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>
16 #include <linux/sched/signal.h>
22 #include <trace/events/f2fs.h>
24 static int gc_thread_func(void *data)
26 struct f2fs_sb_info *sbi = data;
27 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
28 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
31 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 */
46 if (try_to_freeze()) {
47 stat_other_skip_bggc_count(sbi);
50 if (kthread_should_stop())
53 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
54 increase_sleep_time(gc_th, &wait_ms);
55 stat_other_skip_bggc_count(sbi);
59 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
60 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
61 f2fs_stop_checkpoint(sbi, false);
64 if (!sb_start_write_trylock(sbi->sb)) {
65 stat_other_skip_bggc_count(sbi);
70 * [GC triggering condition]
71 * 0. GC is not conducted currently.
72 * 1. There are enough dirty segments.
73 * 2. IO subsystem is idle by checking the # of writeback pages.
74 * 3. IO subsystem is idle by checking the # of requests in
75 * bdev's request list.
77 * Note) We have to avoid triggering GCs frequently.
78 * Because it is possible that some segments can be
79 * invalidated soon after by user update or deletion.
80 * So, I'd like to wait some time to collect dirty segments.
82 if (sbi->gc_mode == GC_URGENT) {
83 wait_ms = gc_th->urgent_sleep_time;
84 down_write(&sbi->gc_lock);
88 if (!down_write_trylock(&sbi->gc_lock)) {
89 stat_other_skip_bggc_count(sbi);
93 if (!is_idle(sbi, GC_TIME)) {
94 increase_sleep_time(gc_th, &wait_ms);
95 up_write(&sbi->gc_lock);
96 stat_io_skip_bggc_count(sbi);
100 if (has_enough_invalid_blocks(sbi))
101 decrease_sleep_time(gc_th, &wait_ms);
103 increase_sleep_time(gc_th, &wait_ms);
105 stat_inc_bggc_count(sbi->stat_info);
107 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
109 /* if return value is not zero, no victim was selected */
110 if (f2fs_gc(sbi, sync_mode, true, NULL_SEGNO))
111 wait_ms = gc_th->no_gc_sleep_time;
113 trace_f2fs_background_gc(sbi->sb, wait_ms,
114 prefree_segments(sbi), free_segments(sbi));
116 /* balancing f2fs's metadata periodically */
117 f2fs_balance_fs_bg(sbi, true);
119 sb_end_write(sbi->sb);
121 } while (!kthread_should_stop());
125 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
127 struct f2fs_gc_kthread *gc_th;
128 dev_t dev = sbi->sb->s_bdev->bd_dev;
131 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
137 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
138 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
139 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
140 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
144 sbi->gc_thread = gc_th;
145 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
146 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
147 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
148 if (IS_ERR(gc_th->f2fs_gc_task)) {
149 err = PTR_ERR(gc_th->f2fs_gc_task);
151 sbi->gc_thread = NULL;
157 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
159 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
162 kthread_stop(gc_th->f2fs_gc_task);
164 sbi->gc_thread = NULL;
167 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
169 int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
171 switch (sbi->gc_mode) {
183 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
184 int type, struct victim_sel_policy *p)
186 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
188 if (p->alloc_mode == SSR) {
189 p->gc_mode = GC_GREEDY;
190 p->dirty_segmap = dirty_i->dirty_segmap[type];
191 p->max_search = dirty_i->nr_dirty[type];
194 p->gc_mode = select_gc_type(sbi, gc_type);
195 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
196 p->max_search = dirty_i->nr_dirty[DIRTY];
197 p->ofs_unit = sbi->segs_per_sec;
201 * adjust candidates range, should select all dirty segments for
202 * foreground GC and urgent GC cases.
204 if (gc_type != FG_GC &&
205 (sbi->gc_mode != GC_URGENT) &&
206 p->max_search > sbi->max_victim_search)
207 p->max_search = sbi->max_victim_search;
209 /* let's select beginning hot/small space first in no_heap mode*/
210 if (test_opt(sbi, NOHEAP) &&
211 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
214 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
217 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
218 struct victim_sel_policy *p)
220 /* SSR allocates in a segment unit */
221 if (p->alloc_mode == SSR)
222 return sbi->blocks_per_seg;
223 if (p->gc_mode == GC_GREEDY)
224 return 2 * sbi->blocks_per_seg * p->ofs_unit;
225 else if (p->gc_mode == GC_CB)
227 else /* No other gc_mode */
231 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
233 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
237 * If the gc_type is FG_GC, we can select victim segments
238 * selected by background GC before.
239 * Those segments guarantee they have small valid blocks.
241 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
242 if (sec_usage_check(sbi, secno))
244 clear_bit(secno, dirty_i->victim_secmap);
245 return GET_SEG_FROM_SEC(sbi, secno);
250 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
252 struct sit_info *sit_i = SIT_I(sbi);
253 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
254 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
255 unsigned long long mtime = 0;
256 unsigned int vblocks;
257 unsigned char age = 0;
261 for (i = 0; i < sbi->segs_per_sec; i++)
262 mtime += get_seg_entry(sbi, start + i)->mtime;
263 vblocks = get_valid_blocks(sbi, segno, true);
265 mtime = div_u64(mtime, sbi->segs_per_sec);
266 vblocks = div_u64(vblocks, sbi->segs_per_sec);
268 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
270 /* Handle if the system time has changed by the user */
271 if (mtime < sit_i->min_mtime)
272 sit_i->min_mtime = mtime;
273 if (mtime > sit_i->max_mtime)
274 sit_i->max_mtime = mtime;
275 if (sit_i->max_mtime != sit_i->min_mtime)
276 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
277 sit_i->max_mtime - sit_i->min_mtime);
279 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
282 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
283 unsigned int segno, struct victim_sel_policy *p)
285 if (p->alloc_mode == SSR)
286 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
288 /* alloc_mode == LFS */
289 if (p->gc_mode == GC_GREEDY)
290 return get_valid_blocks(sbi, segno, true);
292 return get_cb_cost(sbi, segno);
295 static unsigned int count_bits(const unsigned long *addr,
296 unsigned int offset, unsigned int len)
298 unsigned int end = offset + len, sum = 0;
300 while (offset < end) {
301 if (test_bit(offset++, addr))
308 * This function is called from two paths.
309 * One is garbage collection and the other is SSR segment selection.
310 * When it is called during GC, it just gets a victim segment
311 * and it does not remove it from dirty seglist.
312 * When it is called from SSR segment selection, it finds a segment
313 * which has minimum valid blocks and removes it from dirty seglist.
315 static int get_victim_by_default(struct f2fs_sb_info *sbi,
316 unsigned int *result, int gc_type, int type, char alloc_mode)
318 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
319 struct sit_info *sm = SIT_I(sbi);
320 struct victim_sel_policy p;
321 unsigned int secno, last_victim;
322 unsigned int last_segment;
323 unsigned int nsearched = 0;
325 mutex_lock(&dirty_i->seglist_lock);
326 last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
328 p.alloc_mode = alloc_mode;
329 select_policy(sbi, gc_type, type, &p);
331 p.min_segno = NULL_SEGNO;
332 p.min_cost = get_max_cost(sbi, &p);
334 if (*result != NULL_SEGNO) {
335 if (get_valid_blocks(sbi, *result, false) &&
336 !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
337 p.min_segno = *result;
341 if (p.max_search == 0)
344 if (__is_large_section(sbi) && p.alloc_mode == LFS) {
345 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
346 p.min_segno = sbi->next_victim_seg[BG_GC];
347 *result = p.min_segno;
348 sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
351 if (gc_type == FG_GC &&
352 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
353 p.min_segno = sbi->next_victim_seg[FG_GC];
354 *result = p.min_segno;
355 sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
360 last_victim = sm->last_victim[p.gc_mode];
361 if (p.alloc_mode == LFS && gc_type == FG_GC) {
362 p.min_segno = check_bg_victims(sbi);
363 if (p.min_segno != NULL_SEGNO)
371 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
372 if (segno >= last_segment) {
373 if (sm->last_victim[p.gc_mode]) {
375 sm->last_victim[p.gc_mode];
376 sm->last_victim[p.gc_mode] = 0;
383 p.offset = segno + p.ofs_unit;
384 if (p.ofs_unit > 1) {
385 p.offset -= segno % p.ofs_unit;
386 nsearched += count_bits(p.dirty_segmap,
387 p.offset - p.ofs_unit,
393 #ifdef CONFIG_F2FS_CHECK_FS
395 * skip selecting the invalid segno (that is failed due to block
396 * validity check failure during GC) to avoid endless GC loop in
399 if (test_bit(segno, sm->invalid_segmap))
403 secno = GET_SEC_FROM_SEG(sbi, segno);
405 if (sec_usage_check(sbi, secno))
407 /* Don't touch checkpointed data */
408 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
409 get_ckpt_valid_blocks(sbi, segno) &&
410 p.alloc_mode != SSR))
412 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
415 cost = get_gc_cost(sbi, segno, &p);
417 if (p.min_cost > cost) {
422 if (nsearched >= p.max_search) {
423 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
424 sm->last_victim[p.gc_mode] = last_victim + 1;
426 sm->last_victim[p.gc_mode] = segno + 1;
427 sm->last_victim[p.gc_mode] %=
428 (MAIN_SECS(sbi) * sbi->segs_per_sec);
432 if (p.min_segno != NULL_SEGNO) {
434 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
436 if (p.alloc_mode == LFS) {
437 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
438 if (gc_type == FG_GC)
439 sbi->cur_victim_sec = secno;
441 set_bit(secno, dirty_i->victim_secmap);
446 if (p.min_segno != NULL_SEGNO)
447 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
449 prefree_segments(sbi), free_segments(sbi));
450 mutex_unlock(&dirty_i->seglist_lock);
452 return (p.min_segno == NULL_SEGNO) ? 0 : 1;
455 static const struct victim_selection default_v_ops = {
456 .get_victim = get_victim_by_default,
459 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
461 struct inode_entry *ie;
463 ie = radix_tree_lookup(&gc_list->iroot, ino);
469 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
471 struct inode_entry *new_ie;
473 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
477 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
478 new_ie->inode = inode;
480 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
481 list_add_tail(&new_ie->list, &gc_list->ilist);
484 static void put_gc_inode(struct gc_inode_list *gc_list)
486 struct inode_entry *ie, *next_ie;
487 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
488 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
491 kmem_cache_free(f2fs_inode_entry_slab, ie);
495 static int check_valid_map(struct f2fs_sb_info *sbi,
496 unsigned int segno, int offset)
498 struct sit_info *sit_i = SIT_I(sbi);
499 struct seg_entry *sentry;
502 down_read(&sit_i->sentry_lock);
503 sentry = get_seg_entry(sbi, segno);
504 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
505 up_read(&sit_i->sentry_lock);
510 * This function compares node address got in summary with that in NAT.
511 * On validity, copy that node with cold status, otherwise (invalid node)
514 static int gc_node_segment(struct f2fs_sb_info *sbi,
515 struct f2fs_summary *sum, unsigned int segno, int gc_type)
517 struct f2fs_summary *entry;
521 bool fggc = (gc_type == FG_GC);
524 start_addr = START_BLOCK(sbi, segno);
529 if (fggc && phase == 2)
530 atomic_inc(&sbi->wb_sync_req[NODE]);
532 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
533 nid_t nid = le32_to_cpu(entry->nid);
534 struct page *node_page;
538 /* stop BG_GC if there is not enough free sections. */
539 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
542 if (check_valid_map(sbi, segno, off) == 0)
546 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
552 f2fs_ra_node_page(sbi, nid);
557 node_page = f2fs_get_node_page(sbi, nid);
558 if (IS_ERR(node_page))
561 /* block may become invalid during f2fs_get_node_page */
562 if (check_valid_map(sbi, segno, off) == 0) {
563 f2fs_put_page(node_page, 1);
567 if (f2fs_get_node_info(sbi, nid, &ni)) {
568 f2fs_put_page(node_page, 1);
572 if (ni.blk_addr != start_addr + off) {
573 f2fs_put_page(node_page, 1);
577 err = f2fs_move_node_page(node_page, gc_type);
578 if (!err && gc_type == FG_GC)
580 stat_inc_node_blk_count(sbi, 1, gc_type);
587 atomic_dec(&sbi->wb_sync_req[NODE]);
592 * Calculate start block index indicating the given node offset.
593 * Be careful, caller should give this node offset only indicating direct node
594 * blocks. If any node offsets, which point the other types of node blocks such
595 * as indirect or double indirect node blocks, are given, it must be a caller's
598 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
600 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
608 } else if (node_ofs <= indirect_blks) {
609 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
610 bidx = node_ofs - 2 - dec;
612 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
613 bidx = node_ofs - 5 - dec;
615 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
618 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
619 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
621 struct page *node_page;
623 unsigned int ofs_in_node;
624 block_t source_blkaddr;
626 nid = le32_to_cpu(sum->nid);
627 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
629 node_page = f2fs_get_node_page(sbi, nid);
630 if (IS_ERR(node_page))
633 if (f2fs_get_node_info(sbi, nid, dni)) {
634 f2fs_put_page(node_page, 1);
638 if (sum->version != dni->version) {
639 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
641 set_sbi_flag(sbi, SBI_NEED_FSCK);
644 *nofs = ofs_of_node(node_page);
645 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
646 f2fs_put_page(node_page, 1);
648 if (source_blkaddr != blkaddr) {
649 #ifdef CONFIG_F2FS_CHECK_FS
650 unsigned int segno = GET_SEGNO(sbi, blkaddr);
651 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
653 if (unlikely(check_valid_map(sbi, segno, offset))) {
654 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
655 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n",
656 blkaddr, source_blkaddr, segno);
666 static int ra_data_block(struct inode *inode, pgoff_t index)
668 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
669 struct address_space *mapping = inode->i_mapping;
670 struct dnode_of_data dn;
672 struct extent_info ei = {0, 0, 0};
673 struct f2fs_io_info fio = {
680 .encrypted_page = NULL,
686 page = f2fs_grab_cache_page(mapping, index, true);
690 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
691 dn.data_blkaddr = ei.blk + index - ei.fofs;
692 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
693 DATA_GENERIC_ENHANCE_READ))) {
700 set_new_dnode(&dn, inode, NULL, NULL, 0);
701 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
706 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
710 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
711 DATA_GENERIC_ENHANCE))) {
718 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
721 * don't cache encrypted data into meta inode until previous dirty
722 * data were writebacked to avoid racing between GC and flush.
724 f2fs_wait_on_page_writeback(page, DATA, true, true);
726 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
728 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
730 FGP_LOCK | FGP_CREAT, GFP_NOFS);
731 if (!fio.encrypted_page) {
736 err = f2fs_submit_page_bio(&fio);
738 goto put_encrypted_page;
739 f2fs_put_page(fio.encrypted_page, 0);
740 f2fs_put_page(page, 1);
742 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
743 f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
747 f2fs_put_page(fio.encrypted_page, 1);
749 f2fs_put_page(page, 1);
754 * Move data block via META_MAPPING while keeping locked data page.
755 * This can be used to move blocks, aka LBAs, directly on disk.
757 static int move_data_block(struct inode *inode, block_t bidx,
758 int gc_type, unsigned int segno, int off)
760 struct f2fs_io_info fio = {
761 .sbi = F2FS_I_SB(inode),
767 .encrypted_page = NULL,
771 struct dnode_of_data dn;
772 struct f2fs_summary sum;
774 struct page *page, *mpage;
777 bool lfs_mode = f2fs_lfs_mode(fio.sbi);
779 /* do not read out */
780 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
784 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
789 if (f2fs_is_atomic_file(inode)) {
790 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
791 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
796 if (f2fs_is_pinned_file(inode)) {
797 f2fs_pin_file_control(inode, true);
802 set_new_dnode(&dn, inode, NULL, NULL, 0);
803 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
807 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
808 ClearPageUptodate(page);
814 * don't cache encrypted data into meta inode until previous dirty
815 * data were writebacked to avoid racing between GC and flush.
817 f2fs_wait_on_page_writeback(page, DATA, true, true);
819 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
821 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
825 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
829 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
832 down_write(&fio.sbi->io_order_lock);
834 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
835 fio.old_blkaddr, false);
839 fio.encrypted_page = mpage;
841 /* read source block in mpage */
842 if (!PageUptodate(mpage)) {
843 err = f2fs_submit_page_bio(&fio);
845 f2fs_put_page(mpage, 1);
849 f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
850 f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
853 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
854 !PageUptodate(mpage))) {
856 f2fs_put_page(mpage, 1);
861 f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
862 &sum, CURSEG_COLD_DATA, NULL, false);
864 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
865 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
866 if (!fio.encrypted_page) {
868 f2fs_put_page(mpage, 1);
872 /* write target block */
873 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
874 memcpy(page_address(fio.encrypted_page),
875 page_address(mpage), PAGE_SIZE);
876 f2fs_put_page(mpage, 1);
877 invalidate_mapping_pages(META_MAPPING(fio.sbi),
878 fio.old_blkaddr, fio.old_blkaddr);
880 set_page_dirty(fio.encrypted_page);
881 if (clear_page_dirty_for_io(fio.encrypted_page))
882 dec_page_count(fio.sbi, F2FS_DIRTY_META);
884 set_page_writeback(fio.encrypted_page);
885 ClearPageError(page);
887 /* allocate block address */
888 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
890 fio.op = REQ_OP_WRITE;
891 fio.op_flags = REQ_SYNC;
892 fio.new_blkaddr = newaddr;
893 f2fs_submit_page_write(&fio);
896 if (PageWriteback(fio.encrypted_page))
897 end_page_writeback(fio.encrypted_page);
901 f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
903 f2fs_update_data_blkaddr(&dn, newaddr);
904 set_inode_flag(inode, FI_APPEND_WRITE);
905 if (page->index == 0)
906 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
908 f2fs_put_page(fio.encrypted_page, 1);
911 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
915 up_write(&fio.sbi->io_order_lock);
919 f2fs_put_page(page, 1);
923 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
924 unsigned int segno, int off)
929 page = f2fs_get_lock_data_page(inode, bidx, true);
931 return PTR_ERR(page);
933 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
938 if (f2fs_is_atomic_file(inode)) {
939 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
940 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
944 if (f2fs_is_pinned_file(inode)) {
945 if (gc_type == FG_GC)
946 f2fs_pin_file_control(inode, true);
951 if (gc_type == BG_GC) {
952 if (PageWriteback(page)) {
956 set_page_dirty(page);
959 struct f2fs_io_info fio = {
960 .sbi = F2FS_I_SB(inode),
965 .op_flags = REQ_SYNC,
966 .old_blkaddr = NULL_ADDR,
968 .encrypted_page = NULL,
969 .need_lock = LOCK_REQ,
970 .io_type = FS_GC_DATA_IO,
972 bool is_dirty = PageDirty(page);
975 f2fs_wait_on_page_writeback(page, DATA, true, true);
977 set_page_dirty(page);
978 if (clear_page_dirty_for_io(page)) {
979 inode_dec_dirty_pages(inode);
980 f2fs_remove_dirty_inode(inode);
985 err = f2fs_do_write_data_page(&fio);
987 clear_cold_data(page);
988 if (err == -ENOMEM) {
989 congestion_wait(BLK_RW_ASYNC,
994 set_page_dirty(page);
998 f2fs_put_page(page, 1);
1003 * This function tries to get parent node of victim data block, and identifies
1004 * data block validity. If the block is valid, copy that with cold status and
1005 * modify parent node.
1006 * If the parent node is not valid or the data block address is different,
1007 * the victim data block is ignored.
1009 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1010 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
1012 struct super_block *sb = sbi->sb;
1013 struct f2fs_summary *entry;
1019 start_addr = START_BLOCK(sbi, segno);
1024 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
1025 struct page *data_page;
1026 struct inode *inode;
1027 struct node_info dni; /* dnode info for the data */
1028 unsigned int ofs_in_node, nofs;
1030 nid_t nid = le32_to_cpu(entry->nid);
1033 * stop BG_GC if there is not enough free sections.
1034 * Or, stop GC if the segment becomes fully valid caused by
1035 * race condition along with SSR block allocation.
1037 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1038 get_valid_blocks(sbi, segno, true) ==
1042 if (check_valid_map(sbi, segno, off) == 0)
1046 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1052 f2fs_ra_node_page(sbi, nid);
1056 /* Get an inode by ino with checking validity */
1057 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1061 f2fs_ra_node_page(sbi, dni.ino);
1065 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1068 inode = f2fs_iget(sb, dni.ino);
1069 if (IS_ERR(inode) || is_bad_inode(inode)) {
1070 set_sbi_flag(sbi, SBI_NEED_FSCK);
1074 if (!down_write_trylock(
1075 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1077 sbi->skipped_gc_rwsem++;
1081 start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1084 if (f2fs_post_read_required(inode)) {
1085 int err = ra_data_block(inode, start_bidx);
1087 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1092 add_gc_inode(gc_list, inode);
1096 data_page = f2fs_get_read_data_page(inode,
1097 start_bidx, REQ_RAHEAD, true);
1098 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1099 if (IS_ERR(data_page)) {
1104 f2fs_put_page(data_page, 0);
1105 add_gc_inode(gc_list, inode);
1110 inode = find_gc_inode(gc_list, dni.ino);
1112 struct f2fs_inode_info *fi = F2FS_I(inode);
1113 bool locked = false;
1116 if (S_ISREG(inode->i_mode)) {
1117 if (!down_write_trylock(&fi->i_gc_rwsem[READ]))
1119 if (!down_write_trylock(
1120 &fi->i_gc_rwsem[WRITE])) {
1121 sbi->skipped_gc_rwsem++;
1122 up_write(&fi->i_gc_rwsem[READ]);
1127 /* wait for all inflight aio data */
1128 inode_dio_wait(inode);
1131 start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1133 if (f2fs_post_read_required(inode))
1134 err = move_data_block(inode, start_bidx,
1135 gc_type, segno, off);
1137 err = move_data_page(inode, start_bidx, gc_type,
1140 if (!err && (gc_type == FG_GC ||
1141 f2fs_post_read_required(inode)))
1145 up_write(&fi->i_gc_rwsem[WRITE]);
1146 up_write(&fi->i_gc_rwsem[READ]);
1149 stat_inc_data_blk_count(sbi, 1, gc_type);
1159 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1162 struct sit_info *sit_i = SIT_I(sbi);
1165 down_write(&sit_i->sentry_lock);
1166 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1167 NO_CHECK_TYPE, LFS);
1168 up_write(&sit_i->sentry_lock);
1172 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1173 unsigned int start_segno,
1174 struct gc_inode_list *gc_list, int gc_type)
1176 struct page *sum_page;
1177 struct f2fs_summary_block *sum;
1178 struct blk_plug plug;
1179 unsigned int segno = start_segno;
1180 unsigned int end_segno = start_segno + sbi->segs_per_sec;
1181 int seg_freed = 0, migrated = 0;
1182 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1183 SUM_TYPE_DATA : SUM_TYPE_NODE;
1186 if (__is_large_section(sbi))
1187 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1189 /* readahead multi ssa blocks those have contiguous address */
1190 if (__is_large_section(sbi))
1191 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1192 end_segno - segno, META_SSA, true);
1194 /* reference all summary page */
1195 while (segno < end_segno) {
1196 sum_page = f2fs_get_sum_page(sbi, segno++);
1197 if (IS_ERR(sum_page)) {
1198 int err = PTR_ERR(sum_page);
1200 end_segno = segno - 1;
1201 for (segno = start_segno; segno < end_segno; segno++) {
1202 sum_page = find_get_page(META_MAPPING(sbi),
1203 GET_SUM_BLOCK(sbi, segno));
1204 f2fs_put_page(sum_page, 0);
1205 f2fs_put_page(sum_page, 0);
1209 unlock_page(sum_page);
1212 blk_start_plug(&plug);
1214 for (segno = start_segno; segno < end_segno; segno++) {
1216 /* find segment summary of victim */
1217 sum_page = find_get_page(META_MAPPING(sbi),
1218 GET_SUM_BLOCK(sbi, segno));
1219 f2fs_put_page(sum_page, 0);
1221 if (get_valid_blocks(sbi, segno, false) == 0)
1223 if (gc_type == BG_GC && __is_large_section(sbi) &&
1224 migrated >= sbi->migration_granularity)
1226 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1229 sum = page_address(sum_page);
1230 if (type != GET_SUM_TYPE((&sum->footer))) {
1231 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1232 segno, type, GET_SUM_TYPE((&sum->footer)));
1233 set_sbi_flag(sbi, SBI_NEED_FSCK);
1234 f2fs_stop_checkpoint(sbi, false);
1239 * this is to avoid deadlock:
1240 * - lock_page(sum_page) - f2fs_replace_block
1241 * - check_valid_map() - down_write(sentry_lock)
1242 * - down_read(sentry_lock) - change_curseg()
1243 * - lock_page(sum_page)
1245 if (type == SUM_TYPE_NODE)
1246 submitted += gc_node_segment(sbi, sum->entries, segno,
1249 submitted += gc_data_segment(sbi, sum->entries, gc_list,
1252 stat_inc_seg_count(sbi, type, gc_type);
1256 if (gc_type == FG_GC &&
1257 get_valid_blocks(sbi, segno, false) == 0)
1260 if (__is_large_section(sbi) && segno + 1 < end_segno)
1261 sbi->next_victim_seg[gc_type] = segno + 1;
1263 f2fs_put_page(sum_page, 0);
1267 f2fs_submit_merged_write(sbi,
1268 (type == SUM_TYPE_NODE) ? NODE : DATA);
1270 blk_finish_plug(&plug);
1272 stat_inc_call_count(sbi->stat_info);
1277 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1278 bool background, unsigned int segno)
1280 int gc_type = sync ? FG_GC : BG_GC;
1281 int sec_freed = 0, seg_freed = 0, total_freed = 0;
1283 struct cp_control cpc;
1284 unsigned int init_segno = segno;
1285 struct gc_inode_list gc_list = {
1286 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1287 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1289 unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1290 unsigned long long first_skipped;
1291 unsigned int skipped_round = 0, round = 0;
1293 trace_f2fs_gc_begin(sbi->sb, sync, background,
1294 get_pages(sbi, F2FS_DIRTY_NODES),
1295 get_pages(sbi, F2FS_DIRTY_DENTS),
1296 get_pages(sbi, F2FS_DIRTY_IMETA),
1299 reserved_segments(sbi),
1300 prefree_segments(sbi));
1302 cpc.reason = __get_cp_reason(sbi);
1303 sbi->skipped_gc_rwsem = 0;
1304 first_skipped = last_skipped;
1306 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1310 if (unlikely(f2fs_cp_error(sbi))) {
1315 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1317 * For example, if there are many prefree_segments below given
1318 * threshold, we can make them free by checkpoint. Then, we
1319 * secure free segments which doesn't need fggc any more.
1321 if (prefree_segments(sbi) &&
1322 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1323 ret = f2fs_write_checkpoint(sbi, &cpc);
1327 if (has_not_enough_free_secs(sbi, 0, 0))
1331 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1332 if (gc_type == BG_GC && !background) {
1336 if (!__get_victim(sbi, &segno, gc_type)) {
1341 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1342 if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1344 total_freed += seg_freed;
1346 if (gc_type == FG_GC) {
1347 if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1348 sbi->skipped_gc_rwsem)
1350 last_skipped = sbi->skipped_atomic_files[FG_GC];
1354 if (gc_type == FG_GC && seg_freed)
1355 sbi->cur_victim_sec = NULL_SEGNO;
1360 if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1361 if (skipped_round <= MAX_SKIP_GC_COUNT ||
1362 skipped_round * 2 < round) {
1367 if (first_skipped < last_skipped &&
1368 (last_skipped - first_skipped) >
1369 sbi->skipped_gc_rwsem) {
1370 f2fs_drop_inmem_pages_all(sbi, true);
1374 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1375 ret = f2fs_write_checkpoint(sbi, &cpc);
1378 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1379 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1381 trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1382 get_pages(sbi, F2FS_DIRTY_NODES),
1383 get_pages(sbi, F2FS_DIRTY_DENTS),
1384 get_pages(sbi, F2FS_DIRTY_IMETA),
1387 reserved_segments(sbi),
1388 prefree_segments(sbi));
1390 up_write(&sbi->gc_lock);
1392 put_gc_inode(&gc_list);
1395 ret = sec_freed ? 0 : -EAGAIN;
1399 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1401 DIRTY_I(sbi)->v_ops = &default_v_ops;
1403 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1405 /* give warm/cold data area from slower device */
1406 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1407 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1408 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1411 static int free_segment_range(struct f2fs_sb_info *sbi,
1412 unsigned int secs, bool gc_only)
1414 unsigned int segno, next_inuse, start, end;
1415 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1416 int gc_mode, gc_type;
1420 /* Force block allocation for GC */
1421 MAIN_SECS(sbi) -= secs;
1422 start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1423 end = MAIN_SEGS(sbi) - 1;
1425 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1426 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1427 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1428 SIT_I(sbi)->last_victim[gc_mode] = 0;
1430 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1431 if (sbi->next_victim_seg[gc_type] >= start)
1432 sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1433 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1435 /* Move out cursegs from the target range */
1436 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_TYPE; type++)
1437 allocate_segment_for_resize(sbi, type, start, end);
1439 /* do GC to move out valid blocks in the range */
1440 for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1441 struct gc_inode_list gc_list = {
1442 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1443 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1446 do_garbage_collect(sbi, segno, &gc_list, FG_GC);
1447 put_gc_inode(&gc_list);
1449 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1453 if (fatal_signal_pending(current)) {
1461 err = f2fs_write_checkpoint(sbi, &cpc);
1465 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1466 if (next_inuse <= end) {
1467 f2fs_err(sbi, "segno %u should be free but still inuse!",
1469 f2fs_bug_on(sbi, 1);
1472 MAIN_SECS(sbi) += secs;
1476 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1478 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1481 int segment_count_main;
1482 long long block_count;
1483 int segs = secs * sbi->segs_per_sec;
1485 down_write(&sbi->sb_lock);
1487 section_count = le32_to_cpu(raw_sb->section_count);
1488 segment_count = le32_to_cpu(raw_sb->segment_count);
1489 segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1490 block_count = le64_to_cpu(raw_sb->block_count);
1492 raw_sb->section_count = cpu_to_le32(section_count + secs);
1493 raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1494 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1495 raw_sb->block_count = cpu_to_le64(block_count +
1496 (long long)segs * sbi->blocks_per_seg);
1497 if (f2fs_is_multi_device(sbi)) {
1498 int last_dev = sbi->s_ndevs - 1;
1500 le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1502 raw_sb->devs[last_dev].total_segments =
1503 cpu_to_le32(dev_segs + segs);
1506 up_write(&sbi->sb_lock);
1509 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1511 int segs = secs * sbi->segs_per_sec;
1512 long long blks = (long long)segs * sbi->blocks_per_seg;
1513 long long user_block_count =
1514 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1516 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1517 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1518 MAIN_SECS(sbi) += secs;
1519 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1520 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1521 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1523 if (f2fs_is_multi_device(sbi)) {
1524 int last_dev = sbi->s_ndevs - 1;
1526 FDEV(last_dev).total_segments =
1527 (int)FDEV(last_dev).total_segments + segs;
1528 FDEV(last_dev).end_blk =
1529 (long long)FDEV(last_dev).end_blk + blks;
1530 #ifdef CONFIG_BLK_DEV_ZONED
1531 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1532 (int)(blks >> sbi->log_blocks_per_blkz);
1537 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1539 __u64 old_block_count, shrunk_blocks;
1540 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1545 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1546 if (block_count > old_block_count)
1549 if (f2fs_is_multi_device(sbi)) {
1550 int last_dev = sbi->s_ndevs - 1;
1551 __u64 last_segs = FDEV(last_dev).total_segments;
1553 if (block_count + last_segs * sbi->blocks_per_seg <=
1558 /* new fs size should align to section size */
1559 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
1563 if (block_count == old_block_count)
1566 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1567 f2fs_err(sbi, "Should run fsck to repair first.");
1568 return -EFSCORRUPTED;
1571 if (test_opt(sbi, DISABLE_CHECKPOINT)) {
1572 f2fs_err(sbi, "Checkpoint should be enabled.");
1576 shrunk_blocks = old_block_count - block_count;
1577 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
1580 if (!down_write_trylock(&sbi->gc_lock))
1583 /* stop CP to protect MAIN_SEC in free_segment_range */
1585 err = free_segment_range(sbi, secs, true);
1586 f2fs_unlock_op(sbi);
1587 up_write(&sbi->gc_lock);
1591 set_sbi_flag(sbi, SBI_IS_RESIZEFS);
1593 freeze_super(sbi->sb);
1594 down_write(&sbi->gc_lock);
1595 mutex_lock(&sbi->cp_mutex);
1597 spin_lock(&sbi->stat_lock);
1598 if (shrunk_blocks + valid_user_blocks(sbi) +
1599 sbi->current_reserved_blocks + sbi->unusable_block_count +
1600 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
1603 sbi->user_block_count -= shrunk_blocks;
1604 spin_unlock(&sbi->stat_lock);
1608 err = free_segment_range(sbi, secs, false);
1612 update_sb_metadata(sbi, -secs);
1614 err = f2fs_commit_super(sbi, false);
1616 update_sb_metadata(sbi, secs);
1620 update_fs_metadata(sbi, -secs);
1621 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
1622 set_sbi_flag(sbi, SBI_IS_DIRTY);
1624 err = f2fs_write_checkpoint(sbi, &cpc);
1626 update_fs_metadata(sbi, secs);
1627 update_sb_metadata(sbi, secs);
1628 f2fs_commit_super(sbi, false);
1632 set_sbi_flag(sbi, SBI_NEED_FSCK);
1633 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
1635 spin_lock(&sbi->stat_lock);
1636 sbi->user_block_count += shrunk_blocks;
1637 spin_unlock(&sbi->stat_lock);
1640 mutex_unlock(&sbi->cp_mutex);
1641 up_write(&sbi->gc_lock);
1642 thaw_super(sbi->sb);
1643 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);