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.
12 #include <linux/module.h>
13 #include <linux/backing-dev.h>
14 #include <linux/init.h>
15 #include <linux/f2fs_fs.h>
16 #include <linux/kthread.h>
17 #include <linux/delay.h>
18 #include <linux/freezer.h>
24 #include <trace/events/f2fs.h>
26 static int gc_thread_func(void *data)
28 struct f2fs_sb_info *sbi = data;
29 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
30 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
33 wait_ms = gc_th->min_sleep_time;
39 wait_event_interruptible_timeout(*wq,
40 kthread_should_stop(),
41 msecs_to_jiffies(wait_ms));
42 if (kthread_should_stop())
45 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
46 increase_sleep_time(gc_th, &wait_ms);
50 #ifdef CONFIG_F2FS_FAULT_INJECTION
51 if (time_to_inject(sbi, FAULT_CHECKPOINT))
52 f2fs_stop_checkpoint(sbi, false);
56 * [GC triggering condition]
57 * 0. GC is not conducted currently.
58 * 1. There are enough dirty segments.
59 * 2. IO subsystem is idle by checking the # of writeback pages.
60 * 3. IO subsystem is idle by checking the # of requests in
61 * bdev's request list.
63 * Note) We have to avoid triggering GCs frequently.
64 * Because it is possible that some segments can be
65 * invalidated soon after by user update or deletion.
66 * So, I'd like to wait some time to collect dirty segments.
68 if (!mutex_trylock(&sbi->gc_mutex))
72 increase_sleep_time(gc_th, &wait_ms);
73 mutex_unlock(&sbi->gc_mutex);
77 if (has_enough_invalid_blocks(sbi))
78 decrease_sleep_time(gc_th, &wait_ms);
80 increase_sleep_time(gc_th, &wait_ms);
82 stat_inc_bggc_count(sbi);
84 /* if return value is not zero, no victim was selected */
85 if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true))
86 wait_ms = gc_th->no_gc_sleep_time;
88 trace_f2fs_background_gc(sbi->sb, wait_ms,
89 prefree_segments(sbi), free_segments(sbi));
91 /* balancing f2fs's metadata periodically */
92 f2fs_balance_fs_bg(sbi);
94 } while (!kthread_should_stop());
98 int start_gc_thread(struct f2fs_sb_info *sbi)
100 struct f2fs_gc_kthread *gc_th;
101 dev_t dev = sbi->sb->s_bdev->bd_dev;
104 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
110 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
111 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
112 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
116 sbi->gc_thread = gc_th;
117 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
118 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
119 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
120 if (IS_ERR(gc_th->f2fs_gc_task)) {
121 err = PTR_ERR(gc_th->f2fs_gc_task);
123 sbi->gc_thread = NULL;
129 void stop_gc_thread(struct f2fs_sb_info *sbi)
131 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
134 kthread_stop(gc_th->f2fs_gc_task);
136 sbi->gc_thread = NULL;
139 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
141 int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
143 if (gc_th && gc_th->gc_idle) {
144 if (gc_th->gc_idle == 1)
146 else if (gc_th->gc_idle == 2)
152 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
153 int type, struct victim_sel_policy *p)
155 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
157 if (p->alloc_mode == SSR) {
158 p->gc_mode = GC_GREEDY;
159 p->dirty_segmap = dirty_i->dirty_segmap[type];
160 p->max_search = dirty_i->nr_dirty[type];
163 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
164 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
165 p->max_search = dirty_i->nr_dirty[DIRTY];
166 p->ofs_unit = sbi->segs_per_sec;
169 if (p->max_search > sbi->max_victim_search)
170 p->max_search = sbi->max_victim_search;
172 p->offset = sbi->last_victim[p->gc_mode];
175 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
176 struct victim_sel_policy *p)
178 /* SSR allocates in a segment unit */
179 if (p->alloc_mode == SSR)
180 return sbi->blocks_per_seg;
181 if (p->gc_mode == GC_GREEDY)
182 return sbi->blocks_per_seg * p->ofs_unit;
183 else if (p->gc_mode == GC_CB)
185 else /* No other gc_mode */
189 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
191 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
195 * If the gc_type is FG_GC, we can select victim segments
196 * selected by background GC before.
197 * Those segments guarantee they have small valid blocks.
199 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
200 if (sec_usage_check(sbi, secno))
202 clear_bit(secno, dirty_i->victim_secmap);
203 return secno * sbi->segs_per_sec;
208 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
210 struct sit_info *sit_i = SIT_I(sbi);
211 unsigned int secno = GET_SECNO(sbi, segno);
212 unsigned int start = secno * sbi->segs_per_sec;
213 unsigned long long mtime = 0;
214 unsigned int vblocks;
215 unsigned char age = 0;
219 for (i = 0; i < sbi->segs_per_sec; i++)
220 mtime += get_seg_entry(sbi, start + i)->mtime;
221 vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
223 mtime = div_u64(mtime, sbi->segs_per_sec);
224 vblocks = div_u64(vblocks, sbi->segs_per_sec);
226 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
228 /* Handle if the system time has changed by the user */
229 if (mtime < sit_i->min_mtime)
230 sit_i->min_mtime = mtime;
231 if (mtime > sit_i->max_mtime)
232 sit_i->max_mtime = mtime;
233 if (sit_i->max_mtime != sit_i->min_mtime)
234 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
235 sit_i->max_mtime - sit_i->min_mtime);
237 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
240 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
241 unsigned int segno, struct victim_sel_policy *p)
243 if (p->alloc_mode == SSR)
244 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
246 /* alloc_mode == LFS */
247 if (p->gc_mode == GC_GREEDY)
248 return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
250 return get_cb_cost(sbi, segno);
253 static unsigned int count_bits(const unsigned long *addr,
254 unsigned int offset, unsigned int len)
256 unsigned int end = offset + len, sum = 0;
258 while (offset < end) {
259 if (test_bit(offset++, addr))
266 * This function is called from two paths.
267 * One is garbage collection and the other is SSR segment selection.
268 * When it is called during GC, it just gets a victim segment
269 * and it does not remove it from dirty seglist.
270 * When it is called from SSR segment selection, it finds a segment
271 * which has minimum valid blocks and removes it from dirty seglist.
273 static int get_victim_by_default(struct f2fs_sb_info *sbi,
274 unsigned int *result, int gc_type, int type, char alloc_mode)
276 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
277 struct victim_sel_policy p;
278 unsigned int secno, last_victim;
279 unsigned int last_segment = MAIN_SEGS(sbi);
280 unsigned int nsearched = 0;
282 mutex_lock(&dirty_i->seglist_lock);
284 p.alloc_mode = alloc_mode;
285 select_policy(sbi, gc_type, type, &p);
287 p.min_segno = NULL_SEGNO;
288 p.min_cost = get_max_cost(sbi, &p);
290 if (p.max_search == 0)
293 last_victim = sbi->last_victim[p.gc_mode];
294 if (p.alloc_mode == LFS && gc_type == FG_GC) {
295 p.min_segno = check_bg_victims(sbi);
296 if (p.min_segno != NULL_SEGNO)
304 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
305 if (segno >= last_segment) {
306 if (sbi->last_victim[p.gc_mode]) {
307 last_segment = sbi->last_victim[p.gc_mode];
308 sbi->last_victim[p.gc_mode] = 0;
315 p.offset = segno + p.ofs_unit;
316 if (p.ofs_unit > 1) {
317 p.offset -= segno % p.ofs_unit;
318 nsearched += count_bits(p.dirty_segmap,
319 p.offset - p.ofs_unit,
326 secno = GET_SECNO(sbi, segno);
328 if (sec_usage_check(sbi, secno))
330 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
333 cost = get_gc_cost(sbi, segno, &p);
335 if (p.min_cost > cost) {
340 if (nsearched >= p.max_search) {
341 if (!sbi->last_victim[p.gc_mode] && segno <= last_victim)
342 sbi->last_victim[p.gc_mode] = last_victim + 1;
344 sbi->last_victim[p.gc_mode] = segno + 1;
348 if (p.min_segno != NULL_SEGNO) {
350 if (p.alloc_mode == LFS) {
351 secno = GET_SECNO(sbi, p.min_segno);
352 if (gc_type == FG_GC)
353 sbi->cur_victim_sec = secno;
355 set_bit(secno, dirty_i->victim_secmap);
357 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
359 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
361 prefree_segments(sbi), free_segments(sbi));
364 mutex_unlock(&dirty_i->seglist_lock);
366 return (p.min_segno == NULL_SEGNO) ? 0 : 1;
369 static const struct victim_selection default_v_ops = {
370 .get_victim = get_victim_by_default,
373 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
375 struct inode_entry *ie;
377 ie = radix_tree_lookup(&gc_list->iroot, ino);
383 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
385 struct inode_entry *new_ie;
387 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
391 new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
392 new_ie->inode = inode;
394 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
395 list_add_tail(&new_ie->list, &gc_list->ilist);
398 static void put_gc_inode(struct gc_inode_list *gc_list)
400 struct inode_entry *ie, *next_ie;
401 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
402 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
405 kmem_cache_free(inode_entry_slab, ie);
409 static int check_valid_map(struct f2fs_sb_info *sbi,
410 unsigned int segno, int offset)
412 struct sit_info *sit_i = SIT_I(sbi);
413 struct seg_entry *sentry;
416 mutex_lock(&sit_i->sentry_lock);
417 sentry = get_seg_entry(sbi, segno);
418 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
419 mutex_unlock(&sit_i->sentry_lock);
424 * This function compares node address got in summary with that in NAT.
425 * On validity, copy that node with cold status, otherwise (invalid node)
428 static void gc_node_segment(struct f2fs_sb_info *sbi,
429 struct f2fs_summary *sum, unsigned int segno, int gc_type)
431 struct f2fs_summary *entry;
436 start_addr = START_BLOCK(sbi, segno);
441 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
442 nid_t nid = le32_to_cpu(entry->nid);
443 struct page *node_page;
446 /* stop BG_GC if there is not enough free sections. */
447 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
450 if (check_valid_map(sbi, segno, off) == 0)
454 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
460 ra_node_page(sbi, nid);
465 node_page = get_node_page(sbi, nid);
466 if (IS_ERR(node_page))
469 /* block may become invalid during get_node_page */
470 if (check_valid_map(sbi, segno, off) == 0) {
471 f2fs_put_page(node_page, 1);
475 get_node_info(sbi, nid, &ni);
476 if (ni.blk_addr != start_addr + off) {
477 f2fs_put_page(node_page, 1);
481 move_node_page(node_page, gc_type);
482 stat_inc_node_blk_count(sbi, 1, gc_type);
490 * Calculate start block index indicating the given node offset.
491 * Be careful, caller should give this node offset only indicating direct node
492 * blocks. If any node offsets, which point the other types of node blocks such
493 * as indirect or double indirect node blocks, are given, it must be a caller's
496 block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
498 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
506 } else if (node_ofs <= indirect_blks) {
507 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
508 bidx = node_ofs - 2 - dec;
510 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
511 bidx = node_ofs - 5 - dec;
513 return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
516 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
517 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
519 struct page *node_page;
521 unsigned int ofs_in_node;
522 block_t source_blkaddr;
524 nid = le32_to_cpu(sum->nid);
525 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
527 node_page = get_node_page(sbi, nid);
528 if (IS_ERR(node_page))
531 get_node_info(sbi, nid, dni);
533 if (sum->version != dni->version) {
534 f2fs_put_page(node_page, 1);
538 *nofs = ofs_of_node(node_page);
539 source_blkaddr = datablock_addr(node_page, ofs_in_node);
540 f2fs_put_page(node_page, 1);
542 if (source_blkaddr != blkaddr)
547 static void move_encrypted_block(struct inode *inode, block_t bidx,
548 unsigned int segno, int off)
550 struct f2fs_io_info fio = {
551 .sbi = F2FS_I_SB(inode),
554 .op_flags = READ_SYNC,
555 .encrypted_page = NULL,
557 struct dnode_of_data dn;
558 struct f2fs_summary sum;
564 /* do not read out */
565 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
569 if (!check_valid_map(F2FS_I_SB(inode), segno, off))
572 set_new_dnode(&dn, inode, NULL, NULL, 0);
573 err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
577 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
578 ClearPageUptodate(page);
583 * don't cache encrypted data into meta inode until previous dirty
584 * data were writebacked to avoid racing between GC and flush.
586 f2fs_wait_on_page_writeback(page, DATA, true);
588 get_node_info(fio.sbi, dn.nid, &ni);
589 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
593 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
595 allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
596 &sum, CURSEG_COLD_DATA);
598 fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), newaddr,
599 FGP_LOCK | FGP_CREAT, GFP_NOFS);
600 if (!fio.encrypted_page) {
605 err = f2fs_submit_page_bio(&fio);
610 lock_page(fio.encrypted_page);
612 if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
616 if (unlikely(!PageUptodate(fio.encrypted_page))) {
621 set_page_dirty(fio.encrypted_page);
622 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
623 if (clear_page_dirty_for_io(fio.encrypted_page))
624 dec_page_count(fio.sbi, F2FS_DIRTY_META);
626 set_page_writeback(fio.encrypted_page);
628 /* allocate block address */
629 f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
631 fio.op = REQ_OP_WRITE;
632 fio.op_flags = WRITE_SYNC;
633 fio.new_blkaddr = newaddr;
634 f2fs_submit_page_mbio(&fio);
636 f2fs_update_data_blkaddr(&dn, newaddr);
637 set_inode_flag(inode, FI_APPEND_WRITE);
638 if (page->index == 0)
639 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
641 f2fs_put_page(fio.encrypted_page, 1);
644 __f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
649 f2fs_put_page(page, 1);
652 static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
653 unsigned int segno, int off)
657 page = get_lock_data_page(inode, bidx, true);
661 if (!check_valid_map(F2FS_I_SB(inode), segno, off))
664 if (gc_type == BG_GC) {
665 if (PageWriteback(page))
667 set_page_dirty(page);
670 struct f2fs_io_info fio = {
671 .sbi = F2FS_I_SB(inode),
674 .op_flags = WRITE_SYNC,
676 .encrypted_page = NULL,
678 bool is_dirty = PageDirty(page);
682 set_page_dirty(page);
683 f2fs_wait_on_page_writeback(page, DATA, true);
684 if (clear_page_dirty_for_io(page)) {
685 inode_dec_dirty_pages(inode);
686 remove_dirty_inode(inode);
691 err = do_write_data_page(&fio);
692 if (err == -ENOMEM && is_dirty) {
693 congestion_wait(BLK_RW_ASYNC, HZ/50);
698 f2fs_put_page(page, 1);
702 * This function tries to get parent node of victim data block, and identifies
703 * data block validity. If the block is valid, copy that with cold status and
704 * modify parent node.
705 * If the parent node is not valid or the data block address is different,
706 * the victim data block is ignored.
708 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
709 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
711 struct super_block *sb = sbi->sb;
712 struct f2fs_summary *entry;
717 start_addr = START_BLOCK(sbi, segno);
722 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
723 struct page *data_page;
725 struct node_info dni; /* dnode info for the data */
726 unsigned int ofs_in_node, nofs;
728 nid_t nid = le32_to_cpu(entry->nid);
730 /* stop BG_GC if there is not enough free sections. */
731 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
734 if (check_valid_map(sbi, segno, off) == 0)
738 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
744 ra_node_page(sbi, nid);
748 /* Get an inode by ino with checking validity */
749 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
753 ra_node_page(sbi, dni.ino);
757 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
760 inode = f2fs_iget(sb, dni.ino);
761 if (IS_ERR(inode) || is_bad_inode(inode))
764 /* if encrypted inode, let's go phase 3 */
765 if (f2fs_encrypted_inode(inode) &&
766 S_ISREG(inode->i_mode)) {
767 add_gc_inode(gc_list, inode);
771 start_bidx = start_bidx_of_node(nofs, inode);
772 data_page = get_read_data_page(inode,
773 start_bidx + ofs_in_node, REQ_RAHEAD,
775 if (IS_ERR(data_page)) {
780 f2fs_put_page(data_page, 0);
781 add_gc_inode(gc_list, inode);
786 inode = find_gc_inode(gc_list, dni.ino);
788 struct f2fs_inode_info *fi = F2FS_I(inode);
791 if (S_ISREG(inode->i_mode)) {
792 if (!down_write_trylock(&fi->dio_rwsem[READ]))
794 if (!down_write_trylock(
795 &fi->dio_rwsem[WRITE])) {
796 up_write(&fi->dio_rwsem[READ]);
802 start_bidx = start_bidx_of_node(nofs, inode)
804 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
805 move_encrypted_block(inode, start_bidx, segno, off);
807 move_data_page(inode, start_bidx, gc_type, segno, off);
810 up_write(&fi->dio_rwsem[WRITE]);
811 up_write(&fi->dio_rwsem[READ]);
814 stat_inc_data_blk_count(sbi, 1, gc_type);
822 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
825 struct sit_info *sit_i = SIT_I(sbi);
828 mutex_lock(&sit_i->sentry_lock);
829 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
831 mutex_unlock(&sit_i->sentry_lock);
835 static int do_garbage_collect(struct f2fs_sb_info *sbi,
836 unsigned int start_segno,
837 struct gc_inode_list *gc_list, int gc_type)
839 struct page *sum_page;
840 struct f2fs_summary_block *sum;
841 struct blk_plug plug;
842 unsigned int segno = start_segno;
843 unsigned int end_segno = start_segno + sbi->segs_per_sec;
845 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
846 SUM_TYPE_DATA : SUM_TYPE_NODE;
848 /* readahead multi ssa blocks those have contiguous address */
849 if (sbi->segs_per_sec > 1)
850 ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
851 sbi->segs_per_sec, META_SSA, true);
853 /* reference all summary page */
854 while (segno < end_segno) {
855 sum_page = get_sum_page(sbi, segno++);
856 unlock_page(sum_page);
859 blk_start_plug(&plug);
861 for (segno = start_segno; segno < end_segno; segno++) {
863 /* find segment summary of victim */
864 sum_page = find_get_page(META_MAPPING(sbi),
865 GET_SUM_BLOCK(sbi, segno));
866 f2fs_put_page(sum_page, 0);
868 if (get_valid_blocks(sbi, segno, 1) == 0 ||
869 !PageUptodate(sum_page) ||
870 unlikely(f2fs_cp_error(sbi)))
873 sum = page_address(sum_page);
874 f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer)));
877 * this is to avoid deadlock:
878 * - lock_page(sum_page) - f2fs_replace_block
879 * - check_valid_map() - mutex_lock(sentry_lock)
880 * - mutex_lock(sentry_lock) - change_curseg()
881 * - lock_page(sum_page)
884 if (type == SUM_TYPE_NODE)
885 gc_node_segment(sbi, sum->entries, segno, gc_type);
887 gc_data_segment(sbi, sum->entries, gc_list, segno,
890 stat_inc_seg_count(sbi, type, gc_type);
892 f2fs_put_page(sum_page, 0);
895 if (gc_type == FG_GC)
896 f2fs_submit_merged_bio(sbi,
897 (type == SUM_TYPE_NODE) ? NODE : DATA, WRITE);
899 blk_finish_plug(&plug);
901 if (gc_type == FG_GC &&
902 get_valid_blocks(sbi, start_segno, sbi->segs_per_sec) == 0)
905 stat_inc_call_count(sbi->stat_info);
910 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, bool background)
913 int gc_type = sync ? FG_GC : BG_GC;
916 struct cp_control cpc;
917 struct gc_inode_list gc_list = {
918 .ilist = LIST_HEAD_INIT(gc_list.ilist),
919 .iroot = RADIX_TREE_INIT(GFP_NOFS),
922 cpc.reason = __get_cp_reason(sbi);
926 if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE)))
928 if (unlikely(f2fs_cp_error(sbi))) {
933 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, sec_freed, 0)) {
936 * If there is no victim and no prefree segment but still not
937 * enough free sections, we should flush dent/node blocks and do
938 * garbage collections.
940 if (__get_victim(sbi, &segno, gc_type) ||
941 prefree_segments(sbi)) {
942 ret = write_checkpoint(sbi, &cpc);
946 } else if (has_not_enough_free_secs(sbi, 0, 0)) {
947 ret = write_checkpoint(sbi, &cpc);
951 } else if (gc_type == BG_GC && !background) {
952 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
956 if (segno == NULL_SEGNO && !__get_victim(sbi, &segno, gc_type))
960 if (do_garbage_collect(sbi, segno, &gc_list, gc_type) &&
964 if (gc_type == FG_GC)
965 sbi->cur_victim_sec = NULL_SEGNO;
968 if (has_not_enough_free_secs(sbi, sec_freed, 0))
971 if (gc_type == FG_GC)
972 ret = write_checkpoint(sbi, &cpc);
975 mutex_unlock(&sbi->gc_mutex);
977 put_gc_inode(&gc_list);
980 ret = sec_freed ? 0 : -EAGAIN;
984 void build_gc_manager(struct f2fs_sb_info *sbi)
986 DIRTY_I(sbi)->v_ops = &default_v_ops;