8165d7 2192/*
351df4b2
JK		 2193 * Update min, max modified time for cost-benefit GC algorithm
2194 */
2195static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2196{
2197 struct sit_info *sit_i = SIT_I(sbi);
2198 unsigned int segno;
2199
2200 mutex_lock(&sit_i->sentry_lock);
2201
2202 sit_i->min_mtime = LLONG_MAX;
2203
7cd8558b 2204 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2205 unsigned int i;
2206 unsigned long long mtime = 0;
2207
2208 for (i = 0; i < sbi->segs_per_sec; i++)
2209 mtime += get_seg_entry(sbi, segno + i)->mtime;
2210
2211 mtime = div_u64(mtime, sbi->segs_per_sec);
2212
2213 if (sit_i->min_mtime > mtime)
2214 sit_i->min_mtime = mtime;
2215 }
2216 sit_i->max_mtime = get_mtime(sbi);
2217 mutex_unlock(&sit_i->sentry_lock);
2218}
2219
2220int build_segment_manager(struct f2fs_sb_info *sbi)
2221{
2222 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2223 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2224 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2225 int err;
2226
2227 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2228 if (!sm_info)
2229 return -ENOMEM;
2230
2231 /* init sm info */
2232 sbi->sm_info = sm_info;
351df4b2
JK		 2233 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2234 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2235 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2236 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2237 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2238 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2239 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2240 sm_info->rec_prefree_segments = sm_info->main_segments *
2241 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2242 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2243 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2244 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2245
7fd9e544
JK		 2246 INIT_LIST_HEAD(&sm_info->discard_list);
2247 sm_info->nr_discards = 0;
2248 sm_info->max_discards = 0;
2249
bba681cb
JK		 2250 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2251
184a5cd2
CY		 2252 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2253
b270ad6f 2254 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2255 err = create_flush_cmd_control(sbi);
2256 if (err)
a688b9d9 2257 return err;
6b4afdd7
JK		 2258 }
2259
351df4b2
JK		 2260 err = build_sit_info(sbi);
2261 if (err)
2262 return err;
2263 err = build_free_segmap(sbi);
2264 if (err)
2265 return err;
2266 err = build_curseg(sbi);
2267 if (err)
2268 return err;
2269
2270 /* reinit free segmap based on SIT */
2271 build_sit_entries(sbi);
2272
2273 init_free_segmap(sbi);
2274 err = build_dirty_segmap(sbi);
2275 if (err)
2276 return err;
2277
2278 init_min_max_mtime(sbi);
2279 return 0;
2280}
2281
2282static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2283 enum dirty_type dirty_type)
2284{
2285 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2286
2287 mutex_lock(&dirty_i->seglist_lock);
2288 kfree(dirty_i->dirty_segmap[dirty_type]);
2289 dirty_i->nr_dirty[dirty_type] = 0;
2290 mutex_unlock(&dirty_i->seglist_lock);
2291}
2292
5ec4e49f 2293static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2294{
2295 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5ec4e49f 2296 kfree(dirty_i->victim_secmap);
351df4b2
JK		 2297}
2298
2299static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2300{
2301 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2302 int i;
2303
2304 if (!dirty_i)
2305 return;
2306
2307 /* discard pre-free/dirty segments list */
2308 for (i = 0; i < NR_DIRTY_TYPE; i++)
2309 discard_dirty_segmap(sbi, i);
2310
5ec4e49f 2311 destroy_victim_secmap(sbi);
351df4b2
JK		 2312 SM_I(sbi)->dirty_info = NULL;
2313 kfree(dirty_i);
2314}
2315
2316static void destroy_curseg(struct f2fs_sb_info *sbi)
2317{
2318 struct curseg_info *array = SM_I(sbi)->curseg_array;
2319 int i;
2320
2321 if (!array)
2322 return;
2323 SM_I(sbi)->curseg_array = NULL;
2324 for (i = 0; i < NR_CURSEG_TYPE; i++)
2325 kfree(array[i].sum_blk);
2326 kfree(array);
2327}
2328
2329static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2330{
2331 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2332 if (!free_i)
2333 return;
2334 SM_I(sbi)->free_info = NULL;
2335 kfree(free_i->free_segmap);
2336 kfree(free_i->free_secmap);
2337 kfree(free_i);
2338}
2339
2340static void destroy_sit_info(struct f2fs_sb_info *sbi)
2341{
2342 struct sit_info *sit_i = SIT_I(sbi);
2343 unsigned int start;
2344
2345 if (!sit_i)
2346 return;
2347
2348 if (sit_i->sentries) {
7cd8558b 2349 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2350 kfree(sit_i->sentries[start].cur_valid_map);
2351 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2352 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2353 }
2354 }
60a3b782
JK		 2355 kfree(sit_i->tmp_map);
2356
351df4b2
JK		 2357 vfree(sit_i->sentries);
2358 vfree(sit_i->sec_entries);
2359 kfree(sit_i->dirty_sentries_bitmap);
2360
2361 SM_I(sbi)->sit_info = NULL;
2362 kfree(sit_i->sit_bitmap);
2363 kfree(sit_i);
2364}
2365
2366void destroy_segment_manager(struct f2fs_sb_info *sbi)
2367{
2368 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2369
3b03f724
CY		 2370 if (!sm_info)
2371 return;
2163d198 2372 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2373 destroy_dirty_segmap(sbi);
2374 destroy_curseg(sbi);
2375 destroy_free_segmap(sbi);
2376 destroy_sit_info(sbi);
2377 sbi->sm_info = NULL;
2378 kfree(sm_info);
2379}
7fd9e544
JK		 2380
2381int __init create_segment_manager_caches(void)
2382{
2383 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2384 sizeof(struct discard_entry));
7fd9e544 2385 if (!discard_entry_slab)
184a5cd2
CY		 2386 goto fail;
2387
2388 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2389 sizeof(struct sit_entry_set));
184a5cd2
CY		 2390 if (!sit_entry_set_slab)
2391 goto destory_discard_entry;
88b88a66
JK		 2392
2393 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2394 sizeof(struct inmem_pages));
2395 if (!inmem_entry_slab)
2396 goto destroy_sit_entry_set;
7fd9e544 2397 return 0;
184a5cd2 2398
88b88a66
JK		 2399destroy_sit_entry_set:
2400 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2401destory_discard_entry:
2402 kmem_cache_destroy(discard_entry_slab);
2403fail:
2404 return -ENOMEM;
7fd9e544
JK		 2405}
2406
2407void destroy_segment_manager_caches(void)
2408{
184a5cd2 2409 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2410 kmem_cache_destroy(discard_entry_slab);
88b88a66 2411 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2412}
Ubuntu Artful kernel mirror