projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Revert "f2fs: do not skip dentry block writes"
[deliverable/linux.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>
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 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;
9be32d72 200
9e4ded3f 201 f2fs_trace_pid(page);
0722b101 202
decd36b6
CY		 203 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
204 SetPagePrivate(page);
205
88b88a66
JK		 206 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
207
208 /* add atomic page indices to the list */
209 new->page = page;
210 INIT_LIST_HEAD(&new->list);
decd36b6 211
88b88a66
JK		 212 /* increase reference count with clean state */
213 mutex_lock(&fi->inmem_lock);
214 get_page(page);
215 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 216 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 217 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 218
219 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 220}
221
edb27dee 222int commit_inmem_pages(struct inode *inode, bool abort)
88b88a66
JK		 223{
224 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
225 struct f2fs_inode_info *fi = F2FS_I(inode);
226 struct inmem_pages *cur, *tmp;
227 bool submit_bio = false;
228 struct f2fs_io_info fio = {
05ca3632 229 .sbi = sbi,
88b88a66 230 .type = DATA,
1e84371f 231 .rw = WRITE_SYNC | REQ_PRIO,
4375a336 232 .encrypted_page = NULL,
88b88a66 233 };
edb27dee 234 int err = 0;
88b88a66 235
0341845e
JK		 236 /*
237 * The abort is true only when f2fs_evict_inode is called.
238 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
239 * that we don't need to call f2fs_balance_fs.
240 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
241 * inode becomes free by iget_locked in f2fs_iget.
242 */
70c640b1 243 if (!abort) {
0341845e 244 f2fs_balance_fs(sbi);
70c640b1
JK		 245 f2fs_lock_op(sbi);
246 }
88b88a66
JK		 247
248 mutex_lock(&fi->inmem_lock);
249 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
decd36b6 250 lock_page(cur->page);
70c640b1 251 if (!abort) {
70c640b1 252 if (cur->page->mapping == inode->i_mapping) {
6282adbf 253 set_page_dirty(cur->page);
70c640b1
JK		 254 f2fs_wait_on_page_writeback(cur->page, DATA);
255 if (clear_page_dirty_for_io(cur->page))
256 inode_dec_dirty_pages(inode);
8ce67cb0 257 trace_f2fs_commit_inmem_page(cur->page, INMEM);
05ca3632 258 fio.page = cur->page;
edb27dee 259 err = do_write_data_page(&fio);
70c640b1 260 submit_bio = true;
edb27dee
JK		 261 if (err) {
262 unlock_page(cur->page);
263 break;
264 }
70c640b1 265 }
70c640b1 266 } else {
8ce67cb0 267 trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
88b88a66 268 }
decd36b6
CY		 269 set_page_private(cur->page, 0);
270 ClearPagePrivate(cur->page);
271 f2fs_put_page(cur->page, 1);
272
88b88a66
JK		 273 list_del(&cur->list);
274 kmem_cache_free(inmem_entry_slab, cur);
8dcf2ff7 275 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 276 }
88b88a66
JK		 277 mutex_unlock(&fi->inmem_lock);
278
70c640b1
JK		 279 if (!abort) {
280 f2fs_unlock_op(sbi);
281 if (submit_bio)
282 f2fs_submit_merged_bio(sbi, DATA, WRITE);
283 }
edb27dee 284 return err;
88b88a66
JK		 285}
286
0a8165d7 287/*
351df4b2
JK		 288 * This function balances dirty node and dentry pages.
289 * In addition, it controls garbage collection.
290 */
291void f2fs_balance_fs(struct f2fs_sb_info *sbi)
292{
351df4b2 293 /*
029cd28c
JK		 294 * We should do GC or end up with checkpoint, if there are so many dirty
295 * dir/node pages without enough free segments.
351df4b2 296 */
43727527 297 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 298 mutex_lock(&sbi->gc_mutex);
d530d4d8 299 f2fs_gc(sbi, false);
351df4b2
JK		 300 }
301}
302
4660f9c0
JK		 303void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
304{
1dcc336b 305 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 306 if (!available_free_memory(sbi, EXTENT_CACHE))
307 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 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
31696580
CY		 313 if (!available_free_memory(sbi, FREE_NIDS))
314 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
315
1b38dc8e
JK		 316 /* checkpoint is the only way to shrink partial cached entries */
317 if (!available_free_memory(sbi, NAT_ENTRIES) ||
e5e7ea3c 318 excess_prefree_segs(sbi) ||
60b99b48
JK		 319 !available_free_memory(sbi, INO_ENTRIES) ||
320 jiffies > sbi->cp_expires)
4660f9c0
JK		 321 f2fs_sync_fs(sbi->sb, true);
322}
323
2163d198 324static int issue_flush_thread(void *data)
6b4afdd7
JK		 325{
326 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 327 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
328 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 329repeat:
330 if (kthread_should_stop())
331 return 0;
332
721bd4d5 333 if (!llist_empty(&fcc->issue_list)) {
740432f8 334 struct bio *bio;
6b4afdd7
JK		 335 struct flush_cmd *cmd, *next;
336 int ret;
337
740432f8
JK		 338 bio = f2fs_bio_alloc(0);
339
721bd4d5
GZ		 340 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
341 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
342
6b4afdd7
JK		 343 bio->bi_bdev = sbi->sb->s_bdev;
344 ret = submit_bio_wait(WRITE_FLUSH, bio);
345
721bd4d5
GZ		 346 llist_for_each_entry_safe(cmd, next,
347 fcc->dispatch_list, llnode) {
6b4afdd7 348 cmd->ret = ret;
6b4afdd7
JK		 349 complete(&cmd->wait);
350 }
a4ed23f2 351 bio_put(bio);
a688b9d9 352 fcc->dispatch_list = NULL;
6b4afdd7
JK		 353 }
354
a688b9d9 355 wait_event_interruptible(*q,
721bd4d5 356 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 357 goto repeat;
358}
359
360int f2fs_issue_flush(struct f2fs_sb_info *sbi)
361{
a688b9d9 362 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 363 struct flush_cmd cmd;
6b4afdd7 364
24a9ee0f
JK		 365 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
366 test_opt(sbi, FLUSH_MERGE));
367
0f7b2abd
JK		 368 if (test_opt(sbi, NOBARRIER))
369 return 0;
370
740432f8
JK		 371 if (!test_opt(sbi, FLUSH_MERGE)) {
372 struct bio *bio = f2fs_bio_alloc(0);
373 int ret;
374
375 bio->bi_bdev = sbi->sb->s_bdev;
376 ret = submit_bio_wait(WRITE_FLUSH, bio);
377 bio_put(bio);
378 return ret;
379 }
6b4afdd7 380
adf8d90b 381 init_completion(&cmd.wait);
6b4afdd7 382
721bd4d5 383 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 384
a688b9d9
GZ		 385 if (!fcc->dispatch_list)
386 wake_up(&fcc->flush_wait_queue);
6b4afdd7 387
adf8d90b
CY		 388 wait_for_completion(&cmd.wait);
389
390 return cmd.ret;
6b4afdd7
JK		 391}
392
2163d198
GZ		 393int create_flush_cmd_control(struct f2fs_sb_info *sbi)
394{
395 dev_t dev = sbi->sb->s_bdev->bd_dev;
396 struct flush_cmd_control *fcc;
397 int err = 0;
398
399 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
400 if (!fcc)
401 return -ENOMEM;
2163d198 402 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 403 init_llist_head(&fcc->issue_list);
6b2920a5 404 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 405 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
406 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
407 if (IS_ERR(fcc->f2fs_issue_flush)) {
408 err = PTR_ERR(fcc->f2fs_issue_flush);
409 kfree(fcc);
6b2920a5 410 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 411 return err;
412 }
2163d198
GZ		 413
414 return err;
415}
416
417void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
418{
6b2920a5 419 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 420
421 if (fcc && fcc->f2fs_issue_flush)
422 kthread_stop(fcc->f2fs_issue_flush);
423 kfree(fcc);
6b2920a5 424 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 425}
426
351df4b2
JK		 427static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
428 enum dirty_type dirty_type)
429{
430 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
431
432 /* need not be added */
433 if (IS_CURSEG(sbi, segno))
434 return;
435
436 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
437 dirty_i->nr_dirty[dirty_type]++;
438
439 if (dirty_type == DIRTY) {
440 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 441 enum dirty_type t = sentry->type;
b2f2c390 442
ec325b52
JK		 443 if (unlikely(t >= DIRTY)) {
444 f2fs_bug_on(sbi, 1);
445 return;
446 }
4625d6aa
CL		 447 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
448 dirty_i->nr_dirty[t]++;
351df4b2
JK		 449 }
450}
451
452static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
453 enum dirty_type dirty_type)
454{
455 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
456
457 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
458 dirty_i->nr_dirty[dirty_type]--;
459
460 if (dirty_type == DIRTY) {
4625d6aa
CL		 461 struct seg_entry *sentry = get_seg_entry(sbi, segno);
462 enum dirty_type t = sentry->type;
463
464 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
465 dirty_i->nr_dirty[t]--;
b2f2c390 466
5ec4e49f
JK		 467 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
468 clear_bit(GET_SECNO(sbi, segno),
469 dirty_i->victim_secmap);
351df4b2
JK		 470 }
471}
472
0a8165d7 473/*
351df4b2
JK		 474 * Should not occur error such as -ENOMEM.
475 * Adding dirty entry into seglist is not critical operation.
476 * If a given segment is one of current working segments, it won't be added.
477 */
8d8451af 478static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 479{
480 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
481 unsigned short valid_blocks;
482
483 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
484 return;
485
486 mutex_lock(&dirty_i->seglist_lock);
487
488 valid_blocks = get_valid_blocks(sbi, segno, 0);
489
490 if (valid_blocks == 0) {
491 __locate_dirty_segment(sbi, segno, PRE);
492 __remove_dirty_segment(sbi, segno, DIRTY);
493 } else if (valid_blocks < sbi->blocks_per_seg) {
494 __locate_dirty_segment(sbi, segno, DIRTY);
495 } else {
496 /* Recovery routine with SSR needs this */
497 __remove_dirty_segment(sbi, segno, DIRTY);
498 }
499
500 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 501}
502
1e87a78d 503static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 504 block_t blkstart, block_t blklen)
505{
55cf9cb6
CY		 506 sector_t start = SECTOR_FROM_BLOCK(blkstart);
507 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 508 struct seg_entry *se;
509 unsigned int offset;
510 block_t i;
511
512 for (i = blkstart; i < blkstart + blklen; i++) {
513 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
514 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
515
516 if (!f2fs_test_and_set_bit(offset, se->discard_map))
517 sbi->discard_blks--;
518 }
1661d07c 519 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 520 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
521}
522
e90c2d28 523bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 524{
40a02be1
JK		 525 int err = -ENOTSUPP;
526
527 if (test_opt(sbi, DISCARD)) {
528 struct seg_entry *se = get_seg_entry(sbi,
529 GET_SEGNO(sbi, blkaddr));
530 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
531
532 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 533 return false;
40a02be1
JK		 534
535 err = f2fs_issue_discard(sbi, blkaddr, 1);
536 }
537
e90c2d28 538 if (err) {
381722d2 539 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 540 return true;
541 }
542 return false;
37208879
JK		 543}
544
adf4983b 545static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 546 struct cp_control *cpc, struct seg_entry *se,
547 unsigned int start, unsigned int end)
b2955550
JK		 548{
549 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 550 struct discard_entry *new, *last;
551
552 if (!list_empty(head)) {
553 last = list_last_entry(head, struct discard_entry, list);
554 if (START_BLOCK(sbi, cpc->trim_start) + start ==
555 last->blkaddr + last->len) {
556 last->len += end - start;
557 goto done;
558 }
559 }
560
561 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
562 INIT_LIST_HEAD(&new->list);
563 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
564 new->len = end - start;
565 list_add_tail(&new->list, head);
566done:
567 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 568}
569
570static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
571{
b2955550
JK		 572 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
573 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 574 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 575 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
576 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 577 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 578 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 579 unsigned int start = 0, end = -1;
4b2fecc8 580 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 581 int i;
582
a66cdd98 583 if (se->valid_blocks == max_blocks)
b2955550
JK		 584 return;
585
a66cdd98
JK		 586 if (!force) {
587 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 588 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
589 return;
4b2fecc8
JK		 590 }
591
b2955550
JK		 592 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
593 for (i = 0; i < entries; i++)
a66cdd98 594 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 595 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 596
4b2fecc8 597 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 598 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
599 if (start >= max_blocks)
600 break;
601
602 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 603 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 604 }
605}
606
4b2fecc8
JK		 607void release_discard_addrs(struct f2fs_sb_info *sbi)
608{
609 struct list_head *head = &(SM_I(sbi)->discard_list);
610 struct discard_entry *entry, *this;
611
612 /* drop caches */
613 list_for_each_entry_safe(entry, this, head, list) {
614 list_del(&entry->list);
615 kmem_cache_free(discard_entry_slab, entry);
616 }
617}
618
0a8165d7 619/*
351df4b2
JK		 620 * Should call clear_prefree_segments after checkpoint is done.
621 */
622static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
623{
624 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 625 unsigned int segno;
351df4b2
JK		 626
627 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 628 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 629 __set_test_and_free(sbi, segno);
351df4b2
JK		 630 mutex_unlock(&dirty_i->seglist_lock);
631}
632
836b5a63 633void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 634{
b2955550 635 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 636 struct discard_entry *entry, *this;
351df4b2 637 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 638 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 639 unsigned int start = 0, end = -1;
351df4b2
JK		 640
641 mutex_lock(&dirty_i->seglist_lock);
29e59c14 642
351df4b2 643 while (1) {
29e59c14 644 int i;
7cd8558b
JK		 645 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
646 if (start >= MAIN_SEGS(sbi))
351df4b2 647 break;
7cd8558b
JK		 648 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
649 start + 1);
29e59c14
CL		 650
651 for (i = start; i < end; i++)
652 clear_bit(i, prefree_map);
653
654 dirty_i->nr_dirty[PRE] -= end - start;
655
656 if (!test_opt(sbi, DISCARD))
657 continue;
351df4b2 658
37208879
JK		 659 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
660 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 661 }
662 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 663
664 /* send small discards */
2d7b822a 665 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 666 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
667 goto skip;
37208879 668 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 669 cpc->trimmed += entry->len;
836b5a63 670skip:
b2955550
JK		 671 list_del(&entry->list);
672 SM_I(sbi)->nr_discards -= entry->len;
673 kmem_cache_free(discard_entry_slab, entry);
674 }
351df4b2
JK		 675}
676
184a5cd2 677static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 678{
679 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 680
681 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 682 sit_i->dirty_sentries++;
184a5cd2
CY		 683 return false;
684 }
685
686 return true;
351df4b2
JK		 687}
688
689static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
690 unsigned int segno, int modified)
691{
692 struct seg_entry *se = get_seg_entry(sbi, segno);
693 se->type = type;
694 if (modified)
695 __mark_sit_entry_dirty(sbi, segno);
696}
697
698static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
699{
700 struct seg_entry *se;
701 unsigned int segno, offset;
702 long int new_vblocks;
703
704 segno = GET_SEGNO(sbi, blkaddr);
705
706 se = get_seg_entry(sbi, segno);
707 new_vblocks = se->valid_blocks + del;
491c0854 708 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 709
9850cf4a 710 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 711 (new_vblocks > sbi->blocks_per_seg)));
712
713 se->valid_blocks = new_vblocks;
714 se->mtime = get_mtime(sbi);
715 SIT_I(sbi)->max_mtime = se->mtime;
716
717 /* Update valid block bitmap */
718 if (del > 0) {
52aca074 719 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 720 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 721 if (!f2fs_test_and_set_bit(offset, se->discard_map))
722 sbi->discard_blks--;
351df4b2 723 } else {
52aca074 724 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 725 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 726 if (f2fs_test_and_clear_bit(offset, se->discard_map))
727 sbi->discard_blks++;
351df4b2
JK		 728 }
729 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
730 se->ckpt_valid_blocks += del;
731
732 __mark_sit_entry_dirty(sbi, segno);
733
734 /* update total number of valid blocks to be written in ckpt area */
735 SIT_I(sbi)->written_valid_blocks += del;
736
737 if (sbi->segs_per_sec > 1)
738 get_sec_entry(sbi, segno)->valid_blocks += del;
739}
740
5e443818 741void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 742{
5e443818
JK		 743 update_sit_entry(sbi, new, 1);
744 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
745 update_sit_entry(sbi, old, -1);
746
747 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
748 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 749}
750
751void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
752{
753 unsigned int segno = GET_SEGNO(sbi, addr);
754 struct sit_info *sit_i = SIT_I(sbi);
755
9850cf4a 756 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 757 if (addr == NEW_ADDR)
758 return;
759
760 /* add it into sit main buffer */
761 mutex_lock(&sit_i->sentry_lock);
762
763 update_sit_entry(sbi, addr, -1);
764
765 /* add it into dirty seglist */
766 locate_dirty_segment(sbi, segno);
767
768 mutex_unlock(&sit_i->sentry_lock);
769}
770
0a8165d7 771/*
351df4b2
JK		 772 * This function should be resided under the curseg_mutex lock
773 */
774static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 775 struct f2fs_summary *sum)
351df4b2
JK		 776{
777 struct curseg_info *curseg = CURSEG_I(sbi, type);
778 void *addr = curseg->sum_blk;
e79efe3b 779 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 780 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 781}
782
0a8165d7 783/*
351df4b2
JK		 784 * Calculate the number of current summary pages for writing
785 */
3fa06d7b 786int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 787{
351df4b2 788 int valid_sum_count = 0;
9a47938b 789 int i, sum_in_page;
351df4b2
JK		 790
791 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
792 if (sbi->ckpt->alloc_type[i] == SSR)
793 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 794 else {
795 if (for_ra)
796 valid_sum_count += le16_to_cpu(
797 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
798 else
799 valid_sum_count += curseg_blkoff(sbi, i);
800 }
351df4b2
JK		 801 }
802
9a47938b
FL		 803 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
804 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
805 if (valid_sum_count <= sum_in_page)
351df4b2 806 return 1;
9a47938b
FL		 807 else if ((valid_sum_count - sum_in_page) <=
808 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 809 return 2;
810 return 3;
811}
812
0a8165d7 813/*
351df4b2
JK		 814 * Caller should put this summary page
815 */
816struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
817{
818 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
819}
820
381722d2 821void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 822{
823 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 824 void *dst = page_address(page);
825
826 if (src)
827 memcpy(dst, src, PAGE_CACHE_SIZE);
828 else
829 memset(dst, 0, PAGE_CACHE_SIZE);
351df4b2
JK		 830 set_page_dirty(page);
831 f2fs_put_page(page, 1);
832}
833
381722d2
CY		 834static void write_sum_page(struct f2fs_sb_info *sbi,
835 struct f2fs_summary_block *sum_blk, block_t blk_addr)
836{
837 update_meta_page(sbi, (void *)sum_blk, blk_addr);
838}
839
60374688
JK		 840static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
841{
842 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 843 unsigned int segno = curseg->segno + 1;
60374688
JK		 844 struct free_segmap_info *free_i = FREE_I(sbi);
845
7cd8558b 846 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 847 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 848 return 0;
849}
850
0a8165d7 851/*
351df4b2
JK		 852 * Find a new segment from the free segments bitmap to right order
853 * This function should be returned with success, otherwise BUG
854 */
855static void get_new_segment(struct f2fs_sb_info *sbi,
856 unsigned int *newseg, bool new_sec, int dir)
857{
858 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 859 unsigned int segno, secno, zoneno;
7cd8558b 860 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 861 unsigned int hint = *newseg / sbi->segs_per_sec;
862 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
863 unsigned int left_start = hint;
864 bool init = true;
865 int go_left = 0;
866 int i;
867
1a118ccf 868 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 869
870 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
871 segno = find_next_zero_bit(free_i->free_segmap,
7cd8558b 872 MAIN_SEGS(sbi), *newseg + 1);
33afa7fd
JK		 873 if (segno - *newseg < sbi->segs_per_sec -
874 (*newseg % sbi->segs_per_sec))
351df4b2
JK		 875 goto got_it;
876 }
877find_other_zone:
7cd8558b
JK		 878 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
879 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 880 if (dir == ALLOC_RIGHT) {
881 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 882 MAIN_SECS(sbi), 0);
883 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 884 } else {
885 go_left = 1;
886 left_start = hint - 1;
887 }
888 }
889 if (go_left == 0)
890 goto skip_left;
891
892 while (test_bit(left_start, free_i->free_secmap)) {
893 if (left_start > 0) {
894 left_start--;
895 continue;
896 }
897 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 898 MAIN_SECS(sbi), 0);
899 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 900 break;
901 }
902 secno = left_start;
903skip_left:
904 hint = secno;
905 segno = secno * sbi->segs_per_sec;
906 zoneno = secno / sbi->secs_per_zone;
907
908 /* give up on finding another zone */
909 if (!init)
910 goto got_it;
911 if (sbi->secs_per_zone == 1)
912 goto got_it;
913 if (zoneno == old_zoneno)
914 goto got_it;
915 if (dir == ALLOC_LEFT) {
916 if (!go_left && zoneno + 1 >= total_zones)
917 goto got_it;
918 if (go_left && zoneno == 0)
919 goto got_it;
920 }
921 for (i = 0; i < NR_CURSEG_TYPE; i++)
922 if (CURSEG_I(sbi, i)->zone == zoneno)
923 break;
924
925 if (i < NR_CURSEG_TYPE) {
926 /* zone is in user, try another */
927 if (go_left)
928 hint = zoneno * sbi->secs_per_zone - 1;
929 else if (zoneno + 1 >= total_zones)
930 hint = 0;
931 else
932 hint = (zoneno + 1) * sbi->secs_per_zone;
933 init = false;
934 goto find_other_zone;
935 }
936got_it:
937 /* set it as dirty segment in free segmap */
9850cf4a 938 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 939 __set_inuse(sbi, segno);
940 *newseg = segno;
1a118ccf 941 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 942}
943
944static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
945{
946 struct curseg_info *curseg = CURSEG_I(sbi, type);
947 struct summary_footer *sum_footer;
948
949 curseg->segno = curseg->next_segno;
950 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
951 curseg->next_blkoff = 0;
952 curseg->next_segno = NULL_SEGNO;
953
954 sum_footer = &(curseg->sum_blk->footer);
955 memset(sum_footer, 0, sizeof(struct summary_footer));
956 if (IS_DATASEG(type))
957 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
958 if (IS_NODESEG(type))
959 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
960 __set_sit_entry_type(sbi, type, curseg->segno, modified);
961}
962
0a8165d7 963/*
351df4b2
JK		 964 * Allocate a current working segment.
965 * This function always allocates a free segment in LFS manner.
966 */
967static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
968{
969 struct curseg_info *curseg = CURSEG_I(sbi, type);
970 unsigned int segno = curseg->segno;
971 int dir = ALLOC_LEFT;
972
973 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 974 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 975 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
976 dir = ALLOC_RIGHT;
977
978 if (test_opt(sbi, NOHEAP))
979 dir = ALLOC_RIGHT;
980
981 get_new_segment(sbi, &segno, new_sec, dir);
982 curseg->next_segno = segno;
983 reset_curseg(sbi, type, 1);
984 curseg->alloc_type = LFS;
985}
986
987static void __next_free_blkoff(struct f2fs_sb_info *sbi,
988 struct curseg_info *seg, block_t start)
989{
990 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 991 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 992 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 993 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
994 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
995 int i, pos;
996
997 for (i = 0; i < entries; i++)
998 target_map[i] = ckpt_map[i] | cur_map[i];
999
1000 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1001
1002 seg->next_blkoff = pos;
351df4b2
JK		 1003}
1004
0a8165d7 1005/*
351df4b2
JK		 1006 * If a segment is written by LFS manner, next block offset is just obtained
1007 * by increasing the current block offset. However, if a segment is written by
1008 * SSR manner, next block offset obtained by calling __next_free_blkoff
1009 */
1010static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1011 struct curseg_info *seg)
1012{
1013 if (seg->alloc_type == SSR)
1014 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1015 else
1016 seg->next_blkoff++;
1017}
1018
0a8165d7 1019/*
e1c42045 1020 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1021 * manner, so it should recover the existing segment information of valid blocks
1022 */
1023static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1024{
1025 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1026 struct curseg_info *curseg = CURSEG_I(sbi, type);
1027 unsigned int new_segno = curseg->next_segno;
1028 struct f2fs_summary_block *sum_node;
1029 struct page *sum_page;
1030
1031 write_sum_page(sbi, curseg->sum_blk,
1032 GET_SUM_BLOCK(sbi, curseg->segno));
1033 __set_test_and_inuse(sbi, new_segno);
1034
1035 mutex_lock(&dirty_i->seglist_lock);
1036 __remove_dirty_segment(sbi, new_segno, PRE);
1037 __remove_dirty_segment(sbi, new_segno, DIRTY);
1038 mutex_unlock(&dirty_i->seglist_lock);
1039
1040 reset_curseg(sbi, type, 1);
1041 curseg->alloc_type = SSR;
1042 __next_free_blkoff(sbi, curseg, 0);
1043
1044 if (reuse) {
1045 sum_page = get_sum_page(sbi, new_segno);
1046 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1047 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1048 f2fs_put_page(sum_page, 1);
1049 }
1050}
1051
43727527
JK		 1052static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1053{
1054 struct curseg_info *curseg = CURSEG_I(sbi, type);
1055 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1056
1057 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1058 return v_ops->get_victim(sbi,
1059 &(curseg)->next_segno, BG_GC, type, SSR);
1060
1061 /* For data segments, let's do SSR more intensively */
1062 for (; type >= CURSEG_HOT_DATA; type--)
1063 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1064 BG_GC, type, SSR))
1065 return 1;
1066 return 0;
1067}
1068
351df4b2
JK		 1069/*
1070 * flush out current segment and replace it with new segment
1071 * This function should be returned with success, otherwise BUG
1072 */
1073static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1074 int type, bool force)
1075{
1076 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1077
7b405275 1078 if (force)
351df4b2 1079 new_curseg(sbi, type, true);
7b405275 1080 else if (type == CURSEG_WARM_NODE)
351df4b2 1081 new_curseg(sbi, type, false);
60374688
JK		 1082 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1083 new_curseg(sbi, type, false);
351df4b2
JK		 1084 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1085 change_curseg(sbi, type, true);
1086 else
1087 new_curseg(sbi, type, false);
dcdfff65
JK		 1088
1089 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1090}
1091
38aa0889
JK		 1092static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1093{
1094 struct curseg_info *curseg = CURSEG_I(sbi, type);
1095 unsigned int old_segno;
1096
1097 old_segno = curseg->segno;
1098 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1099 locate_dirty_segment(sbi, old_segno);
1100}
1101
351df4b2
JK		 1102void allocate_new_segments(struct f2fs_sb_info *sbi)
1103{
351df4b2
JK		 1104 int i;
1105
38aa0889
JK		 1106 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1107 __allocate_new_segments(sbi, i);
351df4b2
JK		 1108}
1109
1110static const struct segment_allocation default_salloc_ops = {
1111 .allocate_segment = allocate_segment_by_default,
1112};
1113
4b2fecc8
JK		 1114int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1115{
f7ef9b83
JK		 1116 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1117 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1118 unsigned int start_segno, end_segno;
1119 struct cp_control cpc;
1120
836b5a63 1121 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1122 return -EINVAL;
1123
9bd27ae4 1124 cpc.trimmed = 0;
7cd8558b 1125 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1126 goto out;
1127
1128 /* start/end segment number in main_area */
7cd8558b
JK		 1129 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1130 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1131 GET_SEGNO(sbi, end);
4b2fecc8 1132 cpc.reason = CP_DISCARD;
836b5a63 1133 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1134
1135 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1136 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1137 cpc.trim_start = start_segno;
a66cdd98
JK		 1138
1139 if (sbi->discard_blks == 0)
1140 break;
1141 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1142 cpc.trim_end = end_segno;
1143 else
1144 cpc.trim_end = min_t(unsigned int,
1145 rounddown(start_segno +
bba681cb
JK		 1146 BATCHED_TRIM_SEGMENTS(sbi),
1147 sbi->segs_per_sec) - 1, end_segno);
1148
1149 mutex_lock(&sbi->gc_mutex);
1150 write_checkpoint(sbi, &cpc);
1151 mutex_unlock(&sbi->gc_mutex);
1152 }
4b2fecc8 1153out:
f7ef9b83 1154 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
4b2fecc8
JK		 1155 return 0;
1156}
1157
351df4b2
JK		 1158static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1159{
1160 struct curseg_info *curseg = CURSEG_I(sbi, type);
1161 if (curseg->next_blkoff < sbi->blocks_per_seg)
1162 return true;
1163 return false;
1164}
1165
1166static int __get_segment_type_2(struct page *page, enum page_type p_type)
1167{
1168 if (p_type == DATA)
1169 return CURSEG_HOT_DATA;
1170 else
1171 return CURSEG_HOT_NODE;
1172}
1173
1174static int __get_segment_type_4(struct page *page, enum page_type p_type)
1175{
1176 if (p_type == DATA) {
1177 struct inode *inode = page->mapping->host;
1178
1179 if (S_ISDIR(inode->i_mode))
1180 return CURSEG_HOT_DATA;
1181 else
1182 return CURSEG_COLD_DATA;
1183 } else {
a344b9fd
JK		 1184 if (IS_DNODE(page) && is_cold_node(page))
1185 return CURSEG_WARM_NODE;
351df4b2
JK		 1186 else
1187 return CURSEG_COLD_NODE;
1188 }
1189}
1190
1191static int __get_segment_type_6(struct page *page, enum page_type p_type)
1192{
1193 if (p_type == DATA) {
1194 struct inode *inode = page->mapping->host;
1195
1196 if (S_ISDIR(inode->i_mode))
1197 return CURSEG_HOT_DATA;
354a3399 1198 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1199 return CURSEG_COLD_DATA;
1200 else
1201 return CURSEG_WARM_DATA;
1202 } else {
1203 if (IS_DNODE(page))
1204 return is_cold_node(page) ? CURSEG_WARM_NODE :
1205 CURSEG_HOT_NODE;
1206 else
1207 return CURSEG_COLD_NODE;
1208 }
1209}
1210
1211static int __get_segment_type(struct page *page, enum page_type p_type)
1212{
4081363f 1213 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1214 case 2:
1215 return __get_segment_type_2(page, p_type);
1216 case 4:
1217 return __get_segment_type_4(page, p_type);
351df4b2 1218 }
12a67146 1219 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1220 f2fs_bug_on(F2FS_P_SB(page),
1221 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1222 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1223}
1224
bfad7c2d
JK		 1225void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1226 block_t old_blkaddr, block_t *new_blkaddr,
1227 struct f2fs_summary *sum, int type)
351df4b2
JK		 1228{
1229 struct sit_info *sit_i = SIT_I(sbi);
1230 struct curseg_info *curseg;
38aa0889
JK		 1231 bool direct_io = (type == CURSEG_DIRECT_IO);
1232
1233 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1234
351df4b2
JK		 1235 curseg = CURSEG_I(sbi, type);
1236
1237 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1238 mutex_lock(&sit_i->sentry_lock);
351df4b2 1239
38aa0889 1240 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1241 if (direct_io && curseg->next_blkoff &&
1242 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1243 __allocate_new_segments(sbi, type);
1244
351df4b2 1245 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1246
1247 /*
1248 * __add_sum_entry should be resided under the curseg_mutex
1249 * because, this function updates a summary entry in the
1250 * current summary block.
1251 */
e79efe3b 1252 __add_sum_entry(sbi, type, sum);
351df4b2 1253
351df4b2 1254 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1255
1256 stat_inc_block_count(sbi, curseg);
351df4b2 1257
5e443818
JK		 1258 if (!__has_curseg_space(sbi, type))
1259 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1260 /*
1261 * SIT information should be updated before segment allocation,
1262 * since SSR needs latest valid block information.
1263 */
1264 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1265
351df4b2
JK		 1266 mutex_unlock(&sit_i->sentry_lock);
1267
bfad7c2d 1268 if (page && IS_NODESEG(type))
351df4b2
JK		 1269 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1270
bfad7c2d
JK		 1271 mutex_unlock(&curseg->curseg_mutex);
1272}
1273
05ca3632 1274static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1275{
05ca3632 1276 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1277
05ca3632
JK		 1278 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1279 &fio->blk_addr, sum, type);
bfad7c2d 1280
351df4b2 1281 /* writeout dirty page into bdev */
05ca3632 1282 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1283}
1284
577e3495 1285void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1286{
458e6197 1287 struct f2fs_io_info fio = {
05ca3632 1288 .sbi = sbi,
458e6197 1289 .type = META,
cf04e8eb
JK		 1290 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1291 .blk_addr = page->index,
05ca3632 1292 .page = page,
4375a336 1293 .encrypted_page = NULL,
458e6197
JK		 1294 };
1295
351df4b2 1296 set_page_writeback(page);
05ca3632 1297 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1298}
1299
05ca3632 1300void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1301{
1302 struct f2fs_summary sum;
05ca3632 1303
351df4b2 1304 set_summary(&sum, nid, 0, 0);
05ca3632 1305 do_write_page(&sum, fio);
351df4b2
JK		 1306}
1307
05ca3632 1308void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1309{
05ca3632 1310 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1311 struct f2fs_summary sum;
1312 struct node_info ni;
1313
9850cf4a 1314 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1315 get_node_info(sbi, dn->nid, &ni);
1316 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1317 do_write_page(&sum, fio);
e1509cf2 1318 dn->data_blkaddr = fio->blk_addr;
351df4b2
JK		 1319}
1320
05ca3632 1321void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1322{
05ca3632
JK		 1323 stat_inc_inplace_blocks(fio->sbi);
1324 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1325}
1326
528e3459
CY		 1327static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1328 struct f2fs_summary *sum,
19f106bc
CY		 1329 block_t old_blkaddr, block_t new_blkaddr,
1330 bool recover_curseg)
351df4b2
JK		 1331{
1332 struct sit_info *sit_i = SIT_I(sbi);
1333 struct curseg_info *curseg;
1334 unsigned int segno, old_cursegno;
1335 struct seg_entry *se;
1336 int type;
19f106bc 1337 unsigned short old_blkoff;
351df4b2
JK		 1338
1339 segno = GET_SEGNO(sbi, new_blkaddr);
1340 se = get_seg_entry(sbi, segno);
1341 type = se->type;
1342
19f106bc
CY		 1343 if (!recover_curseg) {
1344 /* for recovery flow */
1345 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1346 if (old_blkaddr == NULL_ADDR)
1347 type = CURSEG_COLD_DATA;
1348 else
1349 type = CURSEG_WARM_DATA;
1350 }
1351 } else {
1352 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1353 type = CURSEG_WARM_DATA;
1354 }
19f106bc 1355
351df4b2
JK		 1356 curseg = CURSEG_I(sbi, type);
1357
1358 mutex_lock(&curseg->curseg_mutex);
1359 mutex_lock(&sit_i->sentry_lock);
1360
1361 old_cursegno = curseg->segno;
19f106bc 1362 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1363
1364 /* change the current segment */
1365 if (segno != curseg->segno) {
1366 curseg->next_segno = segno;
1367 change_curseg(sbi, type, true);
1368 }
1369
491c0854 1370 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1371 __add_sum_entry(sbi, type, sum);
351df4b2
JK		 1372
1373 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
351df4b2 1374 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1375
19f106bc
CY		 1376 if (recover_curseg) {
1377 if (old_cursegno != curseg->segno) {
1378 curseg->next_segno = old_cursegno;
1379 change_curseg(sbi, type, true);
1380 }
1381 curseg->next_blkoff = old_blkoff;
1382 }
1383
351df4b2
JK		 1384 mutex_unlock(&sit_i->sentry_lock);
1385 mutex_unlock(&curseg->curseg_mutex);
1386}
1387
528e3459
CY		 1388void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1389 block_t old_addr, block_t new_addr,
1390 unsigned char version, bool recover_curseg)
1391{
1392 struct f2fs_summary sum;
1393
1394 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1395
1396 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1397
1398 dn->data_blkaddr = new_addr;
1399 set_data_blkaddr(dn);
1400 f2fs_update_extent_cache(dn);
1401}
1402
df0f8dc0
CY		 1403static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1404 struct page *page, enum page_type type)
1405{
1406 enum page_type btype = PAGE_TYPE_OF_BIO(type);
1407 struct f2fs_bio_info *io = &sbi->write_io[btype];
df0f8dc0 1408 struct bio_vec *bvec;
4375a336 1409 struct page *target;
df0f8dc0
CY		 1410 int i;
1411
1412 down_read(&io->io_rwsem);
4375a336
JK		 1413 if (!io->bio) {
1414 up_read(&io->io_rwsem);
1415 return false;
1416 }
df0f8dc0 1417
ce23447f 1418 bio_for_each_segment_all(bvec, io->bio, i) {
4375a336
JK		 1419
1420 if (bvec->bv_page->mapping) {
1421 target = bvec->bv_page;
1422 } else {
1423 struct f2fs_crypto_ctx *ctx;
1424
1425 /* encrypted page */
1426 ctx = (struct f2fs_crypto_ctx *)page_private(
1427 bvec->bv_page);
ca40b030 1428 target = ctx->w.control_page;
4375a336
JK		 1429 }
1430
1431 if (page == target) {
df0f8dc0
CY		 1432 up_read(&io->io_rwsem);
1433 return true;
1434 }
1435 }
1436
df0f8dc0
CY		 1437 up_read(&io->io_rwsem);
1438 return false;
1439}
1440
93dfe2ac 1441void f2fs_wait_on_page_writeback(struct page *page,
5514f0aa 1442 enum page_type type)
93dfe2ac 1443{
93dfe2ac 1444 if (PageWriteback(page)) {
4081363f
JK		 1445 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1446
df0f8dc0
CY		 1447 if (is_merged_page(sbi, page, type))
1448 f2fs_submit_merged_bio(sbi, type, WRITE);
93dfe2ac
JK		 1449 wait_on_page_writeback(page);
1450 }
1451}
1452
351df4b2
JK		 1453static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1454{
1455 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1456 struct curseg_info *seg_i;
1457 unsigned char *kaddr;
1458 struct page *page;
1459 block_t start;
1460 int i, j, offset;
1461
1462 start = start_sum_block(sbi);
1463
1464 page = get_meta_page(sbi, start++);
1465 kaddr = (unsigned char *)page_address(page);
1466
1467 /* Step 1: restore nat cache */
1468 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1469 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1470
1471 /* Step 2: restore sit cache */
1472 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1473 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1474 SUM_JOURNAL_SIZE);
1475 offset = 2 * SUM_JOURNAL_SIZE;
1476
1477 /* Step 3: restore summary entries */
1478 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1479 unsigned short blk_off;
1480 unsigned int segno;
1481
1482 seg_i = CURSEG_I(sbi, i);
1483 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1484 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1485 seg_i->next_segno = segno;
1486 reset_curseg(sbi, i, 0);
1487 seg_i->alloc_type = ckpt->alloc_type[i];
1488 seg_i->next_blkoff = blk_off;
1489
1490 if (seg_i->alloc_type == SSR)
1491 blk_off = sbi->blocks_per_seg;
1492
1493 for (j = 0; j < blk_off; j++) {
1494 struct f2fs_summary *s;
1495 s = (struct f2fs_summary *)(kaddr + offset);
1496 seg_i->sum_blk->entries[j] = *s;
1497 offset += SUMMARY_SIZE;
1498 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1499 SUM_FOOTER_SIZE)
1500 continue;
1501
1502 f2fs_put_page(page, 1);
1503 page = NULL;
1504
1505 page = get_meta_page(sbi, start++);
1506 kaddr = (unsigned char *)page_address(page);
1507 offset = 0;
1508 }
1509 }
1510 f2fs_put_page(page, 1);
1511 return 0;
1512}
1513
1514static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1515{
1516 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1517 struct f2fs_summary_block *sum;
1518 struct curseg_info *curseg;
1519 struct page *new;
1520 unsigned short blk_off;
1521 unsigned int segno = 0;
1522 block_t blk_addr = 0;
1523
1524 /* get segment number and block addr */
1525 if (IS_DATASEG(type)) {
1526 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1527 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1528 CURSEG_HOT_DATA]);
119ee914 1529 if (__exist_node_summaries(sbi))
351df4b2
JK		 1530 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1531 else
1532 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1533 } else {
1534 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1535 CURSEG_HOT_NODE]);
1536 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1537 CURSEG_HOT_NODE]);
119ee914 1538 if (__exist_node_summaries(sbi))
351df4b2
JK		 1539 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1540 type - CURSEG_HOT_NODE);
1541 else
1542 blk_addr = GET_SUM_BLOCK(sbi, segno);
1543 }
1544
1545 new = get_meta_page(sbi, blk_addr);
1546 sum = (struct f2fs_summary_block *)page_address(new);
1547
1548 if (IS_NODESEG(type)) {
119ee914 1549 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1550 struct f2fs_summary *ns = &sum->entries[0];
1551 int i;
1552 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1553 ns->version = 0;
1554 ns->ofs_in_node = 0;
1555 }
1556 } else {
d653788a
GZ		 1557 int err;
1558
1559 err = restore_node_summary(sbi, segno, sum);
1560 if (err) {
351df4b2 1561 f2fs_put_page(new, 1);
d653788a 1562 return err;
351df4b2
JK		 1563 }
1564 }
1565 }
1566
1567 /* set uncompleted segment to curseg */
1568 curseg = CURSEG_I(sbi, type);
1569 mutex_lock(&curseg->curseg_mutex);
1570 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1571 curseg->next_segno = segno;
1572 reset_curseg(sbi, type, 0);
1573 curseg->alloc_type = ckpt->alloc_type[type];
1574 curseg->next_blkoff = blk_off;
1575 mutex_unlock(&curseg->curseg_mutex);
1576 f2fs_put_page(new, 1);
1577 return 0;
1578}
1579
1580static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1581{
1582 int type = CURSEG_HOT_DATA;
e4fc5fbf 1583 int err;
351df4b2 1584
25ca923b 1585 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1586 int npages = npages_for_summary_flush(sbi, true);
1587
1588 if (npages >= 2)
1589 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1590 META_CP);
1591
351df4b2
JK		 1592 /* restore for compacted data summary */
1593 if (read_compacted_summaries(sbi))
1594 return -EINVAL;
1595 type = CURSEG_HOT_NODE;
1596 }
1597
119ee914 1598 if (__exist_node_summaries(sbi))
3fa06d7b
CY		 1599 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1600 NR_CURSEG_TYPE - type, META_CP);
1601
e4fc5fbf
CY		 1602 for (; type <= CURSEG_COLD_NODE; type++) {
1603 err = read_normal_summaries(sbi, type);
1604 if (err)
1605 return err;
1606 }
1607
351df4b2
JK		 1608 return 0;
1609}
1610
1611static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1612{
1613 struct page *page;
1614 unsigned char *kaddr;
1615 struct f2fs_summary *summary;
1616 struct curseg_info *seg_i;
1617 int written_size = 0;
1618 int i, j;
1619
1620 page = grab_meta_page(sbi, blkaddr++);
1621 kaddr = (unsigned char *)page_address(page);
1622
1623 /* Step 1: write nat cache */
1624 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1625 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1626 written_size += SUM_JOURNAL_SIZE;
1627
1628 /* Step 2: write sit cache */
1629 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1630 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1631 SUM_JOURNAL_SIZE);
1632 written_size += SUM_JOURNAL_SIZE;
1633
351df4b2
JK		 1634 /* Step 3: write summary entries */
1635 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1636 unsigned short blkoff;
1637 seg_i = CURSEG_I(sbi, i);
1638 if (sbi->ckpt->alloc_type[i] == SSR)
1639 blkoff = sbi->blocks_per_seg;
1640 else
1641 blkoff = curseg_blkoff(sbi, i);
1642
1643 for (j = 0; j < blkoff; j++) {
1644 if (!page) {
1645 page = grab_meta_page(sbi, blkaddr++);
1646 kaddr = (unsigned char *)page_address(page);
1647 written_size = 0;
1648 }
1649 summary = (struct f2fs_summary *)(kaddr + written_size);
1650 *summary = seg_i->sum_blk->entries[j];
1651 written_size += SUMMARY_SIZE;
351df4b2
JK		 1652
1653 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1654 SUM_FOOTER_SIZE)
1655 continue;
1656
e8d61a74 1657 set_page_dirty(page);
351df4b2
JK		 1658 f2fs_put_page(page, 1);
1659 page = NULL;
1660 }
1661 }
e8d61a74
CY		 1662 if (page) {
1663 set_page_dirty(page);
351df4b2 1664 f2fs_put_page(page, 1);
e8d61a74 1665 }
351df4b2
JK		 1666}
1667
1668static void write_normal_summaries(struct f2fs_sb_info *sbi,
1669 block_t blkaddr, int type)
1670{
1671 int i, end;
1672 if (IS_DATASEG(type))
1673 end = type + NR_CURSEG_DATA_TYPE;
1674 else
1675 end = type + NR_CURSEG_NODE_TYPE;
1676
1677 for (i = type; i < end; i++) {
1678 struct curseg_info *sum = CURSEG_I(sbi, i);
1679 mutex_lock(&sum->curseg_mutex);
1680 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1681 mutex_unlock(&sum->curseg_mutex);
1682 }
1683}
1684
1685void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1686{
25ca923b 1687 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1688 write_compacted_summaries(sbi, start_blk);
1689 else
1690 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1691}
1692
1693void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1694{
119ee914 1695 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1696}
1697
1698int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1699 unsigned int val, int alloc)
1700{
1701 int i;
1702
1703 if (type == NAT_JOURNAL) {
1704 for (i = 0; i < nats_in_cursum(sum); i++) {
1705 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1706 return i;
1707 }
1708 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1709 return update_nats_in_cursum(sum, 1);
1710 } else if (type == SIT_JOURNAL) {
1711 for (i = 0; i < sits_in_cursum(sum); i++)
1712 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1713 return i;
1714 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1715 return update_sits_in_cursum(sum, 1);
1716 }
1717 return -1;
1718}
1719
1720static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1721 unsigned int segno)
1722{
2cc22186 1723 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1724}
1725
1726static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1727 unsigned int start)
1728{
1729 struct sit_info *sit_i = SIT_I(sbi);
1730 struct page *src_page, *dst_page;
1731 pgoff_t src_off, dst_off;
1732 void *src_addr, *dst_addr;
1733
1734 src_off = current_sit_addr(sbi, start);
1735 dst_off = next_sit_addr(sbi, src_off);
1736
1737 /* get current sit block page without lock */
1738 src_page = get_meta_page(sbi, src_off);
1739 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1740 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1741
1742 src_addr = page_address(src_page);
1743 dst_addr = page_address(dst_page);
1744 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1745
1746 set_page_dirty(dst_page);
1747 f2fs_put_page(src_page, 1);
1748
1749 set_to_next_sit(sit_i, start);
1750
1751 return dst_page;
1752}
1753
184a5cd2
CY		 1754static struct sit_entry_set *grab_sit_entry_set(void)
1755{
1756 struct sit_entry_set *ses =
80c54505 1757 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1758
1759 ses->entry_cnt = 0;
1760 INIT_LIST_HEAD(&ses->set_list);
1761 return ses;
1762}
1763
1764static void release_sit_entry_set(struct sit_entry_set *ses)
1765{
1766 list_del(&ses->set_list);
1767 kmem_cache_free(sit_entry_set_slab, ses);
1768}
1769
1770static void adjust_sit_entry_set(struct sit_entry_set *ses,
1771 struct list_head *head)
1772{
1773 struct sit_entry_set *next = ses;
1774
1775 if (list_is_last(&ses->set_list, head))
1776 return;
1777
1778 list_for_each_entry_continue(next, head, set_list)
1779 if (ses->entry_cnt <= next->entry_cnt)
1780 break;
1781
1782 list_move_tail(&ses->set_list, &next->set_list);
1783}
1784
1785static void add_sit_entry(unsigned int segno, struct list_head *head)
1786{
1787 struct sit_entry_set *ses;
1788 unsigned int start_segno = START_SEGNO(segno);
1789
1790 list_for_each_entry(ses, head, set_list) {
1791 if (ses->start_segno == start_segno) {
1792 ses->entry_cnt++;
1793 adjust_sit_entry_set(ses, head);
1794 return;
1795 }
1796 }
1797
1798 ses = grab_sit_entry_set();
1799
1800 ses->start_segno = start_segno;
1801 ses->entry_cnt++;
1802 list_add(&ses->set_list, head);
1803}
1804
1805static void add_sits_in_set(struct f2fs_sb_info *sbi)
1806{
1807 struct f2fs_sm_info *sm_info = SM_I(sbi);
1808 struct list_head *set_list = &sm_info->sit_entry_set;
1809 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1810 unsigned int segno;
1811
7cd8558b 1812 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1813 add_sit_entry(segno, set_list);
1814}
1815
1816static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1817{
1818 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1819 struct f2fs_summary_block *sum = curseg->sum_blk;
1820 int i;
1821
184a5cd2
CY		 1822 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1823 unsigned int segno;
1824 bool dirtied;
1825
1826 segno = le32_to_cpu(segno_in_journal(sum, i));
1827 dirtied = __mark_sit_entry_dirty(sbi, segno);
1828
1829 if (!dirtied)
1830 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1831 }
184a5cd2 1832 update_sits_in_cursum(sum, -sits_in_cursum(sum));
351df4b2
JK		 1833}
1834
0a8165d7 1835/*
351df4b2
JK		 1836 * CP calls this function, which flushes SIT entries including sit_journal,
1837 * and moves prefree segs to free segs.
1838 */
4b2fecc8 1839void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1840{
1841 struct sit_info *sit_i = SIT_I(sbi);
1842 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1843 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1844 struct f2fs_summary_block *sum = curseg->sum_blk;
184a5cd2
CY		 1845 struct sit_entry_set *ses, *tmp;
1846 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1847 bool to_journal = true;
4b2fecc8 1848 struct seg_entry *se;
351df4b2
JK		 1849
1850 mutex_lock(&curseg->curseg_mutex);
1851 mutex_lock(&sit_i->sentry_lock);
1852
2b11a74b
WL		 1853 if (!sit_i->dirty_sentries)
1854 goto out;
1855
351df4b2 1856 /*
184a5cd2
CY		 1857 * add and account sit entries of dirty bitmap in sit entry
1858 * set temporarily
351df4b2 1859 */
184a5cd2 1860 add_sits_in_set(sbi);
351df4b2 1861
184a5cd2
CY		 1862 /*
1863 * if there are no enough space in journal to store dirty sit
1864 * entries, remove all entries from journal and add and account
1865 * them in sit entry set.
1866 */
1867 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1868 remove_sits_in_journal(sbi);
b2955550 1869
184a5cd2
CY		 1870 /*
1871 * there are two steps to flush sit entries:
1872 * #1, flush sit entries to journal in current cold data summary block.
1873 * #2, flush sit entries to sit page.
1874 */
1875 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1876 struct page *page = NULL;
184a5cd2
CY		 1877 struct f2fs_sit_block *raw_sit = NULL;
1878 unsigned int start_segno = ses->start_segno;
1879 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1880 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1881 unsigned int segno = start_segno;
1882
1883 if (to_journal &&
1884 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1885 to_journal = false;
1886
1887 if (!to_journal) {
1888 page = get_next_sit_page(sbi, start_segno);
1889 raw_sit = page_address(page);
351df4b2 1890 }
351df4b2 1891
184a5cd2
CY		 1892 /* flush dirty sit entries in region of current sit set */
1893 for_each_set_bit_from(segno, bitmap, end) {
1894 int offset, sit_offset;
4b2fecc8
JK		 1895
1896 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 1897
1898 /* add discard candidates */
d7bc2484 1899 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 1900 cpc->trim_start = segno;
1901 add_discard_addrs(sbi, cpc);
1902 }
184a5cd2
CY		 1903
1904 if (to_journal) {
1905 offset = lookup_journal_in_cursum(sum,
1906 SIT_JOURNAL, segno, 1);
1907 f2fs_bug_on(sbi, offset < 0);
1908 segno_in_journal(sum, offset) =
1909 cpu_to_le32(segno);
1910 seg_info_to_raw_sit(se,
1911 &sit_in_journal(sum, offset));
1912 } else {
1913 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1914 seg_info_to_raw_sit(se,
1915 &raw_sit->entries[sit_offset]);
1916 }
351df4b2 1917
184a5cd2
CY		 1918 __clear_bit(segno, bitmap);
1919 sit_i->dirty_sentries--;
1920 ses->entry_cnt--;
351df4b2
JK		 1921 }
1922
184a5cd2
CY		 1923 if (!to_journal)
1924 f2fs_put_page(page, 1);
1925
1926 f2fs_bug_on(sbi, ses->entry_cnt);
1927 release_sit_entry_set(ses);
351df4b2 1928 }
184a5cd2
CY		 1929
1930 f2fs_bug_on(sbi, !list_empty(head));
1931 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 1932out:
4b2fecc8
JK		 1933 if (cpc->reason == CP_DISCARD) {
1934 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1935 add_discard_addrs(sbi, cpc);
1936 }
351df4b2
JK		 1937 mutex_unlock(&sit_i->sentry_lock);
1938 mutex_unlock(&curseg->curseg_mutex);
1939
351df4b2
JK		 1940 set_prefree_as_free_segments(sbi);
1941}
1942
1943static int build_sit_info(struct f2fs_sb_info *sbi)
1944{
1945 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1946 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1947 struct sit_info *sit_i;
1948 unsigned int sit_segs, start;
1949 char *src_bitmap, *dst_bitmap;
1950 unsigned int bitmap_size;
1951
1952 /* allocate memory for SIT information */
1953 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1954 if (!sit_i)
1955 return -ENOMEM;
1956
1957 SM_I(sbi)->sit_info = sit_i;
1958
39307a8e
JK		 1959 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
1960 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 1961 if (!sit_i->sentries)
1962 return -ENOMEM;
1963
7cd8558b 1964 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 1965 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 1966 if (!sit_i->dirty_sentries_bitmap)
1967 return -ENOMEM;
1968
7cd8558b 1969 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 1970 sit_i->sentries[start].cur_valid_map
1971 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1972 sit_i->sentries[start].ckpt_valid_map
1973 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 1974 sit_i->sentries[start].discard_map
1975 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1976 if (!sit_i->sentries[start].cur_valid_map ||
1977 !sit_i->sentries[start].ckpt_valid_map ||
1978 !sit_i->sentries[start].discard_map)
351df4b2
JK		 1979 return -ENOMEM;
1980 }
1981
60a3b782
JK		 1982 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1983 if (!sit_i->tmp_map)
1984 return -ENOMEM;
1985
351df4b2 1986 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 1987 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
1988 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 1989 if (!sit_i->sec_entries)
1990 return -ENOMEM;
1991 }
1992
1993 /* get information related with SIT */
1994 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1995
1996 /* setup SIT bitmap from ckeckpoint pack */
1997 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1998 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1999
79b5793b 2000 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2001 if (!dst_bitmap)
2002 return -ENOMEM;
351df4b2
JK		 2003
2004 /* init SIT information */
2005 sit_i->s_ops = &default_salloc_ops;
2006
2007 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2008 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2009 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2010 sit_i->sit_bitmap = dst_bitmap;
2011 sit_i->bitmap_size = bitmap_size;
2012 sit_i->dirty_sentries = 0;
2013 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2014 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2015 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2016 mutex_init(&sit_i->sentry_lock);
2017 return 0;
2018}
2019
2020static int build_free_segmap(struct f2fs_sb_info *sbi)
2021{
351df4b2
JK		 2022 struct free_segmap_info *free_i;
2023 unsigned int bitmap_size, sec_bitmap_size;
2024
2025 /* allocate memory for free segmap information */
2026 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2027 if (!free_i)
2028 return -ENOMEM;
2029
2030 SM_I(sbi)->free_info = free_i;
2031
7cd8558b 2032 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2033 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2034 if (!free_i->free_segmap)
2035 return -ENOMEM;
2036
7cd8558b 2037 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2038 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2039 if (!free_i->free_secmap)
2040 return -ENOMEM;
2041
2042 /* set all segments as dirty temporarily */
2043 memset(free_i->free_segmap, 0xff, bitmap_size);
2044 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2045
2046 /* init free segmap information */
7cd8558b 2047 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2048 free_i->free_segments = 0;
2049 free_i->free_sections = 0;
1a118ccf 2050 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2051 return 0;
2052}
2053
2054static int build_curseg(struct f2fs_sb_info *sbi)
2055{
1042d60f 2056 struct curseg_info *array;
351df4b2
JK		 2057 int i;
2058
b434babf 2059 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2060 if (!array)
2061 return -ENOMEM;
2062
2063 SM_I(sbi)->curseg_array = array;
2064
2065 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2066 mutex_init(&array[i].curseg_mutex);
2067 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2068 if (!array[i].sum_blk)
2069 return -ENOMEM;
2070 array[i].segno = NULL_SEGNO;
2071 array[i].next_blkoff = 0;
2072 }
2073 return restore_curseg_summaries(sbi);
2074}
2075
2076static void build_sit_entries(struct f2fs_sb_info *sbi)
2077{
2078 struct sit_info *sit_i = SIT_I(sbi);
2079 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2080 struct f2fs_summary_block *sum = curseg->sum_blk;
74de593a
CY		 2081 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2082 unsigned int i, start, end;
2083 unsigned int readed, start_blk = 0;
90a893c7 2084 int nrpages = MAX_BIO_BLOCKS(sbi);
351df4b2 2085
74de593a 2086 do {
662befda 2087 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
74de593a
CY		 2088
2089 start = start_blk * sit_i->sents_per_block;
2090 end = (start_blk + readed) * sit_i->sents_per_block;
2091
7cd8558b 2092 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2093 struct seg_entry *se = &sit_i->sentries[start];
2094 struct f2fs_sit_block *sit_blk;
2095 struct f2fs_sit_entry sit;
2096 struct page *page;
2097
2098 mutex_lock(&curseg->curseg_mutex);
2099 for (i = 0; i < sits_in_cursum(sum); i++) {
6c311ec6
CF		 2100 if (le32_to_cpu(segno_in_journal(sum, i))
2101 == start) {
74de593a
CY		 2102 sit = sit_in_journal(sum, i);
2103 mutex_unlock(&curseg->curseg_mutex);
2104 goto got_it;
2105 }
351df4b2 2106 }
74de593a
CY		 2107 mutex_unlock(&curseg->curseg_mutex);
2108
2109 page = get_current_sit_page(sbi, start);
2110 sit_blk = (struct f2fs_sit_block *)page_address(page);
2111 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2112 f2fs_put_page(page, 1);
351df4b2 2113got_it:
74de593a
CY		 2114 check_block_count(sbi, start, &sit);
2115 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2116
2117 /* build discard map only one time */
2118 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2119 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2120
74de593a
CY		 2121 if (sbi->segs_per_sec > 1) {
2122 struct sec_entry *e = get_sec_entry(sbi, start);
2123 e->valid_blocks += se->valid_blocks;
2124 }
351df4b2 2125 }
74de593a
CY		 2126 start_blk += readed;
2127 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2128}
2129
2130static void init_free_segmap(struct f2fs_sb_info *sbi)
2131{
2132 unsigned int start;
2133 int type;
2134
7cd8558b 2135 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2136 struct seg_entry *sentry = get_seg_entry(sbi, start);
2137 if (!sentry->valid_blocks)
2138 __set_free(sbi, start);
2139 }
2140
2141 /* set use the current segments */
2142 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2143 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2144 __set_test_and_inuse(sbi, curseg_t->segno);
2145 }
2146}
2147
2148static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2149{
2150 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2151 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2152 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2153 unsigned short valid_blocks;
2154
8736fbf0 2155 while (1) {
351df4b2 2156 /* find dirty segment based on free segmap */
7cd8558b
JK		 2157 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2158 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2159 break;
2160 offset = segno + 1;
2161 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2162 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2163 continue;
ec325b52
JK		 2164 if (valid_blocks > sbi->blocks_per_seg) {
2165 f2fs_bug_on(sbi, 1);
2166 continue;
2167 }
351df4b2
JK		 2168 mutex_lock(&dirty_i->seglist_lock);
2169 __locate_dirty_segment(sbi, segno, DIRTY);
2170 mutex_unlock(&dirty_i->seglist_lock);
2171 }
2172}
2173
5ec4e49f 2174static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2175{
2176 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2177 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2178
39307a8e 2179 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2180 if (!dirty_i->victim_secmap)
351df4b2
JK		 2181 return -ENOMEM;
2182 return 0;
2183}
2184
2185static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2186{
2187 struct dirty_seglist_info *dirty_i;
2188 unsigned int bitmap_size, i;
2189
2190 /* allocate memory for dirty segments list information */
2191 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2192 if (!dirty_i)
2193 return -ENOMEM;
2194
2195 SM_I(sbi)->dirty_info = dirty_i;
2196 mutex_init(&dirty_i->seglist_lock);
2197
7cd8558b 2198 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2199
2200 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2201 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2202 if (!dirty_i->dirty_segmap[i])
2203 return -ENOMEM;
2204 }
2205
2206 init_dirty_segmap(sbi);
5ec4e49f 2207 return init_victim_secmap(sbi);
351df4b2
JK		 2208}
2209
0a8165d7 2210/*
351df4b2
JK		 2211 * Update min, max modified time for cost-benefit GC algorithm
2212 */
2213static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2214{
2215 struct sit_info *sit_i = SIT_I(sbi);
2216 unsigned int segno;
2217
2218 mutex_lock(&sit_i->sentry_lock);
2219
2220 sit_i->min_mtime = LLONG_MAX;
2221
7cd8558b 2222 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2223 unsigned int i;
2224 unsigned long long mtime = 0;
2225
2226 for (i = 0; i < sbi->segs_per_sec; i++)
2227 mtime += get_seg_entry(sbi, segno + i)->mtime;
2228
2229 mtime = div_u64(mtime, sbi->segs_per_sec);
2230
2231 if (sit_i->min_mtime > mtime)
2232 sit_i->min_mtime = mtime;
2233 }
2234 sit_i->max_mtime = get_mtime(sbi);
2235 mutex_unlock(&sit_i->sentry_lock);
2236}
2237
2238int build_segment_manager(struct f2fs_sb_info *sbi)
2239{
2240 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2241 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2242 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2243 int err;
2244
2245 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2246 if (!sm_info)
2247 return -ENOMEM;
2248
2249 /* init sm info */
2250 sbi->sm_info = sm_info;
351df4b2
JK		 2251 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2252 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2253 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2254 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2255 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2256 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2257 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2258 sm_info->rec_prefree_segments = sm_info->main_segments *
2259 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2260 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2261 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2262 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2263
7fd9e544
JK		 2264 INIT_LIST_HEAD(&sm_info->discard_list);
2265 sm_info->nr_discards = 0;
2266 sm_info->max_discards = 0;
2267
bba681cb
JK		 2268 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2269
184a5cd2
CY		 2270 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2271
b270ad6f 2272 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2273 err = create_flush_cmd_control(sbi);
2274 if (err)
a688b9d9 2275 return err;
6b4afdd7
JK		 2276 }
2277
351df4b2
JK		 2278 err = build_sit_info(sbi);
2279 if (err)
2280 return err;
2281 err = build_free_segmap(sbi);
2282 if (err)
2283 return err;
2284 err = build_curseg(sbi);
2285 if (err)
2286 return err;
2287
2288 /* reinit free segmap based on SIT */
2289 build_sit_entries(sbi);
2290
2291 init_free_segmap(sbi);
2292 err = build_dirty_segmap(sbi);
2293 if (err)
2294 return err;
2295
2296 init_min_max_mtime(sbi);
2297 return 0;
2298}
2299
2300static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2301 enum dirty_type dirty_type)
2302{
2303 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2304
2305 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2306 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2307 dirty_i->nr_dirty[dirty_type] = 0;
2308 mutex_unlock(&dirty_i->seglist_lock);
2309}
2310
5ec4e49f 2311static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2312{
2313 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2314 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2315}
2316
2317static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2318{
2319 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2320 int i;
2321
2322 if (!dirty_i)
2323 return;
2324
2325 /* discard pre-free/dirty segments list */
2326 for (i = 0; i < NR_DIRTY_TYPE; i++)
2327 discard_dirty_segmap(sbi, i);
2328
5ec4e49f 2329 destroy_victim_secmap(sbi);
351df4b2
JK		 2330 SM_I(sbi)->dirty_info = NULL;
2331 kfree(dirty_i);
2332}
2333
2334static void destroy_curseg(struct f2fs_sb_info *sbi)
2335{
2336 struct curseg_info *array = SM_I(sbi)->curseg_array;
2337 int i;
2338
2339 if (!array)
2340 return;
2341 SM_I(sbi)->curseg_array = NULL;
2342 for (i = 0; i < NR_CURSEG_TYPE; i++)
2343 kfree(array[i].sum_blk);
2344 kfree(array);
2345}
2346
2347static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2348{
2349 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2350 if (!free_i)
2351 return;
2352 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2353 kvfree(free_i->free_segmap);
2354 kvfree(free_i->free_secmap);
351df4b2
JK		 2355 kfree(free_i);
2356}
2357
2358static void destroy_sit_info(struct f2fs_sb_info *sbi)
2359{
2360 struct sit_info *sit_i = SIT_I(sbi);
2361 unsigned int start;
2362
2363 if (!sit_i)
2364 return;
2365
2366 if (sit_i->sentries) {
7cd8558b 2367 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2368 kfree(sit_i->sentries[start].cur_valid_map);
2369 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2370 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2371 }
2372 }
60a3b782
JK		 2373 kfree(sit_i->tmp_map);
2374
39307a8e
JK		 2375 kvfree(sit_i->sentries);
2376 kvfree(sit_i->sec_entries);
2377 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2378
2379 SM_I(sbi)->sit_info = NULL;
2380 kfree(sit_i->sit_bitmap);
2381 kfree(sit_i);
2382}
2383
2384void destroy_segment_manager(struct f2fs_sb_info *sbi)
2385{
2386 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2387
3b03f724
CY		 2388 if (!sm_info)
2389 return;
2163d198 2390 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2391 destroy_dirty_segmap(sbi);
2392 destroy_curseg(sbi);
2393 destroy_free_segmap(sbi);
2394 destroy_sit_info(sbi);
2395 sbi->sm_info = NULL;
2396 kfree(sm_info);
2397}
7fd9e544
JK		 2398
2399int __init create_segment_manager_caches(void)
2400{
2401 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2402 sizeof(struct discard_entry));
7fd9e544 2403 if (!discard_entry_slab)
184a5cd2
CY		 2404 goto fail;
2405
2406 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2407 sizeof(struct sit_entry_set));
184a5cd2
CY		 2408 if (!sit_entry_set_slab)
2409 goto destory_discard_entry;
88b88a66
JK		 2410
2411 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2412 sizeof(struct inmem_pages));
2413 if (!inmem_entry_slab)
2414 goto destroy_sit_entry_set;
7fd9e544 2415 return 0;
184a5cd2 2416
88b88a66
JK		 2417destroy_sit_entry_set:
2418 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2419destory_discard_entry:
2420 kmem_cache_destroy(discard_entry_slab);
2421fail:
2422 return -ENOMEM;
7fd9e544
JK		 2423}
2424
2425void destroy_segment_manager_caches(void)
2426{
184a5cd2 2427 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2428 kmem_cache_destroy(discard_entry_slab);
88b88a66 2429 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2430}
Linux kernel - Deliverables
This page took 0.246017 seconds and 5 git commands to generate.