HID: hiddev: Fix slab-out-of-bounds write in hiddev_ioctl_usage()
[linux/fpc-iii.git] / fs / f2fs / segment.c
blobe482cca005a61f994f884f7d31351234fde7d354
1 /*
2 * fs/f2fs/segment.c
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
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.
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/swap.h>
18 #include <linux/timer.h>
20 #include "f2fs.h"
21 #include "segment.h"
22 #include "node.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *sit_entry_set_slab;
30 static struct kmem_cache *inmem_entry_slab;
32 static unsigned long __reverse_ulong(unsigned char *str)
34 unsigned long tmp = 0;
35 int shift = 24, idx = 0;
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;
44 return tmp;
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.
51 static inline unsigned long __reverse_ffs(unsigned long word)
53 int num = 0;
55 #if BITS_PER_LONG == 64
56 if ((word & 0xffffffff00000000UL) == 0)
57 num += 32;
58 else
59 word >>= 32;
60 #endif
61 if ((word & 0xffff0000) == 0)
62 num += 16;
63 else
64 word >>= 16;
66 if ((word & 0xff00) == 0)
67 num += 8;
68 else
69 word >>= 8;
71 if ((word & 0xf0) == 0)
72 num += 4;
73 else
74 word >>= 4;
76 if ((word & 0xc) == 0)
77 num += 2;
78 else
79 word >>= 2;
81 if ((word & 0x2) == 0)
82 num += 1;
83 return num;
87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
89 * Example:
90 * MSB <--> LSB
91 * f2fs_set_bit(0, bitmap) => 1000 0000
92 * f2fs_set_bit(7, bitmap) => 0000 0001
94 static unsigned long __find_rev_next_bit(const unsigned long *addr,
95 unsigned long size, unsigned long offset)
97 const unsigned long *p = addr + BIT_WORD(offset);
98 unsigned long result = offset & ~(BITS_PER_LONG - 1);
99 unsigned long tmp;
101 if (offset >= size)
102 return size;
104 size -= result;
105 offset %= BITS_PER_LONG;
106 if (!offset)
107 goto aligned;
109 tmp = __reverse_ulong((unsigned char *)p);
110 tmp &= ~0UL >> offset;
112 if (size < BITS_PER_LONG)
113 goto found_first;
114 if (tmp)
115 goto found_middle;
117 size -= BITS_PER_LONG;
118 result += BITS_PER_LONG;
119 p++;
120 aligned:
121 while (size & ~(BITS_PER_LONG-1)) {
122 tmp = __reverse_ulong((unsigned char *)p);
123 if (tmp)
124 goto found_middle;
125 result += BITS_PER_LONG;
126 size -= BITS_PER_LONG;
127 p++;
129 if (!size)
130 return result;
132 tmp = __reverse_ulong((unsigned char *)p);
133 found_first:
134 tmp &= (~0UL << (BITS_PER_LONG - size));
135 if (!tmp) /* Are any bits set? */
136 return result + size; /* Nope. */
137 found_middle:
138 return result + __reverse_ffs(tmp);
141 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
142 unsigned long size, unsigned long offset)
144 const unsigned long *p = addr + BIT_WORD(offset);
145 unsigned long result = offset & ~(BITS_PER_LONG - 1);
146 unsigned long tmp;
148 if (offset >= size)
149 return size;
151 size -= result;
152 offset %= BITS_PER_LONG;
153 if (!offset)
154 goto aligned;
156 tmp = __reverse_ulong((unsigned char *)p);
157 tmp |= ~((~0UL << offset) >> offset);
159 if (size < BITS_PER_LONG)
160 goto found_first;
161 if (tmp != ~0UL)
162 goto found_middle;
164 size -= BITS_PER_LONG;
165 result += BITS_PER_LONG;
166 p++;
167 aligned:
168 while (size & ~(BITS_PER_LONG - 1)) {
169 tmp = __reverse_ulong((unsigned char *)p);
170 if (tmp != ~0UL)
171 goto found_middle;
172 result += BITS_PER_LONG;
173 size -= BITS_PER_LONG;
174 p++;
176 if (!size)
177 return result;
179 tmp = __reverse_ulong((unsigned char *)p);
180 found_first:
181 tmp |= ~(~0UL << (BITS_PER_LONG - size));
182 if (tmp == ~0UL) /* Are any bits zero? */
183 return result + size; /* Nope. */
184 found_middle:
185 return result + __reverse_ffz(tmp);
188 void register_inmem_page(struct inode *inode, struct page *page)
190 struct f2fs_inode_info *fi = F2FS_I(inode);
191 struct inmem_pages *new;
193 f2fs_trace_pid(page);
195 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 SetPagePrivate(page);
198 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
200 /* add atomic page indices to the list */
201 new->page = page;
202 INIT_LIST_HEAD(&new->list);
204 /* increase reference count with clean state */
205 mutex_lock(&fi->inmem_lock);
206 get_page(page);
207 list_add_tail(&new->list, &fi->inmem_pages);
208 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
209 mutex_unlock(&fi->inmem_lock);
211 trace_f2fs_register_inmem_page(page, INMEM);
214 int commit_inmem_pages(struct inode *inode, bool abort)
216 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
217 struct f2fs_inode_info *fi = F2FS_I(inode);
218 struct inmem_pages *cur, *tmp;
219 bool submit_bio = false;
220 struct f2fs_io_info fio = {
221 .sbi = sbi,
222 .type = DATA,
223 .rw = WRITE_SYNC | REQ_PRIO,
224 .encrypted_page = NULL,
226 int err = 0;
229 * The abort is true only when f2fs_evict_inode is called.
230 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
231 * that we don't need to call f2fs_balance_fs.
232 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
233 * inode becomes free by iget_locked in f2fs_iget.
235 if (!abort) {
236 f2fs_balance_fs(sbi);
237 f2fs_lock_op(sbi);
240 mutex_lock(&fi->inmem_lock);
241 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
242 lock_page(cur->page);
243 if (!abort) {
244 if (cur->page->mapping == inode->i_mapping) {
245 set_page_dirty(cur->page);
246 f2fs_wait_on_page_writeback(cur->page, DATA);
247 if (clear_page_dirty_for_io(cur->page))
248 inode_dec_dirty_pages(inode);
249 trace_f2fs_commit_inmem_page(cur->page, INMEM);
250 fio.page = cur->page;
251 err = do_write_data_page(&fio);
252 if (err) {
253 unlock_page(cur->page);
254 break;
256 clear_cold_data(cur->page);
257 submit_bio = true;
259 } else {
260 trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
262 set_page_private(cur->page, 0);
263 ClearPagePrivate(cur->page);
264 f2fs_put_page(cur->page, 1);
266 list_del(&cur->list);
267 kmem_cache_free(inmem_entry_slab, cur);
268 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
270 mutex_unlock(&fi->inmem_lock);
272 if (!abort) {
273 f2fs_unlock_op(sbi);
274 if (submit_bio)
275 f2fs_submit_merged_bio(sbi, DATA, WRITE);
277 return err;
281 * This function balances dirty node and dentry pages.
282 * In addition, it controls garbage collection.
284 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
287 * We should do GC or end up with checkpoint, if there are so many dirty
288 * dir/node pages without enough free segments.
290 if (has_not_enough_free_secs(sbi, 0)) {
291 mutex_lock(&sbi->gc_mutex);
292 f2fs_gc(sbi, false);
296 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
298 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
299 return;
301 /* try to shrink extent cache when there is no enough memory */
302 if (!available_free_memory(sbi, EXTENT_CACHE))
303 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
305 /* check the # of cached NAT entries */
306 if (!available_free_memory(sbi, NAT_ENTRIES))
307 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
309 if (!available_free_memory(sbi, FREE_NIDS))
310 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
312 /* checkpoint is the only way to shrink partial cached entries */
313 if (!available_free_memory(sbi, NAT_ENTRIES) ||
314 excess_prefree_segs(sbi) ||
315 !available_free_memory(sbi, INO_ENTRIES) ||
316 jiffies > sbi->cp_expires)
317 f2fs_sync_fs(sbi->sb, true);
320 static int issue_flush_thread(void *data)
322 struct f2fs_sb_info *sbi = data;
323 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
324 wait_queue_head_t *q = &fcc->flush_wait_queue;
325 repeat:
326 if (kthread_should_stop())
327 return 0;
329 if (!llist_empty(&fcc->issue_list)) {
330 struct bio *bio;
331 struct flush_cmd *cmd, *next;
332 int ret;
334 bio = f2fs_bio_alloc(0);
336 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
337 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
339 bio->bi_bdev = sbi->sb->s_bdev;
340 ret = submit_bio_wait(WRITE_FLUSH, bio);
342 llist_for_each_entry_safe(cmd, next,
343 fcc->dispatch_list, llnode) {
344 cmd->ret = ret;
345 complete(&cmd->wait);
347 bio_put(bio);
348 fcc->dispatch_list = NULL;
351 wait_event_interruptible(*q,
352 kthread_should_stop() || !llist_empty(&fcc->issue_list));
353 goto repeat;
356 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
358 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
359 struct flush_cmd cmd;
361 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
362 test_opt(sbi, FLUSH_MERGE));
364 if (test_opt(sbi, NOBARRIER))
365 return 0;
367 if (!test_opt(sbi, FLUSH_MERGE)) {
368 struct bio *bio = f2fs_bio_alloc(0);
369 int ret;
371 bio->bi_bdev = sbi->sb->s_bdev;
372 ret = submit_bio_wait(WRITE_FLUSH, bio);
373 bio_put(bio);
374 return ret;
377 init_completion(&cmd.wait);
379 llist_add(&cmd.llnode, &fcc->issue_list);
381 if (!fcc->dispatch_list)
382 wake_up(&fcc->flush_wait_queue);
384 wait_for_completion(&cmd.wait);
386 return cmd.ret;
389 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
391 dev_t dev = sbi->sb->s_bdev->bd_dev;
392 struct flush_cmd_control *fcc;
393 int err = 0;
395 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
396 if (!fcc)
397 return -ENOMEM;
398 init_waitqueue_head(&fcc->flush_wait_queue);
399 init_llist_head(&fcc->issue_list);
400 SM_I(sbi)->cmd_control_info = fcc;
401 if (!test_opt(sbi, FLUSH_MERGE))
402 return err;
404 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
405 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
406 if (IS_ERR(fcc->f2fs_issue_flush)) {
407 err = PTR_ERR(fcc->f2fs_issue_flush);
408 kfree(fcc);
409 SM_I(sbi)->cmd_control_info = NULL;
410 return err;
413 return err;
416 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
418 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
420 if (fcc && fcc->f2fs_issue_flush)
421 kthread_stop(fcc->f2fs_issue_flush);
422 kfree(fcc);
423 SM_I(sbi)->cmd_control_info = NULL;
426 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
427 enum dirty_type dirty_type)
429 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
431 /* need not be added */
432 if (IS_CURSEG(sbi, segno))
433 return;
435 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
436 dirty_i->nr_dirty[dirty_type]++;
438 if (dirty_type == DIRTY) {
439 struct seg_entry *sentry = get_seg_entry(sbi, segno);
440 enum dirty_type t = sentry->type;
442 if (unlikely(t >= DIRTY)) {
443 f2fs_bug_on(sbi, 1);
444 return;
446 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
447 dirty_i->nr_dirty[t]++;
451 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
452 enum dirty_type dirty_type)
454 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
456 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
457 dirty_i->nr_dirty[dirty_type]--;
459 if (dirty_type == DIRTY) {
460 struct seg_entry *sentry = get_seg_entry(sbi, segno);
461 enum dirty_type t = sentry->type;
463 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
464 dirty_i->nr_dirty[t]--;
466 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
467 clear_bit(GET_SECNO(sbi, segno),
468 dirty_i->victim_secmap);
473 * Should not occur error such as -ENOMEM.
474 * Adding dirty entry into seglist is not critical operation.
475 * If a given segment is one of current working segments, it won't be added.
477 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
479 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
480 unsigned short valid_blocks;
482 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
483 return;
485 mutex_lock(&dirty_i->seglist_lock);
487 valid_blocks = get_valid_blocks(sbi, segno, 0);
489 if (valid_blocks == 0) {
490 __locate_dirty_segment(sbi, segno, PRE);
491 __remove_dirty_segment(sbi, segno, DIRTY);
492 } else if (valid_blocks < sbi->blocks_per_seg) {
493 __locate_dirty_segment(sbi, segno, DIRTY);
494 } else {
495 /* Recovery routine with SSR needs this */
496 __remove_dirty_segment(sbi, segno, DIRTY);
499 mutex_unlock(&dirty_i->seglist_lock);
502 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
503 block_t blkstart, block_t blklen)
505 sector_t start = SECTOR_FROM_BLOCK(blkstart);
506 sector_t len = SECTOR_FROM_BLOCK(blklen);
507 struct seg_entry *se;
508 unsigned int offset;
509 block_t i;
511 for (i = blkstart; i < blkstart + blklen; i++) {
512 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
513 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
515 if (!f2fs_test_and_set_bit(offset, se->discard_map))
516 sbi->discard_blks--;
518 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
519 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
522 static void __add_discard_entry(struct f2fs_sb_info *sbi,
523 struct cp_control *cpc, struct seg_entry *se,
524 unsigned int start, unsigned int end)
526 struct list_head *head = &SM_I(sbi)->discard_list;
527 struct discard_entry *new, *last;
529 if (!list_empty(head)) {
530 last = list_last_entry(head, struct discard_entry, list);
531 if (START_BLOCK(sbi, cpc->trim_start) + start ==
532 last->blkaddr + last->len) {
533 last->len += end - start;
534 goto done;
538 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
539 INIT_LIST_HEAD(&new->list);
540 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
541 new->len = end - start;
542 list_add_tail(&new->list, head);
543 done:
544 SM_I(sbi)->nr_discards += end - start;
547 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
549 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
550 int max_blocks = sbi->blocks_per_seg;
551 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
552 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
553 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
554 unsigned long *discard_map = (unsigned long *)se->discard_map;
555 unsigned long *dmap = SIT_I(sbi)->tmp_map;
556 unsigned int start = 0, end = -1;
557 bool force = (cpc->reason == CP_DISCARD);
558 int i;
560 if (se->valid_blocks == max_blocks)
561 return;
563 if (!force) {
564 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
565 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
566 return;
569 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
570 for (i = 0; i < entries; i++)
571 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
572 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
574 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
575 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
576 if (start >= max_blocks)
577 break;
579 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
580 __add_discard_entry(sbi, cpc, se, start, end);
584 void release_discard_addrs(struct f2fs_sb_info *sbi)
586 struct list_head *head = &(SM_I(sbi)->discard_list);
587 struct discard_entry *entry, *this;
589 /* drop caches */
590 list_for_each_entry_safe(entry, this, head, list) {
591 list_del(&entry->list);
592 kmem_cache_free(discard_entry_slab, entry);
597 * Should call clear_prefree_segments after checkpoint is done.
599 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
601 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
602 unsigned int segno;
604 mutex_lock(&dirty_i->seglist_lock);
605 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
606 __set_test_and_free(sbi, segno);
607 mutex_unlock(&dirty_i->seglist_lock);
610 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
612 struct list_head *head = &(SM_I(sbi)->discard_list);
613 struct discard_entry *entry, *this;
614 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
615 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
616 unsigned int start = 0, end = -1;
618 mutex_lock(&dirty_i->seglist_lock);
620 while (1) {
621 int i;
622 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
623 if (start >= MAIN_SEGS(sbi))
624 break;
625 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
626 start + 1);
628 for (i = start; i < end; i++)
629 clear_bit(i, prefree_map);
631 dirty_i->nr_dirty[PRE] -= end - start;
633 if (!test_opt(sbi, DISCARD))
634 continue;
636 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
637 (end - start) << sbi->log_blocks_per_seg);
639 mutex_unlock(&dirty_i->seglist_lock);
641 /* send small discards */
642 list_for_each_entry_safe(entry, this, head, list) {
643 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
644 goto skip;
645 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
646 cpc->trimmed += entry->len;
647 skip:
648 list_del(&entry->list);
649 SM_I(sbi)->nr_discards -= entry->len;
650 kmem_cache_free(discard_entry_slab, entry);
654 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
656 struct sit_info *sit_i = SIT_I(sbi);
658 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
659 sit_i->dirty_sentries++;
660 return false;
663 return true;
666 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
667 unsigned int segno, int modified)
669 struct seg_entry *se = get_seg_entry(sbi, segno);
670 se->type = type;
671 if (modified)
672 __mark_sit_entry_dirty(sbi, segno);
675 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
677 struct seg_entry *se;
678 unsigned int segno, offset;
679 long int new_vblocks;
681 segno = GET_SEGNO(sbi, blkaddr);
683 se = get_seg_entry(sbi, segno);
684 new_vblocks = se->valid_blocks + del;
685 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
687 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
688 (new_vblocks > sbi->blocks_per_seg)));
690 se->valid_blocks = new_vblocks;
691 se->mtime = get_mtime(sbi);
692 SIT_I(sbi)->max_mtime = se->mtime;
694 /* Update valid block bitmap */
695 if (del > 0) {
696 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
697 f2fs_bug_on(sbi, 1);
698 if (!f2fs_test_and_set_bit(offset, se->discard_map))
699 sbi->discard_blks--;
700 } else {
701 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
702 f2fs_bug_on(sbi, 1);
703 if (f2fs_test_and_clear_bit(offset, se->discard_map))
704 sbi->discard_blks++;
706 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
707 se->ckpt_valid_blocks += del;
709 __mark_sit_entry_dirty(sbi, segno);
711 /* update total number of valid blocks to be written in ckpt area */
712 SIT_I(sbi)->written_valid_blocks += del;
714 if (sbi->segs_per_sec > 1)
715 get_sec_entry(sbi, segno)->valid_blocks += del;
718 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
720 update_sit_entry(sbi, new, 1);
721 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
722 update_sit_entry(sbi, old, -1);
724 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
725 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
728 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
730 unsigned int segno = GET_SEGNO(sbi, addr);
731 struct sit_info *sit_i = SIT_I(sbi);
733 f2fs_bug_on(sbi, addr == NULL_ADDR);
734 if (addr == NEW_ADDR)
735 return;
737 /* add it into sit main buffer */
738 mutex_lock(&sit_i->sentry_lock);
740 update_sit_entry(sbi, addr, -1);
742 /* add it into dirty seglist */
743 locate_dirty_segment(sbi, segno);
745 mutex_unlock(&sit_i->sentry_lock);
748 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
750 struct sit_info *sit_i = SIT_I(sbi);
751 unsigned int segno, offset;
752 struct seg_entry *se;
753 bool is_cp = false;
755 if (!is_valid_data_blkaddr(sbi, blkaddr))
756 return true;
758 mutex_lock(&sit_i->sentry_lock);
760 segno = GET_SEGNO(sbi, blkaddr);
761 se = get_seg_entry(sbi, segno);
762 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
764 if (f2fs_test_bit(offset, se->ckpt_valid_map))
765 is_cp = true;
767 mutex_unlock(&sit_i->sentry_lock);
769 return is_cp;
773 * This function should be resided under the curseg_mutex lock
775 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
776 struct f2fs_summary *sum)
778 struct curseg_info *curseg = CURSEG_I(sbi, type);
779 void *addr = curseg->sum_blk;
780 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
781 memcpy(addr, sum, sizeof(struct f2fs_summary));
785 * Calculate the number of current summary pages for writing
787 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
789 int valid_sum_count = 0;
790 int i, sum_in_page;
792 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
793 if (sbi->ckpt->alloc_type[i] == SSR)
794 valid_sum_count += sbi->blocks_per_seg;
795 else {
796 if (for_ra)
797 valid_sum_count += le16_to_cpu(
798 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
799 else
800 valid_sum_count += curseg_blkoff(sbi, i);
804 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
805 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
806 if (valid_sum_count <= sum_in_page)
807 return 1;
808 else if ((valid_sum_count - sum_in_page) <=
809 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
810 return 2;
811 return 3;
815 * Caller should put this summary page
817 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
819 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
822 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
824 struct page *page = grab_meta_page(sbi, blk_addr);
825 void *dst = page_address(page);
827 if (src)
828 memcpy(dst, src, PAGE_CACHE_SIZE);
829 else
830 memset(dst, 0, PAGE_CACHE_SIZE);
831 set_page_dirty(page);
832 f2fs_put_page(page, 1);
835 static void write_sum_page(struct f2fs_sb_info *sbi,
836 struct f2fs_summary_block *sum_blk, block_t blk_addr)
838 update_meta_page(sbi, (void *)sum_blk, blk_addr);
841 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
843 struct curseg_info *curseg = CURSEG_I(sbi, type);
844 unsigned int segno = curseg->segno + 1;
845 struct free_segmap_info *free_i = FREE_I(sbi);
847 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
848 return !test_bit(segno, free_i->free_segmap);
849 return 0;
853 * Find a new segment from the free segments bitmap to right order
854 * This function should be returned with success, otherwise BUG
856 static void get_new_segment(struct f2fs_sb_info *sbi,
857 unsigned int *newseg, bool new_sec, int dir)
859 struct free_segmap_info *free_i = FREE_I(sbi);
860 unsigned int segno, secno, zoneno;
861 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
862 unsigned int hint = *newseg / sbi->segs_per_sec;
863 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
864 unsigned int left_start = hint;
865 bool init = true;
866 int go_left = 0;
867 int i;
869 spin_lock(&free_i->segmap_lock);
871 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
872 segno = find_next_zero_bit(free_i->free_segmap,
873 MAIN_SEGS(sbi), *newseg + 1);
874 if (segno - *newseg < sbi->segs_per_sec -
875 (*newseg % sbi->segs_per_sec))
876 goto got_it;
878 find_other_zone:
879 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
880 if (secno >= MAIN_SECS(sbi)) {
881 if (dir == ALLOC_RIGHT) {
882 secno = find_next_zero_bit(free_i->free_secmap,
883 MAIN_SECS(sbi), 0);
884 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
885 } else {
886 go_left = 1;
887 left_start = hint - 1;
890 if (go_left == 0)
891 goto skip_left;
893 while (test_bit(left_start, free_i->free_secmap)) {
894 if (left_start > 0) {
895 left_start--;
896 continue;
898 left_start = find_next_zero_bit(free_i->free_secmap,
899 MAIN_SECS(sbi), 0);
900 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
901 break;
903 secno = left_start;
904 skip_left:
905 hint = secno;
906 segno = secno * sbi->segs_per_sec;
907 zoneno = secno / sbi->secs_per_zone;
909 /* give up on finding another zone */
910 if (!init)
911 goto got_it;
912 if (sbi->secs_per_zone == 1)
913 goto got_it;
914 if (zoneno == old_zoneno)
915 goto got_it;
916 if (dir == ALLOC_LEFT) {
917 if (!go_left && zoneno + 1 >= total_zones)
918 goto got_it;
919 if (go_left && zoneno == 0)
920 goto got_it;
922 for (i = 0; i < NR_CURSEG_TYPE; i++)
923 if (CURSEG_I(sbi, i)->zone == zoneno)
924 break;
926 if (i < NR_CURSEG_TYPE) {
927 /* zone is in user, try another */
928 if (go_left)
929 hint = zoneno * sbi->secs_per_zone - 1;
930 else if (zoneno + 1 >= total_zones)
931 hint = 0;
932 else
933 hint = (zoneno + 1) * sbi->secs_per_zone;
934 init = false;
935 goto find_other_zone;
937 got_it:
938 /* set it as dirty segment in free segmap */
939 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
940 __set_inuse(sbi, segno);
941 *newseg = segno;
942 spin_unlock(&free_i->segmap_lock);
945 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
947 struct curseg_info *curseg = CURSEG_I(sbi, type);
948 struct summary_footer *sum_footer;
950 curseg->segno = curseg->next_segno;
951 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
952 curseg->next_blkoff = 0;
953 curseg->next_segno = NULL_SEGNO;
955 sum_footer = &(curseg->sum_blk->footer);
956 memset(sum_footer, 0, sizeof(struct summary_footer));
957 if (IS_DATASEG(type))
958 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
959 if (IS_NODESEG(type))
960 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
961 __set_sit_entry_type(sbi, type, curseg->segno, modified);
965 * Allocate a current working segment.
966 * This function always allocates a free segment in LFS manner.
968 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
970 struct curseg_info *curseg = CURSEG_I(sbi, type);
971 unsigned int segno = curseg->segno;
972 int dir = ALLOC_LEFT;
974 write_sum_page(sbi, curseg->sum_blk,
975 GET_SUM_BLOCK(sbi, segno));
976 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
977 dir = ALLOC_RIGHT;
979 if (test_opt(sbi, NOHEAP))
980 dir = ALLOC_RIGHT;
982 get_new_segment(sbi, &segno, new_sec, dir);
983 curseg->next_segno = segno;
984 reset_curseg(sbi, type, 1);
985 curseg->alloc_type = LFS;
988 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
989 struct curseg_info *seg, block_t start)
991 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
992 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
993 unsigned long *target_map = SIT_I(sbi)->tmp_map;
994 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
995 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
996 int i, pos;
998 for (i = 0; i < entries; i++)
999 target_map[i] = ckpt_map[i] | cur_map[i];
1001 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1003 seg->next_blkoff = pos;
1007 * If a segment is written by LFS manner, next block offset is just obtained
1008 * by increasing the current block offset. However, if a segment is written by
1009 * SSR manner, next block offset obtained by calling __next_free_blkoff
1011 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1012 struct curseg_info *seg)
1014 if (seg->alloc_type == SSR)
1015 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1016 else
1017 seg->next_blkoff++;
1021 * This function always allocates a used segment(from dirty seglist) by SSR
1022 * manner, so it should recover the existing segment information of valid blocks
1024 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1026 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1027 struct curseg_info *curseg = CURSEG_I(sbi, type);
1028 unsigned int new_segno = curseg->next_segno;
1029 struct f2fs_summary_block *sum_node;
1030 struct page *sum_page;
1032 write_sum_page(sbi, curseg->sum_blk,
1033 GET_SUM_BLOCK(sbi, curseg->segno));
1034 __set_test_and_inuse(sbi, new_segno);
1036 mutex_lock(&dirty_i->seglist_lock);
1037 __remove_dirty_segment(sbi, new_segno, PRE);
1038 __remove_dirty_segment(sbi, new_segno, DIRTY);
1039 mutex_unlock(&dirty_i->seglist_lock);
1041 reset_curseg(sbi, type, 1);
1042 curseg->alloc_type = SSR;
1043 __next_free_blkoff(sbi, curseg, 0);
1045 if (reuse) {
1046 sum_page = get_sum_page(sbi, new_segno);
1047 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1048 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1049 f2fs_put_page(sum_page, 1);
1053 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1055 struct curseg_info *curseg = CURSEG_I(sbi, type);
1056 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1058 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1059 return v_ops->get_victim(sbi,
1060 &(curseg)->next_segno, BG_GC, type, SSR);
1062 /* For data segments, let's do SSR more intensively */
1063 for (; type >= CURSEG_HOT_DATA; type--)
1064 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1065 BG_GC, type, SSR))
1066 return 1;
1067 return 0;
1071 * flush out current segment and replace it with new segment
1072 * This function should be returned with success, otherwise BUG
1074 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1075 int type, bool force)
1077 struct curseg_info *curseg = CURSEG_I(sbi, type);
1079 if (force)
1080 new_curseg(sbi, type, true);
1081 else if (type == CURSEG_WARM_NODE)
1082 new_curseg(sbi, type, false);
1083 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1084 new_curseg(sbi, type, false);
1085 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1086 change_curseg(sbi, type, true);
1087 else
1088 new_curseg(sbi, type, false);
1090 stat_inc_seg_type(sbi, curseg);
1093 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1095 struct curseg_info *curseg = CURSEG_I(sbi, type);
1096 unsigned int old_segno;
1098 old_segno = curseg->segno;
1099 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1100 locate_dirty_segment(sbi, old_segno);
1103 void allocate_new_segments(struct f2fs_sb_info *sbi)
1105 int i;
1107 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1108 __allocate_new_segments(sbi, i);
1111 static const struct segment_allocation default_salloc_ops = {
1112 .allocate_segment = allocate_segment_by_default,
1115 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1117 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1118 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1119 unsigned int start_segno, end_segno;
1120 struct cp_control cpc;
1122 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1123 return -EINVAL;
1125 cpc.trimmed = 0;
1126 if (end <= MAIN_BLKADDR(sbi))
1127 goto out;
1129 /* start/end segment number in main_area */
1130 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1131 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1132 GET_SEGNO(sbi, end);
1133 cpc.reason = CP_DISCARD;
1134 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1136 /* do checkpoint to issue discard commands safely */
1137 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1138 cpc.trim_start = start_segno;
1140 if (sbi->discard_blks == 0)
1141 break;
1142 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1143 cpc.trim_end = end_segno;
1144 else
1145 cpc.trim_end = min_t(unsigned int,
1146 rounddown(start_segno +
1147 BATCHED_TRIM_SEGMENTS(sbi),
1148 sbi->segs_per_sec) - 1, end_segno);
1150 mutex_lock(&sbi->gc_mutex);
1151 write_checkpoint(sbi, &cpc);
1152 mutex_unlock(&sbi->gc_mutex);
1154 out:
1155 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1156 return 0;
1159 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1161 struct curseg_info *curseg = CURSEG_I(sbi, type);
1162 if (curseg->next_blkoff < sbi->blocks_per_seg)
1163 return true;
1164 return false;
1167 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1169 if (p_type == DATA)
1170 return CURSEG_HOT_DATA;
1171 else
1172 return CURSEG_HOT_NODE;
1175 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1177 if (p_type == DATA) {
1178 struct inode *inode = page->mapping->host;
1180 if (S_ISDIR(inode->i_mode))
1181 return CURSEG_HOT_DATA;
1182 else
1183 return CURSEG_COLD_DATA;
1184 } else {
1185 if (IS_DNODE(page) && is_cold_node(page))
1186 return CURSEG_WARM_NODE;
1187 else
1188 return CURSEG_COLD_NODE;
1192 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1194 if (p_type == DATA) {
1195 struct inode *inode = page->mapping->host;
1197 if (S_ISDIR(inode->i_mode))
1198 return CURSEG_HOT_DATA;
1199 else if (is_cold_data(page) || file_is_cold(inode))
1200 return CURSEG_COLD_DATA;
1201 else
1202 return CURSEG_WARM_DATA;
1203 } else {
1204 if (IS_DNODE(page))
1205 return is_cold_node(page) ? CURSEG_WARM_NODE :
1206 CURSEG_HOT_NODE;
1207 else
1208 return CURSEG_COLD_NODE;
1212 static int __get_segment_type(struct page *page, enum page_type p_type)
1214 switch (F2FS_P_SB(page)->active_logs) {
1215 case 2:
1216 return __get_segment_type_2(page, p_type);
1217 case 4:
1218 return __get_segment_type_4(page, p_type);
1220 /* NR_CURSEG_TYPE(6) logs by default */
1221 f2fs_bug_on(F2FS_P_SB(page),
1222 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1223 return __get_segment_type_6(page, p_type);
1226 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1227 block_t old_blkaddr, block_t *new_blkaddr,
1228 struct f2fs_summary *sum, int type)
1230 struct sit_info *sit_i = SIT_I(sbi);
1231 struct curseg_info *curseg;
1232 bool direct_io = (type == CURSEG_DIRECT_IO);
1234 type = direct_io ? CURSEG_WARM_DATA : type;
1236 curseg = CURSEG_I(sbi, type);
1238 mutex_lock(&curseg->curseg_mutex);
1239 mutex_lock(&sit_i->sentry_lock);
1241 /* direct_io'ed data is aligned to the segment for better performance */
1242 if (direct_io && curseg->next_blkoff &&
1243 !has_not_enough_free_secs(sbi, 0))
1244 __allocate_new_segments(sbi, type);
1246 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1249 * __add_sum_entry should be resided under the curseg_mutex
1250 * because, this function updates a summary entry in the
1251 * current summary block.
1253 __add_sum_entry(sbi, type, sum);
1255 __refresh_next_blkoff(sbi, curseg);
1257 stat_inc_block_count(sbi, curseg);
1259 if (!__has_curseg_space(sbi, type))
1260 sit_i->s_ops->allocate_segment(sbi, type, false);
1262 * SIT information should be updated before segment allocation,
1263 * since SSR needs latest valid block information.
1265 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1267 mutex_unlock(&sit_i->sentry_lock);
1269 if (page && IS_NODESEG(type))
1270 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1272 mutex_unlock(&curseg->curseg_mutex);
1275 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1277 int type = __get_segment_type(fio->page, fio->type);
1279 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1280 &fio->blk_addr, sum, type);
1282 /* writeout dirty page into bdev */
1283 f2fs_submit_page_mbio(fio);
1286 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1288 struct f2fs_io_info fio = {
1289 .sbi = sbi,
1290 .type = META,
1291 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1292 .blk_addr = page->index,
1293 .page = page,
1294 .encrypted_page = NULL,
1297 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1298 fio.rw &= ~REQ_META;
1300 set_page_writeback(page);
1301 f2fs_submit_page_mbio(&fio);
1304 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1306 struct f2fs_summary sum;
1308 set_summary(&sum, nid, 0, 0);
1309 do_write_page(&sum, fio);
1312 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1314 struct f2fs_sb_info *sbi = fio->sbi;
1315 struct f2fs_summary sum;
1316 struct node_info ni;
1318 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1319 get_node_info(sbi, dn->nid, &ni);
1320 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1321 do_write_page(&sum, fio);
1322 dn->data_blkaddr = fio->blk_addr;
1325 void rewrite_data_page(struct f2fs_io_info *fio)
1327 stat_inc_inplace_blocks(fio->sbi);
1328 f2fs_submit_page_mbio(fio);
1331 static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1332 struct f2fs_summary *sum,
1333 block_t old_blkaddr, block_t new_blkaddr,
1334 bool recover_curseg)
1336 struct sit_info *sit_i = SIT_I(sbi);
1337 struct curseg_info *curseg;
1338 unsigned int segno, old_cursegno;
1339 struct seg_entry *se;
1340 int type;
1341 unsigned short old_blkoff;
1343 segno = GET_SEGNO(sbi, new_blkaddr);
1344 se = get_seg_entry(sbi, segno);
1345 type = se->type;
1347 if (!recover_curseg) {
1348 /* for recovery flow */
1349 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1350 if (old_blkaddr == NULL_ADDR)
1351 type = CURSEG_COLD_DATA;
1352 else
1353 type = CURSEG_WARM_DATA;
1355 } else {
1356 if (!IS_CURSEG(sbi, segno))
1357 type = CURSEG_WARM_DATA;
1360 curseg = CURSEG_I(sbi, type);
1362 mutex_lock(&curseg->curseg_mutex);
1363 mutex_lock(&sit_i->sentry_lock);
1365 old_cursegno = curseg->segno;
1366 old_blkoff = curseg->next_blkoff;
1368 /* change the current segment */
1369 if (segno != curseg->segno) {
1370 curseg->next_segno = segno;
1371 change_curseg(sbi, type, true);
1374 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1375 __add_sum_entry(sbi, type, sum);
1377 if (!recover_curseg)
1378 update_sit_entry(sbi, new_blkaddr, 1);
1379 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1380 update_sit_entry(sbi, old_blkaddr, -1);
1382 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1383 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1385 locate_dirty_segment(sbi, old_cursegno);
1387 if (recover_curseg) {
1388 if (old_cursegno != curseg->segno) {
1389 curseg->next_segno = old_cursegno;
1390 change_curseg(sbi, type, true);
1392 curseg->next_blkoff = old_blkoff;
1395 mutex_unlock(&sit_i->sentry_lock);
1396 mutex_unlock(&curseg->curseg_mutex);
1399 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1400 block_t old_addr, block_t new_addr,
1401 unsigned char version, bool recover_curseg)
1403 struct f2fs_summary sum;
1405 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1407 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1409 dn->data_blkaddr = new_addr;
1410 set_data_blkaddr(dn);
1411 f2fs_update_extent_cache(dn);
1414 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1415 struct page *page, enum page_type type)
1417 enum page_type btype = PAGE_TYPE_OF_BIO(type);
1418 struct f2fs_bio_info *io = &sbi->write_io[btype];
1419 struct bio_vec *bvec;
1420 struct page *target;
1421 int i;
1423 down_read(&io->io_rwsem);
1424 if (!io->bio) {
1425 up_read(&io->io_rwsem);
1426 return false;
1429 bio_for_each_segment_all(bvec, io->bio, i) {
1431 if (bvec->bv_page->mapping) {
1432 target = bvec->bv_page;
1433 } else {
1434 struct f2fs_crypto_ctx *ctx;
1436 /* encrypted page */
1437 ctx = (struct f2fs_crypto_ctx *)page_private(
1438 bvec->bv_page);
1439 target = ctx->w.control_page;
1442 if (page == target) {
1443 up_read(&io->io_rwsem);
1444 return true;
1448 up_read(&io->io_rwsem);
1449 return false;
1452 void f2fs_wait_on_page_writeback(struct page *page,
1453 enum page_type type)
1455 if (PageWriteback(page)) {
1456 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1458 if (is_merged_page(sbi, page, type))
1459 f2fs_submit_merged_bio(sbi, type, WRITE);
1460 wait_on_page_writeback(page);
1464 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1465 block_t blkaddr)
1467 struct page *cpage;
1469 if (!is_valid_data_blkaddr(sbi, blkaddr))
1470 return;
1472 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1474 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1475 if (cpage) {
1476 f2fs_wait_on_page_writeback(cpage, DATA);
1477 f2fs_put_page(cpage, 1);
1481 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1483 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1484 struct curseg_info *seg_i;
1485 unsigned char *kaddr;
1486 struct page *page;
1487 block_t start;
1488 int i, j, offset;
1490 start = start_sum_block(sbi);
1492 page = get_meta_page(sbi, start++);
1493 kaddr = (unsigned char *)page_address(page);
1495 /* Step 1: restore nat cache */
1496 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1497 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1499 /* Step 2: restore sit cache */
1500 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1501 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1502 SUM_JOURNAL_SIZE);
1503 offset = 2 * SUM_JOURNAL_SIZE;
1505 /* Step 3: restore summary entries */
1506 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1507 unsigned short blk_off;
1508 unsigned int segno;
1510 seg_i = CURSEG_I(sbi, i);
1511 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1512 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1513 seg_i->next_segno = segno;
1514 reset_curseg(sbi, i, 0);
1515 seg_i->alloc_type = ckpt->alloc_type[i];
1516 seg_i->next_blkoff = blk_off;
1518 if (seg_i->alloc_type == SSR)
1519 blk_off = sbi->blocks_per_seg;
1521 for (j = 0; j < blk_off; j++) {
1522 struct f2fs_summary *s;
1523 s = (struct f2fs_summary *)(kaddr + offset);
1524 seg_i->sum_blk->entries[j] = *s;
1525 offset += SUMMARY_SIZE;
1526 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1527 SUM_FOOTER_SIZE)
1528 continue;
1530 f2fs_put_page(page, 1);
1531 page = NULL;
1533 page = get_meta_page(sbi, start++);
1534 kaddr = (unsigned char *)page_address(page);
1535 offset = 0;
1538 f2fs_put_page(page, 1);
1539 return 0;
1542 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1544 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1545 struct f2fs_summary_block *sum;
1546 struct curseg_info *curseg;
1547 struct page *new;
1548 unsigned short blk_off;
1549 unsigned int segno = 0;
1550 block_t blk_addr = 0;
1552 /* get segment number and block addr */
1553 if (IS_DATASEG(type)) {
1554 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1555 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1556 CURSEG_HOT_DATA]);
1557 if (__exist_node_summaries(sbi))
1558 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1559 else
1560 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1561 } else {
1562 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1563 CURSEG_HOT_NODE]);
1564 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1565 CURSEG_HOT_NODE]);
1566 if (__exist_node_summaries(sbi))
1567 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1568 type - CURSEG_HOT_NODE);
1569 else
1570 blk_addr = GET_SUM_BLOCK(sbi, segno);
1573 new = get_meta_page(sbi, blk_addr);
1574 sum = (struct f2fs_summary_block *)page_address(new);
1576 if (IS_NODESEG(type)) {
1577 if (__exist_node_summaries(sbi)) {
1578 struct f2fs_summary *ns = &sum->entries[0];
1579 int i;
1580 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1581 ns->version = 0;
1582 ns->ofs_in_node = 0;
1584 } else {
1585 int err;
1587 err = restore_node_summary(sbi, segno, sum);
1588 if (err) {
1589 f2fs_put_page(new, 1);
1590 return err;
1595 /* set uncompleted segment to curseg */
1596 curseg = CURSEG_I(sbi, type);
1597 mutex_lock(&curseg->curseg_mutex);
1598 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1599 curseg->next_segno = segno;
1600 reset_curseg(sbi, type, 0);
1601 curseg->alloc_type = ckpt->alloc_type[type];
1602 curseg->next_blkoff = blk_off;
1603 mutex_unlock(&curseg->curseg_mutex);
1604 f2fs_put_page(new, 1);
1605 return 0;
1608 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1610 int type = CURSEG_HOT_DATA;
1611 int err;
1613 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1614 int npages = npages_for_summary_flush(sbi, true);
1616 if (npages >= 2)
1617 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1618 META_CP, true);
1620 /* restore for compacted data summary */
1621 if (read_compacted_summaries(sbi))
1622 return -EINVAL;
1623 type = CURSEG_HOT_NODE;
1626 if (__exist_node_summaries(sbi))
1627 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1628 NR_CURSEG_TYPE - type, META_CP, true);
1630 for (; type <= CURSEG_COLD_NODE; type++) {
1631 err = read_normal_summaries(sbi, type);
1632 if (err)
1633 return err;
1636 return 0;
1639 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1641 struct page *page;
1642 unsigned char *kaddr;
1643 struct f2fs_summary *summary;
1644 struct curseg_info *seg_i;
1645 int written_size = 0;
1646 int i, j;
1648 page = grab_meta_page(sbi, blkaddr++);
1649 kaddr = (unsigned char *)page_address(page);
1651 /* Step 1: write nat cache */
1652 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1653 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1654 written_size += SUM_JOURNAL_SIZE;
1656 /* Step 2: write sit cache */
1657 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1658 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1659 SUM_JOURNAL_SIZE);
1660 written_size += SUM_JOURNAL_SIZE;
1662 /* Step 3: write summary entries */
1663 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1664 unsigned short blkoff;
1665 seg_i = CURSEG_I(sbi, i);
1666 if (sbi->ckpt->alloc_type[i] == SSR)
1667 blkoff = sbi->blocks_per_seg;
1668 else
1669 blkoff = curseg_blkoff(sbi, i);
1671 for (j = 0; j < blkoff; j++) {
1672 if (!page) {
1673 page = grab_meta_page(sbi, blkaddr++);
1674 kaddr = (unsigned char *)page_address(page);
1675 written_size = 0;
1677 summary = (struct f2fs_summary *)(kaddr + written_size);
1678 *summary = seg_i->sum_blk->entries[j];
1679 written_size += SUMMARY_SIZE;
1681 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1682 SUM_FOOTER_SIZE)
1683 continue;
1685 set_page_dirty(page);
1686 f2fs_put_page(page, 1);
1687 page = NULL;
1690 if (page) {
1691 set_page_dirty(page);
1692 f2fs_put_page(page, 1);
1696 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1697 block_t blkaddr, int type)
1699 int i, end;
1700 if (IS_DATASEG(type))
1701 end = type + NR_CURSEG_DATA_TYPE;
1702 else
1703 end = type + NR_CURSEG_NODE_TYPE;
1705 for (i = type; i < end; i++) {
1706 struct curseg_info *sum = CURSEG_I(sbi, i);
1707 mutex_lock(&sum->curseg_mutex);
1708 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1709 mutex_unlock(&sum->curseg_mutex);
1713 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1715 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1716 write_compacted_summaries(sbi, start_blk);
1717 else
1718 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1721 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1723 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1726 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1727 unsigned int val, int alloc)
1729 int i;
1731 if (type == NAT_JOURNAL) {
1732 for (i = 0; i < nats_in_cursum(sum); i++) {
1733 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1734 return i;
1736 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1737 return update_nats_in_cursum(sum, 1);
1738 } else if (type == SIT_JOURNAL) {
1739 for (i = 0; i < sits_in_cursum(sum); i++)
1740 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1741 return i;
1742 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1743 return update_sits_in_cursum(sum, 1);
1745 return -1;
1748 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1749 unsigned int segno)
1751 return get_meta_page(sbi, current_sit_addr(sbi, segno));
1754 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1755 unsigned int start)
1757 struct sit_info *sit_i = SIT_I(sbi);
1758 struct page *src_page, *dst_page;
1759 pgoff_t src_off, dst_off;
1760 void *src_addr, *dst_addr;
1762 src_off = current_sit_addr(sbi, start);
1763 dst_off = next_sit_addr(sbi, src_off);
1765 /* get current sit block page without lock */
1766 src_page = get_meta_page(sbi, src_off);
1767 dst_page = grab_meta_page(sbi, dst_off);
1768 f2fs_bug_on(sbi, PageDirty(src_page));
1770 src_addr = page_address(src_page);
1771 dst_addr = page_address(dst_page);
1772 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1774 set_page_dirty(dst_page);
1775 f2fs_put_page(src_page, 1);
1777 set_to_next_sit(sit_i, start);
1779 return dst_page;
1782 static struct sit_entry_set *grab_sit_entry_set(void)
1784 struct sit_entry_set *ses =
1785 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1787 ses->entry_cnt = 0;
1788 INIT_LIST_HEAD(&ses->set_list);
1789 return ses;
1792 static void release_sit_entry_set(struct sit_entry_set *ses)
1794 list_del(&ses->set_list);
1795 kmem_cache_free(sit_entry_set_slab, ses);
1798 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1799 struct list_head *head)
1801 struct sit_entry_set *next = ses;
1803 if (list_is_last(&ses->set_list, head))
1804 return;
1806 list_for_each_entry_continue(next, head, set_list)
1807 if (ses->entry_cnt <= next->entry_cnt)
1808 break;
1810 list_move_tail(&ses->set_list, &next->set_list);
1813 static void add_sit_entry(unsigned int segno, struct list_head *head)
1815 struct sit_entry_set *ses;
1816 unsigned int start_segno = START_SEGNO(segno);
1818 list_for_each_entry(ses, head, set_list) {
1819 if (ses->start_segno == start_segno) {
1820 ses->entry_cnt++;
1821 adjust_sit_entry_set(ses, head);
1822 return;
1826 ses = grab_sit_entry_set();
1828 ses->start_segno = start_segno;
1829 ses->entry_cnt++;
1830 list_add(&ses->set_list, head);
1833 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1835 struct f2fs_sm_info *sm_info = SM_I(sbi);
1836 struct list_head *set_list = &sm_info->sit_entry_set;
1837 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1838 unsigned int segno;
1840 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1841 add_sit_entry(segno, set_list);
1844 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1846 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1847 struct f2fs_summary_block *sum = curseg->sum_blk;
1848 int i;
1850 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1851 unsigned int segno;
1852 bool dirtied;
1854 segno = le32_to_cpu(segno_in_journal(sum, i));
1855 dirtied = __mark_sit_entry_dirty(sbi, segno);
1857 if (!dirtied)
1858 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1860 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1864 * CP calls this function, which flushes SIT entries including sit_journal,
1865 * and moves prefree segs to free segs.
1867 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1869 struct sit_info *sit_i = SIT_I(sbi);
1870 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1871 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1872 struct f2fs_summary_block *sum = curseg->sum_blk;
1873 struct sit_entry_set *ses, *tmp;
1874 struct list_head *head = &SM_I(sbi)->sit_entry_set;
1875 bool to_journal = true;
1876 struct seg_entry *se;
1878 mutex_lock(&curseg->curseg_mutex);
1879 mutex_lock(&sit_i->sentry_lock);
1881 if (!sit_i->dirty_sentries)
1882 goto out;
1885 * add and account sit entries of dirty bitmap in sit entry
1886 * set temporarily
1888 add_sits_in_set(sbi);
1891 * if there are no enough space in journal to store dirty sit
1892 * entries, remove all entries from journal and add and account
1893 * them in sit entry set.
1895 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1896 remove_sits_in_journal(sbi);
1899 * there are two steps to flush sit entries:
1900 * #1, flush sit entries to journal in current cold data summary block.
1901 * #2, flush sit entries to sit page.
1903 list_for_each_entry_safe(ses, tmp, head, set_list) {
1904 struct page *page = NULL;
1905 struct f2fs_sit_block *raw_sit = NULL;
1906 unsigned int start_segno = ses->start_segno;
1907 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1908 (unsigned long)MAIN_SEGS(sbi));
1909 unsigned int segno = start_segno;
1911 if (to_journal &&
1912 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1913 to_journal = false;
1915 if (!to_journal) {
1916 page = get_next_sit_page(sbi, start_segno);
1917 raw_sit = page_address(page);
1920 /* flush dirty sit entries in region of current sit set */
1921 for_each_set_bit_from(segno, bitmap, end) {
1922 int offset, sit_offset;
1924 se = get_seg_entry(sbi, segno);
1926 /* add discard candidates */
1927 if (cpc->reason != CP_DISCARD) {
1928 cpc->trim_start = segno;
1929 add_discard_addrs(sbi, cpc);
1932 if (to_journal) {
1933 offset = lookup_journal_in_cursum(sum,
1934 SIT_JOURNAL, segno, 1);
1935 f2fs_bug_on(sbi, offset < 0);
1936 segno_in_journal(sum, offset) =
1937 cpu_to_le32(segno);
1938 seg_info_to_raw_sit(se,
1939 &sit_in_journal(sum, offset));
1940 } else {
1941 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1942 seg_info_to_raw_sit(se,
1943 &raw_sit->entries[sit_offset]);
1946 __clear_bit(segno, bitmap);
1947 sit_i->dirty_sentries--;
1948 ses->entry_cnt--;
1951 if (!to_journal)
1952 f2fs_put_page(page, 1);
1954 f2fs_bug_on(sbi, ses->entry_cnt);
1955 release_sit_entry_set(ses);
1958 f2fs_bug_on(sbi, !list_empty(head));
1959 f2fs_bug_on(sbi, sit_i->dirty_sentries);
1960 out:
1961 if (cpc->reason == CP_DISCARD) {
1962 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1963 add_discard_addrs(sbi, cpc);
1965 mutex_unlock(&sit_i->sentry_lock);
1966 mutex_unlock(&curseg->curseg_mutex);
1968 set_prefree_as_free_segments(sbi);
1971 static int build_sit_info(struct f2fs_sb_info *sbi)
1973 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1974 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1975 struct sit_info *sit_i;
1976 unsigned int sit_segs, start;
1977 char *src_bitmap, *dst_bitmap;
1978 unsigned int bitmap_size;
1980 /* allocate memory for SIT information */
1981 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1982 if (!sit_i)
1983 return -ENOMEM;
1985 SM_I(sbi)->sit_info = sit_i;
1987 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
1988 sizeof(struct seg_entry), GFP_KERNEL);
1989 if (!sit_i->sentries)
1990 return -ENOMEM;
1992 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1993 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
1994 if (!sit_i->dirty_sentries_bitmap)
1995 return -ENOMEM;
1997 for (start = 0; start < MAIN_SEGS(sbi); start++) {
1998 sit_i->sentries[start].cur_valid_map
1999 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2000 sit_i->sentries[start].ckpt_valid_map
2001 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2002 sit_i->sentries[start].discard_map
2003 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2004 if (!sit_i->sentries[start].cur_valid_map ||
2005 !sit_i->sentries[start].ckpt_valid_map ||
2006 !sit_i->sentries[start].discard_map)
2007 return -ENOMEM;
2010 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2011 if (!sit_i->tmp_map)
2012 return -ENOMEM;
2014 if (sbi->segs_per_sec > 1) {
2015 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2016 sizeof(struct sec_entry), GFP_KERNEL);
2017 if (!sit_i->sec_entries)
2018 return -ENOMEM;
2021 /* get information related with SIT */
2022 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2024 /* setup SIT bitmap from ckeckpoint pack */
2025 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2026 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2028 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2029 if (!dst_bitmap)
2030 return -ENOMEM;
2032 /* init SIT information */
2033 sit_i->s_ops = &default_salloc_ops;
2035 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2036 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2037 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2038 sit_i->sit_bitmap = dst_bitmap;
2039 sit_i->bitmap_size = bitmap_size;
2040 sit_i->dirty_sentries = 0;
2041 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2042 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2043 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2044 mutex_init(&sit_i->sentry_lock);
2045 return 0;
2048 static int build_free_segmap(struct f2fs_sb_info *sbi)
2050 struct free_segmap_info *free_i;
2051 unsigned int bitmap_size, sec_bitmap_size;
2053 /* allocate memory for free segmap information */
2054 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2055 if (!free_i)
2056 return -ENOMEM;
2058 SM_I(sbi)->free_info = free_i;
2060 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2061 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2062 if (!free_i->free_segmap)
2063 return -ENOMEM;
2065 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2066 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2067 if (!free_i->free_secmap)
2068 return -ENOMEM;
2070 /* set all segments as dirty temporarily */
2071 memset(free_i->free_segmap, 0xff, bitmap_size);
2072 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2074 /* init free segmap information */
2075 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2076 free_i->free_segments = 0;
2077 free_i->free_sections = 0;
2078 spin_lock_init(&free_i->segmap_lock);
2079 return 0;
2082 static int build_curseg(struct f2fs_sb_info *sbi)
2084 struct curseg_info *array;
2085 int i;
2087 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2088 if (!array)
2089 return -ENOMEM;
2091 SM_I(sbi)->curseg_array = array;
2093 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2094 mutex_init(&array[i].curseg_mutex);
2095 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2096 if (!array[i].sum_blk)
2097 return -ENOMEM;
2098 array[i].segno = NULL_SEGNO;
2099 array[i].next_blkoff = 0;
2101 return restore_curseg_summaries(sbi);
2104 static int build_sit_entries(struct f2fs_sb_info *sbi)
2106 struct sit_info *sit_i = SIT_I(sbi);
2107 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2108 struct f2fs_summary_block *sum = curseg->sum_blk;
2109 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2110 unsigned int i, start, end;
2111 unsigned int readed, start_blk = 0;
2112 int nrpages = MAX_BIO_BLOCKS(sbi);
2113 int err = 0;
2115 do {
2116 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2118 start = start_blk * sit_i->sents_per_block;
2119 end = (start_blk + readed) * sit_i->sents_per_block;
2121 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2122 struct seg_entry *se = &sit_i->sentries[start];
2123 struct f2fs_sit_block *sit_blk;
2124 struct f2fs_sit_entry sit;
2125 struct page *page;
2127 page = get_current_sit_page(sbi, start);
2128 sit_blk = (struct f2fs_sit_block *)page_address(page);
2129 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2130 f2fs_put_page(page, 1);
2132 err = check_block_count(sbi, start, &sit);
2133 if (err)
2134 return err;
2135 seg_info_from_raw_sit(se, &sit);
2137 /* build discard map only one time */
2138 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2139 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2141 if (sbi->segs_per_sec > 1)
2142 get_sec_entry(sbi, start)->valid_blocks +=
2143 se->valid_blocks;
2145 start_blk += readed;
2146 } while (start_blk < sit_blk_cnt);
2148 mutex_lock(&curseg->curseg_mutex);
2149 for (i = 0; i < sits_in_cursum(sum); i++) {
2150 struct f2fs_sit_entry sit;
2151 struct seg_entry *se;
2152 unsigned int old_valid_blocks;
2154 start = le32_to_cpu(segno_in_journal(sum, i));
2155 if (start >= MAIN_SEGS(sbi)) {
2156 f2fs_msg(sbi->sb, KERN_ERR,
2157 "Wrong journal entry on segno %u",
2158 start);
2159 set_sbi_flag(sbi, SBI_NEED_FSCK);
2160 err = -EINVAL;
2161 break;
2164 se = &sit_i->sentries[start];
2165 sit = sit_in_journal(sum, i);
2167 old_valid_blocks = se->valid_blocks;
2169 err = check_block_count(sbi, start, &sit);
2170 if (err)
2171 break;
2172 seg_info_from_raw_sit(se, &sit);
2174 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2175 sbi->discard_blks += old_valid_blocks - se->valid_blocks;
2177 if (sbi->segs_per_sec > 1)
2178 get_sec_entry(sbi, start)->valid_blocks +=
2179 se->valid_blocks - old_valid_blocks;
2181 mutex_unlock(&curseg->curseg_mutex);
2182 return err;
2185 static void init_free_segmap(struct f2fs_sb_info *sbi)
2187 unsigned int start;
2188 int type;
2190 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2191 struct seg_entry *sentry = get_seg_entry(sbi, start);
2192 if (!sentry->valid_blocks)
2193 __set_free(sbi, start);
2196 /* set use the current segments */
2197 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2198 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2199 __set_test_and_inuse(sbi, curseg_t->segno);
2203 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2205 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2206 struct free_segmap_info *free_i = FREE_I(sbi);
2207 unsigned int segno = 0, offset = 0;
2208 unsigned short valid_blocks;
2210 while (1) {
2211 /* find dirty segment based on free segmap */
2212 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2213 if (segno >= MAIN_SEGS(sbi))
2214 break;
2215 offset = segno + 1;
2216 valid_blocks = get_valid_blocks(sbi, segno, 0);
2217 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2218 continue;
2219 if (valid_blocks > sbi->blocks_per_seg) {
2220 f2fs_bug_on(sbi, 1);
2221 continue;
2223 mutex_lock(&dirty_i->seglist_lock);
2224 __locate_dirty_segment(sbi, segno, DIRTY);
2225 mutex_unlock(&dirty_i->seglist_lock);
2229 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2231 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2232 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2234 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2235 if (!dirty_i->victim_secmap)
2236 return -ENOMEM;
2237 return 0;
2240 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2242 struct dirty_seglist_info *dirty_i;
2243 unsigned int bitmap_size, i;
2245 /* allocate memory for dirty segments list information */
2246 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2247 if (!dirty_i)
2248 return -ENOMEM;
2250 SM_I(sbi)->dirty_info = dirty_i;
2251 mutex_init(&dirty_i->seglist_lock);
2253 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2255 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2256 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2257 if (!dirty_i->dirty_segmap[i])
2258 return -ENOMEM;
2261 init_dirty_segmap(sbi);
2262 return init_victim_secmap(sbi);
2265 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
2267 int i;
2270 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
2271 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
2273 for (i = 0; i < NO_CHECK_TYPE; i++) {
2274 struct curseg_info *curseg = CURSEG_I(sbi, i);
2275 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
2276 unsigned int blkofs = curseg->next_blkoff;
2278 if (f2fs_test_bit(blkofs, se->cur_valid_map))
2279 goto out;
2281 if (curseg->alloc_type == SSR)
2282 continue;
2284 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
2285 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
2286 continue;
2287 out:
2288 f2fs_msg(sbi->sb, KERN_ERR,
2289 "Current segment's next free block offset is "
2290 "inconsistent with bitmap, logtype:%u, "
2291 "segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
2292 i, curseg->segno, curseg->alloc_type,
2293 curseg->next_blkoff, blkofs);
2294 return -EINVAL;
2297 return 0;
2301 * Update min, max modified time for cost-benefit GC algorithm
2303 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2305 struct sit_info *sit_i = SIT_I(sbi);
2306 unsigned int segno;
2308 mutex_lock(&sit_i->sentry_lock);
2310 sit_i->min_mtime = LLONG_MAX;
2312 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2313 unsigned int i;
2314 unsigned long long mtime = 0;
2316 for (i = 0; i < sbi->segs_per_sec; i++)
2317 mtime += get_seg_entry(sbi, segno + i)->mtime;
2319 mtime = div_u64(mtime, sbi->segs_per_sec);
2321 if (sit_i->min_mtime > mtime)
2322 sit_i->min_mtime = mtime;
2324 sit_i->max_mtime = get_mtime(sbi);
2325 mutex_unlock(&sit_i->sentry_lock);
2328 int build_segment_manager(struct f2fs_sb_info *sbi)
2330 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2331 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2332 struct f2fs_sm_info *sm_info;
2333 int err;
2335 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2336 if (!sm_info)
2337 return -ENOMEM;
2339 /* init sm info */
2340 sbi->sm_info = sm_info;
2341 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2342 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2343 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2344 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2345 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2346 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2347 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2348 sm_info->rec_prefree_segments = sm_info->main_segments *
2349 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2350 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2351 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2352 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2354 INIT_LIST_HEAD(&sm_info->discard_list);
2355 sm_info->nr_discards = 0;
2356 sm_info->max_discards = 0;
2358 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2360 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2362 if (!f2fs_readonly(sbi->sb)) {
2363 err = create_flush_cmd_control(sbi);
2364 if (err)
2365 return err;
2368 err = build_sit_info(sbi);
2369 if (err)
2370 return err;
2371 err = build_free_segmap(sbi);
2372 if (err)
2373 return err;
2374 err = build_curseg(sbi);
2375 if (err)
2376 return err;
2378 /* reinit free segmap based on SIT */
2379 err = build_sit_entries(sbi);
2380 if (err)
2381 return err;
2383 init_free_segmap(sbi);
2384 err = build_dirty_segmap(sbi);
2385 if (err)
2386 return err;
2388 err = sanity_check_curseg(sbi);
2389 if (err)
2390 return err;
2392 init_min_max_mtime(sbi);
2393 return 0;
2396 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2397 enum dirty_type dirty_type)
2399 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2401 mutex_lock(&dirty_i->seglist_lock);
2402 kvfree(dirty_i->dirty_segmap[dirty_type]);
2403 dirty_i->nr_dirty[dirty_type] = 0;
2404 mutex_unlock(&dirty_i->seglist_lock);
2407 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2409 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2410 kvfree(dirty_i->victim_secmap);
2413 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2415 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2416 int i;
2418 if (!dirty_i)
2419 return;
2421 /* discard pre-free/dirty segments list */
2422 for (i = 0; i < NR_DIRTY_TYPE; i++)
2423 discard_dirty_segmap(sbi, i);
2425 destroy_victim_secmap(sbi);
2426 SM_I(sbi)->dirty_info = NULL;
2427 kfree(dirty_i);
2430 static void destroy_curseg(struct f2fs_sb_info *sbi)
2432 struct curseg_info *array = SM_I(sbi)->curseg_array;
2433 int i;
2435 if (!array)
2436 return;
2437 SM_I(sbi)->curseg_array = NULL;
2438 for (i = 0; i < NR_CURSEG_TYPE; i++)
2439 kfree(array[i].sum_blk);
2440 kfree(array);
2443 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2445 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2446 if (!free_i)
2447 return;
2448 SM_I(sbi)->free_info = NULL;
2449 kvfree(free_i->free_segmap);
2450 kvfree(free_i->free_secmap);
2451 kfree(free_i);
2454 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2456 struct sit_info *sit_i = SIT_I(sbi);
2457 unsigned int start;
2459 if (!sit_i)
2460 return;
2462 if (sit_i->sentries) {
2463 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2464 kfree(sit_i->sentries[start].cur_valid_map);
2465 kfree(sit_i->sentries[start].ckpt_valid_map);
2466 kfree(sit_i->sentries[start].discard_map);
2469 kfree(sit_i->tmp_map);
2471 kvfree(sit_i->sentries);
2472 kvfree(sit_i->sec_entries);
2473 kvfree(sit_i->dirty_sentries_bitmap);
2475 SM_I(sbi)->sit_info = NULL;
2476 kfree(sit_i->sit_bitmap);
2477 kfree(sit_i);
2480 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2482 struct f2fs_sm_info *sm_info = SM_I(sbi);
2484 if (!sm_info)
2485 return;
2486 destroy_flush_cmd_control(sbi);
2487 destroy_dirty_segmap(sbi);
2488 destroy_curseg(sbi);
2489 destroy_free_segmap(sbi);
2490 destroy_sit_info(sbi);
2491 sbi->sm_info = NULL;
2492 kfree(sm_info);
2495 int __init create_segment_manager_caches(void)
2497 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2498 sizeof(struct discard_entry));
2499 if (!discard_entry_slab)
2500 goto fail;
2502 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2503 sizeof(struct sit_entry_set));
2504 if (!sit_entry_set_slab)
2505 goto destory_discard_entry;
2507 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2508 sizeof(struct inmem_pages));
2509 if (!inmem_entry_slab)
2510 goto destroy_sit_entry_set;
2511 return 0;
2513 destroy_sit_entry_set:
2514 kmem_cache_destroy(sit_entry_set_slab);
2515 destory_discard_entry:
2516 kmem_cache_destroy(discard_entry_slab);
2517 fail:
2518 return -ENOMEM;
2521 void destroy_segment_manager_caches(void)
2523 kmem_cache_destroy(sit_entry_set_slab);
2524 kmem_cache_destroy(discard_entry_slab);
2525 kmem_cache_destroy(inmem_entry_slab);