]>
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 | ||
688159b6 FL |
248 | static unsigned int count_bits(const unsigned long *addr, |
249 | unsigned int offset, unsigned int len) | |
250 | { | |
251 | unsigned int end = offset + len, sum = 0; | |
252 | ||
253 | while (offset < end) { | |
254 | if (test_bit(offset++, addr)) | |
255 | ++sum; | |
256 | } | |
257 | return sum; | |
258 | } | |
259 | ||
0a8165d7 | 260 | /* |
111d2495 | 261 | * This function is called from two paths. |
7bc09003 JK |
262 | * One is garbage collection and the other is SSR segment selection. |
263 | * When it is called during GC, it just gets a victim segment | |
264 | * and it does not remove it from dirty seglist. | |
265 | * When it is called from SSR segment selection, it finds a segment | |
266 | * which has minimum valid blocks and removes it from dirty seglist. | |
267 | */ | |
268 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
269 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
270 | { | |
271 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
272 | struct victim_sel_policy p; | |
4ce53776 | 273 | unsigned int secno, max_cost, last_victim; |
a43f7ec3 | 274 | unsigned int last_segment = MAIN_SEGS(sbi); |
688159b6 | 275 | unsigned int nsearched = 0; |
7bc09003 | 276 | |
210f41bc CY |
277 | mutex_lock(&dirty_i->seglist_lock); |
278 | ||
7bc09003 JK |
279 | p.alloc_mode = alloc_mode; |
280 | select_policy(sbi, gc_type, type, &p); | |
281 | ||
282 | p.min_segno = NULL_SEGNO; | |
b2b3460a | 283 | p.min_cost = max_cost = get_max_cost(sbi, &p); |
7bc09003 | 284 | |
3342bb30 CY |
285 | if (p.max_search == 0) |
286 | goto out; | |
287 | ||
4ce53776 | 288 | last_victim = sbi->last_victim[p.gc_mode]; |
7bc09003 JK |
289 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
290 | p.min_segno = check_bg_victims(sbi); | |
291 | if (p.min_segno != NULL_SEGNO) | |
292 | goto got_it; | |
293 | } | |
294 | ||
295 | while (1) { | |
296 | unsigned long cost; | |
5ec4e49f | 297 | unsigned int segno; |
7bc09003 | 298 | |
a43f7ec3 CY |
299 | segno = find_next_bit(p.dirty_segmap, last_segment, p.offset); |
300 | if (segno >= last_segment) { | |
7bc09003 | 301 | if (sbi->last_victim[p.gc_mode]) { |
a43f7ec3 | 302 | last_segment = sbi->last_victim[p.gc_mode]; |
7bc09003 JK |
303 | sbi->last_victim[p.gc_mode] = 0; |
304 | p.offset = 0; | |
305 | continue; | |
306 | } | |
307 | break; | |
308 | } | |
a57e564d JX |
309 | |
310 | p.offset = segno + p.ofs_unit; | |
688159b6 | 311 | if (p.ofs_unit > 1) { |
a57e564d | 312 | p.offset -= segno % p.ofs_unit; |
688159b6 FL |
313 | nsearched += count_bits(p.dirty_segmap, |
314 | p.offset - p.ofs_unit, | |
315 | p.ofs_unit); | |
316 | } else { | |
317 | nsearched++; | |
318 | } | |
319 | ||
a57e564d | 320 | |
5ec4e49f | 321 | secno = GET_SECNO(sbi, segno); |
7bc09003 | 322 | |
5ec4e49f | 323 | if (sec_usage_check(sbi, secno)) |
688159b6 | 324 | goto next; |
5ec4e49f | 325 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
688159b6 | 326 | goto next; |
7bc09003 JK |
327 | |
328 | cost = get_gc_cost(sbi, segno, &p); | |
329 | ||
330 | if (p.min_cost > cost) { | |
331 | p.min_segno = segno; | |
332 | p.min_cost = cost; | |
a57e564d | 333 | } |
688159b6 FL |
334 | next: |
335 | if (nsearched >= p.max_search) { | |
4ce53776 JK |
336 | if (!sbi->last_victim[p.gc_mode] && segno <= last_victim) |
337 | sbi->last_victim[p.gc_mode] = last_victim + 1; | |
338 | else | |
339 | sbi->last_victim[p.gc_mode] = segno + 1; | |
7bc09003 JK |
340 | break; |
341 | } | |
342 | } | |
7bc09003 | 343 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 344 | got_it: |
7bc09003 | 345 | if (p.alloc_mode == LFS) { |
5ec4e49f JK |
346 | secno = GET_SECNO(sbi, p.min_segno); |
347 | if (gc_type == FG_GC) | |
348 | sbi->cur_victim_sec = secno; | |
349 | else | |
350 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 351 | } |
5ec4e49f | 352 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
353 | |
354 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
355 | sbi->cur_victim_sec, | |
356 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 | 357 | } |
3342bb30 | 358 | out: |
7bc09003 JK |
359 | mutex_unlock(&dirty_i->seglist_lock); |
360 | ||
361 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
362 | } | |
363 | ||
364 | static const struct victim_selection default_v_ops = { | |
365 | .get_victim = get_victim_by_default, | |
366 | }; | |
367 | ||
7dda2af8 | 368 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 369 | { |
7bc09003 JK |
370 | struct inode_entry *ie; |
371 | ||
7dda2af8 CL |
372 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
373 | if (ie) | |
374 | return ie->inode; | |
7bc09003 JK |
375 | return NULL; |
376 | } | |
377 | ||
7dda2af8 | 378 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 379 | { |
6cc4af56 GZ |
380 | struct inode_entry *new_ie; |
381 | ||
7dda2af8 | 382 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
383 | iput(inode); |
384 | return; | |
7bc09003 | 385 | } |
06292073 | 386 | new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); |
7bc09003 | 387 | new_ie->inode = inode; |
f28e5034 CY |
388 | |
389 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 390 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
391 | } |
392 | ||
7dda2af8 | 393 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
394 | { |
395 | struct inode_entry *ie, *next_ie; | |
7dda2af8 CL |
396 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
397 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
398 | iput(ie->inode); |
399 | list_del(&ie->list); | |
06292073 | 400 | kmem_cache_free(inode_entry_slab, ie); |
7bc09003 JK |
401 | } |
402 | } | |
403 | ||
404 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
405 | unsigned int segno, int offset) | |
406 | { | |
407 | struct sit_info *sit_i = SIT_I(sbi); | |
408 | struct seg_entry *sentry; | |
409 | int ret; | |
410 | ||
411 | mutex_lock(&sit_i->sentry_lock); | |
412 | sentry = get_seg_entry(sbi, segno); | |
413 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
414 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 415 | return ret; |
7bc09003 JK |
416 | } |
417 | ||
0a8165d7 | 418 | /* |
7bc09003 JK |
419 | * This function compares node address got in summary with that in NAT. |
420 | * On validity, copy that node with cold status, otherwise (invalid node) | |
421 | * ignore that. | |
422 | */ | |
718e53fa | 423 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
424 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
425 | { | |
426 | bool initial = true; | |
427 | struct f2fs_summary *entry; | |
26d58599 | 428 | block_t start_addr; |
7bc09003 JK |
429 | int off; |
430 | ||
26d58599 JK |
431 | start_addr = START_BLOCK(sbi, segno); |
432 | ||
7bc09003 JK |
433 | next_step: |
434 | entry = sum; | |
c718379b | 435 | |
7bc09003 JK |
436 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
437 | nid_t nid = le32_to_cpu(entry->nid); | |
438 | struct page *node_page; | |
26d58599 | 439 | struct node_info ni; |
7bc09003 | 440 | |
43727527 JK |
441 | /* stop BG_GC if there is not enough free sections. */ |
442 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
718e53fa | 443 | return; |
7bc09003 | 444 | |
43727527 | 445 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
446 | continue; |
447 | ||
448 | if (initial) { | |
449 | ra_node_page(sbi, nid); | |
450 | continue; | |
451 | } | |
452 | node_page = get_node_page(sbi, nid); | |
453 | if (IS_ERR(node_page)) | |
454 | continue; | |
455 | ||
9a01b56b HY |
456 | /* block may become invalid during get_node_page */ |
457 | if (check_valid_map(sbi, segno, off) == 0) { | |
458 | f2fs_put_page(node_page, 1); | |
459 | continue; | |
26d58599 JK |
460 | } |
461 | ||
462 | get_node_info(sbi, nid, &ni); | |
463 | if (ni.blk_addr != start_addr + off) { | |
464 | f2fs_put_page(node_page, 1); | |
465 | continue; | |
9a01b56b HY |
466 | } |
467 | ||
7bc09003 | 468 | /* set page dirty and write it */ |
4ebefc44 | 469 | if (gc_type == FG_GC) { |
fec1d657 | 470 | f2fs_wait_on_page_writeback(node_page, NODE, true); |
7bc09003 | 471 | set_page_dirty(node_page); |
4ebefc44 JK |
472 | } else { |
473 | if (!PageWriteback(node_page)) | |
474 | set_page_dirty(node_page); | |
475 | } | |
7bc09003 | 476 | f2fs_put_page(node_page, 1); |
e1235983 | 477 | stat_inc_node_blk_count(sbi, 1, gc_type); |
7bc09003 | 478 | } |
c718379b | 479 | |
7bc09003 JK |
480 | if (initial) { |
481 | initial = false; | |
482 | goto next_step; | |
483 | } | |
7bc09003 JK |
484 | } |
485 | ||
0a8165d7 | 486 | /* |
9af45ef5 JK |
487 | * Calculate start block index indicating the given node offset. |
488 | * Be careful, caller should give this node offset only indicating direct node | |
489 | * blocks. If any node offsets, which point the other types of node blocks such | |
490 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
491 | * bug. | |
7bc09003 | 492 | */ |
81ca7350 | 493 | block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode) |
7bc09003 | 494 | { |
ce19a5d4 JK |
495 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
496 | unsigned int bidx; | |
7bc09003 | 497 | |
ce19a5d4 JK |
498 | if (node_ofs == 0) |
499 | return 0; | |
7bc09003 | 500 | |
ce19a5d4 | 501 | if (node_ofs <= 2) { |
7bc09003 JK |
502 | bidx = node_ofs - 1; |
503 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 504 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
505 | bidx = node_ofs - 2 - dec; |
506 | } else { | |
ce19a5d4 | 507 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
508 | bidx = node_ofs - 5 - dec; |
509 | } | |
81ca7350 | 510 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode); |
7bc09003 JK |
511 | } |
512 | ||
c1079892 | 513 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
514 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
515 | { | |
516 | struct page *node_page; | |
517 | nid_t nid; | |
518 | unsigned int ofs_in_node; | |
519 | block_t source_blkaddr; | |
520 | ||
521 | nid = le32_to_cpu(sum->nid); | |
522 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
523 | ||
524 | node_page = get_node_page(sbi, nid); | |
525 | if (IS_ERR(node_page)) | |
c1079892 | 526 | return false; |
7bc09003 JK |
527 | |
528 | get_node_info(sbi, nid, dni); | |
529 | ||
530 | if (sum->version != dni->version) { | |
531 | f2fs_put_page(node_page, 1); | |
c1079892 | 532 | return false; |
7bc09003 JK |
533 | } |
534 | ||
535 | *nofs = ofs_of_node(node_page); | |
536 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
537 | f2fs_put_page(node_page, 1); | |
538 | ||
539 | if (source_blkaddr != blkaddr) | |
c1079892 NK |
540 | return false; |
541 | return true; | |
7bc09003 JK |
542 | } |
543 | ||
4375a336 JK |
544 | static void move_encrypted_block(struct inode *inode, block_t bidx) |
545 | { | |
546 | struct f2fs_io_info fio = { | |
547 | .sbi = F2FS_I_SB(inode), | |
548 | .type = DATA, | |
549 | .rw = READ_SYNC, | |
550 | .encrypted_page = NULL, | |
551 | }; | |
552 | struct dnode_of_data dn; | |
553 | struct f2fs_summary sum; | |
554 | struct node_info ni; | |
555 | struct page *page; | |
4356e48e | 556 | block_t newaddr; |
4375a336 JK |
557 | int err; |
558 | ||
559 | /* do not read out */ | |
a56c7c6f | 560 | page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); |
4375a336 JK |
561 | if (!page) |
562 | return; | |
563 | ||
564 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
565 | err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE); | |
566 | if (err) | |
567 | goto out; | |
568 | ||
08b39fbd CY |
569 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
570 | ClearPageUptodate(page); | |
4375a336 | 571 | goto put_out; |
08b39fbd CY |
572 | } |
573 | ||
574 | /* | |
575 | * don't cache encrypted data into meta inode until previous dirty | |
576 | * data were writebacked to avoid racing between GC and flush. | |
577 | */ | |
fec1d657 | 578 | f2fs_wait_on_page_writeback(page, DATA, true); |
4375a336 JK |
579 | |
580 | get_node_info(fio.sbi, dn.nid, &ni); | |
581 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); | |
582 | ||
583 | /* read page */ | |
584 | fio.page = page; | |
7a9d7548 | 585 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; |
4375a336 | 586 | |
4356e48e CY |
587 | allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, |
588 | &sum, CURSEG_COLD_DATA); | |
589 | ||
590 | fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), newaddr, | |
591 | FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
592 | if (!fio.encrypted_page) { | |
593 | err = -ENOMEM; | |
594 | goto recover_block; | |
595 | } | |
4375a336 | 596 | |
548aedac JK |
597 | err = f2fs_submit_page_bio(&fio); |
598 | if (err) | |
599 | goto put_page_out; | |
600 | ||
601 | /* write page */ | |
602 | lock_page(fio.encrypted_page); | |
603 | ||
4356e48e CY |
604 | if (unlikely(!PageUptodate(fio.encrypted_page))) { |
605 | err = -EIO; | |
548aedac | 606 | goto put_page_out; |
4356e48e CY |
607 | } |
608 | if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) { | |
609 | err = -EIO; | |
548aedac | 610 | goto put_page_out; |
4356e48e | 611 | } |
548aedac | 612 | |
6282adbf | 613 | set_page_dirty(fio.encrypted_page); |
fec1d657 | 614 | f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true); |
6282adbf JK |
615 | if (clear_page_dirty_for_io(fio.encrypted_page)) |
616 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
617 | ||
548aedac | 618 | set_page_writeback(fio.encrypted_page); |
4375a336 JK |
619 | |
620 | /* allocate block address */ | |
fec1d657 | 621 | f2fs_wait_on_page_writeback(dn.node_page, NODE, true); |
4356e48e | 622 | |
4375a336 | 623 | fio.rw = WRITE_SYNC; |
4356e48e | 624 | fio.new_blkaddr = newaddr; |
4375a336 JK |
625 | f2fs_submit_page_mbio(&fio); |
626 | ||
f28b3434 | 627 | f2fs_update_data_blkaddr(&dn, newaddr); |
4375a336 JK |
628 | set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE); |
629 | if (page->index == 0) | |
630 | set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN); | |
548aedac | 631 | put_page_out: |
4375a336 | 632 | f2fs_put_page(fio.encrypted_page, 1); |
4356e48e CY |
633 | recover_block: |
634 | if (err) | |
635 | __f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, | |
636 | true, true); | |
4375a336 JK |
637 | put_out: |
638 | f2fs_put_dnode(&dn); | |
639 | out: | |
640 | f2fs_put_page(page, 1); | |
641 | } | |
642 | ||
c879f90d | 643 | static void move_data_page(struct inode *inode, block_t bidx, int gc_type) |
7bc09003 | 644 | { |
c879f90d JK |
645 | struct page *page; |
646 | ||
a56c7c6f | 647 | page = get_lock_data_page(inode, bidx, true); |
c879f90d JK |
648 | if (IS_ERR(page)) |
649 | return; | |
63a0b7cb | 650 | |
7bc09003 | 651 | if (gc_type == BG_GC) { |
4ebefc44 JK |
652 | if (PageWriteback(page)) |
653 | goto out; | |
7bc09003 JK |
654 | set_page_dirty(page); |
655 | set_cold_data(page); | |
656 | } else { | |
c879f90d JK |
657 | struct f2fs_io_info fio = { |
658 | .sbi = F2FS_I_SB(inode), | |
659 | .type = DATA, | |
660 | .rw = WRITE_SYNC, | |
661 | .page = page, | |
4375a336 | 662 | .encrypted_page = NULL, |
c879f90d | 663 | }; |
6282adbf | 664 | set_page_dirty(page); |
fec1d657 | 665 | f2fs_wait_on_page_writeback(page, DATA, true); |
1fe54f9d | 666 | if (clear_page_dirty_for_io(page)) |
a7ffdbe2 | 667 | inode_dec_dirty_pages(inode); |
7bc09003 | 668 | set_cold_data(page); |
05ca3632 | 669 | do_write_data_page(&fio); |
7bc09003 JK |
670 | clear_cold_data(page); |
671 | } | |
672 | out: | |
673 | f2fs_put_page(page, 1); | |
674 | } | |
675 | ||
0a8165d7 | 676 | /* |
7bc09003 JK |
677 | * This function tries to get parent node of victim data block, and identifies |
678 | * data block validity. If the block is valid, copy that with cold status and | |
679 | * modify parent node. | |
680 | * If the parent node is not valid or the data block address is different, | |
681 | * the victim data block is ignored. | |
682 | */ | |
718e53fa | 683 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dda2af8 | 684 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type) |
7bc09003 JK |
685 | { |
686 | struct super_block *sb = sbi->sb; | |
687 | struct f2fs_summary *entry; | |
688 | block_t start_addr; | |
43727527 | 689 | int off; |
7bc09003 JK |
690 | int phase = 0; |
691 | ||
692 | start_addr = START_BLOCK(sbi, segno); | |
693 | ||
694 | next_step: | |
695 | entry = sum; | |
c718379b | 696 | |
7bc09003 JK |
697 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
698 | struct page *data_page; | |
699 | struct inode *inode; | |
700 | struct node_info dni; /* dnode info for the data */ | |
701 | unsigned int ofs_in_node, nofs; | |
702 | block_t start_bidx; | |
703 | ||
43727527 JK |
704 | /* stop BG_GC if there is not enough free sections. */ |
705 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
718e53fa | 706 | return; |
7bc09003 | 707 | |
43727527 | 708 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
709 | continue; |
710 | ||
711 | if (phase == 0) { | |
712 | ra_node_page(sbi, le32_to_cpu(entry->nid)); | |
713 | continue; | |
714 | } | |
715 | ||
716 | /* Get an inode by ino with checking validity */ | |
c1079892 | 717 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
718 | continue; |
719 | ||
720 | if (phase == 1) { | |
721 | ra_node_page(sbi, dni.ino); | |
722 | continue; | |
723 | } | |
724 | ||
7bc09003 JK |
725 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
726 | ||
727 | if (phase == 2) { | |
d4686d56 | 728 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 729 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
730 | continue; |
731 | ||
4375a336 JK |
732 | /* if encrypted inode, let's go phase 3 */ |
733 | if (f2fs_encrypted_inode(inode) && | |
734 | S_ISREG(inode->i_mode)) { | |
735 | add_gc_inode(gc_list, inode); | |
736 | continue; | |
737 | } | |
738 | ||
81ca7350 | 739 | start_bidx = start_bidx_of_node(nofs, inode); |
43f3eae1 | 740 | data_page = get_read_data_page(inode, |
a56c7c6f | 741 | start_bidx + ofs_in_node, READA, true); |
31a32688 CL |
742 | if (IS_ERR(data_page)) { |
743 | iput(inode); | |
744 | continue; | |
745 | } | |
7bc09003 JK |
746 | |
747 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 748 | add_gc_inode(gc_list, inode); |
31a32688 CL |
749 | continue; |
750 | } | |
751 | ||
752 | /* phase 3 */ | |
7dda2af8 | 753 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 754 | if (inode) { |
81ca7350 | 755 | start_bidx = start_bidx_of_node(nofs, inode) |
c879f90d | 756 | + ofs_in_node; |
4375a336 JK |
757 | if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) |
758 | move_encrypted_block(inode, start_bidx); | |
759 | else | |
760 | move_data_page(inode, start_bidx, gc_type); | |
e1235983 | 761 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 762 | } |
7bc09003 | 763 | } |
c718379b | 764 | |
7bc09003 JK |
765 | if (++phase < 4) |
766 | goto next_step; | |
7bc09003 JK |
767 | } |
768 | ||
769 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 770 | int gc_type) |
7bc09003 JK |
771 | { |
772 | struct sit_info *sit_i = SIT_I(sbi); | |
773 | int ret; | |
8a2d0ace | 774 | |
7bc09003 | 775 | mutex_lock(&sit_i->sentry_lock); |
8a2d0ace GZ |
776 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
777 | NO_CHECK_TYPE, LFS); | |
7bc09003 JK |
778 | mutex_unlock(&sit_i->sentry_lock); |
779 | return ret; | |
780 | } | |
781 | ||
718e53fa CY |
782 | static int do_garbage_collect(struct f2fs_sb_info *sbi, |
783 | unsigned int start_segno, | |
7dda2af8 | 784 | struct gc_inode_list *gc_list, int gc_type) |
7bc09003 JK |
785 | { |
786 | struct page *sum_page; | |
787 | struct f2fs_summary_block *sum; | |
c718379b | 788 | struct blk_plug plug; |
718e53fa CY |
789 | unsigned int segno = start_segno; |
790 | unsigned int end_segno = start_segno + sbi->segs_per_sec; | |
791 | int seg_freed = 0; | |
792 | unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? | |
793 | SUM_TYPE_DATA : SUM_TYPE_NODE; | |
7bc09003 | 794 | |
718e53fa CY |
795 | /* readahead multi ssa blocks those have contiguous address */ |
796 | if (sbi->segs_per_sec > 1) | |
797 | ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), | |
798 | sbi->segs_per_sec, META_SSA, true); | |
799 | ||
800 | /* reference all summary page */ | |
801 | while (segno < end_segno) { | |
802 | sum_page = get_sum_page(sbi, segno++); | |
803 | unlock_page(sum_page); | |
804 | } | |
7bc09003 | 805 | |
c718379b JK |
806 | blk_start_plug(&plug); |
807 | ||
718e53fa CY |
808 | for (segno = start_segno; segno < end_segno; segno++) { |
809 | /* find segment summary of victim */ | |
810 | sum_page = find_get_page(META_MAPPING(sbi), | |
811 | GET_SUM_BLOCK(sbi, segno)); | |
812 | f2fs_bug_on(sbi, !PageUptodate(sum_page)); | |
813 | f2fs_put_page(sum_page, 0); | |
7bc09003 | 814 | |
718e53fa CY |
815 | sum = page_address(sum_page); |
816 | f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer))); | |
817 | ||
818 | /* | |
819 | * this is to avoid deadlock: | |
820 | * - lock_page(sum_page) - f2fs_replace_block | |
821 | * - check_valid_map() - mutex_lock(sentry_lock) | |
822 | * - mutex_lock(sentry_lock) - change_curseg() | |
823 | * - lock_page(sum_page) | |
824 | */ | |
825 | ||
826 | if (type == SUM_TYPE_NODE) | |
827 | gc_node_segment(sbi, sum->entries, segno, gc_type); | |
828 | else | |
829 | gc_data_segment(sbi, sum->entries, gc_list, segno, | |
830 | gc_type); | |
831 | ||
832 | stat_inc_seg_count(sbi, type, gc_type); | |
718e53fa CY |
833 | |
834 | f2fs_put_page(sum_page, 0); | |
835 | } | |
836 | ||
837 | if (gc_type == FG_GC) { | |
838 | if (type == SUM_TYPE_NODE) { | |
839 | struct writeback_control wbc = { | |
840 | .sync_mode = WB_SYNC_ALL, | |
841 | .nr_to_write = LONG_MAX, | |
842 | .for_reclaim = 0, | |
843 | }; | |
844 | sync_node_pages(sbi, 0, &wbc); | |
845 | } else { | |
846 | f2fs_submit_merged_bio(sbi, DATA, WRITE); | |
847 | } | |
7bc09003 | 848 | } |
c718379b | 849 | |
718e53fa | 850 | blk_finish_plug(&plug); |
7bc09003 | 851 | |
718e53fa CY |
852 | if (gc_type == FG_GC) { |
853 | while (start_segno < end_segno) | |
854 | if (get_valid_blocks(sbi, start_segno++, 1) == 0) | |
855 | seg_freed++; | |
856 | } | |
17d899df CY |
857 | |
858 | stat_inc_call_count(sbi->stat_info); | |
859 | ||
718e53fa | 860 | return seg_freed; |
7bc09003 JK |
861 | } |
862 | ||
d530d4d8 | 863 | int f2fs_gc(struct f2fs_sb_info *sbi, bool sync) |
7bc09003 | 864 | { |
718e53fa | 865 | unsigned int segno; |
d530d4d8 | 866 | int gc_type = sync ? FG_GC : BG_GC; |
718e53fa | 867 | int sec_freed = 0, seg_freed; |
d530d4d8 | 868 | int ret = -EINVAL; |
d5053a34 | 869 | struct cp_control cpc; |
7dda2af8 CL |
870 | struct gc_inode_list gc_list = { |
871 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
769ec6e5 | 872 | .iroot = RADIX_TREE_INIT(GFP_NOFS), |
7dda2af8 | 873 | }; |
d5053a34 | 874 | |
119ee914 | 875 | cpc.reason = __get_cp_reason(sbi); |
7bc09003 | 876 | gc_more: |
ab126cfc JK |
877 | segno = NULL_SEGNO; |
878 | ||
6bacf52f | 879 | if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) |
408e9375 | 880 | goto stop; |
6d5a1495 CY |
881 | if (unlikely(f2fs_cp_error(sbi))) { |
882 | ret = -EIO; | |
203681f6 | 883 | goto stop; |
6d5a1495 | 884 | } |
7bc09003 | 885 | |
45fe8492 | 886 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, sec_freed)) { |
408e9375 | 887 | gc_type = FG_GC; |
6e17bfbc JK |
888 | /* |
889 | * If there is no victim and no prefree segment but still not | |
890 | * enough free sections, we should flush dent/node blocks and do | |
891 | * garbage collections. | |
892 | */ | |
798c1b16 JK |
893 | if (__get_victim(sbi, &segno, gc_type) || prefree_segments(sbi)) |
894 | write_checkpoint(sbi, &cpc); | |
6e17bfbc JK |
895 | else if (has_not_enough_free_secs(sbi, 0)) |
896 | write_checkpoint(sbi, &cpc); | |
d64f8047 | 897 | } |
7bc09003 | 898 | |
798c1b16 | 899 | if (segno == NULL_SEGNO && !__get_victim(sbi, &segno, gc_type)) |
408e9375 | 900 | goto stop; |
43727527 | 901 | ret = 0; |
7bc09003 | 902 | |
718e53fa | 903 | seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type); |
45fe8492 | 904 | |
718e53fa | 905 | if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec) |
45fe8492 | 906 | sec_freed++; |
43727527 | 907 | |
5ee5293c | 908 | if (gc_type == FG_GC) |
5ec4e49f | 909 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 910 | |
d530d4d8 CY |
911 | if (!sync) { |
912 | if (has_not_enough_free_secs(sbi, sec_freed)) | |
913 | goto gc_more; | |
43727527 | 914 | |
d530d4d8 CY |
915 | if (gc_type == FG_GC) |
916 | write_checkpoint(sbi, &cpc); | |
917 | } | |
408e9375 | 918 | stop: |
7bc09003 JK |
919 | mutex_unlock(&sbi->gc_mutex); |
920 | ||
7dda2af8 | 921 | put_gc_inode(&gc_list); |
d530d4d8 CY |
922 | |
923 | if (sync) | |
924 | ret = sec_freed ? 0 : -EAGAIN; | |
43727527 | 925 | return ret; |
7bc09003 JK |
926 | } |
927 | ||
928 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
929 | { | |
930 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
931 | } |