~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/f2fs/gc.c

Version: ~ [ linux-5.12-rc7 ] ~ [ linux-5.11.13 ] ~ [ linux-5.10.29 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.111 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.186 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.230 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.266 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.266 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  * fs/f2fs/gc.c
  4  *
  5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
  6  *             http://www.samsung.com/
  7  */
  8 #include <linux/fs.h>
  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 
 17 #include "f2fs.h"
 18 #include "node.h"
 19 #include "segment.h"
 20 #include "gc.h"
 21 #include <trace/events/f2fs.h>
 22 
 23 static int gc_thread_func(void *data)
 24 {
 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;
 28         unsigned int wait_ms;
 29 
 30         wait_ms = gc_th->min_sleep_time;
 31 
 32         set_freezable();
 33         do {
 34                 wait_event_interruptible_timeout(*wq,
 35                                 kthread_should_stop() || freezing(current) ||
 36                                 gc_th->gc_wake,
 37                                 msecs_to_jiffies(wait_ms));
 38 
 39                 /* give it a try one time */
 40                 if (gc_th->gc_wake)
 41                         gc_th->gc_wake = 0;
 42 
 43                 if (try_to_freeze()) {
 44                         stat_other_skip_bggc_count(sbi);
 45                         continue;
 46                 }
 47                 if (kthread_should_stop())
 48                         break;
 49 
 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);
 53                         continue;
 54                 }
 55 
 56                 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 57                         f2fs_show_injection_info(FAULT_CHECKPOINT);
 58                         f2fs_stop_checkpoint(sbi, false);
 59                 }
 60 
 61                 if (!sb_start_write_trylock(sbi->sb)) {
 62                         stat_other_skip_bggc_count(sbi);
 63                         continue;
 64                 }
 65 
 66                 /*
 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.
 73                  *
 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.
 78                  */
 79                 if (sbi->gc_mode == GC_URGENT) {
 80                         wait_ms = gc_th->urgent_sleep_time;
 81                         mutex_lock(&sbi->gc_mutex);
 82                         goto do_gc;
 83                 }
 84 
 85                 if (!mutex_trylock(&sbi->gc_mutex)) {
 86                         stat_other_skip_bggc_count(sbi);
 87                         goto next;
 88                 }
 89 
 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);
 94                         goto next;
 95                 }
 96 
 97                 if (has_enough_invalid_blocks(sbi))
 98                         decrease_sleep_time(gc_th, &wait_ms);
 99                 else
100                         increase_sleep_time(gc_th, &wait_ms);
101 do_gc:
102                 stat_inc_bggc_count(sbi);
103 
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;
107 
108                 trace_f2fs_background_gc(sbi->sb, wait_ms,
109                                 prefree_segments(sbi), free_segments(sbi));
110 
111                 /* balancing f2fs's metadata periodically */
112                 f2fs_balance_fs_bg(sbi);
113 next:
114                 sb_end_write(sbi->sb);
115 
116         } while (!kthread_should_stop());
117         return 0;
118 }
119 
120 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
121 {
122         struct f2fs_gc_kthread *gc_th;
123         dev_t dev = sbi->sb->s_bdev->bd_dev;
124         int err = 0;
125 
126         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
127         if (!gc_th) {
128                 err = -ENOMEM;
129                 goto out;
130         }
131 
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;
136 
137         gc_th->gc_wake= 0;
138 
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);
145                 kvfree(gc_th);
146                 sbi->gc_thread = NULL;
147         }
148 out:
149         return err;
150 }
151 
152 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
153 {
154         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
155         if (!gc_th)
156                 return;
157         kthread_stop(gc_th->f2fs_gc_task);
158         kvfree(gc_th);
159         sbi->gc_thread = NULL;
160 }
161 
162 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
163 {
164         int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
165 
166         switch (sbi->gc_mode) {
167         case GC_IDLE_CB:
168                 gc_mode = GC_CB;
169                 break;
170         case GC_IDLE_GREEDY:
171         case GC_URGENT:
172                 gc_mode = GC_GREEDY;
173                 break;
174         }
175         return gc_mode;
176 }
177 
178 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
179                         int type, struct victim_sel_policy *p)
180 {
181         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
182 
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];
187                 p->ofs_unit = 1;
188         } else {
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;
193         }
194 
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;
200 
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)))
204                 p->offset = 0;
205         else
206                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
207 }
208 
209 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
210                                 struct victim_sel_policy *p)
211 {
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)
218                 return UINT_MAX;
219         else /* No other gc_mode */
220                 return 0;
221 }
222 
223 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
224 {
225         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
226         unsigned int secno;
227 
228         /*
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.
232          */
233         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
234                 if (sec_usage_check(sbi, secno))
235                         continue;
236                 clear_bit(secno, dirty_i->victim_secmap);
237                 return GET_SEG_FROM_SEC(sbi, secno);
238         }
239         return NULL_SEGNO;
240 }
241 
242 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
243 {
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;
250         unsigned char u;
251         unsigned int i;
252 
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);
256 
257         mtime = div_u64(mtime, sbi->segs_per_sec);
258         vblocks = div_u64(vblocks, sbi->segs_per_sec);
259 
260         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
261 
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);
270 
271         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
272 }
273 
274 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
275                         unsigned int segno, struct victim_sel_policy *p)
276 {
277         if (p->alloc_mode == SSR)
278                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
279 
280         /* alloc_mode == LFS */
281         if (p->gc_mode == GC_GREEDY)
282                 return get_valid_blocks(sbi, segno, true);
283         else
284                 return get_cb_cost(sbi, segno);
285 }
286 
287 static unsigned int count_bits(const unsigned long *addr,
288                                 unsigned int offset, unsigned int len)
289 {
290         unsigned int end = offset + len, sum = 0;
291 
292         while (offset < end) {
293                 if (test_bit(offset++, addr))
294                         ++sum;
295         }
296         return sum;
297 }
298 
299 /*
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.
306  */
307 static int get_victim_by_default(struct f2fs_sb_info *sbi,
308                 unsigned int *result, int gc_type, int type, char alloc_mode)
309 {
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;
316 
317         mutex_lock(&dirty_i->seglist_lock);
318         last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
319 
320         p.alloc_mode = alloc_mode;
321         select_policy(sbi, gc_type, type, &p);
322 
323         p.min_segno = NULL_SEGNO;
324         p.min_cost = get_max_cost(sbi, &p);
325 
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;
330                 goto out;
331         }
332 
333         if (p.max_search == 0)
334                 goto out;
335 
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;
341                         goto got_result;
342                 }
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;
348                         goto got_result;
349                 }
350         }
351 
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)
356                         goto got_it;
357         }
358 
359         while (1) {
360                 unsigned long cost;
361                 unsigned int segno;
362 
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]) {
366                                 last_segment =
367                                         sm->last_victim[p.gc_mode];
368                                 sm->last_victim[p.gc_mode] = 0;
369                                 p.offset = 0;
370                                 continue;
371                         }
372                         break;
373                 }
374 
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,
380                                                 p.ofs_unit);
381                 } else {
382                         nsearched++;
383                 }
384 
385                 secno = GET_SEC_FROM_SEG(sbi, segno);
386 
387                 if (sec_usage_check(sbi, secno))
388                         goto next;
389                 /* Don't touch checkpointed data */
390                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
391                                         get_ckpt_valid_blocks(sbi, segno) &&
392                                         p.alloc_mode != SSR))
393                         goto next;
394                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
395                         goto next;
396 
397                 cost = get_gc_cost(sbi, segno, &p);
398 
399                 if (p.min_cost > cost) {
400                         p.min_segno = segno;
401                         p.min_cost = cost;
402                 }
403 next:
404                 if (nsearched >= p.max_search) {
405                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
406                                 sm->last_victim[p.gc_mode] = last_victim + 1;
407                         else
408                                 sm->last_victim[p.gc_mode] = segno + 1;
409                         sm->last_victim[p.gc_mode] %=
410                                 (MAIN_SECS(sbi) * sbi->segs_per_sec);
411                         break;
412                 }
413         }
414         if (p.min_segno != NULL_SEGNO) {
415 got_it:
416                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
417 got_result:
418                 if (p.alloc_mode == LFS) {
419                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
420                         if (gc_type == FG_GC)
421                                 sbi->cur_victim_sec = secno;
422                         else
423                                 set_bit(secno, dirty_i->victim_secmap);
424                 }
425 
426         }
427 out:
428         if (p.min_segno != NULL_SEGNO)
429                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
430                                 sbi->cur_victim_sec,
431                                 prefree_segments(sbi), free_segments(sbi));
432         mutex_unlock(&dirty_i->seglist_lock);
433 
434         return (p.min_segno == NULL_SEGNO) ? 0 : 1;
435 }
436 
437 static const struct victim_selection default_v_ops = {
438         .get_victim = get_victim_by_default,
439 };
440 
441 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
442 {
443         struct inode_entry *ie;
444 
445         ie = radix_tree_lookup(&gc_list->iroot, ino);
446         if (ie)
447                 return ie->inode;
448         return NULL;
449 }
450 
451 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
452 {
453         struct inode_entry *new_ie;
454 
455         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
456                 iput(inode);
457                 return;
458         }
459         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
460         new_ie->inode = inode;
461 
462         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
463         list_add_tail(&new_ie->list, &gc_list->ilist);
464 }
465 
466 static void put_gc_inode(struct gc_inode_list *gc_list)
467 {
468         struct inode_entry *ie, *next_ie;
469         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
470                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
471                 iput(ie->inode);
472                 list_del(&ie->list);
473                 kmem_cache_free(f2fs_inode_entry_slab, ie);
474         }
475 }
476 
477 static int check_valid_map(struct f2fs_sb_info *sbi,
478                                 unsigned int segno, int offset)
479 {
480         struct sit_info *sit_i = SIT_I(sbi);
481         struct seg_entry *sentry;
482         int ret;
483 
484         down_read(&sit_i->sentry_lock);
485         sentry = get_seg_entry(sbi, segno);
486         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
487         up_read(&sit_i->sentry_lock);
488         return ret;
489 }
490 
491 /*
492  * This function compares node address got in summary with that in NAT.
493  * On validity, copy that node with cold status, otherwise (invalid node)
494  * ignore that.
495  */
496 static int gc_node_segment(struct f2fs_sb_info *sbi,
497                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
498 {
499         struct f2fs_summary *entry;
500         block_t start_addr;
501         int off;
502         int phase = 0;
503         bool fggc = (gc_type == FG_GC);
504         int submitted = 0;
505 
506         start_addr = START_BLOCK(sbi, segno);
507 
508 next_step:
509         entry = sum;
510 
511         if (fggc && phase == 2)
512                 atomic_inc(&sbi->wb_sync_req[NODE]);
513 
514         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
515                 nid_t nid = le32_to_cpu(entry->nid);
516                 struct page *node_page;
517                 struct node_info ni;
518                 int err;
519 
520                 /* stop BG_GC if there is not enough free sections. */
521                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
522                         return submitted;
523 
524                 if (check_valid_map(sbi, segno, off) == 0)
525                         continue;
526 
527                 if (phase == 0) {
528                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
529                                                         META_NAT, true);
530                         continue;
531                 }
532 
533                 if (phase == 1) {
534                         f2fs_ra_node_page(sbi, nid);
535                         continue;
536                 }
537 
538                 /* phase == 2 */
539                 node_page = f2fs_get_node_page(sbi, nid);
540                 if (IS_ERR(node_page))
541                         continue;
542 
543                 /* block may become invalid during f2fs_get_node_page */
544                 if (check_valid_map(sbi, segno, off) == 0) {
545                         f2fs_put_page(node_page, 1);
546                         continue;
547                 }
548 
549                 if (f2fs_get_node_info(sbi, nid, &ni)) {
550                         f2fs_put_page(node_page, 1);
551                         continue;
552                 }
553 
554                 if (ni.blk_addr != start_addr + off) {
555                         f2fs_put_page(node_page, 1);
556                         continue;
557                 }
558 
559                 err = f2fs_move_node_page(node_page, gc_type);
560                 if (!err && gc_type == FG_GC)
561                         submitted++;
562                 stat_inc_node_blk_count(sbi, 1, gc_type);
563         }
564 
565         if (++phase < 3)
566                 goto next_step;
567 
568         if (fggc)
569                 atomic_dec(&sbi->wb_sync_req[NODE]);
570         return submitted;
571 }
572 
573 /*
574  * Calculate start block index indicating the given node offset.
575  * Be careful, caller should give this node offset only indicating direct node
576  * blocks. If any node offsets, which point the other types of node blocks such
577  * as indirect or double indirect node blocks, are given, it must be a caller's
578  * bug.
579  */
580 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
581 {
582         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
583         unsigned int bidx;
584 
585         if (node_ofs == 0)
586                 return 0;
587 
588         if (node_ofs <= 2) {
589                 bidx = node_ofs - 1;
590         } else if (node_ofs <= indirect_blks) {
591                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
592                 bidx = node_ofs - 2 - dec;
593         } else {
594                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
595                 bidx = node_ofs - 5 - dec;
596         }
597         return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
598 }
599 
600 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
601                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
602 {
603         struct page *node_page;
604         nid_t nid;
605         unsigned int ofs_in_node;
606         block_t source_blkaddr;
607 
608         nid = le32_to_cpu(sum->nid);
609         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
610 
611         node_page = f2fs_get_node_page(sbi, nid);
612         if (IS_ERR(node_page))
613                 return false;
614 
615         if (f2fs_get_node_info(sbi, nid, dni)) {
616                 f2fs_put_page(node_page, 1);
617                 return false;
618         }
619 
620         if (sum->version != dni->version) {
621                 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
622                           __func__);
623                 set_sbi_flag(sbi, SBI_NEED_FSCK);
624         }
625 
626         *nofs = ofs_of_node(node_page);
627         source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node);
628         f2fs_put_page(node_page, 1);
629 
630         if (source_blkaddr != blkaddr)
631                 return false;
632         return true;
633 }
634 
635 static int ra_data_block(struct inode *inode, pgoff_t index)
636 {
637         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
638         struct address_space *mapping = inode->i_mapping;
639         struct dnode_of_data dn;
640         struct page *page;
641         struct extent_info ei = {0, 0, 0};
642         struct f2fs_io_info fio = {
643                 .sbi = sbi,
644                 .ino = inode->i_ino,
645                 .type = DATA,
646                 .temp = COLD,
647                 .op = REQ_OP_READ,
648                 .op_flags = 0,
649                 .encrypted_page = NULL,
650                 .in_list = false,
651                 .retry = false,
652         };
653         int err;
654 
655         page = f2fs_grab_cache_page(mapping, index, true);
656         if (!page)
657                 return -ENOMEM;
658 
659         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
660                 dn.data_blkaddr = ei.blk + index - ei.fofs;
661                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
662                                                 DATA_GENERIC_ENHANCE_READ))) {
663                         err = -EFSCORRUPTED;
664                         goto put_page;
665                 }
666                 goto got_it;
667         }
668 
669         set_new_dnode(&dn, inode, NULL, NULL, 0);
670         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
671         if (err)
672                 goto put_page;
673         f2fs_put_dnode(&dn);
674 
675         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
676                 err = -ENOENT;
677                 goto put_page;
678         }
679         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
680                                                 DATA_GENERIC_ENHANCE))) {
681                 err = -EFSCORRUPTED;
682                 goto put_page;
683         }
684 got_it:
685         /* read page */
686         fio.page = page;
687         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
688 
689         /*
690          * don't cache encrypted data into meta inode until previous dirty
691          * data were writebacked to avoid racing between GC and flush.
692          */
693         f2fs_wait_on_page_writeback(page, DATA, true, true);
694 
695         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
696 
697         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
698                                         dn.data_blkaddr,
699                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
700         if (!fio.encrypted_page) {
701                 err = -ENOMEM;
702                 goto put_page;
703         }
704 
705         err = f2fs_submit_page_bio(&fio);
706         if (err)
707                 goto put_encrypted_page;
708         f2fs_put_page(fio.encrypted_page, 0);
709         f2fs_put_page(page, 1);
710         return 0;
711 put_encrypted_page:
712         f2fs_put_page(fio.encrypted_page, 1);
713 put_page:
714         f2fs_put_page(page, 1);
715         return err;
716 }
717 
718 /*
719  * Move data block via META_MAPPING while keeping locked data page.
720  * This can be used to move blocks, aka LBAs, directly on disk.
721  */
722 static int move_data_block(struct inode *inode, block_t bidx,
723                                 int gc_type, unsigned int segno, int off)
724 {
725         struct f2fs_io_info fio = {
726                 .sbi = F2FS_I_SB(inode),
727                 .ino = inode->i_ino,
728                 .type = DATA,
729                 .temp = COLD,
730                 .op = REQ_OP_READ,
731                 .op_flags = 0,
732                 .encrypted_page = NULL,
733                 .in_list = false,
734                 .retry = false,
735         };
736         struct dnode_of_data dn;
737         struct f2fs_summary sum;
738         struct node_info ni;
739         struct page *page, *mpage;
740         block_t newaddr;
741         int err = 0;
742         bool lfs_mode = test_opt(fio.sbi, LFS);
743 
744         /* do not read out */
745         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
746         if (!page)
747                 return -ENOMEM;
748 
749         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
750                 err = -ENOENT;
751                 goto out;
752         }
753 
754         if (f2fs_is_atomic_file(inode)) {
755                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
756                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
757                 err = -EAGAIN;
758                 goto out;
759         }
760 
761         if (f2fs_is_pinned_file(inode)) {
762                 f2fs_pin_file_control(inode, true);
763                 err = -EAGAIN;
764                 goto out;
765         }
766 
767         set_new_dnode(&dn, inode, NULL, NULL, 0);
768         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
769         if (err)
770                 goto out;
771 
772         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
773                 ClearPageUptodate(page);
774                 err = -ENOENT;
775                 goto put_out;
776         }
777 
778         /*
779          * don't cache encrypted data into meta inode until previous dirty
780          * data were writebacked to avoid racing between GC and flush.
781          */
782         f2fs_wait_on_page_writeback(page, DATA, true, true);
783 
784         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
785 
786         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
787         if (err)
788                 goto put_out;
789 
790         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
791 
792         /* read page */
793         fio.page = page;
794         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
795 
796         if (lfs_mode)
797                 down_write(&fio.sbi->io_order_lock);
798 
799         mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
800                                         fio.old_blkaddr, false);
801         if (!mpage)
802                 goto up_out;
803 
804         fio.encrypted_page = mpage;
805 
806         /* read source block in mpage */
807         if (!PageUptodate(mpage)) {
808                 err = f2fs_submit_page_bio(&fio);
809                 if (err) {
810                         f2fs_put_page(mpage, 1);
811                         goto up_out;
812                 }
813                 lock_page(mpage);
814                 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
815                                                 !PageUptodate(mpage))) {
816                         err = -EIO;
817                         f2fs_put_page(mpage, 1);
818                         goto up_out;
819                 }
820         }
821 
822         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
823                                         &sum, CURSEG_COLD_DATA, NULL, false);
824 
825         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
826                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
827         if (!fio.encrypted_page) {
828                 err = -ENOMEM;
829                 f2fs_put_page(mpage, 1);
830                 goto recover_block;
831         }
832 
833         /* write target block */
834         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
835         memcpy(page_address(fio.encrypted_page),
836                                 page_address(mpage), PAGE_SIZE);
837         f2fs_put_page(mpage, 1);
838         invalidate_mapping_pages(META_MAPPING(fio.sbi),
839                                 fio.old_blkaddr, fio.old_blkaddr);
840 
841         set_page_dirty(fio.encrypted_page);
842         if (clear_page_dirty_for_io(fio.encrypted_page))
843                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
844 
845         set_page_writeback(fio.encrypted_page);
846         ClearPageError(page);
847 
848         /* allocate block address */
849         f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
850 
851         fio.op = REQ_OP_WRITE;
852         fio.op_flags = REQ_SYNC;
853         fio.new_blkaddr = newaddr;
854         f2fs_submit_page_write(&fio);
855         if (fio.retry) {
856                 err = -EAGAIN;
857                 if (PageWriteback(fio.encrypted_page))
858                         end_page_writeback(fio.encrypted_page);
859                 goto put_page_out;
860         }
861 
862         f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
863 
864         f2fs_update_data_blkaddr(&dn, newaddr);
865         set_inode_flag(inode, FI_APPEND_WRITE);
866         if (page->index == 0)
867                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
868 put_page_out:
869         f2fs_put_page(fio.encrypted_page, 1);
870 recover_block:
871         if (err)
872                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
873                                                                 true, true);
874 up_out:
875         if (lfs_mode)
876                 up_write(&fio.sbi->io_order_lock);
877 put_out:
878         f2fs_put_dnode(&dn);
879 out:
880         f2fs_put_page(page, 1);
881         return err;
882 }
883 
884 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
885                                                         unsigned int segno, int off)
886 {
887         struct page *page;
888         int err = 0;
889 
890         page = f2fs_get_lock_data_page(inode, bidx, true);
891         if (IS_ERR(page))
892                 return PTR_ERR(page);
893 
894         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
895                 err = -ENOENT;
896                 goto out;
897         }
898 
899         if (f2fs_is_atomic_file(inode)) {
900                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
901                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
902                 err = -EAGAIN;
903                 goto out;
904         }
905         if (f2fs_is_pinned_file(inode)) {
906                 if (gc_type == FG_GC)
907                         f2fs_pin_file_control(inode, true);
908                 err = -EAGAIN;
909                 goto out;
910         }
911 
912         if (gc_type == BG_GC) {
913                 if (PageWriteback(page)) {
914                         err = -EAGAIN;
915                         goto out;
916                 }
917                 set_page_dirty(page);
918                 set_cold_data(page);
919         } else {
920                 struct f2fs_io_info fio = {
921                         .sbi = F2FS_I_SB(inode),
922                         .ino = inode->i_ino,
923                         .type = DATA,
924                         .temp = COLD,
925                         .op = REQ_OP_WRITE,
926                         .op_flags = REQ_SYNC,
927                         .old_blkaddr = NULL_ADDR,
928                         .page = page,
929                         .encrypted_page = NULL,
930                         .need_lock = LOCK_REQ,
931                         .io_type = FS_GC_DATA_IO,
932                 };
933                 bool is_dirty = PageDirty(page);
934 
935 retry:
936                 f2fs_wait_on_page_writeback(page, DATA, true, true);
937 
938                 set_page_dirty(page);
939                 if (clear_page_dirty_for_io(page)) {
940                         inode_dec_dirty_pages(inode);
941                         f2fs_remove_dirty_inode(inode);
942                 }
943 
944                 set_cold_data(page);
945 
946                 err = f2fs_do_write_data_page(&fio);
947                 if (err) {
948                         clear_cold_data(page);
949                         if (err == -ENOMEM) {
950                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
951                                 goto retry;
952                         }
953                         if (is_dirty)
954                                 set_page_dirty(page);
955                 }
956         }
957 out:
958         f2fs_put_page(page, 1);
959         return err;
960 }
961 
962 /*
963  * This function tries to get parent node of victim data block, and identifies
964  * data block validity. If the block is valid, copy that with cold status and
965  * modify parent node.
966  * If the parent node is not valid or the data block address is different,
967  * the victim data block is ignored.
968  */
969 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
970                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
971 {
972         struct super_block *sb = sbi->sb;
973         struct f2fs_summary *entry;
974         block_t start_addr;
975         int off;
976         int phase = 0;
977         int submitted = 0;
978 
979         start_addr = START_BLOCK(sbi, segno);
980 
981 next_step:
982         entry = sum;
983 
984         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
985                 struct page *data_page;
986                 struct inode *inode;
987                 struct node_info dni; /* dnode info for the data */
988                 unsigned int ofs_in_node, nofs;
989                 block_t start_bidx;
990                 nid_t nid = le32_to_cpu(entry->nid);
991 
992                 /* stop BG_GC if there is not enough free sections. */
993                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
994                         return submitted;
995 
996                 if (check_valid_map(sbi, segno, off) == 0)
997                         continue;
998 
999                 if (phase == 0) {
1000                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1001                                                         META_NAT, true);
1002                         continue;
1003                 }
1004 
1005                 if (phase == 1) {
1006                         f2fs_ra_node_page(sbi, nid);
1007                         continue;
1008                 }
1009 
1010                 /* Get an inode by ino with checking validity */
1011                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1012                         continue;
1013 
1014                 if (phase == 2) {
1015                         f2fs_ra_node_page(sbi, dni.ino);
1016                         continue;
1017                 }
1018 
1019                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1020 
1021                 if (phase == 3) {
1022                         inode = f2fs_iget(sb, dni.ino);
1023                         if (IS_ERR(inode) || is_bad_inode(inode))
1024                                 continue;
1025 
1026                         if (!down_write_trylock(
1027                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1028                                 iput(inode);
1029                                 sbi->skipped_gc_rwsem++;
1030                                 continue;
1031                         }
1032 
1033                         start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1034                                                                 ofs_in_node;
1035 
1036                         if (f2fs_post_read_required(inode)) {
1037                                 int err = ra_data_block(inode, start_bidx);
1038 
1039                                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1040                                 if (err) {
1041                                         iput(inode);
1042                                         continue;
1043                                 }
1044                                 add_gc_inode(gc_list, inode);
1045                                 continue;
1046                         }
1047 
1048                         data_page = f2fs_get_read_data_page(inode,
1049                                                 start_bidx, REQ_RAHEAD, true);
1050                         up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1051                         if (IS_ERR(data_page)) {
1052                                 iput(inode);
1053                                 continue;
1054                         }
1055 
1056                         f2fs_put_page(data_page, 0);
1057                         add_gc_inode(gc_list, inode);
1058                         continue;
1059                 }
1060 
1061                 /* phase 4 */
1062                 inode = find_gc_inode(gc_list, dni.ino);
1063                 if (inode) {
1064                         struct f2fs_inode_info *fi = F2FS_I(inode);
1065                         bool locked = false;
1066                         int err;
1067 
1068                         if (S_ISREG(inode->i_mode)) {
1069                                 if (!down_write_trylock(&fi->i_gc_rwsem[READ]))
1070                                         continue;
1071                                 if (!down_write_trylock(
1072                                                 &fi->i_gc_rwsem[WRITE])) {
1073                                         sbi->skipped_gc_rwsem++;
1074                                         up_write(&fi->i_gc_rwsem[READ]);
1075                                         continue;
1076                                 }
1077                                 locked = true;
1078 
1079                                 /* wait for all inflight aio data */
1080                                 inode_dio_wait(inode);
1081                         }
1082 
1083                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1084                                                                 + ofs_in_node;
1085                         if (f2fs_post_read_required(inode))
1086                                 err = move_data_block(inode, start_bidx,
1087                                                         gc_type, segno, off);
1088                         else
1089                                 err = move_data_page(inode, start_bidx, gc_type,
1090                                                                 segno, off);
1091 
1092                         if (!err && (gc_type == FG_GC ||
1093                                         f2fs_post_read_required(inode)))
1094                                 submitted++;
1095 
1096                         if (locked) {
1097                                 up_write(&fi->i_gc_rwsem[WRITE]);
1098                                 up_write(&fi->i_gc_rwsem[READ]);
1099                         }
1100 
1101                         stat_inc_data_blk_count(sbi, 1, gc_type);
1102                 }
1103         }
1104 
1105         if (++phase < 5)
1106                 goto next_step;
1107 
1108         return submitted;
1109 }
1110 
1111 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1112                         int gc_type)
1113 {
1114         struct sit_info *sit_i = SIT_I(sbi);
1115         int ret;
1116 
1117         down_write(&sit_i->sentry_lock);
1118         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1119                                               NO_CHECK_TYPE, LFS);
1120         up_write(&sit_i->sentry_lock);
1121         return ret;
1122 }
1123 
1124 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1125                                 unsigned int start_segno,
1126                                 struct gc_inode_list *gc_list, int gc_type)
1127 {
1128         struct page *sum_page;
1129         struct f2fs_summary_block *sum;
1130         struct blk_plug plug;
1131         unsigned int segno = start_segno;
1132         unsigned int end_segno = start_segno + sbi->segs_per_sec;
1133         int seg_freed = 0, migrated = 0;
1134         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1135                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
1136         int submitted = 0;
1137 
1138         if (__is_large_section(sbi))
1139                 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1140 
1141         /* readahead multi ssa blocks those have contiguous address */
1142         if (__is_large_section(sbi))
1143                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1144                                         end_segno - segno, META_SSA, true);
1145 
1146         /* reference all summary page */
1147         while (segno < end_segno) {
1148                 sum_page = f2fs_get_sum_page(sbi, segno++);
1149                 if (IS_ERR(sum_page)) {
1150                         int err = PTR_ERR(sum_page);
1151 
1152                         end_segno = segno - 1;
1153                         for (segno = start_segno; segno < end_segno; segno++) {
1154                                 sum_page = find_get_page(META_MAPPING(sbi),
1155                                                 GET_SUM_BLOCK(sbi, segno));
1156                                 f2fs_put_page(sum_page, 0);
1157                                 f2fs_put_page(sum_page, 0);
1158                         }
1159                         return err;
1160                 }
1161                 unlock_page(sum_page);
1162         }
1163 
1164         blk_start_plug(&plug);
1165 
1166         for (segno = start_segno; segno < end_segno; segno++) {
1167 
1168                 /* find segment summary of victim */
1169                 sum_page = find_get_page(META_MAPPING(sbi),
1170                                         GET_SUM_BLOCK(sbi, segno));
1171                 f2fs_put_page(sum_page, 0);
1172 
1173                 if (get_valid_blocks(sbi, segno, false) == 0)
1174                         goto freed;
1175                 if (__is_large_section(sbi) &&
1176                                 migrated >= sbi->migration_granularity)
1177                         goto skip;
1178                 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1179                         goto skip;
1180 
1181                 sum = page_address(sum_page);
1182                 if (type != GET_SUM_TYPE((&sum->footer))) {
1183                         f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1184                                  segno, type, GET_SUM_TYPE((&sum->footer)));
1185                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1186                         f2fs_stop_checkpoint(sbi, false);
1187                         goto skip;
1188                 }
1189 
1190                 /*
1191                  * this is to avoid deadlock:
1192                  * - lock_page(sum_page)         - f2fs_replace_block
1193                  *  - check_valid_map()            - down_write(sentry_lock)
1194                  *   - down_read(sentry_lock)     - change_curseg()
1195                  *                                  - lock_page(sum_page)
1196                  */
1197                 if (type == SUM_TYPE_NODE)
1198                         submitted += gc_node_segment(sbi, sum->entries, segno,
1199                                                                 gc_type);
1200                 else
1201                         submitted += gc_data_segment(sbi, sum->entries, gc_list,
1202                                                         segno, gc_type);
1203 
1204                 stat_inc_seg_count(sbi, type, gc_type);
1205 
1206 freed:
1207                 if (gc_type == FG_GC &&
1208                                 get_valid_blocks(sbi, segno, false) == 0)
1209                         seg_freed++;
1210                 migrated++;
1211 
1212                 if (__is_large_section(sbi) && segno + 1 < end_segno)
1213                         sbi->next_victim_seg[gc_type] = segno + 1;
1214 skip:
1215                 f2fs_put_page(sum_page, 0);
1216         }
1217 
1218         if (submitted)
1219                 f2fs_submit_merged_write(sbi,
1220                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1221 
1222         blk_finish_plug(&plug);
1223 
1224         stat_inc_call_count(sbi->stat_info);
1225 
1226         return seg_freed;
1227 }
1228 
1229 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1230                         bool background, unsigned int segno)
1231 {
1232         int gc_type = sync ? FG_GC : BG_GC;
1233         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1234         int ret = 0;
1235         struct cp_control cpc;
1236         unsigned int init_segno = segno;
1237         struct gc_inode_list gc_list = {
1238                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1239                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1240         };
1241         unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1242         unsigned long long first_skipped;
1243         unsigned int skipped_round = 0, round = 0;
1244 
1245         trace_f2fs_gc_begin(sbi->sb, sync, background,
1246                                 get_pages(sbi, F2FS_DIRTY_NODES),
1247                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1248                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1249                                 free_sections(sbi),
1250                                 free_segments(sbi),
1251                                 reserved_segments(sbi),
1252                                 prefree_segments(sbi));
1253 
1254         cpc.reason = __get_cp_reason(sbi);
1255         sbi->skipped_gc_rwsem = 0;
1256         first_skipped = last_skipped;
1257 gc_more:
1258         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1259                 ret = -EINVAL;
1260                 goto stop;
1261         }
1262         if (unlikely(f2fs_cp_error(sbi))) {
1263                 ret = -EIO;
1264                 goto stop;
1265         }
1266 
1267         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1268                 /*
1269                  * For example, if there are many prefree_segments below given
1270                  * threshold, we can make them free by checkpoint. Then, we
1271                  * secure free segments which doesn't need fggc any more.
1272                  */
1273                 if (prefree_segments(sbi) &&
1274                                 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1275                         ret = f2fs_write_checkpoint(sbi, &cpc);
1276                         if (ret)
1277                                 goto stop;
1278                 }
1279                 if (has_not_enough_free_secs(sbi, 0, 0))
1280                         gc_type = FG_GC;
1281         }
1282 
1283         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1284         if (gc_type == BG_GC && !background) {
1285                 ret = -EINVAL;
1286                 goto stop;
1287         }
1288         if (!__get_victim(sbi, &segno, gc_type)) {
1289                 ret = -ENODATA;
1290                 goto stop;
1291         }
1292 
1293         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1294         if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1295                 sec_freed++;
1296         total_freed += seg_freed;
1297 
1298         if (gc_type == FG_GC) {
1299                 if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1300                                                 sbi->skipped_gc_rwsem)
1301                         skipped_round++;
1302                 last_skipped = sbi->skipped_atomic_files[FG_GC];
1303                 round++;
1304         }
1305 
1306         if (gc_type == FG_GC)
1307                 sbi->cur_victim_sec = NULL_SEGNO;
1308 
1309         if (sync)
1310                 goto stop;
1311 
1312         if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1313                 if (skipped_round <= MAX_SKIP_GC_COUNT ||
1314                                         skipped_round * 2 < round) {
1315                         segno = NULL_SEGNO;
1316                         goto gc_more;
1317                 }
1318 
1319                 if (first_skipped < last_skipped &&
1320                                 (last_skipped - first_skipped) >
1321                                                 sbi->skipped_gc_rwsem) {
1322                         f2fs_drop_inmem_pages_all(sbi, true);
1323                         segno = NULL_SEGNO;
1324                         goto gc_more;
1325                 }
1326                 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1327                         ret = f2fs_write_checkpoint(sbi, &cpc);
1328         }
1329 stop:
1330         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1331         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1332 
1333         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1334                                 get_pages(sbi, F2FS_DIRTY_NODES),
1335                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1336                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1337                                 free_sections(sbi),
1338                                 free_segments(sbi),
1339                                 reserved_segments(sbi),
1340                                 prefree_segments(sbi));
1341 
1342         mutex_unlock(&sbi->gc_mutex);
1343 
1344         put_gc_inode(&gc_list);
1345 
1346         if (sync && !ret)
1347                 ret = sec_freed ? 0 : -EAGAIN;
1348         return ret;
1349 }
1350 
1351 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1352 {
1353         DIRTY_I(sbi)->v_ops = &default_v_ops;
1354 
1355         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1356 
1357         /* give warm/cold data area from slower device */
1358         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1359                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1360                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1361 }
1362 
1363 static int free_segment_range(struct f2fs_sb_info *sbi, unsigned int start,
1364                                                         unsigned int end)
1365 {
1366         int type;
1367         unsigned int segno, next_inuse;
1368         int err = 0;
1369 
1370         /* Move out cursegs from the target range */
1371         for (type = CURSEG_HOT_DATA; type < NR_CURSEG_TYPE; type++)
1372                 allocate_segment_for_resize(sbi, type, start, end);
1373 
1374         /* do GC to move out valid blocks in the range */
1375         for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1376                 struct gc_inode_list gc_list = {
1377                         .ilist = LIST_HEAD_INIT(gc_list.ilist),
1378                         .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1379                 };
1380 
1381                 mutex_lock(&sbi->gc_mutex);
1382                 do_garbage_collect(sbi, segno, &gc_list, FG_GC);
1383                 mutex_unlock(&sbi->gc_mutex);
1384                 put_gc_inode(&gc_list);
1385 
1386                 if (get_valid_blocks(sbi, segno, true))
1387                         return -EAGAIN;
1388         }
1389 
1390         err = f2fs_sync_fs(sbi->sb, 1);
1391         if (err)
1392                 return err;
1393 
1394         next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1395         if (next_inuse <= end) {
1396                 f2fs_err(sbi, "segno %u should be free but still inuse!",
1397                          next_inuse);
1398                 f2fs_bug_on(sbi, 1);
1399         }
1400         return err;
1401 }
1402 
1403 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1404 {
1405         struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1406         int section_count = le32_to_cpu(raw_sb->section_count);
1407         int segment_count = le32_to_cpu(raw_sb->segment_count);
1408         int segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1409         long long block_count = le64_to_cpu(raw_sb->block_count);
1410         int segs = secs * sbi->segs_per_sec;
1411 
1412         raw_sb->section_count = cpu_to_le32(section_count + secs);
1413         raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1414         raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1415         raw_sb->block_count = cpu_to_le64(block_count +
1416                                         (long long)segs * sbi->blocks_per_seg);
1417 }
1418 
1419 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1420 {
1421         int segs = secs * sbi->segs_per_sec;
1422         long long user_block_count =
1423                                 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1424 
1425         SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1426         MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1427         FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1428         FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1429         F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count +
1430                                         (long long)segs * sbi->blocks_per_seg);
1431 }
1432 
1433 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1434 {
1435         __u64 old_block_count, shrunk_blocks;
1436         unsigned int secs;
1437         int gc_mode, gc_type;
1438         int err = 0;
1439         __u32 rem;
1440 
1441         old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1442         if (block_count > old_block_count)
1443                 return -EINVAL;
1444 
1445         /* new fs size should align to section size */
1446         div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
1447         if (rem)
1448                 return -EINVAL;
1449 
1450         if (block_count == old_block_count)
1451                 return 0;
1452 
1453         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1454                 f2fs_err(sbi, "Should run fsck to repair first.");
1455                 return -EFSCORRUPTED;
1456         }
1457 
1458         if (test_opt(sbi, DISABLE_CHECKPOINT)) {
1459                 f2fs_err(sbi, "Checkpoint should be enabled.");
1460                 return -EINVAL;
1461         }
1462 
1463         freeze_bdev(sbi->sb->s_bdev);
1464 
1465         shrunk_blocks = old_block_count - block_count;
1466         secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
1467         spin_lock(&sbi->stat_lock);
1468         if (shrunk_blocks + valid_user_blocks(sbi) +
1469                 sbi->current_reserved_blocks + sbi->unusable_block_count +
1470                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
1471                 err = -ENOSPC;
1472         else
1473                 sbi->user_block_count -= shrunk_blocks;
1474         spin_unlock(&sbi->stat_lock);
1475         if (err) {
1476                 thaw_bdev(sbi->sb->s_bdev, sbi->sb);
1477                 return err;
1478         }
1479 
1480         mutex_lock(&sbi->resize_mutex);
1481         set_sbi_flag(sbi, SBI_IS_RESIZEFS);
1482 
1483         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1484 
1485         MAIN_SECS(sbi) -= secs;
1486 
1487         for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1488                 if (SIT_I(sbi)->last_victim[gc_mode] >=
1489                                         MAIN_SECS(sbi) * sbi->segs_per_sec)
1490                         SIT_I(sbi)->last_victim[gc_mode] = 0;
1491 
1492         for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1493                 if (sbi->next_victim_seg[gc_type] >=
1494                                         MAIN_SECS(sbi) * sbi->segs_per_sec)
1495                         sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1496 
1497         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1498 
1499         err = free_segment_range(sbi, MAIN_SECS(sbi) * sbi->segs_per_sec,
1500                         MAIN_SEGS(sbi) - 1);
1501         if (err)
1502                 goto out;
1503 
1504         update_sb_metadata(sbi, -secs);
1505 
1506         err = f2fs_commit_super(sbi, false);
1507         if (err) {
1508                 update_sb_metadata(sbi, secs);
1509                 goto out;
1510         }
1511 
1512         update_fs_metadata(sbi, -secs);
1513         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
1514         err = f2fs_sync_fs(sbi->sb, 1);
1515         if (err) {
1516                 update_fs_metadata(sbi, secs);
1517                 update_sb_metadata(sbi, secs);
1518                 f2fs_commit_super(sbi, false);
1519         }
1520 out:
1521         if (err) {
1522                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1523                 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
1524 
1525                 MAIN_SECS(sbi) += secs;
1526                 spin_lock(&sbi->stat_lock);
1527                 sbi->user_block_count += shrunk_blocks;
1528                 spin_unlock(&sbi->stat_lock);
1529         }
1530         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
1531         mutex_unlock(&sbi->resize_mutex);
1532         thaw_bdev(sbi->sb->s_bdev, sbi->sb);
1533         return err;
1534 }
1535 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp