projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: add a tracepoint for f2fs_read_data_pages
[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
6e2c64ad
JK		 771bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
772{
773 struct sit_info *sit_i = SIT_I(sbi);
774 unsigned int segno, offset;
775 struct seg_entry *se;
776 bool is_cp = false;
777
778 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
779 return true;
780
781 mutex_lock(&sit_i->sentry_lock);
782
783 segno = GET_SEGNO(sbi, blkaddr);
784 se = get_seg_entry(sbi, segno);
785 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
786
787 if (f2fs_test_bit(offset, se->ckpt_valid_map))
788 is_cp = true;
789
790 mutex_unlock(&sit_i->sentry_lock);
791
792 return is_cp;
793}
794
0a8165d7 795/*
351df4b2
JK		 796 * This function should be resided under the curseg_mutex lock
797 */
798static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 799 struct f2fs_summary *sum)
351df4b2
JK		 800{
801 struct curseg_info *curseg = CURSEG_I(sbi, type);
802 void *addr = curseg->sum_blk;
e79efe3b 803 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 804 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 805}
806
0a8165d7 807/*
351df4b2
JK		 808 * Calculate the number of current summary pages for writing
809 */
3fa06d7b 810int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 811{
351df4b2 812 int valid_sum_count = 0;
9a47938b 813 int i, sum_in_page;
351df4b2
JK		 814
815 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
816 if (sbi->ckpt->alloc_type[i] == SSR)
817 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 818 else {
819 if (for_ra)
820 valid_sum_count += le16_to_cpu(
821 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
822 else
823 valid_sum_count += curseg_blkoff(sbi, i);
824 }
351df4b2
JK		 825 }
826
9a47938b
FL		 827 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
828 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
829 if (valid_sum_count <= sum_in_page)
351df4b2 830 return 1;
9a47938b
FL		 831 else if ((valid_sum_count - sum_in_page) <=
832 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 833 return 2;
834 return 3;
835}
836
0a8165d7 837/*
351df4b2
JK		 838 * Caller should put this summary page
839 */
840struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
841{
842 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
843}
844
381722d2 845void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 846{
847 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 848 void *dst = page_address(page);
849
850 if (src)
851 memcpy(dst, src, PAGE_CACHE_SIZE);
852 else
853 memset(dst, 0, PAGE_CACHE_SIZE);
351df4b2
JK		 854 set_page_dirty(page);
855 f2fs_put_page(page, 1);
856}
857
381722d2
CY		 858static void write_sum_page(struct f2fs_sb_info *sbi,
859 struct f2fs_summary_block *sum_blk, block_t blk_addr)
860{
861 update_meta_page(sbi, (void *)sum_blk, blk_addr);
862}
863
60374688
JK		 864static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
865{
866 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 867 unsigned int segno = curseg->segno + 1;
60374688
JK		 868 struct free_segmap_info *free_i = FREE_I(sbi);
869
7cd8558b 870 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 871 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 872 return 0;
873}
874
0a8165d7 875/*
351df4b2
JK		 876 * Find a new segment from the free segments bitmap to right order
877 * This function should be returned with success, otherwise BUG
878 */
879static void get_new_segment(struct f2fs_sb_info *sbi,
880 unsigned int *newseg, bool new_sec, int dir)
881{
882 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 883 unsigned int segno, secno, zoneno;
7cd8558b 884 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 885 unsigned int hint = *newseg / sbi->segs_per_sec;
886 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
887 unsigned int left_start = hint;
888 bool init = true;
889 int go_left = 0;
890 int i;
891
1a118ccf 892 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 893
894 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
895 segno = find_next_zero_bit(free_i->free_segmap,
7cd8558b 896 MAIN_SEGS(sbi), *newseg + 1);
33afa7fd
JK		 897 if (segno - *newseg < sbi->segs_per_sec -
898 (*newseg % sbi->segs_per_sec))
351df4b2
JK		 899 goto got_it;
900 }
901find_other_zone:
7cd8558b
JK		 902 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
903 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 904 if (dir == ALLOC_RIGHT) {
905 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 906 MAIN_SECS(sbi), 0);
907 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 908 } else {
909 go_left = 1;
910 left_start = hint - 1;
911 }
912 }
913 if (go_left == 0)
914 goto skip_left;
915
916 while (test_bit(left_start, free_i->free_secmap)) {
917 if (left_start > 0) {
918 left_start--;
919 continue;
920 }
921 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 922 MAIN_SECS(sbi), 0);
923 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 924 break;
925 }
926 secno = left_start;
927skip_left:
928 hint = secno;
929 segno = secno * sbi->segs_per_sec;
930 zoneno = secno / sbi->secs_per_zone;
931
932 /* give up on finding another zone */
933 if (!init)
934 goto got_it;
935 if (sbi->secs_per_zone == 1)
936 goto got_it;
937 if (zoneno == old_zoneno)
938 goto got_it;
939 if (dir == ALLOC_LEFT) {
940 if (!go_left && zoneno + 1 >= total_zones)
941 goto got_it;
942 if (go_left && zoneno == 0)
943 goto got_it;
944 }
945 for (i = 0; i < NR_CURSEG_TYPE; i++)
946 if (CURSEG_I(sbi, i)->zone == zoneno)
947 break;
948
949 if (i < NR_CURSEG_TYPE) {
950 /* zone is in user, try another */
951 if (go_left)
952 hint = zoneno * sbi->secs_per_zone - 1;
953 else if (zoneno + 1 >= total_zones)
954 hint = 0;
955 else
956 hint = (zoneno + 1) * sbi->secs_per_zone;
957 init = false;
958 goto find_other_zone;
959 }
960got_it:
961 /* set it as dirty segment in free segmap */
9850cf4a 962 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 963 __set_inuse(sbi, segno);
964 *newseg = segno;
1a118ccf 965 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 966}
967
968static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
969{
970 struct curseg_info *curseg = CURSEG_I(sbi, type);
971 struct summary_footer *sum_footer;
972
973 curseg->segno = curseg->next_segno;
974 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
975 curseg->next_blkoff = 0;
976 curseg->next_segno = NULL_SEGNO;
977
978 sum_footer = &(curseg->sum_blk->footer);
979 memset(sum_footer, 0, sizeof(struct summary_footer));
980 if (IS_DATASEG(type))
981 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
982 if (IS_NODESEG(type))
983 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
984 __set_sit_entry_type(sbi, type, curseg->segno, modified);
985}
986
0a8165d7 987/*
351df4b2
JK		 988 * Allocate a current working segment.
989 * This function always allocates a free segment in LFS manner.
990 */
991static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
992{
993 struct curseg_info *curseg = CURSEG_I(sbi, type);
994 unsigned int segno = curseg->segno;
995 int dir = ALLOC_LEFT;
996
997 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 998 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 999 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1000 dir = ALLOC_RIGHT;
1001
1002 if (test_opt(sbi, NOHEAP))
1003 dir = ALLOC_RIGHT;
1004
1005 get_new_segment(sbi, &segno, new_sec, dir);
1006 curseg->next_segno = segno;
1007 reset_curseg(sbi, type, 1);
1008 curseg->alloc_type = LFS;
1009}
1010
1011static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1012 struct curseg_info *seg, block_t start)
1013{
1014 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1015 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1016 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1017 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1018 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1019 int i, pos;
1020
1021 for (i = 0; i < entries; i++)
1022 target_map[i] = ckpt_map[i] | cur_map[i];
1023
1024 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1025
1026 seg->next_blkoff = pos;
351df4b2
JK		 1027}
1028
0a8165d7 1029/*
351df4b2
JK		 1030 * If a segment is written by LFS manner, next block offset is just obtained
1031 * by increasing the current block offset. However, if a segment is written by
1032 * SSR manner, next block offset obtained by calling __next_free_blkoff
1033 */
1034static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1035 struct curseg_info *seg)
1036{
1037 if (seg->alloc_type == SSR)
1038 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1039 else
1040 seg->next_blkoff++;
1041}
1042
0a8165d7 1043/*
e1c42045 1044 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1045 * manner, so it should recover the existing segment information of valid blocks
1046 */
1047static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1048{
1049 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1050 struct curseg_info *curseg = CURSEG_I(sbi, type);
1051 unsigned int new_segno = curseg->next_segno;
1052 struct f2fs_summary_block *sum_node;
1053 struct page *sum_page;
1054
1055 write_sum_page(sbi, curseg->sum_blk,
1056 GET_SUM_BLOCK(sbi, curseg->segno));
1057 __set_test_and_inuse(sbi, new_segno);
1058
1059 mutex_lock(&dirty_i->seglist_lock);
1060 __remove_dirty_segment(sbi, new_segno, PRE);
1061 __remove_dirty_segment(sbi, new_segno, DIRTY);
1062 mutex_unlock(&dirty_i->seglist_lock);
1063
1064 reset_curseg(sbi, type, 1);
1065 curseg->alloc_type = SSR;
1066 __next_free_blkoff(sbi, curseg, 0);
1067
1068 if (reuse) {
1069 sum_page = get_sum_page(sbi, new_segno);
1070 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1071 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1072 f2fs_put_page(sum_page, 1);
1073 }
1074}
1075
43727527
JK		 1076static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1077{
1078 struct curseg_info *curseg = CURSEG_I(sbi, type);
1079 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1080
1081 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1082 return v_ops->get_victim(sbi,
1083 &(curseg)->next_segno, BG_GC, type, SSR);
1084
1085 /* For data segments, let's do SSR more intensively */
1086 for (; type >= CURSEG_HOT_DATA; type--)
1087 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1088 BG_GC, type, SSR))
1089 return 1;
1090 return 0;
1091}
1092
351df4b2
JK		 1093/*
1094 * flush out current segment and replace it with new segment
1095 * This function should be returned with success, otherwise BUG
1096 */
1097static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1098 int type, bool force)
1099{
1100 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1101
7b405275 1102 if (force)
351df4b2 1103 new_curseg(sbi, type, true);
7b405275 1104 else if (type == CURSEG_WARM_NODE)
351df4b2 1105 new_curseg(sbi, type, false);
60374688
JK		 1106 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1107 new_curseg(sbi, type, false);
351df4b2
JK		 1108 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1109 change_curseg(sbi, type, true);
1110 else
1111 new_curseg(sbi, type, false);
dcdfff65
JK		 1112
1113 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1114}
1115
38aa0889
JK		 1116static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1117{
1118 struct curseg_info *curseg = CURSEG_I(sbi, type);
1119 unsigned int old_segno;
1120
1121 old_segno = curseg->segno;
1122 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1123 locate_dirty_segment(sbi, old_segno);
1124}
1125
351df4b2
JK		 1126void allocate_new_segments(struct f2fs_sb_info *sbi)
1127{
351df4b2
JK		 1128 int i;
1129
38aa0889
JK		 1130 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1131 __allocate_new_segments(sbi, i);
351df4b2
JK		 1132}
1133
1134static const struct segment_allocation default_salloc_ops = {
1135 .allocate_segment = allocate_segment_by_default,
1136};
1137
4b2fecc8
JK		 1138int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1139{
f7ef9b83
JK		 1140 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1141 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1142 unsigned int start_segno, end_segno;
1143 struct cp_control cpc;
1144
836b5a63 1145 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1146 return -EINVAL;
1147
9bd27ae4 1148 cpc.trimmed = 0;
7cd8558b 1149 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1150 goto out;
1151
1152 /* start/end segment number in main_area */
7cd8558b
JK		 1153 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1154 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1155 GET_SEGNO(sbi, end);
4b2fecc8 1156 cpc.reason = CP_DISCARD;
836b5a63 1157 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1158
1159 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1160 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1161 cpc.trim_start = start_segno;
a66cdd98
JK		 1162
1163 if (sbi->discard_blks == 0)
1164 break;
1165 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1166 cpc.trim_end = end_segno;
1167 else
1168 cpc.trim_end = min_t(unsigned int,
1169 rounddown(start_segno +
bba681cb
JK		 1170 BATCHED_TRIM_SEGMENTS(sbi),
1171 sbi->segs_per_sec) - 1, end_segno);
1172
1173 mutex_lock(&sbi->gc_mutex);
1174 write_checkpoint(sbi, &cpc);
1175 mutex_unlock(&sbi->gc_mutex);
1176 }
4b2fecc8 1177out:
f7ef9b83 1178 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
4b2fecc8
JK		 1179 return 0;
1180}
1181
351df4b2
JK		 1182static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1183{
1184 struct curseg_info *curseg = CURSEG_I(sbi, type);
1185 if (curseg->next_blkoff < sbi->blocks_per_seg)
1186 return true;
1187 return false;
1188}
1189
1190static int __get_segment_type_2(struct page *page, enum page_type p_type)
1191{
1192 if (p_type == DATA)
1193 return CURSEG_HOT_DATA;
1194 else
1195 return CURSEG_HOT_NODE;
1196}
1197
1198static int __get_segment_type_4(struct page *page, enum page_type p_type)
1199{
1200 if (p_type == DATA) {
1201 struct inode *inode = page->mapping->host;
1202
1203 if (S_ISDIR(inode->i_mode))
1204 return CURSEG_HOT_DATA;
1205 else
1206 return CURSEG_COLD_DATA;
1207 } else {
a344b9fd
JK		 1208 if (IS_DNODE(page) && is_cold_node(page))
1209 return CURSEG_WARM_NODE;
351df4b2
JK		 1210 else
1211 return CURSEG_COLD_NODE;
1212 }
1213}
1214
1215static int __get_segment_type_6(struct page *page, enum page_type p_type)
1216{
1217 if (p_type == DATA) {
1218 struct inode *inode = page->mapping->host;
1219
1220 if (S_ISDIR(inode->i_mode))
1221 return CURSEG_HOT_DATA;
354a3399 1222 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1223 return CURSEG_COLD_DATA;
1224 else
1225 return CURSEG_WARM_DATA;
1226 } else {
1227 if (IS_DNODE(page))
1228 return is_cold_node(page) ? CURSEG_WARM_NODE :
1229 CURSEG_HOT_NODE;
1230 else
1231 return CURSEG_COLD_NODE;
1232 }
1233}
1234
1235static int __get_segment_type(struct page *page, enum page_type p_type)
1236{
4081363f 1237 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1238 case 2:
1239 return __get_segment_type_2(page, p_type);
1240 case 4:
1241 return __get_segment_type_4(page, p_type);
351df4b2 1242 }
12a67146 1243 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1244 f2fs_bug_on(F2FS_P_SB(page),
1245 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1246 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1247}
1248
bfad7c2d
JK		 1249void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1250 block_t old_blkaddr, block_t *new_blkaddr,
1251 struct f2fs_summary *sum, int type)
351df4b2
JK		 1252{
1253 struct sit_info *sit_i = SIT_I(sbi);
1254 struct curseg_info *curseg;
38aa0889
JK		 1255 bool direct_io = (type == CURSEG_DIRECT_IO);
1256
1257 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1258
351df4b2
JK		 1259 curseg = CURSEG_I(sbi, type);
1260
1261 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1262 mutex_lock(&sit_i->sentry_lock);
351df4b2 1263
38aa0889 1264 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1265 if (direct_io && curseg->next_blkoff &&
1266 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1267 __allocate_new_segments(sbi, type);
1268
351df4b2 1269 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1270
1271 /*
1272 * __add_sum_entry should be resided under the curseg_mutex
1273 * because, this function updates a summary entry in the
1274 * current summary block.
1275 */
e79efe3b 1276 __add_sum_entry(sbi, type, sum);
351df4b2 1277
351df4b2 1278 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1279
1280 stat_inc_block_count(sbi, curseg);
351df4b2 1281
5e443818
JK		 1282 if (!__has_curseg_space(sbi, type))
1283 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1284 /*
1285 * SIT information should be updated before segment allocation,
1286 * since SSR needs latest valid block information.
1287 */
1288 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1289
351df4b2
JK		 1290 mutex_unlock(&sit_i->sentry_lock);
1291
bfad7c2d 1292 if (page && IS_NODESEG(type))
351df4b2
JK		 1293 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1294
bfad7c2d
JK		 1295 mutex_unlock(&curseg->curseg_mutex);
1296}
1297
05ca3632 1298static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1299{
05ca3632 1300 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1301
05ca3632
JK		 1302 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1303 &fio->blk_addr, sum, type);
bfad7c2d 1304
351df4b2 1305 /* writeout dirty page into bdev */
05ca3632 1306 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1307}
1308
577e3495 1309void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1310{
458e6197 1311 struct f2fs_io_info fio = {
05ca3632 1312 .sbi = sbi,
458e6197 1313 .type = META,
cf04e8eb
JK		 1314 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1315 .blk_addr = page->index,
05ca3632 1316 .page = page,
4375a336 1317 .encrypted_page = NULL,
458e6197
JK		 1318 };
1319
351df4b2 1320 set_page_writeback(page);
05ca3632 1321 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1322}
1323
05ca3632 1324void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1325{
1326 struct f2fs_summary sum;
05ca3632 1327
351df4b2 1328 set_summary(&sum, nid, 0, 0);
05ca3632 1329 do_write_page(&sum, fio);
351df4b2
JK		 1330}
1331
05ca3632 1332void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1333{
05ca3632 1334 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1335 struct f2fs_summary sum;
1336 struct node_info ni;
1337
9850cf4a 1338 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1339 get_node_info(sbi, dn->nid, &ni);
1340 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1341 do_write_page(&sum, fio);
e1509cf2 1342 dn->data_blkaddr = fio->blk_addr;
351df4b2
JK		 1343}
1344
05ca3632 1345void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1346{
05ca3632
JK		 1347 stat_inc_inplace_blocks(fio->sbi);
1348 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1349}
1350
528e3459
CY		 1351static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1352 struct f2fs_summary *sum,
19f106bc
CY		 1353 block_t old_blkaddr, block_t new_blkaddr,
1354 bool recover_curseg)
351df4b2
JK		 1355{
1356 struct sit_info *sit_i = SIT_I(sbi);
1357 struct curseg_info *curseg;
1358 unsigned int segno, old_cursegno;
1359 struct seg_entry *se;
1360 int type;
19f106bc 1361 unsigned short old_blkoff;
351df4b2
JK		 1362
1363 segno = GET_SEGNO(sbi, new_blkaddr);
1364 se = get_seg_entry(sbi, segno);
1365 type = se->type;
1366
19f106bc
CY		 1367 if (!recover_curseg) {
1368 /* for recovery flow */
1369 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1370 if (old_blkaddr == NULL_ADDR)
1371 type = CURSEG_COLD_DATA;
1372 else
1373 type = CURSEG_WARM_DATA;
1374 }
1375 } else {
1376 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1377 type = CURSEG_WARM_DATA;
1378 }
19f106bc 1379
351df4b2
JK		 1380 curseg = CURSEG_I(sbi, type);
1381
1382 mutex_lock(&curseg->curseg_mutex);
1383 mutex_lock(&sit_i->sentry_lock);
1384
1385 old_cursegno = curseg->segno;
19f106bc 1386 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1387
1388 /* change the current segment */
1389 if (segno != curseg->segno) {
1390 curseg->next_segno = segno;
1391 change_curseg(sbi, type, true);
1392 }
1393
491c0854 1394 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1395 __add_sum_entry(sbi, type, sum);
351df4b2 1396
6e2c64ad
JK		 1397 if (!recover_curseg)
1398 update_sit_entry(sbi, new_blkaddr, 1);
1399 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1400 update_sit_entry(sbi, old_blkaddr, -1);
1401
1402 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1403 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1404
351df4b2 1405 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1406
19f106bc
CY		 1407 if (recover_curseg) {
1408 if (old_cursegno != curseg->segno) {
1409 curseg->next_segno = old_cursegno;
1410 change_curseg(sbi, type, true);
1411 }
1412 curseg->next_blkoff = old_blkoff;
1413 }
1414
351df4b2
JK		 1415 mutex_unlock(&sit_i->sentry_lock);
1416 mutex_unlock(&curseg->curseg_mutex);
1417}
1418
528e3459
CY		 1419void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1420 block_t old_addr, block_t new_addr,
1421 unsigned char version, bool recover_curseg)
1422{
1423 struct f2fs_summary sum;
1424
1425 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1426
1427 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1428
1429 dn->data_blkaddr = new_addr;
1430 set_data_blkaddr(dn);
1431 f2fs_update_extent_cache(dn);
1432}
1433
df0f8dc0
CY		 1434static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1435 struct page *page, enum page_type type)
1436{
1437 enum page_type btype = PAGE_TYPE_OF_BIO(type);
1438 struct f2fs_bio_info *io = &sbi->write_io[btype];
df0f8dc0 1439 struct bio_vec *bvec;
4375a336 1440 struct page *target;
df0f8dc0
CY		 1441 int i;
1442
1443 down_read(&io->io_rwsem);
4375a336
JK		 1444 if (!io->bio) {
1445 up_read(&io->io_rwsem);
1446 return false;
1447 }
df0f8dc0 1448
ce23447f 1449 bio_for_each_segment_all(bvec, io->bio, i) {
4375a336
JK		 1450
1451 if (bvec->bv_page->mapping) {
1452 target = bvec->bv_page;
1453 } else {
1454 struct f2fs_crypto_ctx *ctx;
1455
1456 /* encrypted page */
1457 ctx = (struct f2fs_crypto_ctx *)page_private(
1458 bvec->bv_page);
ca40b030 1459 target = ctx->w.control_page;
4375a336
JK		 1460 }
1461
1462 if (page == target) {
df0f8dc0
CY		 1463 up_read(&io->io_rwsem);
1464 return true;
1465 }
1466 }
1467
df0f8dc0
CY		 1468 up_read(&io->io_rwsem);
1469 return false;
1470}
1471
93dfe2ac 1472void f2fs_wait_on_page_writeback(struct page *page,
5514f0aa 1473 enum page_type type)
93dfe2ac 1474{
93dfe2ac 1475 if (PageWriteback(page)) {
4081363f
JK		 1476 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1477
df0f8dc0
CY		 1478 if (is_merged_page(sbi, page, type))
1479 f2fs_submit_merged_bio(sbi, type, WRITE);
93dfe2ac
JK		 1480 wait_on_page_writeback(page);
1481 }
1482}
1483
351df4b2
JK		 1484static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1485{
1486 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1487 struct curseg_info *seg_i;
1488 unsigned char *kaddr;
1489 struct page *page;
1490 block_t start;
1491 int i, j, offset;
1492
1493 start = start_sum_block(sbi);
1494
1495 page = get_meta_page(sbi, start++);
1496 kaddr = (unsigned char *)page_address(page);
1497
1498 /* Step 1: restore nat cache */
1499 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1500 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1501
1502 /* Step 2: restore sit cache */
1503 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1504 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1505 SUM_JOURNAL_SIZE);
1506 offset = 2 * SUM_JOURNAL_SIZE;
1507
1508 /* Step 3: restore summary entries */
1509 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1510 unsigned short blk_off;
1511 unsigned int segno;
1512
1513 seg_i = CURSEG_I(sbi, i);
1514 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1515 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1516 seg_i->next_segno = segno;
1517 reset_curseg(sbi, i, 0);
1518 seg_i->alloc_type = ckpt->alloc_type[i];
1519 seg_i->next_blkoff = blk_off;
1520
1521 if (seg_i->alloc_type == SSR)
1522 blk_off = sbi->blocks_per_seg;
1523
1524 for (j = 0; j < blk_off; j++) {
1525 struct f2fs_summary *s;
1526 s = (struct f2fs_summary *)(kaddr + offset);
1527 seg_i->sum_blk->entries[j] = *s;
1528 offset += SUMMARY_SIZE;
1529 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1530 SUM_FOOTER_SIZE)
1531 continue;
1532
1533 f2fs_put_page(page, 1);
1534 page = NULL;
1535
1536 page = get_meta_page(sbi, start++);
1537 kaddr = (unsigned char *)page_address(page);
1538 offset = 0;
1539 }
1540 }
1541 f2fs_put_page(page, 1);
1542 return 0;
1543}
1544
1545static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1546{
1547 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1548 struct f2fs_summary_block *sum;
1549 struct curseg_info *curseg;
1550 struct page *new;
1551 unsigned short blk_off;
1552 unsigned int segno = 0;
1553 block_t blk_addr = 0;
1554
1555 /* get segment number and block addr */
1556 if (IS_DATASEG(type)) {
1557 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1558 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1559 CURSEG_HOT_DATA]);
119ee914 1560 if (__exist_node_summaries(sbi))
351df4b2
JK		 1561 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1562 else
1563 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1564 } else {
1565 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1566 CURSEG_HOT_NODE]);
1567 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1568 CURSEG_HOT_NODE]);
119ee914 1569 if (__exist_node_summaries(sbi))
351df4b2
JK		 1570 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1571 type - CURSEG_HOT_NODE);
1572 else
1573 blk_addr = GET_SUM_BLOCK(sbi, segno);
1574 }
1575
1576 new = get_meta_page(sbi, blk_addr);
1577 sum = (struct f2fs_summary_block *)page_address(new);
1578
1579 if (IS_NODESEG(type)) {
119ee914 1580 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1581 struct f2fs_summary *ns = &sum->entries[0];
1582 int i;
1583 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1584 ns->version = 0;
1585 ns->ofs_in_node = 0;
1586 }
1587 } else {
d653788a
GZ		 1588 int err;
1589
1590 err = restore_node_summary(sbi, segno, sum);
1591 if (err) {
351df4b2 1592 f2fs_put_page(new, 1);
d653788a 1593 return err;
351df4b2
JK		 1594 }
1595 }
1596 }
1597
1598 /* set uncompleted segment to curseg */
1599 curseg = CURSEG_I(sbi, type);
1600 mutex_lock(&curseg->curseg_mutex);
1601 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1602 curseg->next_segno = segno;
1603 reset_curseg(sbi, type, 0);
1604 curseg->alloc_type = ckpt->alloc_type[type];
1605 curseg->next_blkoff = blk_off;
1606 mutex_unlock(&curseg->curseg_mutex);
1607 f2fs_put_page(new, 1);
1608 return 0;
1609}
1610
1611static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1612{
1613 int type = CURSEG_HOT_DATA;
e4fc5fbf 1614 int err;
351df4b2 1615
25ca923b 1616 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1617 int npages = npages_for_summary_flush(sbi, true);
1618
1619 if (npages >= 2)
1620 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1621 META_CP);
1622
351df4b2
JK		 1623 /* restore for compacted data summary */
1624 if (read_compacted_summaries(sbi))
1625 return -EINVAL;
1626 type = CURSEG_HOT_NODE;
1627 }
1628
119ee914 1629 if (__exist_node_summaries(sbi))
3fa06d7b
CY		 1630 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1631 NR_CURSEG_TYPE - type, META_CP);
1632
e4fc5fbf
CY		 1633 for (; type <= CURSEG_COLD_NODE; type++) {
1634 err = read_normal_summaries(sbi, type);
1635 if (err)
1636 return err;
1637 }
1638
351df4b2
JK		 1639 return 0;
1640}
1641
1642static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1643{
1644 struct page *page;
1645 unsigned char *kaddr;
1646 struct f2fs_summary *summary;
1647 struct curseg_info *seg_i;
1648 int written_size = 0;
1649 int i, j;
1650
1651 page = grab_meta_page(sbi, blkaddr++);
1652 kaddr = (unsigned char *)page_address(page);
1653
1654 /* Step 1: write nat cache */
1655 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1656 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1657 written_size += SUM_JOURNAL_SIZE;
1658
1659 /* Step 2: write sit cache */
1660 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1661 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1662 SUM_JOURNAL_SIZE);
1663 written_size += SUM_JOURNAL_SIZE;
1664
351df4b2
JK		 1665 /* Step 3: write summary entries */
1666 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1667 unsigned short blkoff;
1668 seg_i = CURSEG_I(sbi, i);
1669 if (sbi->ckpt->alloc_type[i] == SSR)
1670 blkoff = sbi->blocks_per_seg;
1671 else
1672 blkoff = curseg_blkoff(sbi, i);
1673
1674 for (j = 0; j < blkoff; j++) {
1675 if (!page) {
1676 page = grab_meta_page(sbi, blkaddr++);
1677 kaddr = (unsigned char *)page_address(page);
1678 written_size = 0;
1679 }
1680 summary = (struct f2fs_summary *)(kaddr + written_size);
1681 *summary = seg_i->sum_blk->entries[j];
1682 written_size += SUMMARY_SIZE;
351df4b2
JK		 1683
1684 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1685 SUM_FOOTER_SIZE)
1686 continue;
1687
e8d61a74 1688 set_page_dirty(page);
351df4b2
JK		 1689 f2fs_put_page(page, 1);
1690 page = NULL;
1691 }
1692 }
e8d61a74
CY		 1693 if (page) {
1694 set_page_dirty(page);
351df4b2 1695 f2fs_put_page(page, 1);
e8d61a74 1696 }
351df4b2
JK		 1697}
1698
1699static void write_normal_summaries(struct f2fs_sb_info *sbi,
1700 block_t blkaddr, int type)
1701{
1702 int i, end;
1703 if (IS_DATASEG(type))
1704 end = type + NR_CURSEG_DATA_TYPE;
1705 else
1706 end = type + NR_CURSEG_NODE_TYPE;
1707
1708 for (i = type; i < end; i++) {
1709 struct curseg_info *sum = CURSEG_I(sbi, i);
1710 mutex_lock(&sum->curseg_mutex);
1711 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1712 mutex_unlock(&sum->curseg_mutex);
1713 }
1714}
1715
1716void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1717{
25ca923b 1718 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1719 write_compacted_summaries(sbi, start_blk);
1720 else
1721 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1722}
1723
1724void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1725{
119ee914 1726 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1727}
1728
1729int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1730 unsigned int val, int alloc)
1731{
1732 int i;
1733
1734 if (type == NAT_JOURNAL) {
1735 for (i = 0; i < nats_in_cursum(sum); i++) {
1736 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1737 return i;
1738 }
1739 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1740 return update_nats_in_cursum(sum, 1);
1741 } else if (type == SIT_JOURNAL) {
1742 for (i = 0; i < sits_in_cursum(sum); i++)
1743 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1744 return i;
1745 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1746 return update_sits_in_cursum(sum, 1);
1747 }
1748 return -1;
1749}
1750
1751static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1752 unsigned int segno)
1753{
2cc22186 1754 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1755}
1756
1757static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1758 unsigned int start)
1759{
1760 struct sit_info *sit_i = SIT_I(sbi);
1761 struct page *src_page, *dst_page;
1762 pgoff_t src_off, dst_off;
1763 void *src_addr, *dst_addr;
1764
1765 src_off = current_sit_addr(sbi, start);
1766 dst_off = next_sit_addr(sbi, src_off);
1767
1768 /* get current sit block page without lock */
1769 src_page = get_meta_page(sbi, src_off);
1770 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1771 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1772
1773 src_addr = page_address(src_page);
1774 dst_addr = page_address(dst_page);
1775 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1776
1777 set_page_dirty(dst_page);
1778 f2fs_put_page(src_page, 1);
1779
1780 set_to_next_sit(sit_i, start);
1781
1782 return dst_page;
1783}
1784
184a5cd2
CY		 1785static struct sit_entry_set *grab_sit_entry_set(void)
1786{
1787 struct sit_entry_set *ses =
80c54505 1788 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1789
1790 ses->entry_cnt = 0;
1791 INIT_LIST_HEAD(&ses->set_list);
1792 return ses;
1793}
1794
1795static void release_sit_entry_set(struct sit_entry_set *ses)
1796{
1797 list_del(&ses->set_list);
1798 kmem_cache_free(sit_entry_set_slab, ses);
1799}
1800
1801static void adjust_sit_entry_set(struct sit_entry_set *ses,
1802 struct list_head *head)
1803{
1804 struct sit_entry_set *next = ses;
1805
1806 if (list_is_last(&ses->set_list, head))
1807 return;
1808
1809 list_for_each_entry_continue(next, head, set_list)
1810 if (ses->entry_cnt <= next->entry_cnt)
1811 break;
1812
1813 list_move_tail(&ses->set_list, &next->set_list);
1814}
1815
1816static void add_sit_entry(unsigned int segno, struct list_head *head)
1817{
1818 struct sit_entry_set *ses;
1819 unsigned int start_segno = START_SEGNO(segno);
1820
1821 list_for_each_entry(ses, head, set_list) {
1822 if (ses->start_segno == start_segno) {
1823 ses->entry_cnt++;
1824 adjust_sit_entry_set(ses, head);
1825 return;
1826 }
1827 }
1828
1829 ses = grab_sit_entry_set();
1830
1831 ses->start_segno = start_segno;
1832 ses->entry_cnt++;
1833 list_add(&ses->set_list, head);
1834}
1835
1836static void add_sits_in_set(struct f2fs_sb_info *sbi)
1837{
1838 struct f2fs_sm_info *sm_info = SM_I(sbi);
1839 struct list_head *set_list = &sm_info->sit_entry_set;
1840 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1841 unsigned int segno;
1842
7cd8558b 1843 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1844 add_sit_entry(segno, set_list);
1845}
1846
1847static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1848{
1849 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1850 struct f2fs_summary_block *sum = curseg->sum_blk;
1851 int i;
1852
184a5cd2
CY		 1853 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1854 unsigned int segno;
1855 bool dirtied;
1856
1857 segno = le32_to_cpu(segno_in_journal(sum, i));
1858 dirtied = __mark_sit_entry_dirty(sbi, segno);
1859
1860 if (!dirtied)
1861 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1862 }
184a5cd2 1863 update_sits_in_cursum(sum, -sits_in_cursum(sum));
351df4b2
JK		 1864}
1865
0a8165d7 1866/*
351df4b2
JK		 1867 * CP calls this function, which flushes SIT entries including sit_journal,
1868 * and moves prefree segs to free segs.
1869 */
4b2fecc8 1870void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1871{
1872 struct sit_info *sit_i = SIT_I(sbi);
1873 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1874 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1875 struct f2fs_summary_block *sum = curseg->sum_blk;
184a5cd2
CY		 1876 struct sit_entry_set *ses, *tmp;
1877 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1878 bool to_journal = true;
4b2fecc8 1879 struct seg_entry *se;
351df4b2
JK		 1880
1881 mutex_lock(&curseg->curseg_mutex);
1882 mutex_lock(&sit_i->sentry_lock);
1883
2b11a74b
WL		 1884 if (!sit_i->dirty_sentries)
1885 goto out;
1886
351df4b2 1887 /*
184a5cd2
CY		 1888 * add and account sit entries of dirty bitmap in sit entry
1889 * set temporarily
351df4b2 1890 */
184a5cd2 1891 add_sits_in_set(sbi);
351df4b2 1892
184a5cd2
CY		 1893 /*
1894 * if there are no enough space in journal to store dirty sit
1895 * entries, remove all entries from journal and add and account
1896 * them in sit entry set.
1897 */
1898 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1899 remove_sits_in_journal(sbi);
b2955550 1900
184a5cd2
CY		 1901 /*
1902 * there are two steps to flush sit entries:
1903 * #1, flush sit entries to journal in current cold data summary block.
1904 * #2, flush sit entries to sit page.
1905 */
1906 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1907 struct page *page = NULL;
184a5cd2
CY		 1908 struct f2fs_sit_block *raw_sit = NULL;
1909 unsigned int start_segno = ses->start_segno;
1910 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1911 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1912 unsigned int segno = start_segno;
1913
1914 if (to_journal &&
1915 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1916 to_journal = false;
1917
1918 if (!to_journal) {
1919 page = get_next_sit_page(sbi, start_segno);
1920 raw_sit = page_address(page);
351df4b2 1921 }
351df4b2 1922
184a5cd2
CY		 1923 /* flush dirty sit entries in region of current sit set */
1924 for_each_set_bit_from(segno, bitmap, end) {
1925 int offset, sit_offset;
4b2fecc8
JK		 1926
1927 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 1928
1929 /* add discard candidates */
d7bc2484 1930 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 1931 cpc->trim_start = segno;
1932 add_discard_addrs(sbi, cpc);
1933 }
184a5cd2
CY		 1934
1935 if (to_journal) {
1936 offset = lookup_journal_in_cursum(sum,
1937 SIT_JOURNAL, segno, 1);
1938 f2fs_bug_on(sbi, offset < 0);
1939 segno_in_journal(sum, offset) =
1940 cpu_to_le32(segno);
1941 seg_info_to_raw_sit(se,
1942 &sit_in_journal(sum, offset));
1943 } else {
1944 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1945 seg_info_to_raw_sit(se,
1946 &raw_sit->entries[sit_offset]);
1947 }
351df4b2 1948
184a5cd2
CY		 1949 __clear_bit(segno, bitmap);
1950 sit_i->dirty_sentries--;
1951 ses->entry_cnt--;
351df4b2
JK		 1952 }
1953
184a5cd2
CY		 1954 if (!to_journal)
1955 f2fs_put_page(page, 1);
1956
1957 f2fs_bug_on(sbi, ses->entry_cnt);
1958 release_sit_entry_set(ses);
351df4b2 1959 }
184a5cd2
CY		 1960
1961 f2fs_bug_on(sbi, !list_empty(head));
1962 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 1963out:
4b2fecc8
JK		 1964 if (cpc->reason == CP_DISCARD) {
1965 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1966 add_discard_addrs(sbi, cpc);
1967 }
351df4b2
JK		 1968 mutex_unlock(&sit_i->sentry_lock);
1969 mutex_unlock(&curseg->curseg_mutex);
1970
351df4b2
JK		 1971 set_prefree_as_free_segments(sbi);
1972}
1973
1974static int build_sit_info(struct f2fs_sb_info *sbi)
1975{
1976 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1977 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1978 struct sit_info *sit_i;
1979 unsigned int sit_segs, start;
1980 char *src_bitmap, *dst_bitmap;
1981 unsigned int bitmap_size;
1982
1983 /* allocate memory for SIT information */
1984 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1985 if (!sit_i)
1986 return -ENOMEM;
1987
1988 SM_I(sbi)->sit_info = sit_i;
1989
39307a8e
JK		 1990 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
1991 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 1992 if (!sit_i->sentries)
1993 return -ENOMEM;
1994
7cd8558b 1995 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 1996 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 1997 if (!sit_i->dirty_sentries_bitmap)
1998 return -ENOMEM;
1999
7cd8558b 2000 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2001 sit_i->sentries[start].cur_valid_map
2002 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2003 sit_i->sentries[start].ckpt_valid_map
2004 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 2005 sit_i->sentries[start].discard_map
2006 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2007 if (!sit_i->sentries[start].cur_valid_map ||
2008 !sit_i->sentries[start].ckpt_valid_map ||
2009 !sit_i->sentries[start].discard_map)
351df4b2
JK		 2010 return -ENOMEM;
2011 }
2012
60a3b782
JK		 2013 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2014 if (!sit_i->tmp_map)
2015 return -ENOMEM;
2016
351df4b2 2017 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2018 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2019 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2020 if (!sit_i->sec_entries)
2021 return -ENOMEM;
2022 }
2023
2024 /* get information related with SIT */
2025 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2026
2027 /* setup SIT bitmap from ckeckpoint pack */
2028 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2029 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2030
79b5793b 2031 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2032 if (!dst_bitmap)
2033 return -ENOMEM;
351df4b2
JK		 2034
2035 /* init SIT information */
2036 sit_i->s_ops = &default_salloc_ops;
2037
2038 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2039 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2040 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2041 sit_i->sit_bitmap = dst_bitmap;
2042 sit_i->bitmap_size = bitmap_size;
2043 sit_i->dirty_sentries = 0;
2044 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2045 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2046 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2047 mutex_init(&sit_i->sentry_lock);
2048 return 0;
2049}
2050
2051static int build_free_segmap(struct f2fs_sb_info *sbi)
2052{
351df4b2
JK		 2053 struct free_segmap_info *free_i;
2054 unsigned int bitmap_size, sec_bitmap_size;
2055
2056 /* allocate memory for free segmap information */
2057 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2058 if (!free_i)
2059 return -ENOMEM;
2060
2061 SM_I(sbi)->free_info = free_i;
2062
7cd8558b 2063 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2064 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2065 if (!free_i->free_segmap)
2066 return -ENOMEM;
2067
7cd8558b 2068 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2069 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2070 if (!free_i->free_secmap)
2071 return -ENOMEM;
2072
2073 /* set all segments as dirty temporarily */
2074 memset(free_i->free_segmap, 0xff, bitmap_size);
2075 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2076
2077 /* init free segmap information */
7cd8558b 2078 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2079 free_i->free_segments = 0;
2080 free_i->free_sections = 0;
1a118ccf 2081 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2082 return 0;
2083}
2084
2085static int build_curseg(struct f2fs_sb_info *sbi)
2086{
1042d60f 2087 struct curseg_info *array;
351df4b2
JK		 2088 int i;
2089
b434babf 2090 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2091 if (!array)
2092 return -ENOMEM;
2093
2094 SM_I(sbi)->curseg_array = array;
2095
2096 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2097 mutex_init(&array[i].curseg_mutex);
2098 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2099 if (!array[i].sum_blk)
2100 return -ENOMEM;
2101 array[i].segno = NULL_SEGNO;
2102 array[i].next_blkoff = 0;
2103 }
2104 return restore_curseg_summaries(sbi);
2105}
2106
2107static void build_sit_entries(struct f2fs_sb_info *sbi)
2108{
2109 struct sit_info *sit_i = SIT_I(sbi);
2110 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2111 struct f2fs_summary_block *sum = curseg->sum_blk;
74de593a
CY		 2112 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2113 unsigned int i, start, end;
2114 unsigned int readed, start_blk = 0;
90a893c7 2115 int nrpages = MAX_BIO_BLOCKS(sbi);
351df4b2 2116
74de593a 2117 do {
662befda 2118 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
74de593a
CY		 2119
2120 start = start_blk * sit_i->sents_per_block;
2121 end = (start_blk + readed) * sit_i->sents_per_block;
2122
7cd8558b 2123 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2124 struct seg_entry *se = &sit_i->sentries[start];
2125 struct f2fs_sit_block *sit_blk;
2126 struct f2fs_sit_entry sit;
2127 struct page *page;
2128
2129 mutex_lock(&curseg->curseg_mutex);
2130 for (i = 0; i < sits_in_cursum(sum); i++) {
6c311ec6
CF		 2131 if (le32_to_cpu(segno_in_journal(sum, i))
2132 == start) {
74de593a
CY		 2133 sit = sit_in_journal(sum, i);
2134 mutex_unlock(&curseg->curseg_mutex);
2135 goto got_it;
2136 }
351df4b2 2137 }
74de593a
CY		 2138 mutex_unlock(&curseg->curseg_mutex);
2139
2140 page = get_current_sit_page(sbi, start);
2141 sit_blk = (struct f2fs_sit_block *)page_address(page);
2142 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2143 f2fs_put_page(page, 1);
351df4b2 2144got_it:
74de593a
CY		 2145 check_block_count(sbi, start, &sit);
2146 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2147
2148 /* build discard map only one time */
2149 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2150 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2151
74de593a
CY		 2152 if (sbi->segs_per_sec > 1) {
2153 struct sec_entry *e = get_sec_entry(sbi, start);
2154 e->valid_blocks += se->valid_blocks;
2155 }
351df4b2 2156 }
74de593a
CY		 2157 start_blk += readed;
2158 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2159}
2160
2161static void init_free_segmap(struct f2fs_sb_info *sbi)
2162{
2163 unsigned int start;
2164 int type;
2165
7cd8558b 2166 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2167 struct seg_entry *sentry = get_seg_entry(sbi, start);
2168 if (!sentry->valid_blocks)
2169 __set_free(sbi, start);
2170 }
2171
2172 /* set use the current segments */
2173 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2174 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2175 __set_test_and_inuse(sbi, curseg_t->segno);
2176 }
2177}
2178
2179static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2180{
2181 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2182 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2183 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2184 unsigned short valid_blocks;
2185
8736fbf0 2186 while (1) {
351df4b2 2187 /* find dirty segment based on free segmap */
7cd8558b
JK		 2188 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2189 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2190 break;
2191 offset = segno + 1;
2192 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2193 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2194 continue;
ec325b52
JK		 2195 if (valid_blocks > sbi->blocks_per_seg) {
2196 f2fs_bug_on(sbi, 1);
2197 continue;
2198 }
351df4b2
JK		 2199 mutex_lock(&dirty_i->seglist_lock);
2200 __locate_dirty_segment(sbi, segno, DIRTY);
2201 mutex_unlock(&dirty_i->seglist_lock);
2202 }
2203}
2204
5ec4e49f 2205static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2206{
2207 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2208 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2209
39307a8e 2210 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2211 if (!dirty_i->victim_secmap)
351df4b2
JK		 2212 return -ENOMEM;
2213 return 0;
2214}
2215
2216static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2217{
2218 struct dirty_seglist_info *dirty_i;
2219 unsigned int bitmap_size, i;
2220
2221 /* allocate memory for dirty segments list information */
2222 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2223 if (!dirty_i)
2224 return -ENOMEM;
2225
2226 SM_I(sbi)->dirty_info = dirty_i;
2227 mutex_init(&dirty_i->seglist_lock);
2228
7cd8558b 2229 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2230
2231 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2232 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2233 if (!dirty_i->dirty_segmap[i])
2234 return -ENOMEM;
2235 }
2236
2237 init_dirty_segmap(sbi);
5ec4e49f 2238 return init_victim_secmap(sbi);
351df4b2
JK		 2239}
2240
0a8165d7 2241/*
351df4b2
JK		 2242 * Update min, max modified time for cost-benefit GC algorithm
2243 */
2244static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2245{
2246 struct sit_info *sit_i = SIT_I(sbi);
2247 unsigned int segno;
2248
2249 mutex_lock(&sit_i->sentry_lock);
2250
2251 sit_i->min_mtime = LLONG_MAX;
2252
7cd8558b 2253 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2254 unsigned int i;
2255 unsigned long long mtime = 0;
2256
2257 for (i = 0; i < sbi->segs_per_sec; i++)
2258 mtime += get_seg_entry(sbi, segno + i)->mtime;
2259
2260 mtime = div_u64(mtime, sbi->segs_per_sec);
2261
2262 if (sit_i->min_mtime > mtime)
2263 sit_i->min_mtime = mtime;
2264 }
2265 sit_i->max_mtime = get_mtime(sbi);
2266 mutex_unlock(&sit_i->sentry_lock);
2267}
2268
2269int build_segment_manager(struct f2fs_sb_info *sbi)
2270{
2271 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2272 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2273 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2274 int err;
2275
2276 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2277 if (!sm_info)
2278 return -ENOMEM;
2279
2280 /* init sm info */
2281 sbi->sm_info = sm_info;
351df4b2
JK		 2282 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2283 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2284 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2285 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2286 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2287 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2288 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2289 sm_info->rec_prefree_segments = sm_info->main_segments *
2290 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2291 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2292 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2293 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2294
7fd9e544
JK		 2295 INIT_LIST_HEAD(&sm_info->discard_list);
2296 sm_info->nr_discards = 0;
2297 sm_info->max_discards = 0;
2298
bba681cb
JK		 2299 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2300
184a5cd2
CY		 2301 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2302
b270ad6f 2303 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2304 err = create_flush_cmd_control(sbi);
2305 if (err)
a688b9d9 2306 return err;
6b4afdd7
JK		 2307 }
2308
351df4b2
JK		 2309 err = build_sit_info(sbi);
2310 if (err)
2311 return err;
2312 err = build_free_segmap(sbi);
2313 if (err)
2314 return err;
2315 err = build_curseg(sbi);
2316 if (err)
2317 return err;
2318
2319 /* reinit free segmap based on SIT */
2320 build_sit_entries(sbi);
2321
2322 init_free_segmap(sbi);
2323 err = build_dirty_segmap(sbi);
2324 if (err)
2325 return err;
2326
2327 init_min_max_mtime(sbi);
2328 return 0;
2329}
2330
2331static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2332 enum dirty_type dirty_type)
2333{
2334 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2335
2336 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2337 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2338 dirty_i->nr_dirty[dirty_type] = 0;
2339 mutex_unlock(&dirty_i->seglist_lock);
2340}
2341
5ec4e49f 2342static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2343{
2344 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2345 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2346}
2347
2348static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2349{
2350 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2351 int i;
2352
2353 if (!dirty_i)
2354 return;
2355
2356 /* discard pre-free/dirty segments list */
2357 for (i = 0; i < NR_DIRTY_TYPE; i++)
2358 discard_dirty_segmap(sbi, i);
2359
5ec4e49f 2360 destroy_victim_secmap(sbi);
351df4b2
JK		 2361 SM_I(sbi)->dirty_info = NULL;
2362 kfree(dirty_i);
2363}
2364
2365static void destroy_curseg(struct f2fs_sb_info *sbi)
2366{
2367 struct curseg_info *array = SM_I(sbi)->curseg_array;
2368 int i;
2369
2370 if (!array)
2371 return;
2372 SM_I(sbi)->curseg_array = NULL;
2373 for (i = 0; i < NR_CURSEG_TYPE; i++)
2374 kfree(array[i].sum_blk);
2375 kfree(array);
2376}
2377
2378static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2379{
2380 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2381 if (!free_i)
2382 return;
2383 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2384 kvfree(free_i->free_segmap);
2385 kvfree(free_i->free_secmap);
351df4b2
JK		 2386 kfree(free_i);
2387}
2388
2389static void destroy_sit_info(struct f2fs_sb_info *sbi)
2390{
2391 struct sit_info *sit_i = SIT_I(sbi);
2392 unsigned int start;
2393
2394 if (!sit_i)
2395 return;
2396
2397 if (sit_i->sentries) {
7cd8558b 2398 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2399 kfree(sit_i->sentries[start].cur_valid_map);
2400 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2401 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2402 }
2403 }
60a3b782
JK		 2404 kfree(sit_i->tmp_map);
2405
39307a8e
JK		 2406 kvfree(sit_i->sentries);
2407 kvfree(sit_i->sec_entries);
2408 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2409
2410 SM_I(sbi)->sit_info = NULL;
2411 kfree(sit_i->sit_bitmap);
2412 kfree(sit_i);
2413}
2414
2415void destroy_segment_manager(struct f2fs_sb_info *sbi)
2416{
2417 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2418
3b03f724
CY		 2419 if (!sm_info)
2420 return;
2163d198 2421 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2422 destroy_dirty_segmap(sbi);
2423 destroy_curseg(sbi);
2424 destroy_free_segmap(sbi);
2425 destroy_sit_info(sbi);
2426 sbi->sm_info = NULL;
2427 kfree(sm_info);
2428}
7fd9e544
JK		 2429
2430int __init create_segment_manager_caches(void)
2431{
2432 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2433 sizeof(struct discard_entry));
7fd9e544 2434 if (!discard_entry_slab)
184a5cd2
CY		 2435 goto fail;
2436
2437 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2438 sizeof(struct sit_entry_set));
184a5cd2
CY		 2439 if (!sit_entry_set_slab)
2440 goto destory_discard_entry;
88b88a66
JK		 2441
2442 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2443 sizeof(struct inmem_pages));
2444 if (!inmem_entry_slab)
2445 goto destroy_sit_entry_set;
7fd9e544 2446 return 0;
184a5cd2 2447
88b88a66
JK		 2448destroy_sit_entry_set:
2449 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2450destory_discard_entry:
2451 kmem_cache_destroy(discard_entry_slab);
2452fail:
2453 return -ENOMEM;
7fd9e544
JK		 2454}
2455
2456void destroy_segment_manager_caches(void)
2457{
184a5cd2 2458 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2459 kmem_cache_destroy(discard_entry_slab);
88b88a66 2460 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2461}
Linux kernel - Deliverables
This page took 0.245724 seconds and 5 git commands to generate.