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;
37 wait_event_interruptible_timeout(*wq,
38 kthread_should_stop() || freezing(current) ||
40 msecs_to_jiffies(wait_ms));
42 /* give it a try one time */
48 if (kthread_should_stop())
51 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
52 increase_sleep_time(gc_th, &wait_ms);
56 #ifdef CONFIG_F2FS_FAULT_INJECTION
57 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
58 f2fs_show_injection_info(FAULT_CHECKPOINT);
59 f2fs_stop_checkpoint(sbi, false);
63 if (!sb_start_write_trylock(sbi->sb))
67 * [GC triggering condition]
68 * 0. GC is not conducted currently.
69 * 1. There are enough dirty segments.
70 * 2. IO subsystem is idle by checking the # of writeback pages.
71 * 3. IO subsystem is idle by checking the # of requests in
72 * bdev's request list.
74 * Note) We have to avoid triggering GCs frequently.
75 * Because it is possible that some segments can be
76 * invalidated soon after by user update or deletion.
77 * So, I'd like to wait some time to collect dirty segments.
79 if (!mutex_trylock(&sbi->gc_mutex))
82 if (gc_th->gc_urgent) {
83 wait_ms = gc_th->urgent_sleep_time;
88 increase_sleep_time(gc_th, &wait_ms);
89 mutex_unlock(&sbi->gc_mutex);
93 if (has_enough_invalid_blocks(sbi))
94 decrease_sleep_time(gc_th, &wait_ms);
96 increase_sleep_time(gc_th, &wait_ms);
98 stat_inc_bggc_count(sbi);
100 /* if return value is not zero, no victim was selected */
101 if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO))
102 wait_ms = gc_th->no_gc_sleep_time;
104 trace_f2fs_background_gc(sbi->sb, wait_ms,
105 prefree_segments(sbi), free_segments(sbi));
107 /* balancing f2fs's metadata periodically */
108 f2fs_balance_fs_bg(sbi);
110 sb_end_write(sbi->sb);
112 } while (!kthread_should_stop());
116 int start_gc_thread(struct f2fs_sb_info *sbi)
118 struct f2fs_gc_kthread *gc_th;
119 dev_t dev = sbi->sb->s_bdev->bd_dev;
122 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
128 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
129 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
130 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
131 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
134 gc_th->gc_urgent = 0;
137 sbi->gc_thread = gc_th;
138 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
139 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
140 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
141 if (IS_ERR(gc_th->f2fs_gc_task)) {
142 err = PTR_ERR(gc_th->f2fs_gc_task);
144 sbi->gc_thread = NULL;
150 void stop_gc_thread(struct f2fs_sb_info *sbi)
152 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
155 kthread_stop(gc_th->f2fs_gc_task);
157 sbi->gc_thread = NULL;
160 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
162 int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
164 if (gc_th && gc_th->gc_idle) {
165 if (gc_th->gc_idle == 1)
167 else if (gc_th->gc_idle == 2)
173 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
174 int type, struct victim_sel_policy *p)
176 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
178 if (p->alloc_mode == SSR) {
179 p->gc_mode = GC_GREEDY;
180 p->dirty_segmap = dirty_i->dirty_segmap[type];
181 p->max_search = dirty_i->nr_dirty[type];
184 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
185 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
186 p->max_search = dirty_i->nr_dirty[DIRTY];
187 p->ofs_unit = sbi->segs_per_sec;
190 /* we need to check every dirty segments in the FG_GC case */
191 if (gc_type != FG_GC && p->max_search > sbi->max_victim_search)
192 p->max_search = sbi->max_victim_search;
194 /* let's select beginning hot/small space first in no_heap mode*/
195 if (test_opt(sbi, NOHEAP) &&
196 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
199 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
202 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
203 struct victim_sel_policy *p)
205 /* SSR allocates in a segment unit */
206 if (p->alloc_mode == SSR)
207 return sbi->blocks_per_seg;
208 if (p->gc_mode == GC_GREEDY)
209 return 2 * sbi->blocks_per_seg * p->ofs_unit;
210 else if (p->gc_mode == GC_CB)
212 else /* No other gc_mode */
216 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
218 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
222 * If the gc_type is FG_GC, we can select victim segments
223 * selected by background GC before.
224 * Those segments guarantee they have small valid blocks.
226 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
227 if (sec_usage_check(sbi, secno))
230 if (no_fggc_candidate(sbi, secno))
233 clear_bit(secno, dirty_i->victim_secmap);
234 return GET_SEG_FROM_SEC(sbi, secno);
239 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
241 struct sit_info *sit_i = SIT_I(sbi);
242 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
243 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
244 unsigned long long mtime = 0;
245 unsigned int vblocks;
246 unsigned char age = 0;
250 for (i = 0; i < sbi->segs_per_sec; i++)
251 mtime += get_seg_entry(sbi, start + i)->mtime;
252 vblocks = get_valid_blocks(sbi, segno, true);
254 mtime = div_u64(mtime, sbi->segs_per_sec);
255 vblocks = div_u64(vblocks, sbi->segs_per_sec);
257 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
259 /* Handle if the system time has changed by the user */
260 if (mtime < sit_i->min_mtime)
261 sit_i->min_mtime = mtime;
262 if (mtime > sit_i->max_mtime)
263 sit_i->max_mtime = mtime;
264 if (sit_i->max_mtime != sit_i->min_mtime)
265 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
266 sit_i->max_mtime - sit_i->min_mtime);
268 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
271 static unsigned int get_greedy_cost(struct f2fs_sb_info *sbi,
274 unsigned int valid_blocks =
275 get_valid_blocks(sbi, segno, true);
277 return IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
278 valid_blocks * 2 : valid_blocks;
281 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
282 unsigned int segno, struct victim_sel_policy *p)
284 if (p->alloc_mode == SSR)
285 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
287 /* alloc_mode == LFS */
288 if (p->gc_mode == GC_GREEDY)
289 return get_greedy_cost(sbi, segno);
291 return get_cb_cost(sbi, segno);
294 static unsigned int count_bits(const unsigned long *addr,
295 unsigned int offset, unsigned int len)
297 unsigned int end = offset + len, sum = 0;
299 while (offset < end) {
300 if (test_bit(offset++, addr))
307 * This function is called from two paths.
308 * One is garbage collection and the other is SSR segment selection.
309 * When it is called during GC, it just gets a victim segment
310 * and it does not remove it from dirty seglist.
311 * When it is called from SSR segment selection, it finds a segment
312 * which has minimum valid blocks and removes it from dirty seglist.
314 static int get_victim_by_default(struct f2fs_sb_info *sbi,
315 unsigned int *result, int gc_type, int type, char alloc_mode)
317 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
318 struct sit_info *sm = SIT_I(sbi);
319 struct victim_sel_policy p;
320 unsigned int secno, last_victim;
321 unsigned int last_segment = MAIN_SEGS(sbi);
322 unsigned int nsearched = 0;
324 mutex_lock(&dirty_i->seglist_lock);
326 p.alloc_mode = alloc_mode;
327 select_policy(sbi, gc_type, type, &p);
329 p.min_segno = NULL_SEGNO;
330 p.min_cost = get_max_cost(sbi, &p);
332 if (*result != NULL_SEGNO) {
333 if (get_valid_blocks(sbi, *result, false) &&
334 !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
335 p.min_segno = *result;
339 if (p.max_search == 0)
342 last_victim = sm->last_victim[p.gc_mode];
343 if (p.alloc_mode == LFS && gc_type == FG_GC) {
344 p.min_segno = check_bg_victims(sbi);
345 if (p.min_segno != NULL_SEGNO)
353 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
354 if (segno >= last_segment) {
355 if (sm->last_victim[p.gc_mode]) {
357 sm->last_victim[p.gc_mode];
358 sm->last_victim[p.gc_mode] = 0;
365 p.offset = segno + p.ofs_unit;
366 if (p.ofs_unit > 1) {
367 p.offset -= segno % p.ofs_unit;
368 nsearched += count_bits(p.dirty_segmap,
369 p.offset - p.ofs_unit,
375 secno = GET_SEC_FROM_SEG(sbi, segno);
377 if (sec_usage_check(sbi, secno))
379 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
381 if (gc_type == FG_GC && p.alloc_mode == LFS &&
382 no_fggc_candidate(sbi, secno))
385 cost = get_gc_cost(sbi, segno, &p);
387 if (p.min_cost > cost) {
392 if (nsearched >= p.max_search) {
393 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
394 sm->last_victim[p.gc_mode] = last_victim + 1;
396 sm->last_victim[p.gc_mode] = segno + 1;
397 sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi);
401 if (p.min_segno != NULL_SEGNO) {
403 if (p.alloc_mode == LFS) {
404 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
405 if (gc_type == FG_GC)
406 sbi->cur_victim_sec = secno;
408 set_bit(secno, dirty_i->victim_secmap);
410 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
412 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
414 prefree_segments(sbi), free_segments(sbi));
417 mutex_unlock(&dirty_i->seglist_lock);
419 return (p.min_segno == NULL_SEGNO) ? 0 : 1;
422 static const struct victim_selection default_v_ops = {
423 .get_victim = get_victim_by_default,
426 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
428 struct inode_entry *ie;
430 ie = radix_tree_lookup(&gc_list->iroot, ino);
436 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
438 struct inode_entry *new_ie;
440 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
444 new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
445 new_ie->inode = inode;
447 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
448 list_add_tail(&new_ie->list, &gc_list->ilist);
451 static void put_gc_inode(struct gc_inode_list *gc_list)
453 struct inode_entry *ie, *next_ie;
454 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
455 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
458 kmem_cache_free(inode_entry_slab, ie);
462 static int check_valid_map(struct f2fs_sb_info *sbi,
463 unsigned int segno, int offset)
465 struct sit_info *sit_i = SIT_I(sbi);
466 struct seg_entry *sentry;
469 mutex_lock(&sit_i->sentry_lock);
470 sentry = get_seg_entry(sbi, segno);
471 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
472 mutex_unlock(&sit_i->sentry_lock);
477 * This function compares node address got in summary with that in NAT.
478 * On validity, copy that node with cold status, otherwise (invalid node)
481 static void gc_node_segment(struct f2fs_sb_info *sbi,
482 struct f2fs_summary *sum, unsigned int segno, int gc_type)
484 struct f2fs_summary *entry;
489 start_addr = START_BLOCK(sbi, segno);
494 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
495 nid_t nid = le32_to_cpu(entry->nid);
496 struct page *node_page;
499 /* stop BG_GC if there is not enough free sections. */
500 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
503 if (check_valid_map(sbi, segno, off) == 0)
507 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
513 ra_node_page(sbi, nid);
518 node_page = get_node_page(sbi, nid);
519 if (IS_ERR(node_page))
522 /* block may become invalid during get_node_page */
523 if (check_valid_map(sbi, segno, off) == 0) {
524 f2fs_put_page(node_page, 1);
528 get_node_info(sbi, nid, &ni);
529 if (ni.blk_addr != start_addr + off) {
530 f2fs_put_page(node_page, 1);
534 move_node_page(node_page, gc_type);
535 stat_inc_node_blk_count(sbi, 1, gc_type);
543 * Calculate start block index indicating the given node offset.
544 * Be careful, caller should give this node offset only indicating direct node
545 * blocks. If any node offsets, which point the other types of node blocks such
546 * as indirect or double indirect node blocks, are given, it must be a caller's
549 block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
551 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
559 } else if (node_ofs <= indirect_blks) {
560 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
561 bidx = node_ofs - 2 - dec;
563 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
564 bidx = node_ofs - 5 - dec;
566 return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
569 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
570 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
572 struct page *node_page;
574 unsigned int ofs_in_node;
575 block_t source_blkaddr;
577 nid = le32_to_cpu(sum->nid);
578 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
580 node_page = get_node_page(sbi, nid);
581 if (IS_ERR(node_page))
584 get_node_info(sbi, nid, dni);
586 if (sum->version != dni->version) {
587 f2fs_msg(sbi->sb, KERN_WARNING,
588 "%s: valid data with mismatched node version.",
590 set_sbi_flag(sbi, SBI_NEED_FSCK);
593 *nofs = ofs_of_node(node_page);
594 source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node);
595 f2fs_put_page(node_page, 1);
597 if (source_blkaddr != blkaddr)
603 * Move data block via META_MAPPING while keeping locked data page.
604 * This can be used to move blocks, aka LBAs, directly on disk.
606 static void move_data_block(struct inode *inode, block_t bidx,
607 unsigned int segno, int off)
609 struct f2fs_io_info fio = {
610 .sbi = F2FS_I_SB(inode),
615 .encrypted_page = NULL,
618 struct dnode_of_data dn;
619 struct f2fs_summary sum;
625 /* do not read out */
626 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
630 if (!check_valid_map(F2FS_I_SB(inode), segno, off))
633 if (f2fs_is_atomic_file(inode))
636 set_new_dnode(&dn, inode, NULL, NULL, 0);
637 err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
641 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
642 ClearPageUptodate(page);
647 * don't cache encrypted data into meta inode until previous dirty
648 * data were writebacked to avoid racing between GC and flush.
650 f2fs_wait_on_page_writeback(page, DATA, true);
652 get_node_info(fio.sbi, dn.nid, &ni);
653 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
657 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
659 allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
660 &sum, CURSEG_COLD_DATA, NULL, false);
662 fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), newaddr,
663 FGP_LOCK | FGP_CREAT, GFP_NOFS);
664 if (!fio.encrypted_page) {
669 err = f2fs_submit_page_bio(&fio);
674 lock_page(fio.encrypted_page);
676 if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
680 if (unlikely(!PageUptodate(fio.encrypted_page))) {
685 set_page_dirty(fio.encrypted_page);
686 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
687 if (clear_page_dirty_for_io(fio.encrypted_page))
688 dec_page_count(fio.sbi, F2FS_DIRTY_META);
690 set_page_writeback(fio.encrypted_page);
692 /* allocate block address */
693 f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
695 fio.op = REQ_OP_WRITE;
696 fio.op_flags = REQ_SYNC;
697 fio.new_blkaddr = newaddr;
698 err = f2fs_submit_page_write(&fio);
700 if (PageWriteback(fio.encrypted_page))
701 end_page_writeback(fio.encrypted_page);
705 f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
707 f2fs_update_data_blkaddr(&dn, newaddr);
708 set_inode_flag(inode, FI_APPEND_WRITE);
709 if (page->index == 0)
710 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
712 f2fs_put_page(fio.encrypted_page, 1);
715 __f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
720 f2fs_put_page(page, 1);
723 static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
724 unsigned int segno, int off)
728 page = get_lock_data_page(inode, bidx, true);
732 if (!check_valid_map(F2FS_I_SB(inode), segno, off))
735 if (f2fs_is_atomic_file(inode))
738 if (gc_type == BG_GC) {
739 if (PageWriteback(page))
741 set_page_dirty(page);
744 struct f2fs_io_info fio = {
745 .sbi = F2FS_I_SB(inode),
749 .op_flags = REQ_SYNC,
750 .old_blkaddr = NULL_ADDR,
752 .encrypted_page = NULL,
753 .need_lock = LOCK_REQ,
754 .io_type = FS_GC_DATA_IO,
756 bool is_dirty = PageDirty(page);
760 set_page_dirty(page);
761 f2fs_wait_on_page_writeback(page, DATA, true);
762 if (clear_page_dirty_for_io(page)) {
763 inode_dec_dirty_pages(inode);
764 remove_dirty_inode(inode);
769 err = do_write_data_page(&fio);
771 clear_cold_data(page);
772 if (err == -ENOMEM) {
773 congestion_wait(BLK_RW_ASYNC, HZ/50);
777 set_page_dirty(page);
781 f2fs_put_page(page, 1);
785 * This function tries to get parent node of victim data block, and identifies
786 * data block validity. If the block is valid, copy that with cold status and
787 * modify parent node.
788 * If the parent node is not valid or the data block address is different,
789 * the victim data block is ignored.
791 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
792 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
794 struct super_block *sb = sbi->sb;
795 struct f2fs_summary *entry;
800 start_addr = START_BLOCK(sbi, segno);
805 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
806 struct page *data_page;
808 struct node_info dni; /* dnode info for the data */
809 unsigned int ofs_in_node, nofs;
811 nid_t nid = le32_to_cpu(entry->nid);
813 /* stop BG_GC if there is not enough free sections. */
814 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
817 if (check_valid_map(sbi, segno, off) == 0)
821 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
827 ra_node_page(sbi, nid);
831 /* Get an inode by ino with checking validity */
832 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
836 ra_node_page(sbi, dni.ino);
840 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
843 inode = f2fs_iget(sb, dni.ino);
844 if (IS_ERR(inode) || is_bad_inode(inode))
847 /* if encrypted inode, let's go phase 3 */
848 if (f2fs_encrypted_file(inode)) {
849 add_gc_inode(gc_list, inode);
853 start_bidx = start_bidx_of_node(nofs, inode);
854 data_page = get_read_data_page(inode,
855 start_bidx + ofs_in_node, REQ_RAHEAD,
857 if (IS_ERR(data_page)) {
862 f2fs_put_page(data_page, 0);
863 add_gc_inode(gc_list, inode);
868 inode = find_gc_inode(gc_list, dni.ino);
870 struct f2fs_inode_info *fi = F2FS_I(inode);
873 if (S_ISREG(inode->i_mode)) {
874 if (!down_write_trylock(&fi->dio_rwsem[READ]))
876 if (!down_write_trylock(
877 &fi->dio_rwsem[WRITE])) {
878 up_write(&fi->dio_rwsem[READ]);
883 /* wait for all inflight aio data */
884 inode_dio_wait(inode);
887 start_bidx = start_bidx_of_node(nofs, inode)
889 if (f2fs_encrypted_file(inode))
890 move_data_block(inode, start_bidx, segno, off);
892 move_data_page(inode, start_bidx, gc_type,
896 up_write(&fi->dio_rwsem[WRITE]);
897 up_write(&fi->dio_rwsem[READ]);
900 stat_inc_data_blk_count(sbi, 1, gc_type);
908 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
911 struct sit_info *sit_i = SIT_I(sbi);
914 mutex_lock(&sit_i->sentry_lock);
915 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
917 mutex_unlock(&sit_i->sentry_lock);
921 static int do_garbage_collect(struct f2fs_sb_info *sbi,
922 unsigned int start_segno,
923 struct gc_inode_list *gc_list, int gc_type)
925 struct page *sum_page;
926 struct f2fs_summary_block *sum;
927 struct blk_plug plug;
928 unsigned int segno = start_segno;
929 unsigned int end_segno = start_segno + sbi->segs_per_sec;
931 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
932 SUM_TYPE_DATA : SUM_TYPE_NODE;
934 /* readahead multi ssa blocks those have contiguous address */
935 if (sbi->segs_per_sec > 1)
936 ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
937 sbi->segs_per_sec, META_SSA, true);
939 /* reference all summary page */
940 while (segno < end_segno) {
941 sum_page = get_sum_page(sbi, segno++);
942 unlock_page(sum_page);
945 blk_start_plug(&plug);
947 for (segno = start_segno; segno < end_segno; segno++) {
949 /* find segment summary of victim */
950 sum_page = find_get_page(META_MAPPING(sbi),
951 GET_SUM_BLOCK(sbi, segno));
952 f2fs_put_page(sum_page, 0);
954 if (get_valid_blocks(sbi, segno, false) == 0)
956 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
959 sum = page_address(sum_page);
960 if (type != GET_SUM_TYPE((&sum->footer))) {
961 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent segment (%u) "
962 "type [%d, %d] in SSA and SIT",
963 segno, type, GET_SUM_TYPE((&sum->footer)));
964 set_sbi_flag(sbi, SBI_NEED_FSCK);
969 * this is to avoid deadlock:
970 * - lock_page(sum_page) - f2fs_replace_block
971 * - check_valid_map() - mutex_lock(sentry_lock)
972 * - mutex_lock(sentry_lock) - change_curseg()
973 * - lock_page(sum_page)
975 if (type == SUM_TYPE_NODE)
976 gc_node_segment(sbi, sum->entries, segno, gc_type);
978 gc_data_segment(sbi, sum->entries, gc_list, segno,
981 stat_inc_seg_count(sbi, type, gc_type);
984 if (gc_type == FG_GC &&
985 get_valid_blocks(sbi, segno, false) == 0)
988 f2fs_put_page(sum_page, 0);
991 if (gc_type == FG_GC)
992 f2fs_submit_merged_write(sbi,
993 (type == SUM_TYPE_NODE) ? NODE : DATA);
995 blk_finish_plug(&plug);
997 stat_inc_call_count(sbi->stat_info);
1002 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1003 bool background, unsigned int segno)
1005 int gc_type = sync ? FG_GC : BG_GC;
1006 int sec_freed = 0, seg_freed = 0, total_freed = 0;
1008 struct cp_control cpc;
1009 unsigned int init_segno = segno;
1010 struct gc_inode_list gc_list = {
1011 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1012 .iroot = RADIX_TREE_INIT(GFP_NOFS),
1015 trace_f2fs_gc_begin(sbi->sb, sync, background,
1016 get_pages(sbi, F2FS_DIRTY_NODES),
1017 get_pages(sbi, F2FS_DIRTY_DENTS),
1018 get_pages(sbi, F2FS_DIRTY_IMETA),
1021 reserved_segments(sbi),
1022 prefree_segments(sbi));
1024 cpc.reason = __get_cp_reason(sbi);
1026 if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) {
1030 if (unlikely(f2fs_cp_error(sbi))) {
1035 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1037 * For example, if there are many prefree_segments below given
1038 * threshold, we can make them free by checkpoint. Then, we
1039 * secure free segments which doesn't need fggc any more.
1041 if (prefree_segments(sbi)) {
1042 ret = write_checkpoint(sbi, &cpc);
1046 if (has_not_enough_free_secs(sbi, 0, 0))
1050 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1051 if (gc_type == BG_GC && !background) {
1055 if (!__get_victim(sbi, &segno, gc_type)) {
1060 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1061 if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1063 total_freed += seg_freed;
1065 if (gc_type == FG_GC)
1066 sbi->cur_victim_sec = NULL_SEGNO;
1069 if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1074 if (gc_type == FG_GC)
1075 ret = write_checkpoint(sbi, &cpc);
1078 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1079 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1081 trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1082 get_pages(sbi, F2FS_DIRTY_NODES),
1083 get_pages(sbi, F2FS_DIRTY_DENTS),
1084 get_pages(sbi, F2FS_DIRTY_IMETA),
1087 reserved_segments(sbi),
1088 prefree_segments(sbi));
1090 mutex_unlock(&sbi->gc_mutex);
1092 put_gc_inode(&gc_list);
1095 ret = sec_freed ? 0 : -EAGAIN;
1099 void build_gc_manager(struct f2fs_sb_info *sbi)
1101 u64 main_count, resv_count, ovp_count;
1103 DIRTY_I(sbi)->v_ops = &default_v_ops;
1105 /* threshold of # of valid blocks in a section for victims of FG_GC */
1106 main_count = SM_I(sbi)->main_segments << sbi->log_blocks_per_seg;
1107 resv_count = SM_I(sbi)->reserved_segments << sbi->log_blocks_per_seg;
1108 ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg;
1110 sbi->fggc_threshold = div64_u64((main_count - ovp_count) *
1111 BLKS_PER_SEC(sbi), (main_count - resv_count));
1113 /* give warm/cold data area from slower device */
1114 if (f2fs_is_multi_device(sbi) && sbi->segs_per_sec == 1)
1115 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1116 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;