projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: avoid garbage lenghs in dentries
[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);
fec1d657
JK		 226 f2fs_wait_on_page_writeback(cur->page, DATA,
227 true);
70c640b1
JK		 228 if (clear_page_dirty_for_io(cur->page))
229 inode_dec_dirty_pages(inode);
8ce67cb0 230 trace_f2fs_commit_inmem_page(cur->page, INMEM);
05ca3632 231 fio.page = cur->page;
edb27dee 232 err = do_write_data_page(&fio);
edb27dee
JK		 233 if (err) {
234 unlock_page(cur->page);
235 break;
236 }
7fee7406 237 clear_cold_data(cur->page);
2b246fb0 238 submit_bio = true;
70c640b1 239 }
70c640b1 240 } else {
f478f43f 241 ClearPageUptodate(cur->page);
8ce67cb0 242 trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
88b88a66 243 }
decd36b6
CY		 244 set_page_private(cur->page, 0);
245 ClearPagePrivate(cur->page);
246 f2fs_put_page(cur->page, 1);
247
88b88a66
JK		 248 list_del(&cur->list);
249 kmem_cache_free(inmem_entry_slab, cur);
8dcf2ff7 250 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 251 }
88b88a66
JK		 252 mutex_unlock(&fi->inmem_lock);
253
70c640b1
JK		 254 if (!abort) {
255 f2fs_unlock_op(sbi);
256 if (submit_bio)
0c3a5797
CY		 257 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0,
258 DATA, WRITE);
70c640b1 259 }
edb27dee 260 return err;
88b88a66
JK		 261}
262
0a8165d7 263/*
351df4b2
JK		 264 * This function balances dirty node and dentry pages.
265 * In addition, it controls garbage collection.
266 */
2c4db1a6 267void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 268{
2c4db1a6
JK		 269 if (!need)
270 return;
351df4b2 271 /*
029cd28c
JK		 272 * We should do GC or end up with checkpoint, if there are so many dirty
273 * dir/node pages without enough free segments.
351df4b2 274 */
43727527 275 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 276 mutex_lock(&sbi->gc_mutex);
d530d4d8 277 f2fs_gc(sbi, false);
351df4b2
JK		 278 }
279}
280
4660f9c0
JK		 281void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
282{
1dcc336b 283 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 284 if (!available_free_memory(sbi, EXTENT_CACHE))
285 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 286
1b38dc8e
JK		 287 /* check the # of cached NAT entries */
288 if (!available_free_memory(sbi, NAT_ENTRIES))
289 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
290
31696580
CY		 291 if (!available_free_memory(sbi, FREE_NIDS))
292 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
293
1b38dc8e
JK		 294 /* checkpoint is the only way to shrink partial cached entries */
295 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 296 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 297 excess_prefree_segs(sbi) ||
298 excess_dirty_nats(sbi) ||
d0239e1b 299 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
36b35a0d
CY		 300 if (test_opt(sbi, DATA_FLUSH))
301 sync_dirty_inodes(sbi, FILE_INODE);
4660f9c0 302 f2fs_sync_fs(sbi->sb, true);
42190d2a 303 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 304 }
4660f9c0
JK		 305}
306
2163d198 307static int issue_flush_thread(void *data)
6b4afdd7
JK		 308{
309 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 310 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
311 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 312repeat:
313 if (kthread_should_stop())
314 return 0;
315
721bd4d5 316 if (!llist_empty(&fcc->issue_list)) {
740432f8 317 struct bio *bio;
6b4afdd7
JK		 318 struct flush_cmd *cmd, *next;
319 int ret;
320
740432f8
JK		 321 bio = f2fs_bio_alloc(0);
322
721bd4d5
GZ		 323 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
324 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
325
6b4afdd7
JK		 326 bio->bi_bdev = sbi->sb->s_bdev;
327 ret = submit_bio_wait(WRITE_FLUSH, bio);
328
721bd4d5
GZ		 329 llist_for_each_entry_safe(cmd, next,
330 fcc->dispatch_list, llnode) {
6b4afdd7 331 cmd->ret = ret;
6b4afdd7
JK		 332 complete(&cmd->wait);
333 }
a4ed23f2 334 bio_put(bio);
a688b9d9 335 fcc->dispatch_list = NULL;
6b4afdd7
JK		 336 }
337
a688b9d9 338 wait_event_interruptible(*q,
721bd4d5 339 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 340 goto repeat;
341}
342
343int f2fs_issue_flush(struct f2fs_sb_info *sbi)
344{
a688b9d9 345 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 346 struct flush_cmd cmd;
6b4afdd7 347
24a9ee0f
JK		 348 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
349 test_opt(sbi, FLUSH_MERGE));
350
0f7b2abd
JK		 351 if (test_opt(sbi, NOBARRIER))
352 return 0;
353
740432f8
JK		 354 if (!test_opt(sbi, FLUSH_MERGE)) {
355 struct bio *bio = f2fs_bio_alloc(0);
356 int ret;
357
358 bio->bi_bdev = sbi->sb->s_bdev;
359 ret = submit_bio_wait(WRITE_FLUSH, bio);
360 bio_put(bio);
361 return ret;
362 }
6b4afdd7 363
adf8d90b 364 init_completion(&cmd.wait);
6b4afdd7 365
721bd4d5 366 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 367
a688b9d9
GZ		 368 if (!fcc->dispatch_list)
369 wake_up(&fcc->flush_wait_queue);
6b4afdd7 370
adf8d90b
CY		 371 wait_for_completion(&cmd.wait);
372
373 return cmd.ret;
6b4afdd7
JK		 374}
375
2163d198
GZ		 376int create_flush_cmd_control(struct f2fs_sb_info *sbi)
377{
378 dev_t dev = sbi->sb->s_bdev->bd_dev;
379 struct flush_cmd_control *fcc;
380 int err = 0;
381
382 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
383 if (!fcc)
384 return -ENOMEM;
2163d198 385 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 386 init_llist_head(&fcc->issue_list);
6b2920a5 387 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 388 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
389 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
390 if (IS_ERR(fcc->f2fs_issue_flush)) {
391 err = PTR_ERR(fcc->f2fs_issue_flush);
392 kfree(fcc);
6b2920a5 393 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 394 return err;
395 }
2163d198
GZ		 396
397 return err;
398}
399
400void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
401{
6b2920a5 402 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 403
404 if (fcc && fcc->f2fs_issue_flush)
405 kthread_stop(fcc->f2fs_issue_flush);
406 kfree(fcc);
6b2920a5 407 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 408}
409
351df4b2
JK		 410static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
411 enum dirty_type dirty_type)
412{
413 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
414
415 /* need not be added */
416 if (IS_CURSEG(sbi, segno))
417 return;
418
419 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
420 dirty_i->nr_dirty[dirty_type]++;
421
422 if (dirty_type == DIRTY) {
423 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 424 enum dirty_type t = sentry->type;
b2f2c390 425
ec325b52
JK		 426 if (unlikely(t >= DIRTY)) {
427 f2fs_bug_on(sbi, 1);
428 return;
429 }
4625d6aa
CL		 430 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
431 dirty_i->nr_dirty[t]++;
351df4b2
JK		 432 }
433}
434
435static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
436 enum dirty_type dirty_type)
437{
438 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
439
440 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
441 dirty_i->nr_dirty[dirty_type]--;
442
443 if (dirty_type == DIRTY) {
4625d6aa
CL		 444 struct seg_entry *sentry = get_seg_entry(sbi, segno);
445 enum dirty_type t = sentry->type;
446
447 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
448 dirty_i->nr_dirty[t]--;
b2f2c390 449
5ec4e49f
JK		 450 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
451 clear_bit(GET_SECNO(sbi, segno),
452 dirty_i->victim_secmap);
351df4b2
JK		 453 }
454}
455
0a8165d7 456/*
351df4b2
JK		 457 * Should not occur error such as -ENOMEM.
458 * Adding dirty entry into seglist is not critical operation.
459 * If a given segment is one of current working segments, it won't be added.
460 */
8d8451af 461static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 462{
463 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
464 unsigned short valid_blocks;
465
466 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
467 return;
468
469 mutex_lock(&dirty_i->seglist_lock);
470
471 valid_blocks = get_valid_blocks(sbi, segno, 0);
472
473 if (valid_blocks == 0) {
474 __locate_dirty_segment(sbi, segno, PRE);
475 __remove_dirty_segment(sbi, segno, DIRTY);
476 } else if (valid_blocks < sbi->blocks_per_seg) {
477 __locate_dirty_segment(sbi, segno, DIRTY);
478 } else {
479 /* Recovery routine with SSR needs this */
480 __remove_dirty_segment(sbi, segno, DIRTY);
481 }
482
483 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 484}
485
1e87a78d 486static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 487 block_t blkstart, block_t blklen)
488{
55cf9cb6
CY		 489 sector_t start = SECTOR_FROM_BLOCK(blkstart);
490 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 491 struct seg_entry *se;
492 unsigned int offset;
493 block_t i;
494
495 for (i = blkstart; i < blkstart + blklen; i++) {
496 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
497 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
498
499 if (!f2fs_test_and_set_bit(offset, se->discard_map))
500 sbi->discard_blks--;
501 }
1661d07c 502 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 503 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
504}
505
e90c2d28 506bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 507{
60b286c4 508 int err = -EOPNOTSUPP;
40a02be1
JK		 509
510 if (test_opt(sbi, DISCARD)) {
511 struct seg_entry *se = get_seg_entry(sbi,
512 GET_SEGNO(sbi, blkaddr));
513 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
514
515 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 516 return false;
40a02be1
JK		 517
518 err = f2fs_issue_discard(sbi, blkaddr, 1);
519 }
520
e90c2d28 521 if (err) {
381722d2 522 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 523 return true;
524 }
525 return false;
37208879
JK		 526}
527
adf4983b 528static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 529 struct cp_control *cpc, struct seg_entry *se,
530 unsigned int start, unsigned int end)
b2955550
JK		 531{
532 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 533 struct discard_entry *new, *last;
534
535 if (!list_empty(head)) {
536 last = list_last_entry(head, struct discard_entry, list);
537 if (START_BLOCK(sbi, cpc->trim_start) + start ==
538 last->blkaddr + last->len) {
539 last->len += end - start;
540 goto done;
541 }
542 }
543
544 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
545 INIT_LIST_HEAD(&new->list);
546 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
547 new->len = end - start;
548 list_add_tail(&new->list, head);
549done:
550 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 551}
552
553static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
554{
b2955550
JK		 555 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
556 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 557 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 558 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
559 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 560 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 561 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 562 unsigned int start = 0, end = -1;
4b2fecc8 563 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 564 int i;
565
a66cdd98 566 if (se->valid_blocks == max_blocks)
b2955550
JK		 567 return;
568
a66cdd98
JK		 569 if (!force) {
570 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 571 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
572 return;
4b2fecc8
JK		 573 }
574
b2955550
JK		 575 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
576 for (i = 0; i < entries; i++)
a66cdd98 577 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 578 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 579
4b2fecc8 580 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 581 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
582 if (start >= max_blocks)
583 break;
584
585 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 586 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 587 }
588}
589
4b2fecc8
JK		 590void release_discard_addrs(struct f2fs_sb_info *sbi)
591{
592 struct list_head *head = &(SM_I(sbi)->discard_list);
593 struct discard_entry *entry, *this;
594
595 /* drop caches */
596 list_for_each_entry_safe(entry, this, head, list) {
597 list_del(&entry->list);
598 kmem_cache_free(discard_entry_slab, entry);
599 }
600}
601
0a8165d7 602/*
351df4b2
JK		 603 * Should call clear_prefree_segments after checkpoint is done.
604 */
605static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
606{
607 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 608 unsigned int segno;
351df4b2
JK		 609
610 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 611 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 612 __set_test_and_free(sbi, segno);
351df4b2
JK		 613 mutex_unlock(&dirty_i->seglist_lock);
614}
615
836b5a63 616void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 617{
b2955550 618 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 619 struct discard_entry *entry, *this;
351df4b2 620 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 621 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 622 unsigned int start = 0, end = -1;
351df4b2
JK		 623
624 mutex_lock(&dirty_i->seglist_lock);
29e59c14 625
351df4b2 626 while (1) {
29e59c14 627 int i;
7cd8558b
JK		 628 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
629 if (start >= MAIN_SEGS(sbi))
351df4b2 630 break;
7cd8558b
JK		 631 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
632 start + 1);
29e59c14
CL		 633
634 for (i = start; i < end; i++)
635 clear_bit(i, prefree_map);
636
637 dirty_i->nr_dirty[PRE] -= end - start;
638
639 if (!test_opt(sbi, DISCARD))
640 continue;
351df4b2 641
37208879
JK		 642 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
643 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 644 }
645 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 646
647 /* send small discards */
2d7b822a 648 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 649 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
650 goto skip;
37208879 651 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 652 cpc->trimmed += entry->len;
836b5a63 653skip:
b2955550
JK		 654 list_del(&entry->list);
655 SM_I(sbi)->nr_discards -= entry->len;
656 kmem_cache_free(discard_entry_slab, entry);
657 }
351df4b2
JK		 658}
659
184a5cd2 660static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 661{
662 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 663
664 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 665 sit_i->dirty_sentries++;
184a5cd2
CY		 666 return false;
667 }
668
669 return true;
351df4b2
JK		 670}
671
672static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
673 unsigned int segno, int modified)
674{
675 struct seg_entry *se = get_seg_entry(sbi, segno);
676 se->type = type;
677 if (modified)
678 __mark_sit_entry_dirty(sbi, segno);
679}
680
681static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
682{
683 struct seg_entry *se;
684 unsigned int segno, offset;
685 long int new_vblocks;
686
687 segno = GET_SEGNO(sbi, blkaddr);
688
689 se = get_seg_entry(sbi, segno);
690 new_vblocks = se->valid_blocks + del;
491c0854 691 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 692
9850cf4a 693 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 694 (new_vblocks > sbi->blocks_per_seg)));
695
696 se->valid_blocks = new_vblocks;
697 se->mtime = get_mtime(sbi);
698 SIT_I(sbi)->max_mtime = se->mtime;
699
700 /* Update valid block bitmap */
701 if (del > 0) {
52aca074 702 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 703 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 704 if (!f2fs_test_and_set_bit(offset, se->discard_map))
705 sbi->discard_blks--;
351df4b2 706 } else {
52aca074 707 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 708 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 709 if (f2fs_test_and_clear_bit(offset, se->discard_map))
710 sbi->discard_blks++;
351df4b2
JK		 711 }
712 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
713 se->ckpt_valid_blocks += del;
714
715 __mark_sit_entry_dirty(sbi, segno);
716
717 /* update total number of valid blocks to be written in ckpt area */
718 SIT_I(sbi)->written_valid_blocks += del;
719
720 if (sbi->segs_per_sec > 1)
721 get_sec_entry(sbi, segno)->valid_blocks += del;
722}
723
5e443818 724void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 725{
5e443818
JK		 726 update_sit_entry(sbi, new, 1);
727 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
728 update_sit_entry(sbi, old, -1);
729
730 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
731 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 732}
733
734void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
735{
736 unsigned int segno = GET_SEGNO(sbi, addr);
737 struct sit_info *sit_i = SIT_I(sbi);
738
9850cf4a 739 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 740 if (addr == NEW_ADDR)
741 return;
742
743 /* add it into sit main buffer */
744 mutex_lock(&sit_i->sentry_lock);
745
746 update_sit_entry(sbi, addr, -1);
747
748 /* add it into dirty seglist */
749 locate_dirty_segment(sbi, segno);
750
751 mutex_unlock(&sit_i->sentry_lock);
752}
753
6e2c64ad
JK		 754bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
755{
756 struct sit_info *sit_i = SIT_I(sbi);
757 unsigned int segno, offset;
758 struct seg_entry *se;
759 bool is_cp = false;
760
761 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
762 return true;
763
764 mutex_lock(&sit_i->sentry_lock);
765
766 segno = GET_SEGNO(sbi, blkaddr);
767 se = get_seg_entry(sbi, segno);
768 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
769
770 if (f2fs_test_bit(offset, se->ckpt_valid_map))
771 is_cp = true;
772
773 mutex_unlock(&sit_i->sentry_lock);
774
775 return is_cp;
776}
777
0a8165d7 778/*
351df4b2
JK		 779 * This function should be resided under the curseg_mutex lock
780 */
781static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 782 struct f2fs_summary *sum)
351df4b2
JK		 783{
784 struct curseg_info *curseg = CURSEG_I(sbi, type);
785 void *addr = curseg->sum_blk;
e79efe3b 786 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 787 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 788}
789
0a8165d7 790/*
351df4b2
JK		 791 * Calculate the number of current summary pages for writing
792 */
3fa06d7b 793int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 794{
351df4b2 795 int valid_sum_count = 0;
9a47938b 796 int i, sum_in_page;
351df4b2
JK		 797
798 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
799 if (sbi->ckpt->alloc_type[i] == SSR)
800 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 801 else {
802 if (for_ra)
803 valid_sum_count += le16_to_cpu(
804 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
805 else
806 valid_sum_count += curseg_blkoff(sbi, i);
807 }
351df4b2
JK		 808 }
809
9a47938b
FL		 810 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
811 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
812 if (valid_sum_count <= sum_in_page)
351df4b2 813 return 1;
9a47938b
FL		 814 else if ((valid_sum_count - sum_in_page) <=
815 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 816 return 2;
817 return 3;
818}
819
0a8165d7 820/*
351df4b2
JK		 821 * Caller should put this summary page
822 */
823struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
824{
825 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
826}
827
381722d2 828void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 829{
830 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 831 void *dst = page_address(page);
832
833 if (src)
834 memcpy(dst, src, PAGE_CACHE_SIZE);
835 else
836 memset(dst, 0, PAGE_CACHE_SIZE);
351df4b2
JK		 837 set_page_dirty(page);
838 f2fs_put_page(page, 1);
839}
840
381722d2
CY		 841static void write_sum_page(struct f2fs_sb_info *sbi,
842 struct f2fs_summary_block *sum_blk, block_t blk_addr)
843{
844 update_meta_page(sbi, (void *)sum_blk, blk_addr);
845}
846
60374688
JK		 847static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
848{
849 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 850 unsigned int segno = curseg->segno + 1;
60374688
JK		 851 struct free_segmap_info *free_i = FREE_I(sbi);
852
7cd8558b 853 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 854 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 855 return 0;
856}
857
0a8165d7 858/*
351df4b2
JK		 859 * Find a new segment from the free segments bitmap to right order
860 * This function should be returned with success, otherwise BUG
861 */
862static void get_new_segment(struct f2fs_sb_info *sbi,
863 unsigned int *newseg, bool new_sec, int dir)
864{
865 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 866 unsigned int segno, secno, zoneno;
7cd8558b 867 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 868 unsigned int hint = *newseg / sbi->segs_per_sec;
869 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
870 unsigned int left_start = hint;
871 bool init = true;
872 int go_left = 0;
873 int i;
874
1a118ccf 875 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 876
877 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
878 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 879 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
880 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 881 goto got_it;
882 }
883find_other_zone:
7cd8558b
JK		 884 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
885 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 886 if (dir == ALLOC_RIGHT) {
887 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 888 MAIN_SECS(sbi), 0);
889 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 890 } else {
891 go_left = 1;
892 left_start = hint - 1;
893 }
894 }
895 if (go_left == 0)
896 goto skip_left;
897
898 while (test_bit(left_start, free_i->free_secmap)) {
899 if (left_start > 0) {
900 left_start--;
901 continue;
902 }
903 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 904 MAIN_SECS(sbi), 0);
905 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 906 break;
907 }
908 secno = left_start;
909skip_left:
910 hint = secno;
911 segno = secno * sbi->segs_per_sec;
912 zoneno = secno / sbi->secs_per_zone;
913
914 /* give up on finding another zone */
915 if (!init)
916 goto got_it;
917 if (sbi->secs_per_zone == 1)
918 goto got_it;
919 if (zoneno == old_zoneno)
920 goto got_it;
921 if (dir == ALLOC_LEFT) {
922 if (!go_left && zoneno + 1 >= total_zones)
923 goto got_it;
924 if (go_left && zoneno == 0)
925 goto got_it;
926 }
927 for (i = 0; i < NR_CURSEG_TYPE; i++)
928 if (CURSEG_I(sbi, i)->zone == zoneno)
929 break;
930
931 if (i < NR_CURSEG_TYPE) {
932 /* zone is in user, try another */
933 if (go_left)
934 hint = zoneno * sbi->secs_per_zone - 1;
935 else if (zoneno + 1 >= total_zones)
936 hint = 0;
937 else
938 hint = (zoneno + 1) * sbi->secs_per_zone;
939 init = false;
940 goto find_other_zone;
941 }
942got_it:
943 /* set it as dirty segment in free segmap */
9850cf4a 944 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 945 __set_inuse(sbi, segno);
946 *newseg = segno;
1a118ccf 947 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 948}
949
950static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
951{
952 struct curseg_info *curseg = CURSEG_I(sbi, type);
953 struct summary_footer *sum_footer;
954
955 curseg->segno = curseg->next_segno;
956 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
957 curseg->next_blkoff = 0;
958 curseg->next_segno = NULL_SEGNO;
959
960 sum_footer = &(curseg->sum_blk->footer);
961 memset(sum_footer, 0, sizeof(struct summary_footer));
962 if (IS_DATASEG(type))
963 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
964 if (IS_NODESEG(type))
965 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
966 __set_sit_entry_type(sbi, type, curseg->segno, modified);
967}
968
0a8165d7 969/*
351df4b2
JK		 970 * Allocate a current working segment.
971 * This function always allocates a free segment in LFS manner.
972 */
973static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
974{
975 struct curseg_info *curseg = CURSEG_I(sbi, type);
976 unsigned int segno = curseg->segno;
977 int dir = ALLOC_LEFT;
978
979 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 980 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 981 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
982 dir = ALLOC_RIGHT;
983
984 if (test_opt(sbi, NOHEAP))
985 dir = ALLOC_RIGHT;
986
987 get_new_segment(sbi, &segno, new_sec, dir);
988 curseg->next_segno = segno;
989 reset_curseg(sbi, type, 1);
990 curseg->alloc_type = LFS;
991}
992
993static void __next_free_blkoff(struct f2fs_sb_info *sbi,
994 struct curseg_info *seg, block_t start)
995{
996 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 997 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 998 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 999 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1000 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1001 int i, pos;
1002
1003 for (i = 0; i < entries; i++)
1004 target_map[i] = ckpt_map[i] | cur_map[i];
1005
1006 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1007
1008 seg->next_blkoff = pos;
351df4b2
JK		 1009}
1010
0a8165d7 1011/*
351df4b2
JK		 1012 * If a segment is written by LFS manner, next block offset is just obtained
1013 * by increasing the current block offset. However, if a segment is written by
1014 * SSR manner, next block offset obtained by calling __next_free_blkoff
1015 */
1016static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1017 struct curseg_info *seg)
1018{
1019 if (seg->alloc_type == SSR)
1020 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1021 else
1022 seg->next_blkoff++;
1023}
1024
0a8165d7 1025/*
e1c42045 1026 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1027 * manner, so it should recover the existing segment information of valid blocks
1028 */
1029static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1030{
1031 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1032 struct curseg_info *curseg = CURSEG_I(sbi, type);
1033 unsigned int new_segno = curseg->next_segno;
1034 struct f2fs_summary_block *sum_node;
1035 struct page *sum_page;
1036
1037 write_sum_page(sbi, curseg->sum_blk,
1038 GET_SUM_BLOCK(sbi, curseg->segno));
1039 __set_test_and_inuse(sbi, new_segno);
1040
1041 mutex_lock(&dirty_i->seglist_lock);
1042 __remove_dirty_segment(sbi, new_segno, PRE);
1043 __remove_dirty_segment(sbi, new_segno, DIRTY);
1044 mutex_unlock(&dirty_i->seglist_lock);
1045
1046 reset_curseg(sbi, type, 1);
1047 curseg->alloc_type = SSR;
1048 __next_free_blkoff(sbi, curseg, 0);
1049
1050 if (reuse) {
1051 sum_page = get_sum_page(sbi, new_segno);
1052 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1053 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1054 f2fs_put_page(sum_page, 1);
1055 }
1056}
1057
43727527
JK		 1058static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1059{
1060 struct curseg_info *curseg = CURSEG_I(sbi, type);
1061 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1062
1063 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1064 return v_ops->get_victim(sbi,
1065 &(curseg)->next_segno, BG_GC, type, SSR);
1066
1067 /* For data segments, let's do SSR more intensively */
1068 for (; type >= CURSEG_HOT_DATA; type--)
1069 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1070 BG_GC, type, SSR))
1071 return 1;
1072 return 0;
1073}
1074
351df4b2
JK		 1075/*
1076 * flush out current segment and replace it with new segment
1077 * This function should be returned with success, otherwise BUG
1078 */
1079static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1080 int type, bool force)
1081{
1082 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1083
7b405275 1084 if (force)
351df4b2 1085 new_curseg(sbi, type, true);
7b405275 1086 else if (type == CURSEG_WARM_NODE)
351df4b2 1087 new_curseg(sbi, type, false);
60374688
JK		 1088 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1089 new_curseg(sbi, type, false);
351df4b2
JK		 1090 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1091 change_curseg(sbi, type, true);
1092 else
1093 new_curseg(sbi, type, false);
dcdfff65
JK		 1094
1095 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1096}
1097
38aa0889
JK		 1098static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1099{
1100 struct curseg_info *curseg = CURSEG_I(sbi, type);
1101 unsigned int old_segno;
1102
1103 old_segno = curseg->segno;
1104 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1105 locate_dirty_segment(sbi, old_segno);
1106}
1107
351df4b2
JK		 1108void allocate_new_segments(struct f2fs_sb_info *sbi)
1109{
351df4b2
JK		 1110 int i;
1111
38aa0889
JK		 1112 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1113 __allocate_new_segments(sbi, i);
351df4b2
JK		 1114}
1115
1116static const struct segment_allocation default_salloc_ops = {
1117 .allocate_segment = allocate_segment_by_default,
1118};
1119
4b2fecc8
JK		 1120int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1121{
f7ef9b83
JK		 1122 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1123 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1124 unsigned int start_segno, end_segno;
1125 struct cp_control cpc;
c34f42e2 1126 int err = 0;
4b2fecc8 1127
836b5a63 1128 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1129 return -EINVAL;
1130
9bd27ae4 1131 cpc.trimmed = 0;
7cd8558b 1132 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1133 goto out;
1134
1135 /* start/end segment number in main_area */
7cd8558b
JK		 1136 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1137 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1138 GET_SEGNO(sbi, end);
4b2fecc8 1139 cpc.reason = CP_DISCARD;
836b5a63 1140 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1141
1142 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1143 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1144 cpc.trim_start = start_segno;
a66cdd98
JK		 1145
1146 if (sbi->discard_blks == 0)
1147 break;
1148 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1149 cpc.trim_end = end_segno;
1150 else
1151 cpc.trim_end = min_t(unsigned int,
1152 rounddown(start_segno +
bba681cb
JK		 1153 BATCHED_TRIM_SEGMENTS(sbi),
1154 sbi->segs_per_sec) - 1, end_segno);
1155
1156 mutex_lock(&sbi->gc_mutex);
c34f42e2 1157 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1158 mutex_unlock(&sbi->gc_mutex);
1159 }
4b2fecc8 1160out:
f7ef9b83 1161 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1162 return err;
4b2fecc8
JK		 1163}
1164
351df4b2
JK		 1165static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1166{
1167 struct curseg_info *curseg = CURSEG_I(sbi, type);
1168 if (curseg->next_blkoff < sbi->blocks_per_seg)
1169 return true;
1170 return false;
1171}
1172
1173static int __get_segment_type_2(struct page *page, enum page_type p_type)
1174{
1175 if (p_type == DATA)
1176 return CURSEG_HOT_DATA;
1177 else
1178 return CURSEG_HOT_NODE;
1179}
1180
1181static int __get_segment_type_4(struct page *page, enum page_type p_type)
1182{
1183 if (p_type == DATA) {
1184 struct inode *inode = page->mapping->host;
1185
1186 if (S_ISDIR(inode->i_mode))
1187 return CURSEG_HOT_DATA;
1188 else
1189 return CURSEG_COLD_DATA;
1190 } else {
a344b9fd
JK		 1191 if (IS_DNODE(page) && is_cold_node(page))
1192 return CURSEG_WARM_NODE;
351df4b2
JK		 1193 else
1194 return CURSEG_COLD_NODE;
1195 }
1196}
1197
1198static int __get_segment_type_6(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;
354a3399 1205 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1206 return CURSEG_COLD_DATA;
1207 else
1208 return CURSEG_WARM_DATA;
1209 } else {
1210 if (IS_DNODE(page))
1211 return is_cold_node(page) ? CURSEG_WARM_NODE :
1212 CURSEG_HOT_NODE;
1213 else
1214 return CURSEG_COLD_NODE;
1215 }
1216}
1217
1218static int __get_segment_type(struct page *page, enum page_type p_type)
1219{
4081363f 1220 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1221 case 2:
1222 return __get_segment_type_2(page, p_type);
1223 case 4:
1224 return __get_segment_type_4(page, p_type);
351df4b2 1225 }
12a67146 1226 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1227 f2fs_bug_on(F2FS_P_SB(page),
1228 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1229 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1230}
1231
bfad7c2d
JK		 1232void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1233 block_t old_blkaddr, block_t *new_blkaddr,
1234 struct f2fs_summary *sum, int type)
351df4b2
JK		 1235{
1236 struct sit_info *sit_i = SIT_I(sbi);
1237 struct curseg_info *curseg;
38aa0889
JK		 1238 bool direct_io = (type == CURSEG_DIRECT_IO);
1239
1240 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1241
351df4b2
JK		 1242 curseg = CURSEG_I(sbi, type);
1243
1244 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1245 mutex_lock(&sit_i->sentry_lock);
351df4b2 1246
38aa0889 1247 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1248 if (direct_io && curseg->next_blkoff &&
1249 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1250 __allocate_new_segments(sbi, type);
1251
351df4b2 1252 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1253
1254 /*
1255 * __add_sum_entry should be resided under the curseg_mutex
1256 * because, this function updates a summary entry in the
1257 * current summary block.
1258 */
e79efe3b 1259 __add_sum_entry(sbi, type, sum);
351df4b2 1260
351df4b2 1261 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1262
1263 stat_inc_block_count(sbi, curseg);
351df4b2 1264
5e443818
JK		 1265 if (!__has_curseg_space(sbi, type))
1266 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1267 /*
1268 * SIT information should be updated before segment allocation,
1269 * since SSR needs latest valid block information.
1270 */
1271 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1272
351df4b2
JK		 1273 mutex_unlock(&sit_i->sentry_lock);
1274
bfad7c2d 1275 if (page && IS_NODESEG(type))
351df4b2
JK		 1276 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1277
bfad7c2d
JK		 1278 mutex_unlock(&curseg->curseg_mutex);
1279}
1280
05ca3632 1281static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1282{
05ca3632 1283 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1284
05ca3632
JK		 1285 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1286 &fio->blk_addr, sum, type);
bfad7c2d 1287
351df4b2 1288 /* writeout dirty page into bdev */
05ca3632 1289 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1290}
1291
577e3495 1292void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1293{
458e6197 1294 struct f2fs_io_info fio = {
05ca3632 1295 .sbi = sbi,
458e6197 1296 .type = META,
cf04e8eb
JK		 1297 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1298 .blk_addr = page->index,
05ca3632 1299 .page = page,
4375a336 1300 .encrypted_page = NULL,
458e6197
JK		 1301 };
1302
2b947003
CY		 1303 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1304 fio.rw &= ~REQ_META;
1305
351df4b2 1306 set_page_writeback(page);
05ca3632 1307 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1308}
1309
05ca3632 1310void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1311{
1312 struct f2fs_summary sum;
05ca3632 1313
351df4b2 1314 set_summary(&sum, nid, 0, 0);
05ca3632 1315 do_write_page(&sum, fio);
351df4b2
JK		 1316}
1317
05ca3632 1318void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1319{
05ca3632 1320 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1321 struct f2fs_summary sum;
1322 struct node_info ni;
1323
9850cf4a 1324 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1325 get_node_info(sbi, dn->nid, &ni);
1326 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1327 do_write_page(&sum, fio);
e1509cf2 1328 dn->data_blkaddr = fio->blk_addr;
351df4b2
JK		 1329}
1330
05ca3632 1331void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1332{
05ca3632
JK		 1333 stat_inc_inplace_blocks(fio->sbi);
1334 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1335}
1336
528e3459
CY		 1337static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1338 struct f2fs_summary *sum,
19f106bc
CY		 1339 block_t old_blkaddr, block_t new_blkaddr,
1340 bool recover_curseg)
351df4b2
JK		 1341{
1342 struct sit_info *sit_i = SIT_I(sbi);
1343 struct curseg_info *curseg;
1344 unsigned int segno, old_cursegno;
1345 struct seg_entry *se;
1346 int type;
19f106bc 1347 unsigned short old_blkoff;
351df4b2
JK		 1348
1349 segno = GET_SEGNO(sbi, new_blkaddr);
1350 se = get_seg_entry(sbi, segno);
1351 type = se->type;
1352
19f106bc
CY		 1353 if (!recover_curseg) {
1354 /* for recovery flow */
1355 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1356 if (old_blkaddr == NULL_ADDR)
1357 type = CURSEG_COLD_DATA;
1358 else
1359 type = CURSEG_WARM_DATA;
1360 }
1361 } else {
1362 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1363 type = CURSEG_WARM_DATA;
1364 }
19f106bc 1365
351df4b2
JK		 1366 curseg = CURSEG_I(sbi, type);
1367
1368 mutex_lock(&curseg->curseg_mutex);
1369 mutex_lock(&sit_i->sentry_lock);
1370
1371 old_cursegno = curseg->segno;
19f106bc 1372 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1373
1374 /* change the current segment */
1375 if (segno != curseg->segno) {
1376 curseg->next_segno = segno;
1377 change_curseg(sbi, type, true);
1378 }
1379
491c0854 1380 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1381 __add_sum_entry(sbi, type, sum);
351df4b2 1382
6e2c64ad
JK		 1383 if (!recover_curseg)
1384 update_sit_entry(sbi, new_blkaddr, 1);
1385 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1386 update_sit_entry(sbi, old_blkaddr, -1);
1387
1388 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1389 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1390
351df4b2 1391 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1392
19f106bc
CY		 1393 if (recover_curseg) {
1394 if (old_cursegno != curseg->segno) {
1395 curseg->next_segno = old_cursegno;
1396 change_curseg(sbi, type, true);
1397 }
1398 curseg->next_blkoff = old_blkoff;
1399 }
1400
351df4b2
JK		 1401 mutex_unlock(&sit_i->sentry_lock);
1402 mutex_unlock(&curseg->curseg_mutex);
1403}
1404
528e3459
CY		 1405void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1406 block_t old_addr, block_t new_addr,
1407 unsigned char version, bool recover_curseg)
1408{
1409 struct f2fs_summary sum;
1410
1411 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1412
1413 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1414
1415 dn->data_blkaddr = new_addr;
1416 set_data_blkaddr(dn);
1417 f2fs_update_extent_cache(dn);
1418}
1419
93dfe2ac 1420void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1421 enum page_type type, bool ordered)
93dfe2ac 1422{
93dfe2ac 1423 if (PageWriteback(page)) {
4081363f
JK		 1424 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1425
0c3a5797 1426 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1427 if (ordered)
1428 wait_on_page_writeback(page);
1429 else
1430 wait_for_stable_page(page);
93dfe2ac
JK		 1431 }
1432}
1433
08b39fbd
CY		 1434void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1435 block_t blkaddr)
1436{
1437 struct page *cpage;
1438
1439 if (blkaddr == NEW_ADDR)
1440 return;
1441
1442 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1443
1444 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1445 if (cpage) {
fec1d657 1446 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1447 f2fs_put_page(cpage, 1);
1448 }
1449}
1450
351df4b2
JK		 1451static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1452{
1453 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1454 struct curseg_info *seg_i;
1455 unsigned char *kaddr;
1456 struct page *page;
1457 block_t start;
1458 int i, j, offset;
1459
1460 start = start_sum_block(sbi);
1461
1462 page = get_meta_page(sbi, start++);
1463 kaddr = (unsigned char *)page_address(page);
1464
1465 /* Step 1: restore nat cache */
1466 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1467 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1468
1469 /* Step 2: restore sit cache */
1470 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1471 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1472 SUM_JOURNAL_SIZE);
1473 offset = 2 * SUM_JOURNAL_SIZE;
1474
1475 /* Step 3: restore summary entries */
1476 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1477 unsigned short blk_off;
1478 unsigned int segno;
1479
1480 seg_i = CURSEG_I(sbi, i);
1481 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1482 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1483 seg_i->next_segno = segno;
1484 reset_curseg(sbi, i, 0);
1485 seg_i->alloc_type = ckpt->alloc_type[i];
1486 seg_i->next_blkoff = blk_off;
1487
1488 if (seg_i->alloc_type == SSR)
1489 blk_off = sbi->blocks_per_seg;
1490
1491 for (j = 0; j < blk_off; j++) {
1492 struct f2fs_summary *s;
1493 s = (struct f2fs_summary *)(kaddr + offset);
1494 seg_i->sum_blk->entries[j] = *s;
1495 offset += SUMMARY_SIZE;
1496 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1497 SUM_FOOTER_SIZE)
1498 continue;
1499
1500 f2fs_put_page(page, 1);
1501 page = NULL;
1502
1503 page = get_meta_page(sbi, start++);
1504 kaddr = (unsigned char *)page_address(page);
1505 offset = 0;
1506 }
1507 }
1508 f2fs_put_page(page, 1);
1509 return 0;
1510}
1511
1512static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1513{
1514 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1515 struct f2fs_summary_block *sum;
1516 struct curseg_info *curseg;
1517 struct page *new;
1518 unsigned short blk_off;
1519 unsigned int segno = 0;
1520 block_t blk_addr = 0;
1521
1522 /* get segment number and block addr */
1523 if (IS_DATASEG(type)) {
1524 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1525 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1526 CURSEG_HOT_DATA]);
119ee914 1527 if (__exist_node_summaries(sbi))
351df4b2
JK		 1528 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1529 else
1530 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1531 } else {
1532 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1533 CURSEG_HOT_NODE]);
1534 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1535 CURSEG_HOT_NODE]);
119ee914 1536 if (__exist_node_summaries(sbi))
351df4b2
JK		 1537 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1538 type - CURSEG_HOT_NODE);
1539 else
1540 blk_addr = GET_SUM_BLOCK(sbi, segno);
1541 }
1542
1543 new = get_meta_page(sbi, blk_addr);
1544 sum = (struct f2fs_summary_block *)page_address(new);
1545
1546 if (IS_NODESEG(type)) {
119ee914 1547 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1548 struct f2fs_summary *ns = &sum->entries[0];
1549 int i;
1550 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1551 ns->version = 0;
1552 ns->ofs_in_node = 0;
1553 }
1554 } else {
d653788a
GZ		 1555 int err;
1556
1557 err = restore_node_summary(sbi, segno, sum);
1558 if (err) {
351df4b2 1559 f2fs_put_page(new, 1);
d653788a 1560 return err;
351df4b2
JK		 1561 }
1562 }
1563 }
1564
1565 /* set uncompleted segment to curseg */
1566 curseg = CURSEG_I(sbi, type);
1567 mutex_lock(&curseg->curseg_mutex);
1568 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1569 curseg->next_segno = segno;
1570 reset_curseg(sbi, type, 0);
1571 curseg->alloc_type = ckpt->alloc_type[type];
1572 curseg->next_blkoff = blk_off;
1573 mutex_unlock(&curseg->curseg_mutex);
1574 f2fs_put_page(new, 1);
1575 return 0;
1576}
1577
1578static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1579{
1580 int type = CURSEG_HOT_DATA;
e4fc5fbf 1581 int err;
351df4b2 1582
25ca923b 1583 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1584 int npages = npages_for_summary_flush(sbi, true);
1585
1586 if (npages >= 2)
1587 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1588 META_CP, true);
3fa06d7b 1589
351df4b2
JK		 1590 /* restore for compacted data summary */
1591 if (read_compacted_summaries(sbi))
1592 return -EINVAL;
1593 type = CURSEG_HOT_NODE;
1594 }
1595
119ee914 1596 if (__exist_node_summaries(sbi))
3fa06d7b 1597 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1598 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1599
e4fc5fbf
CY		 1600 for (; type <= CURSEG_COLD_NODE; type++) {
1601 err = read_normal_summaries(sbi, type);
1602 if (err)
1603 return err;
1604 }
1605
351df4b2
JK		 1606 return 0;
1607}
1608
1609static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1610{
1611 struct page *page;
1612 unsigned char *kaddr;
1613 struct f2fs_summary *summary;
1614 struct curseg_info *seg_i;
1615 int written_size = 0;
1616 int i, j;
1617
1618 page = grab_meta_page(sbi, blkaddr++);
1619 kaddr = (unsigned char *)page_address(page);
1620
1621 /* Step 1: write nat cache */
1622 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1623 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1624 written_size += SUM_JOURNAL_SIZE;
1625
1626 /* Step 2: write sit cache */
1627 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1628 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1629 SUM_JOURNAL_SIZE);
1630 written_size += SUM_JOURNAL_SIZE;
1631
351df4b2
JK		 1632 /* Step 3: write summary entries */
1633 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1634 unsigned short blkoff;
1635 seg_i = CURSEG_I(sbi, i);
1636 if (sbi->ckpt->alloc_type[i] == SSR)
1637 blkoff = sbi->blocks_per_seg;
1638 else
1639 blkoff = curseg_blkoff(sbi, i);
1640
1641 for (j = 0; j < blkoff; j++) {
1642 if (!page) {
1643 page = grab_meta_page(sbi, blkaddr++);
1644 kaddr = (unsigned char *)page_address(page);
1645 written_size = 0;
1646 }
1647 summary = (struct f2fs_summary *)(kaddr + written_size);
1648 *summary = seg_i->sum_blk->entries[j];
1649 written_size += SUMMARY_SIZE;
351df4b2
JK		 1650
1651 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1652 SUM_FOOTER_SIZE)
1653 continue;
1654
e8d61a74 1655 set_page_dirty(page);
351df4b2
JK		 1656 f2fs_put_page(page, 1);
1657 page = NULL;
1658 }
1659 }
e8d61a74
CY		 1660 if (page) {
1661 set_page_dirty(page);
351df4b2 1662 f2fs_put_page(page, 1);
e8d61a74 1663 }
351df4b2
JK		 1664}
1665
1666static void write_normal_summaries(struct f2fs_sb_info *sbi,
1667 block_t blkaddr, int type)
1668{
1669 int i, end;
1670 if (IS_DATASEG(type))
1671 end = type + NR_CURSEG_DATA_TYPE;
1672 else
1673 end = type + NR_CURSEG_NODE_TYPE;
1674
1675 for (i = type; i < end; i++) {
1676 struct curseg_info *sum = CURSEG_I(sbi, i);
1677 mutex_lock(&sum->curseg_mutex);
1678 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1679 mutex_unlock(&sum->curseg_mutex);
1680 }
1681}
1682
1683void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1684{
25ca923b 1685 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1686 write_compacted_summaries(sbi, start_blk);
1687 else
1688 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1689}
1690
1691void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1692{
119ee914 1693 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1694}
1695
1696int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1697 unsigned int val, int alloc)
1698{
1699 int i;
1700
1701 if (type == NAT_JOURNAL) {
1702 for (i = 0; i < nats_in_cursum(sum); i++) {
1703 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1704 return i;
1705 }
855639de 1706 if (alloc && __has_cursum_space(sum, 1, NAT_JOURNAL))
351df4b2
JK		 1707 return update_nats_in_cursum(sum, 1);
1708 } else if (type == SIT_JOURNAL) {
1709 for (i = 0; i < sits_in_cursum(sum); i++)
1710 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1711 return i;
855639de 1712 if (alloc && __has_cursum_space(sum, 1, SIT_JOURNAL))
351df4b2
JK		 1713 return update_sits_in_cursum(sum, 1);
1714 }
1715 return -1;
1716}
1717
1718static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1719 unsigned int segno)
1720{
2cc22186 1721 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1722}
1723
1724static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1725 unsigned int start)
1726{
1727 struct sit_info *sit_i = SIT_I(sbi);
1728 struct page *src_page, *dst_page;
1729 pgoff_t src_off, dst_off;
1730 void *src_addr, *dst_addr;
1731
1732 src_off = current_sit_addr(sbi, start);
1733 dst_off = next_sit_addr(sbi, src_off);
1734
1735 /* get current sit block page without lock */
1736 src_page = get_meta_page(sbi, src_off);
1737 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1738 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1739
1740 src_addr = page_address(src_page);
1741 dst_addr = page_address(dst_page);
1742 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1743
1744 set_page_dirty(dst_page);
1745 f2fs_put_page(src_page, 1);
1746
1747 set_to_next_sit(sit_i, start);
1748
1749 return dst_page;
1750}
1751
184a5cd2
CY		 1752static struct sit_entry_set *grab_sit_entry_set(void)
1753{
1754 struct sit_entry_set *ses =
80c54505 1755 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1756
1757 ses->entry_cnt = 0;
1758 INIT_LIST_HEAD(&ses->set_list);
1759 return ses;
1760}
1761
1762static void release_sit_entry_set(struct sit_entry_set *ses)
1763{
1764 list_del(&ses->set_list);
1765 kmem_cache_free(sit_entry_set_slab, ses);
1766}
1767
1768static void adjust_sit_entry_set(struct sit_entry_set *ses,
1769 struct list_head *head)
1770{
1771 struct sit_entry_set *next = ses;
1772
1773 if (list_is_last(&ses->set_list, head))
1774 return;
1775
1776 list_for_each_entry_continue(next, head, set_list)
1777 if (ses->entry_cnt <= next->entry_cnt)
1778 break;
1779
1780 list_move_tail(&ses->set_list, &next->set_list);
1781}
1782
1783static void add_sit_entry(unsigned int segno, struct list_head *head)
1784{
1785 struct sit_entry_set *ses;
1786 unsigned int start_segno = START_SEGNO(segno);
1787
1788 list_for_each_entry(ses, head, set_list) {
1789 if (ses->start_segno == start_segno) {
1790 ses->entry_cnt++;
1791 adjust_sit_entry_set(ses, head);
1792 return;
1793 }
1794 }
1795
1796 ses = grab_sit_entry_set();
1797
1798 ses->start_segno = start_segno;
1799 ses->entry_cnt++;
1800 list_add(&ses->set_list, head);
1801}
1802
1803static void add_sits_in_set(struct f2fs_sb_info *sbi)
1804{
1805 struct f2fs_sm_info *sm_info = SM_I(sbi);
1806 struct list_head *set_list = &sm_info->sit_entry_set;
1807 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1808 unsigned int segno;
1809
7cd8558b 1810 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1811 add_sit_entry(segno, set_list);
1812}
1813
1814static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1815{
1816 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1817 struct f2fs_summary_block *sum = curseg->sum_blk;
1818 int i;
1819
184a5cd2
CY		 1820 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1821 unsigned int segno;
1822 bool dirtied;
1823
1824 segno = le32_to_cpu(segno_in_journal(sum, i));
1825 dirtied = __mark_sit_entry_dirty(sbi, segno);
1826
1827 if (!dirtied)
1828 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1829 }
184a5cd2 1830 update_sits_in_cursum(sum, -sits_in_cursum(sum));
351df4b2
JK		 1831}
1832
0a8165d7 1833/*
351df4b2
JK		 1834 * CP calls this function, which flushes SIT entries including sit_journal,
1835 * and moves prefree segs to free segs.
1836 */
4b2fecc8 1837void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1838{
1839 struct sit_info *sit_i = SIT_I(sbi);
1840 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1841 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1842 struct f2fs_summary_block *sum = curseg->sum_blk;
184a5cd2
CY		 1843 struct sit_entry_set *ses, *tmp;
1844 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1845 bool to_journal = true;
4b2fecc8 1846 struct seg_entry *se;
351df4b2
JK		 1847
1848 mutex_lock(&curseg->curseg_mutex);
1849 mutex_lock(&sit_i->sentry_lock);
1850
2b11a74b
WL		 1851 if (!sit_i->dirty_sentries)
1852 goto out;
1853
351df4b2 1854 /*
184a5cd2
CY		 1855 * add and account sit entries of dirty bitmap in sit entry
1856 * set temporarily
351df4b2 1857 */
184a5cd2 1858 add_sits_in_set(sbi);
351df4b2 1859
184a5cd2
CY		 1860 /*
1861 * if there are no enough space in journal to store dirty sit
1862 * entries, remove all entries from journal and add and account
1863 * them in sit entry set.
1864 */
1865 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1866 remove_sits_in_journal(sbi);
b2955550 1867
184a5cd2
CY		 1868 /*
1869 * there are two steps to flush sit entries:
1870 * #1, flush sit entries to journal in current cold data summary block.
1871 * #2, flush sit entries to sit page.
1872 */
1873 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1874 struct page *page = NULL;
184a5cd2
CY		 1875 struct f2fs_sit_block *raw_sit = NULL;
1876 unsigned int start_segno = ses->start_segno;
1877 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1878 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1879 unsigned int segno = start_segno;
1880
1881 if (to_journal &&
1882 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1883 to_journal = false;
1884
1885 if (!to_journal) {
1886 page = get_next_sit_page(sbi, start_segno);
1887 raw_sit = page_address(page);
351df4b2 1888 }
351df4b2 1889
184a5cd2
CY		 1890 /* flush dirty sit entries in region of current sit set */
1891 for_each_set_bit_from(segno, bitmap, end) {
1892 int offset, sit_offset;
4b2fecc8
JK		 1893
1894 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 1895
1896 /* add discard candidates */
d7bc2484 1897 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 1898 cpc->trim_start = segno;
1899 add_discard_addrs(sbi, cpc);
1900 }
184a5cd2
CY		 1901
1902 if (to_journal) {
1903 offset = lookup_journal_in_cursum(sum,
1904 SIT_JOURNAL, segno, 1);
1905 f2fs_bug_on(sbi, offset < 0);
1906 segno_in_journal(sum, offset) =
1907 cpu_to_le32(segno);
1908 seg_info_to_raw_sit(se,
1909 &sit_in_journal(sum, offset));
1910 } else {
1911 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1912 seg_info_to_raw_sit(se,
1913 &raw_sit->entries[sit_offset]);
1914 }
351df4b2 1915
184a5cd2
CY		 1916 __clear_bit(segno, bitmap);
1917 sit_i->dirty_sentries--;
1918 ses->entry_cnt--;
351df4b2
JK		 1919 }
1920
184a5cd2
CY		 1921 if (!to_journal)
1922 f2fs_put_page(page, 1);
1923
1924 f2fs_bug_on(sbi, ses->entry_cnt);
1925 release_sit_entry_set(ses);
351df4b2 1926 }
184a5cd2
CY		 1927
1928 f2fs_bug_on(sbi, !list_empty(head));
1929 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 1930out:
4b2fecc8
JK		 1931 if (cpc->reason == CP_DISCARD) {
1932 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1933 add_discard_addrs(sbi, cpc);
1934 }
351df4b2
JK		 1935 mutex_unlock(&sit_i->sentry_lock);
1936 mutex_unlock(&curseg->curseg_mutex);
1937
351df4b2
JK		 1938 set_prefree_as_free_segments(sbi);
1939}
1940
1941static int build_sit_info(struct f2fs_sb_info *sbi)
1942{
1943 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1944 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1945 struct sit_info *sit_i;
1946 unsigned int sit_segs, start;
1947 char *src_bitmap, *dst_bitmap;
1948 unsigned int bitmap_size;
1949
1950 /* allocate memory for SIT information */
1951 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1952 if (!sit_i)
1953 return -ENOMEM;
1954
1955 SM_I(sbi)->sit_info = sit_i;
1956
39307a8e
JK		 1957 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
1958 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 1959 if (!sit_i->sentries)
1960 return -ENOMEM;
1961
7cd8558b 1962 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 1963 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 1964 if (!sit_i->dirty_sentries_bitmap)
1965 return -ENOMEM;
1966
7cd8558b 1967 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 1968 sit_i->sentries[start].cur_valid_map
1969 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1970 sit_i->sentries[start].ckpt_valid_map
1971 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 1972 sit_i->sentries[start].discard_map
1973 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1974 if (!sit_i->sentries[start].cur_valid_map ||
1975 !sit_i->sentries[start].ckpt_valid_map ||
1976 !sit_i->sentries[start].discard_map)
351df4b2
JK		 1977 return -ENOMEM;
1978 }
1979
60a3b782
JK		 1980 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1981 if (!sit_i->tmp_map)
1982 return -ENOMEM;
1983
351df4b2 1984 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 1985 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
1986 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 1987 if (!sit_i->sec_entries)
1988 return -ENOMEM;
1989 }
1990
1991 /* get information related with SIT */
1992 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1993
1994 /* setup SIT bitmap from ckeckpoint pack */
1995 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1996 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1997
79b5793b 1998 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 1999 if (!dst_bitmap)
2000 return -ENOMEM;
351df4b2
JK		 2001
2002 /* init SIT information */
2003 sit_i->s_ops = &default_salloc_ops;
2004
2005 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2006 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2007 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2008 sit_i->sit_bitmap = dst_bitmap;
2009 sit_i->bitmap_size = bitmap_size;
2010 sit_i->dirty_sentries = 0;
2011 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2012 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2013 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2014 mutex_init(&sit_i->sentry_lock);
2015 return 0;
2016}
2017
2018static int build_free_segmap(struct f2fs_sb_info *sbi)
2019{
351df4b2
JK		 2020 struct free_segmap_info *free_i;
2021 unsigned int bitmap_size, sec_bitmap_size;
2022
2023 /* allocate memory for free segmap information */
2024 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2025 if (!free_i)
2026 return -ENOMEM;
2027
2028 SM_I(sbi)->free_info = free_i;
2029
7cd8558b 2030 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2031 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2032 if (!free_i->free_segmap)
2033 return -ENOMEM;
2034
7cd8558b 2035 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2036 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2037 if (!free_i->free_secmap)
2038 return -ENOMEM;
2039
2040 /* set all segments as dirty temporarily */
2041 memset(free_i->free_segmap, 0xff, bitmap_size);
2042 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2043
2044 /* init free segmap information */
7cd8558b 2045 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2046 free_i->free_segments = 0;
2047 free_i->free_sections = 0;
1a118ccf 2048 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2049 return 0;
2050}
2051
2052static int build_curseg(struct f2fs_sb_info *sbi)
2053{
1042d60f 2054 struct curseg_info *array;
351df4b2
JK		 2055 int i;
2056
b434babf 2057 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2058 if (!array)
2059 return -ENOMEM;
2060
2061 SM_I(sbi)->curseg_array = array;
2062
2063 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2064 mutex_init(&array[i].curseg_mutex);
2065 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2066 if (!array[i].sum_blk)
2067 return -ENOMEM;
2068 array[i].segno = NULL_SEGNO;
2069 array[i].next_blkoff = 0;
2070 }
2071 return restore_curseg_summaries(sbi);
2072}
2073
2074static void build_sit_entries(struct f2fs_sb_info *sbi)
2075{
2076 struct sit_info *sit_i = SIT_I(sbi);
2077 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2078 struct f2fs_summary_block *sum = curseg->sum_blk;
74de593a
CY		 2079 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2080 unsigned int i, start, end;
2081 unsigned int readed, start_blk = 0;
90a893c7 2082 int nrpages = MAX_BIO_BLOCKS(sbi);
351df4b2 2083
74de593a 2084 do {
26879fb1 2085 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2086
2087 start = start_blk * sit_i->sents_per_block;
2088 end = (start_blk + readed) * sit_i->sents_per_block;
2089
7cd8558b 2090 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2091 struct seg_entry *se = &sit_i->sentries[start];
2092 struct f2fs_sit_block *sit_blk;
2093 struct f2fs_sit_entry sit;
2094 struct page *page;
2095
2096 mutex_lock(&curseg->curseg_mutex);
2097 for (i = 0; i < sits_in_cursum(sum); i++) {
6c311ec6
CF		 2098 if (le32_to_cpu(segno_in_journal(sum, i))
2099 == start) {
74de593a
CY		 2100 sit = sit_in_journal(sum, i);
2101 mutex_unlock(&curseg->curseg_mutex);
2102 goto got_it;
2103 }
351df4b2 2104 }
74de593a
CY		 2105 mutex_unlock(&curseg->curseg_mutex);
2106
2107 page = get_current_sit_page(sbi, start);
2108 sit_blk = (struct f2fs_sit_block *)page_address(page);
2109 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2110 f2fs_put_page(page, 1);
351df4b2 2111got_it:
74de593a
CY		 2112 check_block_count(sbi, start, &sit);
2113 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2114
2115 /* build discard map only one time */
2116 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2117 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2118
74de593a
CY		 2119 if (sbi->segs_per_sec > 1) {
2120 struct sec_entry *e = get_sec_entry(sbi, start);
2121 e->valid_blocks += se->valid_blocks;
2122 }
351df4b2 2123 }
74de593a
CY		 2124 start_blk += readed;
2125 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2126}
2127
2128static void init_free_segmap(struct f2fs_sb_info *sbi)
2129{
2130 unsigned int start;
2131 int type;
2132
7cd8558b 2133 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2134 struct seg_entry *sentry = get_seg_entry(sbi, start);
2135 if (!sentry->valid_blocks)
2136 __set_free(sbi, start);
2137 }
2138
2139 /* set use the current segments */
2140 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2141 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2142 __set_test_and_inuse(sbi, curseg_t->segno);
2143 }
2144}
2145
2146static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2147{
2148 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2149 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2150 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2151 unsigned short valid_blocks;
2152
8736fbf0 2153 while (1) {
351df4b2 2154 /* find dirty segment based on free segmap */
7cd8558b
JK		 2155 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2156 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2157 break;
2158 offset = segno + 1;
2159 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2160 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2161 continue;
ec325b52
JK		 2162 if (valid_blocks > sbi->blocks_per_seg) {
2163 f2fs_bug_on(sbi, 1);
2164 continue;
2165 }
351df4b2
JK		 2166 mutex_lock(&dirty_i->seglist_lock);
2167 __locate_dirty_segment(sbi, segno, DIRTY);
2168 mutex_unlock(&dirty_i->seglist_lock);
2169 }
2170}
2171
5ec4e49f 2172static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2173{
2174 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2175 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2176
39307a8e 2177 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2178 if (!dirty_i->victim_secmap)
351df4b2
JK		 2179 return -ENOMEM;
2180 return 0;
2181}
2182
2183static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2184{
2185 struct dirty_seglist_info *dirty_i;
2186 unsigned int bitmap_size, i;
2187
2188 /* allocate memory for dirty segments list information */
2189 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2190 if (!dirty_i)
2191 return -ENOMEM;
2192
2193 SM_I(sbi)->dirty_info = dirty_i;
2194 mutex_init(&dirty_i->seglist_lock);
2195
7cd8558b 2196 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2197
2198 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2199 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2200 if (!dirty_i->dirty_segmap[i])
2201 return -ENOMEM;
2202 }
2203
2204 init_dirty_segmap(sbi);
5ec4e49f 2205 return init_victim_secmap(sbi);
351df4b2
JK		 2206}
2207
0a8165d7 2208/*
351df4b2
JK		 2209 * Update min, max modified time for cost-benefit GC algorithm
2210 */
2211static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2212{
2213 struct sit_info *sit_i = SIT_I(sbi);
2214 unsigned int segno;
2215
2216 mutex_lock(&sit_i->sentry_lock);
2217
2218 sit_i->min_mtime = LLONG_MAX;
2219
7cd8558b 2220 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2221 unsigned int i;
2222 unsigned long long mtime = 0;
2223
2224 for (i = 0; i < sbi->segs_per_sec; i++)
2225 mtime += get_seg_entry(sbi, segno + i)->mtime;
2226
2227 mtime = div_u64(mtime, sbi->segs_per_sec);
2228
2229 if (sit_i->min_mtime > mtime)
2230 sit_i->min_mtime = mtime;
2231 }
2232 sit_i->max_mtime = get_mtime(sbi);
2233 mutex_unlock(&sit_i->sentry_lock);
2234}
2235
2236int build_segment_manager(struct f2fs_sb_info *sbi)
2237{
2238 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2239 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2240 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2241 int err;
2242
2243 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2244 if (!sm_info)
2245 return -ENOMEM;
2246
2247 /* init sm info */
2248 sbi->sm_info = sm_info;
351df4b2
JK		 2249 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2250 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2251 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2252 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2253 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2254 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2255 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2256 sm_info->rec_prefree_segments = sm_info->main_segments *
2257 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2258 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2259 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2260 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2261
7fd9e544
JK		 2262 INIT_LIST_HEAD(&sm_info->discard_list);
2263 sm_info->nr_discards = 0;
2264 sm_info->max_discards = 0;
2265
bba681cb
JK		 2266 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2267
184a5cd2
CY		 2268 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2269
b270ad6f 2270 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2271 err = create_flush_cmd_control(sbi);
2272 if (err)
a688b9d9 2273 return err;
6b4afdd7
JK		 2274 }
2275
351df4b2
JK		 2276 err = build_sit_info(sbi);
2277 if (err)
2278 return err;
2279 err = build_free_segmap(sbi);
2280 if (err)
2281 return err;
2282 err = build_curseg(sbi);
2283 if (err)
2284 return err;
2285
2286 /* reinit free segmap based on SIT */
2287 build_sit_entries(sbi);
2288
2289 init_free_segmap(sbi);
2290 err = build_dirty_segmap(sbi);
2291 if (err)
2292 return err;
2293
2294 init_min_max_mtime(sbi);
2295 return 0;
2296}
2297
2298static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2299 enum dirty_type dirty_type)
2300{
2301 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2302
2303 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2304 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2305 dirty_i->nr_dirty[dirty_type] = 0;
2306 mutex_unlock(&dirty_i->seglist_lock);
2307}
2308
5ec4e49f 2309static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2310{
2311 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2312 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2313}
2314
2315static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2316{
2317 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2318 int i;
2319
2320 if (!dirty_i)
2321 return;
2322
2323 /* discard pre-free/dirty segments list */
2324 for (i = 0; i < NR_DIRTY_TYPE; i++)
2325 discard_dirty_segmap(sbi, i);
2326
5ec4e49f 2327 destroy_victim_secmap(sbi);
351df4b2
JK		 2328 SM_I(sbi)->dirty_info = NULL;
2329 kfree(dirty_i);
2330}
2331
2332static void destroy_curseg(struct f2fs_sb_info *sbi)
2333{
2334 struct curseg_info *array = SM_I(sbi)->curseg_array;
2335 int i;
2336
2337 if (!array)
2338 return;
2339 SM_I(sbi)->curseg_array = NULL;
2340 for (i = 0; i < NR_CURSEG_TYPE; i++)
2341 kfree(array[i].sum_blk);
2342 kfree(array);
2343}
2344
2345static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2346{
2347 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2348 if (!free_i)
2349 return;
2350 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2351 kvfree(free_i->free_segmap);
2352 kvfree(free_i->free_secmap);
351df4b2
JK		 2353 kfree(free_i);
2354}
2355
2356static void destroy_sit_info(struct f2fs_sb_info *sbi)
2357{
2358 struct sit_info *sit_i = SIT_I(sbi);
2359 unsigned int start;
2360
2361 if (!sit_i)
2362 return;
2363
2364 if (sit_i->sentries) {
7cd8558b 2365 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2366 kfree(sit_i->sentries[start].cur_valid_map);
2367 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2368 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2369 }
2370 }
60a3b782
JK		 2371 kfree(sit_i->tmp_map);
2372
39307a8e
JK		 2373 kvfree(sit_i->sentries);
2374 kvfree(sit_i->sec_entries);
2375 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2376
2377 SM_I(sbi)->sit_info = NULL;
2378 kfree(sit_i->sit_bitmap);
2379 kfree(sit_i);
2380}
2381
2382void destroy_segment_manager(struct f2fs_sb_info *sbi)
2383{
2384 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2385
3b03f724
CY		 2386 if (!sm_info)
2387 return;
2163d198 2388 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2389 destroy_dirty_segmap(sbi);
2390 destroy_curseg(sbi);
2391 destroy_free_segmap(sbi);
2392 destroy_sit_info(sbi);
2393 sbi->sm_info = NULL;
2394 kfree(sm_info);
2395}
7fd9e544
JK		 2396
2397int __init create_segment_manager_caches(void)
2398{
2399 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2400 sizeof(struct discard_entry));
7fd9e544 2401 if (!discard_entry_slab)
184a5cd2
CY		 2402 goto fail;
2403
2404 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2405 sizeof(struct sit_entry_set));
184a5cd2
CY		 2406 if (!sit_entry_set_slab)
2407 goto destory_discard_entry;
88b88a66
JK		 2408
2409 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2410 sizeof(struct inmem_pages));
2411 if (!inmem_entry_slab)
2412 goto destroy_sit_entry_set;
7fd9e544 2413 return 0;
184a5cd2 2414
88b88a66
JK		 2415destroy_sit_entry_set:
2416 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2417destory_discard_entry:
2418 kmem_cache_destroy(discard_entry_slab);
2419fail:
2420 return -ENOMEM;
7fd9e544
JK		 2421}
2422
2423void destroy_segment_manager_caches(void)
2424{
184a5cd2 2425 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2426 kmem_cache_destroy(discard_entry_slab);
88b88a66 2427 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2428}
Linux kernel - Deliverables
This page took 0.258133 seconds and 5 git commands to generate.