Commit | Line | Data |
---|---|---|
0a8165d7 | 1 | /* |
7bc09003 JK |
2 | * fs/f2fs/gc.c |
3 | * | |
4 | * Copyright (c) 2012 Samsung Electronics Co., Ltd. | |
5 | * http://www.samsung.com/ | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | */ | |
11 | #include <linux/fs.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/backing-dev.h> | |
7bc09003 JK |
14 | #include <linux/init.h> |
15 | #include <linux/f2fs_fs.h> | |
16 | #include <linux/kthread.h> | |
17 | #include <linux/delay.h> | |
18 | #include <linux/freezer.h> | |
7bc09003 JK |
19 | |
20 | #include "f2fs.h" | |
21 | #include "node.h" | |
22 | #include "segment.h" | |
23 | #include "gc.h" | |
8e46b3ed | 24 | #include <trace/events/f2fs.h> |
7bc09003 | 25 | |
7bc09003 JK |
26 | static int gc_thread_func(void *data) |
27 | { | |
28 | struct f2fs_sb_info *sbi = data; | |
b59d0bae | 29 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; |
7bc09003 JK |
30 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; |
31 | long wait_ms; | |
32 | ||
b59d0bae | 33 | wait_ms = gc_th->min_sleep_time; |
7bc09003 JK |
34 | |
35 | do { | |
36 | if (try_to_freeze()) | |
37 | continue; | |
38 | else | |
39 | wait_event_interruptible_timeout(*wq, | |
40 | kthread_should_stop(), | |
41 | msecs_to_jiffies(wait_ms)); | |
42 | if (kthread_should_stop()) | |
43 | break; | |
44 | ||
d6212a5f | 45 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
88dd8934 | 46 | increase_sleep_time(gc_th, &wait_ms); |
d6212a5f CL |
47 | continue; |
48 | } | |
49 | ||
7bc09003 JK |
50 | /* |
51 | * [GC triggering condition] | |
52 | * 0. GC is not conducted currently. | |
53 | * 1. There are enough dirty segments. | |
54 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
55 | * 3. IO subsystem is idle by checking the # of requests in | |
56 | * bdev's request list. | |
57 | * | |
e1c42045 | 58 | * Note) We have to avoid triggering GCs frequently. |
7bc09003 JK |
59 | * Because it is possible that some segments can be |
60 | * invalidated soon after by user update or deletion. | |
61 | * So, I'd like to wait some time to collect dirty segments. | |
62 | */ | |
63 | if (!mutex_trylock(&sbi->gc_mutex)) | |
64 | continue; | |
65 | ||
66 | if (!is_idle(sbi)) { | |
88dd8934 | 67 | increase_sleep_time(gc_th, &wait_ms); |
7bc09003 JK |
68 | mutex_unlock(&sbi->gc_mutex); |
69 | continue; | |
70 | } | |
71 | ||
72 | if (has_enough_invalid_blocks(sbi)) | |
88dd8934 | 73 | decrease_sleep_time(gc_th, &wait_ms); |
7bc09003 | 74 | else |
88dd8934 | 75 | increase_sleep_time(gc_th, &wait_ms); |
7bc09003 | 76 | |
dcdfff65 | 77 | stat_inc_bggc_count(sbi); |
7bc09003 | 78 | |
43727527 | 79 | /* if return value is not zero, no victim was selected */ |
6aefd93b | 80 | if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC))) |
b59d0bae | 81 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 82 | |
84e4214f JK |
83 | trace_f2fs_background_gc(sbi->sb, wait_ms, |
84 | prefree_segments(sbi), free_segments(sbi)); | |
85 | ||
4660f9c0 JK |
86 | /* balancing f2fs's metadata periodically */ |
87 | f2fs_balance_fs_bg(sbi); | |
81eb8d6e | 88 | |
7bc09003 JK |
89 | } while (!kthread_should_stop()); |
90 | return 0; | |
91 | } | |
92 | ||
93 | int start_gc_thread(struct f2fs_sb_info *sbi) | |
94 | { | |
1042d60f | 95 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 96 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 97 | int err = 0; |
7bc09003 JK |
98 | |
99 | gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL); | |
7a267f8d NJ |
100 | if (!gc_th) { |
101 | err = -ENOMEM; | |
102 | goto out; | |
103 | } | |
7bc09003 | 104 | |
b59d0bae NJ |
105 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
106 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
107 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
108 | ||
d2dc095f NJ |
109 | gc_th->gc_idle = 0; |
110 | ||
7bc09003 JK |
111 | sbi->gc_thread = gc_th; |
112 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
113 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 114 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 115 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 116 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 117 | kfree(gc_th); |
25718423 | 118 | sbi->gc_thread = NULL; |
7bc09003 | 119 | } |
7a267f8d NJ |
120 | out: |
121 | return err; | |
7bc09003 JK |
122 | } |
123 | ||
124 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
125 | { | |
126 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
127 | if (!gc_th) | |
128 | return; | |
129 | kthread_stop(gc_th->f2fs_gc_task); | |
130 | kfree(gc_th); | |
131 | sbi->gc_thread = NULL; | |
132 | } | |
133 | ||
d2dc095f | 134 | static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type) |
7bc09003 | 135 | { |
d2dc095f NJ |
136 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
137 | ||
138 | if (gc_th && gc_th->gc_idle) { | |
139 | if (gc_th->gc_idle == 1) | |
140 | gc_mode = GC_CB; | |
141 | else if (gc_th->gc_idle == 2) | |
142 | gc_mode = GC_GREEDY; | |
143 | } | |
144 | return gc_mode; | |
7bc09003 JK |
145 | } |
146 | ||
147 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
148 | int type, struct victim_sel_policy *p) | |
149 | { | |
150 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
151 | ||
4ebefc44 | 152 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
153 | p->gc_mode = GC_GREEDY; |
154 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
a26b7c8a | 155 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 JK |
156 | p->ofs_unit = 1; |
157 | } else { | |
d2dc095f | 158 | p->gc_mode = select_gc_type(sbi->gc_thread, gc_type); |
7bc09003 | 159 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
a26b7c8a | 160 | p->max_search = dirty_i->nr_dirty[DIRTY]; |
7bc09003 JK |
161 | p->ofs_unit = sbi->segs_per_sec; |
162 | } | |
a26b7c8a | 163 | |
b1c57c1c JK |
164 | if (p->max_search > sbi->max_victim_search) |
165 | p->max_search = sbi->max_victim_search; | |
a26b7c8a | 166 | |
7bc09003 JK |
167 | p->offset = sbi->last_victim[p->gc_mode]; |
168 | } | |
169 | ||
170 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
171 | struct victim_sel_policy *p) | |
172 | { | |
b7250d2d JK |
173 | /* SSR allocates in a segment unit */ |
174 | if (p->alloc_mode == SSR) | |
3519e3f9 | 175 | return sbi->blocks_per_seg; |
7bc09003 | 176 | if (p->gc_mode == GC_GREEDY) |
3519e3f9 | 177 | return sbi->blocks_per_seg * p->ofs_unit; |
7bc09003 JK |
178 | else if (p->gc_mode == GC_CB) |
179 | return UINT_MAX; | |
180 | else /* No other gc_mode */ | |
181 | return 0; | |
182 | } | |
183 | ||
184 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
185 | { | |
186 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 187 | unsigned int secno; |
7bc09003 JK |
188 | |
189 | /* | |
190 | * If the gc_type is FG_GC, we can select victim segments | |
191 | * selected by background GC before. | |
192 | * Those segments guarantee they have small valid blocks. | |
193 | */ | |
7cd8558b | 194 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 195 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 196 | continue; |
5ec4e49f JK |
197 | clear_bit(secno, dirty_i->victim_secmap); |
198 | return secno * sbi->segs_per_sec; | |
7bc09003 JK |
199 | } |
200 | return NULL_SEGNO; | |
201 | } | |
202 | ||
203 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
204 | { | |
205 | struct sit_info *sit_i = SIT_I(sbi); | |
206 | unsigned int secno = GET_SECNO(sbi, segno); | |
207 | unsigned int start = secno * sbi->segs_per_sec; | |
208 | unsigned long long mtime = 0; | |
209 | unsigned int vblocks; | |
210 | unsigned char age = 0; | |
211 | unsigned char u; | |
212 | unsigned int i; | |
213 | ||
214 | for (i = 0; i < sbi->segs_per_sec; i++) | |
215 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
216 | vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
217 | ||
218 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
219 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
220 | ||
221 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
222 | ||
e1c42045 | 223 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
224 | if (mtime < sit_i->min_mtime) |
225 | sit_i->min_mtime = mtime; | |
226 | if (mtime > sit_i->max_mtime) | |
227 | sit_i->max_mtime = mtime; | |
228 | if (sit_i->max_mtime != sit_i->min_mtime) | |
229 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
230 | sit_i->max_mtime - sit_i->min_mtime); | |
231 | ||
232 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
233 | } | |
234 | ||
a57e564d JX |
235 | static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, |
236 | unsigned int segno, struct victim_sel_policy *p) | |
7bc09003 JK |
237 | { |
238 | if (p->alloc_mode == SSR) | |
239 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; | |
240 | ||
241 | /* alloc_mode == LFS */ | |
242 | if (p->gc_mode == GC_GREEDY) | |
243 | return get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
244 | else | |
245 | return get_cb_cost(sbi, segno); | |
246 | } | |
247 | ||
0a8165d7 | 248 | /* |
111d2495 | 249 | * This function is called from two paths. |
7bc09003 JK |
250 | * One is garbage collection and the other is SSR segment selection. |
251 | * When it is called during GC, it just gets a victim segment | |
252 | * and it does not remove it from dirty seglist. | |
253 | * When it is called from SSR segment selection, it finds a segment | |
254 | * which has minimum valid blocks and removes it from dirty seglist. | |
255 | */ | |
256 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
257 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
258 | { | |
259 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
260 | struct victim_sel_policy p; | |
b2b3460a | 261 | unsigned int secno, max_cost; |
a43f7ec3 | 262 | unsigned int last_segment = MAIN_SEGS(sbi); |
7bc09003 JK |
263 | int nsearched = 0; |
264 | ||
210f41bc CY |
265 | mutex_lock(&dirty_i->seglist_lock); |
266 | ||
7bc09003 JK |
267 | p.alloc_mode = alloc_mode; |
268 | select_policy(sbi, gc_type, type, &p); | |
269 | ||
270 | p.min_segno = NULL_SEGNO; | |
b2b3460a | 271 | p.min_cost = max_cost = get_max_cost(sbi, &p); |
7bc09003 | 272 | |
3342bb30 CY |
273 | if (p.max_search == 0) |
274 | goto out; | |
275 | ||
7bc09003 JK |
276 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
277 | p.min_segno = check_bg_victims(sbi); | |
278 | if (p.min_segno != NULL_SEGNO) | |
279 | goto got_it; | |
280 | } | |
281 | ||
282 | while (1) { | |
283 | unsigned long cost; | |
5ec4e49f | 284 | unsigned int segno; |
7bc09003 | 285 | |
a43f7ec3 CY |
286 | segno = find_next_bit(p.dirty_segmap, last_segment, p.offset); |
287 | if (segno >= last_segment) { | |
7bc09003 | 288 | if (sbi->last_victim[p.gc_mode]) { |
a43f7ec3 | 289 | last_segment = sbi->last_victim[p.gc_mode]; |
7bc09003 JK |
290 | sbi->last_victim[p.gc_mode] = 0; |
291 | p.offset = 0; | |
292 | continue; | |
293 | } | |
294 | break; | |
295 | } | |
a57e564d JX |
296 | |
297 | p.offset = segno + p.ofs_unit; | |
298 | if (p.ofs_unit > 1) | |
299 | p.offset -= segno % p.ofs_unit; | |
300 | ||
5ec4e49f | 301 | secno = GET_SECNO(sbi, segno); |
7bc09003 | 302 | |
5ec4e49f | 303 | if (sec_usage_check(sbi, secno)) |
7bc09003 | 304 | continue; |
5ec4e49f | 305 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
7bc09003 JK |
306 | continue; |
307 | ||
308 | cost = get_gc_cost(sbi, segno, &p); | |
309 | ||
310 | if (p.min_cost > cost) { | |
311 | p.min_segno = segno; | |
312 | p.min_cost = cost; | |
a57e564d | 313 | } else if (unlikely(cost == max_cost)) { |
7bc09003 | 314 | continue; |
a57e564d | 315 | } |
7bc09003 | 316 | |
a26b7c8a | 317 | if (nsearched++ >= p.max_search) { |
7bc09003 JK |
318 | sbi->last_victim[p.gc_mode] = segno; |
319 | break; | |
320 | } | |
321 | } | |
7bc09003 | 322 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 323 | got_it: |
7bc09003 | 324 | if (p.alloc_mode == LFS) { |
5ec4e49f JK |
325 | secno = GET_SECNO(sbi, p.min_segno); |
326 | if (gc_type == FG_GC) | |
327 | sbi->cur_victim_sec = secno; | |
328 | else | |
329 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 330 | } |
5ec4e49f | 331 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
332 | |
333 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
334 | sbi->cur_victim_sec, | |
335 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 | 336 | } |
3342bb30 | 337 | out: |
7bc09003 JK |
338 | mutex_unlock(&dirty_i->seglist_lock); |
339 | ||
340 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
341 | } | |
342 | ||
343 | static const struct victim_selection default_v_ops = { | |
344 | .get_victim = get_victim_by_default, | |
345 | }; | |
346 | ||
7dda2af8 | 347 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 348 | { |
7bc09003 JK |
349 | struct inode_entry *ie; |
350 | ||
7dda2af8 CL |
351 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
352 | if (ie) | |
353 | return ie->inode; | |
7bc09003 JK |
354 | return NULL; |
355 | } | |
356 | ||
7dda2af8 | 357 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 358 | { |
6cc4af56 GZ |
359 | struct inode_entry *new_ie; |
360 | ||
7dda2af8 | 361 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
362 | iput(inode); |
363 | return; | |
7bc09003 | 364 | } |
06292073 | 365 | new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); |
7bc09003 | 366 | new_ie->inode = inode; |
f28e5034 CY |
367 | |
368 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 369 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
370 | } |
371 | ||
7dda2af8 | 372 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
373 | { |
374 | struct inode_entry *ie, *next_ie; | |
7dda2af8 CL |
375 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
376 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
377 | iput(ie->inode); |
378 | list_del(&ie->list); | |
06292073 | 379 | kmem_cache_free(inode_entry_slab, ie); |
7bc09003 JK |
380 | } |
381 | } | |
382 | ||
383 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
384 | unsigned int segno, int offset) | |
385 | { | |
386 | struct sit_info *sit_i = SIT_I(sbi); | |
387 | struct seg_entry *sentry; | |
388 | int ret; | |
389 | ||
390 | mutex_lock(&sit_i->sentry_lock); | |
391 | sentry = get_seg_entry(sbi, segno); | |
392 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
393 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 394 | return ret; |
7bc09003 JK |
395 | } |
396 | ||
0a8165d7 | 397 | /* |
7bc09003 JK |
398 | * This function compares node address got in summary with that in NAT. |
399 | * On validity, copy that node with cold status, otherwise (invalid node) | |
400 | * ignore that. | |
401 | */ | |
5ee5293c | 402 | static int gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
403 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
404 | { | |
405 | bool initial = true; | |
406 | struct f2fs_summary *entry; | |
26d58599 | 407 | block_t start_addr; |
7bc09003 JK |
408 | int off; |
409 | ||
26d58599 JK |
410 | start_addr = START_BLOCK(sbi, segno); |
411 | ||
7bc09003 JK |
412 | next_step: |
413 | entry = sum; | |
c718379b | 414 | |
7bc09003 JK |
415 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
416 | nid_t nid = le32_to_cpu(entry->nid); | |
417 | struct page *node_page; | |
26d58599 | 418 | struct node_info ni; |
7bc09003 | 419 | |
43727527 JK |
420 | /* stop BG_GC if there is not enough free sections. */ |
421 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
5ee5293c | 422 | return 0; |
7bc09003 | 423 | |
43727527 | 424 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
425 | continue; |
426 | ||
427 | if (initial) { | |
428 | ra_node_page(sbi, nid); | |
429 | continue; | |
430 | } | |
431 | node_page = get_node_page(sbi, nid); | |
432 | if (IS_ERR(node_page)) | |
433 | continue; | |
434 | ||
9a01b56b HY |
435 | /* block may become invalid during get_node_page */ |
436 | if (check_valid_map(sbi, segno, off) == 0) { | |
437 | f2fs_put_page(node_page, 1); | |
438 | continue; | |
26d58599 JK |
439 | } |
440 | ||
441 | get_node_info(sbi, nid, &ni); | |
442 | if (ni.blk_addr != start_addr + off) { | |
443 | f2fs_put_page(node_page, 1); | |
444 | continue; | |
9a01b56b HY |
445 | } |
446 | ||
7bc09003 | 447 | /* set page dirty and write it */ |
4ebefc44 | 448 | if (gc_type == FG_GC) { |
5514f0aa | 449 | f2fs_wait_on_page_writeback(node_page, NODE); |
7bc09003 | 450 | set_page_dirty(node_page); |
4ebefc44 JK |
451 | } else { |
452 | if (!PageWriteback(node_page)) | |
453 | set_page_dirty(node_page); | |
454 | } | |
7bc09003 | 455 | f2fs_put_page(node_page, 1); |
e1235983 | 456 | stat_inc_node_blk_count(sbi, 1, gc_type); |
7bc09003 | 457 | } |
c718379b | 458 | |
7bc09003 JK |
459 | if (initial) { |
460 | initial = false; | |
461 | goto next_step; | |
462 | } | |
463 | ||
464 | if (gc_type == FG_GC) { | |
465 | struct writeback_control wbc = { | |
466 | .sync_mode = WB_SYNC_ALL, | |
467 | .nr_to_write = LONG_MAX, | |
468 | .for_reclaim = 0, | |
469 | }; | |
470 | sync_node_pages(sbi, 0, &wbc); | |
4ebefc44 | 471 | |
5ee5293c JK |
472 | /* return 1 only if FG_GC succefully reclaimed one */ |
473 | if (get_valid_blocks(sbi, segno, 1) == 0) | |
474 | return 1; | |
7bc09003 | 475 | } |
5ee5293c | 476 | return 0; |
7bc09003 JK |
477 | } |
478 | ||
0a8165d7 | 479 | /* |
9af45ef5 JK |
480 | * Calculate start block index indicating the given node offset. |
481 | * Be careful, caller should give this node offset only indicating direct node | |
482 | * blocks. If any node offsets, which point the other types of node blocks such | |
483 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
484 | * bug. | |
7bc09003 | 485 | */ |
de93653f | 486 | block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi) |
7bc09003 | 487 | { |
ce19a5d4 JK |
488 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
489 | unsigned int bidx; | |
7bc09003 | 490 | |
ce19a5d4 JK |
491 | if (node_ofs == 0) |
492 | return 0; | |
7bc09003 | 493 | |
ce19a5d4 | 494 | if (node_ofs <= 2) { |
7bc09003 JK |
495 | bidx = node_ofs - 1; |
496 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 497 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
498 | bidx = node_ofs - 2 - dec; |
499 | } else { | |
ce19a5d4 | 500 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
501 | bidx = node_ofs - 5 - dec; |
502 | } | |
de93653f | 503 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi); |
7bc09003 JK |
504 | } |
505 | ||
c1079892 | 506 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
507 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
508 | { | |
509 | struct page *node_page; | |
510 | nid_t nid; | |
511 | unsigned int ofs_in_node; | |
512 | block_t source_blkaddr; | |
513 | ||
514 | nid = le32_to_cpu(sum->nid); | |
515 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
516 | ||
517 | node_page = get_node_page(sbi, nid); | |
518 | if (IS_ERR(node_page)) | |
c1079892 | 519 | return false; |
7bc09003 JK |
520 | |
521 | get_node_info(sbi, nid, dni); | |
522 | ||
523 | if (sum->version != dni->version) { | |
524 | f2fs_put_page(node_page, 1); | |
c1079892 | 525 | return false; |
7bc09003 JK |
526 | } |
527 | ||
528 | *nofs = ofs_of_node(node_page); | |
529 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
530 | f2fs_put_page(node_page, 1); | |
531 | ||
532 | if (source_blkaddr != blkaddr) | |
c1079892 NK |
533 | return false; |
534 | return true; | |
7bc09003 JK |
535 | } |
536 | ||
4375a336 JK |
537 | static void move_encrypted_block(struct inode *inode, block_t bidx) |
538 | { | |
539 | struct f2fs_io_info fio = { | |
540 | .sbi = F2FS_I_SB(inode), | |
541 | .type = DATA, | |
542 | .rw = READ_SYNC, | |
543 | .encrypted_page = NULL, | |
544 | }; | |
545 | struct dnode_of_data dn; | |
546 | struct f2fs_summary sum; | |
547 | struct node_info ni; | |
548 | struct page *page; | |
549 | int err; | |
550 | ||
551 | /* do not read out */ | |
a56c7c6f | 552 | page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); |
4375a336 JK |
553 | if (!page) |
554 | return; | |
555 | ||
556 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
557 | err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE); | |
558 | if (err) | |
559 | goto out; | |
560 | ||
08b39fbd CY |
561 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
562 | ClearPageUptodate(page); | |
4375a336 | 563 | goto put_out; |
08b39fbd CY |
564 | } |
565 | ||
566 | /* | |
567 | * don't cache encrypted data into meta inode until previous dirty | |
568 | * data were writebacked to avoid racing between GC and flush. | |
569 | */ | |
570 | f2fs_wait_on_page_writeback(page, DATA); | |
4375a336 JK |
571 | |
572 | get_node_info(fio.sbi, dn.nid, &ni); | |
573 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); | |
574 | ||
575 | /* read page */ | |
576 | fio.page = page; | |
577 | fio.blk_addr = dn.data_blkaddr; | |
578 | ||
1b77c416 JK |
579 | fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), |
580 | fio.blk_addr, | |
581 | FGP_LOCK|FGP_CREAT, | |
582 | GFP_NOFS); | |
4375a336 JK |
583 | if (!fio.encrypted_page) |
584 | goto put_out; | |
585 | ||
548aedac JK |
586 | err = f2fs_submit_page_bio(&fio); |
587 | if (err) | |
588 | goto put_page_out; | |
589 | ||
590 | /* write page */ | |
591 | lock_page(fio.encrypted_page); | |
592 | ||
593 | if (unlikely(!PageUptodate(fio.encrypted_page))) | |
594 | goto put_page_out; | |
595 | if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) | |
596 | goto put_page_out; | |
597 | ||
6282adbf | 598 | set_page_dirty(fio.encrypted_page); |
08b39fbd | 599 | f2fs_wait_on_page_writeback(fio.encrypted_page, DATA); |
6282adbf JK |
600 | if (clear_page_dirty_for_io(fio.encrypted_page)) |
601 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
602 | ||
548aedac | 603 | set_page_writeback(fio.encrypted_page); |
4375a336 JK |
604 | |
605 | /* allocate block address */ | |
606 | f2fs_wait_on_page_writeback(dn.node_page, NODE); | |
4375a336 JK |
607 | allocate_data_block(fio.sbi, NULL, fio.blk_addr, |
608 | &fio.blk_addr, &sum, CURSEG_COLD_DATA); | |
4375a336 JK |
609 | fio.rw = WRITE_SYNC; |
610 | f2fs_submit_page_mbio(&fio); | |
611 | ||
548aedac | 612 | dn.data_blkaddr = fio.blk_addr; |
4375a336 JK |
613 | set_data_blkaddr(&dn); |
614 | f2fs_update_extent_cache(&dn); | |
615 | set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE); | |
616 | if (page->index == 0) | |
617 | set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN); | |
548aedac | 618 | put_page_out: |
4375a336 JK |
619 | f2fs_put_page(fio.encrypted_page, 1); |
620 | put_out: | |
621 | f2fs_put_dnode(&dn); | |
622 | out: | |
623 | f2fs_put_page(page, 1); | |
624 | } | |
625 | ||
c879f90d | 626 | static void move_data_page(struct inode *inode, block_t bidx, int gc_type) |
7bc09003 | 627 | { |
c879f90d JK |
628 | struct page *page; |
629 | ||
a56c7c6f | 630 | page = get_lock_data_page(inode, bidx, true); |
c879f90d JK |
631 | if (IS_ERR(page)) |
632 | return; | |
63a0b7cb | 633 | |
7bc09003 | 634 | if (gc_type == BG_GC) { |
4ebefc44 JK |
635 | if (PageWriteback(page)) |
636 | goto out; | |
7bc09003 JK |
637 | set_page_dirty(page); |
638 | set_cold_data(page); | |
639 | } else { | |
c879f90d JK |
640 | struct f2fs_io_info fio = { |
641 | .sbi = F2FS_I_SB(inode), | |
642 | .type = DATA, | |
643 | .rw = WRITE_SYNC, | |
644 | .page = page, | |
4375a336 | 645 | .encrypted_page = NULL, |
c879f90d | 646 | }; |
6282adbf | 647 | set_page_dirty(page); |
5514f0aa | 648 | f2fs_wait_on_page_writeback(page, DATA); |
1fe54f9d | 649 | if (clear_page_dirty_for_io(page)) |
a7ffdbe2 | 650 | inode_dec_dirty_pages(inode); |
7bc09003 | 651 | set_cold_data(page); |
05ca3632 | 652 | do_write_data_page(&fio); |
7bc09003 JK |
653 | clear_cold_data(page); |
654 | } | |
655 | out: | |
656 | f2fs_put_page(page, 1); | |
657 | } | |
658 | ||
0a8165d7 | 659 | /* |
7bc09003 JK |
660 | * This function tries to get parent node of victim data block, and identifies |
661 | * data block validity. If the block is valid, copy that with cold status and | |
662 | * modify parent node. | |
663 | * If the parent node is not valid or the data block address is different, | |
664 | * the victim data block is ignored. | |
665 | */ | |
5ee5293c | 666 | static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dda2af8 | 667 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type) |
7bc09003 JK |
668 | { |
669 | struct super_block *sb = sbi->sb; | |
670 | struct f2fs_summary *entry; | |
671 | block_t start_addr; | |
43727527 | 672 | int off; |
7bc09003 JK |
673 | int phase = 0; |
674 | ||
675 | start_addr = START_BLOCK(sbi, segno); | |
676 | ||
677 | next_step: | |
678 | entry = sum; | |
c718379b | 679 | |
7bc09003 JK |
680 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
681 | struct page *data_page; | |
682 | struct inode *inode; | |
683 | struct node_info dni; /* dnode info for the data */ | |
684 | unsigned int ofs_in_node, nofs; | |
685 | block_t start_bidx; | |
686 | ||
43727527 JK |
687 | /* stop BG_GC if there is not enough free sections. */ |
688 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
5ee5293c | 689 | return 0; |
7bc09003 | 690 | |
43727527 | 691 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
692 | continue; |
693 | ||
694 | if (phase == 0) { | |
695 | ra_node_page(sbi, le32_to_cpu(entry->nid)); | |
696 | continue; | |
697 | } | |
698 | ||
699 | /* Get an inode by ino with checking validity */ | |
c1079892 | 700 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
701 | continue; |
702 | ||
703 | if (phase == 1) { | |
704 | ra_node_page(sbi, dni.ino); | |
705 | continue; | |
706 | } | |
707 | ||
7bc09003 JK |
708 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
709 | ||
710 | if (phase == 2) { | |
d4686d56 | 711 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 712 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
713 | continue; |
714 | ||
4375a336 JK |
715 | /* if encrypted inode, let's go phase 3 */ |
716 | if (f2fs_encrypted_inode(inode) && | |
717 | S_ISREG(inode->i_mode)) { | |
718 | add_gc_inode(gc_list, inode); | |
719 | continue; | |
720 | } | |
721 | ||
de93653f | 722 | start_bidx = start_bidx_of_node(nofs, F2FS_I(inode)); |
43f3eae1 | 723 | data_page = get_read_data_page(inode, |
a56c7c6f | 724 | start_bidx + ofs_in_node, READA, true); |
31a32688 CL |
725 | if (IS_ERR(data_page)) { |
726 | iput(inode); | |
727 | continue; | |
728 | } | |
7bc09003 JK |
729 | |
730 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 731 | add_gc_inode(gc_list, inode); |
31a32688 CL |
732 | continue; |
733 | } | |
734 | ||
735 | /* phase 3 */ | |
7dda2af8 | 736 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 737 | if (inode) { |
c879f90d JK |
738 | start_bidx = start_bidx_of_node(nofs, F2FS_I(inode)) |
739 | + ofs_in_node; | |
4375a336 JK |
740 | if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) |
741 | move_encrypted_block(inode, start_bidx); | |
742 | else | |
743 | move_data_page(inode, start_bidx, gc_type); | |
e1235983 | 744 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 745 | } |
7bc09003 | 746 | } |
c718379b | 747 | |
7bc09003 JK |
748 | if (++phase < 4) |
749 | goto next_step; | |
43727527 | 750 | |
4ebefc44 | 751 | if (gc_type == FG_GC) { |
458e6197 | 752 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
4ebefc44 | 753 | |
5ee5293c JK |
754 | /* return 1 only if FG_GC succefully reclaimed one */ |
755 | if (get_valid_blocks(sbi, segno, 1) == 0) | |
756 | return 1; | |
4ebefc44 | 757 | } |
5ee5293c | 758 | return 0; |
7bc09003 JK |
759 | } |
760 | ||
761 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 762 | int gc_type) |
7bc09003 JK |
763 | { |
764 | struct sit_info *sit_i = SIT_I(sbi); | |
765 | int ret; | |
8a2d0ace | 766 | |
7bc09003 | 767 | mutex_lock(&sit_i->sentry_lock); |
8a2d0ace GZ |
768 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
769 | NO_CHECK_TYPE, LFS); | |
7bc09003 JK |
770 | mutex_unlock(&sit_i->sentry_lock); |
771 | return ret; | |
772 | } | |
773 | ||
5ee5293c | 774 | static int do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno, |
7dda2af8 | 775 | struct gc_inode_list *gc_list, int gc_type) |
7bc09003 JK |
776 | { |
777 | struct page *sum_page; | |
778 | struct f2fs_summary_block *sum; | |
c718379b | 779 | struct blk_plug plug; |
5ee5293c | 780 | int nfree = 0; |
7bc09003 JK |
781 | |
782 | /* read segment summary of victim */ | |
783 | sum_page = get_sum_page(sbi, segno); | |
7bc09003 | 784 | |
c718379b JK |
785 | blk_start_plug(&plug); |
786 | ||
7bc09003 JK |
787 | sum = page_address(sum_page); |
788 | ||
9236cac5 JK |
789 | /* |
790 | * this is to avoid deadlock: | |
791 | * - lock_page(sum_page) - f2fs_replace_block | |
792 | * - check_valid_map() - mutex_lock(sentry_lock) | |
793 | * - mutex_lock(sentry_lock) - change_curseg() | |
794 | * - lock_page(sum_page) | |
795 | */ | |
796 | unlock_page(sum_page); | |
797 | ||
7bc09003 JK |
798 | switch (GET_SUM_TYPE((&sum->footer))) { |
799 | case SUM_TYPE_NODE: | |
5ee5293c | 800 | nfree = gc_node_segment(sbi, sum->entries, segno, gc_type); |
7bc09003 JK |
801 | break; |
802 | case SUM_TYPE_DATA: | |
5ee5293c JK |
803 | nfree = gc_data_segment(sbi, sum->entries, gc_list, |
804 | segno, gc_type); | |
7bc09003 JK |
805 | break; |
806 | } | |
c718379b JK |
807 | blk_finish_plug(&plug); |
808 | ||
e1235983 | 809 | stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)), gc_type); |
7bc09003 JK |
810 | stat_inc_call_count(sbi->stat_info); |
811 | ||
9236cac5 | 812 | f2fs_put_page(sum_page, 0); |
5ee5293c | 813 | return nfree; |
7bc09003 JK |
814 | } |
815 | ||
d530d4d8 | 816 | int f2fs_gc(struct f2fs_sb_info *sbi, bool sync) |
7bc09003 | 817 | { |
ab126cfc | 818 | unsigned int segno, i; |
d530d4d8 | 819 | int gc_type = sync ? FG_GC : BG_GC; |
45fe8492 | 820 | int sec_freed = 0; |
d530d4d8 | 821 | int ret = -EINVAL; |
d5053a34 | 822 | struct cp_control cpc; |
7dda2af8 CL |
823 | struct gc_inode_list gc_list = { |
824 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
769ec6e5 | 825 | .iroot = RADIX_TREE_INIT(GFP_NOFS), |
7dda2af8 | 826 | }; |
d5053a34 | 827 | |
119ee914 | 828 | cpc.reason = __get_cp_reason(sbi); |
7bc09003 | 829 | gc_more: |
ab126cfc JK |
830 | segno = NULL_SEGNO; |
831 | ||
6bacf52f | 832 | if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) |
408e9375 | 833 | goto stop; |
6d5a1495 CY |
834 | if (unlikely(f2fs_cp_error(sbi))) { |
835 | ret = -EIO; | |
203681f6 | 836 | goto stop; |
6d5a1495 | 837 | } |
7bc09003 | 838 | |
45fe8492 | 839 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, sec_freed)) { |
408e9375 | 840 | gc_type = FG_GC; |
798c1b16 JK |
841 | if (__get_victim(sbi, &segno, gc_type) || prefree_segments(sbi)) |
842 | write_checkpoint(sbi, &cpc); | |
d64f8047 | 843 | } |
7bc09003 | 844 | |
798c1b16 | 845 | if (segno == NULL_SEGNO && !__get_victim(sbi, &segno, gc_type)) |
408e9375 | 846 | goto stop; |
43727527 | 847 | ret = 0; |
7bc09003 | 848 | |
81c1a0f1 CY |
849 | /* readahead multi ssa blocks those have contiguous address */ |
850 | if (sbi->segs_per_sec > 1) | |
851 | ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), sbi->segs_per_sec, | |
26879fb1 | 852 | META_SSA, true); |
81c1a0f1 | 853 | |
45fe8492 CY |
854 | for (i = 0; i < sbi->segs_per_sec; i++) { |
855 | /* | |
856 | * for FG_GC case, halt gcing left segments once failed one | |
857 | * of segments in selected section to avoid long latency. | |
858 | */ | |
859 | if (!do_garbage_collect(sbi, segno + i, &gc_list, gc_type) && | |
860 | gc_type == FG_GC) | |
861 | break; | |
862 | } | |
863 | ||
864 | if (i == sbi->segs_per_sec && gc_type == FG_GC) | |
865 | sec_freed++; | |
43727527 | 866 | |
5ee5293c | 867 | if (gc_type == FG_GC) |
5ec4e49f | 868 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 869 | |
d530d4d8 CY |
870 | if (!sync) { |
871 | if (has_not_enough_free_secs(sbi, sec_freed)) | |
872 | goto gc_more; | |
43727527 | 873 | |
d530d4d8 CY |
874 | if (gc_type == FG_GC) |
875 | write_checkpoint(sbi, &cpc); | |
876 | } | |
408e9375 | 877 | stop: |
7bc09003 JK |
878 | mutex_unlock(&sbi->gc_mutex); |
879 | ||
7dda2af8 | 880 | put_gc_inode(&gc_list); |
d530d4d8 CY |
881 | |
882 | if (sync) | |
883 | ret = sec_freed ? 0 : -EAGAIN; | |
43727527 | 884 | return ret; |
7bc09003 JK |
885 | } |
886 | ||
887 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
888 | { | |
889 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
890 | } |