]>
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> | |
19 | #include <linux/blkdev.h> | |
20 | ||
21 | #include "f2fs.h" | |
22 | #include "node.h" | |
23 | #include "segment.h" | |
24 | #include "gc.h" | |
8e46b3ed | 25 | #include <trace/events/f2fs.h> |
7bc09003 | 26 | |
7bc09003 JK |
27 | static int gc_thread_func(void *data) |
28 | { | |
29 | struct f2fs_sb_info *sbi = data; | |
b59d0bae | 30 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; |
7bc09003 JK |
31 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; |
32 | long wait_ms; | |
33 | ||
b59d0bae | 34 | wait_ms = gc_th->min_sleep_time; |
7bc09003 JK |
35 | |
36 | do { | |
37 | if (try_to_freeze()) | |
38 | continue; | |
39 | else | |
40 | wait_event_interruptible_timeout(*wq, | |
41 | kthread_should_stop(), | |
42 | msecs_to_jiffies(wait_ms)); | |
43 | if (kthread_should_stop()) | |
44 | break; | |
45 | ||
d6212a5f | 46 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
88dd8934 | 47 | increase_sleep_time(gc_th, &wait_ms); |
d6212a5f CL |
48 | continue; |
49 | } | |
50 | ||
7bc09003 JK |
51 | /* |
52 | * [GC triggering condition] | |
53 | * 0. GC is not conducted currently. | |
54 | * 1. There are enough dirty segments. | |
55 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
56 | * 3. IO subsystem is idle by checking the # of requests in | |
57 | * bdev's request list. | |
58 | * | |
e1c42045 | 59 | * Note) We have to avoid triggering GCs frequently. |
7bc09003 JK |
60 | * Because it is possible that some segments can be |
61 | * invalidated soon after by user update or deletion. | |
62 | * So, I'd like to wait some time to collect dirty segments. | |
63 | */ | |
64 | if (!mutex_trylock(&sbi->gc_mutex)) | |
65 | continue; | |
66 | ||
67 | if (!is_idle(sbi)) { | |
88dd8934 | 68 | increase_sleep_time(gc_th, &wait_ms); |
7bc09003 JK |
69 | mutex_unlock(&sbi->gc_mutex); |
70 | continue; | |
71 | } | |
72 | ||
73 | if (has_enough_invalid_blocks(sbi)) | |
88dd8934 | 74 | decrease_sleep_time(gc_th, &wait_ms); |
7bc09003 | 75 | else |
88dd8934 | 76 | increase_sleep_time(gc_th, &wait_ms); |
7bc09003 | 77 | |
dcdfff65 | 78 | stat_inc_bggc_count(sbi); |
7bc09003 | 79 | |
43727527 | 80 | /* if return value is not zero, no victim was selected */ |
6aefd93b | 81 | if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC))) |
b59d0bae | 82 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 83 | |
84e4214f JK |
84 | trace_f2fs_background_gc(sbi->sb, wait_ms, |
85 | prefree_segments(sbi), free_segments(sbi)); | |
86 | ||
4660f9c0 JK |
87 | /* balancing f2fs's metadata periodically */ |
88 | f2fs_balance_fs_bg(sbi); | |
81eb8d6e | 89 | |
7bc09003 JK |
90 | } while (!kthread_should_stop()); |
91 | return 0; | |
92 | } | |
93 | ||
94 | int start_gc_thread(struct f2fs_sb_info *sbi) | |
95 | { | |
1042d60f | 96 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 97 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 98 | int err = 0; |
7bc09003 JK |
99 | |
100 | gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL); | |
7a267f8d NJ |
101 | if (!gc_th) { |
102 | err = -ENOMEM; | |
103 | goto out; | |
104 | } | |
7bc09003 | 105 | |
b59d0bae NJ |
106 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
107 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
108 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
109 | ||
d2dc095f NJ |
110 | gc_th->gc_idle = 0; |
111 | ||
7bc09003 JK |
112 | sbi->gc_thread = gc_th; |
113 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
114 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 115 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 116 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 117 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 118 | kfree(gc_th); |
25718423 | 119 | sbi->gc_thread = NULL; |
7bc09003 | 120 | } |
7a267f8d NJ |
121 | out: |
122 | return err; | |
7bc09003 JK |
123 | } |
124 | ||
125 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
126 | { | |
127 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
128 | if (!gc_th) | |
129 | return; | |
130 | kthread_stop(gc_th->f2fs_gc_task); | |
131 | kfree(gc_th); | |
132 | sbi->gc_thread = NULL; | |
133 | } | |
134 | ||
d2dc095f | 135 | static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type) |
7bc09003 | 136 | { |
d2dc095f NJ |
137 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
138 | ||
139 | if (gc_th && gc_th->gc_idle) { | |
140 | if (gc_th->gc_idle == 1) | |
141 | gc_mode = GC_CB; | |
142 | else if (gc_th->gc_idle == 2) | |
143 | gc_mode = GC_GREEDY; | |
144 | } | |
145 | return gc_mode; | |
7bc09003 JK |
146 | } |
147 | ||
148 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
149 | int type, struct victim_sel_policy *p) | |
150 | { | |
151 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
152 | ||
4ebefc44 | 153 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
154 | p->gc_mode = GC_GREEDY; |
155 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
a26b7c8a | 156 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 JK |
157 | p->ofs_unit = 1; |
158 | } else { | |
d2dc095f | 159 | p->gc_mode = select_gc_type(sbi->gc_thread, gc_type); |
7bc09003 | 160 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
a26b7c8a | 161 | p->max_search = dirty_i->nr_dirty[DIRTY]; |
7bc09003 JK |
162 | p->ofs_unit = sbi->segs_per_sec; |
163 | } | |
a26b7c8a | 164 | |
b1c57c1c JK |
165 | if (p->max_search > sbi->max_victim_search) |
166 | p->max_search = sbi->max_victim_search; | |
a26b7c8a | 167 | |
7bc09003 JK |
168 | p->offset = sbi->last_victim[p->gc_mode]; |
169 | } | |
170 | ||
171 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
172 | struct victim_sel_policy *p) | |
173 | { | |
b7250d2d JK |
174 | /* SSR allocates in a segment unit */ |
175 | if (p->alloc_mode == SSR) | |
176 | return 1 << sbi->log_blocks_per_seg; | |
7bc09003 JK |
177 | if (p->gc_mode == GC_GREEDY) |
178 | return (1 << sbi->log_blocks_per_seg) * p->ofs_unit; | |
179 | else if (p->gc_mode == GC_CB) | |
180 | return UINT_MAX; | |
181 | else /* No other gc_mode */ | |
182 | return 0; | |
183 | } | |
184 | ||
185 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
186 | { | |
187 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 188 | unsigned int secno; |
7bc09003 JK |
189 | |
190 | /* | |
191 | * If the gc_type is FG_GC, we can select victim segments | |
192 | * selected by background GC before. | |
193 | * Those segments guarantee they have small valid blocks. | |
194 | */ | |
7cd8558b | 195 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 196 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 197 | continue; |
5ec4e49f JK |
198 | clear_bit(secno, dirty_i->victim_secmap); |
199 | return secno * sbi->segs_per_sec; | |
7bc09003 JK |
200 | } |
201 | return NULL_SEGNO; | |
202 | } | |
203 | ||
204 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
205 | { | |
206 | struct sit_info *sit_i = SIT_I(sbi); | |
207 | unsigned int secno = GET_SECNO(sbi, segno); | |
208 | unsigned int start = secno * sbi->segs_per_sec; | |
209 | unsigned long long mtime = 0; | |
210 | unsigned int vblocks; | |
211 | unsigned char age = 0; | |
212 | unsigned char u; | |
213 | unsigned int i; | |
214 | ||
215 | for (i = 0; i < sbi->segs_per_sec; i++) | |
216 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
217 | vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
218 | ||
219 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
220 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
221 | ||
222 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
223 | ||
e1c42045 | 224 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
225 | if (mtime < sit_i->min_mtime) |
226 | sit_i->min_mtime = mtime; | |
227 | if (mtime > sit_i->max_mtime) | |
228 | sit_i->max_mtime = mtime; | |
229 | if (sit_i->max_mtime != sit_i->min_mtime) | |
230 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
231 | sit_i->max_mtime - sit_i->min_mtime); | |
232 | ||
233 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
234 | } | |
235 | ||
a57e564d JX |
236 | static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, |
237 | unsigned int segno, struct victim_sel_policy *p) | |
7bc09003 JK |
238 | { |
239 | if (p->alloc_mode == SSR) | |
240 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; | |
241 | ||
242 | /* alloc_mode == LFS */ | |
243 | if (p->gc_mode == GC_GREEDY) | |
244 | return get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
245 | else | |
246 | return get_cb_cost(sbi, segno); | |
247 | } | |
248 | ||
0a8165d7 | 249 | /* |
111d2495 | 250 | * This function is called from two paths. |
7bc09003 JK |
251 | * One is garbage collection and the other is SSR segment selection. |
252 | * When it is called during GC, it just gets a victim segment | |
253 | * and it does not remove it from dirty seglist. | |
254 | * When it is called from SSR segment selection, it finds a segment | |
255 | * which has minimum valid blocks and removes it from dirty seglist. | |
256 | */ | |
257 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
258 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
259 | { | |
260 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
261 | struct victim_sel_policy p; | |
b2b3460a | 262 | unsigned int secno, max_cost; |
a43f7ec3 | 263 | unsigned int last_segment = MAIN_SEGS(sbi); |
7bc09003 JK |
264 | int nsearched = 0; |
265 | ||
210f41bc CY |
266 | mutex_lock(&dirty_i->seglist_lock); |
267 | ||
7bc09003 JK |
268 | p.alloc_mode = alloc_mode; |
269 | select_policy(sbi, gc_type, type, &p); | |
270 | ||
271 | p.min_segno = NULL_SEGNO; | |
b2b3460a | 272 | p.min_cost = max_cost = get_max_cost(sbi, &p); |
7bc09003 | 273 | |
3342bb30 CY |
274 | if (p.max_search == 0) |
275 | goto out; | |
276 | ||
7bc09003 JK |
277 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
278 | p.min_segno = check_bg_victims(sbi); | |
279 | if (p.min_segno != NULL_SEGNO) | |
280 | goto got_it; | |
281 | } | |
282 | ||
283 | while (1) { | |
284 | unsigned long cost; | |
5ec4e49f | 285 | unsigned int segno; |
7bc09003 | 286 | |
a43f7ec3 CY |
287 | segno = find_next_bit(p.dirty_segmap, last_segment, p.offset); |
288 | if (segno >= last_segment) { | |
7bc09003 | 289 | if (sbi->last_victim[p.gc_mode]) { |
a43f7ec3 | 290 | last_segment = sbi->last_victim[p.gc_mode]; |
7bc09003 JK |
291 | sbi->last_victim[p.gc_mode] = 0; |
292 | p.offset = 0; | |
293 | continue; | |
294 | } | |
295 | break; | |
296 | } | |
a57e564d JX |
297 | |
298 | p.offset = segno + p.ofs_unit; | |
299 | if (p.ofs_unit > 1) | |
300 | p.offset -= segno % p.ofs_unit; | |
301 | ||
5ec4e49f | 302 | secno = GET_SECNO(sbi, segno); |
7bc09003 | 303 | |
5ec4e49f | 304 | if (sec_usage_check(sbi, secno)) |
7bc09003 | 305 | continue; |
5ec4e49f | 306 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
7bc09003 JK |
307 | continue; |
308 | ||
309 | cost = get_gc_cost(sbi, segno, &p); | |
310 | ||
311 | if (p.min_cost > cost) { | |
312 | p.min_segno = segno; | |
313 | p.min_cost = cost; | |
a57e564d | 314 | } else if (unlikely(cost == max_cost)) { |
7bc09003 | 315 | continue; |
a57e564d | 316 | } |
7bc09003 | 317 | |
a26b7c8a | 318 | if (nsearched++ >= p.max_search) { |
7bc09003 JK |
319 | sbi->last_victim[p.gc_mode] = segno; |
320 | break; | |
321 | } | |
322 | } | |
7bc09003 | 323 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 324 | got_it: |
7bc09003 | 325 | if (p.alloc_mode == LFS) { |
5ec4e49f JK |
326 | secno = GET_SECNO(sbi, p.min_segno); |
327 | if (gc_type == FG_GC) | |
328 | sbi->cur_victim_sec = secno; | |
329 | else | |
330 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 331 | } |
5ec4e49f | 332 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
333 | |
334 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
335 | sbi->cur_victim_sec, | |
336 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 | 337 | } |
3342bb30 | 338 | out: |
7bc09003 JK |
339 | mutex_unlock(&dirty_i->seglist_lock); |
340 | ||
341 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
342 | } | |
343 | ||
344 | static const struct victim_selection default_v_ops = { | |
345 | .get_victim = get_victim_by_default, | |
346 | }; | |
347 | ||
7dda2af8 | 348 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 349 | { |
7bc09003 JK |
350 | struct inode_entry *ie; |
351 | ||
7dda2af8 CL |
352 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
353 | if (ie) | |
354 | return ie->inode; | |
7bc09003 JK |
355 | return NULL; |
356 | } | |
357 | ||
7dda2af8 | 358 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 359 | { |
6cc4af56 GZ |
360 | struct inode_entry *new_ie; |
361 | ||
7dda2af8 | 362 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
363 | iput(inode); |
364 | return; | |
7bc09003 | 365 | } |
06292073 | 366 | new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); |
7bc09003 | 367 | new_ie->inode = inode; |
f28e5034 CY |
368 | |
369 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 370 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
371 | } |
372 | ||
7dda2af8 | 373 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
374 | { |
375 | struct inode_entry *ie, *next_ie; | |
7dda2af8 CL |
376 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
377 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
378 | iput(ie->inode); |
379 | list_del(&ie->list); | |
06292073 | 380 | kmem_cache_free(inode_entry_slab, ie); |
7bc09003 JK |
381 | } |
382 | } | |
383 | ||
384 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
385 | unsigned int segno, int offset) | |
386 | { | |
387 | struct sit_info *sit_i = SIT_I(sbi); | |
388 | struct seg_entry *sentry; | |
389 | int ret; | |
390 | ||
391 | mutex_lock(&sit_i->sentry_lock); | |
392 | sentry = get_seg_entry(sbi, segno); | |
393 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
394 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 395 | return ret; |
7bc09003 JK |
396 | } |
397 | ||
0a8165d7 | 398 | /* |
7bc09003 JK |
399 | * This function compares node address got in summary with that in NAT. |
400 | * On validity, copy that node with cold status, otherwise (invalid node) | |
401 | * ignore that. | |
402 | */ | |
5ee5293c | 403 | static int gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
404 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
405 | { | |
406 | bool initial = true; | |
407 | struct f2fs_summary *entry; | |
26d58599 | 408 | block_t start_addr; |
7bc09003 JK |
409 | int off; |
410 | ||
26d58599 JK |
411 | start_addr = START_BLOCK(sbi, segno); |
412 | ||
7bc09003 JK |
413 | next_step: |
414 | entry = sum; | |
c718379b | 415 | |
7bc09003 JK |
416 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
417 | nid_t nid = le32_to_cpu(entry->nid); | |
418 | struct page *node_page; | |
26d58599 | 419 | struct node_info ni; |
7bc09003 | 420 | |
43727527 JK |
421 | /* stop BG_GC if there is not enough free sections. */ |
422 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
5ee5293c | 423 | return 0; |
7bc09003 | 424 | |
43727527 | 425 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
426 | continue; |
427 | ||
428 | if (initial) { | |
429 | ra_node_page(sbi, nid); | |
430 | continue; | |
431 | } | |
432 | node_page = get_node_page(sbi, nid); | |
433 | if (IS_ERR(node_page)) | |
434 | continue; | |
435 | ||
9a01b56b HY |
436 | /* block may become invalid during get_node_page */ |
437 | if (check_valid_map(sbi, segno, off) == 0) { | |
438 | f2fs_put_page(node_page, 1); | |
439 | continue; | |
26d58599 JK |
440 | } |
441 | ||
442 | get_node_info(sbi, nid, &ni); | |
443 | if (ni.blk_addr != start_addr + off) { | |
444 | f2fs_put_page(node_page, 1); | |
445 | continue; | |
9a01b56b HY |
446 | } |
447 | ||
7bc09003 | 448 | /* set page dirty and write it */ |
4ebefc44 | 449 | if (gc_type == FG_GC) { |
5514f0aa | 450 | f2fs_wait_on_page_writeback(node_page, NODE); |
7bc09003 | 451 | set_page_dirty(node_page); |
4ebefc44 JK |
452 | } else { |
453 | if (!PageWriteback(node_page)) | |
454 | set_page_dirty(node_page); | |
455 | } | |
7bc09003 | 456 | f2fs_put_page(node_page, 1); |
e1235983 | 457 | stat_inc_node_blk_count(sbi, 1, gc_type); |
7bc09003 | 458 | } |
c718379b | 459 | |
7bc09003 JK |
460 | if (initial) { |
461 | initial = false; | |
462 | goto next_step; | |
463 | } | |
464 | ||
465 | if (gc_type == FG_GC) { | |
466 | struct writeback_control wbc = { | |
467 | .sync_mode = WB_SYNC_ALL, | |
468 | .nr_to_write = LONG_MAX, | |
469 | .for_reclaim = 0, | |
470 | }; | |
471 | sync_node_pages(sbi, 0, &wbc); | |
4ebefc44 | 472 | |
5ee5293c JK |
473 | /* return 1 only if FG_GC succefully reclaimed one */ |
474 | if (get_valid_blocks(sbi, segno, 1) == 0) | |
475 | return 1; | |
7bc09003 | 476 | } |
5ee5293c | 477 | return 0; |
7bc09003 JK |
478 | } |
479 | ||
0a8165d7 | 480 | /* |
9af45ef5 JK |
481 | * Calculate start block index indicating the given node offset. |
482 | * Be careful, caller should give this node offset only indicating direct node | |
483 | * blocks. If any node offsets, which point the other types of node blocks such | |
484 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
485 | * bug. | |
7bc09003 | 486 | */ |
de93653f | 487 | block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi) |
7bc09003 | 488 | { |
ce19a5d4 JK |
489 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
490 | unsigned int bidx; | |
7bc09003 | 491 | |
ce19a5d4 JK |
492 | if (node_ofs == 0) |
493 | return 0; | |
7bc09003 | 494 | |
ce19a5d4 | 495 | if (node_ofs <= 2) { |
7bc09003 JK |
496 | bidx = node_ofs - 1; |
497 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 498 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
499 | bidx = node_ofs - 2 - dec; |
500 | } else { | |
ce19a5d4 | 501 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
502 | bidx = node_ofs - 5 - dec; |
503 | } | |
de93653f | 504 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi); |
7bc09003 JK |
505 | } |
506 | ||
c1079892 | 507 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
508 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
509 | { | |
510 | struct page *node_page; | |
511 | nid_t nid; | |
512 | unsigned int ofs_in_node; | |
513 | block_t source_blkaddr; | |
514 | ||
515 | nid = le32_to_cpu(sum->nid); | |
516 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
517 | ||
518 | node_page = get_node_page(sbi, nid); | |
519 | if (IS_ERR(node_page)) | |
c1079892 | 520 | return false; |
7bc09003 JK |
521 | |
522 | get_node_info(sbi, nid, dni); | |
523 | ||
524 | if (sum->version != dni->version) { | |
525 | f2fs_put_page(node_page, 1); | |
c1079892 | 526 | return false; |
7bc09003 JK |
527 | } |
528 | ||
529 | *nofs = ofs_of_node(node_page); | |
530 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
531 | f2fs_put_page(node_page, 1); | |
532 | ||
533 | if (source_blkaddr != blkaddr) | |
c1079892 NK |
534 | return false; |
535 | return true; | |
7bc09003 JK |
536 | } |
537 | ||
4375a336 JK |
538 | static void move_encrypted_block(struct inode *inode, block_t bidx) |
539 | { | |
540 | struct f2fs_io_info fio = { | |
541 | .sbi = F2FS_I_SB(inode), | |
542 | .type = DATA, | |
543 | .rw = READ_SYNC, | |
544 | .encrypted_page = NULL, | |
545 | }; | |
546 | struct dnode_of_data dn; | |
547 | struct f2fs_summary sum; | |
548 | struct node_info ni; | |
549 | struct page *page; | |
550 | int err; | |
551 | ||
552 | /* do not read out */ | |
a56c7c6f | 553 | page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); |
4375a336 JK |
554 | if (!page) |
555 | return; | |
556 | ||
557 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
558 | err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE); | |
559 | if (err) | |
560 | goto out; | |
561 | ||
08b39fbd CY |
562 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
563 | ClearPageUptodate(page); | |
4375a336 | 564 | goto put_out; |
08b39fbd CY |
565 | } |
566 | ||
567 | /* | |
568 | * don't cache encrypted data into meta inode until previous dirty | |
569 | * data were writebacked to avoid racing between GC and flush. | |
570 | */ | |
571 | f2fs_wait_on_page_writeback(page, DATA); | |
4375a336 JK |
572 | |
573 | get_node_info(fio.sbi, dn.nid, &ni); | |
574 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); | |
575 | ||
576 | /* read page */ | |
577 | fio.page = page; | |
578 | fio.blk_addr = dn.data_blkaddr; | |
579 | ||
1b77c416 JK |
580 | fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), |
581 | fio.blk_addr, | |
582 | FGP_LOCK|FGP_CREAT, | |
583 | GFP_NOFS); | |
4375a336 JK |
584 | if (!fio.encrypted_page) |
585 | goto put_out; | |
586 | ||
548aedac JK |
587 | err = f2fs_submit_page_bio(&fio); |
588 | if (err) | |
589 | goto put_page_out; | |
590 | ||
591 | /* write page */ | |
592 | lock_page(fio.encrypted_page); | |
593 | ||
594 | if (unlikely(!PageUptodate(fio.encrypted_page))) | |
595 | goto put_page_out; | |
596 | if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) | |
597 | goto put_page_out; | |
598 | ||
6282adbf | 599 | set_page_dirty(fio.encrypted_page); |
08b39fbd | 600 | f2fs_wait_on_page_writeback(fio.encrypted_page, DATA); |
6282adbf JK |
601 | if (clear_page_dirty_for_io(fio.encrypted_page)) |
602 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
603 | ||
548aedac | 604 | set_page_writeback(fio.encrypted_page); |
4375a336 JK |
605 | |
606 | /* allocate block address */ | |
607 | f2fs_wait_on_page_writeback(dn.node_page, NODE); | |
4375a336 JK |
608 | allocate_data_block(fio.sbi, NULL, fio.blk_addr, |
609 | &fio.blk_addr, &sum, CURSEG_COLD_DATA); | |
4375a336 JK |
610 | fio.rw = WRITE_SYNC; |
611 | f2fs_submit_page_mbio(&fio); | |
612 | ||
548aedac | 613 | dn.data_blkaddr = fio.blk_addr; |
4375a336 JK |
614 | set_data_blkaddr(&dn); |
615 | f2fs_update_extent_cache(&dn); | |
616 | set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE); | |
617 | if (page->index == 0) | |
618 | set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN); | |
548aedac | 619 | put_page_out: |
4375a336 JK |
620 | f2fs_put_page(fio.encrypted_page, 1); |
621 | put_out: | |
622 | f2fs_put_dnode(&dn); | |
623 | out: | |
624 | f2fs_put_page(page, 1); | |
625 | } | |
626 | ||
c879f90d | 627 | static void move_data_page(struct inode *inode, block_t bidx, int gc_type) |
7bc09003 | 628 | { |
c879f90d JK |
629 | struct page *page; |
630 | ||
a56c7c6f | 631 | page = get_lock_data_page(inode, bidx, true); |
c879f90d JK |
632 | if (IS_ERR(page)) |
633 | return; | |
63a0b7cb | 634 | |
7bc09003 | 635 | if (gc_type == BG_GC) { |
4ebefc44 JK |
636 | if (PageWriteback(page)) |
637 | goto out; | |
7bc09003 JK |
638 | set_page_dirty(page); |
639 | set_cold_data(page); | |
640 | } else { | |
c879f90d JK |
641 | struct f2fs_io_info fio = { |
642 | .sbi = F2FS_I_SB(inode), | |
643 | .type = DATA, | |
644 | .rw = WRITE_SYNC, | |
645 | .page = page, | |
4375a336 | 646 | .encrypted_page = NULL, |
c879f90d | 647 | }; |
6282adbf | 648 | set_page_dirty(page); |
5514f0aa | 649 | f2fs_wait_on_page_writeback(page, DATA); |
1fe54f9d | 650 | if (clear_page_dirty_for_io(page)) |
a7ffdbe2 | 651 | inode_dec_dirty_pages(inode); |
7bc09003 | 652 | set_cold_data(page); |
05ca3632 | 653 | do_write_data_page(&fio); |
7bc09003 JK |
654 | clear_cold_data(page); |
655 | } | |
656 | out: | |
657 | f2fs_put_page(page, 1); | |
658 | } | |
659 | ||
0a8165d7 | 660 | /* |
7bc09003 JK |
661 | * This function tries to get parent node of victim data block, and identifies |
662 | * data block validity. If the block is valid, copy that with cold status and | |
663 | * modify parent node. | |
664 | * If the parent node is not valid or the data block address is different, | |
665 | * the victim data block is ignored. | |
666 | */ | |
5ee5293c | 667 | static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dda2af8 | 668 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type) |
7bc09003 JK |
669 | { |
670 | struct super_block *sb = sbi->sb; | |
671 | struct f2fs_summary *entry; | |
672 | block_t start_addr; | |
43727527 | 673 | int off; |
7bc09003 JK |
674 | int phase = 0; |
675 | ||
676 | start_addr = START_BLOCK(sbi, segno); | |
677 | ||
678 | next_step: | |
679 | entry = sum; | |
c718379b | 680 | |
7bc09003 JK |
681 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
682 | struct page *data_page; | |
683 | struct inode *inode; | |
684 | struct node_info dni; /* dnode info for the data */ | |
685 | unsigned int ofs_in_node, nofs; | |
686 | block_t start_bidx; | |
687 | ||
43727527 JK |
688 | /* stop BG_GC if there is not enough free sections. */ |
689 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
5ee5293c | 690 | return 0; |
7bc09003 | 691 | |
43727527 | 692 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
693 | continue; |
694 | ||
695 | if (phase == 0) { | |
696 | ra_node_page(sbi, le32_to_cpu(entry->nid)); | |
697 | continue; | |
698 | } | |
699 | ||
700 | /* Get an inode by ino with checking validity */ | |
c1079892 | 701 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
702 | continue; |
703 | ||
704 | if (phase == 1) { | |
705 | ra_node_page(sbi, dni.ino); | |
706 | continue; | |
707 | } | |
708 | ||
7bc09003 JK |
709 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
710 | ||
711 | if (phase == 2) { | |
d4686d56 | 712 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 713 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
714 | continue; |
715 | ||
4375a336 JK |
716 | /* if encrypted inode, let's go phase 3 */ |
717 | if (f2fs_encrypted_inode(inode) && | |
718 | S_ISREG(inode->i_mode)) { | |
719 | add_gc_inode(gc_list, inode); | |
720 | continue; | |
721 | } | |
722 | ||
de93653f | 723 | start_bidx = start_bidx_of_node(nofs, F2FS_I(inode)); |
43f3eae1 | 724 | data_page = get_read_data_page(inode, |
a56c7c6f | 725 | start_bidx + ofs_in_node, READA, true); |
31a32688 CL |
726 | if (IS_ERR(data_page)) { |
727 | iput(inode); | |
728 | continue; | |
729 | } | |
7bc09003 JK |
730 | |
731 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 732 | add_gc_inode(gc_list, inode); |
31a32688 CL |
733 | continue; |
734 | } | |
735 | ||
736 | /* phase 3 */ | |
7dda2af8 | 737 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 738 | if (inode) { |
c879f90d JK |
739 | start_bidx = start_bidx_of_node(nofs, F2FS_I(inode)) |
740 | + ofs_in_node; | |
4375a336 JK |
741 | if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) |
742 | move_encrypted_block(inode, start_bidx); | |
743 | else | |
744 | move_data_page(inode, start_bidx, gc_type); | |
e1235983 | 745 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 746 | } |
7bc09003 | 747 | } |
c718379b | 748 | |
7bc09003 JK |
749 | if (++phase < 4) |
750 | goto next_step; | |
43727527 | 751 | |
4ebefc44 | 752 | if (gc_type == FG_GC) { |
458e6197 | 753 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
4ebefc44 | 754 | |
5ee5293c JK |
755 | /* return 1 only if FG_GC succefully reclaimed one */ |
756 | if (get_valid_blocks(sbi, segno, 1) == 0) | |
757 | return 1; | |
4ebefc44 | 758 | } |
5ee5293c | 759 | return 0; |
7bc09003 JK |
760 | } |
761 | ||
762 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 763 | int gc_type) |
7bc09003 JK |
764 | { |
765 | struct sit_info *sit_i = SIT_I(sbi); | |
766 | int ret; | |
8a2d0ace | 767 | |
7bc09003 | 768 | mutex_lock(&sit_i->sentry_lock); |
8a2d0ace GZ |
769 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
770 | NO_CHECK_TYPE, LFS); | |
7bc09003 JK |
771 | mutex_unlock(&sit_i->sentry_lock); |
772 | return ret; | |
773 | } | |
774 | ||
5ee5293c | 775 | static int do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno, |
7dda2af8 | 776 | struct gc_inode_list *gc_list, int gc_type) |
7bc09003 JK |
777 | { |
778 | struct page *sum_page; | |
779 | struct f2fs_summary_block *sum; | |
c718379b | 780 | struct blk_plug plug; |
5ee5293c | 781 | int nfree = 0; |
7bc09003 JK |
782 | |
783 | /* read segment summary of victim */ | |
784 | sum_page = get_sum_page(sbi, segno); | |
7bc09003 | 785 | |
c718379b JK |
786 | blk_start_plug(&plug); |
787 | ||
7bc09003 JK |
788 | sum = page_address(sum_page); |
789 | ||
9236cac5 JK |
790 | /* |
791 | * this is to avoid deadlock: | |
792 | * - lock_page(sum_page) - f2fs_replace_block | |
793 | * - check_valid_map() - mutex_lock(sentry_lock) | |
794 | * - mutex_lock(sentry_lock) - change_curseg() | |
795 | * - lock_page(sum_page) | |
796 | */ | |
797 | unlock_page(sum_page); | |
798 | ||
7bc09003 JK |
799 | switch (GET_SUM_TYPE((&sum->footer))) { |
800 | case SUM_TYPE_NODE: | |
5ee5293c | 801 | nfree = gc_node_segment(sbi, sum->entries, segno, gc_type); |
7bc09003 JK |
802 | break; |
803 | case SUM_TYPE_DATA: | |
5ee5293c JK |
804 | nfree = gc_data_segment(sbi, sum->entries, gc_list, |
805 | segno, gc_type); | |
7bc09003 JK |
806 | break; |
807 | } | |
c718379b JK |
808 | blk_finish_plug(&plug); |
809 | ||
e1235983 | 810 | stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)), gc_type); |
7bc09003 JK |
811 | stat_inc_call_count(sbi->stat_info); |
812 | ||
9236cac5 | 813 | f2fs_put_page(sum_page, 0); |
5ee5293c | 814 | return nfree; |
7bc09003 JK |
815 | } |
816 | ||
d530d4d8 | 817 | int f2fs_gc(struct f2fs_sb_info *sbi, bool sync) |
7bc09003 | 818 | { |
ab126cfc | 819 | unsigned int segno, i; |
d530d4d8 | 820 | int gc_type = sync ? FG_GC : BG_GC; |
45fe8492 | 821 | int sec_freed = 0; |
d530d4d8 | 822 | int ret = -EINVAL; |
d5053a34 | 823 | struct cp_control cpc; |
7dda2af8 CL |
824 | struct gc_inode_list gc_list = { |
825 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
769ec6e5 | 826 | .iroot = RADIX_TREE_INIT(GFP_NOFS), |
7dda2af8 | 827 | }; |
d5053a34 | 828 | |
119ee914 | 829 | cpc.reason = __get_cp_reason(sbi); |
7bc09003 | 830 | gc_more: |
ab126cfc JK |
831 | segno = NULL_SEGNO; |
832 | ||
6bacf52f | 833 | if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) |
408e9375 | 834 | goto stop; |
1e968fdf | 835 | if (unlikely(f2fs_cp_error(sbi))) |
203681f6 | 836 | goto stop; |
7bc09003 | 837 | |
45fe8492 | 838 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, sec_freed)) { |
408e9375 | 839 | gc_type = FG_GC; |
798c1b16 JK |
840 | if (__get_victim(sbi, &segno, gc_type) || prefree_segments(sbi)) |
841 | write_checkpoint(sbi, &cpc); | |
d64f8047 | 842 | } |
7bc09003 | 843 | |
798c1b16 | 844 | if (segno == NULL_SEGNO && !__get_victim(sbi, &segno, gc_type)) |
408e9375 | 845 | goto stop; |
43727527 | 846 | ret = 0; |
7bc09003 | 847 | |
81c1a0f1 CY |
848 | /* readahead multi ssa blocks those have contiguous address */ |
849 | if (sbi->segs_per_sec > 1) | |
850 | ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), sbi->segs_per_sec, | |
26879fb1 | 851 | META_SSA, true); |
81c1a0f1 | 852 | |
45fe8492 CY |
853 | for (i = 0; i < sbi->segs_per_sec; i++) { |
854 | /* | |
855 | * for FG_GC case, halt gcing left segments once failed one | |
856 | * of segments in selected section to avoid long latency. | |
857 | */ | |
858 | if (!do_garbage_collect(sbi, segno + i, &gc_list, gc_type) && | |
859 | gc_type == FG_GC) | |
860 | break; | |
861 | } | |
862 | ||
863 | if (i == sbi->segs_per_sec && gc_type == FG_GC) | |
864 | sec_freed++; | |
43727527 | 865 | |
5ee5293c | 866 | if (gc_type == FG_GC) |
5ec4e49f | 867 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 868 | |
d530d4d8 CY |
869 | if (!sync) { |
870 | if (has_not_enough_free_secs(sbi, sec_freed)) | |
871 | goto gc_more; | |
43727527 | 872 | |
d530d4d8 CY |
873 | if (gc_type == FG_GC) |
874 | write_checkpoint(sbi, &cpc); | |
875 | } | |
408e9375 | 876 | stop: |
7bc09003 JK |
877 | mutex_unlock(&sbi->gc_mutex); |
878 | ||
7dda2af8 | 879 | put_gc_inode(&gc_list); |
d530d4d8 CY |
880 | |
881 | if (sync) | |
882 | ret = sec_freed ? 0 : -EAGAIN; | |
43727527 | 883 | return ret; |
7bc09003 JK |
884 | } |
885 | ||
886 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
887 | { | |
888 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
889 | } |