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;
36 wait_event_interruptible_timeout(*wq,
37 kthread_should_stop() || freezing(current) ||
39 msecs_to_jiffies(wait_ms));
41 /* give it a try one time */
45 if (try_to_freeze()) {
46 stat_other_skip_bggc_count(sbi);
49 if (kthread_should_stop())
52 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
53 increase_sleep_time(gc_th, &wait_ms);
54 stat_other_skip_bggc_count(sbi);
58 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
59 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
60 f2fs_stop_checkpoint(sbi, false);
63 if (!sb_start_write_trylock(sbi->sb)) {
64 stat_other_skip_bggc_count(sbi);
69 * [GC triggering condition]
70 * 0. GC is not conducted currently.
71 * 1. There are enough dirty segments.
72 * 2. IO subsystem is idle by checking the # of writeback pages.
73 * 3. IO subsystem is idle by checking the # of requests in
74 * bdev's request list.
76 * Note) We have to avoid triggering GCs frequently.
77 * Because it is possible that some segments can be
78 * invalidated soon after by user update or deletion.
79 * So, I'd like to wait some time to collect dirty segments.
81 if (sbi->gc_mode == GC_URGENT) {
82 wait_ms = gc_th->urgent_sleep_time;
83 down_write(&sbi->gc_lock);
87 if (!down_write_trylock(&sbi->gc_lock)) {
88 stat_other_skip_bggc_count(sbi);
92 if (!is_idle(sbi, GC_TIME)) {
93 increase_sleep_time(gc_th, &wait_ms);
94 up_write(&sbi->gc_lock);
95 stat_io_skip_bggc_count(sbi);
99 if (has_enough_invalid_blocks(sbi))
100 decrease_sleep_time(gc_th, &wait_ms);
102 increase_sleep_time(gc_th, &wait_ms);
104 stat_inc_bggc_count(sbi->stat_info);
106 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
108 /* if return value is not zero, no victim was selected */
109 if (f2fs_gc(sbi, sync_mode, true, NULL_SEGNO))
110 wait_ms = gc_th->no_gc_sleep_time;
112 trace_f2fs_background_gc(sbi->sb, wait_ms,
113 prefree_segments(sbi), free_segments(sbi));
115 /* balancing f2fs's metadata periodically */
116 f2fs_balance_fs_bg(sbi, true);
118 sb_end_write(sbi->sb);
120 } while (!kthread_should_stop());
124 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
126 struct f2fs_gc_kthread *gc_th;
127 dev_t dev = sbi->sb->s_bdev->bd_dev;
130 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
136 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
137 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
138 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
139 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
143 sbi->gc_thread = gc_th;
144 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
145 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
146 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
147 if (IS_ERR(gc_th->f2fs_gc_task)) {
148 err = PTR_ERR(gc_th->f2fs_gc_task);
150 sbi->gc_thread = NULL;
156 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
158 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
161 kthread_stop(gc_th->f2fs_gc_task);
163 sbi->gc_thread = NULL;
166 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
168 int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
170 switch (sbi->gc_mode) {
182 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
183 int type, struct victim_sel_policy *p)
185 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
187 if (p->alloc_mode == SSR) {
188 p->gc_mode = GC_GREEDY;
189 p->dirty_segmap = dirty_i->dirty_segmap[type];
190 p->max_search = dirty_i->nr_dirty[type];
193 p->gc_mode = select_gc_type(sbi, gc_type);
194 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
195 p->max_search = dirty_i->nr_dirty[DIRTY];
196 p->ofs_unit = sbi->segs_per_sec;
200 * adjust candidates range, should select all dirty segments for
201 * foreground GC and urgent GC cases.
203 if (gc_type != FG_GC &&
204 (sbi->gc_mode != GC_URGENT) &&
205 p->max_search > sbi->max_victim_search)
206 p->max_search = sbi->max_victim_search;
208 /* let's select beginning hot/small space first in no_heap mode*/
209 if (test_opt(sbi, NOHEAP) &&
210 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
213 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
216 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
217 struct victim_sel_policy *p)
219 /* SSR allocates in a segment unit */
220 if (p->alloc_mode == SSR)
221 return sbi->blocks_per_seg;
222 if (p->gc_mode == GC_GREEDY)
223 return 2 * sbi->blocks_per_seg * p->ofs_unit;
224 else if (p->gc_mode == GC_CB)
226 else /* No other gc_mode */
230 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
232 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
236 * If the gc_type is FG_GC, we can select victim segments
237 * selected by background GC before.
238 * Those segments guarantee they have small valid blocks.
240 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
241 if (sec_usage_check(sbi, secno))
243 clear_bit(secno, dirty_i->victim_secmap);
244 return GET_SEG_FROM_SEC(sbi, secno);
249 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
251 struct sit_info *sit_i = SIT_I(sbi);
252 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
253 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
254 unsigned long long mtime = 0;
255 unsigned int vblocks;
256 unsigned char age = 0;
260 for (i = 0; i < sbi->segs_per_sec; i++)
261 mtime += get_seg_entry(sbi, start + i)->mtime;
262 vblocks = get_valid_blocks(sbi, segno, true);
264 mtime = div_u64(mtime, sbi->segs_per_sec);
265 vblocks = div_u64(vblocks, sbi->segs_per_sec);
267 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
269 /* Handle if the system time has changed by the user */
270 if (mtime < sit_i->min_mtime)
271 sit_i->min_mtime = mtime;
272 if (mtime > sit_i->max_mtime)
273 sit_i->max_mtime = mtime;
274 if (sit_i->max_mtime != sit_i->min_mtime)
275 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
276 sit_i->max_mtime - sit_i->min_mtime);
278 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
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_valid_blocks(sbi, segno, true);
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;
322 unsigned int nsearched = 0;
324 mutex_lock(&dirty_i->seglist_lock);
325 last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
327 p.alloc_mode = alloc_mode;
328 select_policy(sbi, gc_type, type, &p);
330 p.min_segno = NULL_SEGNO;
331 p.min_cost = get_max_cost(sbi, &p);
333 if (*result != NULL_SEGNO) {
334 if (get_valid_blocks(sbi, *result, false) &&
335 !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
336 p.min_segno = *result;
340 if (p.max_search == 0)
343 if (__is_large_section(sbi) && p.alloc_mode == LFS) {
344 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
345 p.min_segno = sbi->next_victim_seg[BG_GC];
346 *result = p.min_segno;
347 sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
350 if (gc_type == FG_GC &&
351 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
352 p.min_segno = sbi->next_victim_seg[FG_GC];
353 *result = p.min_segno;
354 sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
359 last_victim = sm->last_victim[p.gc_mode];
360 if (p.alloc_mode == LFS && gc_type == FG_GC) {
361 p.min_segno = check_bg_victims(sbi);
362 if (p.min_segno != NULL_SEGNO)
370 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
371 if (segno >= last_segment) {
372 if (sm->last_victim[p.gc_mode]) {
374 sm->last_victim[p.gc_mode];
375 sm->last_victim[p.gc_mode] = 0;
382 p.offset = segno + p.ofs_unit;
383 if (p.ofs_unit > 1) {
384 p.offset -= segno % p.ofs_unit;
385 nsearched += count_bits(p.dirty_segmap,
386 p.offset - p.ofs_unit,
392 #ifdef CONFIG_F2FS_CHECK_FS
394 * skip selecting the invalid segno (that is failed due to block
395 * validity check failure during GC) to avoid endless GC loop in
398 if (test_bit(segno, sm->invalid_segmap))
402 secno = GET_SEC_FROM_SEG(sbi, segno);
404 if (sec_usage_check(sbi, secno))
406 /* Don't touch checkpointed data */
407 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
408 get_ckpt_valid_blocks(sbi, segno) &&
409 p.alloc_mode != SSR))
411 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
414 cost = get_gc_cost(sbi, segno, &p);
416 if (p.min_cost > cost) {
421 if (nsearched >= p.max_search) {
422 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
423 sm->last_victim[p.gc_mode] = last_victim + 1;
425 sm->last_victim[p.gc_mode] = segno + 1;
426 sm->last_victim[p.gc_mode] %=
427 (MAIN_SECS(sbi) * sbi->segs_per_sec);
431 if (p.min_segno != NULL_SEGNO) {
433 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
435 if (p.alloc_mode == LFS) {
436 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
437 if (gc_type == FG_GC)
438 sbi->cur_victim_sec = secno;
440 set_bit(secno, dirty_i->victim_secmap);
445 if (p.min_segno != NULL_SEGNO)
446 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
448 prefree_segments(sbi), free_segments(sbi));
449 mutex_unlock(&dirty_i->seglist_lock);
451 return (p.min_segno == NULL_SEGNO) ? 0 : 1;
454 static const struct victim_selection default_v_ops = {
455 .get_victim = get_victim_by_default,
458 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
460 struct inode_entry *ie;
462 ie = radix_tree_lookup(&gc_list->iroot, ino);
468 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
470 struct inode_entry *new_ie;
472 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
476 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
477 new_ie->inode = inode;
479 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
480 list_add_tail(&new_ie->list, &gc_list->ilist);
483 static void put_gc_inode(struct gc_inode_list *gc_list)
485 struct inode_entry *ie, *next_ie;
486 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
487 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
490 kmem_cache_free(f2fs_inode_entry_slab, ie);
494 static int check_valid_map(struct f2fs_sb_info *sbi,
495 unsigned int segno, int offset)
497 struct sit_info *sit_i = SIT_I(sbi);
498 struct seg_entry *sentry;
501 down_read(&sit_i->sentry_lock);
502 sentry = get_seg_entry(sbi, segno);
503 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
504 up_read(&sit_i->sentry_lock);
509 * This function compares node address got in summary with that in NAT.
510 * On validity, copy that node with cold status, otherwise (invalid node)
513 static int gc_node_segment(struct f2fs_sb_info *sbi,
514 struct f2fs_summary *sum, unsigned int segno, int gc_type)
516 struct f2fs_summary *entry;
520 bool fggc = (gc_type == FG_GC);
523 start_addr = START_BLOCK(sbi, segno);
528 if (fggc && phase == 2)
529 atomic_inc(&sbi->wb_sync_req[NODE]);
531 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
532 nid_t nid = le32_to_cpu(entry->nid);
533 struct page *node_page;
537 /* stop BG_GC if there is not enough free sections. */
538 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
541 if (check_valid_map(sbi, segno, off) == 0)
545 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
551 f2fs_ra_node_page(sbi, nid);
556 node_page = f2fs_get_node_page(sbi, nid);
557 if (IS_ERR(node_page))
560 /* block may become invalid during f2fs_get_node_page */
561 if (check_valid_map(sbi, segno, off) == 0) {
562 f2fs_put_page(node_page, 1);
566 if (f2fs_get_node_info(sbi, nid, &ni)) {
567 f2fs_put_page(node_page, 1);
571 if (ni.blk_addr != start_addr + off) {
572 f2fs_put_page(node_page, 1);
576 err = f2fs_move_node_page(node_page, gc_type);
577 if (!err && gc_type == FG_GC)
579 stat_inc_node_blk_count(sbi, 1, gc_type);
586 atomic_dec(&sbi->wb_sync_req[NODE]);
591 * Calculate start block index indicating the given node offset.
592 * Be careful, caller should give this node offset only indicating direct node
593 * blocks. If any node offsets, which point the other types of node blocks such
594 * as indirect or double indirect node blocks, are given, it must be a caller's
597 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
599 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
607 } else if (node_ofs <= indirect_blks) {
608 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
609 bidx = node_ofs - 2 - dec;
611 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
612 bidx = node_ofs - 5 - dec;
614 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
617 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
618 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
620 struct page *node_page;
622 unsigned int ofs_in_node;
623 block_t source_blkaddr;
625 nid = le32_to_cpu(sum->nid);
626 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
628 node_page = f2fs_get_node_page(sbi, nid);
629 if (IS_ERR(node_page))
632 if (f2fs_get_node_info(sbi, nid, dni)) {
633 f2fs_put_page(node_page, 1);
637 if (sum->version != dni->version) {
638 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
640 set_sbi_flag(sbi, SBI_NEED_FSCK);
643 *nofs = ofs_of_node(node_page);
644 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
645 f2fs_put_page(node_page, 1);
647 if (source_blkaddr != blkaddr) {
648 #ifdef CONFIG_F2FS_CHECK_FS
649 unsigned int segno = GET_SEGNO(sbi, blkaddr);
650 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
652 if (unlikely(check_valid_map(sbi, segno, offset))) {
653 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
654 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n",
655 blkaddr, source_blkaddr, segno);
665 static int ra_data_block(struct inode *inode, pgoff_t index)
667 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
668 struct address_space *mapping = inode->i_mapping;
669 struct dnode_of_data dn;
671 struct extent_info ei = {0, 0, 0};
672 struct f2fs_io_info fio = {
679 .encrypted_page = NULL,
685 page = f2fs_grab_cache_page(mapping, index, true);
689 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
690 dn.data_blkaddr = ei.blk + index - ei.fofs;
691 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
692 DATA_GENERIC_ENHANCE_READ))) {
699 set_new_dnode(&dn, inode, NULL, NULL, 0);
700 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
705 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
709 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
710 DATA_GENERIC_ENHANCE))) {
717 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
720 * don't cache encrypted data into meta inode until previous dirty
721 * data were writebacked to avoid racing between GC and flush.
723 f2fs_wait_on_page_writeback(page, DATA, true, true);
725 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
727 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
729 FGP_LOCK | FGP_CREAT, GFP_NOFS);
730 if (!fio.encrypted_page) {
735 err = f2fs_submit_page_bio(&fio);
737 goto put_encrypted_page;
738 f2fs_put_page(fio.encrypted_page, 0);
739 f2fs_put_page(page, 1);
742 f2fs_put_page(fio.encrypted_page, 1);
744 f2fs_put_page(page, 1);
749 * Move data block via META_MAPPING while keeping locked data page.
750 * This can be used to move blocks, aka LBAs, directly on disk.
752 static int move_data_block(struct inode *inode, block_t bidx,
753 int gc_type, unsigned int segno, int off)
755 struct f2fs_io_info fio = {
756 .sbi = F2FS_I_SB(inode),
762 .encrypted_page = NULL,
766 struct dnode_of_data dn;
767 struct f2fs_summary sum;
769 struct page *page, *mpage;
772 bool lfs_mode = f2fs_lfs_mode(fio.sbi);
774 /* do not read out */
775 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
779 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
784 if (f2fs_is_atomic_file(inode)) {
785 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
786 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
791 if (f2fs_is_pinned_file(inode)) {
792 f2fs_pin_file_control(inode, true);
797 set_new_dnode(&dn, inode, NULL, NULL, 0);
798 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
802 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
803 ClearPageUptodate(page);
809 * don't cache encrypted data into meta inode until previous dirty
810 * data were writebacked to avoid racing between GC and flush.
812 f2fs_wait_on_page_writeback(page, DATA, true, true);
814 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
816 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
820 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
824 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
827 down_write(&fio.sbi->io_order_lock);
829 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
830 fio.old_blkaddr, false);
834 fio.encrypted_page = mpage;
836 /* read source block in mpage */
837 if (!PageUptodate(mpage)) {
838 err = f2fs_submit_page_bio(&fio);
840 f2fs_put_page(mpage, 1);
844 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
845 !PageUptodate(mpage))) {
847 f2fs_put_page(mpage, 1);
852 f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
853 &sum, CURSEG_COLD_DATA, NULL, false);
855 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
856 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
857 if (!fio.encrypted_page) {
859 f2fs_put_page(mpage, 1);
863 /* write target block */
864 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
865 memcpy(page_address(fio.encrypted_page),
866 page_address(mpage), PAGE_SIZE);
867 f2fs_put_page(mpage, 1);
868 invalidate_mapping_pages(META_MAPPING(fio.sbi),
869 fio.old_blkaddr, fio.old_blkaddr);
871 set_page_dirty(fio.encrypted_page);
872 if (clear_page_dirty_for_io(fio.encrypted_page))
873 dec_page_count(fio.sbi, F2FS_DIRTY_META);
875 set_page_writeback(fio.encrypted_page);
876 ClearPageError(page);
878 /* allocate block address */
879 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
881 fio.op = REQ_OP_WRITE;
882 fio.op_flags = REQ_SYNC;
883 fio.new_blkaddr = newaddr;
884 f2fs_submit_page_write(&fio);
887 if (PageWriteback(fio.encrypted_page))
888 end_page_writeback(fio.encrypted_page);
892 f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
894 f2fs_update_data_blkaddr(&dn, newaddr);
895 set_inode_flag(inode, FI_APPEND_WRITE);
896 if (page->index == 0)
897 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
899 f2fs_put_page(fio.encrypted_page, 1);
902 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
906 up_write(&fio.sbi->io_order_lock);
910 f2fs_put_page(page, 1);
914 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
915 unsigned int segno, int off)
920 page = f2fs_get_lock_data_page(inode, bidx, true);
922 return PTR_ERR(page);
924 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
929 if (f2fs_is_atomic_file(inode)) {
930 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
931 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
935 if (f2fs_is_pinned_file(inode)) {
936 if (gc_type == FG_GC)
937 f2fs_pin_file_control(inode, true);
942 if (gc_type == BG_GC) {
943 if (PageWriteback(page)) {
947 set_page_dirty(page);
950 struct f2fs_io_info fio = {
951 .sbi = F2FS_I_SB(inode),
956 .op_flags = REQ_SYNC,
957 .old_blkaddr = NULL_ADDR,
959 .encrypted_page = NULL,
960 .need_lock = LOCK_REQ,
961 .io_type = FS_GC_DATA_IO,
963 bool is_dirty = PageDirty(page);
966 f2fs_wait_on_page_writeback(page, DATA, true, true);
968 set_page_dirty(page);
969 if (clear_page_dirty_for_io(page)) {
970 inode_dec_dirty_pages(inode);
971 f2fs_remove_dirty_inode(inode);
976 err = f2fs_do_write_data_page(&fio);
978 clear_cold_data(page);
979 if (err == -ENOMEM) {
980 congestion_wait(BLK_RW_ASYNC,
985 set_page_dirty(page);
989 f2fs_put_page(page, 1);
994 * This function tries to get parent node of victim data block, and identifies
995 * data block validity. If the block is valid, copy that with cold status and
996 * modify parent node.
997 * If the parent node is not valid or the data block address is different,
998 * the victim data block is ignored.
1000 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1001 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
1003 struct super_block *sb = sbi->sb;
1004 struct f2fs_summary *entry;
1010 start_addr = START_BLOCK(sbi, segno);
1015 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
1016 struct page *data_page;
1017 struct inode *inode;
1018 struct node_info dni; /* dnode info for the data */
1019 unsigned int ofs_in_node, nofs;
1021 nid_t nid = le32_to_cpu(entry->nid);
1024 * stop BG_GC if there is not enough free sections.
1025 * Or, stop GC if the segment becomes fully valid caused by
1026 * race condition along with SSR block allocation.
1028 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1029 get_valid_blocks(sbi, segno, true) ==
1033 if (check_valid_map(sbi, segno, off) == 0)
1037 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1043 f2fs_ra_node_page(sbi, nid);
1047 /* Get an inode by ino with checking validity */
1048 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1052 f2fs_ra_node_page(sbi, dni.ino);
1056 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1059 inode = f2fs_iget(sb, dni.ino);
1060 if (IS_ERR(inode) || is_bad_inode(inode)) {
1061 set_sbi_flag(sbi, SBI_NEED_FSCK);
1065 if (!down_write_trylock(
1066 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1068 sbi->skipped_gc_rwsem++;
1072 start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1075 if (f2fs_post_read_required(inode)) {
1076 int err = ra_data_block(inode, start_bidx);
1078 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1083 add_gc_inode(gc_list, inode);
1087 data_page = f2fs_get_read_data_page(inode,
1088 start_bidx, REQ_RAHEAD, true);
1089 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1090 if (IS_ERR(data_page)) {
1095 f2fs_put_page(data_page, 0);
1096 add_gc_inode(gc_list, inode);
1101 inode = find_gc_inode(gc_list, dni.ino);
1103 struct f2fs_inode_info *fi = F2FS_I(inode);
1104 bool locked = false;
1107 if (S_ISREG(inode->i_mode)) {
1108 if (!down_write_trylock(&fi->i_gc_rwsem[READ]))
1110 if (!down_write_trylock(
1111 &fi->i_gc_rwsem[WRITE])) {
1112 sbi->skipped_gc_rwsem++;
1113 up_write(&fi->i_gc_rwsem[READ]);
1118 /* wait for all inflight aio data */
1119 inode_dio_wait(inode);
1122 start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1124 if (f2fs_post_read_required(inode))
1125 err = move_data_block(inode, start_bidx,
1126 gc_type, segno, off);
1128 err = move_data_page(inode, start_bidx, gc_type,
1131 if (!err && (gc_type == FG_GC ||
1132 f2fs_post_read_required(inode)))
1136 up_write(&fi->i_gc_rwsem[WRITE]);
1137 up_write(&fi->i_gc_rwsem[READ]);
1140 stat_inc_data_blk_count(sbi, 1, gc_type);
1150 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1153 struct sit_info *sit_i = SIT_I(sbi);
1156 down_write(&sit_i->sentry_lock);
1157 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1158 NO_CHECK_TYPE, LFS);
1159 up_write(&sit_i->sentry_lock);
1163 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1164 unsigned int start_segno,
1165 struct gc_inode_list *gc_list, int gc_type)
1167 struct page *sum_page;
1168 struct f2fs_summary_block *sum;
1169 struct blk_plug plug;
1170 unsigned int segno = start_segno;
1171 unsigned int end_segno = start_segno + sbi->segs_per_sec;
1172 int seg_freed = 0, migrated = 0;
1173 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1174 SUM_TYPE_DATA : SUM_TYPE_NODE;
1177 if (__is_large_section(sbi))
1178 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1180 /* readahead multi ssa blocks those have contiguous address */
1181 if (__is_large_section(sbi))
1182 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1183 end_segno - segno, META_SSA, true);
1185 /* reference all summary page */
1186 while (segno < end_segno) {
1187 sum_page = f2fs_get_sum_page(sbi, segno++);
1188 if (IS_ERR(sum_page)) {
1189 int err = PTR_ERR(sum_page);
1191 end_segno = segno - 1;
1192 for (segno = start_segno; segno < end_segno; segno++) {
1193 sum_page = find_get_page(META_MAPPING(sbi),
1194 GET_SUM_BLOCK(sbi, segno));
1195 f2fs_put_page(sum_page, 0);
1196 f2fs_put_page(sum_page, 0);
1200 unlock_page(sum_page);
1203 blk_start_plug(&plug);
1205 for (segno = start_segno; segno < end_segno; segno++) {
1207 /* find segment summary of victim */
1208 sum_page = find_get_page(META_MAPPING(sbi),
1209 GET_SUM_BLOCK(sbi, segno));
1210 f2fs_put_page(sum_page, 0);
1212 if (get_valid_blocks(sbi, segno, false) == 0)
1214 if (gc_type == BG_GC && __is_large_section(sbi) &&
1215 migrated >= sbi->migration_granularity)
1217 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1220 sum = page_address(sum_page);
1221 if (type != GET_SUM_TYPE((&sum->footer))) {
1222 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1223 segno, type, GET_SUM_TYPE((&sum->footer)));
1224 set_sbi_flag(sbi, SBI_NEED_FSCK);
1225 f2fs_stop_checkpoint(sbi, false);
1230 * this is to avoid deadlock:
1231 * - lock_page(sum_page) - f2fs_replace_block
1232 * - check_valid_map() - down_write(sentry_lock)
1233 * - down_read(sentry_lock) - change_curseg()
1234 * - lock_page(sum_page)
1236 if (type == SUM_TYPE_NODE)
1237 submitted += gc_node_segment(sbi, sum->entries, segno,
1240 submitted += gc_data_segment(sbi, sum->entries, gc_list,
1243 stat_inc_seg_count(sbi, type, gc_type);
1247 if (gc_type == FG_GC &&
1248 get_valid_blocks(sbi, segno, false) == 0)
1251 if (__is_large_section(sbi) && segno + 1 < end_segno)
1252 sbi->next_victim_seg[gc_type] = segno + 1;
1254 f2fs_put_page(sum_page, 0);
1258 f2fs_submit_merged_write(sbi,
1259 (type == SUM_TYPE_NODE) ? NODE : DATA);
1261 blk_finish_plug(&plug);
1263 stat_inc_call_count(sbi->stat_info);
1268 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1269 bool background, unsigned int segno)
1271 int gc_type = sync ? FG_GC : BG_GC;
1272 int sec_freed = 0, seg_freed = 0, total_freed = 0;
1274 struct cp_control cpc;
1275 unsigned int init_segno = segno;
1276 struct gc_inode_list gc_list = {
1277 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1278 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1280 unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1281 unsigned long long first_skipped;
1282 unsigned int skipped_round = 0, round = 0;
1284 trace_f2fs_gc_begin(sbi->sb, sync, background,
1285 get_pages(sbi, F2FS_DIRTY_NODES),
1286 get_pages(sbi, F2FS_DIRTY_DENTS),
1287 get_pages(sbi, F2FS_DIRTY_IMETA),
1290 reserved_segments(sbi),
1291 prefree_segments(sbi));
1293 cpc.reason = __get_cp_reason(sbi);
1294 sbi->skipped_gc_rwsem = 0;
1295 first_skipped = last_skipped;
1297 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1301 if (unlikely(f2fs_cp_error(sbi))) {
1306 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1308 * For example, if there are many prefree_segments below given
1309 * threshold, we can make them free by checkpoint. Then, we
1310 * secure free segments which doesn't need fggc any more.
1312 if (prefree_segments(sbi) &&
1313 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1314 ret = f2fs_write_checkpoint(sbi, &cpc);
1318 if (has_not_enough_free_secs(sbi, 0, 0))
1322 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1323 if (gc_type == BG_GC && !background) {
1327 if (!__get_victim(sbi, &segno, gc_type)) {
1332 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1333 if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1335 total_freed += seg_freed;
1337 if (gc_type == FG_GC) {
1338 if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1339 sbi->skipped_gc_rwsem)
1341 last_skipped = sbi->skipped_atomic_files[FG_GC];
1345 if (gc_type == FG_GC && seg_freed)
1346 sbi->cur_victim_sec = NULL_SEGNO;
1351 if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1352 if (skipped_round <= MAX_SKIP_GC_COUNT ||
1353 skipped_round * 2 < round) {
1358 if (first_skipped < last_skipped &&
1359 (last_skipped - first_skipped) >
1360 sbi->skipped_gc_rwsem) {
1361 f2fs_drop_inmem_pages_all(sbi, true);
1365 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1366 ret = f2fs_write_checkpoint(sbi, &cpc);
1369 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1370 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1372 trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1373 get_pages(sbi, F2FS_DIRTY_NODES),
1374 get_pages(sbi, F2FS_DIRTY_DENTS),
1375 get_pages(sbi, F2FS_DIRTY_IMETA),
1378 reserved_segments(sbi),
1379 prefree_segments(sbi));
1381 up_write(&sbi->gc_lock);
1383 put_gc_inode(&gc_list);
1386 ret = sec_freed ? 0 : -EAGAIN;
1390 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1392 DIRTY_I(sbi)->v_ops = &default_v_ops;
1394 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1396 /* give warm/cold data area from slower device */
1397 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1398 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1399 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1402 static int free_segment_range(struct f2fs_sb_info *sbi, unsigned int start,
1406 unsigned int segno, next_inuse;
1409 /* Move out cursegs from the target range */
1410 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_TYPE; type++)
1411 allocate_segment_for_resize(sbi, type, start, end);
1413 /* do GC to move out valid blocks in the range */
1414 for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1415 struct gc_inode_list gc_list = {
1416 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1417 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1420 down_write(&sbi->gc_lock);
1421 do_garbage_collect(sbi, segno, &gc_list, FG_GC);
1422 up_write(&sbi->gc_lock);
1423 put_gc_inode(&gc_list);
1425 if (get_valid_blocks(sbi, segno, true))
1429 err = f2fs_sync_fs(sbi->sb, 1);
1433 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1434 if (next_inuse <= end) {
1435 f2fs_err(sbi, "segno %u should be free but still inuse!",
1437 f2fs_bug_on(sbi, 1);
1442 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1444 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1447 int segment_count_main;
1448 long long block_count;
1449 int segs = secs * sbi->segs_per_sec;
1451 down_write(&sbi->sb_lock);
1453 section_count = le32_to_cpu(raw_sb->section_count);
1454 segment_count = le32_to_cpu(raw_sb->segment_count);
1455 segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1456 block_count = le64_to_cpu(raw_sb->block_count);
1458 raw_sb->section_count = cpu_to_le32(section_count + secs);
1459 raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1460 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1461 raw_sb->block_count = cpu_to_le64(block_count +
1462 (long long)segs * sbi->blocks_per_seg);
1463 if (f2fs_is_multi_device(sbi)) {
1464 int last_dev = sbi->s_ndevs - 1;
1466 le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1468 raw_sb->devs[last_dev].total_segments =
1469 cpu_to_le32(dev_segs + segs);
1472 up_write(&sbi->sb_lock);
1475 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1477 int segs = secs * sbi->segs_per_sec;
1478 long long blks = (long long)segs * sbi->blocks_per_seg;
1479 long long user_block_count =
1480 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1482 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1483 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1484 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1485 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1486 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1488 if (f2fs_is_multi_device(sbi)) {
1489 int last_dev = sbi->s_ndevs - 1;
1491 FDEV(last_dev).total_segments =
1492 (int)FDEV(last_dev).total_segments + segs;
1493 FDEV(last_dev).end_blk =
1494 (long long)FDEV(last_dev).end_blk + blks;
1495 #ifdef CONFIG_BLK_DEV_ZONED
1496 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1497 (int)(blks >> sbi->log_blocks_per_blkz);
1502 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1504 __u64 old_block_count, shrunk_blocks;
1506 int gc_mode, gc_type;
1510 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1511 if (block_count > old_block_count)
1514 if (f2fs_is_multi_device(sbi)) {
1515 int last_dev = sbi->s_ndevs - 1;
1516 __u64 last_segs = FDEV(last_dev).total_segments;
1518 if (block_count + last_segs * sbi->blocks_per_seg <=
1523 /* new fs size should align to section size */
1524 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
1528 if (block_count == old_block_count)
1531 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1532 f2fs_err(sbi, "Should run fsck to repair first.");
1533 return -EFSCORRUPTED;
1536 if (test_opt(sbi, DISABLE_CHECKPOINT)) {
1537 f2fs_err(sbi, "Checkpoint should be enabled.");
1541 freeze_bdev(sbi->sb->s_bdev);
1543 shrunk_blocks = old_block_count - block_count;
1544 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
1545 spin_lock(&sbi->stat_lock);
1546 if (shrunk_blocks + valid_user_blocks(sbi) +
1547 sbi->current_reserved_blocks + sbi->unusable_block_count +
1548 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
1551 sbi->user_block_count -= shrunk_blocks;
1552 spin_unlock(&sbi->stat_lock);
1554 thaw_bdev(sbi->sb->s_bdev, sbi->sb);
1558 mutex_lock(&sbi->resize_mutex);
1559 set_sbi_flag(sbi, SBI_IS_RESIZEFS);
1561 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1563 MAIN_SECS(sbi) -= secs;
1565 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1566 if (SIT_I(sbi)->last_victim[gc_mode] >=
1567 MAIN_SECS(sbi) * sbi->segs_per_sec)
1568 SIT_I(sbi)->last_victim[gc_mode] = 0;
1570 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1571 if (sbi->next_victim_seg[gc_type] >=
1572 MAIN_SECS(sbi) * sbi->segs_per_sec)
1573 sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1575 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1577 err = free_segment_range(sbi, MAIN_SECS(sbi) * sbi->segs_per_sec,
1578 MAIN_SEGS(sbi) - 1);
1582 update_sb_metadata(sbi, -secs);
1584 err = f2fs_commit_super(sbi, false);
1586 update_sb_metadata(sbi, secs);
1590 mutex_lock(&sbi->cp_mutex);
1591 update_fs_metadata(sbi, -secs);
1592 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
1593 set_sbi_flag(sbi, SBI_IS_DIRTY);
1594 mutex_unlock(&sbi->cp_mutex);
1596 err = f2fs_sync_fs(sbi->sb, 1);
1598 mutex_lock(&sbi->cp_mutex);
1599 update_fs_metadata(sbi, secs);
1600 mutex_unlock(&sbi->cp_mutex);
1601 update_sb_metadata(sbi, secs);
1602 f2fs_commit_super(sbi, false);
1606 set_sbi_flag(sbi, SBI_NEED_FSCK);
1607 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
1609 MAIN_SECS(sbi) += secs;
1610 spin_lock(&sbi->stat_lock);
1611 sbi->user_block_count += shrunk_blocks;
1612 spin_unlock(&sbi->stat_lock);
1614 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
1615 mutex_unlock(&sbi->resize_mutex);
1616 thaw_bdev(sbi->sb->s_bdev, sbi->sb);