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 JK |
80 | /* if return value is not zero, no victim was selected */ |
81 | if (f2fs_gc(sbi)) | |
b59d0bae | 82 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 83 | |
4660f9c0 JK |
84 | /* balancing f2fs's metadata periodically */ |
85 | f2fs_balance_fs_bg(sbi); | |
81eb8d6e | 86 | |
7bc09003 JK |
87 | } while (!kthread_should_stop()); |
88 | return 0; | |
89 | } | |
90 | ||
91 | int start_gc_thread(struct f2fs_sb_info *sbi) | |
92 | { | |
1042d60f | 93 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 94 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 95 | int err = 0; |
7bc09003 JK |
96 | |
97 | gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL); | |
7a267f8d NJ |
98 | if (!gc_th) { |
99 | err = -ENOMEM; | |
100 | goto out; | |
101 | } | |
7bc09003 | 102 | |
b59d0bae NJ |
103 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
104 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
105 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
106 | ||
d2dc095f NJ |
107 | gc_th->gc_idle = 0; |
108 | ||
7bc09003 JK |
109 | sbi->gc_thread = gc_th; |
110 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
111 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 112 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 113 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 114 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 115 | kfree(gc_th); |
25718423 | 116 | sbi->gc_thread = NULL; |
7bc09003 | 117 | } |
7a267f8d NJ |
118 | out: |
119 | return err; | |
7bc09003 JK |
120 | } |
121 | ||
122 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
123 | { | |
124 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
125 | if (!gc_th) | |
126 | return; | |
127 | kthread_stop(gc_th->f2fs_gc_task); | |
128 | kfree(gc_th); | |
129 | sbi->gc_thread = NULL; | |
130 | } | |
131 | ||
d2dc095f | 132 | static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type) |
7bc09003 | 133 | { |
d2dc095f NJ |
134 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
135 | ||
136 | if (gc_th && gc_th->gc_idle) { | |
137 | if (gc_th->gc_idle == 1) | |
138 | gc_mode = GC_CB; | |
139 | else if (gc_th->gc_idle == 2) | |
140 | gc_mode = GC_GREEDY; | |
141 | } | |
142 | return gc_mode; | |
7bc09003 JK |
143 | } |
144 | ||
145 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
146 | int type, struct victim_sel_policy *p) | |
147 | { | |
148 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
149 | ||
4ebefc44 | 150 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
151 | p->gc_mode = GC_GREEDY; |
152 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
a26b7c8a | 153 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 JK |
154 | p->ofs_unit = 1; |
155 | } else { | |
d2dc095f | 156 | p->gc_mode = select_gc_type(sbi->gc_thread, gc_type); |
7bc09003 | 157 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
a26b7c8a | 158 | p->max_search = dirty_i->nr_dirty[DIRTY]; |
7bc09003 JK |
159 | p->ofs_unit = sbi->segs_per_sec; |
160 | } | |
a26b7c8a | 161 | |
b1c57c1c JK |
162 | if (p->max_search > sbi->max_victim_search) |
163 | p->max_search = sbi->max_victim_search; | |
a26b7c8a | 164 | |
7bc09003 JK |
165 | p->offset = sbi->last_victim[p->gc_mode]; |
166 | } | |
167 | ||
168 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
169 | struct victim_sel_policy *p) | |
170 | { | |
b7250d2d JK |
171 | /* SSR allocates in a segment unit */ |
172 | if (p->alloc_mode == SSR) | |
173 | return 1 << sbi->log_blocks_per_seg; | |
7bc09003 JK |
174 | if (p->gc_mode == GC_GREEDY) |
175 | return (1 << sbi->log_blocks_per_seg) * p->ofs_unit; | |
176 | else if (p->gc_mode == GC_CB) | |
177 | return UINT_MAX; | |
178 | else /* No other gc_mode */ | |
179 | return 0; | |
180 | } | |
181 | ||
182 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
183 | { | |
184 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 185 | unsigned int secno; |
7bc09003 JK |
186 | |
187 | /* | |
188 | * If the gc_type is FG_GC, we can select victim segments | |
189 | * selected by background GC before. | |
190 | * Those segments guarantee they have small valid blocks. | |
191 | */ | |
7cd8558b | 192 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 193 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 194 | continue; |
5ec4e49f JK |
195 | clear_bit(secno, dirty_i->victim_secmap); |
196 | return secno * sbi->segs_per_sec; | |
7bc09003 JK |
197 | } |
198 | return NULL_SEGNO; | |
199 | } | |
200 | ||
201 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
202 | { | |
203 | struct sit_info *sit_i = SIT_I(sbi); | |
204 | unsigned int secno = GET_SECNO(sbi, segno); | |
205 | unsigned int start = secno * sbi->segs_per_sec; | |
206 | unsigned long long mtime = 0; | |
207 | unsigned int vblocks; | |
208 | unsigned char age = 0; | |
209 | unsigned char u; | |
210 | unsigned int i; | |
211 | ||
212 | for (i = 0; i < sbi->segs_per_sec; i++) | |
213 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
214 | vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
215 | ||
216 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
217 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
218 | ||
219 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
220 | ||
e1c42045 | 221 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
222 | if (mtime < sit_i->min_mtime) |
223 | sit_i->min_mtime = mtime; | |
224 | if (mtime > sit_i->max_mtime) | |
225 | sit_i->max_mtime = mtime; | |
226 | if (sit_i->max_mtime != sit_i->min_mtime) | |
227 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
228 | sit_i->max_mtime - sit_i->min_mtime); | |
229 | ||
230 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
231 | } | |
232 | ||
a57e564d JX |
233 | static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, |
234 | unsigned int segno, struct victim_sel_policy *p) | |
7bc09003 JK |
235 | { |
236 | if (p->alloc_mode == SSR) | |
237 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; | |
238 | ||
239 | /* alloc_mode == LFS */ | |
240 | if (p->gc_mode == GC_GREEDY) | |
241 | return get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
242 | else | |
243 | return get_cb_cost(sbi, segno); | |
244 | } | |
245 | ||
0a8165d7 | 246 | /* |
111d2495 | 247 | * This function is called from two paths. |
7bc09003 JK |
248 | * One is garbage collection and the other is SSR segment selection. |
249 | * When it is called during GC, it just gets a victim segment | |
250 | * and it does not remove it from dirty seglist. | |
251 | * When it is called from SSR segment selection, it finds a segment | |
252 | * which has minimum valid blocks and removes it from dirty seglist. | |
253 | */ | |
254 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
255 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
256 | { | |
257 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
258 | struct victim_sel_policy p; | |
b2b3460a | 259 | unsigned int secno, max_cost; |
a43f7ec3 | 260 | unsigned int last_segment = MAIN_SEGS(sbi); |
7bc09003 JK |
261 | int nsearched = 0; |
262 | ||
210f41bc CY |
263 | mutex_lock(&dirty_i->seglist_lock); |
264 | ||
7bc09003 JK |
265 | p.alloc_mode = alloc_mode; |
266 | select_policy(sbi, gc_type, type, &p); | |
267 | ||
268 | p.min_segno = NULL_SEGNO; | |
b2b3460a | 269 | p.min_cost = max_cost = get_max_cost(sbi, &p); |
7bc09003 | 270 | |
7bc09003 JK |
271 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
272 | p.min_segno = check_bg_victims(sbi); | |
273 | if (p.min_segno != NULL_SEGNO) | |
274 | goto got_it; | |
275 | } | |
276 | ||
277 | while (1) { | |
278 | unsigned long cost; | |
5ec4e49f | 279 | unsigned int segno; |
7bc09003 | 280 | |
a43f7ec3 CY |
281 | segno = find_next_bit(p.dirty_segmap, last_segment, p.offset); |
282 | if (segno >= last_segment) { | |
7bc09003 | 283 | if (sbi->last_victim[p.gc_mode]) { |
a43f7ec3 | 284 | last_segment = sbi->last_victim[p.gc_mode]; |
7bc09003 JK |
285 | sbi->last_victim[p.gc_mode] = 0; |
286 | p.offset = 0; | |
287 | continue; | |
288 | } | |
289 | break; | |
290 | } | |
a57e564d JX |
291 | |
292 | p.offset = segno + p.ofs_unit; | |
293 | if (p.ofs_unit > 1) | |
294 | p.offset -= segno % p.ofs_unit; | |
295 | ||
5ec4e49f | 296 | secno = GET_SECNO(sbi, segno); |
7bc09003 | 297 | |
5ec4e49f | 298 | if (sec_usage_check(sbi, secno)) |
7bc09003 | 299 | continue; |
5ec4e49f | 300 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
7bc09003 JK |
301 | continue; |
302 | ||
303 | cost = get_gc_cost(sbi, segno, &p); | |
304 | ||
305 | if (p.min_cost > cost) { | |
306 | p.min_segno = segno; | |
307 | p.min_cost = cost; | |
a57e564d | 308 | } else if (unlikely(cost == max_cost)) { |
7bc09003 | 309 | continue; |
a57e564d | 310 | } |
7bc09003 | 311 | |
a26b7c8a | 312 | if (nsearched++ >= p.max_search) { |
7bc09003 JK |
313 | sbi->last_victim[p.gc_mode] = segno; |
314 | break; | |
315 | } | |
316 | } | |
7bc09003 | 317 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 318 | got_it: |
7bc09003 | 319 | if (p.alloc_mode == LFS) { |
5ec4e49f JK |
320 | secno = GET_SECNO(sbi, p.min_segno); |
321 | if (gc_type == FG_GC) | |
322 | sbi->cur_victim_sec = secno; | |
323 | else | |
324 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 325 | } |
5ec4e49f | 326 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
327 | |
328 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
329 | sbi->cur_victim_sec, | |
330 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 JK |
331 | } |
332 | mutex_unlock(&dirty_i->seglist_lock); | |
333 | ||
334 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
335 | } | |
336 | ||
337 | static const struct victim_selection default_v_ops = { | |
338 | .get_victim = get_victim_by_default, | |
339 | }; | |
340 | ||
7dda2af8 | 341 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 342 | { |
7bc09003 JK |
343 | struct inode_entry *ie; |
344 | ||
7dda2af8 CL |
345 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
346 | if (ie) | |
347 | return ie->inode; | |
7bc09003 JK |
348 | return NULL; |
349 | } | |
350 | ||
7dda2af8 | 351 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 352 | { |
6cc4af56 GZ |
353 | struct inode_entry *new_ie; |
354 | ||
7dda2af8 | 355 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
356 | iput(inode); |
357 | return; | |
7bc09003 | 358 | } |
06292073 | 359 | new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); |
7bc09003 | 360 | new_ie->inode = inode; |
f28e5034 CY |
361 | |
362 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 363 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
364 | } |
365 | ||
7dda2af8 | 366 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
367 | { |
368 | struct inode_entry *ie, *next_ie; | |
7dda2af8 CL |
369 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
370 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
371 | iput(ie->inode); |
372 | list_del(&ie->list); | |
06292073 | 373 | kmem_cache_free(inode_entry_slab, ie); |
7bc09003 JK |
374 | } |
375 | } | |
376 | ||
377 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
378 | unsigned int segno, int offset) | |
379 | { | |
380 | struct sit_info *sit_i = SIT_I(sbi); | |
381 | struct seg_entry *sentry; | |
382 | int ret; | |
383 | ||
384 | mutex_lock(&sit_i->sentry_lock); | |
385 | sentry = get_seg_entry(sbi, segno); | |
386 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
387 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 388 | return ret; |
7bc09003 JK |
389 | } |
390 | ||
0a8165d7 | 391 | /* |
7bc09003 JK |
392 | * This function compares node address got in summary with that in NAT. |
393 | * On validity, copy that node with cold status, otherwise (invalid node) | |
394 | * ignore that. | |
395 | */ | |
5ee5293c | 396 | static int gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
397 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
398 | { | |
399 | bool initial = true; | |
400 | struct f2fs_summary *entry; | |
26d58599 | 401 | block_t start_addr; |
7bc09003 JK |
402 | int off; |
403 | ||
26d58599 JK |
404 | start_addr = START_BLOCK(sbi, segno); |
405 | ||
7bc09003 JK |
406 | next_step: |
407 | entry = sum; | |
c718379b | 408 | |
7bc09003 JK |
409 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
410 | nid_t nid = le32_to_cpu(entry->nid); | |
411 | struct page *node_page; | |
26d58599 | 412 | struct node_info ni; |
7bc09003 | 413 | |
43727527 JK |
414 | /* stop BG_GC if there is not enough free sections. */ |
415 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
5ee5293c | 416 | return 0; |
7bc09003 | 417 | |
43727527 | 418 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
419 | continue; |
420 | ||
421 | if (initial) { | |
422 | ra_node_page(sbi, nid); | |
423 | continue; | |
424 | } | |
425 | node_page = get_node_page(sbi, nid); | |
426 | if (IS_ERR(node_page)) | |
427 | continue; | |
428 | ||
9a01b56b HY |
429 | /* block may become invalid during get_node_page */ |
430 | if (check_valid_map(sbi, segno, off) == 0) { | |
431 | f2fs_put_page(node_page, 1); | |
432 | continue; | |
26d58599 JK |
433 | } |
434 | ||
435 | get_node_info(sbi, nid, &ni); | |
436 | if (ni.blk_addr != start_addr + off) { | |
437 | f2fs_put_page(node_page, 1); | |
438 | continue; | |
9a01b56b HY |
439 | } |
440 | ||
7bc09003 | 441 | /* set page dirty and write it */ |
4ebefc44 | 442 | if (gc_type == FG_GC) { |
5514f0aa | 443 | f2fs_wait_on_page_writeback(node_page, NODE); |
7bc09003 | 444 | set_page_dirty(node_page); |
4ebefc44 JK |
445 | } else { |
446 | if (!PageWriteback(node_page)) | |
447 | set_page_dirty(node_page); | |
448 | } | |
7bc09003 | 449 | f2fs_put_page(node_page, 1); |
e1235983 | 450 | stat_inc_node_blk_count(sbi, 1, gc_type); |
7bc09003 | 451 | } |
c718379b | 452 | |
7bc09003 JK |
453 | if (initial) { |
454 | initial = false; | |
455 | goto next_step; | |
456 | } | |
457 | ||
458 | if (gc_type == FG_GC) { | |
459 | struct writeback_control wbc = { | |
460 | .sync_mode = WB_SYNC_ALL, | |
461 | .nr_to_write = LONG_MAX, | |
462 | .for_reclaim = 0, | |
463 | }; | |
464 | sync_node_pages(sbi, 0, &wbc); | |
4ebefc44 | 465 | |
5ee5293c JK |
466 | /* return 1 only if FG_GC succefully reclaimed one */ |
467 | if (get_valid_blocks(sbi, segno, 1) == 0) | |
468 | return 1; | |
7bc09003 | 469 | } |
5ee5293c | 470 | return 0; |
7bc09003 JK |
471 | } |
472 | ||
0a8165d7 | 473 | /* |
9af45ef5 JK |
474 | * Calculate start block index indicating the given node offset. |
475 | * Be careful, caller should give this node offset only indicating direct node | |
476 | * blocks. If any node offsets, which point the other types of node blocks such | |
477 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
478 | * bug. | |
7bc09003 | 479 | */ |
de93653f | 480 | block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi) |
7bc09003 | 481 | { |
ce19a5d4 JK |
482 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
483 | unsigned int bidx; | |
7bc09003 | 484 | |
ce19a5d4 JK |
485 | if (node_ofs == 0) |
486 | return 0; | |
7bc09003 | 487 | |
ce19a5d4 | 488 | if (node_ofs <= 2) { |
7bc09003 JK |
489 | bidx = node_ofs - 1; |
490 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 491 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
492 | bidx = node_ofs - 2 - dec; |
493 | } else { | |
ce19a5d4 | 494 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
495 | bidx = node_ofs - 5 - dec; |
496 | } | |
de93653f | 497 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi); |
7bc09003 JK |
498 | } |
499 | ||
c1079892 | 500 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
501 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
502 | { | |
503 | struct page *node_page; | |
504 | nid_t nid; | |
505 | unsigned int ofs_in_node; | |
506 | block_t source_blkaddr; | |
507 | ||
508 | nid = le32_to_cpu(sum->nid); | |
509 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
510 | ||
511 | node_page = get_node_page(sbi, nid); | |
512 | if (IS_ERR(node_page)) | |
c1079892 | 513 | return false; |
7bc09003 JK |
514 | |
515 | get_node_info(sbi, nid, dni); | |
516 | ||
517 | if (sum->version != dni->version) { | |
518 | f2fs_put_page(node_page, 1); | |
c1079892 | 519 | return false; |
7bc09003 JK |
520 | } |
521 | ||
522 | *nofs = ofs_of_node(node_page); | |
523 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
524 | f2fs_put_page(node_page, 1); | |
525 | ||
526 | if (source_blkaddr != blkaddr) | |
c1079892 NK |
527 | return false; |
528 | return true; | |
7bc09003 JK |
529 | } |
530 | ||
4375a336 JK |
531 | static void move_encrypted_block(struct inode *inode, block_t bidx) |
532 | { | |
533 | struct f2fs_io_info fio = { | |
534 | .sbi = F2FS_I_SB(inode), | |
535 | .type = DATA, | |
536 | .rw = READ_SYNC, | |
537 | .encrypted_page = NULL, | |
538 | }; | |
539 | struct dnode_of_data dn; | |
540 | struct f2fs_summary sum; | |
541 | struct node_info ni; | |
542 | struct page *page; | |
543 | int err; | |
544 | ||
545 | /* do not read out */ | |
546 | page = grab_cache_page(inode->i_mapping, bidx); | |
547 | if (!page) | |
548 | return; | |
549 | ||
550 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
551 | err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE); | |
552 | if (err) | |
553 | goto out; | |
554 | ||
555 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) | |
556 | goto put_out; | |
557 | ||
558 | get_node_info(fio.sbi, dn.nid, &ni); | |
559 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); | |
560 | ||
561 | /* read page */ | |
562 | fio.page = page; | |
563 | fio.blk_addr = dn.data_blkaddr; | |
564 | ||
1b77c416 JK |
565 | fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), |
566 | fio.blk_addr, | |
567 | FGP_LOCK|FGP_CREAT, | |
568 | GFP_NOFS); | |
4375a336 JK |
569 | if (!fio.encrypted_page) |
570 | goto put_out; | |
571 | ||
548aedac JK |
572 | err = f2fs_submit_page_bio(&fio); |
573 | if (err) | |
574 | goto put_page_out; | |
575 | ||
576 | /* write page */ | |
577 | lock_page(fio.encrypted_page); | |
578 | ||
579 | if (unlikely(!PageUptodate(fio.encrypted_page))) | |
580 | goto put_page_out; | |
581 | if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) | |
582 | goto put_page_out; | |
583 | ||
6282adbf JK |
584 | set_page_dirty(fio.encrypted_page); |
585 | f2fs_wait_on_page_writeback(fio.encrypted_page, META); | |
586 | if (clear_page_dirty_for_io(fio.encrypted_page)) | |
587 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
588 | ||
548aedac | 589 | set_page_writeback(fio.encrypted_page); |
4375a336 JK |
590 | |
591 | /* allocate block address */ | |
592 | f2fs_wait_on_page_writeback(dn.node_page, NODE); | |
4375a336 JK |
593 | allocate_data_block(fio.sbi, NULL, fio.blk_addr, |
594 | &fio.blk_addr, &sum, CURSEG_COLD_DATA); | |
4375a336 JK |
595 | fio.rw = WRITE_SYNC; |
596 | f2fs_submit_page_mbio(&fio); | |
597 | ||
548aedac | 598 | dn.data_blkaddr = fio.blk_addr; |
4375a336 JK |
599 | set_data_blkaddr(&dn); |
600 | f2fs_update_extent_cache(&dn); | |
601 | set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE); | |
602 | if (page->index == 0) | |
603 | set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN); | |
548aedac | 604 | put_page_out: |
4375a336 JK |
605 | f2fs_put_page(fio.encrypted_page, 1); |
606 | put_out: | |
607 | f2fs_put_dnode(&dn); | |
608 | out: | |
609 | f2fs_put_page(page, 1); | |
610 | } | |
611 | ||
c879f90d | 612 | static void move_data_page(struct inode *inode, block_t bidx, int gc_type) |
7bc09003 | 613 | { |
c879f90d JK |
614 | struct page *page; |
615 | ||
616 | page = get_lock_data_page(inode, bidx); | |
617 | if (IS_ERR(page)) | |
618 | return; | |
63a0b7cb | 619 | |
7bc09003 | 620 | if (gc_type == BG_GC) { |
4ebefc44 JK |
621 | if (PageWriteback(page)) |
622 | goto out; | |
7bc09003 JK |
623 | set_page_dirty(page); |
624 | set_cold_data(page); | |
625 | } else { | |
c879f90d JK |
626 | struct f2fs_io_info fio = { |
627 | .sbi = F2FS_I_SB(inode), | |
628 | .type = DATA, | |
629 | .rw = WRITE_SYNC, | |
630 | .page = page, | |
4375a336 | 631 | .encrypted_page = NULL, |
c879f90d | 632 | }; |
6282adbf | 633 | set_page_dirty(page); |
5514f0aa | 634 | f2fs_wait_on_page_writeback(page, DATA); |
1fe54f9d | 635 | if (clear_page_dirty_for_io(page)) |
a7ffdbe2 | 636 | inode_dec_dirty_pages(inode); |
7bc09003 | 637 | set_cold_data(page); |
05ca3632 | 638 | do_write_data_page(&fio); |
7bc09003 JK |
639 | clear_cold_data(page); |
640 | } | |
641 | out: | |
642 | f2fs_put_page(page, 1); | |
643 | } | |
644 | ||
0a8165d7 | 645 | /* |
7bc09003 JK |
646 | * This function tries to get parent node of victim data block, and identifies |
647 | * data block validity. If the block is valid, copy that with cold status and | |
648 | * modify parent node. | |
649 | * If the parent node is not valid or the data block address is different, | |
650 | * the victim data block is ignored. | |
651 | */ | |
5ee5293c | 652 | static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dda2af8 | 653 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type) |
7bc09003 JK |
654 | { |
655 | struct super_block *sb = sbi->sb; | |
656 | struct f2fs_summary *entry; | |
657 | block_t start_addr; | |
43727527 | 658 | int off; |
7bc09003 JK |
659 | int phase = 0; |
660 | ||
661 | start_addr = START_BLOCK(sbi, segno); | |
662 | ||
663 | next_step: | |
664 | entry = sum; | |
c718379b | 665 | |
7bc09003 JK |
666 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
667 | struct page *data_page; | |
668 | struct inode *inode; | |
669 | struct node_info dni; /* dnode info for the data */ | |
670 | unsigned int ofs_in_node, nofs; | |
671 | block_t start_bidx; | |
672 | ||
43727527 JK |
673 | /* stop BG_GC if there is not enough free sections. */ |
674 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
5ee5293c | 675 | return 0; |
7bc09003 | 676 | |
43727527 | 677 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
678 | continue; |
679 | ||
680 | if (phase == 0) { | |
681 | ra_node_page(sbi, le32_to_cpu(entry->nid)); | |
682 | continue; | |
683 | } | |
684 | ||
685 | /* Get an inode by ino with checking validity */ | |
c1079892 | 686 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
687 | continue; |
688 | ||
689 | if (phase == 1) { | |
690 | ra_node_page(sbi, dni.ino); | |
691 | continue; | |
692 | } | |
693 | ||
7bc09003 JK |
694 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
695 | ||
696 | if (phase == 2) { | |
d4686d56 | 697 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 698 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
699 | continue; |
700 | ||
4375a336 JK |
701 | /* if encrypted inode, let's go phase 3 */ |
702 | if (f2fs_encrypted_inode(inode) && | |
703 | S_ISREG(inode->i_mode)) { | |
704 | add_gc_inode(gc_list, inode); | |
705 | continue; | |
706 | } | |
707 | ||
de93653f | 708 | start_bidx = start_bidx_of_node(nofs, F2FS_I(inode)); |
43f3eae1 JK |
709 | data_page = get_read_data_page(inode, |
710 | start_bidx + ofs_in_node, READA); | |
31a32688 CL |
711 | if (IS_ERR(data_page)) { |
712 | iput(inode); | |
713 | continue; | |
714 | } | |
7bc09003 JK |
715 | |
716 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 717 | add_gc_inode(gc_list, inode); |
31a32688 CL |
718 | continue; |
719 | } | |
720 | ||
721 | /* phase 3 */ | |
7dda2af8 | 722 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 723 | if (inode) { |
c879f90d JK |
724 | start_bidx = start_bidx_of_node(nofs, F2FS_I(inode)) |
725 | + ofs_in_node; | |
4375a336 JK |
726 | if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) |
727 | move_encrypted_block(inode, start_bidx); | |
728 | else | |
729 | move_data_page(inode, start_bidx, gc_type); | |
e1235983 | 730 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 731 | } |
7bc09003 | 732 | } |
c718379b | 733 | |
7bc09003 JK |
734 | if (++phase < 4) |
735 | goto next_step; | |
43727527 | 736 | |
4ebefc44 | 737 | if (gc_type == FG_GC) { |
458e6197 | 738 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
4ebefc44 | 739 | |
5ee5293c JK |
740 | /* return 1 only if FG_GC succefully reclaimed one */ |
741 | if (get_valid_blocks(sbi, segno, 1) == 0) | |
742 | return 1; | |
4ebefc44 | 743 | } |
5ee5293c | 744 | return 0; |
7bc09003 JK |
745 | } |
746 | ||
747 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 748 | int gc_type) |
7bc09003 JK |
749 | { |
750 | struct sit_info *sit_i = SIT_I(sbi); | |
751 | int ret; | |
8a2d0ace | 752 | |
7bc09003 | 753 | mutex_lock(&sit_i->sentry_lock); |
8a2d0ace GZ |
754 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
755 | NO_CHECK_TYPE, LFS); | |
7bc09003 JK |
756 | mutex_unlock(&sit_i->sentry_lock); |
757 | return ret; | |
758 | } | |
759 | ||
5ee5293c | 760 | static int do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno, |
7dda2af8 | 761 | struct gc_inode_list *gc_list, int gc_type) |
7bc09003 JK |
762 | { |
763 | struct page *sum_page; | |
764 | struct f2fs_summary_block *sum; | |
c718379b | 765 | struct blk_plug plug; |
5ee5293c | 766 | int nfree = 0; |
7bc09003 JK |
767 | |
768 | /* read segment summary of victim */ | |
769 | sum_page = get_sum_page(sbi, segno); | |
7bc09003 | 770 | |
c718379b JK |
771 | blk_start_plug(&plug); |
772 | ||
7bc09003 JK |
773 | sum = page_address(sum_page); |
774 | ||
9236cac5 JK |
775 | /* |
776 | * this is to avoid deadlock: | |
777 | * - lock_page(sum_page) - f2fs_replace_block | |
778 | * - check_valid_map() - mutex_lock(sentry_lock) | |
779 | * - mutex_lock(sentry_lock) - change_curseg() | |
780 | * - lock_page(sum_page) | |
781 | */ | |
782 | unlock_page(sum_page); | |
783 | ||
7bc09003 JK |
784 | switch (GET_SUM_TYPE((&sum->footer))) { |
785 | case SUM_TYPE_NODE: | |
5ee5293c | 786 | nfree = gc_node_segment(sbi, sum->entries, segno, gc_type); |
7bc09003 JK |
787 | break; |
788 | case SUM_TYPE_DATA: | |
5ee5293c JK |
789 | nfree = gc_data_segment(sbi, sum->entries, gc_list, |
790 | segno, gc_type); | |
7bc09003 JK |
791 | break; |
792 | } | |
c718379b JK |
793 | blk_finish_plug(&plug); |
794 | ||
e1235983 | 795 | stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)), gc_type); |
7bc09003 JK |
796 | stat_inc_call_count(sbi->stat_info); |
797 | ||
9236cac5 | 798 | f2fs_put_page(sum_page, 0); |
5ee5293c | 799 | return nfree; |
7bc09003 JK |
800 | } |
801 | ||
408e9375 | 802 | int f2fs_gc(struct f2fs_sb_info *sbi) |
7bc09003 | 803 | { |
ab126cfc | 804 | unsigned int segno, i; |
7bc09003 | 805 | int gc_type = BG_GC; |
45fe8492 | 806 | int sec_freed = 0; |
43727527 | 807 | int ret = -1; |
d5053a34 | 808 | struct cp_control cpc; |
7dda2af8 CL |
809 | struct gc_inode_list gc_list = { |
810 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
769ec6e5 | 811 | .iroot = RADIX_TREE_INIT(GFP_NOFS), |
7dda2af8 | 812 | }; |
d5053a34 | 813 | |
119ee914 | 814 | cpc.reason = __get_cp_reason(sbi); |
7bc09003 | 815 | gc_more: |
ab126cfc JK |
816 | segno = NULL_SEGNO; |
817 | ||
6bacf52f | 818 | if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) |
408e9375 | 819 | goto stop; |
1e968fdf | 820 | if (unlikely(f2fs_cp_error(sbi))) |
203681f6 | 821 | goto stop; |
7bc09003 | 822 | |
45fe8492 | 823 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, sec_freed)) { |
408e9375 | 824 | gc_type = FG_GC; |
798c1b16 JK |
825 | if (__get_victim(sbi, &segno, gc_type) || prefree_segments(sbi)) |
826 | write_checkpoint(sbi, &cpc); | |
d64f8047 | 827 | } |
7bc09003 | 828 | |
798c1b16 | 829 | if (segno == NULL_SEGNO && !__get_victim(sbi, &segno, gc_type)) |
408e9375 | 830 | goto stop; |
43727527 | 831 | ret = 0; |
7bc09003 | 832 | |
81c1a0f1 CY |
833 | /* readahead multi ssa blocks those have contiguous address */ |
834 | if (sbi->segs_per_sec > 1) | |
835 | ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), sbi->segs_per_sec, | |
836 | META_SSA); | |
837 | ||
45fe8492 CY |
838 | for (i = 0; i < sbi->segs_per_sec; i++) { |
839 | /* | |
840 | * for FG_GC case, halt gcing left segments once failed one | |
841 | * of segments in selected section to avoid long latency. | |
842 | */ | |
843 | if (!do_garbage_collect(sbi, segno + i, &gc_list, gc_type) && | |
844 | gc_type == FG_GC) | |
845 | break; | |
846 | } | |
847 | ||
848 | if (i == sbi->segs_per_sec && gc_type == FG_GC) | |
849 | sec_freed++; | |
43727527 | 850 | |
5ee5293c | 851 | if (gc_type == FG_GC) |
5ec4e49f | 852 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 853 | |
45fe8492 | 854 | if (has_not_enough_free_secs(sbi, sec_freed)) |
43727527 JK |
855 | goto gc_more; |
856 | ||
857 | if (gc_type == FG_GC) | |
75ab4cb8 | 858 | write_checkpoint(sbi, &cpc); |
408e9375 | 859 | stop: |
7bc09003 JK |
860 | mutex_unlock(&sbi->gc_mutex); |
861 | ||
7dda2af8 | 862 | put_gc_inode(&gc_list); |
43727527 | 863 | return ret; |
7bc09003 JK |
864 | } |
865 | ||
866 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
867 | { | |
868 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
869 | } |