1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/module.h>
10 #include <linux/backing-dev.h>
11 #include <linux/init.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/kthread.h>
14 #include <linux/delay.h>
15 #include <linux/freezer.h>
21 #include <trace/events/f2fs.h>
23 static int gc_thread_func(void *data)
25 struct f2fs_sb_info *sbi = data;
26 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
27 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
30 wait_ms = gc_th->min_sleep_time;
34 wait_event_interruptible_timeout(*wq,
35 kthread_should_stop() || freezing(current) ||
37 msecs_to_jiffies(wait_ms));
39 /* give it a try one time */
43 if (try_to_freeze()) {
44 stat_other_skip_bggc_count(sbi);
47 if (kthread_should_stop())
50 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
51 increase_sleep_time(gc_th, &wait_ms);
52 stat_other_skip_bggc_count(sbi);
56 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
57 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
58 f2fs_stop_checkpoint(sbi, false);
61 if (!sb_start_write_trylock(sbi->sb)) {
62 stat_other_skip_bggc_count(sbi);
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 (sbi->gc_mode == GC_URGENT) {
80 wait_ms = gc_th->urgent_sleep_time;
81 mutex_lock(&sbi->gc_mutex);
85 if (!mutex_trylock(&sbi->gc_mutex)) {
86 stat_other_skip_bggc_count(sbi);
90 if (!is_idle(sbi, GC_TIME)) {
91 increase_sleep_time(gc_th, &wait_ms);
92 mutex_unlock(&sbi->gc_mutex);
93 stat_io_skip_bggc_count(sbi);
97 if (has_enough_invalid_blocks(sbi))
98 decrease_sleep_time(gc_th, &wait_ms);
100 increase_sleep_time(gc_th, &wait_ms);
102 stat_inc_bggc_count(sbi);
104 /* if return value is not zero, no victim was selected */
105 if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO))
106 wait_ms = gc_th->no_gc_sleep_time;
108 trace_f2fs_background_gc(sbi->sb, wait_ms,
109 prefree_segments(sbi), free_segments(sbi));
111 /* balancing f2fs's metadata periodically */
112 f2fs_balance_fs_bg(sbi);
114 sb_end_write(sbi->sb);
116 } while (!kthread_should_stop());
120 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
122 struct f2fs_gc_kthread *gc_th;
123 dev_t dev = sbi->sb->s_bdev->bd_dev;
126 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
132 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
133 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
134 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
135 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
139 sbi->gc_thread = gc_th;
140 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
141 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
142 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
143 if (IS_ERR(gc_th->f2fs_gc_task)) {
144 err = PTR_ERR(gc_th->f2fs_gc_task);
146 sbi->gc_thread = NULL;
152 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
154 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
157 kthread_stop(gc_th->f2fs_gc_task);
159 sbi->gc_thread = NULL;
162 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
164 int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
166 switch (sbi->gc_mode) {
178 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
179 int type, struct victim_sel_policy *p)
181 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
183 if (p->alloc_mode == SSR) {
184 p->gc_mode = GC_GREEDY;
185 p->dirty_segmap = dirty_i->dirty_segmap[type];
186 p->max_search = dirty_i->nr_dirty[type];
189 p->gc_mode = select_gc_type(sbi, gc_type);
190 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
191 p->max_search = dirty_i->nr_dirty[DIRTY];
192 p->ofs_unit = sbi->segs_per_sec;
195 /* we need to check every dirty segments in the FG_GC case */
196 if (gc_type != FG_GC &&
197 (sbi->gc_mode != GC_URGENT) &&
198 p->max_search > sbi->max_victim_search)
199 p->max_search = sbi->max_victim_search;
201 /* let's select beginning hot/small space first in no_heap mode*/
202 if (test_opt(sbi, NOHEAP) &&
203 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
206 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
209 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
210 struct victim_sel_policy *p)
212 /* SSR allocates in a segment unit */
213 if (p->alloc_mode == SSR)
214 return sbi->blocks_per_seg;
215 if (p->gc_mode == GC_GREEDY)
216 return 2 * sbi->blocks_per_seg * p->ofs_unit;
217 else if (p->gc_mode == GC_CB)
219 else /* No other gc_mode */
223 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
225 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
229 * If the gc_type is FG_GC, we can select victim segments
230 * selected by background GC before.
231 * Those segments guarantee they have small valid blocks.
233 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
234 if (sec_usage_check(sbi, secno))
236 clear_bit(secno, dirty_i->victim_secmap);
237 return GET_SEG_FROM_SEC(sbi, secno);
242 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
244 struct sit_info *sit_i = SIT_I(sbi);
245 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
246 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
247 unsigned long long mtime = 0;
248 unsigned int vblocks;
249 unsigned char age = 0;
253 for (i = 0; i < sbi->segs_per_sec; i++)
254 mtime += get_seg_entry(sbi, start + i)->mtime;
255 vblocks = get_valid_blocks(sbi, segno, true);
257 mtime = div_u64(mtime, sbi->segs_per_sec);
258 vblocks = div_u64(vblocks, sbi->segs_per_sec);
260 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
262 /* Handle if the system time has changed by the user */
263 if (mtime < sit_i->min_mtime)
264 sit_i->min_mtime = mtime;
265 if (mtime > sit_i->max_mtime)
266 sit_i->max_mtime = mtime;
267 if (sit_i->max_mtime != sit_i->min_mtime)
268 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
269 sit_i->max_mtime - sit_i->min_mtime);
271 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
274 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
275 unsigned int segno, struct victim_sel_policy *p)
277 if (p->alloc_mode == SSR)
278 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
280 /* alloc_mode == LFS */
281 if (p->gc_mode == GC_GREEDY)
282 return get_valid_blocks(sbi, segno, true);
284 return get_cb_cost(sbi, segno);
287 static unsigned int count_bits(const unsigned long *addr,
288 unsigned int offset, unsigned int len)
290 unsigned int end = offset + len, sum = 0;
292 while (offset < end) {
293 if (test_bit(offset++, addr))
300 * This function is called from two paths.
301 * One is garbage collection and the other is SSR segment selection.
302 * When it is called during GC, it just gets a victim segment
303 * and it does not remove it from dirty seglist.
304 * When it is called from SSR segment selection, it finds a segment
305 * which has minimum valid blocks and removes it from dirty seglist.
307 static int get_victim_by_default(struct f2fs_sb_info *sbi,
308 unsigned int *result, int gc_type, int type, char alloc_mode)
310 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
311 struct sit_info *sm = SIT_I(sbi);
312 struct victim_sel_policy p;
313 unsigned int secno, last_victim;
314 unsigned int last_segment;
315 unsigned int nsearched = 0;
317 mutex_lock(&dirty_i->seglist_lock);
318 last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
320 p.alloc_mode = alloc_mode;
321 select_policy(sbi, gc_type, type, &p);
323 p.min_segno = NULL_SEGNO;
324 p.min_cost = get_max_cost(sbi, &p);
326 if (*result != NULL_SEGNO) {
327 if (get_valid_blocks(sbi, *result, false) &&
328 !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
329 p.min_segno = *result;
333 if (p.max_search == 0)
336 if (__is_large_section(sbi) && p.alloc_mode == LFS) {
337 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
338 p.min_segno = sbi->next_victim_seg[BG_GC];
339 *result = p.min_segno;
340 sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
343 if (gc_type == FG_GC &&
344 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
345 p.min_segno = sbi->next_victim_seg[FG_GC];
346 *result = p.min_segno;
347 sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
352 last_victim = sm->last_victim[p.gc_mode];
353 if (p.alloc_mode == LFS && gc_type == FG_GC) {
354 p.min_segno = check_bg_victims(sbi);
355 if (p.min_segno != NULL_SEGNO)
363 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
364 if (segno >= last_segment) {
365 if (sm->last_victim[p.gc_mode]) {
367 sm->last_victim[p.gc_mode];
368 sm->last_victim[p.gc_mode] = 0;
375 p.offset = segno + p.ofs_unit;
376 if (p.ofs_unit > 1) {
377 p.offset -= segno % p.ofs_unit;
378 nsearched += count_bits(p.dirty_segmap,
379 p.offset - p.ofs_unit,
385 #ifdef CONFIG_F2FS_CHECK_FS
387 * skip selecting the invalid segno (that is failed due to block
388 * validity check failure during GC) to avoid endless GC loop in
391 if (test_bit(segno, sm->invalid_segmap))
395 secno = GET_SEC_FROM_SEG(sbi, segno);
397 if (sec_usage_check(sbi, secno))
399 /* Don't touch checkpointed data */
400 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
401 get_ckpt_valid_blocks(sbi, segno) &&
402 p.alloc_mode != SSR))
404 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
407 cost = get_gc_cost(sbi, segno, &p);
409 if (p.min_cost > cost) {
414 if (nsearched >= p.max_search) {
415 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
416 sm->last_victim[p.gc_mode] = last_victim + 1;
418 sm->last_victim[p.gc_mode] = segno + 1;
419 sm->last_victim[p.gc_mode] %=
420 (MAIN_SECS(sbi) * sbi->segs_per_sec);
424 if (p.min_segno != NULL_SEGNO) {
426 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
428 if (p.alloc_mode == LFS) {
429 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
430 if (gc_type == FG_GC)
431 sbi->cur_victim_sec = secno;
433 set_bit(secno, dirty_i->victim_secmap);
438 if (p.min_segno != NULL_SEGNO)
439 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
441 prefree_segments(sbi), free_segments(sbi));
442 mutex_unlock(&dirty_i->seglist_lock);
444 return (p.min_segno == NULL_SEGNO) ? 0 : 1;
447 static const struct victim_selection default_v_ops = {
448 .get_victim = get_victim_by_default,
451 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
453 struct inode_entry *ie;
455 ie = radix_tree_lookup(&gc_list->iroot, ino);
461 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
463 struct inode_entry *new_ie;
465 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
469 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
470 new_ie->inode = inode;
472 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
473 list_add_tail(&new_ie->list, &gc_list->ilist);
476 static void put_gc_inode(struct gc_inode_list *gc_list)
478 struct inode_entry *ie, *next_ie;
479 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
480 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
483 kmem_cache_free(f2fs_inode_entry_slab, ie);
487 static int check_valid_map(struct f2fs_sb_info *sbi,
488 unsigned int segno, int offset)
490 struct sit_info *sit_i = SIT_I(sbi);
491 struct seg_entry *sentry;
494 down_read(&sit_i->sentry_lock);
495 sentry = get_seg_entry(sbi, segno);
496 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
497 up_read(&sit_i->sentry_lock);
502 * This function compares node address got in summary with that in NAT.
503 * On validity, copy that node with cold status, otherwise (invalid node)
506 static int gc_node_segment(struct f2fs_sb_info *sbi,
507 struct f2fs_summary *sum, unsigned int segno, int gc_type)
509 struct f2fs_summary *entry;
513 bool fggc = (gc_type == FG_GC);
516 start_addr = START_BLOCK(sbi, segno);
521 if (fggc && phase == 2)
522 atomic_inc(&sbi->wb_sync_req[NODE]);
524 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
525 nid_t nid = le32_to_cpu(entry->nid);
526 struct page *node_page;
530 /* stop BG_GC if there is not enough free sections. */
531 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
534 if (check_valid_map(sbi, segno, off) == 0)
538 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
544 f2fs_ra_node_page(sbi, nid);
549 node_page = f2fs_get_node_page(sbi, nid);
550 if (IS_ERR(node_page))
553 /* block may become invalid during f2fs_get_node_page */
554 if (check_valid_map(sbi, segno, off) == 0) {
555 f2fs_put_page(node_page, 1);
559 if (f2fs_get_node_info(sbi, nid, &ni)) {
560 f2fs_put_page(node_page, 1);
564 if (ni.blk_addr != start_addr + off) {
565 f2fs_put_page(node_page, 1);
569 err = f2fs_move_node_page(node_page, gc_type);
570 if (!err && gc_type == FG_GC)
572 stat_inc_node_blk_count(sbi, 1, gc_type);
579 atomic_dec(&sbi->wb_sync_req[NODE]);
584 * Calculate start block index indicating the given node offset.
585 * Be careful, caller should give this node offset only indicating direct node
586 * blocks. If any node offsets, which point the other types of node blocks such
587 * as indirect or double indirect node blocks, are given, it must be a caller's
590 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
592 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
600 } else if (node_ofs <= indirect_blks) {
601 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
602 bidx = node_ofs - 2 - dec;
604 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
605 bidx = node_ofs - 5 - dec;
607 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
610 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
611 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
613 struct page *node_page;
615 unsigned int ofs_in_node;
616 block_t source_blkaddr;
618 nid = le32_to_cpu(sum->nid);
619 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
621 node_page = f2fs_get_node_page(sbi, nid);
622 if (IS_ERR(node_page))
625 if (f2fs_get_node_info(sbi, nid, dni)) {
626 f2fs_put_page(node_page, 1);
630 if (sum->version != dni->version) {
631 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
633 set_sbi_flag(sbi, SBI_NEED_FSCK);
636 *nofs = ofs_of_node(node_page);
637 source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node);
638 f2fs_put_page(node_page, 1);
640 if (source_blkaddr != blkaddr) {
641 #ifdef CONFIG_F2FS_CHECK_FS
642 unsigned int segno = GET_SEGNO(sbi, blkaddr);
643 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
645 if (unlikely(check_valid_map(sbi, segno, offset))) {
646 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
647 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n",
648 blkaddr, source_blkaddr, segno);
658 static int ra_data_block(struct inode *inode, pgoff_t index)
660 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
661 struct address_space *mapping = inode->i_mapping;
662 struct dnode_of_data dn;
664 struct extent_info ei = {0, 0, 0};
665 struct f2fs_io_info fio = {
672 .encrypted_page = NULL,
678 page = f2fs_grab_cache_page(mapping, index, true);
682 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
683 dn.data_blkaddr = ei.blk + index - ei.fofs;
684 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
685 DATA_GENERIC_ENHANCE_READ))) {
692 set_new_dnode(&dn, inode, NULL, NULL, 0);
693 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
698 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
702 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
703 DATA_GENERIC_ENHANCE))) {
710 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
713 * don't cache encrypted data into meta inode until previous dirty
714 * data were writebacked to avoid racing between GC and flush.
716 f2fs_wait_on_page_writeback(page, DATA, true, true);
718 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
720 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
722 FGP_LOCK | FGP_CREAT, GFP_NOFS);
723 if (!fio.encrypted_page) {
728 err = f2fs_submit_page_bio(&fio);
730 goto put_encrypted_page;
731 f2fs_put_page(fio.encrypted_page, 0);
732 f2fs_put_page(page, 1);
735 f2fs_put_page(fio.encrypted_page, 1);
737 f2fs_put_page(page, 1);
742 * Move data block via META_MAPPING while keeping locked data page.
743 * This can be used to move blocks, aka LBAs, directly on disk.
745 static int move_data_block(struct inode *inode, block_t bidx,
746 int gc_type, unsigned int segno, int off)
748 struct f2fs_io_info fio = {
749 .sbi = F2FS_I_SB(inode),
755 .encrypted_page = NULL,
759 struct dnode_of_data dn;
760 struct f2fs_summary sum;
762 struct page *page, *mpage;
765 bool lfs_mode = test_opt(fio.sbi, LFS);
767 /* do not read out */
768 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
772 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
777 if (f2fs_is_atomic_file(inode)) {
778 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
779 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
784 if (f2fs_is_pinned_file(inode)) {
785 f2fs_pin_file_control(inode, true);
790 set_new_dnode(&dn, inode, NULL, NULL, 0);
791 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
795 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
796 ClearPageUptodate(page);
802 * don't cache encrypted data into meta inode until previous dirty
803 * data were writebacked to avoid racing between GC and flush.
805 f2fs_wait_on_page_writeback(page, DATA, true, true);
807 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
809 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
813 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
817 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
820 down_write(&fio.sbi->io_order_lock);
822 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
823 fio.old_blkaddr, false);
827 fio.encrypted_page = mpage;
829 /* read source block in mpage */
830 if (!PageUptodate(mpage)) {
831 err = f2fs_submit_page_bio(&fio);
833 f2fs_put_page(mpage, 1);
837 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
838 !PageUptodate(mpage))) {
840 f2fs_put_page(mpage, 1);
845 f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
846 &sum, CURSEG_COLD_DATA, NULL, false);
848 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
849 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
850 if (!fio.encrypted_page) {
852 f2fs_put_page(mpage, 1);
856 /* write target block */
857 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
858 memcpy(page_address(fio.encrypted_page),
859 page_address(mpage), PAGE_SIZE);
860 f2fs_put_page(mpage, 1);
861 invalidate_mapping_pages(META_MAPPING(fio.sbi),
862 fio.old_blkaddr, fio.old_blkaddr);
864 set_page_dirty(fio.encrypted_page);
865 if (clear_page_dirty_for_io(fio.encrypted_page))
866 dec_page_count(fio.sbi, F2FS_DIRTY_META);
868 set_page_writeback(fio.encrypted_page);
869 ClearPageError(page);
871 /* allocate block address */
872 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
874 fio.op = REQ_OP_WRITE;
875 fio.op_flags = REQ_SYNC;
876 fio.new_blkaddr = newaddr;
877 f2fs_submit_page_write(&fio);
880 if (PageWriteback(fio.encrypted_page))
881 end_page_writeback(fio.encrypted_page);
885 f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
887 f2fs_update_data_blkaddr(&dn, newaddr);
888 set_inode_flag(inode, FI_APPEND_WRITE);
889 if (page->index == 0)
890 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
892 f2fs_put_page(fio.encrypted_page, 1);
895 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
899 up_write(&fio.sbi->io_order_lock);
903 f2fs_put_page(page, 1);
907 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
908 unsigned int segno, int off)
913 page = f2fs_get_lock_data_page(inode, bidx, true);
915 return PTR_ERR(page);
917 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
922 if (f2fs_is_atomic_file(inode)) {
923 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
924 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
928 if (f2fs_is_pinned_file(inode)) {
929 if (gc_type == FG_GC)
930 f2fs_pin_file_control(inode, true);
935 if (gc_type == BG_GC) {
936 if (PageWriteback(page)) {
940 set_page_dirty(page);
943 struct f2fs_io_info fio = {
944 .sbi = F2FS_I_SB(inode),
949 .op_flags = REQ_SYNC,
950 .old_blkaddr = NULL_ADDR,
952 .encrypted_page = NULL,
953 .need_lock = LOCK_REQ,
954 .io_type = FS_GC_DATA_IO,
956 bool is_dirty = PageDirty(page);
959 f2fs_wait_on_page_writeback(page, DATA, true, true);
961 set_page_dirty(page);
962 if (clear_page_dirty_for_io(page)) {
963 inode_dec_dirty_pages(inode);
964 f2fs_remove_dirty_inode(inode);
969 err = f2fs_do_write_data_page(&fio);
971 clear_cold_data(page);
972 if (err == -ENOMEM) {
973 congestion_wait(BLK_RW_ASYNC, HZ/50);
977 set_page_dirty(page);
981 f2fs_put_page(page, 1);
986 * This function tries to get parent node of victim data block, and identifies
987 * data block validity. If the block is valid, copy that with cold status and
988 * modify parent node.
989 * If the parent node is not valid or the data block address is different,
990 * the victim data block is ignored.
992 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
993 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
995 struct super_block *sb = sbi->sb;
996 struct f2fs_summary *entry;
1002 start_addr = START_BLOCK(sbi, segno);
1007 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
1008 struct page *data_page;
1009 struct inode *inode;
1010 struct node_info dni; /* dnode info for the data */
1011 unsigned int ofs_in_node, nofs;
1013 nid_t nid = le32_to_cpu(entry->nid);
1016 * stop BG_GC if there is not enough free sections.
1017 * Or, stop GC if the segment becomes fully valid caused by
1018 * race condition along with SSR block allocation.
1020 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1021 get_valid_blocks(sbi, segno, false) ==
1022 sbi->blocks_per_seg)
1025 if (check_valid_map(sbi, segno, off) == 0)
1029 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1035 f2fs_ra_node_page(sbi, nid);
1039 /* Get an inode by ino with checking validity */
1040 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1044 f2fs_ra_node_page(sbi, dni.ino);
1048 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1051 inode = f2fs_iget(sb, dni.ino);
1052 if (IS_ERR(inode) || is_bad_inode(inode))
1055 if (!down_write_trylock(
1056 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1058 sbi->skipped_gc_rwsem++;
1062 start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1065 if (f2fs_post_read_required(inode)) {
1066 int err = ra_data_block(inode, start_bidx);
1068 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1073 add_gc_inode(gc_list, inode);
1077 data_page = f2fs_get_read_data_page(inode,
1078 start_bidx, REQ_RAHEAD, true);
1079 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1080 if (IS_ERR(data_page)) {
1085 f2fs_put_page(data_page, 0);
1086 add_gc_inode(gc_list, inode);
1091 inode = find_gc_inode(gc_list, dni.ino);
1093 struct f2fs_inode_info *fi = F2FS_I(inode);
1094 bool locked = false;
1097 if (S_ISREG(inode->i_mode)) {
1098 if (!down_write_trylock(&fi->i_gc_rwsem[READ]))
1100 if (!down_write_trylock(
1101 &fi->i_gc_rwsem[WRITE])) {
1102 sbi->skipped_gc_rwsem++;
1103 up_write(&fi->i_gc_rwsem[READ]);
1108 /* wait for all inflight aio data */
1109 inode_dio_wait(inode);
1112 start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1114 if (f2fs_post_read_required(inode))
1115 err = move_data_block(inode, start_bidx,
1116 gc_type, segno, off);
1118 err = move_data_page(inode, start_bidx, gc_type,
1121 if (!err && (gc_type == FG_GC ||
1122 f2fs_post_read_required(inode)))
1126 up_write(&fi->i_gc_rwsem[WRITE]);
1127 up_write(&fi->i_gc_rwsem[READ]);
1130 stat_inc_data_blk_count(sbi, 1, gc_type);
1140 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1143 struct sit_info *sit_i = SIT_I(sbi);
1146 down_write(&sit_i->sentry_lock);
1147 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1148 NO_CHECK_TYPE, LFS);
1149 up_write(&sit_i->sentry_lock);
1153 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1154 unsigned int start_segno,
1155 struct gc_inode_list *gc_list, int gc_type)
1157 struct page *sum_page;
1158 struct f2fs_summary_block *sum;
1159 struct blk_plug plug;
1160 unsigned int segno = start_segno;
1161 unsigned int end_segno = start_segno + sbi->segs_per_sec;
1162 int seg_freed = 0, migrated = 0;
1163 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1164 SUM_TYPE_DATA : SUM_TYPE_NODE;
1167 if (__is_large_section(sbi))
1168 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1170 /* readahead multi ssa blocks those have contiguous address */
1171 if (__is_large_section(sbi))
1172 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1173 end_segno - segno, META_SSA, true);
1175 /* reference all summary page */
1176 while (segno < end_segno) {
1177 sum_page = f2fs_get_sum_page(sbi, segno++);
1178 if (IS_ERR(sum_page)) {
1179 int err = PTR_ERR(sum_page);
1181 end_segno = segno - 1;
1182 for (segno = start_segno; segno < end_segno; segno++) {
1183 sum_page = find_get_page(META_MAPPING(sbi),
1184 GET_SUM_BLOCK(sbi, segno));
1185 f2fs_put_page(sum_page, 0);
1186 f2fs_put_page(sum_page, 0);
1190 unlock_page(sum_page);
1193 blk_start_plug(&plug);
1195 for (segno = start_segno; segno < end_segno; segno++) {
1197 /* find segment summary of victim */
1198 sum_page = find_get_page(META_MAPPING(sbi),
1199 GET_SUM_BLOCK(sbi, segno));
1200 f2fs_put_page(sum_page, 0);
1202 if (get_valid_blocks(sbi, segno, false) == 0)
1204 if (__is_large_section(sbi) &&
1205 migrated >= sbi->migration_granularity)
1207 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1210 sum = page_address(sum_page);
1211 if (type != GET_SUM_TYPE((&sum->footer))) {
1212 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1213 segno, type, GET_SUM_TYPE((&sum->footer)));
1214 set_sbi_flag(sbi, SBI_NEED_FSCK);
1215 f2fs_stop_checkpoint(sbi, false);
1220 * this is to avoid deadlock:
1221 * - lock_page(sum_page) - f2fs_replace_block
1222 * - check_valid_map() - down_write(sentry_lock)
1223 * - down_read(sentry_lock) - change_curseg()
1224 * - lock_page(sum_page)
1226 if (type == SUM_TYPE_NODE)
1227 submitted += gc_node_segment(sbi, sum->entries, segno,
1230 submitted += gc_data_segment(sbi, sum->entries, gc_list,
1233 stat_inc_seg_count(sbi, type, gc_type);
1236 if (gc_type == FG_GC &&
1237 get_valid_blocks(sbi, segno, false) == 0)
1241 if (__is_large_section(sbi) && segno + 1 < end_segno)
1242 sbi->next_victim_seg[gc_type] = segno + 1;
1244 f2fs_put_page(sum_page, 0);
1248 f2fs_submit_merged_write(sbi,
1249 (type == SUM_TYPE_NODE) ? NODE : DATA);
1251 blk_finish_plug(&plug);
1253 stat_inc_call_count(sbi->stat_info);
1258 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1259 bool background, unsigned int segno)
1261 int gc_type = sync ? FG_GC : BG_GC;
1262 int sec_freed = 0, seg_freed = 0, total_freed = 0;
1264 struct cp_control cpc;
1265 unsigned int init_segno = segno;
1266 struct gc_inode_list gc_list = {
1267 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1268 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1270 unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1271 unsigned long long first_skipped;
1272 unsigned int skipped_round = 0, round = 0;
1274 trace_f2fs_gc_begin(sbi->sb, sync, background,
1275 get_pages(sbi, F2FS_DIRTY_NODES),
1276 get_pages(sbi, F2FS_DIRTY_DENTS),
1277 get_pages(sbi, F2FS_DIRTY_IMETA),
1280 reserved_segments(sbi),
1281 prefree_segments(sbi));
1283 cpc.reason = __get_cp_reason(sbi);
1284 sbi->skipped_gc_rwsem = 0;
1285 first_skipped = last_skipped;
1287 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1291 if (unlikely(f2fs_cp_error(sbi))) {
1296 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1298 * For example, if there are many prefree_segments below given
1299 * threshold, we can make them free by checkpoint. Then, we
1300 * secure free segments which doesn't need fggc any more.
1302 if (prefree_segments(sbi) &&
1303 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1304 ret = f2fs_write_checkpoint(sbi, &cpc);
1308 if (has_not_enough_free_secs(sbi, 0, 0))
1312 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1313 if (gc_type == BG_GC && !background) {
1317 if (!__get_victim(sbi, &segno, gc_type)) {
1322 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1323 if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1325 total_freed += seg_freed;
1327 if (gc_type == FG_GC) {
1328 if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1329 sbi->skipped_gc_rwsem)
1331 last_skipped = sbi->skipped_atomic_files[FG_GC];
1335 if (gc_type == FG_GC && seg_freed)
1336 sbi->cur_victim_sec = NULL_SEGNO;
1341 if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1342 if (skipped_round <= MAX_SKIP_GC_COUNT ||
1343 skipped_round * 2 < round) {
1348 if (first_skipped < last_skipped &&
1349 (last_skipped - first_skipped) >
1350 sbi->skipped_gc_rwsem) {
1351 f2fs_drop_inmem_pages_all(sbi, true);
1355 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1356 ret = f2fs_write_checkpoint(sbi, &cpc);
1359 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1360 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1362 trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1363 get_pages(sbi, F2FS_DIRTY_NODES),
1364 get_pages(sbi, F2FS_DIRTY_DENTS),
1365 get_pages(sbi, F2FS_DIRTY_IMETA),
1368 reserved_segments(sbi),
1369 prefree_segments(sbi));
1371 mutex_unlock(&sbi->gc_mutex);
1373 put_gc_inode(&gc_list);
1376 ret = sec_freed ? 0 : -EAGAIN;
1380 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1382 DIRTY_I(sbi)->v_ops = &default_v_ops;
1384 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1386 /* give warm/cold data area from slower device */
1387 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1388 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1389 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1392 static int free_segment_range(struct f2fs_sb_info *sbi, unsigned int start,
1396 unsigned int segno, next_inuse;
1399 /* Move out cursegs from the target range */
1400 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_TYPE; type++)
1401 allocate_segment_for_resize(sbi, type, start, end);
1403 /* do GC to move out valid blocks in the range */
1404 for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1405 struct gc_inode_list gc_list = {
1406 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1407 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1410 mutex_lock(&sbi->gc_mutex);
1411 do_garbage_collect(sbi, segno, &gc_list, FG_GC);
1412 mutex_unlock(&sbi->gc_mutex);
1413 put_gc_inode(&gc_list);
1415 if (get_valid_blocks(sbi, segno, true))
1419 err = f2fs_sync_fs(sbi->sb, 1);
1423 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1424 if (next_inuse <= end) {
1425 f2fs_err(sbi, "segno %u should be free but still inuse!",
1427 f2fs_bug_on(sbi, 1);
1432 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1434 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1435 int section_count = le32_to_cpu(raw_sb->section_count);
1436 int segment_count = le32_to_cpu(raw_sb->segment_count);
1437 int segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1438 long long block_count = le64_to_cpu(raw_sb->block_count);
1439 int segs = secs * sbi->segs_per_sec;
1441 raw_sb->section_count = cpu_to_le32(section_count + secs);
1442 raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1443 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1444 raw_sb->block_count = cpu_to_le64(block_count +
1445 (long long)segs * sbi->blocks_per_seg);
1446 if (f2fs_is_multi_device(sbi)) {
1447 int last_dev = sbi->s_ndevs - 1;
1449 le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1451 raw_sb->devs[last_dev].total_segments =
1452 cpu_to_le32(dev_segs + segs);
1456 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1458 int segs = secs * sbi->segs_per_sec;
1459 long long blks = (long long)segs * sbi->blocks_per_seg;
1460 long long user_block_count =
1461 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1463 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1464 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1465 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1466 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1467 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1469 if (f2fs_is_multi_device(sbi)) {
1470 int last_dev = sbi->s_ndevs - 1;
1472 FDEV(last_dev).total_segments =
1473 (int)FDEV(last_dev).total_segments + segs;
1474 FDEV(last_dev).end_blk =
1475 (long long)FDEV(last_dev).end_blk + blks;
1476 #ifdef CONFIG_BLK_DEV_ZONED
1477 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1478 (int)(blks >> sbi->log_blocks_per_blkz);
1483 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1485 __u64 old_block_count, shrunk_blocks;
1487 int gc_mode, gc_type;
1491 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1492 if (block_count > old_block_count)
1495 if (f2fs_is_multi_device(sbi)) {
1496 int last_dev = sbi->s_ndevs - 1;
1497 __u64 last_segs = FDEV(last_dev).total_segments;
1499 if (block_count + last_segs * sbi->blocks_per_seg <=
1504 /* new fs size should align to section size */
1505 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
1509 if (block_count == old_block_count)
1512 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1513 f2fs_err(sbi, "Should run fsck to repair first.");
1514 return -EFSCORRUPTED;
1517 if (test_opt(sbi, DISABLE_CHECKPOINT)) {
1518 f2fs_err(sbi, "Checkpoint should be enabled.");
1522 freeze_bdev(sbi->sb->s_bdev);
1524 shrunk_blocks = old_block_count - block_count;
1525 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
1526 spin_lock(&sbi->stat_lock);
1527 if (shrunk_blocks + valid_user_blocks(sbi) +
1528 sbi->current_reserved_blocks + sbi->unusable_block_count +
1529 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
1532 sbi->user_block_count -= shrunk_blocks;
1533 spin_unlock(&sbi->stat_lock);
1535 thaw_bdev(sbi->sb->s_bdev, sbi->sb);
1539 mutex_lock(&sbi->resize_mutex);
1540 set_sbi_flag(sbi, SBI_IS_RESIZEFS);
1542 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1544 MAIN_SECS(sbi) -= secs;
1546 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1547 if (SIT_I(sbi)->last_victim[gc_mode] >=
1548 MAIN_SECS(sbi) * sbi->segs_per_sec)
1549 SIT_I(sbi)->last_victim[gc_mode] = 0;
1551 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1552 if (sbi->next_victim_seg[gc_type] >=
1553 MAIN_SECS(sbi) * sbi->segs_per_sec)
1554 sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1556 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1558 err = free_segment_range(sbi, MAIN_SECS(sbi) * sbi->segs_per_sec,
1559 MAIN_SEGS(sbi) - 1);
1563 update_sb_metadata(sbi, -secs);
1565 err = f2fs_commit_super(sbi, false);
1567 update_sb_metadata(sbi, secs);
1571 update_fs_metadata(sbi, -secs);
1572 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
1573 err = f2fs_sync_fs(sbi->sb, 1);
1575 update_fs_metadata(sbi, secs);
1576 update_sb_metadata(sbi, secs);
1577 f2fs_commit_super(sbi, false);
1581 set_sbi_flag(sbi, SBI_NEED_FSCK);
1582 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
1584 MAIN_SECS(sbi) += secs;
1585 spin_lock(&sbi->stat_lock);
1586 sbi->user_block_count += shrunk_blocks;
1587 spin_unlock(&sbi->stat_lock);
1589 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
1590 mutex_unlock(&sbi->resize_mutex);
1591 thaw_bdev(sbi->sb->s_bdev, sbi->sb);