projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: increase the number of max hard links
[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>
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;
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);
408e9375 299 f2fs_gc(sbi);
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) ||
88a70a69 319 !available_free_memory(sbi, INO_ENTRIES))
4660f9c0
JK		 320 f2fs_sync_fs(sbi->sb, true);
321}
322
2163d198 323static int issue_flush_thread(void *data)
6b4afdd7
JK		 324{
325 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 326 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
327 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 328repeat:
329 if (kthread_should_stop())
330 return 0;
331
721bd4d5 332 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 333 struct bio *bio = bio_alloc(GFP_NOIO, 0);
334 struct flush_cmd *cmd, *next;
335 int ret;
336
721bd4d5
GZ		 337 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
338 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
339
6b4afdd7
JK		 340 bio->bi_bdev = sbi->sb->s_bdev;
341 ret = submit_bio_wait(WRITE_FLUSH, bio);
342
721bd4d5
GZ		 343 llist_for_each_entry_safe(cmd, next,
344 fcc->dispatch_list, llnode) {
6b4afdd7 345 cmd->ret = ret;
6b4afdd7
JK		 346 complete(&cmd->wait);
347 }
a4ed23f2 348 bio_put(bio);
a688b9d9 349 fcc->dispatch_list = NULL;
6b4afdd7
JK		 350 }
351
a688b9d9 352 wait_event_interruptible(*q,
721bd4d5 353 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 354 goto repeat;
355}
356
357int f2fs_issue_flush(struct f2fs_sb_info *sbi)
358{
a688b9d9 359 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 360 struct flush_cmd cmd;
6b4afdd7 361
24a9ee0f
JK		 362 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
363 test_opt(sbi, FLUSH_MERGE));
364
0f7b2abd
JK		 365 if (test_opt(sbi, NOBARRIER))
366 return 0;
367
6b4afdd7
JK		 368 if (!test_opt(sbi, FLUSH_MERGE))
369 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
370
adf8d90b 371 init_completion(&cmd.wait);
6b4afdd7 372
721bd4d5 373 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 374
a688b9d9
GZ		 375 if (!fcc->dispatch_list)
376 wake_up(&fcc->flush_wait_queue);
6b4afdd7 377
adf8d90b
CY		 378 wait_for_completion(&cmd.wait);
379
380 return cmd.ret;
6b4afdd7
JK		 381}
382
2163d198
GZ		 383int create_flush_cmd_control(struct f2fs_sb_info *sbi)
384{
385 dev_t dev = sbi->sb->s_bdev->bd_dev;
386 struct flush_cmd_control *fcc;
387 int err = 0;
388
389 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
390 if (!fcc)
391 return -ENOMEM;
2163d198 392 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 393 init_llist_head(&fcc->issue_list);
6b2920a5 394 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 395 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
396 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
397 if (IS_ERR(fcc->f2fs_issue_flush)) {
398 err = PTR_ERR(fcc->f2fs_issue_flush);
399 kfree(fcc);
6b2920a5 400 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 401 return err;
402 }
2163d198
GZ		 403
404 return err;
405}
406
407void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
408{
6b2920a5 409 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 410
411 if (fcc && fcc->f2fs_issue_flush)
412 kthread_stop(fcc->f2fs_issue_flush);
413 kfree(fcc);
6b2920a5 414 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 415}
416
351df4b2
JK		 417static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
418 enum dirty_type dirty_type)
419{
420 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
421
422 /* need not be added */
423 if (IS_CURSEG(sbi, segno))
424 return;
425
426 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
427 dirty_i->nr_dirty[dirty_type]++;
428
429 if (dirty_type == DIRTY) {
430 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 431 enum dirty_type t = sentry->type;
b2f2c390 432
ec325b52
JK		 433 if (unlikely(t >= DIRTY)) {
434 f2fs_bug_on(sbi, 1);
435 return;
436 }
4625d6aa
CL		 437 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
438 dirty_i->nr_dirty[t]++;
351df4b2
JK		 439 }
440}
441
442static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
443 enum dirty_type dirty_type)
444{
445 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
446
447 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
448 dirty_i->nr_dirty[dirty_type]--;
449
450 if (dirty_type == DIRTY) {
4625d6aa
CL		 451 struct seg_entry *sentry = get_seg_entry(sbi, segno);
452 enum dirty_type t = sentry->type;
453
454 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
455 dirty_i->nr_dirty[t]--;
b2f2c390 456
5ec4e49f
JK		 457 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
458 clear_bit(GET_SECNO(sbi, segno),
459 dirty_i->victim_secmap);
351df4b2
JK		 460 }
461}
462
0a8165d7 463/*
351df4b2
JK		 464 * Should not occur error such as -ENOMEM.
465 * Adding dirty entry into seglist is not critical operation.
466 * If a given segment is one of current working segments, it won't be added.
467 */
8d8451af 468static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 469{
470 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
471 unsigned short valid_blocks;
472
473 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
474 return;
475
476 mutex_lock(&dirty_i->seglist_lock);
477
478 valid_blocks = get_valid_blocks(sbi, segno, 0);
479
480 if (valid_blocks == 0) {
481 __locate_dirty_segment(sbi, segno, PRE);
482 __remove_dirty_segment(sbi, segno, DIRTY);
483 } else if (valid_blocks < sbi->blocks_per_seg) {
484 __locate_dirty_segment(sbi, segno, DIRTY);
485 } else {
486 /* Recovery routine with SSR needs this */
487 __remove_dirty_segment(sbi, segno, DIRTY);
488 }
489
490 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 491}
492
1e87a78d 493static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 494 block_t blkstart, block_t blklen)
495{
55cf9cb6
CY		 496 sector_t start = SECTOR_FROM_BLOCK(blkstart);
497 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 498 struct seg_entry *se;
499 unsigned int offset;
500 block_t i;
501
502 for (i = blkstart; i < blkstart + blklen; i++) {
503 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
504 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
505
506 if (!f2fs_test_and_set_bit(offset, se->discard_map))
507 sbi->discard_blks--;
508 }
1661d07c 509 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 510 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
511}
512
e90c2d28 513bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 514{
40a02be1
JK		 515 int err = -ENOTSUPP;
516
517 if (test_opt(sbi, DISCARD)) {
518 struct seg_entry *se = get_seg_entry(sbi,
519 GET_SEGNO(sbi, blkaddr));
520 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
521
522 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 523 return false;
40a02be1
JK		 524
525 err = f2fs_issue_discard(sbi, blkaddr, 1);
526 }
527
e90c2d28 528 if (err) {
381722d2 529 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 530 return true;
531 }
532 return false;
37208879
JK		 533}
534
adf4983b 535static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 536 struct cp_control *cpc, struct seg_entry *se,
537 unsigned int start, unsigned int end)
b2955550
JK		 538{
539 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 540 struct discard_entry *new, *last;
541
542 if (!list_empty(head)) {
543 last = list_last_entry(head, struct discard_entry, list);
544 if (START_BLOCK(sbi, cpc->trim_start) + start ==
545 last->blkaddr + last->len) {
546 last->len += end - start;
547 goto done;
548 }
549 }
550
551 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
552 INIT_LIST_HEAD(&new->list);
553 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
554 new->len = end - start;
555 list_add_tail(&new->list, head);
556done:
557 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 558}
559
560static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
561{
b2955550
JK		 562 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
563 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 564 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 565 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
566 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 567 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 568 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 569 unsigned int start = 0, end = -1;
4b2fecc8 570 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 571 int i;
572
a66cdd98 573 if (se->valid_blocks == max_blocks)
b2955550
JK		 574 return;
575
a66cdd98
JK		 576 if (!force) {
577 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 578 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
579 return;
4b2fecc8
JK		 580 }
581
b2955550
JK		 582 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
583 for (i = 0; i < entries; i++)
a66cdd98 584 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 585 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 586
4b2fecc8 587 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 588 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
589 if (start >= max_blocks)
590 break;
591
592 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 593 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 594 }
595}
596
4b2fecc8
JK		 597void release_discard_addrs(struct f2fs_sb_info *sbi)
598{
599 struct list_head *head = &(SM_I(sbi)->discard_list);
600 struct discard_entry *entry, *this;
601
602 /* drop caches */
603 list_for_each_entry_safe(entry, this, head, list) {
604 list_del(&entry->list);
605 kmem_cache_free(discard_entry_slab, entry);
606 }
607}
608
0a8165d7 609/*
351df4b2
JK		 610 * Should call clear_prefree_segments after checkpoint is done.
611 */
612static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
613{
614 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 615 unsigned int segno;
351df4b2
JK		 616
617 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 618 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 619 __set_test_and_free(sbi, segno);
351df4b2
JK		 620 mutex_unlock(&dirty_i->seglist_lock);
621}
622
836b5a63 623void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 624{
b2955550 625 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 626 struct discard_entry *entry, *this;
351df4b2 627 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 628 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 629 unsigned int start = 0, end = -1;
351df4b2
JK		 630
631 mutex_lock(&dirty_i->seglist_lock);
29e59c14 632
351df4b2 633 while (1) {
29e59c14 634 int i;
7cd8558b
JK		 635 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
636 if (start >= MAIN_SEGS(sbi))
351df4b2 637 break;
7cd8558b
JK		 638 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
639 start + 1);
29e59c14
CL		 640
641 for (i = start; i < end; i++)
642 clear_bit(i, prefree_map);
643
644 dirty_i->nr_dirty[PRE] -= end - start;
645
646 if (!test_opt(sbi, DISCARD))
647 continue;
351df4b2 648
37208879
JK		 649 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
650 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 651 }
652 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 653
654 /* send small discards */
2d7b822a 655 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 656 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
657 goto skip;
37208879 658 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 659 cpc->trimmed += entry->len;
836b5a63 660skip:
b2955550
JK		 661 list_del(&entry->list);
662 SM_I(sbi)->nr_discards -= entry->len;
663 kmem_cache_free(discard_entry_slab, entry);
664 }
351df4b2
JK		 665}
666
184a5cd2 667static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 668{
669 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 670
671 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 672 sit_i->dirty_sentries++;
184a5cd2
CY		 673 return false;
674 }
675
676 return true;
351df4b2
JK		 677}
678
679static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
680 unsigned int segno, int modified)
681{
682 struct seg_entry *se = get_seg_entry(sbi, segno);
683 se->type = type;
684 if (modified)
685 __mark_sit_entry_dirty(sbi, segno);
686}
687
688static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
689{
690 struct seg_entry *se;
691 unsigned int segno, offset;
692 long int new_vblocks;
693
694 segno = GET_SEGNO(sbi, blkaddr);
695
696 se = get_seg_entry(sbi, segno);
697 new_vblocks = se->valid_blocks + del;
491c0854 698 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 699
9850cf4a 700 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 701 (new_vblocks > sbi->blocks_per_seg)));
702
703 se->valid_blocks = new_vblocks;
704 se->mtime = get_mtime(sbi);
705 SIT_I(sbi)->max_mtime = se->mtime;
706
707 /* Update valid block bitmap */
708 if (del > 0) {
52aca074 709 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 710 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 711 if (!f2fs_test_and_set_bit(offset, se->discard_map))
712 sbi->discard_blks--;
351df4b2 713 } else {
52aca074 714 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 715 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 716 if (f2fs_test_and_clear_bit(offset, se->discard_map))
717 sbi->discard_blks++;
351df4b2
JK		 718 }
719 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
720 se->ckpt_valid_blocks += del;
721
722 __mark_sit_entry_dirty(sbi, segno);
723
724 /* update total number of valid blocks to be written in ckpt area */
725 SIT_I(sbi)->written_valid_blocks += del;
726
727 if (sbi->segs_per_sec > 1)
728 get_sec_entry(sbi, segno)->valid_blocks += del;
729}
730
5e443818 731void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 732{
5e443818
JK		 733 update_sit_entry(sbi, new, 1);
734 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
735 update_sit_entry(sbi, old, -1);
736
737 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
738 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 739}
740
741void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
742{
743 unsigned int segno = GET_SEGNO(sbi, addr);
744 struct sit_info *sit_i = SIT_I(sbi);
745
9850cf4a 746 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 747 if (addr == NEW_ADDR)
748 return;
749
750 /* add it into sit main buffer */
751 mutex_lock(&sit_i->sentry_lock);
752
753 update_sit_entry(sbi, addr, -1);
754
755 /* add it into dirty seglist */
756 locate_dirty_segment(sbi, segno);
757
758 mutex_unlock(&sit_i->sentry_lock);
759}
760
0a8165d7 761/*
351df4b2
JK		 762 * This function should be resided under the curseg_mutex lock
763 */
764static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 765 struct f2fs_summary *sum)
351df4b2
JK		 766{
767 struct curseg_info *curseg = CURSEG_I(sbi, type);
768 void *addr = curseg->sum_blk;
e79efe3b 769 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 770 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 771}
772
0a8165d7 773/*
351df4b2
JK		 774 * Calculate the number of current summary pages for writing
775 */
3fa06d7b 776int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 777{
351df4b2 778 int valid_sum_count = 0;
9a47938b 779 int i, sum_in_page;
351df4b2
JK		 780
781 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
782 if (sbi->ckpt->alloc_type[i] == SSR)
783 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 784 else {
785 if (for_ra)
786 valid_sum_count += le16_to_cpu(
787 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
788 else
789 valid_sum_count += curseg_blkoff(sbi, i);
790 }
351df4b2
JK		 791 }
792
9a47938b
FL		 793 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
794 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
795 if (valid_sum_count <= sum_in_page)
351df4b2 796 return 1;
9a47938b
FL		 797 else if ((valid_sum_count - sum_in_page) <=
798 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 799 return 2;
800 return 3;
801}
802
0a8165d7 803/*
351df4b2
JK		 804 * Caller should put this summary page
805 */
806struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
807{
808 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
809}
810
381722d2 811void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 812{
813 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 814 void *dst = page_address(page);
815
816 if (src)
817 memcpy(dst, src, PAGE_CACHE_SIZE);
818 else
819 memset(dst, 0, PAGE_CACHE_SIZE);
351df4b2
JK		 820 set_page_dirty(page);
821 f2fs_put_page(page, 1);
822}
823
381722d2
CY		 824static void write_sum_page(struct f2fs_sb_info *sbi,
825 struct f2fs_summary_block *sum_blk, block_t blk_addr)
826{
827 update_meta_page(sbi, (void *)sum_blk, blk_addr);
828}
829
60374688
JK		 830static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
831{
832 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 833 unsigned int segno = curseg->segno + 1;
60374688
JK		 834 struct free_segmap_info *free_i = FREE_I(sbi);
835
7cd8558b 836 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 837 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 838 return 0;
839}
840
0a8165d7 841/*
351df4b2
JK		 842 * Find a new segment from the free segments bitmap to right order
843 * This function should be returned with success, otherwise BUG
844 */
845static void get_new_segment(struct f2fs_sb_info *sbi,
846 unsigned int *newseg, bool new_sec, int dir)
847{
848 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 849 unsigned int segno, secno, zoneno;
7cd8558b 850 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 851 unsigned int hint = *newseg / sbi->segs_per_sec;
852 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
853 unsigned int left_start = hint;
854 bool init = true;
855 int go_left = 0;
856 int i;
857
1a118ccf 858 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 859
860 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
861 segno = find_next_zero_bit(free_i->free_segmap,
7cd8558b 862 MAIN_SEGS(sbi), *newseg + 1);
33afa7fd
JK		 863 if (segno - *newseg < sbi->segs_per_sec -
864 (*newseg % sbi->segs_per_sec))
351df4b2
JK		 865 goto got_it;
866 }
867find_other_zone:
7cd8558b
JK		 868 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
869 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 870 if (dir == ALLOC_RIGHT) {
871 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 872 MAIN_SECS(sbi), 0);
873 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 874 } else {
875 go_left = 1;
876 left_start = hint - 1;
877 }
878 }
879 if (go_left == 0)
880 goto skip_left;
881
882 while (test_bit(left_start, free_i->free_secmap)) {
883 if (left_start > 0) {
884 left_start--;
885 continue;
886 }
887 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 888 MAIN_SECS(sbi), 0);
889 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 890 break;
891 }
892 secno = left_start;
893skip_left:
894 hint = secno;
895 segno = secno * sbi->segs_per_sec;
896 zoneno = secno / sbi->secs_per_zone;
897
898 /* give up on finding another zone */
899 if (!init)
900 goto got_it;
901 if (sbi->secs_per_zone == 1)
902 goto got_it;
903 if (zoneno == old_zoneno)
904 goto got_it;
905 if (dir == ALLOC_LEFT) {
906 if (!go_left && zoneno + 1 >= total_zones)
907 goto got_it;
908 if (go_left && zoneno == 0)
909 goto got_it;
910 }
911 for (i = 0; i < NR_CURSEG_TYPE; i++)
912 if (CURSEG_I(sbi, i)->zone == zoneno)
913 break;
914
915 if (i < NR_CURSEG_TYPE) {
916 /* zone is in user, try another */
917 if (go_left)
918 hint = zoneno * sbi->secs_per_zone - 1;
919 else if (zoneno + 1 >= total_zones)
920 hint = 0;
921 else
922 hint = (zoneno + 1) * sbi->secs_per_zone;
923 init = false;
924 goto find_other_zone;
925 }
926got_it:
927 /* set it as dirty segment in free segmap */
9850cf4a 928 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 929 __set_inuse(sbi, segno);
930 *newseg = segno;
1a118ccf 931 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 932}
933
934static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
935{
936 struct curseg_info *curseg = CURSEG_I(sbi, type);
937 struct summary_footer *sum_footer;
938
939 curseg->segno = curseg->next_segno;
940 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
941 curseg->next_blkoff = 0;
942 curseg->next_segno = NULL_SEGNO;
943
944 sum_footer = &(curseg->sum_blk->footer);
945 memset(sum_footer, 0, sizeof(struct summary_footer));
946 if (IS_DATASEG(type))
947 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
948 if (IS_NODESEG(type))
949 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
950 __set_sit_entry_type(sbi, type, curseg->segno, modified);
951}
952
0a8165d7 953/*
351df4b2
JK		 954 * Allocate a current working segment.
955 * This function always allocates a free segment in LFS manner.
956 */
957static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
958{
959 struct curseg_info *curseg = CURSEG_I(sbi, type);
960 unsigned int segno = curseg->segno;
961 int dir = ALLOC_LEFT;
962
963 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 964 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 965 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
966 dir = ALLOC_RIGHT;
967
968 if (test_opt(sbi, NOHEAP))
969 dir = ALLOC_RIGHT;
970
971 get_new_segment(sbi, &segno, new_sec, dir);
972 curseg->next_segno = segno;
973 reset_curseg(sbi, type, 1);
974 curseg->alloc_type = LFS;
975}
976
977static void __next_free_blkoff(struct f2fs_sb_info *sbi,
978 struct curseg_info *seg, block_t start)
979{
980 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 981 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 982 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 983 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
984 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
985 int i, pos;
986
987 for (i = 0; i < entries; i++)
988 target_map[i] = ckpt_map[i] | cur_map[i];
989
990 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
991
992 seg->next_blkoff = pos;
351df4b2
JK		 993}
994
0a8165d7 995/*
351df4b2
JK		 996 * If a segment is written by LFS manner, next block offset is just obtained
997 * by increasing the current block offset. However, if a segment is written by
998 * SSR manner, next block offset obtained by calling __next_free_blkoff
999 */
1000static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1001 struct curseg_info *seg)
1002{
1003 if (seg->alloc_type == SSR)
1004 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1005 else
1006 seg->next_blkoff++;
1007}
1008
0a8165d7 1009/*
e1c42045 1010 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1011 * manner, so it should recover the existing segment information of valid blocks
1012 */
1013static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1014{
1015 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1016 struct curseg_info *curseg = CURSEG_I(sbi, type);
1017 unsigned int new_segno = curseg->next_segno;
1018 struct f2fs_summary_block *sum_node;
1019 struct page *sum_page;
1020
1021 write_sum_page(sbi, curseg->sum_blk,
1022 GET_SUM_BLOCK(sbi, curseg->segno));
1023 __set_test_and_inuse(sbi, new_segno);
1024
1025 mutex_lock(&dirty_i->seglist_lock);
1026 __remove_dirty_segment(sbi, new_segno, PRE);
1027 __remove_dirty_segment(sbi, new_segno, DIRTY);
1028 mutex_unlock(&dirty_i->seglist_lock);
1029
1030 reset_curseg(sbi, type, 1);
1031 curseg->alloc_type = SSR;
1032 __next_free_blkoff(sbi, curseg, 0);
1033
1034 if (reuse) {
1035 sum_page = get_sum_page(sbi, new_segno);
1036 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1037 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1038 f2fs_put_page(sum_page, 1);
1039 }
1040}
1041
43727527
JK		 1042static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1043{
1044 struct curseg_info *curseg = CURSEG_I(sbi, type);
1045 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1046
1047 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1048 return v_ops->get_victim(sbi,
1049 &(curseg)->next_segno, BG_GC, type, SSR);
1050
1051 /* For data segments, let's do SSR more intensively */
1052 for (; type >= CURSEG_HOT_DATA; type--)
1053 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1054 BG_GC, type, SSR))
1055 return 1;
1056 return 0;
1057}
1058
351df4b2
JK		 1059/*
1060 * flush out current segment and replace it with new segment
1061 * This function should be returned with success, otherwise BUG
1062 */
1063static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1064 int type, bool force)
1065{
1066 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1067
7b405275 1068 if (force)
351df4b2 1069 new_curseg(sbi, type, true);
7b405275 1070 else if (type == CURSEG_WARM_NODE)
351df4b2 1071 new_curseg(sbi, type, false);
60374688
JK		 1072 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1073 new_curseg(sbi, type, false);
351df4b2
JK		 1074 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1075 change_curseg(sbi, type, true);
1076 else
1077 new_curseg(sbi, type, false);
dcdfff65
JK		 1078
1079 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1080}
1081
38aa0889
JK		 1082static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1083{
1084 struct curseg_info *curseg = CURSEG_I(sbi, type);
1085 unsigned int old_segno;
1086
1087 old_segno = curseg->segno;
1088 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1089 locate_dirty_segment(sbi, old_segno);
1090}
1091
351df4b2
JK		 1092void allocate_new_segments(struct f2fs_sb_info *sbi)
1093{
351df4b2
JK		 1094 int i;
1095
38aa0889
JK		 1096 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1097 __allocate_new_segments(sbi, i);
351df4b2
JK		 1098}
1099
1100static const struct segment_allocation default_salloc_ops = {
1101 .allocate_segment = allocate_segment_by_default,
1102};
1103
4b2fecc8
JK		 1104int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1105{
f7ef9b83
JK		 1106 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1107 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1108 unsigned int start_segno, end_segno;
1109 struct cp_control cpc;
1110
836b5a63 1111 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1112 return -EINVAL;
1113
9bd27ae4 1114 cpc.trimmed = 0;
7cd8558b 1115 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1116 goto out;
1117
1118 /* start/end segment number in main_area */
7cd8558b
JK		 1119 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1120 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1121 GET_SEGNO(sbi, end);
4b2fecc8 1122 cpc.reason = CP_DISCARD;
836b5a63 1123 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1124
1125 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1126 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1127 cpc.trim_start = start_segno;
a66cdd98
JK		 1128
1129 if (sbi->discard_blks == 0)
1130 break;
1131 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1132 cpc.trim_end = end_segno;
1133 else
1134 cpc.trim_end = min_t(unsigned int,
1135 rounddown(start_segno +
bba681cb
JK		 1136 BATCHED_TRIM_SEGMENTS(sbi),
1137 sbi->segs_per_sec) - 1, end_segno);
1138
1139 mutex_lock(&sbi->gc_mutex);
1140 write_checkpoint(sbi, &cpc);
1141 mutex_unlock(&sbi->gc_mutex);
1142 }
4b2fecc8 1143out:
f7ef9b83 1144 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
4b2fecc8
JK		 1145 return 0;
1146}
1147
351df4b2
JK		 1148static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1149{
1150 struct curseg_info *curseg = CURSEG_I(sbi, type);
1151 if (curseg->next_blkoff < sbi->blocks_per_seg)
1152 return true;
1153 return false;
1154}
1155
1156static int __get_segment_type_2(struct page *page, enum page_type p_type)
1157{
1158 if (p_type == DATA)
1159 return CURSEG_HOT_DATA;
1160 else
1161 return CURSEG_HOT_NODE;
1162}
1163
1164static int __get_segment_type_4(struct page *page, enum page_type p_type)
1165{
1166 if (p_type == DATA) {
1167 struct inode *inode = page->mapping->host;
1168
1169 if (S_ISDIR(inode->i_mode))
1170 return CURSEG_HOT_DATA;
1171 else
1172 return CURSEG_COLD_DATA;
1173 } else {
a344b9fd
JK		 1174 if (IS_DNODE(page) && is_cold_node(page))
1175 return CURSEG_WARM_NODE;
351df4b2
JK		 1176 else
1177 return CURSEG_COLD_NODE;
1178 }
1179}
1180
1181static int __get_segment_type_6(struct page *page, enum page_type p_type)
1182{
1183 if (p_type == DATA) {
1184 struct inode *inode = page->mapping->host;
1185
1186 if (S_ISDIR(inode->i_mode))
1187 return CURSEG_HOT_DATA;
354a3399 1188 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1189 return CURSEG_COLD_DATA;
1190 else
1191 return CURSEG_WARM_DATA;
1192 } else {
1193 if (IS_DNODE(page))
1194 return is_cold_node(page) ? CURSEG_WARM_NODE :
1195 CURSEG_HOT_NODE;
1196 else
1197 return CURSEG_COLD_NODE;
1198 }
1199}
1200
1201static int __get_segment_type(struct page *page, enum page_type p_type)
1202{
4081363f 1203 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1204 case 2:
1205 return __get_segment_type_2(page, p_type);
1206 case 4:
1207 return __get_segment_type_4(page, p_type);
351df4b2 1208 }
12a67146 1209 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1210 f2fs_bug_on(F2FS_P_SB(page),
1211 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1212 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1213}
1214
bfad7c2d
JK		 1215void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1216 block_t old_blkaddr, block_t *new_blkaddr,
1217 struct f2fs_summary *sum, int type)
351df4b2
JK		 1218{
1219 struct sit_info *sit_i = SIT_I(sbi);
1220 struct curseg_info *curseg;
38aa0889
JK		 1221 bool direct_io = (type == CURSEG_DIRECT_IO);
1222
1223 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1224
351df4b2
JK		 1225 curseg = CURSEG_I(sbi, type);
1226
1227 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1228 mutex_lock(&sit_i->sentry_lock);
351df4b2 1229
38aa0889 1230 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1231 if (direct_io && curseg->next_blkoff &&
1232 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1233 __allocate_new_segments(sbi, type);
1234
351df4b2 1235 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1236
1237 /*
1238 * __add_sum_entry should be resided under the curseg_mutex
1239 * because, this function updates a summary entry in the
1240 * current summary block.
1241 */
e79efe3b 1242 __add_sum_entry(sbi, type, sum);
351df4b2 1243
351df4b2 1244 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1245
1246 stat_inc_block_count(sbi, curseg);
351df4b2 1247
5e443818
JK		 1248 if (!__has_curseg_space(sbi, type))
1249 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1250 /*
1251 * SIT information should be updated before segment allocation,
1252 * since SSR needs latest valid block information.
1253 */
1254 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1255
351df4b2
JK		 1256 mutex_unlock(&sit_i->sentry_lock);
1257
bfad7c2d 1258 if (page && IS_NODESEG(type))
351df4b2
JK		 1259 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1260
bfad7c2d
JK		 1261 mutex_unlock(&curseg->curseg_mutex);
1262}
1263
05ca3632 1264static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1265{
05ca3632 1266 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1267
05ca3632
JK		 1268 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1269 &fio->blk_addr, sum, type);
bfad7c2d 1270
351df4b2 1271 /* writeout dirty page into bdev */
05ca3632 1272 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1273}
1274
577e3495 1275void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1276{
458e6197 1277 struct f2fs_io_info fio = {
05ca3632 1278 .sbi = sbi,
458e6197 1279 .type = META,
cf04e8eb
JK		 1280 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1281 .blk_addr = page->index,
05ca3632 1282 .page = page,
4375a336 1283 .encrypted_page = NULL,
458e6197
JK		 1284 };
1285
351df4b2 1286 set_page_writeback(page);
05ca3632 1287 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1288}
1289
05ca3632 1290void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1291{
1292 struct f2fs_summary sum;
05ca3632 1293
351df4b2 1294 set_summary(&sum, nid, 0, 0);
05ca3632 1295 do_write_page(&sum, fio);
351df4b2
JK		 1296}
1297
05ca3632 1298void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1299{
05ca3632 1300 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1301 struct f2fs_summary sum;
1302 struct node_info ni;
1303
9850cf4a 1304 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1305 get_node_info(sbi, dn->nid, &ni);
1306 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1307 do_write_page(&sum, fio);
e1509cf2 1308 dn->data_blkaddr = fio->blk_addr;
351df4b2
JK		 1309}
1310
05ca3632 1311void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1312{
05ca3632
JK		 1313 stat_inc_inplace_blocks(fio->sbi);
1314 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1315}
1316
528e3459
CY		 1317static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1318 struct f2fs_summary *sum,
19f106bc
CY		 1319 block_t old_blkaddr, block_t new_blkaddr,
1320 bool recover_curseg)
351df4b2
JK		 1321{
1322 struct sit_info *sit_i = SIT_I(sbi);
1323 struct curseg_info *curseg;
1324 unsigned int segno, old_cursegno;
1325 struct seg_entry *se;
1326 int type;
19f106bc 1327 unsigned short old_blkoff;
351df4b2
JK		 1328
1329 segno = GET_SEGNO(sbi, new_blkaddr);
1330 se = get_seg_entry(sbi, segno);
1331 type = se->type;
1332
19f106bc
CY		 1333 if (!recover_curseg) {
1334 /* for recovery flow */
1335 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1336 if (old_blkaddr == NULL_ADDR)
1337 type = CURSEG_COLD_DATA;
1338 else
1339 type = CURSEG_WARM_DATA;
1340 }
1341 } else {
1342 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1343 type = CURSEG_WARM_DATA;
1344 }
19f106bc 1345
351df4b2
JK		 1346 curseg = CURSEG_I(sbi, type);
1347
1348 mutex_lock(&curseg->curseg_mutex);
1349 mutex_lock(&sit_i->sentry_lock);
1350
1351 old_cursegno = curseg->segno;
19f106bc 1352 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1353
1354 /* change the current segment */
1355 if (segno != curseg->segno) {
1356 curseg->next_segno = segno;
1357 change_curseg(sbi, type, true);
1358 }
1359
491c0854 1360 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1361 __add_sum_entry(sbi, type, sum);
351df4b2
JK		 1362
1363 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
351df4b2 1364 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1365
19f106bc
CY		 1366 if (recover_curseg) {
1367 if (old_cursegno != curseg->segno) {
1368 curseg->next_segno = old_cursegno;
1369 change_curseg(sbi, type, true);
1370 }
1371 curseg->next_blkoff = old_blkoff;
1372 }
1373
351df4b2
JK		 1374 mutex_unlock(&sit_i->sentry_lock);
1375 mutex_unlock(&curseg->curseg_mutex);
1376}
1377
528e3459
CY		 1378void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1379 block_t old_addr, block_t new_addr,
1380 unsigned char version, bool recover_curseg)
1381{
1382 struct f2fs_summary sum;
1383
1384 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1385
1386 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1387
1388 dn->data_blkaddr = new_addr;
1389 set_data_blkaddr(dn);
1390 f2fs_update_extent_cache(dn);
1391}
1392
df0f8dc0
CY		 1393static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1394 struct page *page, enum page_type type)
1395{
1396 enum page_type btype = PAGE_TYPE_OF_BIO(type);
1397 struct f2fs_bio_info *io = &sbi->write_io[btype];
df0f8dc0 1398 struct bio_vec *bvec;
4375a336 1399 struct page *target;
df0f8dc0
CY		 1400 int i;
1401
1402 down_read(&io->io_rwsem);
4375a336
JK		 1403 if (!io->bio) {
1404 up_read(&io->io_rwsem);
1405 return false;
1406 }
df0f8dc0 1407
ce23447f 1408 bio_for_each_segment_all(bvec, io->bio, i) {
4375a336
JK		 1409
1410 if (bvec->bv_page->mapping) {
1411 target = bvec->bv_page;
1412 } else {
1413 struct f2fs_crypto_ctx *ctx;
1414
1415 /* encrypted page */
1416 ctx = (struct f2fs_crypto_ctx *)page_private(
1417 bvec->bv_page);
ca40b030 1418 target = ctx->w.control_page;
4375a336
JK		 1419 }
1420
1421 if (page == target) {
df0f8dc0
CY		 1422 up_read(&io->io_rwsem);
1423 return true;
1424 }
1425 }
1426
df0f8dc0
CY		 1427 up_read(&io->io_rwsem);
1428 return false;
1429}
1430
93dfe2ac 1431void f2fs_wait_on_page_writeback(struct page *page,
5514f0aa 1432 enum page_type type)
93dfe2ac 1433{
93dfe2ac 1434 if (PageWriteback(page)) {
4081363f
JK		 1435 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1436
df0f8dc0
CY		 1437 if (is_merged_page(sbi, page, type))
1438 f2fs_submit_merged_bio(sbi, type, WRITE);
93dfe2ac
JK		 1439 wait_on_page_writeback(page);
1440 }
1441}
1442
351df4b2
JK		 1443static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1444{
1445 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1446 struct curseg_info *seg_i;
1447 unsigned char *kaddr;
1448 struct page *page;
1449 block_t start;
1450 int i, j, offset;
1451
1452 start = start_sum_block(sbi);
1453
1454 page = get_meta_page(sbi, start++);
1455 kaddr = (unsigned char *)page_address(page);
1456
1457 /* Step 1: restore nat cache */
1458 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1459 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1460
1461 /* Step 2: restore sit cache */
1462 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1463 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1464 SUM_JOURNAL_SIZE);
1465 offset = 2 * SUM_JOURNAL_SIZE;
1466
1467 /* Step 3: restore summary entries */
1468 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1469 unsigned short blk_off;
1470 unsigned int segno;
1471
1472 seg_i = CURSEG_I(sbi, i);
1473 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1474 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1475 seg_i->next_segno = segno;
1476 reset_curseg(sbi, i, 0);
1477 seg_i->alloc_type = ckpt->alloc_type[i];
1478 seg_i->next_blkoff = blk_off;
1479
1480 if (seg_i->alloc_type == SSR)
1481 blk_off = sbi->blocks_per_seg;
1482
1483 for (j = 0; j < blk_off; j++) {
1484 struct f2fs_summary *s;
1485 s = (struct f2fs_summary *)(kaddr + offset);
1486 seg_i->sum_blk->entries[j] = *s;
1487 offset += SUMMARY_SIZE;
1488 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1489 SUM_FOOTER_SIZE)
1490 continue;
1491
1492 f2fs_put_page(page, 1);
1493 page = NULL;
1494
1495 page = get_meta_page(sbi, start++);
1496 kaddr = (unsigned char *)page_address(page);
1497 offset = 0;
1498 }
1499 }
1500 f2fs_put_page(page, 1);
1501 return 0;
1502}
1503
1504static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1505{
1506 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1507 struct f2fs_summary_block *sum;
1508 struct curseg_info *curseg;
1509 struct page *new;
1510 unsigned short blk_off;
1511 unsigned int segno = 0;
1512 block_t blk_addr = 0;
1513
1514 /* get segment number and block addr */
1515 if (IS_DATASEG(type)) {
1516 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1517 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1518 CURSEG_HOT_DATA]);
119ee914 1519 if (__exist_node_summaries(sbi))
351df4b2
JK		 1520 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1521 else
1522 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1523 } else {
1524 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1525 CURSEG_HOT_NODE]);
1526 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1527 CURSEG_HOT_NODE]);
119ee914 1528 if (__exist_node_summaries(sbi))
351df4b2
JK		 1529 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1530 type - CURSEG_HOT_NODE);
1531 else
1532 blk_addr = GET_SUM_BLOCK(sbi, segno);
1533 }
1534
1535 new = get_meta_page(sbi, blk_addr);
1536 sum = (struct f2fs_summary_block *)page_address(new);
1537
1538 if (IS_NODESEG(type)) {
119ee914 1539 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1540 struct f2fs_summary *ns = &sum->entries[0];
1541 int i;
1542 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1543 ns->version = 0;
1544 ns->ofs_in_node = 0;
1545 }
1546 } else {
d653788a
GZ		 1547 int err;
1548
1549 err = restore_node_summary(sbi, segno, sum);
1550 if (err) {
351df4b2 1551 f2fs_put_page(new, 1);
d653788a 1552 return err;
351df4b2
JK		 1553 }
1554 }
1555 }
1556
1557 /* set uncompleted segment to curseg */
1558 curseg = CURSEG_I(sbi, type);
1559 mutex_lock(&curseg->curseg_mutex);
1560 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1561 curseg->next_segno = segno;
1562 reset_curseg(sbi, type, 0);
1563 curseg->alloc_type = ckpt->alloc_type[type];
1564 curseg->next_blkoff = blk_off;
1565 mutex_unlock(&curseg->curseg_mutex);
1566 f2fs_put_page(new, 1);
1567 return 0;
1568}
1569
1570static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1571{
1572 int type = CURSEG_HOT_DATA;
e4fc5fbf 1573 int err;
351df4b2 1574
25ca923b 1575 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1576 int npages = npages_for_summary_flush(sbi, true);
1577
1578 if (npages >= 2)
1579 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1580 META_CP);
1581
351df4b2
JK		 1582 /* restore for compacted data summary */
1583 if (read_compacted_summaries(sbi))
1584 return -EINVAL;
1585 type = CURSEG_HOT_NODE;
1586 }
1587
119ee914 1588 if (__exist_node_summaries(sbi))
3fa06d7b
CY		 1589 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1590 NR_CURSEG_TYPE - type, META_CP);
1591
e4fc5fbf
CY		 1592 for (; type <= CURSEG_COLD_NODE; type++) {
1593 err = read_normal_summaries(sbi, type);
1594 if (err)
1595 return err;
1596 }
1597
351df4b2
JK		 1598 return 0;
1599}
1600
1601static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1602{
1603 struct page *page;
1604 unsigned char *kaddr;
1605 struct f2fs_summary *summary;
1606 struct curseg_info *seg_i;
1607 int written_size = 0;
1608 int i, j;
1609
1610 page = grab_meta_page(sbi, blkaddr++);
1611 kaddr = (unsigned char *)page_address(page);
1612
1613 /* Step 1: write nat cache */
1614 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1615 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1616 written_size += SUM_JOURNAL_SIZE;
1617
1618 /* Step 2: write sit cache */
1619 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1620 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1621 SUM_JOURNAL_SIZE);
1622 written_size += SUM_JOURNAL_SIZE;
1623
351df4b2
JK		 1624 /* Step 3: write summary entries */
1625 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1626 unsigned short blkoff;
1627 seg_i = CURSEG_I(sbi, i);
1628 if (sbi->ckpt->alloc_type[i] == SSR)
1629 blkoff = sbi->blocks_per_seg;
1630 else
1631 blkoff = curseg_blkoff(sbi, i);
1632
1633 for (j = 0; j < blkoff; j++) {
1634 if (!page) {
1635 page = grab_meta_page(sbi, blkaddr++);
1636 kaddr = (unsigned char *)page_address(page);
1637 written_size = 0;
1638 }
1639 summary = (struct f2fs_summary *)(kaddr + written_size);
1640 *summary = seg_i->sum_blk->entries[j];
1641 written_size += SUMMARY_SIZE;
351df4b2
JK		 1642
1643 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1644 SUM_FOOTER_SIZE)
1645 continue;
1646
e8d61a74 1647 set_page_dirty(page);
351df4b2
JK		 1648 f2fs_put_page(page, 1);
1649 page = NULL;
1650 }
1651 }
e8d61a74
CY		 1652 if (page) {
1653 set_page_dirty(page);
351df4b2 1654 f2fs_put_page(page, 1);
e8d61a74 1655 }
351df4b2
JK		 1656}
1657
1658static void write_normal_summaries(struct f2fs_sb_info *sbi,
1659 block_t blkaddr, int type)
1660{
1661 int i, end;
1662 if (IS_DATASEG(type))
1663 end = type + NR_CURSEG_DATA_TYPE;
1664 else
1665 end = type + NR_CURSEG_NODE_TYPE;
1666
1667 for (i = type; i < end; i++) {
1668 struct curseg_info *sum = CURSEG_I(sbi, i);
1669 mutex_lock(&sum->curseg_mutex);
1670 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1671 mutex_unlock(&sum->curseg_mutex);
1672 }
1673}
1674
1675void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1676{
25ca923b 1677 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1678 write_compacted_summaries(sbi, start_blk);
1679 else
1680 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1681}
1682
1683void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1684{
119ee914 1685 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1686}
1687
1688int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1689 unsigned int val, int alloc)
1690{
1691 int i;
1692
1693 if (type == NAT_JOURNAL) {
1694 for (i = 0; i < nats_in_cursum(sum); i++) {
1695 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1696 return i;
1697 }
1698 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1699 return update_nats_in_cursum(sum, 1);
1700 } else if (type == SIT_JOURNAL) {
1701 for (i = 0; i < sits_in_cursum(sum); i++)
1702 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1703 return i;
1704 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1705 return update_sits_in_cursum(sum, 1);
1706 }
1707 return -1;
1708}
1709
1710static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1711 unsigned int segno)
1712{
2cc22186 1713 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1714}
1715
1716static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1717 unsigned int start)
1718{
1719 struct sit_info *sit_i = SIT_I(sbi);
1720 struct page *src_page, *dst_page;
1721 pgoff_t src_off, dst_off;
1722 void *src_addr, *dst_addr;
1723
1724 src_off = current_sit_addr(sbi, start);
1725 dst_off = next_sit_addr(sbi, src_off);
1726
1727 /* get current sit block page without lock */
1728 src_page = get_meta_page(sbi, src_off);
1729 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1730 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1731
1732 src_addr = page_address(src_page);
1733 dst_addr = page_address(dst_page);
1734 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1735
1736 set_page_dirty(dst_page);
1737 f2fs_put_page(src_page, 1);
1738
1739 set_to_next_sit(sit_i, start);
1740
1741 return dst_page;
1742}
1743
184a5cd2
CY		 1744static struct sit_entry_set *grab_sit_entry_set(void)
1745{
1746 struct sit_entry_set *ses =
1747 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1748
1749 ses->entry_cnt = 0;
1750 INIT_LIST_HEAD(&ses->set_list);
1751 return ses;
1752}
1753
1754static void release_sit_entry_set(struct sit_entry_set *ses)
1755{
1756 list_del(&ses->set_list);
1757 kmem_cache_free(sit_entry_set_slab, ses);
1758}
1759
1760static void adjust_sit_entry_set(struct sit_entry_set *ses,
1761 struct list_head *head)
1762{
1763 struct sit_entry_set *next = ses;
1764
1765 if (list_is_last(&ses->set_list, head))
1766 return;
1767
1768 list_for_each_entry_continue(next, head, set_list)
1769 if (ses->entry_cnt <= next->entry_cnt)
1770 break;
1771
1772 list_move_tail(&ses->set_list, &next->set_list);
1773}
1774
1775static void add_sit_entry(unsigned int segno, struct list_head *head)
1776{
1777 struct sit_entry_set *ses;
1778 unsigned int start_segno = START_SEGNO(segno);
1779
1780 list_for_each_entry(ses, head, set_list) {
1781 if (ses->start_segno == start_segno) {
1782 ses->entry_cnt++;
1783 adjust_sit_entry_set(ses, head);
1784 return;
1785 }
1786 }
1787
1788 ses = grab_sit_entry_set();
1789
1790 ses->start_segno = start_segno;
1791 ses->entry_cnt++;
1792 list_add(&ses->set_list, head);
1793}
1794
1795static void add_sits_in_set(struct f2fs_sb_info *sbi)
1796{
1797 struct f2fs_sm_info *sm_info = SM_I(sbi);
1798 struct list_head *set_list = &sm_info->sit_entry_set;
1799 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1800 unsigned int segno;
1801
7cd8558b 1802 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1803 add_sit_entry(segno, set_list);
1804}
1805
1806static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1807{
1808 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1809 struct f2fs_summary_block *sum = curseg->sum_blk;
1810 int i;
1811
184a5cd2
CY		 1812 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1813 unsigned int segno;
1814 bool dirtied;
1815
1816 segno = le32_to_cpu(segno_in_journal(sum, i));
1817 dirtied = __mark_sit_entry_dirty(sbi, segno);
1818
1819 if (!dirtied)
1820 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1821 }
184a5cd2 1822 update_sits_in_cursum(sum, -sits_in_cursum(sum));
351df4b2
JK		 1823}
1824
0a8165d7 1825/*
351df4b2
JK		 1826 * CP calls this function, which flushes SIT entries including sit_journal,
1827 * and moves prefree segs to free segs.
1828 */
4b2fecc8 1829void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1830{
1831 struct sit_info *sit_i = SIT_I(sbi);
1832 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1833 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1834 struct f2fs_summary_block *sum = curseg->sum_blk;
184a5cd2
CY		 1835 struct sit_entry_set *ses, *tmp;
1836 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1837 bool to_journal = true;
4b2fecc8 1838 struct seg_entry *se;
351df4b2
JK		 1839
1840 mutex_lock(&curseg->curseg_mutex);
1841 mutex_lock(&sit_i->sentry_lock);
1842
2b11a74b
WL		 1843 if (!sit_i->dirty_sentries)
1844 goto out;
1845
351df4b2 1846 /*
184a5cd2
CY		 1847 * add and account sit entries of dirty bitmap in sit entry
1848 * set temporarily
351df4b2 1849 */
184a5cd2 1850 add_sits_in_set(sbi);
351df4b2 1851
184a5cd2
CY		 1852 /*
1853 * if there are no enough space in journal to store dirty sit
1854 * entries, remove all entries from journal and add and account
1855 * them in sit entry set.
1856 */
1857 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1858 remove_sits_in_journal(sbi);
b2955550 1859
184a5cd2
CY		 1860 /*
1861 * there are two steps to flush sit entries:
1862 * #1, flush sit entries to journal in current cold data summary block.
1863 * #2, flush sit entries to sit page.
1864 */
1865 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1866 struct page *page = NULL;
184a5cd2
CY		 1867 struct f2fs_sit_block *raw_sit = NULL;
1868 unsigned int start_segno = ses->start_segno;
1869 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1870 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1871 unsigned int segno = start_segno;
1872
1873 if (to_journal &&
1874 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1875 to_journal = false;
1876
1877 if (!to_journal) {
1878 page = get_next_sit_page(sbi, start_segno);
1879 raw_sit = page_address(page);
351df4b2 1880 }
351df4b2 1881
184a5cd2
CY		 1882 /* flush dirty sit entries in region of current sit set */
1883 for_each_set_bit_from(segno, bitmap, end) {
1884 int offset, sit_offset;
4b2fecc8
JK		 1885
1886 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 1887
1888 /* add discard candidates */
d7bc2484 1889 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 1890 cpc->trim_start = segno;
1891 add_discard_addrs(sbi, cpc);
1892 }
184a5cd2
CY		 1893
1894 if (to_journal) {
1895 offset = lookup_journal_in_cursum(sum,
1896 SIT_JOURNAL, segno, 1);
1897 f2fs_bug_on(sbi, offset < 0);
1898 segno_in_journal(sum, offset) =
1899 cpu_to_le32(segno);
1900 seg_info_to_raw_sit(se,
1901 &sit_in_journal(sum, offset));
1902 } else {
1903 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1904 seg_info_to_raw_sit(se,
1905 &raw_sit->entries[sit_offset]);
1906 }
351df4b2 1907
184a5cd2
CY		 1908 __clear_bit(segno, bitmap);
1909 sit_i->dirty_sentries--;
1910 ses->entry_cnt--;
351df4b2
JK		 1911 }
1912
184a5cd2
CY		 1913 if (!to_journal)
1914 f2fs_put_page(page, 1);
1915
1916 f2fs_bug_on(sbi, ses->entry_cnt);
1917 release_sit_entry_set(ses);
351df4b2 1918 }
184a5cd2
CY		 1919
1920 f2fs_bug_on(sbi, !list_empty(head));
1921 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 1922out:
4b2fecc8
JK		 1923 if (cpc->reason == CP_DISCARD) {
1924 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1925 add_discard_addrs(sbi, cpc);
1926 }
351df4b2
JK		 1927 mutex_unlock(&sit_i->sentry_lock);
1928 mutex_unlock(&curseg->curseg_mutex);
1929
351df4b2
JK		 1930 set_prefree_as_free_segments(sbi);
1931}
1932
1933static int build_sit_info(struct f2fs_sb_info *sbi)
1934{
1935 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1936 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1937 struct sit_info *sit_i;
1938 unsigned int sit_segs, start;
1939 char *src_bitmap, *dst_bitmap;
1940 unsigned int bitmap_size;
1941
1942 /* allocate memory for SIT information */
1943 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1944 if (!sit_i)
1945 return -ENOMEM;
1946
1947 SM_I(sbi)->sit_info = sit_i;
1948
7cd8558b 1949 sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
351df4b2
JK		 1950 if (!sit_i->sentries)
1951 return -ENOMEM;
1952
7cd8558b 1953 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 1954 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1955 if (!sit_i->dirty_sentries_bitmap)
1956 return -ENOMEM;
1957
7cd8558b 1958 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 1959 sit_i->sentries[start].cur_valid_map
1960 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1961 sit_i->sentries[start].ckpt_valid_map
1962 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 1963 sit_i->sentries[start].discard_map
1964 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1965 if (!sit_i->sentries[start].cur_valid_map ||
1966 !sit_i->sentries[start].ckpt_valid_map ||
1967 !sit_i->sentries[start].discard_map)
351df4b2
JK		 1968 return -ENOMEM;
1969 }
1970
60a3b782
JK		 1971 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1972 if (!sit_i->tmp_map)
1973 return -ENOMEM;
1974
351df4b2 1975 if (sbi->segs_per_sec > 1) {
7cd8558b 1976 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
351df4b2
JK		 1977 sizeof(struct sec_entry));
1978 if (!sit_i->sec_entries)
1979 return -ENOMEM;
1980 }
1981
1982 /* get information related with SIT */
1983 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1984
1985 /* setup SIT bitmap from ckeckpoint pack */
1986 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1987 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1988
79b5793b 1989 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 1990 if (!dst_bitmap)
1991 return -ENOMEM;
351df4b2
JK		 1992
1993 /* init SIT information */
1994 sit_i->s_ops = &default_salloc_ops;
1995
1996 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1997 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1998 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1999 sit_i->sit_bitmap = dst_bitmap;
2000 sit_i->bitmap_size = bitmap_size;
2001 sit_i->dirty_sentries = 0;
2002 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2003 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2004 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2005 mutex_init(&sit_i->sentry_lock);
2006 return 0;
2007}
2008
2009static int build_free_segmap(struct f2fs_sb_info *sbi)
2010{
351df4b2
JK		 2011 struct free_segmap_info *free_i;
2012 unsigned int bitmap_size, sec_bitmap_size;
2013
2014 /* allocate memory for free segmap information */
2015 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2016 if (!free_i)
2017 return -ENOMEM;
2018
2019 SM_I(sbi)->free_info = free_i;
2020
7cd8558b 2021 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2022 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
2023 if (!free_i->free_segmap)
2024 return -ENOMEM;
2025
7cd8558b 2026 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2
JK		 2027 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
2028 if (!free_i->free_secmap)
2029 return -ENOMEM;
2030
2031 /* set all segments as dirty temporarily */
2032 memset(free_i->free_segmap, 0xff, bitmap_size);
2033 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2034
2035 /* init free segmap information */
7cd8558b 2036 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2037 free_i->free_segments = 0;
2038 free_i->free_sections = 0;
1a118ccf 2039 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2040 return 0;
2041}
2042
2043static int build_curseg(struct f2fs_sb_info *sbi)
2044{
1042d60f 2045 struct curseg_info *array;
351df4b2
JK		 2046 int i;
2047
b434babf 2048 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2049 if (!array)
2050 return -ENOMEM;
2051
2052 SM_I(sbi)->curseg_array = array;
2053
2054 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2055 mutex_init(&array[i].curseg_mutex);
2056 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2057 if (!array[i].sum_blk)
2058 return -ENOMEM;
2059 array[i].segno = NULL_SEGNO;
2060 array[i].next_blkoff = 0;
2061 }
2062 return restore_curseg_summaries(sbi);
2063}
2064
2065static void build_sit_entries(struct f2fs_sb_info *sbi)
2066{
2067 struct sit_info *sit_i = SIT_I(sbi);
2068 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2069 struct f2fs_summary_block *sum = curseg->sum_blk;
74de593a
CY		 2070 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2071 unsigned int i, start, end;
2072 unsigned int readed, start_blk = 0;
90a893c7 2073 int nrpages = MAX_BIO_BLOCKS(sbi);
351df4b2 2074
74de593a 2075 do {
662befda 2076 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
74de593a
CY		 2077
2078 start = start_blk * sit_i->sents_per_block;
2079 end = (start_blk + readed) * sit_i->sents_per_block;
2080
7cd8558b 2081 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2082 struct seg_entry *se = &sit_i->sentries[start];
2083 struct f2fs_sit_block *sit_blk;
2084 struct f2fs_sit_entry sit;
2085 struct page *page;
2086
2087 mutex_lock(&curseg->curseg_mutex);
2088 for (i = 0; i < sits_in_cursum(sum); i++) {
6c311ec6
CF		 2089 if (le32_to_cpu(segno_in_journal(sum, i))
2090 == start) {
74de593a
CY		 2091 sit = sit_in_journal(sum, i);
2092 mutex_unlock(&curseg->curseg_mutex);
2093 goto got_it;
2094 }
351df4b2 2095 }
74de593a
CY		 2096 mutex_unlock(&curseg->curseg_mutex);
2097
2098 page = get_current_sit_page(sbi, start);
2099 sit_blk = (struct f2fs_sit_block *)page_address(page);
2100 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2101 f2fs_put_page(page, 1);
351df4b2 2102got_it:
74de593a
CY		 2103 check_block_count(sbi, start, &sit);
2104 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2105
2106 /* build discard map only one time */
2107 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2108 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2109
74de593a
CY		 2110 if (sbi->segs_per_sec > 1) {
2111 struct sec_entry *e = get_sec_entry(sbi, start);
2112 e->valid_blocks += se->valid_blocks;
2113 }
351df4b2 2114 }
74de593a
CY		 2115 start_blk += readed;
2116 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2117}
2118
2119static void init_free_segmap(struct f2fs_sb_info *sbi)
2120{
2121 unsigned int start;
2122 int type;
2123
7cd8558b 2124 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2125 struct seg_entry *sentry = get_seg_entry(sbi, start);
2126 if (!sentry->valid_blocks)
2127 __set_free(sbi, start);
2128 }
2129
2130 /* set use the current segments */
2131 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2132 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2133 __set_test_and_inuse(sbi, curseg_t->segno);
2134 }
2135}
2136
2137static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2138{
2139 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2140 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2141 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2142 unsigned short valid_blocks;
2143
8736fbf0 2144 while (1) {
351df4b2 2145 /* find dirty segment based on free segmap */
7cd8558b
JK		 2146 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2147 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2148 break;
2149 offset = segno + 1;
2150 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2151 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2152 continue;
ec325b52
JK		 2153 if (valid_blocks > sbi->blocks_per_seg) {
2154 f2fs_bug_on(sbi, 1);
2155 continue;
2156 }
351df4b2
JK		 2157 mutex_lock(&dirty_i->seglist_lock);
2158 __locate_dirty_segment(sbi, segno, DIRTY);
2159 mutex_unlock(&dirty_i->seglist_lock);
2160 }
2161}
2162
5ec4e49f 2163static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2164{
2165 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2166 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2167
5ec4e49f
JK		 2168 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2169 if (!dirty_i->victim_secmap)
351df4b2
JK		 2170 return -ENOMEM;
2171 return 0;
2172}
2173
2174static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2175{
2176 struct dirty_seglist_info *dirty_i;
2177 unsigned int bitmap_size, i;
2178
2179 /* allocate memory for dirty segments list information */
2180 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2181 if (!dirty_i)
2182 return -ENOMEM;
2183
2184 SM_I(sbi)->dirty_info = dirty_i;
2185 mutex_init(&dirty_i->seglist_lock);
2186
7cd8558b 2187 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2188
2189 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2190 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2191 if (!dirty_i->dirty_segmap[i])
2192 return -ENOMEM;
2193 }
2194
2195 init_dirty_segmap(sbi);
5ec4e49f 2196 return init_victim_secmap(sbi);
351df4b2
JK		 2197}
2198
0a8165d7 2199/*
351df4b2
JK		 2200 * Update min, max modified time for cost-benefit GC algorithm
2201 */
2202static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2203{
2204 struct sit_info *sit_i = SIT_I(sbi);
2205 unsigned int segno;
2206
2207 mutex_lock(&sit_i->sentry_lock);
2208
2209 sit_i->min_mtime = LLONG_MAX;
2210
7cd8558b 2211 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2212 unsigned int i;
2213 unsigned long long mtime = 0;
2214
2215 for (i = 0; i < sbi->segs_per_sec; i++)
2216 mtime += get_seg_entry(sbi, segno + i)->mtime;
2217
2218 mtime = div_u64(mtime, sbi->segs_per_sec);
2219
2220 if (sit_i->min_mtime > mtime)
2221 sit_i->min_mtime = mtime;
2222 }
2223 sit_i->max_mtime = get_mtime(sbi);
2224 mutex_unlock(&sit_i->sentry_lock);
2225}
2226
2227int build_segment_manager(struct f2fs_sb_info *sbi)
2228{
2229 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2230 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2231 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2232 int err;
2233
2234 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2235 if (!sm_info)
2236 return -ENOMEM;
2237
2238 /* init sm info */
2239 sbi->sm_info = sm_info;
351df4b2
JK		 2240 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2241 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2242 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2243 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2244 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2245 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2246 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2247 sm_info->rec_prefree_segments = sm_info->main_segments *
2248 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2249 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2250 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2251 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2252
7fd9e544
JK		 2253 INIT_LIST_HEAD(&sm_info->discard_list);
2254 sm_info->nr_discards = 0;
2255 sm_info->max_discards = 0;
2256
bba681cb
JK		 2257 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2258
184a5cd2
CY		 2259 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2260
b270ad6f 2261 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2262 err = create_flush_cmd_control(sbi);
2263 if (err)
a688b9d9 2264 return err;
6b4afdd7
JK		 2265 }
2266
351df4b2
JK		 2267 err = build_sit_info(sbi);
2268 if (err)
2269 return err;
2270 err = build_free_segmap(sbi);
2271 if (err)
2272 return err;
2273 err = build_curseg(sbi);
2274 if (err)
2275 return err;
2276
2277 /* reinit free segmap based on SIT */
2278 build_sit_entries(sbi);
2279
2280 init_free_segmap(sbi);
2281 err = build_dirty_segmap(sbi);
2282 if (err)
2283 return err;
2284
2285 init_min_max_mtime(sbi);
2286 return 0;
2287}
2288
2289static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2290 enum dirty_type dirty_type)
2291{
2292 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2293
2294 mutex_lock(&dirty_i->seglist_lock);
2295 kfree(dirty_i->dirty_segmap[dirty_type]);
2296 dirty_i->nr_dirty[dirty_type] = 0;
2297 mutex_unlock(&dirty_i->seglist_lock);
2298}
2299
5ec4e49f 2300static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2301{
2302 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5ec4e49f 2303 kfree(dirty_i->victim_secmap);
351df4b2
JK		 2304}
2305
2306static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2307{
2308 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2309 int i;
2310
2311 if (!dirty_i)
2312 return;
2313
2314 /* discard pre-free/dirty segments list */
2315 for (i = 0; i < NR_DIRTY_TYPE; i++)
2316 discard_dirty_segmap(sbi, i);
2317
5ec4e49f 2318 destroy_victim_secmap(sbi);
351df4b2
JK		 2319 SM_I(sbi)->dirty_info = NULL;
2320 kfree(dirty_i);
2321}
2322
2323static void destroy_curseg(struct f2fs_sb_info *sbi)
2324{
2325 struct curseg_info *array = SM_I(sbi)->curseg_array;
2326 int i;
2327
2328 if (!array)
2329 return;
2330 SM_I(sbi)->curseg_array = NULL;
2331 for (i = 0; i < NR_CURSEG_TYPE; i++)
2332 kfree(array[i].sum_blk);
2333 kfree(array);
2334}
2335
2336static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2337{
2338 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2339 if (!free_i)
2340 return;
2341 SM_I(sbi)->free_info = NULL;
2342 kfree(free_i->free_segmap);
2343 kfree(free_i->free_secmap);
2344 kfree(free_i);
2345}
2346
2347static void destroy_sit_info(struct f2fs_sb_info *sbi)
2348{
2349 struct sit_info *sit_i = SIT_I(sbi);
2350 unsigned int start;
2351
2352 if (!sit_i)
2353 return;
2354
2355 if (sit_i->sentries) {
7cd8558b 2356 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2357 kfree(sit_i->sentries[start].cur_valid_map);
2358 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2359 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2360 }
2361 }
60a3b782
JK		 2362 kfree(sit_i->tmp_map);
2363
351df4b2
JK		 2364 vfree(sit_i->sentries);
2365 vfree(sit_i->sec_entries);
2366 kfree(sit_i->dirty_sentries_bitmap);
2367
2368 SM_I(sbi)->sit_info = NULL;
2369 kfree(sit_i->sit_bitmap);
2370 kfree(sit_i);
2371}
2372
2373void destroy_segment_manager(struct f2fs_sb_info *sbi)
2374{
2375 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2376
3b03f724
CY		 2377 if (!sm_info)
2378 return;
2163d198 2379 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2380 destroy_dirty_segmap(sbi);
2381 destroy_curseg(sbi);
2382 destroy_free_segmap(sbi);
2383 destroy_sit_info(sbi);
2384 sbi->sm_info = NULL;
2385 kfree(sm_info);
2386}
7fd9e544
JK		 2387
2388int __init create_segment_manager_caches(void)
2389{
2390 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2391 sizeof(struct discard_entry));
7fd9e544 2392 if (!discard_entry_slab)
184a5cd2
CY		 2393 goto fail;
2394
2395 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2396 sizeof(struct sit_entry_set));
184a5cd2
CY		 2397 if (!sit_entry_set_slab)
2398 goto destory_discard_entry;
88b88a66
JK		 2399
2400 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2401 sizeof(struct inmem_pages));
2402 if (!inmem_entry_slab)
2403 goto destroy_sit_entry_set;
7fd9e544 2404 return 0;
184a5cd2 2405
88b88a66
JK		 2406destroy_sit_entry_set:
2407 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2408destory_discard_entry:
2409 kmem_cache_destroy(discard_entry_slab);
2410fail:
2411 return -ENOMEM;
7fd9e544
JK		 2412}
2413
2414void destroy_segment_manager_caches(void)
2415{
184a5cd2 2416 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2417 kmem_cache_destroy(discard_entry_slab);
88b88a66 2418 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2419}
Linux kernel - Deliverables
This page took 0.233516 seconds and 5 git commands to generate.