Commit | Line | Data |
---|---|---|
7c1a000d | 1 | // SPDX-License-Identifier: GPL-2.0 |
0a8165d7 | 2 | /* |
7bc09003 JK |
3 | * fs/f2fs/gc.c |
4 | * | |
5 | * Copyright (c) 2012 Samsung Electronics Co., Ltd. | |
6 | * http://www.samsung.com/ | |
7bc09003 JK |
7 | */ |
8 | #include <linux/fs.h> | |
9 | #include <linux/module.h> | |
7bc09003 JK |
10 | #include <linux/init.h> |
11 | #include <linux/f2fs_fs.h> | |
12 | #include <linux/kthread.h> | |
13 | #include <linux/delay.h> | |
14 | #include <linux/freezer.h> | |
b4b10061 | 15 | #include <linux/sched/signal.h> |
6691d940 | 16 | #include <linux/random.h> |
4034247a | 17 | #include <linux/sched/mm.h> |
7bc09003 JK |
18 | |
19 | #include "f2fs.h" | |
20 | #include "node.h" | |
21 | #include "segment.h" | |
22 | #include "gc.h" | |
52118743 | 23 | #include "iostat.h" |
8e46b3ed | 24 | #include <trace/events/f2fs.h> |
7bc09003 | 25 | |
093749e2 CY |
26 | static struct kmem_cache *victim_entry_slab; |
27 | ||
da52f8ad JQ |
28 | static unsigned int count_bits(const unsigned long *addr, |
29 | unsigned int offset, unsigned int len); | |
30 | ||
7bc09003 JK |
31 | static int gc_thread_func(void *data) |
32 | { | |
33 | struct f2fs_sb_info *sbi = data; | |
b59d0bae | 34 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; |
7bc09003 | 35 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; |
5911d2d1 | 36 | wait_queue_head_t *fggc_wq = &sbi->gc_thread->fggc_wq; |
b8c502b8 | 37 | unsigned int wait_ms; |
d147ea4a JK |
38 | struct f2fs_gc_control gc_control = { |
39 | .victim_segno = NULL_SEGNO, | |
c58d7c55 JK |
40 | .should_migrate_blocks = false, |
41 | .err_gc_skipped = false }; | |
7bc09003 | 42 | |
b59d0bae | 43 | wait_ms = gc_th->min_sleep_time; |
7bc09003 | 44 | |
1d7be270 | 45 | set_freezable(); |
7bc09003 | 46 | do { |
5911d2d1 | 47 | bool sync_mode, foreground = false; |
bbbc34fd | 48 | |
1d7be270 | 49 | wait_event_interruptible_timeout(*wq, |
d9872a69 | 50 | kthread_should_stop() || freezing(current) || |
5911d2d1 | 51 | waitqueue_active(fggc_wq) || |
d9872a69 | 52 | gc_th->gc_wake, |
1d7be270 JK |
53 | msecs_to_jiffies(wait_ms)); |
54 | ||
5911d2d1 CY |
55 | if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq)) |
56 | foreground = true; | |
57 | ||
d9872a69 JK |
58 | /* give it a try one time */ |
59 | if (gc_th->gc_wake) | |
60 | gc_th->gc_wake = 0; | |
61 | ||
274bd9ba CY |
62 | if (try_to_freeze()) { |
63 | stat_other_skip_bggc_count(sbi); | |
7bc09003 | 64 | continue; |
274bd9ba | 65 | } |
7bc09003 JK |
66 | if (kthread_should_stop()) |
67 | break; | |
68 | ||
d6212a5f | 69 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
88dd8934 | 70 | increase_sleep_time(gc_th, &wait_ms); |
274bd9ba | 71 | stat_other_skip_bggc_count(sbi); |
d6212a5f CL |
72 | continue; |
73 | } | |
74 | ||
55523519 | 75 | if (time_to_inject(sbi, FAULT_CHECKPOINT)) { |
c45d6002 | 76 | f2fs_show_injection_info(sbi, FAULT_CHECKPOINT); |
a9cfee0e CY |
77 | f2fs_stop_checkpoint(sbi, false, |
78 | STOP_CP_REASON_FAULT_INJECT); | |
55523519 | 79 | } |
0f348028 | 80 | |
274bd9ba CY |
81 | if (!sb_start_write_trylock(sbi->sb)) { |
82 | stat_other_skip_bggc_count(sbi); | |
dc6febb6 | 83 | continue; |
274bd9ba | 84 | } |
dc6febb6 | 85 | |
7bc09003 JK |
86 | /* |
87 | * [GC triggering condition] | |
88 | * 0. GC is not conducted currently. | |
89 | * 1. There are enough dirty segments. | |
90 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
91 | * 3. IO subsystem is idle by checking the # of requests in | |
92 | * bdev's request list. | |
93 | * | |
e1c42045 | 94 | * Note) We have to avoid triggering GCs frequently. |
7bc09003 JK |
95 | * Because it is possible that some segments can be |
96 | * invalidated soon after by user update or deletion. | |
97 | * So, I'd like to wait some time to collect dirty segments. | |
98 | */ | |
d98af5f4 DJ |
99 | if (sbi->gc_mode == GC_URGENT_HIGH || |
100 | sbi->gc_mode == GC_URGENT_MID) { | |
d9872a69 | 101 | wait_ms = gc_th->urgent_sleep_time; |
e4544b63 | 102 | f2fs_down_write(&sbi->gc_lock); |
d9872a69 JK |
103 | goto do_gc; |
104 | } | |
105 | ||
5911d2d1 | 106 | if (foreground) { |
e4544b63 | 107 | f2fs_down_write(&sbi->gc_lock); |
5911d2d1 | 108 | goto do_gc; |
e4544b63 | 109 | } else if (!f2fs_down_write_trylock(&sbi->gc_lock)) { |
274bd9ba | 110 | stat_other_skip_bggc_count(sbi); |
69babac0 | 111 | goto next; |
274bd9ba | 112 | } |
69babac0 | 113 | |
a7d10cf3 | 114 | if (!is_idle(sbi, GC_TIME)) { |
88dd8934 | 115 | increase_sleep_time(gc_th, &wait_ms); |
e4544b63 | 116 | f2fs_up_write(&sbi->gc_lock); |
274bd9ba | 117 | stat_io_skip_bggc_count(sbi); |
dc6febb6 | 118 | goto next; |
7bc09003 JK |
119 | } |
120 | ||
121 | if (has_enough_invalid_blocks(sbi)) | |
88dd8934 | 122 | decrease_sleep_time(gc_th, &wait_ms); |
7bc09003 | 123 | else |
88dd8934 | 124 | increase_sleep_time(gc_th, &wait_ms); |
d9872a69 | 125 | do_gc: |
5911d2d1 CY |
126 | if (!foreground) |
127 | stat_inc_bggc_count(sbi->stat_info); | |
7bc09003 | 128 | |
bbbc34fd CY |
129 | sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC; |
130 | ||
5911d2d1 CY |
131 | /* foreground GC was been triggered via f2fs_balance_fs() */ |
132 | if (foreground) | |
133 | sync_mode = false; | |
134 | ||
d147ea4a JK |
135 | gc_control.init_gc_type = sync_mode ? FG_GC : BG_GC; |
136 | gc_control.no_bg_gc = foreground; | |
c81d5bae | 137 | gc_control.nr_free_secs = foreground ? 1 : 0; |
d147ea4a | 138 | |
43727527 | 139 | /* if return value is not zero, no victim was selected */ |
1adaa71e | 140 | if (f2fs_gc(sbi, &gc_control)) { |
141 | /* don't bother wait_ms by foreground gc */ | |
142 | if (!foreground) | |
143 | wait_ms = gc_th->no_gc_sleep_time; | |
144 | } | |
81eb8d6e | 145 | |
5911d2d1 CY |
146 | if (foreground) |
147 | wake_up_all(&gc_th->fggc_wq); | |
148 | ||
84e4214f JK |
149 | trace_f2fs_background_gc(sbi->sb, wait_ms, |
150 | prefree_segments(sbi), free_segments(sbi)); | |
151 | ||
4660f9c0 | 152 | /* balancing f2fs's metadata periodically */ |
7bcd0cfa | 153 | f2fs_balance_fs_bg(sbi, true); |
dc6febb6 | 154 | next: |
e5a0db6a YL |
155 | if (sbi->gc_mode != GC_NORMAL) { |
156 | spin_lock(&sbi->gc_remaining_trials_lock); | |
157 | if (sbi->gc_remaining_trials) { | |
158 | sbi->gc_remaining_trials--; | |
159 | if (!sbi->gc_remaining_trials) | |
6359a1aa YL |
160 | sbi->gc_mode = GC_NORMAL; |
161 | } | |
e5a0db6a | 162 | spin_unlock(&sbi->gc_remaining_trials_lock); |
6359a1aa | 163 | } |
dc6febb6 | 164 | sb_end_write(sbi->sb); |
81eb8d6e | 165 | |
7bc09003 JK |
166 | } while (!kthread_should_stop()); |
167 | return 0; | |
168 | } | |
169 | ||
4d57b86d | 170 | int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) |
7bc09003 | 171 | { |
1042d60f | 172 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 173 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7bc09003 | 174 | |
1ecc0c5c | 175 | gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
146dbcbf YL |
176 | if (!gc_th) |
177 | return -ENOMEM; | |
7bc09003 | 178 | |
d9872a69 | 179 | gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME; |
b59d0bae NJ |
180 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
181 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
182 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
183 | ||
5f029c04 | 184 | gc_th->gc_wake = 0; |
d2dc095f | 185 | |
7bc09003 JK |
186 | sbi->gc_thread = gc_th; |
187 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
5911d2d1 | 188 | init_waitqueue_head(&sbi->gc_thread->fggc_wq); |
7bc09003 | 189 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, |
ec7b1f2d | 190 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 191 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
146dbcbf YL |
192 | int err = PTR_ERR(gc_th->f2fs_gc_task); |
193 | ||
c8eb7024 | 194 | kfree(gc_th); |
25718423 | 195 | sbi->gc_thread = NULL; |
146dbcbf | 196 | return err; |
7bc09003 | 197 | } |
146dbcbf YL |
198 | |
199 | return 0; | |
7bc09003 JK |
200 | } |
201 | ||
4d57b86d | 202 | void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) |
7bc09003 JK |
203 | { |
204 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
5f029c04 | 205 | |
7bc09003 JK |
206 | if (!gc_th) |
207 | return; | |
208 | kthread_stop(gc_th->f2fs_gc_task); | |
5911d2d1 | 209 | wake_up_all(&gc_th->fggc_wq); |
c8eb7024 | 210 | kfree(gc_th); |
7bc09003 JK |
211 | sbi->gc_thread = NULL; |
212 | } | |
213 | ||
5b0e9539 | 214 | static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) |
7bc09003 | 215 | { |
093749e2 CY |
216 | int gc_mode; |
217 | ||
218 | if (gc_type == BG_GC) { | |
219 | if (sbi->am.atgc_enabled) | |
220 | gc_mode = GC_AT; | |
221 | else | |
222 | gc_mode = GC_CB; | |
223 | } else { | |
224 | gc_mode = GC_GREEDY; | |
225 | } | |
d2dc095f | 226 | |
5b0e9539 JK |
227 | switch (sbi->gc_mode) { |
228 | case GC_IDLE_CB: | |
229 | gc_mode = GC_CB; | |
230 | break; | |
231 | case GC_IDLE_GREEDY: | |
0e5e8111 | 232 | case GC_URGENT_HIGH: |
b27bc809 | 233 | gc_mode = GC_GREEDY; |
5b0e9539 | 234 | break; |
093749e2 CY |
235 | case GC_IDLE_AT: |
236 | gc_mode = GC_AT; | |
237 | break; | |
5b0e9539 | 238 | } |
093749e2 | 239 | |
d2dc095f | 240 | return gc_mode; |
7bc09003 JK |
241 | } |
242 | ||
243 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
244 | int type, struct victim_sel_policy *p) | |
245 | { | |
246 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
247 | ||
4ebefc44 | 248 | if (p->alloc_mode == SSR) { |
7bc09003 | 249 | p->gc_mode = GC_GREEDY; |
da52f8ad | 250 | p->dirty_bitmap = dirty_i->dirty_segmap[type]; |
a26b7c8a | 251 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 | 252 | p->ofs_unit = 1; |
093749e2 CY |
253 | } else if (p->alloc_mode == AT_SSR) { |
254 | p->gc_mode = GC_GREEDY; | |
255 | p->dirty_bitmap = dirty_i->dirty_segmap[type]; | |
256 | p->max_search = dirty_i->nr_dirty[type]; | |
257 | p->ofs_unit = 1; | |
7bc09003 | 258 | } else { |
5b0e9539 | 259 | p->gc_mode = select_gc_type(sbi, gc_type); |
7bc09003 | 260 | p->ofs_unit = sbi->segs_per_sec; |
da52f8ad JQ |
261 | if (__is_large_section(sbi)) { |
262 | p->dirty_bitmap = dirty_i->dirty_secmap; | |
263 | p->max_search = count_bits(p->dirty_bitmap, | |
264 | 0, MAIN_SECS(sbi)); | |
265 | } else { | |
266 | p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY]; | |
267 | p->max_search = dirty_i->nr_dirty[DIRTY]; | |
268 | } | |
7bc09003 | 269 | } |
a26b7c8a | 270 | |
7a88ddb5 CY |
271 | /* |
272 | * adjust candidates range, should select all dirty segments for | |
273 | * foreground GC and urgent GC cases. | |
274 | */ | |
b27bc809 | 275 | if (gc_type != FG_GC && |
0e5e8111 | 276 | (sbi->gc_mode != GC_URGENT_HIGH) && |
093749e2 | 277 | (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) && |
b27bc809 | 278 | p->max_search > sbi->max_victim_search) |
b1c57c1c | 279 | p->max_search = sbi->max_victim_search; |
a26b7c8a | 280 | |
b94929d9 | 281 | /* let's select beginning hot/small space first in no_heap mode*/ |
6691d940 | 282 | if (f2fs_need_rand_seg(sbi)) |
81895a65 | 283 | p->offset = prandom_u32_max(MAIN_SECS(sbi) * sbi->segs_per_sec); |
6691d940 | 284 | else if (test_opt(sbi, NOHEAP) && |
b94929d9 | 285 | (type == CURSEG_HOT_DATA || IS_NODESEG(type))) |
7a20b8a6 JK |
286 | p->offset = 0; |
287 | else | |
e066b83c | 288 | p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; |
7bc09003 JK |
289 | } |
290 | ||
291 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
292 | struct victim_sel_policy *p) | |
293 | { | |
b7250d2d JK |
294 | /* SSR allocates in a segment unit */ |
295 | if (p->alloc_mode == SSR) | |
3519e3f9 | 296 | return sbi->blocks_per_seg; |
093749e2 CY |
297 | else if (p->alloc_mode == AT_SSR) |
298 | return UINT_MAX; | |
299 | ||
300 | /* LFS */ | |
7bc09003 | 301 | if (p->gc_mode == GC_GREEDY) |
c541a51b | 302 | return 2 * sbi->blocks_per_seg * p->ofs_unit; |
7bc09003 JK |
303 | else if (p->gc_mode == GC_CB) |
304 | return UINT_MAX; | |
093749e2 CY |
305 | else if (p->gc_mode == GC_AT) |
306 | return UINT_MAX; | |
7bc09003 JK |
307 | else /* No other gc_mode */ |
308 | return 0; | |
309 | } | |
310 | ||
311 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
312 | { | |
313 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 314 | unsigned int secno; |
7bc09003 JK |
315 | |
316 | /* | |
317 | * If the gc_type is FG_GC, we can select victim segments | |
318 | * selected by background GC before. | |
319 | * Those segments guarantee they have small valid blocks. | |
320 | */ | |
7cd8558b | 321 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 322 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 323 | continue; |
5ec4e49f | 324 | clear_bit(secno, dirty_i->victim_secmap); |
4ddb1a4d | 325 | return GET_SEG_FROM_SEC(sbi, secno); |
7bc09003 JK |
326 | } |
327 | return NULL_SEGNO; | |
328 | } | |
329 | ||
330 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
331 | { | |
332 | struct sit_info *sit_i = SIT_I(sbi); | |
4ddb1a4d JK |
333 | unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); |
334 | unsigned int start = GET_SEG_FROM_SEC(sbi, secno); | |
7bc09003 JK |
335 | unsigned long long mtime = 0; |
336 | unsigned int vblocks; | |
337 | unsigned char age = 0; | |
338 | unsigned char u; | |
339 | unsigned int i; | |
de881df9 | 340 | unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno); |
7bc09003 | 341 | |
de881df9 | 342 | for (i = 0; i < usable_segs_per_sec; i++) |
7bc09003 | 343 | mtime += get_seg_entry(sbi, start + i)->mtime; |
302bd348 | 344 | vblocks = get_valid_blocks(sbi, segno, true); |
7bc09003 | 345 | |
de881df9 AR |
346 | mtime = div_u64(mtime, usable_segs_per_sec); |
347 | vblocks = div_u64(vblocks, usable_segs_per_sec); | |
7bc09003 JK |
348 | |
349 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
350 | ||
e1c42045 | 351 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
352 | if (mtime < sit_i->min_mtime) |
353 | sit_i->min_mtime = mtime; | |
354 | if (mtime > sit_i->max_mtime) | |
355 | sit_i->max_mtime = mtime; | |
356 | if (sit_i->max_mtime != sit_i->min_mtime) | |
357 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
358 | sit_i->max_mtime - sit_i->min_mtime); | |
359 | ||
360 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
361 | } | |
362 | ||
a57e564d JX |
363 | static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, |
364 | unsigned int segno, struct victim_sel_policy *p) | |
7bc09003 JK |
365 | { |
366 | if (p->alloc_mode == SSR) | |
2afce76a | 367 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; |
7bc09003 JK |
368 | |
369 | /* alloc_mode == LFS */ | |
370 | if (p->gc_mode == GC_GREEDY) | |
91f4382b | 371 | return get_valid_blocks(sbi, segno, true); |
093749e2 | 372 | else if (p->gc_mode == GC_CB) |
7bc09003 | 373 | return get_cb_cost(sbi, segno); |
093749e2 CY |
374 | |
375 | f2fs_bug_on(sbi, 1); | |
376 | return 0; | |
7bc09003 JK |
377 | } |
378 | ||
688159b6 FL |
379 | static unsigned int count_bits(const unsigned long *addr, |
380 | unsigned int offset, unsigned int len) | |
381 | { | |
382 | unsigned int end = offset + len, sum = 0; | |
383 | ||
384 | while (offset < end) { | |
385 | if (test_bit(offset++, addr)) | |
386 | ++sum; | |
387 | } | |
388 | return sum; | |
389 | } | |
390 | ||
093749e2 CY |
391 | static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi, |
392 | unsigned long long mtime, unsigned int segno, | |
393 | struct rb_node *parent, struct rb_node **p, | |
394 | bool left_most) | |
395 | { | |
396 | struct atgc_management *am = &sbi->am; | |
397 | struct victim_entry *ve; | |
398 | ||
32410577 CY |
399 | ve = f2fs_kmem_cache_alloc(victim_entry_slab, |
400 | GFP_NOFS, true, NULL); | |
093749e2 CY |
401 | |
402 | ve->mtime = mtime; | |
403 | ve->segno = segno; | |
404 | ||
405 | rb_link_node(&ve->rb_node, parent, p); | |
406 | rb_insert_color_cached(&ve->rb_node, &am->root, left_most); | |
407 | ||
408 | list_add_tail(&ve->list, &am->victim_list); | |
409 | ||
410 | am->victim_count++; | |
411 | ||
412 | return ve; | |
413 | } | |
414 | ||
415 | static void insert_victim_entry(struct f2fs_sb_info *sbi, | |
416 | unsigned long long mtime, unsigned int segno) | |
417 | { | |
418 | struct atgc_management *am = &sbi->am; | |
419 | struct rb_node **p; | |
420 | struct rb_node *parent = NULL; | |
421 | bool left_most = true; | |
422 | ||
423 | p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most); | |
424 | attach_victim_entry(sbi, mtime, segno, parent, p, left_most); | |
425 | } | |
426 | ||
427 | static void add_victim_entry(struct f2fs_sb_info *sbi, | |
428 | struct victim_sel_policy *p, unsigned int segno) | |
429 | { | |
430 | struct sit_info *sit_i = SIT_I(sbi); | |
431 | unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); | |
432 | unsigned int start = GET_SEG_FROM_SEC(sbi, secno); | |
433 | unsigned long long mtime = 0; | |
434 | unsigned int i; | |
435 | ||
436 | if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { | |
437 | if (p->gc_mode == GC_AT && | |
438 | get_valid_blocks(sbi, segno, true) == 0) | |
439 | return; | |
093749e2 CY |
440 | } |
441 | ||
442 | for (i = 0; i < sbi->segs_per_sec; i++) | |
443 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
444 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
445 | ||
446 | /* Handle if the system time has changed by the user */ | |
447 | if (mtime < sit_i->min_mtime) | |
448 | sit_i->min_mtime = mtime; | |
449 | if (mtime > sit_i->max_mtime) | |
450 | sit_i->max_mtime = mtime; | |
451 | if (mtime < sit_i->dirty_min_mtime) | |
452 | sit_i->dirty_min_mtime = mtime; | |
453 | if (mtime > sit_i->dirty_max_mtime) | |
454 | sit_i->dirty_max_mtime = mtime; | |
455 | ||
456 | /* don't choose young section as candidate */ | |
457 | if (sit_i->dirty_max_mtime - mtime < p->age_threshold) | |
458 | return; | |
459 | ||
460 | insert_victim_entry(sbi, mtime, segno); | |
461 | } | |
462 | ||
463 | static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi, | |
464 | struct victim_sel_policy *p) | |
465 | { | |
466 | struct atgc_management *am = &sbi->am; | |
467 | struct rb_node *parent = NULL; | |
468 | bool left_most; | |
469 | ||
470 | f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most); | |
471 | ||
472 | return parent; | |
473 | } | |
474 | ||
475 | static void atgc_lookup_victim(struct f2fs_sb_info *sbi, | |
476 | struct victim_sel_policy *p) | |
477 | { | |
478 | struct sit_info *sit_i = SIT_I(sbi); | |
479 | struct atgc_management *am = &sbi->am; | |
480 | struct rb_root_cached *root = &am->root; | |
481 | struct rb_node *node; | |
482 | struct rb_entry *re; | |
483 | struct victim_entry *ve; | |
484 | unsigned long long total_time; | |
485 | unsigned long long age, u, accu; | |
486 | unsigned long long max_mtime = sit_i->dirty_max_mtime; | |
487 | unsigned long long min_mtime = sit_i->dirty_min_mtime; | |
074b5ea2 | 488 | unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi); |
093749e2 CY |
489 | unsigned int vblocks; |
490 | unsigned int dirty_threshold = max(am->max_candidate_count, | |
491 | am->candidate_ratio * | |
492 | am->victim_count / 100); | |
493 | unsigned int age_weight = am->age_weight; | |
494 | unsigned int cost; | |
495 | unsigned int iter = 0; | |
496 | ||
497 | if (max_mtime < min_mtime) | |
498 | return; | |
499 | ||
500 | max_mtime += 1; | |
501 | total_time = max_mtime - min_mtime; | |
502 | ||
503 | accu = div64_u64(ULLONG_MAX, total_time); | |
504 | accu = min_t(unsigned long long, div_u64(accu, 100), | |
505 | DEFAULT_ACCURACY_CLASS); | |
506 | ||
507 | node = rb_first_cached(root); | |
508 | next: | |
509 | re = rb_entry_safe(node, struct rb_entry, rb_node); | |
510 | if (!re) | |
511 | return; | |
512 | ||
513 | ve = (struct victim_entry *)re; | |
514 | ||
515 | if (ve->mtime >= max_mtime || ve->mtime < min_mtime) | |
516 | goto skip; | |
517 | ||
518 | /* age = 10000 * x% * 60 */ | |
519 | age = div64_u64(accu * (max_mtime - ve->mtime), total_time) * | |
520 | age_weight; | |
521 | ||
522 | vblocks = get_valid_blocks(sbi, ve->segno, true); | |
523 | f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks); | |
524 | ||
525 | /* u = 10000 * x% * 40 */ | |
526 | u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) * | |
527 | (100 - age_weight); | |
528 | ||
529 | f2fs_bug_on(sbi, age + u >= UINT_MAX); | |
530 | ||
531 | cost = UINT_MAX - (age + u); | |
532 | iter++; | |
533 | ||
534 | if (cost < p->min_cost || | |
535 | (cost == p->min_cost && age > p->oldest_age)) { | |
536 | p->min_cost = cost; | |
537 | p->oldest_age = age; | |
538 | p->min_segno = ve->segno; | |
539 | } | |
540 | skip: | |
541 | if (iter < dirty_threshold) { | |
542 | node = rb_next(node); | |
543 | goto next; | |
544 | } | |
545 | } | |
546 | ||
547 | /* | |
548 | * select candidates around source section in range of | |
549 | * [target - dirty_threshold, target + dirty_threshold] | |
550 | */ | |
551 | static void atssr_lookup_victim(struct f2fs_sb_info *sbi, | |
552 | struct victim_sel_policy *p) | |
553 | { | |
554 | struct sit_info *sit_i = SIT_I(sbi); | |
555 | struct atgc_management *am = &sbi->am; | |
556 | struct rb_node *node; | |
557 | struct rb_entry *re; | |
558 | struct victim_entry *ve; | |
559 | unsigned long long age; | |
560 | unsigned long long max_mtime = sit_i->dirty_max_mtime; | |
561 | unsigned long long min_mtime = sit_i->dirty_min_mtime; | |
562 | unsigned int seg_blocks = sbi->blocks_per_seg; | |
563 | unsigned int vblocks; | |
564 | unsigned int dirty_threshold = max(am->max_candidate_count, | |
565 | am->candidate_ratio * | |
566 | am->victim_count / 100); | |
567 | unsigned int cost; | |
568 | unsigned int iter = 0; | |
569 | int stage = 0; | |
570 | ||
571 | if (max_mtime < min_mtime) | |
572 | return; | |
573 | max_mtime += 1; | |
574 | next_stage: | |
575 | node = lookup_central_victim(sbi, p); | |
576 | next_node: | |
577 | re = rb_entry_safe(node, struct rb_entry, rb_node); | |
578 | if (!re) { | |
579 | if (stage == 0) | |
580 | goto skip_stage; | |
581 | return; | |
582 | } | |
583 | ||
584 | ve = (struct victim_entry *)re; | |
585 | ||
586 | if (ve->mtime >= max_mtime || ve->mtime < min_mtime) | |
587 | goto skip_node; | |
588 | ||
589 | age = max_mtime - ve->mtime; | |
590 | ||
591 | vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks; | |
592 | f2fs_bug_on(sbi, !vblocks); | |
593 | ||
594 | /* rare case */ | |
595 | if (vblocks == seg_blocks) | |
596 | goto skip_node; | |
597 | ||
598 | iter++; | |
599 | ||
600 | age = max_mtime - abs(p->age - age); | |
601 | cost = UINT_MAX - vblocks; | |
602 | ||
603 | if (cost < p->min_cost || | |
604 | (cost == p->min_cost && age > p->oldest_age)) { | |
605 | p->min_cost = cost; | |
606 | p->oldest_age = age; | |
607 | p->min_segno = ve->segno; | |
608 | } | |
609 | skip_node: | |
610 | if (iter < dirty_threshold) { | |
611 | if (stage == 0) | |
612 | node = rb_prev(node); | |
613 | else if (stage == 1) | |
614 | node = rb_next(node); | |
615 | goto next_node; | |
616 | } | |
617 | skip_stage: | |
618 | if (stage < 1) { | |
619 | stage++; | |
620 | iter = 0; | |
621 | goto next_stage; | |
622 | } | |
623 | } | |
624 | static void lookup_victim_by_age(struct f2fs_sb_info *sbi, | |
625 | struct victim_sel_policy *p) | |
626 | { | |
627 | f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, | |
628 | &sbi->am.root, true)); | |
629 | ||
630 | if (p->gc_mode == GC_AT) | |
631 | atgc_lookup_victim(sbi, p); | |
632 | else if (p->alloc_mode == AT_SSR) | |
633 | atssr_lookup_victim(sbi, p); | |
634 | else | |
635 | f2fs_bug_on(sbi, 1); | |
636 | } | |
637 | ||
638 | static void release_victim_entry(struct f2fs_sb_info *sbi) | |
639 | { | |
640 | struct atgc_management *am = &sbi->am; | |
641 | struct victim_entry *ve, *tmp; | |
642 | ||
643 | list_for_each_entry_safe(ve, tmp, &am->victim_list, list) { | |
644 | list_del(&ve->list); | |
645 | kmem_cache_free(victim_entry_slab, ve); | |
646 | am->victim_count--; | |
647 | } | |
648 | ||
649 | am->root = RB_ROOT_CACHED; | |
650 | ||
651 | f2fs_bug_on(sbi, am->victim_count); | |
652 | f2fs_bug_on(sbi, !list_empty(&am->victim_list)); | |
653 | } | |
654 | ||
71419129 CY |
655 | static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno) |
656 | { | |
657 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
658 | unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); | |
659 | ||
660 | if (!dirty_i->enable_pin_section) | |
661 | return false; | |
662 | if (!test_and_set_bit(secno, dirty_i->pinned_secmap)) | |
663 | dirty_i->pinned_secmap_cnt++; | |
664 | return true; | |
665 | } | |
666 | ||
667 | static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i) | |
668 | { | |
669 | return dirty_i->pinned_secmap_cnt; | |
670 | } | |
671 | ||
672 | static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i, | |
673 | unsigned int secno) | |
674 | { | |
675 | return dirty_i->enable_pin_section && | |
676 | f2fs_pinned_section_exists(dirty_i) && | |
677 | test_bit(secno, dirty_i->pinned_secmap); | |
678 | } | |
679 | ||
680 | static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable) | |
681 | { | |
682 | unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); | |
683 | ||
684 | if (f2fs_pinned_section_exists(DIRTY_I(sbi))) { | |
685 | memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size); | |
686 | DIRTY_I(sbi)->pinned_secmap_cnt = 0; | |
687 | } | |
688 | DIRTY_I(sbi)->enable_pin_section = enable; | |
689 | } | |
690 | ||
691 | static int f2fs_gc_pinned_control(struct inode *inode, int gc_type, | |
692 | unsigned int segno) | |
693 | { | |
694 | if (!f2fs_is_pinned_file(inode)) | |
695 | return 0; | |
696 | if (gc_type != FG_GC) | |
697 | return -EBUSY; | |
698 | if (!f2fs_pin_section(F2FS_I_SB(inode), segno)) | |
699 | f2fs_pin_file_control(inode, true); | |
700 | return -EAGAIN; | |
701 | } | |
702 | ||
0a8165d7 | 703 | /* |
111d2495 | 704 | * This function is called from two paths. |
7bc09003 JK |
705 | * One is garbage collection and the other is SSR segment selection. |
706 | * When it is called during GC, it just gets a victim segment | |
707 | * and it does not remove it from dirty seglist. | |
708 | * When it is called from SSR segment selection, it finds a segment | |
709 | * which has minimum valid blocks and removes it from dirty seglist. | |
710 | */ | |
711 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
093749e2 CY |
712 | unsigned int *result, int gc_type, int type, |
713 | char alloc_mode, unsigned long long age) | |
7bc09003 JK |
714 | { |
715 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
e066b83c | 716 | struct sit_info *sm = SIT_I(sbi); |
7bc09003 | 717 | struct victim_sel_policy p; |
3fa56503 | 718 | unsigned int secno, last_victim; |
04f0b2ea | 719 | unsigned int last_segment; |
093749e2 CY |
720 | unsigned int nsearched; |
721 | bool is_atgc; | |
97767500 | 722 | int ret = 0; |
7bc09003 | 723 | |
210f41bc | 724 | mutex_lock(&dirty_i->seglist_lock); |
04f0b2ea | 725 | last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec; |
210f41bc | 726 | |
7bc09003 | 727 | p.alloc_mode = alloc_mode; |
093749e2 CY |
728 | p.age = age; |
729 | p.age_threshold = sbi->am.age_threshold; | |
7bc09003 | 730 | |
093749e2 CY |
731 | retry: |
732 | select_policy(sbi, gc_type, type, &p); | |
7bc09003 | 733 | p.min_segno = NULL_SEGNO; |
093749e2 | 734 | p.oldest_age = 0; |
3fa56503 | 735 | p.min_cost = get_max_cost(sbi, &p); |
7bc09003 | 736 | |
093749e2 CY |
737 | is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR); |
738 | nsearched = 0; | |
739 | ||
740 | if (is_atgc) | |
741 | SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX; | |
742 | ||
e066b83c | 743 | if (*result != NULL_SEGNO) { |
97767500 QZ |
744 | if (!get_valid_blocks(sbi, *result, false)) { |
745 | ret = -ENODATA; | |
746 | goto out; | |
747 | } | |
748 | ||
749 | if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) | |
750 | ret = -EBUSY; | |
751 | else | |
e066b83c JK |
752 | p.min_segno = *result; |
753 | goto out; | |
754 | } | |
755 | ||
97767500 | 756 | ret = -ENODATA; |
3342bb30 CY |
757 | if (p.max_search == 0) |
758 | goto out; | |
759 | ||
e3080b01 CY |
760 | if (__is_large_section(sbi) && p.alloc_mode == LFS) { |
761 | if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) { | |
762 | p.min_segno = sbi->next_victim_seg[BG_GC]; | |
763 | *result = p.min_segno; | |
764 | sbi->next_victim_seg[BG_GC] = NULL_SEGNO; | |
765 | goto got_result; | |
766 | } | |
767 | if (gc_type == FG_GC && | |
768 | sbi->next_victim_seg[FG_GC] != NULL_SEGNO) { | |
769 | p.min_segno = sbi->next_victim_seg[FG_GC]; | |
770 | *result = p.min_segno; | |
771 | sbi->next_victim_seg[FG_GC] = NULL_SEGNO; | |
772 | goto got_result; | |
773 | } | |
774 | } | |
775 | ||
e066b83c | 776 | last_victim = sm->last_victim[p.gc_mode]; |
7bc09003 JK |
777 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
778 | p.min_segno = check_bg_victims(sbi); | |
779 | if (p.min_segno != NULL_SEGNO) | |
780 | goto got_it; | |
781 | } | |
782 | ||
783 | while (1) { | |
da52f8ad JQ |
784 | unsigned long cost, *dirty_bitmap; |
785 | unsigned int unit_no, segno; | |
786 | ||
787 | dirty_bitmap = p.dirty_bitmap; | |
788 | unit_no = find_next_bit(dirty_bitmap, | |
789 | last_segment / p.ofs_unit, | |
790 | p.offset / p.ofs_unit); | |
791 | segno = unit_no * p.ofs_unit; | |
a43f7ec3 | 792 | if (segno >= last_segment) { |
e066b83c JK |
793 | if (sm->last_victim[p.gc_mode]) { |
794 | last_segment = | |
795 | sm->last_victim[p.gc_mode]; | |
796 | sm->last_victim[p.gc_mode] = 0; | |
7bc09003 JK |
797 | p.offset = 0; |
798 | continue; | |
799 | } | |
800 | break; | |
801 | } | |
a57e564d JX |
802 | |
803 | p.offset = segno + p.ofs_unit; | |
da52f8ad | 804 | nsearched++; |
688159b6 | 805 | |
bbf9f7d9 ST |
806 | #ifdef CONFIG_F2FS_CHECK_FS |
807 | /* | |
808 | * skip selecting the invalid segno (that is failed due to block | |
809 | * validity check failure during GC) to avoid endless GC loop in | |
810 | * such cases. | |
811 | */ | |
812 | if (test_bit(segno, sm->invalid_segmap)) | |
813 | goto next; | |
814 | #endif | |
815 | ||
4ddb1a4d | 816 | secno = GET_SEC_FROM_SEG(sbi, segno); |
7bc09003 | 817 | |
5ec4e49f | 818 | if (sec_usage_check(sbi, secno)) |
688159b6 | 819 | goto next; |
61461fc9 | 820 | |
4354994f | 821 | /* Don't touch checkpointed data */ |
61461fc9 CY |
822 | if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { |
823 | if (p.alloc_mode == LFS) { | |
824 | /* | |
825 | * LFS is set to find source section during GC. | |
826 | * The victim should have no checkpointed data. | |
827 | */ | |
828 | if (get_ckpt_valid_blocks(sbi, segno, true)) | |
829 | goto next; | |
830 | } else { | |
831 | /* | |
832 | * SSR | AT_SSR are set to find target segment | |
833 | * for writes which can be full by checkpointed | |
834 | * and newly written blocks. | |
835 | */ | |
836 | if (!f2fs_segment_has_free_slot(sbi, segno)) | |
837 | goto next; | |
838 | } | |
839 | } | |
840 | ||
5ec4e49f | 841 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
688159b6 | 842 | goto next; |
7bc09003 | 843 | |
71419129 CY |
844 | if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno)) |
845 | goto next; | |
846 | ||
093749e2 CY |
847 | if (is_atgc) { |
848 | add_victim_entry(sbi, &p, segno); | |
849 | goto next; | |
850 | } | |
851 | ||
7bc09003 JK |
852 | cost = get_gc_cost(sbi, segno, &p); |
853 | ||
854 | if (p.min_cost > cost) { | |
855 | p.min_segno = segno; | |
856 | p.min_cost = cost; | |
a57e564d | 857 | } |
688159b6 FL |
858 | next: |
859 | if (nsearched >= p.max_search) { | |
e066b83c | 860 | if (!sm->last_victim[p.gc_mode] && segno <= last_victim) |
da52f8ad JQ |
861 | sm->last_victim[p.gc_mode] = |
862 | last_victim + p.ofs_unit; | |
4ce53776 | 863 | else |
da52f8ad | 864 | sm->last_victim[p.gc_mode] = segno + p.ofs_unit; |
04f0b2ea QS |
865 | sm->last_victim[p.gc_mode] %= |
866 | (MAIN_SECS(sbi) * sbi->segs_per_sec); | |
7bc09003 JK |
867 | break; |
868 | } | |
869 | } | |
093749e2 CY |
870 | |
871 | /* get victim for GC_AT/AT_SSR */ | |
872 | if (is_atgc) { | |
873 | lookup_victim_by_age(sbi, &p); | |
874 | release_victim_entry(sbi); | |
875 | } | |
876 | ||
877 | if (is_atgc && p.min_segno == NULL_SEGNO && | |
878 | sm->elapsed_time < p.age_threshold) { | |
879 | p.age_threshold = 0; | |
880 | goto retry; | |
881 | } | |
882 | ||
7bc09003 | 883 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 884 | got_it: |
e3080b01 CY |
885 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
886 | got_result: | |
7bc09003 | 887 | if (p.alloc_mode == LFS) { |
4ddb1a4d | 888 | secno = GET_SEC_FROM_SEG(sbi, p.min_segno); |
5ec4e49f JK |
889 | if (gc_type == FG_GC) |
890 | sbi->cur_victim_sec = secno; | |
891 | else | |
892 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 893 | } |
97767500 | 894 | ret = 0; |
8e46b3ed | 895 | |
e3c59108 ST |
896 | } |
897 | out: | |
898 | if (p.min_segno != NULL_SEGNO) | |
8e46b3ed NJ |
899 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, |
900 | sbi->cur_victim_sec, | |
901 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 JK |
902 | mutex_unlock(&dirty_i->seglist_lock); |
903 | ||
97767500 | 904 | return ret; |
7bc09003 JK |
905 | } |
906 | ||
907 | static const struct victim_selection default_v_ops = { | |
908 | .get_victim = get_victim_by_default, | |
909 | }; | |
910 | ||
7dda2af8 | 911 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 912 | { |
7bc09003 JK |
913 | struct inode_entry *ie; |
914 | ||
7dda2af8 CL |
915 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
916 | if (ie) | |
917 | return ie->inode; | |
7bc09003 JK |
918 | return NULL; |
919 | } | |
920 | ||
7dda2af8 | 921 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 922 | { |
6cc4af56 GZ |
923 | struct inode_entry *new_ie; |
924 | ||
7dda2af8 | 925 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
926 | iput(inode); |
927 | return; | |
7bc09003 | 928 | } |
32410577 CY |
929 | new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, |
930 | GFP_NOFS, true, NULL); | |
7bc09003 | 931 | new_ie->inode = inode; |
f28e5034 CY |
932 | |
933 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 934 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
935 | } |
936 | ||
7dda2af8 | 937 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
938 | { |
939 | struct inode_entry *ie, *next_ie; | |
5f029c04 | 940 | |
7dda2af8 CL |
941 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
942 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
943 | iput(ie->inode); |
944 | list_del(&ie->list); | |
4d57b86d | 945 | kmem_cache_free(f2fs_inode_entry_slab, ie); |
7bc09003 JK |
946 | } |
947 | } | |
948 | ||
949 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
950 | unsigned int segno, int offset) | |
951 | { | |
952 | struct sit_info *sit_i = SIT_I(sbi); | |
953 | struct seg_entry *sentry; | |
954 | int ret; | |
955 | ||
3d26fa6b | 956 | down_read(&sit_i->sentry_lock); |
7bc09003 JK |
957 | sentry = get_seg_entry(sbi, segno); |
958 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
3d26fa6b | 959 | up_read(&sit_i->sentry_lock); |
43727527 | 960 | return ret; |
7bc09003 JK |
961 | } |
962 | ||
0a8165d7 | 963 | /* |
7bc09003 JK |
964 | * This function compares node address got in summary with that in NAT. |
965 | * On validity, copy that node with cold status, otherwise (invalid node) | |
966 | * ignore that. | |
967 | */ | |
48018b4c | 968 | static int gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
969 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
970 | { | |
7bc09003 | 971 | struct f2fs_summary *entry; |
26d58599 | 972 | block_t start_addr; |
7bc09003 | 973 | int off; |
7ea984b0 | 974 | int phase = 0; |
c29fd0c0 | 975 | bool fggc = (gc_type == FG_GC); |
48018b4c | 976 | int submitted = 0; |
de881df9 | 977 | unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); |
7bc09003 | 978 | |
26d58599 JK |
979 | start_addr = START_BLOCK(sbi, segno); |
980 | ||
7bc09003 JK |
981 | next_step: |
982 | entry = sum; | |
c718379b | 983 | |
c29fd0c0 CY |
984 | if (fggc && phase == 2) |
985 | atomic_inc(&sbi->wb_sync_req[NODE]); | |
986 | ||
de881df9 | 987 | for (off = 0; off < usable_blks_in_seg; off++, entry++) { |
7bc09003 JK |
988 | nid_t nid = le32_to_cpu(entry->nid); |
989 | struct page *node_page; | |
26d58599 | 990 | struct node_info ni; |
48018b4c | 991 | int err; |
7bc09003 | 992 | |
43727527 | 993 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 994 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
48018b4c | 995 | return submitted; |
7bc09003 | 996 | |
43727527 | 997 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
998 | continue; |
999 | ||
7ea984b0 | 1000 | if (phase == 0) { |
4d57b86d | 1001 | f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
7ea984b0 CY |
1002 | META_NAT, true); |
1003 | continue; | |
1004 | } | |
1005 | ||
1006 | if (phase == 1) { | |
4d57b86d | 1007 | f2fs_ra_node_page(sbi, nid); |
7bc09003 JK |
1008 | continue; |
1009 | } | |
7ea984b0 CY |
1010 | |
1011 | /* phase == 2 */ | |
4d57b86d | 1012 | node_page = f2fs_get_node_page(sbi, nid); |
7bc09003 JK |
1013 | if (IS_ERR(node_page)) |
1014 | continue; | |
1015 | ||
4d57b86d | 1016 | /* block may become invalid during f2fs_get_node_page */ |
9a01b56b HY |
1017 | if (check_valid_map(sbi, segno, off) == 0) { |
1018 | f2fs_put_page(node_page, 1); | |
1019 | continue; | |
26d58599 JK |
1020 | } |
1021 | ||
a9419b63 | 1022 | if (f2fs_get_node_info(sbi, nid, &ni, false)) { |
7735730d CY |
1023 | f2fs_put_page(node_page, 1); |
1024 | continue; | |
1025 | } | |
1026 | ||
26d58599 JK |
1027 | if (ni.blk_addr != start_addr + off) { |
1028 | f2fs_put_page(node_page, 1); | |
1029 | continue; | |
9a01b56b HY |
1030 | } |
1031 | ||
48018b4c CY |
1032 | err = f2fs_move_node_page(node_page, gc_type); |
1033 | if (!err && gc_type == FG_GC) | |
1034 | submitted++; | |
e1235983 | 1035 | stat_inc_node_blk_count(sbi, 1, gc_type); |
7bc09003 | 1036 | } |
c718379b | 1037 | |
7ea984b0 | 1038 | if (++phase < 3) |
7bc09003 | 1039 | goto next_step; |
c29fd0c0 CY |
1040 | |
1041 | if (fggc) | |
1042 | atomic_dec(&sbi->wb_sync_req[NODE]); | |
48018b4c | 1043 | return submitted; |
7bc09003 JK |
1044 | } |
1045 | ||
0a8165d7 | 1046 | /* |
9af45ef5 JK |
1047 | * Calculate start block index indicating the given node offset. |
1048 | * Be careful, caller should give this node offset only indicating direct node | |
1049 | * blocks. If any node offsets, which point the other types of node blocks such | |
1050 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
1051 | * bug. | |
7bc09003 | 1052 | */ |
4d57b86d | 1053 | block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode) |
7bc09003 | 1054 | { |
ce19a5d4 JK |
1055 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
1056 | unsigned int bidx; | |
7bc09003 | 1057 | |
ce19a5d4 JK |
1058 | if (node_ofs == 0) |
1059 | return 0; | |
7bc09003 | 1060 | |
ce19a5d4 | 1061 | if (node_ofs <= 2) { |
7bc09003 JK |
1062 | bidx = node_ofs - 1; |
1063 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 1064 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
5f029c04 | 1065 | |
7bc09003 JK |
1066 | bidx = node_ofs - 2 - dec; |
1067 | } else { | |
ce19a5d4 | 1068 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
5f029c04 | 1069 | |
7bc09003 JK |
1070 | bidx = node_ofs - 5 - dec; |
1071 | } | |
d02a6e61 | 1072 | return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode); |
7bc09003 JK |
1073 | } |
1074 | ||
c1079892 | 1075 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
1076 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
1077 | { | |
1078 | struct page *node_page; | |
1079 | nid_t nid; | |
c6ad7fd1 | 1080 | unsigned int ofs_in_node, max_addrs; |
7bc09003 JK |
1081 | block_t source_blkaddr; |
1082 | ||
1083 | nid = le32_to_cpu(sum->nid); | |
1084 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
1085 | ||
4d57b86d | 1086 | node_page = f2fs_get_node_page(sbi, nid); |
7bc09003 | 1087 | if (IS_ERR(node_page)) |
c1079892 | 1088 | return false; |
7bc09003 | 1089 | |
a9419b63 | 1090 | if (f2fs_get_node_info(sbi, nid, dni, false)) { |
7735730d CY |
1091 | f2fs_put_page(node_page, 1); |
1092 | return false; | |
1093 | } | |
7bc09003 JK |
1094 | |
1095 | if (sum->version != dni->version) { | |
dcbb4c10 JP |
1096 | f2fs_warn(sbi, "%s: valid data with mismatched node version.", |
1097 | __func__); | |
c13ff37e | 1098 | set_sbi_flag(sbi, SBI_NEED_FSCK); |
7bc09003 JK |
1099 | } |
1100 | ||
6d18762e CY |
1101 | if (f2fs_check_nid_range(sbi, dni->ino)) { |
1102 | f2fs_put_page(node_page, 1); | |
77900c45 | 1103 | return false; |
6d18762e | 1104 | } |
77900c45 | 1105 | |
c6ad7fd1 CY |
1106 | max_addrs = IS_INODE(node_page) ? DEF_ADDRS_PER_INODE : |
1107 | DEF_ADDRS_PER_BLOCK; | |
1108 | if (ofs_in_node >= max_addrs) { | |
1109 | f2fs_err(sbi, "Inconsistent ofs_in_node:%u in summary, ino:%u, nid:%u, max:%u", | |
1110 | ofs_in_node, dni->ino, dni->nid, max_addrs); | |
c3db3c2f | 1111 | f2fs_put_page(node_page, 1); |
c6ad7fd1 CY |
1112 | return false; |
1113 | } | |
1114 | ||
7bc09003 | 1115 | *nofs = ofs_of_node(node_page); |
a2ced1ce | 1116 | source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node); |
7bc09003 JK |
1117 | f2fs_put_page(node_page, 1); |
1118 | ||
bbf9f7d9 ST |
1119 | if (source_blkaddr != blkaddr) { |
1120 | #ifdef CONFIG_F2FS_CHECK_FS | |
1121 | unsigned int segno = GET_SEGNO(sbi, blkaddr); | |
1122 | unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); | |
1123 | ||
1124 | if (unlikely(check_valid_map(sbi, segno, offset))) { | |
1125 | if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) { | |
833dcd35 JP |
1126 | f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u", |
1127 | blkaddr, source_blkaddr, segno); | |
f6db4307 | 1128 | set_sbi_flag(sbi, SBI_NEED_FSCK); |
bbf9f7d9 ST |
1129 | } |
1130 | } | |
1131 | #endif | |
c1079892 | 1132 | return false; |
bbf9f7d9 | 1133 | } |
c1079892 | 1134 | return true; |
7bc09003 JK |
1135 | } |
1136 | ||
6aa58d8a CY |
1137 | static int ra_data_block(struct inode *inode, pgoff_t index) |
1138 | { | |
1139 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
1140 | struct address_space *mapping = inode->i_mapping; | |
1141 | struct dnode_of_data dn; | |
1142 | struct page *page; | |
1143 | struct extent_info ei = {0, 0, 0}; | |
1144 | struct f2fs_io_info fio = { | |
1145 | .sbi = sbi, | |
1146 | .ino = inode->i_ino, | |
1147 | .type = DATA, | |
1148 | .temp = COLD, | |
1149 | .op = REQ_OP_READ, | |
1150 | .op_flags = 0, | |
1151 | .encrypted_page = NULL, | |
1152 | .in_list = false, | |
1153 | .retry = false, | |
1154 | }; | |
1155 | int err; | |
1156 | ||
1157 | page = f2fs_grab_cache_page(mapping, index, true); | |
1158 | if (!page) | |
1159 | return -ENOMEM; | |
1160 | ||
1161 | if (f2fs_lookup_extent_cache(inode, index, &ei)) { | |
1162 | dn.data_blkaddr = ei.blk + index - ei.fofs; | |
93770ab7 CY |
1163 | if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, |
1164 | DATA_GENERIC_ENHANCE_READ))) { | |
10f966bb | 1165 | err = -EFSCORRUPTED; |
95fa90c9 | 1166 | f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); |
93770ab7 CY |
1167 | goto put_page; |
1168 | } | |
6aa58d8a CY |
1169 | goto got_it; |
1170 | } | |
1171 | ||
1172 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
1173 | err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); | |
1174 | if (err) | |
1175 | goto put_page; | |
1176 | f2fs_put_dnode(&dn); | |
1177 | ||
93770ab7 CY |
1178 | if (!__is_valid_data_blkaddr(dn.data_blkaddr)) { |
1179 | err = -ENOENT; | |
1180 | goto put_page; | |
1181 | } | |
6aa58d8a | 1182 | if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, |
93770ab7 | 1183 | DATA_GENERIC_ENHANCE))) { |
10f966bb | 1184 | err = -EFSCORRUPTED; |
95fa90c9 | 1185 | f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); |
6aa58d8a CY |
1186 | goto put_page; |
1187 | } | |
1188 | got_it: | |
1189 | /* read page */ | |
1190 | fio.page = page; | |
1191 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; | |
1192 | ||
9bf1a3f7 YS |
1193 | /* |
1194 | * don't cache encrypted data into meta inode until previous dirty | |
1195 | * data were writebacked to avoid racing between GC and flush. | |
1196 | */ | |
bae0ee7a | 1197 | f2fs_wait_on_page_writeback(page, DATA, true, true); |
9bf1a3f7 YS |
1198 | |
1199 | f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); | |
1200 | ||
6aa58d8a CY |
1201 | fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi), |
1202 | dn.data_blkaddr, | |
1203 | FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
1204 | if (!fio.encrypted_page) { | |
1205 | err = -ENOMEM; | |
1206 | goto put_page; | |
1207 | } | |
1208 | ||
1209 | err = f2fs_submit_page_bio(&fio); | |
1210 | if (err) | |
1211 | goto put_encrypted_page; | |
1212 | f2fs_put_page(fio.encrypted_page, 0); | |
1213 | f2fs_put_page(page, 1); | |
8b83ac81 | 1214 | |
34a23525 CY |
1215 | f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE); |
1216 | f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE); | |
8b83ac81 | 1217 | |
6aa58d8a CY |
1218 | return 0; |
1219 | put_encrypted_page: | |
1220 | f2fs_put_page(fio.encrypted_page, 1); | |
1221 | put_page: | |
1222 | f2fs_put_page(page, 1); | |
1223 | return err; | |
1224 | } | |
1225 | ||
d4c759ee JK |
1226 | /* |
1227 | * Move data block via META_MAPPING while keeping locked data page. | |
1228 | * This can be used to move blocks, aka LBAs, directly on disk. | |
1229 | */ | |
48018b4c | 1230 | static int move_data_block(struct inode *inode, block_t bidx, |
2ef79ecb | 1231 | int gc_type, unsigned int segno, int off) |
4375a336 JK |
1232 | { |
1233 | struct f2fs_io_info fio = { | |
1234 | .sbi = F2FS_I_SB(inode), | |
39d787be | 1235 | .ino = inode->i_ino, |
4375a336 | 1236 | .type = DATA, |
a912b54d | 1237 | .temp = COLD, |
04d328de | 1238 | .op = REQ_OP_READ, |
70fd7614 | 1239 | .op_flags = 0, |
4375a336 | 1240 | .encrypted_page = NULL, |
fb830fc5 | 1241 | .in_list = false, |
fe16efe6 | 1242 | .retry = false, |
4375a336 JK |
1243 | }; |
1244 | struct dnode_of_data dn; | |
1245 | struct f2fs_summary sum; | |
1246 | struct node_info ni; | |
6aa58d8a | 1247 | struct page *page, *mpage; |
4356e48e | 1248 | block_t newaddr; |
48018b4c | 1249 | int err = 0; |
b0332a0f | 1250 | bool lfs_mode = f2fs_lfs_mode(fio.sbi); |
ac2d750b WG |
1251 | int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) && |
1252 | (fio.sbi->gc_mode != GC_URGENT_HIGH) ? | |
093749e2 | 1253 | CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA; |
4375a336 JK |
1254 | |
1255 | /* do not read out */ | |
a56c7c6f | 1256 | page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); |
4375a336 | 1257 | if (!page) |
48018b4c | 1258 | return -ENOMEM; |
4375a336 | 1259 | |
48018b4c CY |
1260 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { |
1261 | err = -ENOENT; | |
20614711 | 1262 | goto out; |
48018b4c | 1263 | } |
20614711 | 1264 | |
71419129 CY |
1265 | err = f2fs_gc_pinned_control(inode, gc_type, segno); |
1266 | if (err) | |
1ad71a27 | 1267 | goto out; |
1ad71a27 | 1268 | |
4375a336 | 1269 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
4d57b86d | 1270 | err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE); |
4375a336 JK |
1271 | if (err) |
1272 | goto out; | |
1273 | ||
08b39fbd CY |
1274 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
1275 | ClearPageUptodate(page); | |
48018b4c | 1276 | err = -ENOENT; |
4375a336 | 1277 | goto put_out; |
08b39fbd CY |
1278 | } |
1279 | ||
1280 | /* | |
1281 | * don't cache encrypted data into meta inode until previous dirty | |
1282 | * data were writebacked to avoid racing between GC and flush. | |
1283 | */ | |
bae0ee7a | 1284 | f2fs_wait_on_page_writeback(page, DATA, true, true); |
4375a336 | 1285 | |
9bf1a3f7 YS |
1286 | f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); |
1287 | ||
a9419b63 | 1288 | err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false); |
7735730d CY |
1289 | if (err) |
1290 | goto put_out; | |
1291 | ||
4375a336 JK |
1292 | /* read page */ |
1293 | fio.page = page; | |
7a9d7548 | 1294 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; |
4375a336 | 1295 | |
107a805d | 1296 | if (lfs_mode) |
e4544b63 | 1297 | f2fs_down_write(&fio.sbi->io_order_lock); |
107a805d | 1298 | |
543b8c46 JK |
1299 | mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi), |
1300 | fio.old_blkaddr, false); | |
d7cd3702 CY |
1301 | if (!mpage) { |
1302 | err = -ENOMEM; | |
543b8c46 | 1303 | goto up_out; |
d7cd3702 | 1304 | } |
543b8c46 JK |
1305 | |
1306 | fio.encrypted_page = mpage; | |
1307 | ||
1308 | /* read source block in mpage */ | |
1309 | if (!PageUptodate(mpage)) { | |
1310 | err = f2fs_submit_page_bio(&fio); | |
1311 | if (err) { | |
1312 | f2fs_put_page(mpage, 1); | |
1313 | goto up_out; | |
1314 | } | |
8b83ac81 | 1315 | |
34a23525 CY |
1316 | f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO, |
1317 | F2FS_BLKSIZE); | |
1318 | f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO, | |
1319 | F2FS_BLKSIZE); | |
8b83ac81 | 1320 | |
543b8c46 JK |
1321 | lock_page(mpage); |
1322 | if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) || | |
1323 | !PageUptodate(mpage))) { | |
1324 | err = -EIO; | |
1325 | f2fs_put_page(mpage, 1); | |
1326 | goto up_out; | |
1327 | } | |
1328 | } | |
1329 | ||
cf740403 CY |
1330 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); |
1331 | ||
1332 | /* allocate block address */ | |
4d57b86d | 1333 | f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, |
093749e2 | 1334 | &sum, type, NULL); |
4356e48e | 1335 | |
01eccef7 CY |
1336 | fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), |
1337 | newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
4356e48e CY |
1338 | if (!fio.encrypted_page) { |
1339 | err = -ENOMEM; | |
6aa58d8a | 1340 | f2fs_put_page(mpage, 1); |
543b8c46 | 1341 | goto recover_block; |
4356e48e | 1342 | } |
548aedac | 1343 | |
543b8c46 | 1344 | /* write target block */ |
bae0ee7a | 1345 | f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true); |
543b8c46 JK |
1346 | memcpy(page_address(fio.encrypted_page), |
1347 | page_address(mpage), PAGE_SIZE); | |
1348 | f2fs_put_page(mpage, 1); | |
1349 | invalidate_mapping_pages(META_MAPPING(fio.sbi), | |
1350 | fio.old_blkaddr, fio.old_blkaddr); | |
6ce19aff | 1351 | f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr); |
543b8c46 | 1352 | |
8d64d365 | 1353 | set_page_dirty(fio.encrypted_page); |
6282adbf JK |
1354 | if (clear_page_dirty_for_io(fio.encrypted_page)) |
1355 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
1356 | ||
548aedac | 1357 | set_page_writeback(fio.encrypted_page); |
17c50035 | 1358 | ClearPageError(page); |
4375a336 | 1359 | |
04d328de | 1360 | fio.op = REQ_OP_WRITE; |
70fd7614 | 1361 | fio.op_flags = REQ_SYNC; |
4356e48e | 1362 | fio.new_blkaddr = newaddr; |
fe16efe6 CY |
1363 | f2fs_submit_page_write(&fio); |
1364 | if (fio.retry) { | |
48018b4c | 1365 | err = -EAGAIN; |
a9d572c7 SY |
1366 | if (PageWriteback(fio.encrypted_page)) |
1367 | end_page_writeback(fio.encrypted_page); | |
1368 | goto put_page_out; | |
1369 | } | |
4375a336 | 1370 | |
34a23525 | 1371 | f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE); |
b0af6d49 | 1372 | |
f28b3434 | 1373 | f2fs_update_data_blkaddr(&dn, newaddr); |
91942321 | 1374 | set_inode_flag(inode, FI_APPEND_WRITE); |
4375a336 | 1375 | if (page->index == 0) |
91942321 | 1376 | set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); |
548aedac | 1377 | put_page_out: |
4375a336 | 1378 | f2fs_put_page(fio.encrypted_page, 1); |
4356e48e CY |
1379 | recover_block: |
1380 | if (err) | |
4d57b86d | 1381 | f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, |
c5d02785 | 1382 | true, true, true); |
543b8c46 JK |
1383 | up_out: |
1384 | if (lfs_mode) | |
e4544b63 | 1385 | f2fs_up_write(&fio.sbi->io_order_lock); |
4375a336 JK |
1386 | put_out: |
1387 | f2fs_put_dnode(&dn); | |
1388 | out: | |
1389 | f2fs_put_page(page, 1); | |
48018b4c | 1390 | return err; |
4375a336 JK |
1391 | } |
1392 | ||
48018b4c | 1393 | static int move_data_page(struct inode *inode, block_t bidx, int gc_type, |
20614711 | 1394 | unsigned int segno, int off) |
7bc09003 | 1395 | { |
c879f90d | 1396 | struct page *page; |
48018b4c | 1397 | int err = 0; |
c879f90d | 1398 | |
4d57b86d | 1399 | page = f2fs_get_lock_data_page(inode, bidx, true); |
c879f90d | 1400 | if (IS_ERR(page)) |
48018b4c | 1401 | return PTR_ERR(page); |
63a0b7cb | 1402 | |
48018b4c CY |
1403 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { |
1404 | err = -ENOENT; | |
20614711 | 1405 | goto out; |
48018b4c | 1406 | } |
20614711 | 1407 | |
71419129 CY |
1408 | err = f2fs_gc_pinned_control(inode, gc_type, segno); |
1409 | if (err) | |
1ad71a27 | 1410 | goto out; |
5fe45743 | 1411 | |
7bc09003 | 1412 | if (gc_type == BG_GC) { |
48018b4c CY |
1413 | if (PageWriteback(page)) { |
1414 | err = -EAGAIN; | |
4ebefc44 | 1415 | goto out; |
48018b4c | 1416 | } |
7bc09003 | 1417 | set_page_dirty(page); |
b763f3be | 1418 | set_page_private_gcing(page); |
7bc09003 | 1419 | } else { |
c879f90d JK |
1420 | struct f2fs_io_info fio = { |
1421 | .sbi = F2FS_I_SB(inode), | |
39d787be | 1422 | .ino = inode->i_ino, |
c879f90d | 1423 | .type = DATA, |
a912b54d | 1424 | .temp = COLD, |
04d328de | 1425 | .op = REQ_OP_WRITE, |
70fd7614 | 1426 | .op_flags = REQ_SYNC, |
e959c8f5 | 1427 | .old_blkaddr = NULL_ADDR, |
c879f90d | 1428 | .page = page, |
4375a336 | 1429 | .encrypted_page = NULL, |
cc15620b | 1430 | .need_lock = LOCK_REQ, |
b0af6d49 | 1431 | .io_type = FS_GC_DATA_IO, |
c879f90d | 1432 | }; |
72e1c797 | 1433 | bool is_dirty = PageDirty(page); |
72e1c797 CY |
1434 | |
1435 | retry: | |
bae0ee7a | 1436 | f2fs_wait_on_page_writeback(page, DATA, true, true); |
8d64d365 CY |
1437 | |
1438 | set_page_dirty(page); | |
933439c8 | 1439 | if (clear_page_dirty_for_io(page)) { |
a7ffdbe2 | 1440 | inode_dec_dirty_pages(inode); |
4d57b86d | 1441 | f2fs_remove_dirty_inode(inode); |
933439c8 | 1442 | } |
72e1c797 | 1443 | |
b763f3be | 1444 | set_page_private_gcing(page); |
72e1c797 | 1445 | |
4d57b86d | 1446 | err = f2fs_do_write_data_page(&fio); |
14a28559 | 1447 | if (err) { |
b763f3be | 1448 | clear_page_private_gcing(page); |
14a28559 | 1449 | if (err == -ENOMEM) { |
4034247a | 1450 | memalloc_retry_wait(GFP_NOFS); |
14a28559 CY |
1451 | goto retry; |
1452 | } | |
1453 | if (is_dirty) | |
1454 | set_page_dirty(page); | |
72e1c797 | 1455 | } |
7bc09003 JK |
1456 | } |
1457 | out: | |
1458 | f2fs_put_page(page, 1); | |
48018b4c | 1459 | return err; |
7bc09003 JK |
1460 | } |
1461 | ||
0a8165d7 | 1462 | /* |
7bc09003 JK |
1463 | * This function tries to get parent node of victim data block, and identifies |
1464 | * data block validity. If the block is valid, copy that with cold status and | |
1465 | * modify parent node. | |
1466 | * If the parent node is not valid or the data block address is different, | |
1467 | * the victim data block is ignored. | |
1468 | */ | |
48018b4c | 1469 | static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dede886 CY |
1470 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type, |
1471 | bool force_migrate) | |
7bc09003 JK |
1472 | { |
1473 | struct super_block *sb = sbi->sb; | |
1474 | struct f2fs_summary *entry; | |
1475 | block_t start_addr; | |
43727527 | 1476 | int off; |
7bc09003 | 1477 | int phase = 0; |
48018b4c | 1478 | int submitted = 0; |
de881df9 | 1479 | unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); |
7bc09003 JK |
1480 | |
1481 | start_addr = START_BLOCK(sbi, segno); | |
1482 | ||
1483 | next_step: | |
1484 | entry = sum; | |
c718379b | 1485 | |
de881df9 | 1486 | for (off = 0; off < usable_blks_in_seg; off++, entry++) { |
7bc09003 JK |
1487 | struct page *data_page; |
1488 | struct inode *inode; | |
1489 | struct node_info dni; /* dnode info for the data */ | |
1490 | unsigned int ofs_in_node, nofs; | |
1491 | block_t start_bidx; | |
7ea984b0 | 1492 | nid_t nid = le32_to_cpu(entry->nid); |
7bc09003 | 1493 | |
803e74be JK |
1494 | /* |
1495 | * stop BG_GC if there is not enough free sections. | |
1496 | * Or, stop GC if the segment becomes fully valid caused by | |
1497 | * race condition along with SSR block allocation. | |
1498 | */ | |
1499 | if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) || | |
7dede886 | 1500 | (!force_migrate && get_valid_blocks(sbi, segno, true) == |
074b5ea2 | 1501 | CAP_BLKS_PER_SEC(sbi))) |
48018b4c | 1502 | return submitted; |
7bc09003 | 1503 | |
43727527 | 1504 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
1505 | continue; |
1506 | ||
1507 | if (phase == 0) { | |
4d57b86d | 1508 | f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
7ea984b0 CY |
1509 | META_NAT, true); |
1510 | continue; | |
1511 | } | |
1512 | ||
1513 | if (phase == 1) { | |
4d57b86d | 1514 | f2fs_ra_node_page(sbi, nid); |
7bc09003 JK |
1515 | continue; |
1516 | } | |
1517 | ||
1518 | /* Get an inode by ino with checking validity */ | |
c1079892 | 1519 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
1520 | continue; |
1521 | ||
7ea984b0 | 1522 | if (phase == 2) { |
4d57b86d | 1523 | f2fs_ra_node_page(sbi, dni.ino); |
7bc09003 JK |
1524 | continue; |
1525 | } | |
1526 | ||
7bc09003 JK |
1527 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
1528 | ||
7ea984b0 | 1529 | if (phase == 3) { |
71419129 CY |
1530 | int err; |
1531 | ||
d4686d56 | 1532 | inode = f2fs_iget(sb, dni.ino); |
9056d648 CY |
1533 | if (IS_ERR(inode) || is_bad_inode(inode) || |
1534 | special_file(inode->i_mode)) | |
7bc09003 JK |
1535 | continue; |
1536 | ||
71419129 CY |
1537 | err = f2fs_gc_pinned_control(inode, gc_type, segno); |
1538 | if (err == -EAGAIN) { | |
a22bb552 CY |
1539 | iput(inode); |
1540 | return submitted; | |
1541 | } | |
1542 | ||
e4544b63 | 1543 | if (!f2fs_down_write_trylock( |
b2532c69 | 1544 | &F2FS_I(inode)->i_gc_rwsem[WRITE])) { |
bb06664a | 1545 | iput(inode); |
6f8d4455 | 1546 | sbi->skipped_gc_rwsem++; |
bb06664a CY |
1547 | continue; |
1548 | } | |
1549 | ||
6aa58d8a CY |
1550 | start_bidx = f2fs_start_bidx_of_node(nofs, inode) + |
1551 | ofs_in_node; | |
1552 | ||
1553 | if (f2fs_post_read_required(inode)) { | |
1554 | int err = ra_data_block(inode, start_bidx); | |
1555 | ||
e4544b63 | 1556 | f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); |
6aa58d8a CY |
1557 | if (err) { |
1558 | iput(inode); | |
1559 | continue; | |
1560 | } | |
1561 | add_gc_inode(gc_list, inode); | |
1562 | continue; | |
1563 | } | |
1564 | ||
4d57b86d | 1565 | data_page = f2fs_get_read_data_page(inode, |
6aa58d8a | 1566 | start_bidx, REQ_RAHEAD, true); |
e4544b63 | 1567 | f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); |
31a32688 CL |
1568 | if (IS_ERR(data_page)) { |
1569 | iput(inode); | |
1570 | continue; | |
1571 | } | |
7bc09003 JK |
1572 | |
1573 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 1574 | add_gc_inode(gc_list, inode); |
31a32688 CL |
1575 | continue; |
1576 | } | |
1577 | ||
7ea984b0 | 1578 | /* phase 4 */ |
7dda2af8 | 1579 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 1580 | if (inode) { |
82e0a5aa CY |
1581 | struct f2fs_inode_info *fi = F2FS_I(inode); |
1582 | bool locked = false; | |
48018b4c | 1583 | int err; |
82e0a5aa CY |
1584 | |
1585 | if (S_ISREG(inode->i_mode)) { | |
e4544b63 | 1586 | if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[READ])) { |
ad126ebd | 1587 | sbi->skipped_gc_rwsem++; |
82e0a5aa | 1588 | continue; |
ad126ebd | 1589 | } |
e4544b63 | 1590 | if (!f2fs_down_write_trylock( |
b2532c69 | 1591 | &fi->i_gc_rwsem[WRITE])) { |
6f8d4455 | 1592 | sbi->skipped_gc_rwsem++; |
e4544b63 | 1593 | f2fs_up_write(&fi->i_gc_rwsem[READ]); |
82e0a5aa CY |
1594 | continue; |
1595 | } | |
1596 | locked = true; | |
73ac2f4e CY |
1597 | |
1598 | /* wait for all inflight aio data */ | |
1599 | inode_dio_wait(inode); | |
82e0a5aa CY |
1600 | } |
1601 | ||
4d57b86d | 1602 | start_bidx = f2fs_start_bidx_of_node(nofs, inode) |
c879f90d | 1603 | + ofs_in_node; |
6dbb1796 | 1604 | if (f2fs_post_read_required(inode)) |
48018b4c CY |
1605 | err = move_data_block(inode, start_bidx, |
1606 | gc_type, segno, off); | |
4375a336 | 1607 | else |
48018b4c | 1608 | err = move_data_page(inode, start_bidx, gc_type, |
d4c759ee | 1609 | segno, off); |
82e0a5aa | 1610 | |
48018b4c CY |
1611 | if (!err && (gc_type == FG_GC || |
1612 | f2fs_post_read_required(inode))) | |
1613 | submitted++; | |
1614 | ||
82e0a5aa | 1615 | if (locked) { |
e4544b63 TM |
1616 | f2fs_up_write(&fi->i_gc_rwsem[WRITE]); |
1617 | f2fs_up_write(&fi->i_gc_rwsem[READ]); | |
82e0a5aa CY |
1618 | } |
1619 | ||
e1235983 | 1620 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 1621 | } |
7bc09003 | 1622 | } |
c718379b | 1623 | |
7ea984b0 | 1624 | if (++phase < 5) |
7bc09003 | 1625 | goto next_step; |
48018b4c CY |
1626 | |
1627 | return submitted; | |
7bc09003 JK |
1628 | } |
1629 | ||
1630 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 1631 | int gc_type) |
7bc09003 JK |
1632 | { |
1633 | struct sit_info *sit_i = SIT_I(sbi); | |
1634 | int ret; | |
8a2d0ace | 1635 | |
3d26fa6b | 1636 | down_write(&sit_i->sentry_lock); |
8a2d0ace | 1637 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
093749e2 | 1638 | NO_CHECK_TYPE, LFS, 0); |
3d26fa6b | 1639 | up_write(&sit_i->sentry_lock); |
7bc09003 JK |
1640 | return ret; |
1641 | } | |
1642 | ||
718e53fa CY |
1643 | static int do_garbage_collect(struct f2fs_sb_info *sbi, |
1644 | unsigned int start_segno, | |
7dede886 CY |
1645 | struct gc_inode_list *gc_list, int gc_type, |
1646 | bool force_migrate) | |
7bc09003 JK |
1647 | { |
1648 | struct page *sum_page; | |
1649 | struct f2fs_summary_block *sum; | |
c718379b | 1650 | struct blk_plug plug; |
718e53fa CY |
1651 | unsigned int segno = start_segno; |
1652 | unsigned int end_segno = start_segno + sbi->segs_per_sec; | |
e3080b01 | 1653 | int seg_freed = 0, migrated = 0; |
718e53fa CY |
1654 | unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? |
1655 | SUM_TYPE_DATA : SUM_TYPE_NODE; | |
48018b4c | 1656 | int submitted = 0; |
7bc09003 | 1657 | |
e3080b01 CY |
1658 | if (__is_large_section(sbi)) |
1659 | end_segno = rounddown(end_segno, sbi->segs_per_sec); | |
1660 | ||
de881df9 AR |
1661 | /* |
1662 | * zone-capacity can be less than zone-size in zoned devices, | |
1663 | * resulting in less than expected usable segments in the zone, | |
1664 | * calculate the end segno in the zone which can be garbage collected | |
1665 | */ | |
1666 | if (f2fs_sb_has_blkzoned(sbi)) | |
1667 | end_segno -= sbi->segs_per_sec - | |
1668 | f2fs_usable_segs_in_sec(sbi, segno); | |
1669 | ||
093749e2 CY |
1670 | sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type); |
1671 | ||
718e53fa | 1672 | /* readahead multi ssa blocks those have contiguous address */ |
2c70c5e3 | 1673 | if (__is_large_section(sbi)) |
4d57b86d | 1674 | f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), |
e3080b01 | 1675 | end_segno - segno, META_SSA, true); |
718e53fa CY |
1676 | |
1677 | /* reference all summary page */ | |
1678 | while (segno < end_segno) { | |
4d57b86d | 1679 | sum_page = f2fs_get_sum_page(sbi, segno++); |
edc55aaf JK |
1680 | if (IS_ERR(sum_page)) { |
1681 | int err = PTR_ERR(sum_page); | |
1682 | ||
1683 | end_segno = segno - 1; | |
1684 | for (segno = start_segno; segno < end_segno; segno++) { | |
1685 | sum_page = find_get_page(META_MAPPING(sbi), | |
1686 | GET_SUM_BLOCK(sbi, segno)); | |
1687 | f2fs_put_page(sum_page, 0); | |
1688 | f2fs_put_page(sum_page, 0); | |
1689 | } | |
1690 | return err; | |
1691 | } | |
718e53fa CY |
1692 | unlock_page(sum_page); |
1693 | } | |
7bc09003 | 1694 | |
c718379b JK |
1695 | blk_start_plug(&plug); |
1696 | ||
718e53fa | 1697 | for (segno = start_segno; segno < end_segno; segno++) { |
aa987273 | 1698 | |
718e53fa CY |
1699 | /* find segment summary of victim */ |
1700 | sum_page = find_get_page(META_MAPPING(sbi), | |
1701 | GET_SUM_BLOCK(sbi, segno)); | |
718e53fa | 1702 | f2fs_put_page(sum_page, 0); |
7bc09003 | 1703 | |
d6c66cd1 YS |
1704 | if (get_valid_blocks(sbi, segno, false) == 0) |
1705 | goto freed; | |
dabfbbc8 | 1706 | if (gc_type == BG_GC && __is_large_section(sbi) && |
e3080b01 CY |
1707 | migrated >= sbi->migration_granularity) |
1708 | goto skip; | |
d6c66cd1 | 1709 | if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi))) |
e3080b01 | 1710 | goto skip; |
de0dcc40 | 1711 | |
718e53fa | 1712 | sum = page_address(sum_page); |
10d255c3 | 1713 | if (type != GET_SUM_TYPE((&sum->footer))) { |
dcbb4c10 JP |
1714 | f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT", |
1715 | segno, type, GET_SUM_TYPE((&sum->footer))); | |
10d255c3 | 1716 | set_sbi_flag(sbi, SBI_NEED_FSCK); |
a9cfee0e CY |
1717 | f2fs_stop_checkpoint(sbi, false, |
1718 | STOP_CP_REASON_CORRUPTED_SUMMARY); | |
e3080b01 | 1719 | goto skip; |
10d255c3 | 1720 | } |
718e53fa CY |
1721 | |
1722 | /* | |
1723 | * this is to avoid deadlock: | |
1724 | * - lock_page(sum_page) - f2fs_replace_block | |
3d26fa6b CY |
1725 | * - check_valid_map() - down_write(sentry_lock) |
1726 | * - down_read(sentry_lock) - change_curseg() | |
718e53fa CY |
1727 | * - lock_page(sum_page) |
1728 | */ | |
718e53fa | 1729 | if (type == SUM_TYPE_NODE) |
48018b4c | 1730 | submitted += gc_node_segment(sbi, sum->entries, segno, |
718e53fa | 1731 | gc_type); |
48018b4c CY |
1732 | else |
1733 | submitted += gc_data_segment(sbi, sum->entries, gc_list, | |
7dede886 CY |
1734 | segno, gc_type, |
1735 | force_migrate); | |
718e53fa CY |
1736 | |
1737 | stat_inc_seg_count(sbi, type, gc_type); | |
07c6b593 | 1738 | sbi->gc_reclaimed_segs[sbi->gc_mode]++; |
8c7b9ac1 | 1739 | migrated++; |
c56f16da | 1740 | |
d6c66cd1 | 1741 | freed: |
c56f16da CY |
1742 | if (gc_type == FG_GC && |
1743 | get_valid_blocks(sbi, segno, false) == 0) | |
1744 | seg_freed++; | |
e3080b01 CY |
1745 | |
1746 | if (__is_large_section(sbi) && segno + 1 < end_segno) | |
1747 | sbi->next_victim_seg[gc_type] = segno + 1; | |
1748 | skip: | |
718e53fa CY |
1749 | f2fs_put_page(sum_page, 0); |
1750 | } | |
1751 | ||
48018b4c | 1752 | if (submitted) |
b9109b0e JK |
1753 | f2fs_submit_merged_write(sbi, |
1754 | (type == SUM_TYPE_NODE) ? NODE : DATA); | |
c718379b | 1755 | |
718e53fa | 1756 | blk_finish_plug(&plug); |
7bc09003 | 1757 | |
17d899df CY |
1758 | stat_inc_call_count(sbi->stat_info); |
1759 | ||
c56f16da | 1760 | return seg_freed; |
7bc09003 JK |
1761 | } |
1762 | ||
d147ea4a | 1763 | int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control) |
7bc09003 | 1764 | { |
d147ea4a JK |
1765 | int gc_type = gc_control->init_gc_type; |
1766 | unsigned int segno = gc_control->victim_segno; | |
c56f16da CY |
1767 | int sec_freed = 0, seg_freed = 0, total_freed = 0; |
1768 | int ret = 0; | |
d5053a34 | 1769 | struct cp_control cpc; |
7dda2af8 CL |
1770 | struct gc_inode_list gc_list = { |
1771 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
f6bb2a2c | 1772 | .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), |
7dda2af8 | 1773 | }; |
2ef79ecb | 1774 | unsigned int skipped_round = 0, round = 0; |
d5053a34 | 1775 | |
d147ea4a | 1776 | trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc, |
c81d5bae | 1777 | gc_control->nr_free_secs, |
c56f16da CY |
1778 | get_pages(sbi, F2FS_DIRTY_NODES), |
1779 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1780 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1781 | free_sections(sbi), | |
1782 | free_segments(sbi), | |
1783 | reserved_segments(sbi), | |
1784 | prefree_segments(sbi)); | |
1785 | ||
119ee914 | 1786 | cpc.reason = __get_cp_reason(sbi); |
6f8d4455 | 1787 | sbi->skipped_gc_rwsem = 0; |
7bc09003 | 1788 | gc_more: |
1751e8a6 | 1789 | if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { |
e5dbd956 | 1790 | ret = -EINVAL; |
408e9375 | 1791 | goto stop; |
e5dbd956 | 1792 | } |
6d5a1495 CY |
1793 | if (unlikely(f2fs_cp_error(sbi))) { |
1794 | ret = -EIO; | |
203681f6 | 1795 | goto stop; |
6d5a1495 | 1796 | } |
7bc09003 | 1797 | |
19f4e688 | 1798 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) { |
6e17bfbc | 1799 | /* |
19f4e688 HP |
1800 | * For example, if there are many prefree_segments below given |
1801 | * threshold, we can make them free by checkpoint. Then, we | |
1802 | * secure free segments which doesn't need fggc any more. | |
6e17bfbc | 1803 | */ |
d147ea4a | 1804 | if (prefree_segments(sbi)) { |
4d57b86d | 1805 | ret = f2fs_write_checkpoint(sbi, &cpc); |
8fd5a37e JK |
1806 | if (ret) |
1807 | goto stop; | |
1808 | } | |
19f4e688 HP |
1809 | if (has_not_enough_free_secs(sbi, 0, 0)) |
1810 | gc_type = FG_GC; | |
d64f8047 | 1811 | } |
7bc09003 | 1812 | |
19f4e688 | 1813 | /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ |
d147ea4a | 1814 | if (gc_type == BG_GC && gc_control->no_bg_gc) { |
c56f16da | 1815 | ret = -EINVAL; |
19f4e688 | 1816 | goto stop; |
c56f16da | 1817 | } |
71419129 | 1818 | retry: |
97767500 | 1819 | ret = __get_victim(sbi, &segno, gc_type); |
71419129 CY |
1820 | if (ret) { |
1821 | /* allow to search victim from sections has pinned data */ | |
1822 | if (ret == -ENODATA && gc_type == FG_GC && | |
1823 | f2fs_pinned_section_exists(DIRTY_I(sbi))) { | |
1824 | f2fs_unpin_all_sections(sbi, false); | |
1825 | goto retry; | |
1826 | } | |
408e9375 | 1827 | goto stop; |
71419129 | 1828 | } |
7bc09003 | 1829 | |
d147ea4a JK |
1830 | seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, |
1831 | gc_control->should_migrate_blocks); | |
c56f16da | 1832 | total_freed += seg_freed; |
43727527 | 1833 | |
d147ea4a JK |
1834 | if (seg_freed == f2fs_usable_segs_in_sec(sbi, segno)) |
1835 | sec_freed++; | |
2ef79ecb | 1836 | |
10d0786b | 1837 | if (gc_type == FG_GC) |
5ec4e49f | 1838 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 1839 | |
c81d5bae JK |
1840 | if (gc_control->init_gc_type == FG_GC || |
1841 | !has_not_enough_free_secs(sbi, | |
1842 | (gc_type == FG_GC) ? sec_freed : 0, 0)) { | |
1843 | if (gc_type == FG_GC && sec_freed < gc_control->nr_free_secs) | |
1844 | goto go_gc_more; | |
a9163b94 | 1845 | goto stop; |
c81d5bae | 1846 | } |
43727527 | 1847 | |
d147ea4a JK |
1848 | /* FG_GC stops GC by skip_count */ |
1849 | if (gc_type == FG_GC) { | |
1850 | if (sbi->skipped_gc_rwsem) | |
1851 | skipped_round++; | |
1852 | round++; | |
1853 | if (skipped_round > MAX_SKIP_GC_COUNT && | |
1854 | skipped_round * 2 >= round) { | |
4d57b86d | 1855 | ret = f2fs_write_checkpoint(sbi, &cpc); |
d147ea4a | 1856 | goto stop; |
a9163b94 | 1857 | } |
a9163b94 | 1858 | } |
d147ea4a JK |
1859 | |
1860 | /* Write checkpoint to reclaim prefree segments */ | |
1861 | if (free_sections(sbi) < NR_CURSEG_PERSIST_TYPE && | |
1862 | prefree_segments(sbi)) { | |
a9163b94 | 1863 | ret = f2fs_write_checkpoint(sbi, &cpc); |
d147ea4a JK |
1864 | if (ret) |
1865 | goto stop; | |
1866 | } | |
c81d5bae | 1867 | go_gc_more: |
d147ea4a JK |
1868 | segno = NULL_SEGNO; |
1869 | goto gc_more; | |
1870 | ||
408e9375 | 1871 | stop: |
e066b83c | 1872 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; |
d147ea4a | 1873 | SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno; |
c56f16da | 1874 | |
71419129 CY |
1875 | if (gc_type == FG_GC) |
1876 | f2fs_unpin_all_sections(sbi, true); | |
1877 | ||
c56f16da CY |
1878 | trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed, |
1879 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1880 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1881 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1882 | free_sections(sbi), | |
1883 | free_segments(sbi), | |
1884 | reserved_segments(sbi), | |
1885 | prefree_segments(sbi)); | |
1886 | ||
e4544b63 | 1887 | f2fs_up_write(&sbi->gc_lock); |
7bc09003 | 1888 | |
7dda2af8 | 1889 | put_gc_inode(&gc_list); |
d530d4d8 | 1890 | |
d147ea4a | 1891 | if (gc_control->err_gc_skipped && !ret) |
d530d4d8 | 1892 | ret = sec_freed ? 0 : -EAGAIN; |
43727527 | 1893 | return ret; |
7bc09003 JK |
1894 | } |
1895 | ||
093749e2 CY |
1896 | int __init f2fs_create_garbage_collection_cache(void) |
1897 | { | |
1898 | victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry", | |
1899 | sizeof(struct victim_entry)); | |
1900 | if (!victim_entry_slab) | |
1901 | return -ENOMEM; | |
1902 | return 0; | |
1903 | } | |
1904 | ||
1905 | void f2fs_destroy_garbage_collection_cache(void) | |
1906 | { | |
1907 | kmem_cache_destroy(victim_entry_slab); | |
1908 | } | |
1909 | ||
1910 | static void init_atgc_management(struct f2fs_sb_info *sbi) | |
1911 | { | |
1912 | struct atgc_management *am = &sbi->am; | |
1913 | ||
1914 | if (test_opt(sbi, ATGC) && | |
1915 | SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD) | |
1916 | am->atgc_enabled = true; | |
1917 | ||
1918 | am->root = RB_ROOT_CACHED; | |
1919 | INIT_LIST_HEAD(&am->victim_list); | |
1920 | am->victim_count = 0; | |
1921 | ||
1922 | am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO; | |
1923 | am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT; | |
1924 | am->age_weight = DEF_GC_THREAD_AGE_WEIGHT; | |
89e53ff1 | 1925 | am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD; |
093749e2 CY |
1926 | } |
1927 | ||
4d57b86d | 1928 | void f2fs_build_gc_manager(struct f2fs_sb_info *sbi) |
7bc09003 JK |
1929 | { |
1930 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
e93b9865 | 1931 | |
1ad71a27 | 1932 | sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES; |
d5793249 JK |
1933 | |
1934 | /* give warm/cold data area from slower device */ | |
0916878d | 1935 | if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi)) |
d5793249 JK |
1936 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = |
1937 | GET_SEGNO(sbi, FDEV(0).end_blk) + 1; | |
093749e2 CY |
1938 | |
1939 | init_atgc_management(sbi); | |
7bc09003 | 1940 | } |
04f0b2ea | 1941 | |
b4b10061 JK |
1942 | static int free_segment_range(struct f2fs_sb_info *sbi, |
1943 | unsigned int secs, bool gc_only) | |
04f0b2ea | 1944 | { |
b4b10061 JK |
1945 | unsigned int segno, next_inuse, start, end; |
1946 | struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; | |
1947 | int gc_mode, gc_type; | |
04f0b2ea | 1948 | int err = 0; |
b4b10061 JK |
1949 | int type; |
1950 | ||
1951 | /* Force block allocation for GC */ | |
1952 | MAIN_SECS(sbi) -= secs; | |
1953 | start = MAIN_SECS(sbi) * sbi->segs_per_sec; | |
1954 | end = MAIN_SEGS(sbi) - 1; | |
1955 | ||
1956 | mutex_lock(&DIRTY_I(sbi)->seglist_lock); | |
1957 | for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++) | |
1958 | if (SIT_I(sbi)->last_victim[gc_mode] >= start) | |
1959 | SIT_I(sbi)->last_victim[gc_mode] = 0; | |
1960 | ||
1961 | for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++) | |
1962 | if (sbi->next_victim_seg[gc_type] >= start) | |
1963 | sbi->next_victim_seg[gc_type] = NULL_SEGNO; | |
1964 | mutex_unlock(&DIRTY_I(sbi)->seglist_lock); | |
04f0b2ea QS |
1965 | |
1966 | /* Move out cursegs from the target range */ | |
d0b9e42a | 1967 | for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++) |
0ef81833 | 1968 | f2fs_allocate_segment_for_resize(sbi, type, start, end); |
04f0b2ea QS |
1969 | |
1970 | /* do GC to move out valid blocks in the range */ | |
1971 | for (segno = start; segno <= end; segno += sbi->segs_per_sec) { | |
1972 | struct gc_inode_list gc_list = { | |
1973 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
1974 | .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), | |
1975 | }; | |
1976 | ||
7dede886 | 1977 | do_garbage_collect(sbi, segno, &gc_list, FG_GC, true); |
04f0b2ea QS |
1978 | put_gc_inode(&gc_list); |
1979 | ||
b4b10061 JK |
1980 | if (!gc_only && get_valid_blocks(sbi, segno, true)) { |
1981 | err = -EAGAIN; | |
1982 | goto out; | |
1983 | } | |
1984 | if (fatal_signal_pending(current)) { | |
1985 | err = -ERESTARTSYS; | |
1986 | goto out; | |
1987 | } | |
04f0b2ea | 1988 | } |
b4b10061 JK |
1989 | if (gc_only) |
1990 | goto out; | |
04f0b2ea | 1991 | |
b4b10061 | 1992 | err = f2fs_write_checkpoint(sbi, &cpc); |
04f0b2ea | 1993 | if (err) |
b4b10061 | 1994 | goto out; |
04f0b2ea QS |
1995 | |
1996 | next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start); | |
1997 | if (next_inuse <= end) { | |
dcbb4c10 JP |
1998 | f2fs_err(sbi, "segno %u should be free but still inuse!", |
1999 | next_inuse); | |
04f0b2ea QS |
2000 | f2fs_bug_on(sbi, 1); |
2001 | } | |
b4b10061 JK |
2002 | out: |
2003 | MAIN_SECS(sbi) += secs; | |
04f0b2ea QS |
2004 | return err; |
2005 | } | |
2006 | ||
2007 | static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs) | |
2008 | { | |
2009 | struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi); | |
a4ba5dfc CY |
2010 | int section_count; |
2011 | int segment_count; | |
2012 | int segment_count_main; | |
2013 | long long block_count; | |
04f0b2ea QS |
2014 | int segs = secs * sbi->segs_per_sec; |
2015 | ||
e4544b63 | 2016 | f2fs_down_write(&sbi->sb_lock); |
a4ba5dfc CY |
2017 | |
2018 | section_count = le32_to_cpu(raw_sb->section_count); | |
2019 | segment_count = le32_to_cpu(raw_sb->segment_count); | |
2020 | segment_count_main = le32_to_cpu(raw_sb->segment_count_main); | |
2021 | block_count = le64_to_cpu(raw_sb->block_count); | |
2022 | ||
04f0b2ea QS |
2023 | raw_sb->section_count = cpu_to_le32(section_count + secs); |
2024 | raw_sb->segment_count = cpu_to_le32(segment_count + segs); | |
2025 | raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs); | |
2026 | raw_sb->block_count = cpu_to_le64(block_count + | |
2027 | (long long)segs * sbi->blocks_per_seg); | |
46d9ce19 QS |
2028 | if (f2fs_is_multi_device(sbi)) { |
2029 | int last_dev = sbi->s_ndevs - 1; | |
2030 | int dev_segs = | |
2031 | le32_to_cpu(raw_sb->devs[last_dev].total_segments); | |
2032 | ||
2033 | raw_sb->devs[last_dev].total_segments = | |
2034 | cpu_to_le32(dev_segs + segs); | |
2035 | } | |
a4ba5dfc | 2036 | |
e4544b63 | 2037 | f2fs_up_write(&sbi->sb_lock); |
04f0b2ea QS |
2038 | } |
2039 | ||
2040 | static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs) | |
2041 | { | |
2042 | int segs = secs * sbi->segs_per_sec; | |
46d9ce19 | 2043 | long long blks = (long long)segs * sbi->blocks_per_seg; |
04f0b2ea QS |
2044 | long long user_block_count = |
2045 | le64_to_cpu(F2FS_CKPT(sbi)->user_block_count); | |
2046 | ||
2047 | SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs; | |
2048 | MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs; | |
b4b10061 | 2049 | MAIN_SECS(sbi) += secs; |
04f0b2ea QS |
2050 | FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs; |
2051 | FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs; | |
46d9ce19 QS |
2052 | F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks); |
2053 | ||
2054 | if (f2fs_is_multi_device(sbi)) { | |
2055 | int last_dev = sbi->s_ndevs - 1; | |
2056 | ||
2057 | FDEV(last_dev).total_segments = | |
2058 | (int)FDEV(last_dev).total_segments + segs; | |
2059 | FDEV(last_dev).end_blk = | |
2060 | (long long)FDEV(last_dev).end_blk + blks; | |
2061 | #ifdef CONFIG_BLK_DEV_ZONED | |
2062 | FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz + | |
2063 | (int)(blks >> sbi->log_blocks_per_blkz); | |
2064 | #endif | |
2065 | } | |
04f0b2ea QS |
2066 | } |
2067 | ||
2068 | int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count) | |
2069 | { | |
2070 | __u64 old_block_count, shrunk_blocks; | |
b4b10061 | 2071 | struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; |
04f0b2ea | 2072 | unsigned int secs; |
04f0b2ea QS |
2073 | int err = 0; |
2074 | __u32 rem; | |
2075 | ||
2076 | old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count); | |
2077 | if (block_count > old_block_count) | |
2078 | return -EINVAL; | |
2079 | ||
46d9ce19 QS |
2080 | if (f2fs_is_multi_device(sbi)) { |
2081 | int last_dev = sbi->s_ndevs - 1; | |
2082 | __u64 last_segs = FDEV(last_dev).total_segments; | |
2083 | ||
2084 | if (block_count + last_segs * sbi->blocks_per_seg <= | |
2085 | old_block_count) | |
2086 | return -EINVAL; | |
2087 | } | |
2088 | ||
04f0b2ea QS |
2089 | /* new fs size should align to section size */ |
2090 | div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem); | |
2091 | if (rem) | |
2092 | return -EINVAL; | |
2093 | ||
2094 | if (block_count == old_block_count) | |
2095 | return 0; | |
2096 | ||
2097 | if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { | |
dcbb4c10 | 2098 | f2fs_err(sbi, "Should run fsck to repair first."); |
10f966bb | 2099 | return -EFSCORRUPTED; |
04f0b2ea QS |
2100 | } |
2101 | ||
2102 | if (test_opt(sbi, DISABLE_CHECKPOINT)) { | |
dcbb4c10 | 2103 | f2fs_err(sbi, "Checkpoint should be enabled."); |
04f0b2ea QS |
2104 | return -EINVAL; |
2105 | } | |
2106 | ||
04f0b2ea QS |
2107 | shrunk_blocks = old_block_count - block_count; |
2108 | secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi)); | |
b4b10061 JK |
2109 | |
2110 | /* stop other GC */ | |
e4544b63 | 2111 | if (!f2fs_down_write_trylock(&sbi->gc_lock)) |
b4b10061 JK |
2112 | return -EAGAIN; |
2113 | ||
2114 | /* stop CP to protect MAIN_SEC in free_segment_range */ | |
2115 | f2fs_lock_op(sbi); | |
3ab0598e CY |
2116 | |
2117 | spin_lock(&sbi->stat_lock); | |
2118 | if (shrunk_blocks + valid_user_blocks(sbi) + | |
2119 | sbi->current_reserved_blocks + sbi->unusable_block_count + | |
2120 | F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) | |
2121 | err = -ENOSPC; | |
2122 | spin_unlock(&sbi->stat_lock); | |
2123 | ||
2124 | if (err) | |
2125 | goto out_unlock; | |
2126 | ||
b4b10061 | 2127 | err = free_segment_range(sbi, secs, true); |
3ab0598e CY |
2128 | |
2129 | out_unlock: | |
b4b10061 | 2130 | f2fs_unlock_op(sbi); |
e4544b63 | 2131 | f2fs_up_write(&sbi->gc_lock); |
b4b10061 JK |
2132 | if (err) |
2133 | return err; | |
2134 | ||
b4b10061 | 2135 | freeze_super(sbi->sb); |
e4544b63 TM |
2136 | f2fs_down_write(&sbi->gc_lock); |
2137 | f2fs_down_write(&sbi->cp_global_sem); | |
b4b10061 | 2138 | |
04f0b2ea QS |
2139 | spin_lock(&sbi->stat_lock); |
2140 | if (shrunk_blocks + valid_user_blocks(sbi) + | |
2141 | sbi->current_reserved_blocks + sbi->unusable_block_count + | |
2142 | F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) | |
2143 | err = -ENOSPC; | |
2144 | else | |
2145 | sbi->user_block_count -= shrunk_blocks; | |
2146 | spin_unlock(&sbi->stat_lock); | |
b4b10061 JK |
2147 | if (err) |
2148 | goto out_err; | |
04f0b2ea | 2149 | |
28fc4e90 | 2150 | set_sbi_flag(sbi, SBI_IS_RESIZEFS); |
b4b10061 | 2151 | err = free_segment_range(sbi, secs, false); |
04f0b2ea | 2152 | if (err) |
b4b10061 | 2153 | goto recover_out; |
04f0b2ea QS |
2154 | |
2155 | update_sb_metadata(sbi, -secs); | |
2156 | ||
2157 | err = f2fs_commit_super(sbi, false); | |
2158 | if (err) { | |
2159 | update_sb_metadata(sbi, secs); | |
b4b10061 | 2160 | goto recover_out; |
04f0b2ea QS |
2161 | } |
2162 | ||
2163 | update_fs_metadata(sbi, -secs); | |
2164 | clear_sbi_flag(sbi, SBI_IS_RESIZEFS); | |
68275682 | 2165 | set_sbi_flag(sbi, SBI_IS_DIRTY); |
68275682 | 2166 | |
b4b10061 | 2167 | err = f2fs_write_checkpoint(sbi, &cpc); |
04f0b2ea QS |
2168 | if (err) { |
2169 | update_fs_metadata(sbi, secs); | |
2170 | update_sb_metadata(sbi, secs); | |
2171 | f2fs_commit_super(sbi, false); | |
2172 | } | |
b4b10061 | 2173 | recover_out: |
28fc4e90 | 2174 | clear_sbi_flag(sbi, SBI_IS_RESIZEFS); |
04f0b2ea QS |
2175 | if (err) { |
2176 | set_sbi_flag(sbi, SBI_NEED_FSCK); | |
dcbb4c10 | 2177 | f2fs_err(sbi, "resize_fs failed, should run fsck to repair!"); |
04f0b2ea | 2178 | |
04f0b2ea QS |
2179 | spin_lock(&sbi->stat_lock); |
2180 | sbi->user_block_count += shrunk_blocks; | |
2181 | spin_unlock(&sbi->stat_lock); | |
2182 | } | |
b4b10061 | 2183 | out_err: |
e4544b63 TM |
2184 | f2fs_up_write(&sbi->cp_global_sem); |
2185 | f2fs_up_write(&sbi->gc_lock); | |
b4b10061 | 2186 | thaw_super(sbi->sb); |
04f0b2ea QS |
2187 | return err; |
2188 | } |