projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: detect idle time depending on user behavior
[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
f96999c3
JK		 32static unsigned long __reverse_ulong(unsigned char *str)
33{
34 unsigned long tmp = 0;
35 int shift = 24, idx = 0;
36
37#if BITS_PER_LONG == 64
38 shift = 56;
39#endif
40 while (shift >= 0) {
41 tmp |= (unsigned long)str[idx++] << shift;
42 shift -= BITS_PER_BYTE;
43 }
44 return tmp;
45}
46
9a7f143a
CL		 47/*
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
50 */
51static inline unsigned long __reverse_ffs(unsigned long word)
52{
53 int num = 0;
54
55#if BITS_PER_LONG == 64
f96999c3 56 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 57 num += 32;
f96999c3 58 else
9a7f143a 59 word >>= 32;
9a7f143a 60#endif
f96999c3 61 if ((word & 0xffff0000) == 0)
9a7f143a 62 num += 16;
f96999c3 63 else
9a7f143a 64 word >>= 16;
f96999c3
JK		 65
66 if ((word & 0xff00) == 0)
9a7f143a 67 num += 8;
f96999c3 68 else
9a7f143a 69 word >>= 8;
f96999c3 70
9a7f143a
CL		 71 if ((word & 0xf0) == 0)
72 num += 4;
73 else
74 word >>= 4;
f96999c3 75
9a7f143a
CL		 76 if ((word & 0xc) == 0)
77 num += 2;
78 else
79 word >>= 2;
f96999c3 80
9a7f143a
CL		 81 if ((word & 0x2) == 0)
82 num += 1;
83 return num;
84}
85
86/*
e1c42045 87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 89 * @size must be integral times of unsigned long.
9a7f143a 90 * Example:
f96999c3
JK		 91 * MSB <--> LSB
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 94 */
95static unsigned long __find_rev_next_bit(const unsigned long *addr,
96 unsigned long size, unsigned long offset)
97{
98 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 99 unsigned long result = size;
9a7f143a 100 unsigned long tmp;
9a7f143a
CL		 101
102 if (offset >= size)
103 return size;
104
692223d1 105 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 106 offset %= BITS_PER_LONG;
f96999c3 107
692223d1
FL		 108 while (1) {
109 if (*p == 0)
110 goto pass;
9a7f143a 111
f96999c3 112 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 113
114 tmp &= ~0UL >> offset;
115 if (size < BITS_PER_LONG)
116 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 117 if (tmp)
692223d1
FL		 118 goto found;
119pass:
120 if (size <= BITS_PER_LONG)
121 break;
9a7f143a 122 size -= BITS_PER_LONG;
692223d1 123 offset = 0;
f96999c3 124 p++;
9a7f143a 125 }
692223d1
FL		 126 return result;
127found:
128 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 129}
130
131static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
132 unsigned long size, unsigned long offset)
133{
134 const unsigned long *p = addr + BIT_WORD(offset);
80609448 135 unsigned long result = size;
9a7f143a 136 unsigned long tmp;
9a7f143a
CL		 137
138 if (offset >= size)
139 return size;
140
80609448 141 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 142 offset %= BITS_PER_LONG;
80609448
JK		 143
144 while (1) {
145 if (*p == ~0UL)
146 goto pass;
147
f96999c3 148 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 149
150 if (offset)
151 tmp |= ~0UL << (BITS_PER_LONG - offset);
152 if (size < BITS_PER_LONG)
153 tmp |= ~0UL >> size;
f96999c3 154 if (tmp != ~0UL)
80609448
JK		 155 goto found;
156pass:
157 if (size <= BITS_PER_LONG)
158 break;
9a7f143a 159 size -= BITS_PER_LONG;
80609448 160 offset = 0;
f96999c3 161 p++;
9a7f143a 162 }
80609448
JK		 163 return result;
164found:
165 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 166}
167
88b88a66
JK		 168void register_inmem_page(struct inode *inode, struct page *page)
169{
170 struct f2fs_inode_info *fi = F2FS_I(inode);
171 struct inmem_pages *new;
9be32d72 172
9e4ded3f 173 f2fs_trace_pid(page);
0722b101 174
decd36b6
CY		 175 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
176 SetPagePrivate(page);
177
88b88a66
JK		 178 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
179
180 /* add atomic page indices to the list */
181 new->page = page;
182 INIT_LIST_HEAD(&new->list);
decd36b6 183
88b88a66
JK		 184 /* increase reference count with clean state */
185 mutex_lock(&fi->inmem_lock);
186 get_page(page);
187 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 188 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 189 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 190
191 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 192}
193
edb27dee 194int commit_inmem_pages(struct inode *inode, bool abort)
88b88a66
JK		 195{
196 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
197 struct f2fs_inode_info *fi = F2FS_I(inode);
198 struct inmem_pages *cur, *tmp;
199 bool submit_bio = false;
200 struct f2fs_io_info fio = {
05ca3632 201 .sbi = sbi,
88b88a66 202 .type = DATA,
1e84371f 203 .rw = WRITE_SYNC | REQ_PRIO,
4375a336 204 .encrypted_page = NULL,
88b88a66 205 };
edb27dee 206 int err = 0;
88b88a66 207
0341845e
JK		 208 /*
209 * The abort is true only when f2fs_evict_inode is called.
210 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
211 * that we don't need to call f2fs_balance_fs.
212 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
213 * inode becomes free by iget_locked in f2fs_iget.
214 */
70c640b1 215 if (!abort) {
2c4db1a6 216 f2fs_balance_fs(sbi, true);
70c640b1
JK		 217 f2fs_lock_op(sbi);
218 }
88b88a66
JK		 219
220 mutex_lock(&fi->inmem_lock);
221 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
decd36b6 222 lock_page(cur->page);
70c640b1 223 if (!abort) {
70c640b1 224 if (cur->page->mapping == inode->i_mapping) {
6282adbf 225 set_page_dirty(cur->page);
70c640b1
JK		 226 f2fs_wait_on_page_writeback(cur->page, DATA);
227 if (clear_page_dirty_for_io(cur->page))
228 inode_dec_dirty_pages(inode);
8ce67cb0 229 trace_f2fs_commit_inmem_page(cur->page, INMEM);
05ca3632 230 fio.page = cur->page;
edb27dee 231 err = do_write_data_page(&fio);
edb27dee
JK		 232 if (err) {
233 unlock_page(cur->page);
234 break;
235 }
7fee7406 236 clear_cold_data(cur->page);
2b246fb0 237 submit_bio = true;
70c640b1 238 }
70c640b1 239 } else {
f478f43f 240 ClearPageUptodate(cur->page);
8ce67cb0 241 trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
88b88a66 242 }
decd36b6
CY		 243 set_page_private(cur->page, 0);
244 ClearPagePrivate(cur->page);
245 f2fs_put_page(cur->page, 1);
246
88b88a66
JK		 247 list_del(&cur->list);
248 kmem_cache_free(inmem_entry_slab, cur);
8dcf2ff7 249 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 250 }
88b88a66
JK		 251 mutex_unlock(&fi->inmem_lock);
252
70c640b1
JK		 253 if (!abort) {
254 f2fs_unlock_op(sbi);
255 if (submit_bio)
256 f2fs_submit_merged_bio(sbi, DATA, WRITE);
257 }
edb27dee 258 return err;
88b88a66
JK		 259}
260
0a8165d7 261/*
351df4b2
JK		 262 * This function balances dirty node and dentry pages.
263 * In addition, it controls garbage collection.
264 */
2c4db1a6 265void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 266{
2c4db1a6
JK		 267 if (!need)
268 return;
351df4b2 269 /*
029cd28c
JK		 270 * We should do GC or end up with checkpoint, if there are so many dirty
271 * dir/node pages without enough free segments.
351df4b2 272 */
43727527 273 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 274 mutex_lock(&sbi->gc_mutex);
d530d4d8 275 f2fs_gc(sbi, false);
351df4b2
JK		 276 }
277}
278
4660f9c0
JK		 279void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
280{
1dcc336b 281 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 282 if (!available_free_memory(sbi, EXTENT_CACHE))
283 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 284
1b38dc8e
JK		 285 /* check the # of cached NAT entries */
286 if (!available_free_memory(sbi, NAT_ENTRIES))
287 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
288
31696580
CY		 289 if (!available_free_memory(sbi, FREE_NIDS))
290 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
291
1b38dc8e
JK		 292 /* checkpoint is the only way to shrink partial cached entries */
293 if (!available_free_memory(sbi, NAT_ENTRIES) ||
e5e7ea3c 294 excess_prefree_segs(sbi) ||
60b99b48 295 !available_free_memory(sbi, INO_ENTRIES) ||
d0239e1b 296 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
36b35a0d
CY		 297 if (test_opt(sbi, DATA_FLUSH))
298 sync_dirty_inodes(sbi, FILE_INODE);
4660f9c0 299 f2fs_sync_fs(sbi->sb, true);
36b35a0d 300 }
4660f9c0
JK		 301}
302
2163d198 303static int issue_flush_thread(void *data)
6b4afdd7
JK		 304{
305 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 306 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
307 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 308repeat:
309 if (kthread_should_stop())
310 return 0;
311
721bd4d5 312 if (!llist_empty(&fcc->issue_list)) {
740432f8 313 struct bio *bio;
6b4afdd7
JK		 314 struct flush_cmd *cmd, *next;
315 int ret;
316
740432f8
JK		 317 bio = f2fs_bio_alloc(0);
318
721bd4d5
GZ		 319 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
320 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
321
6b4afdd7
JK		 322 bio->bi_bdev = sbi->sb->s_bdev;
323 ret = submit_bio_wait(WRITE_FLUSH, bio);
324
721bd4d5
GZ		 325 llist_for_each_entry_safe(cmd, next,
326 fcc->dispatch_list, llnode) {
6b4afdd7 327 cmd->ret = ret;
6b4afdd7
JK		 328 complete(&cmd->wait);
329 }
a4ed23f2 330 bio_put(bio);
a688b9d9 331 fcc->dispatch_list = NULL;
6b4afdd7
JK		 332 }
333
a688b9d9 334 wait_event_interruptible(*q,
721bd4d5 335 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 336 goto repeat;
337}
338
339int f2fs_issue_flush(struct f2fs_sb_info *sbi)
340{
a688b9d9 341 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 342 struct flush_cmd cmd;
6b4afdd7 343
24a9ee0f
JK		 344 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
345 test_opt(sbi, FLUSH_MERGE));
346
0f7b2abd
JK		 347 if (test_opt(sbi, NOBARRIER))
348 return 0;
349
740432f8
JK		 350 if (!test_opt(sbi, FLUSH_MERGE)) {
351 struct bio *bio = f2fs_bio_alloc(0);
352 int ret;
353
354 bio->bi_bdev = sbi->sb->s_bdev;
355 ret = submit_bio_wait(WRITE_FLUSH, bio);
356 bio_put(bio);
357 return ret;
358 }
6b4afdd7 359
adf8d90b 360 init_completion(&cmd.wait);
6b4afdd7 361
721bd4d5 362 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 363
a688b9d9
GZ		 364 if (!fcc->dispatch_list)
365 wake_up(&fcc->flush_wait_queue);
6b4afdd7 366
adf8d90b
CY		 367 wait_for_completion(&cmd.wait);
368
369 return cmd.ret;
6b4afdd7
JK		 370}
371
2163d198
GZ		 372int create_flush_cmd_control(struct f2fs_sb_info *sbi)
373{
374 dev_t dev = sbi->sb->s_bdev->bd_dev;
375 struct flush_cmd_control *fcc;
376 int err = 0;
377
378 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
379 if (!fcc)
380 return -ENOMEM;
2163d198 381 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 382 init_llist_head(&fcc->issue_list);
6b2920a5 383 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 384 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
385 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
386 if (IS_ERR(fcc->f2fs_issue_flush)) {
387 err = PTR_ERR(fcc->f2fs_issue_flush);
388 kfree(fcc);
6b2920a5 389 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 390 return err;
391 }
2163d198
GZ		 392
393 return err;
394}
395
396void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
397{
6b2920a5 398 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 399
400 if (fcc && fcc->f2fs_issue_flush)
401 kthread_stop(fcc->f2fs_issue_flush);
402 kfree(fcc);
6b2920a5 403 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 404}
405
351df4b2
JK		 406static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
407 enum dirty_type dirty_type)
408{
409 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
410
411 /* need not be added */
412 if (IS_CURSEG(sbi, segno))
413 return;
414
415 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
416 dirty_i->nr_dirty[dirty_type]++;
417
418 if (dirty_type == DIRTY) {
419 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 420 enum dirty_type t = sentry->type;
b2f2c390 421
ec325b52
JK		 422 if (unlikely(t >= DIRTY)) {
423 f2fs_bug_on(sbi, 1);
424 return;
425 }
4625d6aa
CL		 426 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
427 dirty_i->nr_dirty[t]++;
351df4b2
JK		 428 }
429}
430
431static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
432 enum dirty_type dirty_type)
433{
434 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
435
436 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
437 dirty_i->nr_dirty[dirty_type]--;
438
439 if (dirty_type == DIRTY) {
4625d6aa
CL		 440 struct seg_entry *sentry = get_seg_entry(sbi, segno);
441 enum dirty_type t = sentry->type;
442
443 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
444 dirty_i->nr_dirty[t]--;
b2f2c390 445
5ec4e49f
JK		 446 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
447 clear_bit(GET_SECNO(sbi, segno),
448 dirty_i->victim_secmap);
351df4b2
JK		 449 }
450}
451
0a8165d7 452/*
351df4b2
JK		 453 * Should not occur error such as -ENOMEM.
454 * Adding dirty entry into seglist is not critical operation.
455 * If a given segment is one of current working segments, it won't be added.
456 */
8d8451af 457static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 458{
459 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
460 unsigned short valid_blocks;
461
462 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
463 return;
464
465 mutex_lock(&dirty_i->seglist_lock);
466
467 valid_blocks = get_valid_blocks(sbi, segno, 0);
468
469 if (valid_blocks == 0) {
470 __locate_dirty_segment(sbi, segno, PRE);
471 __remove_dirty_segment(sbi, segno, DIRTY);
472 } else if (valid_blocks < sbi->blocks_per_seg) {
473 __locate_dirty_segment(sbi, segno, DIRTY);
474 } else {
475 /* Recovery routine with SSR needs this */
476 __remove_dirty_segment(sbi, segno, DIRTY);
477 }
478
479 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 480}
481
1e87a78d 482static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 483 block_t blkstart, block_t blklen)
484{
55cf9cb6
CY		 485 sector_t start = SECTOR_FROM_BLOCK(blkstart);
486 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 487 struct seg_entry *se;
488 unsigned int offset;
489 block_t i;
490
491 for (i = blkstart; i < blkstart + blklen; i++) {
492 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
493 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
494
495 if (!f2fs_test_and_set_bit(offset, se->discard_map))
496 sbi->discard_blks--;
497 }
1661d07c 498 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 499 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
500}
501
e90c2d28 502bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 503{
40a02be1
JK		 504 int err = -ENOTSUPP;
505
506 if (test_opt(sbi, DISCARD)) {
507 struct seg_entry *se = get_seg_entry(sbi,
508 GET_SEGNO(sbi, blkaddr));
509 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
510
511 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 512 return false;
40a02be1
JK		 513
514 err = f2fs_issue_discard(sbi, blkaddr, 1);
515 }
516
e90c2d28 517 if (err) {
381722d2 518 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 519 return true;
520 }
521 return false;
37208879
JK		 522}
523
adf4983b 524static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 525 struct cp_control *cpc, struct seg_entry *se,
526 unsigned int start, unsigned int end)
b2955550
JK		 527{
528 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 529 struct discard_entry *new, *last;
530
531 if (!list_empty(head)) {
532 last = list_last_entry(head, struct discard_entry, list);
533 if (START_BLOCK(sbi, cpc->trim_start) + start ==
534 last->blkaddr + last->len) {
535 last->len += end - start;
536 goto done;
537 }
538 }
539
540 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
541 INIT_LIST_HEAD(&new->list);
542 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
543 new->len = end - start;
544 list_add_tail(&new->list, head);
545done:
546 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 547}
548
549static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
550{
b2955550
JK		 551 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
552 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 553 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 554 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
555 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 556 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 557 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 558 unsigned int start = 0, end = -1;
4b2fecc8 559 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 560 int i;
561
a66cdd98 562 if (se->valid_blocks == max_blocks)
b2955550
JK		 563 return;
564
a66cdd98
JK		 565 if (!force) {
566 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 567 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
568 return;
4b2fecc8
JK		 569 }
570
b2955550
JK		 571 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
572 for (i = 0; i < entries; i++)
a66cdd98 573 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 574 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 575
4b2fecc8 576 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 577 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
578 if (start >= max_blocks)
579 break;
580
581 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 582 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 583 }
584}
585
4b2fecc8
JK		 586void release_discard_addrs(struct f2fs_sb_info *sbi)
587{
588 struct list_head *head = &(SM_I(sbi)->discard_list);
589 struct discard_entry *entry, *this;
590
591 /* drop caches */
592 list_for_each_entry_safe(entry, this, head, list) {
593 list_del(&entry->list);
594 kmem_cache_free(discard_entry_slab, entry);
595 }
596}
597
0a8165d7 598/*
351df4b2
JK		 599 * Should call clear_prefree_segments after checkpoint is done.
600 */
601static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
602{
603 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 604 unsigned int segno;
351df4b2
JK		 605
606 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 607 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 608 __set_test_and_free(sbi, segno);
351df4b2
JK		 609 mutex_unlock(&dirty_i->seglist_lock);
610}
611
836b5a63 612void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 613{
b2955550 614 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 615 struct discard_entry *entry, *this;
351df4b2 616 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 617 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 618 unsigned int start = 0, end = -1;
351df4b2
JK		 619
620 mutex_lock(&dirty_i->seglist_lock);
29e59c14 621
351df4b2 622 while (1) {
29e59c14 623 int i;
7cd8558b
JK		 624 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
625 if (start >= MAIN_SEGS(sbi))
351df4b2 626 break;
7cd8558b
JK		 627 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
628 start + 1);
29e59c14
CL		 629
630 for (i = start; i < end; i++)
631 clear_bit(i, prefree_map);
632
633 dirty_i->nr_dirty[PRE] -= end - start;
634
635 if (!test_opt(sbi, DISCARD))
636 continue;
351df4b2 637
37208879
JK		 638 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
639 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 640 }
641 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 642
643 /* send small discards */
2d7b822a 644 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 645 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
646 goto skip;
37208879 647 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 648 cpc->trimmed += entry->len;
836b5a63 649skip:
b2955550
JK		 650 list_del(&entry->list);
651 SM_I(sbi)->nr_discards -= entry->len;
652 kmem_cache_free(discard_entry_slab, entry);
653 }
351df4b2
JK		 654}
655
184a5cd2 656static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 657{
658 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 659
660 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 661 sit_i->dirty_sentries++;
184a5cd2
CY		 662 return false;
663 }
664
665 return true;
351df4b2
JK		 666}
667
668static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
669 unsigned int segno, int modified)
670{
671 struct seg_entry *se = get_seg_entry(sbi, segno);
672 se->type = type;
673 if (modified)
674 __mark_sit_entry_dirty(sbi, segno);
675}
676
677static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
678{
679 struct seg_entry *se;
680 unsigned int segno, offset;
681 long int new_vblocks;
682
683 segno = GET_SEGNO(sbi, blkaddr);
684
685 se = get_seg_entry(sbi, segno);
686 new_vblocks = se->valid_blocks + del;
491c0854 687 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 688
9850cf4a 689 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 690 (new_vblocks > sbi->blocks_per_seg)));
691
692 se->valid_blocks = new_vblocks;
693 se->mtime = get_mtime(sbi);
694 SIT_I(sbi)->max_mtime = se->mtime;
695
696 /* Update valid block bitmap */
697 if (del > 0) {
52aca074 698 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 699 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 700 if (!f2fs_test_and_set_bit(offset, se->discard_map))
701 sbi->discard_blks--;
351df4b2 702 } else {
52aca074 703 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 704 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 705 if (f2fs_test_and_clear_bit(offset, se->discard_map))
706 sbi->discard_blks++;
351df4b2
JK		 707 }
708 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
709 se->ckpt_valid_blocks += del;
710
711 __mark_sit_entry_dirty(sbi, segno);
712
713 /* update total number of valid blocks to be written in ckpt area */
714 SIT_I(sbi)->written_valid_blocks += del;
715
716 if (sbi->segs_per_sec > 1)
717 get_sec_entry(sbi, segno)->valid_blocks += del;
718}
719
5e443818 720void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 721{
5e443818
JK		 722 update_sit_entry(sbi, new, 1);
723 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
724 update_sit_entry(sbi, old, -1);
725
726 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
727 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 728}
729
730void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
731{
732 unsigned int segno = GET_SEGNO(sbi, addr);
733 struct sit_info *sit_i = SIT_I(sbi);
734
9850cf4a 735 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 736 if (addr == NEW_ADDR)
737 return;
738
739 /* add it into sit main buffer */
740 mutex_lock(&sit_i->sentry_lock);
741
742 update_sit_entry(sbi, addr, -1);
743
744 /* add it into dirty seglist */
745 locate_dirty_segment(sbi, segno);
746
747 mutex_unlock(&sit_i->sentry_lock);
748}
749
6e2c64ad
JK		 750bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
751{
752 struct sit_info *sit_i = SIT_I(sbi);
753 unsigned int segno, offset;
754 struct seg_entry *se;
755 bool is_cp = false;
756
757 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
758 return true;
759
760 mutex_lock(&sit_i->sentry_lock);
761
762 segno = GET_SEGNO(sbi, blkaddr);
763 se = get_seg_entry(sbi, segno);
764 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
765
766 if (f2fs_test_bit(offset, se->ckpt_valid_map))
767 is_cp = true;
768
769 mutex_unlock(&sit_i->sentry_lock);
770
771 return is_cp;
772}
773
0a8165d7 774/*
351df4b2
JK		 775 * This function should be resided under the curseg_mutex lock
776 */
777static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 778 struct f2fs_summary *sum)
351df4b2
JK		 779{
780 struct curseg_info *curseg = CURSEG_I(sbi, type);
781 void *addr = curseg->sum_blk;
e79efe3b 782 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 783 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 784}
785
0a8165d7 786/*
351df4b2
JK		 787 * Calculate the number of current summary pages for writing
788 */
3fa06d7b 789int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 790{
351df4b2 791 int valid_sum_count = 0;
9a47938b 792 int i, sum_in_page;
351df4b2
JK		 793
794 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
795 if (sbi->ckpt->alloc_type[i] == SSR)
796 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 797 else {
798 if (for_ra)
799 valid_sum_count += le16_to_cpu(
800 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
801 else
802 valid_sum_count += curseg_blkoff(sbi, i);
803 }
351df4b2
JK		 804 }
805
9a47938b
FL		 806 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
807 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
808 if (valid_sum_count <= sum_in_page)
351df4b2 809 return 1;
9a47938b
FL		 810 else if ((valid_sum_count - sum_in_page) <=
811 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 812 return 2;
813 return 3;
814}
815
0a8165d7 816/*
351df4b2
JK		 817 * Caller should put this summary page
818 */
819struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
820{
821 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
822}
823
381722d2 824void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 825{
826 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 827 void *dst = page_address(page);
828
829 if (src)
830 memcpy(dst, src, PAGE_CACHE_SIZE);
831 else
832 memset(dst, 0, PAGE_CACHE_SIZE);
351df4b2
JK		 833 set_page_dirty(page);
834 f2fs_put_page(page, 1);
835}
836
381722d2
CY		 837static void write_sum_page(struct f2fs_sb_info *sbi,
838 struct f2fs_summary_block *sum_blk, block_t blk_addr)
839{
840 update_meta_page(sbi, (void *)sum_blk, blk_addr);
841}
842
60374688
JK		 843static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
844{
845 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 846 unsigned int segno = curseg->segno + 1;
60374688
JK		 847 struct free_segmap_info *free_i = FREE_I(sbi);
848
7cd8558b 849 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 850 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 851 return 0;
852}
853
0a8165d7 854/*
351df4b2
JK		 855 * Find a new segment from the free segments bitmap to right order
856 * This function should be returned with success, otherwise BUG
857 */
858static void get_new_segment(struct f2fs_sb_info *sbi,
859 unsigned int *newseg, bool new_sec, int dir)
860{
861 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 862 unsigned int segno, secno, zoneno;
7cd8558b 863 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 864 unsigned int hint = *newseg / sbi->segs_per_sec;
865 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
866 unsigned int left_start = hint;
867 bool init = true;
868 int go_left = 0;
869 int i;
870
1a118ccf 871 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 872
873 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
874 segno = find_next_zero_bit(free_i->free_segmap,
7cd8558b 875 MAIN_SEGS(sbi), *newseg + 1);
33afa7fd
JK		 876 if (segno - *newseg < sbi->segs_per_sec -
877 (*newseg % sbi->segs_per_sec))
351df4b2
JK		 878 goto got_it;
879 }
880find_other_zone:
7cd8558b
JK		 881 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
882 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 883 if (dir == ALLOC_RIGHT) {
884 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 885 MAIN_SECS(sbi), 0);
886 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 887 } else {
888 go_left = 1;
889 left_start = hint - 1;
890 }
891 }
892 if (go_left == 0)
893 goto skip_left;
894
895 while (test_bit(left_start, free_i->free_secmap)) {
896 if (left_start > 0) {
897 left_start--;
898 continue;
899 }
900 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 901 MAIN_SECS(sbi), 0);
902 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 903 break;
904 }
905 secno = left_start;
906skip_left:
907 hint = secno;
908 segno = secno * sbi->segs_per_sec;
909 zoneno = secno / sbi->secs_per_zone;
910
911 /* give up on finding another zone */
912 if (!init)
913 goto got_it;
914 if (sbi->secs_per_zone == 1)
915 goto got_it;
916 if (zoneno == old_zoneno)
917 goto got_it;
918 if (dir == ALLOC_LEFT) {
919 if (!go_left && zoneno + 1 >= total_zones)
920 goto got_it;
921 if (go_left && zoneno == 0)
922 goto got_it;
923 }
924 for (i = 0; i < NR_CURSEG_TYPE; i++)
925 if (CURSEG_I(sbi, i)->zone == zoneno)
926 break;
927
928 if (i < NR_CURSEG_TYPE) {
929 /* zone is in user, try another */
930 if (go_left)
931 hint = zoneno * sbi->secs_per_zone - 1;
932 else if (zoneno + 1 >= total_zones)
933 hint = 0;
934 else
935 hint = (zoneno + 1) * sbi->secs_per_zone;
936 init = false;
937 goto find_other_zone;
938 }
939got_it:
940 /* set it as dirty segment in free segmap */
9850cf4a 941 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 942 __set_inuse(sbi, segno);
943 *newseg = segno;
1a118ccf 944 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 945}
946
947static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
948{
949 struct curseg_info *curseg = CURSEG_I(sbi, type);
950 struct summary_footer *sum_footer;
951
952 curseg->segno = curseg->next_segno;
953 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
954 curseg->next_blkoff = 0;
955 curseg->next_segno = NULL_SEGNO;
956
957 sum_footer = &(curseg->sum_blk->footer);
958 memset(sum_footer, 0, sizeof(struct summary_footer));
959 if (IS_DATASEG(type))
960 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
961 if (IS_NODESEG(type))
962 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
963 __set_sit_entry_type(sbi, type, curseg->segno, modified);
964}
965
0a8165d7 966/*
351df4b2
JK		 967 * Allocate a current working segment.
968 * This function always allocates a free segment in LFS manner.
969 */
970static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
971{
972 struct curseg_info *curseg = CURSEG_I(sbi, type);
973 unsigned int segno = curseg->segno;
974 int dir = ALLOC_LEFT;
975
976 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 977 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 978 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
979 dir = ALLOC_RIGHT;
980
981 if (test_opt(sbi, NOHEAP))
982 dir = ALLOC_RIGHT;
983
984 get_new_segment(sbi, &segno, new_sec, dir);
985 curseg->next_segno = segno;
986 reset_curseg(sbi, type, 1);
987 curseg->alloc_type = LFS;
988}
989
990static void __next_free_blkoff(struct f2fs_sb_info *sbi,
991 struct curseg_info *seg, block_t start)
992{
993 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 994 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 995 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 996 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
997 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
998 int i, pos;
999
1000 for (i = 0; i < entries; i++)
1001 target_map[i] = ckpt_map[i] | cur_map[i];
1002
1003 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1004
1005 seg->next_blkoff = pos;
351df4b2
JK		 1006}
1007
0a8165d7 1008/*
351df4b2
JK		 1009 * If a segment is written by LFS manner, next block offset is just obtained
1010 * by increasing the current block offset. However, if a segment is written by
1011 * SSR manner, next block offset obtained by calling __next_free_blkoff
1012 */
1013static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1014 struct curseg_info *seg)
1015{
1016 if (seg->alloc_type == SSR)
1017 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1018 else
1019 seg->next_blkoff++;
1020}
1021
0a8165d7 1022/*
e1c42045 1023 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1024 * manner, so it should recover the existing segment information of valid blocks
1025 */
1026static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1027{
1028 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1029 struct curseg_info *curseg = CURSEG_I(sbi, type);
1030 unsigned int new_segno = curseg->next_segno;
1031 struct f2fs_summary_block *sum_node;
1032 struct page *sum_page;
1033
1034 write_sum_page(sbi, curseg->sum_blk,
1035 GET_SUM_BLOCK(sbi, curseg->segno));
1036 __set_test_and_inuse(sbi, new_segno);
1037
1038 mutex_lock(&dirty_i->seglist_lock);
1039 __remove_dirty_segment(sbi, new_segno, PRE);
1040 __remove_dirty_segment(sbi, new_segno, DIRTY);
1041 mutex_unlock(&dirty_i->seglist_lock);
1042
1043 reset_curseg(sbi, type, 1);
1044 curseg->alloc_type = SSR;
1045 __next_free_blkoff(sbi, curseg, 0);
1046
1047 if (reuse) {
1048 sum_page = get_sum_page(sbi, new_segno);
1049 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1050 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1051 f2fs_put_page(sum_page, 1);
1052 }
1053}
1054
43727527
JK		 1055static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1056{
1057 struct curseg_info *curseg = CURSEG_I(sbi, type);
1058 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1059
1060 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1061 return v_ops->get_victim(sbi,
1062 &(curseg)->next_segno, BG_GC, type, SSR);
1063
1064 /* For data segments, let's do SSR more intensively */
1065 for (; type >= CURSEG_HOT_DATA; type--)
1066 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1067 BG_GC, type, SSR))
1068 return 1;
1069 return 0;
1070}
1071
351df4b2
JK		 1072/*
1073 * flush out current segment and replace it with new segment
1074 * This function should be returned with success, otherwise BUG
1075 */
1076static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1077 int type, bool force)
1078{
1079 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1080
7b405275 1081 if (force)
351df4b2 1082 new_curseg(sbi, type, true);
7b405275 1083 else if (type == CURSEG_WARM_NODE)
351df4b2 1084 new_curseg(sbi, type, false);
60374688
JK		 1085 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1086 new_curseg(sbi, type, false);
351df4b2
JK		 1087 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1088 change_curseg(sbi, type, true);
1089 else
1090 new_curseg(sbi, type, false);
dcdfff65
JK		 1091
1092 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1093}
1094
38aa0889
JK		 1095static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1096{
1097 struct curseg_info *curseg = CURSEG_I(sbi, type);
1098 unsigned int old_segno;
1099
1100 old_segno = curseg->segno;
1101 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1102 locate_dirty_segment(sbi, old_segno);
1103}
1104
351df4b2
JK		 1105void allocate_new_segments(struct f2fs_sb_info *sbi)
1106{
351df4b2
JK		 1107 int i;
1108
38aa0889
JK		 1109 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1110 __allocate_new_segments(sbi, i);
351df4b2
JK		 1111}
1112
1113static const struct segment_allocation default_salloc_ops = {
1114 .allocate_segment = allocate_segment_by_default,
1115};
1116
4b2fecc8
JK		 1117int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1118{
f7ef9b83
JK		 1119 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1120 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1121 unsigned int start_segno, end_segno;
1122 struct cp_control cpc;
c34f42e2 1123 int err = 0;
4b2fecc8 1124
836b5a63 1125 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1126 return -EINVAL;
1127
9bd27ae4 1128 cpc.trimmed = 0;
7cd8558b 1129 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1130 goto out;
1131
1132 /* start/end segment number in main_area */
7cd8558b
JK		 1133 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1134 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1135 GET_SEGNO(sbi, end);
4b2fecc8 1136 cpc.reason = CP_DISCARD;
836b5a63 1137 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1138
1139 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1140 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1141 cpc.trim_start = start_segno;
a66cdd98
JK		 1142
1143 if (sbi->discard_blks == 0)
1144 break;
1145 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1146 cpc.trim_end = end_segno;
1147 else
1148 cpc.trim_end = min_t(unsigned int,
1149 rounddown(start_segno +
bba681cb
JK		 1150 BATCHED_TRIM_SEGMENTS(sbi),
1151 sbi->segs_per_sec) - 1, end_segno);
1152
1153 mutex_lock(&sbi->gc_mutex);
c34f42e2 1154 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1155 mutex_unlock(&sbi->gc_mutex);
1156 }
4b2fecc8 1157out:
f7ef9b83 1158 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1159 return err;
4b2fecc8
JK		 1160}
1161
351df4b2
JK		 1162static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1163{
1164 struct curseg_info *curseg = CURSEG_I(sbi, type);
1165 if (curseg->next_blkoff < sbi->blocks_per_seg)
1166 return true;
1167 return false;
1168}
1169
1170static int __get_segment_type_2(struct page *page, enum page_type p_type)
1171{
1172 if (p_type == DATA)
1173 return CURSEG_HOT_DATA;
1174 else
1175 return CURSEG_HOT_NODE;
1176}
1177
1178static int __get_segment_type_4(struct page *page, enum page_type p_type)
1179{
1180 if (p_type == DATA) {
1181 struct inode *inode = page->mapping->host;
1182
1183 if (S_ISDIR(inode->i_mode))
1184 return CURSEG_HOT_DATA;
1185 else
1186 return CURSEG_COLD_DATA;
1187 } else {
a344b9fd
JK		 1188 if (IS_DNODE(page) && is_cold_node(page))
1189 return CURSEG_WARM_NODE;
351df4b2
JK		 1190 else
1191 return CURSEG_COLD_NODE;
1192 }
1193}
1194
1195static int __get_segment_type_6(struct page *page, enum page_type p_type)
1196{
1197 if (p_type == DATA) {
1198 struct inode *inode = page->mapping->host;
1199
1200 if (S_ISDIR(inode->i_mode))
1201 return CURSEG_HOT_DATA;
354a3399 1202 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1203 return CURSEG_COLD_DATA;
1204 else
1205 return CURSEG_WARM_DATA;
1206 } else {
1207 if (IS_DNODE(page))
1208 return is_cold_node(page) ? CURSEG_WARM_NODE :
1209 CURSEG_HOT_NODE;
1210 else
1211 return CURSEG_COLD_NODE;
1212 }
1213}
1214
1215static int __get_segment_type(struct page *page, enum page_type p_type)
1216{
4081363f 1217 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1218 case 2:
1219 return __get_segment_type_2(page, p_type);
1220 case 4:
1221 return __get_segment_type_4(page, p_type);
351df4b2 1222 }
12a67146 1223 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1224 f2fs_bug_on(F2FS_P_SB(page),
1225 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1226 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1227}
1228
bfad7c2d
JK		 1229void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1230 block_t old_blkaddr, block_t *new_blkaddr,
1231 struct f2fs_summary *sum, int type)
351df4b2
JK		 1232{
1233 struct sit_info *sit_i = SIT_I(sbi);
1234 struct curseg_info *curseg;
38aa0889
JK		 1235 bool direct_io = (type == CURSEG_DIRECT_IO);
1236
1237 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1238
351df4b2
JK		 1239 curseg = CURSEG_I(sbi, type);
1240
1241 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1242 mutex_lock(&sit_i->sentry_lock);
351df4b2 1243
38aa0889 1244 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1245 if (direct_io && curseg->next_blkoff &&
1246 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1247 __allocate_new_segments(sbi, type);
1248
351df4b2 1249 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1250
1251 /*
1252 * __add_sum_entry should be resided under the curseg_mutex
1253 * because, this function updates a summary entry in the
1254 * current summary block.
1255 */
e79efe3b 1256 __add_sum_entry(sbi, type, sum);
351df4b2 1257
351df4b2 1258 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1259
1260 stat_inc_block_count(sbi, curseg);
351df4b2 1261
5e443818
JK		 1262 if (!__has_curseg_space(sbi, type))
1263 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1264 /*
1265 * SIT information should be updated before segment allocation,
1266 * since SSR needs latest valid block information.
1267 */
1268 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1269
351df4b2
JK		 1270 mutex_unlock(&sit_i->sentry_lock);
1271
bfad7c2d 1272 if (page && IS_NODESEG(type))
351df4b2
JK		 1273 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1274
bfad7c2d
JK		 1275 mutex_unlock(&curseg->curseg_mutex);
1276}
1277
05ca3632 1278static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1279{
05ca3632 1280 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1281
05ca3632
JK		 1282 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1283 &fio->blk_addr, sum, type);
bfad7c2d 1284
351df4b2 1285 /* writeout dirty page into bdev */
05ca3632 1286 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1287}
1288
577e3495 1289void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1290{
458e6197 1291 struct f2fs_io_info fio = {
05ca3632 1292 .sbi = sbi,
458e6197 1293 .type = META,
cf04e8eb
JK		 1294 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1295 .blk_addr = page->index,
05ca3632 1296 .page = page,
4375a336 1297 .encrypted_page = NULL,
458e6197
JK		 1298 };
1299
2b947003
CY		 1300 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1301 fio.rw &= ~REQ_META;
1302
351df4b2 1303 set_page_writeback(page);
05ca3632 1304 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1305}
1306
05ca3632 1307void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1308{
1309 struct f2fs_summary sum;
05ca3632 1310
351df4b2 1311 set_summary(&sum, nid, 0, 0);
05ca3632 1312 do_write_page(&sum, fio);
351df4b2
JK		 1313}
1314
05ca3632 1315void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1316{
05ca3632 1317 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1318 struct f2fs_summary sum;
1319 struct node_info ni;
1320
9850cf4a 1321 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1322 get_node_info(sbi, dn->nid, &ni);
1323 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1324 do_write_page(&sum, fio);
e1509cf2 1325 dn->data_blkaddr = fio->blk_addr;
351df4b2
JK		 1326}
1327
05ca3632 1328void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1329{
05ca3632
JK		 1330 stat_inc_inplace_blocks(fio->sbi);
1331 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1332}
1333
528e3459
CY		 1334static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1335 struct f2fs_summary *sum,
19f106bc
CY		 1336 block_t old_blkaddr, block_t new_blkaddr,
1337 bool recover_curseg)
351df4b2
JK		 1338{
1339 struct sit_info *sit_i = SIT_I(sbi);
1340 struct curseg_info *curseg;
1341 unsigned int segno, old_cursegno;
1342 struct seg_entry *se;
1343 int type;
19f106bc 1344 unsigned short old_blkoff;
351df4b2
JK		 1345
1346 segno = GET_SEGNO(sbi, new_blkaddr);
1347 se = get_seg_entry(sbi, segno);
1348 type = se->type;
1349
19f106bc
CY		 1350 if (!recover_curseg) {
1351 /* for recovery flow */
1352 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1353 if (old_blkaddr == NULL_ADDR)
1354 type = CURSEG_COLD_DATA;
1355 else
1356 type = CURSEG_WARM_DATA;
1357 }
1358 } else {
1359 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1360 type = CURSEG_WARM_DATA;
1361 }
19f106bc 1362
351df4b2
JK		 1363 curseg = CURSEG_I(sbi, type);
1364
1365 mutex_lock(&curseg->curseg_mutex);
1366 mutex_lock(&sit_i->sentry_lock);
1367
1368 old_cursegno = curseg->segno;
19f106bc 1369 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1370
1371 /* change the current segment */
1372 if (segno != curseg->segno) {
1373 curseg->next_segno = segno;
1374 change_curseg(sbi, type, true);
1375 }
1376
491c0854 1377 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1378 __add_sum_entry(sbi, type, sum);
351df4b2 1379
6e2c64ad
JK		 1380 if (!recover_curseg)
1381 update_sit_entry(sbi, new_blkaddr, 1);
1382 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1383 update_sit_entry(sbi, old_blkaddr, -1);
1384
1385 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1386 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1387
351df4b2 1388 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1389
19f106bc
CY		 1390 if (recover_curseg) {
1391 if (old_cursegno != curseg->segno) {
1392 curseg->next_segno = old_cursegno;
1393 change_curseg(sbi, type, true);
1394 }
1395 curseg->next_blkoff = old_blkoff;
1396 }
1397
351df4b2
JK		 1398 mutex_unlock(&sit_i->sentry_lock);
1399 mutex_unlock(&curseg->curseg_mutex);
1400}
1401
528e3459
CY		 1402void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1403 block_t old_addr, block_t new_addr,
1404 unsigned char version, bool recover_curseg)
1405{
1406 struct f2fs_summary sum;
1407
1408 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1409
1410 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1411
1412 dn->data_blkaddr = new_addr;
1413 set_data_blkaddr(dn);
1414 f2fs_update_extent_cache(dn);
1415}
1416
df0f8dc0
CY		 1417static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1418 struct page *page, enum page_type type)
1419{
1420 enum page_type btype = PAGE_TYPE_OF_BIO(type);
1421 struct f2fs_bio_info *io = &sbi->write_io[btype];
df0f8dc0 1422 struct bio_vec *bvec;
4375a336 1423 struct page *target;
df0f8dc0
CY		 1424 int i;
1425
1426 down_read(&io->io_rwsem);
4375a336
JK		 1427 if (!io->bio) {
1428 up_read(&io->io_rwsem);
1429 return false;
1430 }
df0f8dc0 1431
ce23447f 1432 bio_for_each_segment_all(bvec, io->bio, i) {
4375a336
JK		 1433
1434 if (bvec->bv_page->mapping) {
1435 target = bvec->bv_page;
1436 } else {
1437 struct f2fs_crypto_ctx *ctx;
1438
1439 /* encrypted page */
1440 ctx = (struct f2fs_crypto_ctx *)page_private(
1441 bvec->bv_page);
ca40b030 1442 target = ctx->w.control_page;
4375a336
JK		 1443 }
1444
1445 if (page == target) {
df0f8dc0
CY		 1446 up_read(&io->io_rwsem);
1447 return true;
1448 }
1449 }
1450
df0f8dc0
CY		 1451 up_read(&io->io_rwsem);
1452 return false;
1453}
1454
93dfe2ac 1455void f2fs_wait_on_page_writeback(struct page *page,
5514f0aa 1456 enum page_type type)
93dfe2ac 1457{
93dfe2ac 1458 if (PageWriteback(page)) {
4081363f
JK		 1459 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1460
df0f8dc0
CY		 1461 if (is_merged_page(sbi, page, type))
1462 f2fs_submit_merged_bio(sbi, type, WRITE);
93dfe2ac
JK		 1463 wait_on_page_writeback(page);
1464 }
1465}
1466
08b39fbd
CY		 1467void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1468 block_t blkaddr)
1469{
1470 struct page *cpage;
1471
1472 if (blkaddr == NEW_ADDR)
1473 return;
1474
1475 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1476
1477 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1478 if (cpage) {
1479 f2fs_wait_on_page_writeback(cpage, DATA);
1480 f2fs_put_page(cpage, 1);
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,
26879fb1 1621 META_CP, true);
3fa06d7b 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 1630 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1631 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 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 }
855639de 1739 if (alloc && __has_cursum_space(sum, 1, NAT_JOURNAL))
351df4b2
JK		 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;
855639de 1745 if (alloc && __has_cursum_space(sum, 1, SIT_JOURNAL))
351df4b2
JK		 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 {
26879fb1 2118 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
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 0.246954 seconds and 5 git commands to generate.