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> | |
10 | #include <linux/backing-dev.h> | |
7bc09003 JK |
11 | #include <linux/init.h> |
12 | #include <linux/f2fs_fs.h> | |
13 | #include <linux/kthread.h> | |
14 | #include <linux/delay.h> | |
15 | #include <linux/freezer.h> | |
b4b10061 | 16 | #include <linux/sched/signal.h> |
7bc09003 JK |
17 | |
18 | #include "f2fs.h" | |
19 | #include "node.h" | |
20 | #include "segment.h" | |
21 | #include "gc.h" | |
8e46b3ed | 22 | #include <trace/events/f2fs.h> |
7bc09003 | 23 | |
093749e2 CY |
24 | static struct kmem_cache *victim_entry_slab; |
25 | ||
da52f8ad JQ |
26 | static unsigned int count_bits(const unsigned long *addr, |
27 | unsigned int offset, unsigned int len); | |
28 | ||
7bc09003 JK |
29 | static int gc_thread_func(void *data) |
30 | { | |
31 | struct f2fs_sb_info *sbi = data; | |
b59d0bae | 32 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; |
7bc09003 | 33 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; |
5911d2d1 | 34 | wait_queue_head_t *fggc_wq = &sbi->gc_thread->fggc_wq; |
b8c502b8 | 35 | unsigned int wait_ms; |
7bc09003 | 36 | |
b59d0bae | 37 | wait_ms = gc_th->min_sleep_time; |
7bc09003 | 38 | |
1d7be270 | 39 | set_freezable(); |
7bc09003 | 40 | do { |
5911d2d1 | 41 | bool sync_mode, foreground = false; |
bbbc34fd | 42 | |
1d7be270 | 43 | wait_event_interruptible_timeout(*wq, |
d9872a69 | 44 | kthread_should_stop() || freezing(current) || |
5911d2d1 | 45 | waitqueue_active(fggc_wq) || |
d9872a69 | 46 | gc_th->gc_wake, |
1d7be270 JK |
47 | msecs_to_jiffies(wait_ms)); |
48 | ||
5911d2d1 CY |
49 | if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq)) |
50 | foreground = true; | |
51 | ||
d9872a69 JK |
52 | /* give it a try one time */ |
53 | if (gc_th->gc_wake) | |
54 | gc_th->gc_wake = 0; | |
55 | ||
274bd9ba CY |
56 | if (try_to_freeze()) { |
57 | stat_other_skip_bggc_count(sbi); | |
7bc09003 | 58 | continue; |
274bd9ba | 59 | } |
7bc09003 JK |
60 | if (kthread_should_stop()) |
61 | break; | |
62 | ||
d6212a5f | 63 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
88dd8934 | 64 | increase_sleep_time(gc_th, &wait_ms); |
274bd9ba | 65 | stat_other_skip_bggc_count(sbi); |
d6212a5f CL |
66 | continue; |
67 | } | |
68 | ||
55523519 | 69 | if (time_to_inject(sbi, FAULT_CHECKPOINT)) { |
c45d6002 | 70 | f2fs_show_injection_info(sbi, FAULT_CHECKPOINT); |
0f348028 | 71 | f2fs_stop_checkpoint(sbi, false); |
55523519 | 72 | } |
0f348028 | 73 | |
274bd9ba CY |
74 | if (!sb_start_write_trylock(sbi->sb)) { |
75 | stat_other_skip_bggc_count(sbi); | |
dc6febb6 | 76 | continue; |
274bd9ba | 77 | } |
dc6febb6 | 78 | |
7bc09003 JK |
79 | /* |
80 | * [GC triggering condition] | |
81 | * 0. GC is not conducted currently. | |
82 | * 1. There are enough dirty segments. | |
83 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
84 | * 3. IO subsystem is idle by checking the # of requests in | |
85 | * bdev's request list. | |
86 | * | |
e1c42045 | 87 | * Note) We have to avoid triggering GCs frequently. |
7bc09003 JK |
88 | * Because it is possible that some segments can be |
89 | * invalidated soon after by user update or deletion. | |
90 | * So, I'd like to wait some time to collect dirty segments. | |
91 | */ | |
0e5e8111 | 92 | if (sbi->gc_mode == GC_URGENT_HIGH) { |
d9872a69 | 93 | wait_ms = gc_th->urgent_sleep_time; |
fb24fea7 | 94 | down_write(&sbi->gc_lock); |
d9872a69 JK |
95 | goto do_gc; |
96 | } | |
97 | ||
5911d2d1 CY |
98 | if (foreground) { |
99 | down_write(&sbi->gc_lock); | |
100 | goto do_gc; | |
101 | } else if (!down_write_trylock(&sbi->gc_lock)) { | |
274bd9ba | 102 | stat_other_skip_bggc_count(sbi); |
69babac0 | 103 | goto next; |
274bd9ba | 104 | } |
69babac0 | 105 | |
a7d10cf3 | 106 | if (!is_idle(sbi, GC_TIME)) { |
88dd8934 | 107 | increase_sleep_time(gc_th, &wait_ms); |
fb24fea7 | 108 | up_write(&sbi->gc_lock); |
274bd9ba | 109 | stat_io_skip_bggc_count(sbi); |
dc6febb6 | 110 | goto next; |
7bc09003 JK |
111 | } |
112 | ||
113 | if (has_enough_invalid_blocks(sbi)) | |
88dd8934 | 114 | decrease_sleep_time(gc_th, &wait_ms); |
7bc09003 | 115 | else |
88dd8934 | 116 | increase_sleep_time(gc_th, &wait_ms); |
d9872a69 | 117 | do_gc: |
5911d2d1 CY |
118 | if (!foreground) |
119 | stat_inc_bggc_count(sbi->stat_info); | |
7bc09003 | 120 | |
bbbc34fd CY |
121 | sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC; |
122 | ||
5911d2d1 CY |
123 | /* foreground GC was been triggered via f2fs_balance_fs() */ |
124 | if (foreground) | |
125 | sync_mode = false; | |
126 | ||
43727527 | 127 | /* if return value is not zero, no victim was selected */ |
5911d2d1 | 128 | if (f2fs_gc(sbi, sync_mode, !foreground, false, NULL_SEGNO)) |
b59d0bae | 129 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 130 | |
5911d2d1 CY |
131 | if (foreground) |
132 | wake_up_all(&gc_th->fggc_wq); | |
133 | ||
84e4214f JK |
134 | trace_f2fs_background_gc(sbi->sb, wait_ms, |
135 | prefree_segments(sbi), free_segments(sbi)); | |
136 | ||
4660f9c0 | 137 | /* balancing f2fs's metadata periodically */ |
7bcd0cfa | 138 | f2fs_balance_fs_bg(sbi, true); |
dc6febb6 CY |
139 | next: |
140 | sb_end_write(sbi->sb); | |
81eb8d6e | 141 | |
7bc09003 JK |
142 | } while (!kthread_should_stop()); |
143 | return 0; | |
144 | } | |
145 | ||
4d57b86d | 146 | int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) |
7bc09003 | 147 | { |
1042d60f | 148 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 149 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 150 | int err = 0; |
7bc09003 | 151 | |
1ecc0c5c | 152 | gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
7a267f8d NJ |
153 | if (!gc_th) { |
154 | err = -ENOMEM; | |
155 | goto out; | |
156 | } | |
7bc09003 | 157 | |
d9872a69 | 158 | gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME; |
b59d0bae NJ |
159 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
160 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
161 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
162 | ||
5f029c04 | 163 | gc_th->gc_wake = 0; |
d2dc095f | 164 | |
7bc09003 JK |
165 | sbi->gc_thread = gc_th; |
166 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
5911d2d1 | 167 | init_waitqueue_head(&sbi->gc_thread->fggc_wq); |
7bc09003 | 168 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, |
ec7b1f2d | 169 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 170 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 171 | err = PTR_ERR(gc_th->f2fs_gc_task); |
c8eb7024 | 172 | kfree(gc_th); |
25718423 | 173 | sbi->gc_thread = NULL; |
7bc09003 | 174 | } |
7a267f8d NJ |
175 | out: |
176 | return err; | |
7bc09003 JK |
177 | } |
178 | ||
4d57b86d | 179 | void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) |
7bc09003 JK |
180 | { |
181 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
5f029c04 | 182 | |
7bc09003 JK |
183 | if (!gc_th) |
184 | return; | |
185 | kthread_stop(gc_th->f2fs_gc_task); | |
5911d2d1 | 186 | wake_up_all(&gc_th->fggc_wq); |
c8eb7024 | 187 | kfree(gc_th); |
7bc09003 JK |
188 | sbi->gc_thread = NULL; |
189 | } | |
190 | ||
5b0e9539 | 191 | static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) |
7bc09003 | 192 | { |
093749e2 CY |
193 | int gc_mode; |
194 | ||
195 | if (gc_type == BG_GC) { | |
196 | if (sbi->am.atgc_enabled) | |
197 | gc_mode = GC_AT; | |
198 | else | |
199 | gc_mode = GC_CB; | |
200 | } else { | |
201 | gc_mode = GC_GREEDY; | |
202 | } | |
d2dc095f | 203 | |
5b0e9539 JK |
204 | switch (sbi->gc_mode) { |
205 | case GC_IDLE_CB: | |
206 | gc_mode = GC_CB; | |
207 | break; | |
208 | case GC_IDLE_GREEDY: | |
0e5e8111 | 209 | case GC_URGENT_HIGH: |
b27bc809 | 210 | gc_mode = GC_GREEDY; |
5b0e9539 | 211 | break; |
093749e2 CY |
212 | case GC_IDLE_AT: |
213 | gc_mode = GC_AT; | |
214 | break; | |
5b0e9539 | 215 | } |
093749e2 | 216 | |
d2dc095f | 217 | return gc_mode; |
7bc09003 JK |
218 | } |
219 | ||
220 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
221 | int type, struct victim_sel_policy *p) | |
222 | { | |
223 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
224 | ||
4ebefc44 | 225 | if (p->alloc_mode == SSR) { |
7bc09003 | 226 | p->gc_mode = GC_GREEDY; |
da52f8ad | 227 | p->dirty_bitmap = dirty_i->dirty_segmap[type]; |
a26b7c8a | 228 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 | 229 | p->ofs_unit = 1; |
093749e2 CY |
230 | } else if (p->alloc_mode == AT_SSR) { |
231 | p->gc_mode = GC_GREEDY; | |
232 | p->dirty_bitmap = dirty_i->dirty_segmap[type]; | |
233 | p->max_search = dirty_i->nr_dirty[type]; | |
234 | p->ofs_unit = 1; | |
7bc09003 | 235 | } else { |
5b0e9539 | 236 | p->gc_mode = select_gc_type(sbi, gc_type); |
7bc09003 | 237 | p->ofs_unit = sbi->segs_per_sec; |
da52f8ad JQ |
238 | if (__is_large_section(sbi)) { |
239 | p->dirty_bitmap = dirty_i->dirty_secmap; | |
240 | p->max_search = count_bits(p->dirty_bitmap, | |
241 | 0, MAIN_SECS(sbi)); | |
242 | } else { | |
243 | p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY]; | |
244 | p->max_search = dirty_i->nr_dirty[DIRTY]; | |
245 | } | |
7bc09003 | 246 | } |
a26b7c8a | 247 | |
7a88ddb5 CY |
248 | /* |
249 | * adjust candidates range, should select all dirty segments for | |
250 | * foreground GC and urgent GC cases. | |
251 | */ | |
b27bc809 | 252 | if (gc_type != FG_GC && |
0e5e8111 | 253 | (sbi->gc_mode != GC_URGENT_HIGH) && |
093749e2 | 254 | (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) && |
b27bc809 | 255 | p->max_search > sbi->max_victim_search) |
b1c57c1c | 256 | p->max_search = sbi->max_victim_search; |
a26b7c8a | 257 | |
b94929d9 YS |
258 | /* let's select beginning hot/small space first in no_heap mode*/ |
259 | if (test_opt(sbi, NOHEAP) && | |
260 | (type == CURSEG_HOT_DATA || IS_NODESEG(type))) | |
7a20b8a6 JK |
261 | p->offset = 0; |
262 | else | |
e066b83c | 263 | p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; |
7bc09003 JK |
264 | } |
265 | ||
266 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
267 | struct victim_sel_policy *p) | |
268 | { | |
b7250d2d JK |
269 | /* SSR allocates in a segment unit */ |
270 | if (p->alloc_mode == SSR) | |
3519e3f9 | 271 | return sbi->blocks_per_seg; |
093749e2 CY |
272 | else if (p->alloc_mode == AT_SSR) |
273 | return UINT_MAX; | |
274 | ||
275 | /* LFS */ | |
7bc09003 | 276 | if (p->gc_mode == GC_GREEDY) |
c541a51b | 277 | return 2 * sbi->blocks_per_seg * p->ofs_unit; |
7bc09003 JK |
278 | else if (p->gc_mode == GC_CB) |
279 | return UINT_MAX; | |
093749e2 CY |
280 | else if (p->gc_mode == GC_AT) |
281 | return UINT_MAX; | |
7bc09003 JK |
282 | else /* No other gc_mode */ |
283 | return 0; | |
284 | } | |
285 | ||
286 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
287 | { | |
288 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 289 | unsigned int secno; |
7bc09003 JK |
290 | |
291 | /* | |
292 | * If the gc_type is FG_GC, we can select victim segments | |
293 | * selected by background GC before. | |
294 | * Those segments guarantee they have small valid blocks. | |
295 | */ | |
7cd8558b | 296 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 297 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 298 | continue; |
5ec4e49f | 299 | clear_bit(secno, dirty_i->victim_secmap); |
4ddb1a4d | 300 | return GET_SEG_FROM_SEC(sbi, secno); |
7bc09003 JK |
301 | } |
302 | return NULL_SEGNO; | |
303 | } | |
304 | ||
305 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
306 | { | |
307 | struct sit_info *sit_i = SIT_I(sbi); | |
4ddb1a4d JK |
308 | unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); |
309 | unsigned int start = GET_SEG_FROM_SEC(sbi, secno); | |
7bc09003 JK |
310 | unsigned long long mtime = 0; |
311 | unsigned int vblocks; | |
312 | unsigned char age = 0; | |
313 | unsigned char u; | |
314 | unsigned int i; | |
de881df9 | 315 | unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno); |
7bc09003 | 316 | |
de881df9 | 317 | for (i = 0; i < usable_segs_per_sec; i++) |
7bc09003 | 318 | mtime += get_seg_entry(sbi, start + i)->mtime; |
302bd348 | 319 | vblocks = get_valid_blocks(sbi, segno, true); |
7bc09003 | 320 | |
de881df9 AR |
321 | mtime = div_u64(mtime, usable_segs_per_sec); |
322 | vblocks = div_u64(vblocks, usable_segs_per_sec); | |
7bc09003 JK |
323 | |
324 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
325 | ||
e1c42045 | 326 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
327 | if (mtime < sit_i->min_mtime) |
328 | sit_i->min_mtime = mtime; | |
329 | if (mtime > sit_i->max_mtime) | |
330 | sit_i->max_mtime = mtime; | |
331 | if (sit_i->max_mtime != sit_i->min_mtime) | |
332 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
333 | sit_i->max_mtime - sit_i->min_mtime); | |
334 | ||
335 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
336 | } | |
337 | ||
a57e564d JX |
338 | static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, |
339 | unsigned int segno, struct victim_sel_policy *p) | |
7bc09003 JK |
340 | { |
341 | if (p->alloc_mode == SSR) | |
2afce76a | 342 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; |
7bc09003 JK |
343 | |
344 | /* alloc_mode == LFS */ | |
345 | if (p->gc_mode == GC_GREEDY) | |
91f4382b | 346 | return get_valid_blocks(sbi, segno, true); |
093749e2 | 347 | else if (p->gc_mode == GC_CB) |
7bc09003 | 348 | return get_cb_cost(sbi, segno); |
093749e2 CY |
349 | |
350 | f2fs_bug_on(sbi, 1); | |
351 | return 0; | |
7bc09003 JK |
352 | } |
353 | ||
688159b6 FL |
354 | static unsigned int count_bits(const unsigned long *addr, |
355 | unsigned int offset, unsigned int len) | |
356 | { | |
357 | unsigned int end = offset + len, sum = 0; | |
358 | ||
359 | while (offset < end) { | |
360 | if (test_bit(offset++, addr)) | |
361 | ++sum; | |
362 | } | |
363 | return sum; | |
364 | } | |
365 | ||
093749e2 CY |
366 | static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi, |
367 | unsigned long long mtime, unsigned int segno, | |
368 | struct rb_node *parent, struct rb_node **p, | |
369 | bool left_most) | |
370 | { | |
371 | struct atgc_management *am = &sbi->am; | |
372 | struct victim_entry *ve; | |
373 | ||
374 | ve = f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS); | |
375 | ||
376 | ve->mtime = mtime; | |
377 | ve->segno = segno; | |
378 | ||
379 | rb_link_node(&ve->rb_node, parent, p); | |
380 | rb_insert_color_cached(&ve->rb_node, &am->root, left_most); | |
381 | ||
382 | list_add_tail(&ve->list, &am->victim_list); | |
383 | ||
384 | am->victim_count++; | |
385 | ||
386 | return ve; | |
387 | } | |
388 | ||
389 | static void insert_victim_entry(struct f2fs_sb_info *sbi, | |
390 | unsigned long long mtime, unsigned int segno) | |
391 | { | |
392 | struct atgc_management *am = &sbi->am; | |
393 | struct rb_node **p; | |
394 | struct rb_node *parent = NULL; | |
395 | bool left_most = true; | |
396 | ||
397 | p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most); | |
398 | attach_victim_entry(sbi, mtime, segno, parent, p, left_most); | |
399 | } | |
400 | ||
401 | static void add_victim_entry(struct f2fs_sb_info *sbi, | |
402 | struct victim_sel_policy *p, unsigned int segno) | |
403 | { | |
404 | struct sit_info *sit_i = SIT_I(sbi); | |
405 | unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); | |
406 | unsigned int start = GET_SEG_FROM_SEC(sbi, secno); | |
407 | unsigned long long mtime = 0; | |
408 | unsigned int i; | |
409 | ||
410 | if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { | |
411 | if (p->gc_mode == GC_AT && | |
412 | get_valid_blocks(sbi, segno, true) == 0) | |
413 | return; | |
093749e2 CY |
414 | } |
415 | ||
416 | for (i = 0; i < sbi->segs_per_sec; i++) | |
417 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
418 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
419 | ||
420 | /* Handle if the system time has changed by the user */ | |
421 | if (mtime < sit_i->min_mtime) | |
422 | sit_i->min_mtime = mtime; | |
423 | if (mtime > sit_i->max_mtime) | |
424 | sit_i->max_mtime = mtime; | |
425 | if (mtime < sit_i->dirty_min_mtime) | |
426 | sit_i->dirty_min_mtime = mtime; | |
427 | if (mtime > sit_i->dirty_max_mtime) | |
428 | sit_i->dirty_max_mtime = mtime; | |
429 | ||
430 | /* don't choose young section as candidate */ | |
431 | if (sit_i->dirty_max_mtime - mtime < p->age_threshold) | |
432 | return; | |
433 | ||
434 | insert_victim_entry(sbi, mtime, segno); | |
435 | } | |
436 | ||
437 | static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi, | |
438 | struct victim_sel_policy *p) | |
439 | { | |
440 | struct atgc_management *am = &sbi->am; | |
441 | struct rb_node *parent = NULL; | |
442 | bool left_most; | |
443 | ||
444 | f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most); | |
445 | ||
446 | return parent; | |
447 | } | |
448 | ||
449 | static void atgc_lookup_victim(struct f2fs_sb_info *sbi, | |
450 | struct victim_sel_policy *p) | |
451 | { | |
452 | struct sit_info *sit_i = SIT_I(sbi); | |
453 | struct atgc_management *am = &sbi->am; | |
454 | struct rb_root_cached *root = &am->root; | |
455 | struct rb_node *node; | |
456 | struct rb_entry *re; | |
457 | struct victim_entry *ve; | |
458 | unsigned long long total_time; | |
459 | unsigned long long age, u, accu; | |
460 | unsigned long long max_mtime = sit_i->dirty_max_mtime; | |
461 | unsigned long long min_mtime = sit_i->dirty_min_mtime; | |
462 | unsigned int sec_blocks = BLKS_PER_SEC(sbi); | |
463 | unsigned int vblocks; | |
464 | unsigned int dirty_threshold = max(am->max_candidate_count, | |
465 | am->candidate_ratio * | |
466 | am->victim_count / 100); | |
467 | unsigned int age_weight = am->age_weight; | |
468 | unsigned int cost; | |
469 | unsigned int iter = 0; | |
470 | ||
471 | if (max_mtime < min_mtime) | |
472 | return; | |
473 | ||
474 | max_mtime += 1; | |
475 | total_time = max_mtime - min_mtime; | |
476 | ||
477 | accu = div64_u64(ULLONG_MAX, total_time); | |
478 | accu = min_t(unsigned long long, div_u64(accu, 100), | |
479 | DEFAULT_ACCURACY_CLASS); | |
480 | ||
481 | node = rb_first_cached(root); | |
482 | next: | |
483 | re = rb_entry_safe(node, struct rb_entry, rb_node); | |
484 | if (!re) | |
485 | return; | |
486 | ||
487 | ve = (struct victim_entry *)re; | |
488 | ||
489 | if (ve->mtime >= max_mtime || ve->mtime < min_mtime) | |
490 | goto skip; | |
491 | ||
492 | /* age = 10000 * x% * 60 */ | |
493 | age = div64_u64(accu * (max_mtime - ve->mtime), total_time) * | |
494 | age_weight; | |
495 | ||
496 | vblocks = get_valid_blocks(sbi, ve->segno, true); | |
497 | f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks); | |
498 | ||
499 | /* u = 10000 * x% * 40 */ | |
500 | u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) * | |
501 | (100 - age_weight); | |
502 | ||
503 | f2fs_bug_on(sbi, age + u >= UINT_MAX); | |
504 | ||
505 | cost = UINT_MAX - (age + u); | |
506 | iter++; | |
507 | ||
508 | if (cost < p->min_cost || | |
509 | (cost == p->min_cost && age > p->oldest_age)) { | |
510 | p->min_cost = cost; | |
511 | p->oldest_age = age; | |
512 | p->min_segno = ve->segno; | |
513 | } | |
514 | skip: | |
515 | if (iter < dirty_threshold) { | |
516 | node = rb_next(node); | |
517 | goto next; | |
518 | } | |
519 | } | |
520 | ||
521 | /* | |
522 | * select candidates around source section in range of | |
523 | * [target - dirty_threshold, target + dirty_threshold] | |
524 | */ | |
525 | static void atssr_lookup_victim(struct f2fs_sb_info *sbi, | |
526 | struct victim_sel_policy *p) | |
527 | { | |
528 | struct sit_info *sit_i = SIT_I(sbi); | |
529 | struct atgc_management *am = &sbi->am; | |
530 | struct rb_node *node; | |
531 | struct rb_entry *re; | |
532 | struct victim_entry *ve; | |
533 | unsigned long long age; | |
534 | unsigned long long max_mtime = sit_i->dirty_max_mtime; | |
535 | unsigned long long min_mtime = sit_i->dirty_min_mtime; | |
536 | unsigned int seg_blocks = sbi->blocks_per_seg; | |
537 | unsigned int vblocks; | |
538 | unsigned int dirty_threshold = max(am->max_candidate_count, | |
539 | am->candidate_ratio * | |
540 | am->victim_count / 100); | |
541 | unsigned int cost; | |
542 | unsigned int iter = 0; | |
543 | int stage = 0; | |
544 | ||
545 | if (max_mtime < min_mtime) | |
546 | return; | |
547 | max_mtime += 1; | |
548 | next_stage: | |
549 | node = lookup_central_victim(sbi, p); | |
550 | next_node: | |
551 | re = rb_entry_safe(node, struct rb_entry, rb_node); | |
552 | if (!re) { | |
553 | if (stage == 0) | |
554 | goto skip_stage; | |
555 | return; | |
556 | } | |
557 | ||
558 | ve = (struct victim_entry *)re; | |
559 | ||
560 | if (ve->mtime >= max_mtime || ve->mtime < min_mtime) | |
561 | goto skip_node; | |
562 | ||
563 | age = max_mtime - ve->mtime; | |
564 | ||
565 | vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks; | |
566 | f2fs_bug_on(sbi, !vblocks); | |
567 | ||
568 | /* rare case */ | |
569 | if (vblocks == seg_blocks) | |
570 | goto skip_node; | |
571 | ||
572 | iter++; | |
573 | ||
574 | age = max_mtime - abs(p->age - age); | |
575 | cost = UINT_MAX - vblocks; | |
576 | ||
577 | if (cost < p->min_cost || | |
578 | (cost == p->min_cost && age > p->oldest_age)) { | |
579 | p->min_cost = cost; | |
580 | p->oldest_age = age; | |
581 | p->min_segno = ve->segno; | |
582 | } | |
583 | skip_node: | |
584 | if (iter < dirty_threshold) { | |
585 | if (stage == 0) | |
586 | node = rb_prev(node); | |
587 | else if (stage == 1) | |
588 | node = rb_next(node); | |
589 | goto next_node; | |
590 | } | |
591 | skip_stage: | |
592 | if (stage < 1) { | |
593 | stage++; | |
594 | iter = 0; | |
595 | goto next_stage; | |
596 | } | |
597 | } | |
598 | static void lookup_victim_by_age(struct f2fs_sb_info *sbi, | |
599 | struct victim_sel_policy *p) | |
600 | { | |
601 | f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, | |
602 | &sbi->am.root, true)); | |
603 | ||
604 | if (p->gc_mode == GC_AT) | |
605 | atgc_lookup_victim(sbi, p); | |
606 | else if (p->alloc_mode == AT_SSR) | |
607 | atssr_lookup_victim(sbi, p); | |
608 | else | |
609 | f2fs_bug_on(sbi, 1); | |
610 | } | |
611 | ||
612 | static void release_victim_entry(struct f2fs_sb_info *sbi) | |
613 | { | |
614 | struct atgc_management *am = &sbi->am; | |
615 | struct victim_entry *ve, *tmp; | |
616 | ||
617 | list_for_each_entry_safe(ve, tmp, &am->victim_list, list) { | |
618 | list_del(&ve->list); | |
619 | kmem_cache_free(victim_entry_slab, ve); | |
620 | am->victim_count--; | |
621 | } | |
622 | ||
623 | am->root = RB_ROOT_CACHED; | |
624 | ||
625 | f2fs_bug_on(sbi, am->victim_count); | |
626 | f2fs_bug_on(sbi, !list_empty(&am->victim_list)); | |
627 | } | |
628 | ||
0a8165d7 | 629 | /* |
111d2495 | 630 | * This function is called from two paths. |
7bc09003 JK |
631 | * One is garbage collection and the other is SSR segment selection. |
632 | * When it is called during GC, it just gets a victim segment | |
633 | * and it does not remove it from dirty seglist. | |
634 | * When it is called from SSR segment selection, it finds a segment | |
635 | * which has minimum valid blocks and removes it from dirty seglist. | |
636 | */ | |
637 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
093749e2 CY |
638 | unsigned int *result, int gc_type, int type, |
639 | char alloc_mode, unsigned long long age) | |
7bc09003 JK |
640 | { |
641 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
e066b83c | 642 | struct sit_info *sm = SIT_I(sbi); |
7bc09003 | 643 | struct victim_sel_policy p; |
3fa56503 | 644 | unsigned int secno, last_victim; |
04f0b2ea | 645 | unsigned int last_segment; |
093749e2 CY |
646 | unsigned int nsearched; |
647 | bool is_atgc; | |
97767500 | 648 | int ret = 0; |
7bc09003 | 649 | |
210f41bc | 650 | mutex_lock(&dirty_i->seglist_lock); |
04f0b2ea | 651 | last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec; |
210f41bc | 652 | |
7bc09003 | 653 | p.alloc_mode = alloc_mode; |
093749e2 CY |
654 | p.age = age; |
655 | p.age_threshold = sbi->am.age_threshold; | |
7bc09003 | 656 | |
093749e2 CY |
657 | retry: |
658 | select_policy(sbi, gc_type, type, &p); | |
7bc09003 | 659 | p.min_segno = NULL_SEGNO; |
093749e2 | 660 | p.oldest_age = 0; |
3fa56503 | 661 | p.min_cost = get_max_cost(sbi, &p); |
7bc09003 | 662 | |
093749e2 CY |
663 | is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR); |
664 | nsearched = 0; | |
665 | ||
666 | if (is_atgc) | |
667 | SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX; | |
668 | ||
e066b83c | 669 | if (*result != NULL_SEGNO) { |
97767500 QZ |
670 | if (!get_valid_blocks(sbi, *result, false)) { |
671 | ret = -ENODATA; | |
672 | goto out; | |
673 | } | |
674 | ||
675 | if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) | |
676 | ret = -EBUSY; | |
677 | else | |
e066b83c JK |
678 | p.min_segno = *result; |
679 | goto out; | |
680 | } | |
681 | ||
97767500 | 682 | ret = -ENODATA; |
3342bb30 CY |
683 | if (p.max_search == 0) |
684 | goto out; | |
685 | ||
e3080b01 CY |
686 | if (__is_large_section(sbi) && p.alloc_mode == LFS) { |
687 | if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) { | |
688 | p.min_segno = sbi->next_victim_seg[BG_GC]; | |
689 | *result = p.min_segno; | |
690 | sbi->next_victim_seg[BG_GC] = NULL_SEGNO; | |
691 | goto got_result; | |
692 | } | |
693 | if (gc_type == FG_GC && | |
694 | sbi->next_victim_seg[FG_GC] != NULL_SEGNO) { | |
695 | p.min_segno = sbi->next_victim_seg[FG_GC]; | |
696 | *result = p.min_segno; | |
697 | sbi->next_victim_seg[FG_GC] = NULL_SEGNO; | |
698 | goto got_result; | |
699 | } | |
700 | } | |
701 | ||
e066b83c | 702 | last_victim = sm->last_victim[p.gc_mode]; |
7bc09003 JK |
703 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
704 | p.min_segno = check_bg_victims(sbi); | |
705 | if (p.min_segno != NULL_SEGNO) | |
706 | goto got_it; | |
707 | } | |
708 | ||
709 | while (1) { | |
da52f8ad JQ |
710 | unsigned long cost, *dirty_bitmap; |
711 | unsigned int unit_no, segno; | |
712 | ||
713 | dirty_bitmap = p.dirty_bitmap; | |
714 | unit_no = find_next_bit(dirty_bitmap, | |
715 | last_segment / p.ofs_unit, | |
716 | p.offset / p.ofs_unit); | |
717 | segno = unit_no * p.ofs_unit; | |
a43f7ec3 | 718 | if (segno >= last_segment) { |
e066b83c JK |
719 | if (sm->last_victim[p.gc_mode]) { |
720 | last_segment = | |
721 | sm->last_victim[p.gc_mode]; | |
722 | sm->last_victim[p.gc_mode] = 0; | |
7bc09003 JK |
723 | p.offset = 0; |
724 | continue; | |
725 | } | |
726 | break; | |
727 | } | |
a57e564d JX |
728 | |
729 | p.offset = segno + p.ofs_unit; | |
da52f8ad | 730 | nsearched++; |
688159b6 | 731 | |
bbf9f7d9 ST |
732 | #ifdef CONFIG_F2FS_CHECK_FS |
733 | /* | |
734 | * skip selecting the invalid segno (that is failed due to block | |
735 | * validity check failure during GC) to avoid endless GC loop in | |
736 | * such cases. | |
737 | */ | |
738 | if (test_bit(segno, sm->invalid_segmap)) | |
739 | goto next; | |
740 | #endif | |
741 | ||
4ddb1a4d | 742 | secno = GET_SEC_FROM_SEG(sbi, segno); |
7bc09003 | 743 | |
5ec4e49f | 744 | if (sec_usage_check(sbi, secno)) |
688159b6 | 745 | goto next; |
61461fc9 | 746 | |
4354994f | 747 | /* Don't touch checkpointed data */ |
61461fc9 CY |
748 | if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { |
749 | if (p.alloc_mode == LFS) { | |
750 | /* | |
751 | * LFS is set to find source section during GC. | |
752 | * The victim should have no checkpointed data. | |
753 | */ | |
754 | if (get_ckpt_valid_blocks(sbi, segno, true)) | |
755 | goto next; | |
756 | } else { | |
757 | /* | |
758 | * SSR | AT_SSR are set to find target segment | |
759 | * for writes which can be full by checkpointed | |
760 | * and newly written blocks. | |
761 | */ | |
762 | if (!f2fs_segment_has_free_slot(sbi, segno)) | |
763 | goto next; | |
764 | } | |
765 | } | |
766 | ||
5ec4e49f | 767 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
688159b6 | 768 | goto next; |
7bc09003 | 769 | |
093749e2 CY |
770 | if (is_atgc) { |
771 | add_victim_entry(sbi, &p, segno); | |
772 | goto next; | |
773 | } | |
774 | ||
7bc09003 JK |
775 | cost = get_gc_cost(sbi, segno, &p); |
776 | ||
777 | if (p.min_cost > cost) { | |
778 | p.min_segno = segno; | |
779 | p.min_cost = cost; | |
a57e564d | 780 | } |
688159b6 FL |
781 | next: |
782 | if (nsearched >= p.max_search) { | |
e066b83c | 783 | if (!sm->last_victim[p.gc_mode] && segno <= last_victim) |
da52f8ad JQ |
784 | sm->last_victim[p.gc_mode] = |
785 | last_victim + p.ofs_unit; | |
4ce53776 | 786 | else |
da52f8ad | 787 | sm->last_victim[p.gc_mode] = segno + p.ofs_unit; |
04f0b2ea QS |
788 | sm->last_victim[p.gc_mode] %= |
789 | (MAIN_SECS(sbi) * sbi->segs_per_sec); | |
7bc09003 JK |
790 | break; |
791 | } | |
792 | } | |
093749e2 CY |
793 | |
794 | /* get victim for GC_AT/AT_SSR */ | |
795 | if (is_atgc) { | |
796 | lookup_victim_by_age(sbi, &p); | |
797 | release_victim_entry(sbi); | |
798 | } | |
799 | ||
800 | if (is_atgc && p.min_segno == NULL_SEGNO && | |
801 | sm->elapsed_time < p.age_threshold) { | |
802 | p.age_threshold = 0; | |
803 | goto retry; | |
804 | } | |
805 | ||
7bc09003 | 806 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 807 | got_it: |
e3080b01 CY |
808 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
809 | got_result: | |
7bc09003 | 810 | if (p.alloc_mode == LFS) { |
4ddb1a4d | 811 | secno = GET_SEC_FROM_SEG(sbi, p.min_segno); |
5ec4e49f JK |
812 | if (gc_type == FG_GC) |
813 | sbi->cur_victim_sec = secno; | |
814 | else | |
815 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 816 | } |
97767500 | 817 | ret = 0; |
8e46b3ed | 818 | |
e3c59108 ST |
819 | } |
820 | out: | |
821 | if (p.min_segno != NULL_SEGNO) | |
8e46b3ed NJ |
822 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, |
823 | sbi->cur_victim_sec, | |
824 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 JK |
825 | mutex_unlock(&dirty_i->seglist_lock); |
826 | ||
97767500 | 827 | return ret; |
7bc09003 JK |
828 | } |
829 | ||
830 | static const struct victim_selection default_v_ops = { | |
831 | .get_victim = get_victim_by_default, | |
832 | }; | |
833 | ||
7dda2af8 | 834 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 835 | { |
7bc09003 JK |
836 | struct inode_entry *ie; |
837 | ||
7dda2af8 CL |
838 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
839 | if (ie) | |
840 | return ie->inode; | |
7bc09003 JK |
841 | return NULL; |
842 | } | |
843 | ||
7dda2af8 | 844 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 845 | { |
6cc4af56 GZ |
846 | struct inode_entry *new_ie; |
847 | ||
7dda2af8 | 848 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
849 | iput(inode); |
850 | return; | |
7bc09003 | 851 | } |
4d57b86d | 852 | new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS); |
7bc09003 | 853 | new_ie->inode = inode; |
f28e5034 CY |
854 | |
855 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 856 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
857 | } |
858 | ||
7dda2af8 | 859 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
860 | { |
861 | struct inode_entry *ie, *next_ie; | |
5f029c04 | 862 | |
7dda2af8 CL |
863 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
864 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
865 | iput(ie->inode); |
866 | list_del(&ie->list); | |
4d57b86d | 867 | kmem_cache_free(f2fs_inode_entry_slab, ie); |
7bc09003 JK |
868 | } |
869 | } | |
870 | ||
871 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
872 | unsigned int segno, int offset) | |
873 | { | |
874 | struct sit_info *sit_i = SIT_I(sbi); | |
875 | struct seg_entry *sentry; | |
876 | int ret; | |
877 | ||
3d26fa6b | 878 | down_read(&sit_i->sentry_lock); |
7bc09003 JK |
879 | sentry = get_seg_entry(sbi, segno); |
880 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
3d26fa6b | 881 | up_read(&sit_i->sentry_lock); |
43727527 | 882 | return ret; |
7bc09003 JK |
883 | } |
884 | ||
0a8165d7 | 885 | /* |
7bc09003 JK |
886 | * This function compares node address got in summary with that in NAT. |
887 | * On validity, copy that node with cold status, otherwise (invalid node) | |
888 | * ignore that. | |
889 | */ | |
48018b4c | 890 | static int gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
891 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
892 | { | |
7bc09003 | 893 | struct f2fs_summary *entry; |
26d58599 | 894 | block_t start_addr; |
7bc09003 | 895 | int off; |
7ea984b0 | 896 | int phase = 0; |
c29fd0c0 | 897 | bool fggc = (gc_type == FG_GC); |
48018b4c | 898 | int submitted = 0; |
de881df9 | 899 | unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); |
7bc09003 | 900 | |
26d58599 JK |
901 | start_addr = START_BLOCK(sbi, segno); |
902 | ||
7bc09003 JK |
903 | next_step: |
904 | entry = sum; | |
c718379b | 905 | |
c29fd0c0 CY |
906 | if (fggc && phase == 2) |
907 | atomic_inc(&sbi->wb_sync_req[NODE]); | |
908 | ||
de881df9 | 909 | for (off = 0; off < usable_blks_in_seg; off++, entry++) { |
7bc09003 JK |
910 | nid_t nid = le32_to_cpu(entry->nid); |
911 | struct page *node_page; | |
26d58599 | 912 | struct node_info ni; |
48018b4c | 913 | int err; |
7bc09003 | 914 | |
43727527 | 915 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 916 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
48018b4c | 917 | return submitted; |
7bc09003 | 918 | |
43727527 | 919 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
920 | continue; |
921 | ||
7ea984b0 | 922 | if (phase == 0) { |
4d57b86d | 923 | f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
7ea984b0 CY |
924 | META_NAT, true); |
925 | continue; | |
926 | } | |
927 | ||
928 | if (phase == 1) { | |
4d57b86d | 929 | f2fs_ra_node_page(sbi, nid); |
7bc09003 JK |
930 | continue; |
931 | } | |
7ea984b0 CY |
932 | |
933 | /* phase == 2 */ | |
4d57b86d | 934 | node_page = f2fs_get_node_page(sbi, nid); |
7bc09003 JK |
935 | if (IS_ERR(node_page)) |
936 | continue; | |
937 | ||
4d57b86d | 938 | /* block may become invalid during f2fs_get_node_page */ |
9a01b56b HY |
939 | if (check_valid_map(sbi, segno, off) == 0) { |
940 | f2fs_put_page(node_page, 1); | |
941 | continue; | |
26d58599 JK |
942 | } |
943 | ||
7735730d CY |
944 | if (f2fs_get_node_info(sbi, nid, &ni)) { |
945 | f2fs_put_page(node_page, 1); | |
946 | continue; | |
947 | } | |
948 | ||
26d58599 JK |
949 | if (ni.blk_addr != start_addr + off) { |
950 | f2fs_put_page(node_page, 1); | |
951 | continue; | |
9a01b56b HY |
952 | } |
953 | ||
48018b4c CY |
954 | err = f2fs_move_node_page(node_page, gc_type); |
955 | if (!err && gc_type == FG_GC) | |
956 | submitted++; | |
e1235983 | 957 | stat_inc_node_blk_count(sbi, 1, gc_type); |
7bc09003 | 958 | } |
c718379b | 959 | |
7ea984b0 | 960 | if (++phase < 3) |
7bc09003 | 961 | goto next_step; |
c29fd0c0 CY |
962 | |
963 | if (fggc) | |
964 | atomic_dec(&sbi->wb_sync_req[NODE]); | |
48018b4c | 965 | return submitted; |
7bc09003 JK |
966 | } |
967 | ||
0a8165d7 | 968 | /* |
9af45ef5 JK |
969 | * Calculate start block index indicating the given node offset. |
970 | * Be careful, caller should give this node offset only indicating direct node | |
971 | * blocks. If any node offsets, which point the other types of node blocks such | |
972 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
973 | * bug. | |
7bc09003 | 974 | */ |
4d57b86d | 975 | block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode) |
7bc09003 | 976 | { |
ce19a5d4 JK |
977 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
978 | unsigned int bidx; | |
7bc09003 | 979 | |
ce19a5d4 JK |
980 | if (node_ofs == 0) |
981 | return 0; | |
7bc09003 | 982 | |
ce19a5d4 | 983 | if (node_ofs <= 2) { |
7bc09003 JK |
984 | bidx = node_ofs - 1; |
985 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 986 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
5f029c04 | 987 | |
7bc09003 JK |
988 | bidx = node_ofs - 2 - dec; |
989 | } else { | |
ce19a5d4 | 990 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
5f029c04 | 991 | |
7bc09003 JK |
992 | bidx = node_ofs - 5 - dec; |
993 | } | |
d02a6e61 | 994 | return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode); |
7bc09003 JK |
995 | } |
996 | ||
c1079892 | 997 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
998 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
999 | { | |
1000 | struct page *node_page; | |
1001 | nid_t nid; | |
1002 | unsigned int ofs_in_node; | |
1003 | block_t source_blkaddr; | |
1004 | ||
1005 | nid = le32_to_cpu(sum->nid); | |
1006 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
1007 | ||
4d57b86d | 1008 | node_page = f2fs_get_node_page(sbi, nid); |
7bc09003 | 1009 | if (IS_ERR(node_page)) |
c1079892 | 1010 | return false; |
7bc09003 | 1011 | |
7735730d CY |
1012 | if (f2fs_get_node_info(sbi, nid, dni)) { |
1013 | f2fs_put_page(node_page, 1); | |
1014 | return false; | |
1015 | } | |
7bc09003 JK |
1016 | |
1017 | if (sum->version != dni->version) { | |
dcbb4c10 JP |
1018 | f2fs_warn(sbi, "%s: valid data with mismatched node version.", |
1019 | __func__); | |
c13ff37e | 1020 | set_sbi_flag(sbi, SBI_NEED_FSCK); |
7bc09003 JK |
1021 | } |
1022 | ||
1023 | *nofs = ofs_of_node(node_page); | |
a2ced1ce | 1024 | source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node); |
7bc09003 JK |
1025 | f2fs_put_page(node_page, 1); |
1026 | ||
bbf9f7d9 ST |
1027 | if (source_blkaddr != blkaddr) { |
1028 | #ifdef CONFIG_F2FS_CHECK_FS | |
1029 | unsigned int segno = GET_SEGNO(sbi, blkaddr); | |
1030 | unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); | |
1031 | ||
1032 | if (unlikely(check_valid_map(sbi, segno, offset))) { | |
1033 | if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) { | |
1034 | f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n", | |
1035 | blkaddr, source_blkaddr, segno); | |
1036 | f2fs_bug_on(sbi, 1); | |
1037 | } | |
1038 | } | |
1039 | #endif | |
c1079892 | 1040 | return false; |
bbf9f7d9 | 1041 | } |
c1079892 | 1042 | return true; |
7bc09003 JK |
1043 | } |
1044 | ||
6aa58d8a CY |
1045 | static int ra_data_block(struct inode *inode, pgoff_t index) |
1046 | { | |
1047 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
1048 | struct address_space *mapping = inode->i_mapping; | |
1049 | struct dnode_of_data dn; | |
1050 | struct page *page; | |
1051 | struct extent_info ei = {0, 0, 0}; | |
1052 | struct f2fs_io_info fio = { | |
1053 | .sbi = sbi, | |
1054 | .ino = inode->i_ino, | |
1055 | .type = DATA, | |
1056 | .temp = COLD, | |
1057 | .op = REQ_OP_READ, | |
1058 | .op_flags = 0, | |
1059 | .encrypted_page = NULL, | |
1060 | .in_list = false, | |
1061 | .retry = false, | |
1062 | }; | |
1063 | int err; | |
1064 | ||
1065 | page = f2fs_grab_cache_page(mapping, index, true); | |
1066 | if (!page) | |
1067 | return -ENOMEM; | |
1068 | ||
1069 | if (f2fs_lookup_extent_cache(inode, index, &ei)) { | |
1070 | dn.data_blkaddr = ei.blk + index - ei.fofs; | |
93770ab7 CY |
1071 | if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, |
1072 | DATA_GENERIC_ENHANCE_READ))) { | |
10f966bb | 1073 | err = -EFSCORRUPTED; |
93770ab7 CY |
1074 | goto put_page; |
1075 | } | |
6aa58d8a CY |
1076 | goto got_it; |
1077 | } | |
1078 | ||
1079 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
1080 | err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); | |
1081 | if (err) | |
1082 | goto put_page; | |
1083 | f2fs_put_dnode(&dn); | |
1084 | ||
93770ab7 CY |
1085 | if (!__is_valid_data_blkaddr(dn.data_blkaddr)) { |
1086 | err = -ENOENT; | |
1087 | goto put_page; | |
1088 | } | |
6aa58d8a | 1089 | if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, |
93770ab7 | 1090 | DATA_GENERIC_ENHANCE))) { |
10f966bb | 1091 | err = -EFSCORRUPTED; |
6aa58d8a CY |
1092 | goto put_page; |
1093 | } | |
1094 | got_it: | |
1095 | /* read page */ | |
1096 | fio.page = page; | |
1097 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; | |
1098 | ||
9bf1a3f7 YS |
1099 | /* |
1100 | * don't cache encrypted data into meta inode until previous dirty | |
1101 | * data were writebacked to avoid racing between GC and flush. | |
1102 | */ | |
bae0ee7a | 1103 | f2fs_wait_on_page_writeback(page, DATA, true, true); |
9bf1a3f7 YS |
1104 | |
1105 | f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); | |
1106 | ||
6aa58d8a CY |
1107 | fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi), |
1108 | dn.data_blkaddr, | |
1109 | FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
1110 | if (!fio.encrypted_page) { | |
1111 | err = -ENOMEM; | |
1112 | goto put_page; | |
1113 | } | |
1114 | ||
1115 | err = f2fs_submit_page_bio(&fio); | |
1116 | if (err) | |
1117 | goto put_encrypted_page; | |
1118 | f2fs_put_page(fio.encrypted_page, 0); | |
1119 | f2fs_put_page(page, 1); | |
8b83ac81 CY |
1120 | |
1121 | f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE); | |
9c122384 | 1122 | f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE); |
8b83ac81 | 1123 | |
6aa58d8a CY |
1124 | return 0; |
1125 | put_encrypted_page: | |
1126 | f2fs_put_page(fio.encrypted_page, 1); | |
1127 | put_page: | |
1128 | f2fs_put_page(page, 1); | |
1129 | return err; | |
1130 | } | |
1131 | ||
d4c759ee JK |
1132 | /* |
1133 | * Move data block via META_MAPPING while keeping locked data page. | |
1134 | * This can be used to move blocks, aka LBAs, directly on disk. | |
1135 | */ | |
48018b4c | 1136 | static int move_data_block(struct inode *inode, block_t bidx, |
2ef79ecb | 1137 | int gc_type, unsigned int segno, int off) |
4375a336 JK |
1138 | { |
1139 | struct f2fs_io_info fio = { | |
1140 | .sbi = F2FS_I_SB(inode), | |
39d787be | 1141 | .ino = inode->i_ino, |
4375a336 | 1142 | .type = DATA, |
a912b54d | 1143 | .temp = COLD, |
04d328de | 1144 | .op = REQ_OP_READ, |
70fd7614 | 1145 | .op_flags = 0, |
4375a336 | 1146 | .encrypted_page = NULL, |
fb830fc5 | 1147 | .in_list = false, |
fe16efe6 | 1148 | .retry = false, |
4375a336 JK |
1149 | }; |
1150 | struct dnode_of_data dn; | |
1151 | struct f2fs_summary sum; | |
1152 | struct node_info ni; | |
6aa58d8a | 1153 | struct page *page, *mpage; |
4356e48e | 1154 | block_t newaddr; |
48018b4c | 1155 | int err = 0; |
b0332a0f | 1156 | bool lfs_mode = f2fs_lfs_mode(fio.sbi); |
ac2d750b WG |
1157 | int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) && |
1158 | (fio.sbi->gc_mode != GC_URGENT_HIGH) ? | |
093749e2 | 1159 | CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA; |
4375a336 JK |
1160 | |
1161 | /* do not read out */ | |
a56c7c6f | 1162 | page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); |
4375a336 | 1163 | if (!page) |
48018b4c | 1164 | return -ENOMEM; |
4375a336 | 1165 | |
48018b4c CY |
1166 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { |
1167 | err = -ENOENT; | |
20614711 | 1168 | goto out; |
48018b4c | 1169 | } |
20614711 | 1170 | |
2ef79ecb CY |
1171 | if (f2fs_is_atomic_file(inode)) { |
1172 | F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; | |
1173 | F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; | |
48018b4c | 1174 | err = -EAGAIN; |
5fe45743 | 1175 | goto out; |
2ef79ecb | 1176 | } |
5fe45743 | 1177 | |
1ad71a27 JK |
1178 | if (f2fs_is_pinned_file(inode)) { |
1179 | f2fs_pin_file_control(inode, true); | |
48018b4c | 1180 | err = -EAGAIN; |
1ad71a27 JK |
1181 | goto out; |
1182 | } | |
1183 | ||
4375a336 | 1184 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
4d57b86d | 1185 | err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE); |
4375a336 JK |
1186 | if (err) |
1187 | goto out; | |
1188 | ||
08b39fbd CY |
1189 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
1190 | ClearPageUptodate(page); | |
48018b4c | 1191 | err = -ENOENT; |
4375a336 | 1192 | goto put_out; |
08b39fbd CY |
1193 | } |
1194 | ||
1195 | /* | |
1196 | * don't cache encrypted data into meta inode until previous dirty | |
1197 | * data were writebacked to avoid racing between GC and flush. | |
1198 | */ | |
bae0ee7a | 1199 | f2fs_wait_on_page_writeback(page, DATA, true, true); |
4375a336 | 1200 | |
9bf1a3f7 YS |
1201 | f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); |
1202 | ||
7735730d CY |
1203 | err = f2fs_get_node_info(fio.sbi, dn.nid, &ni); |
1204 | if (err) | |
1205 | goto put_out; | |
1206 | ||
4375a336 JK |
1207 | /* read page */ |
1208 | fio.page = page; | |
7a9d7548 | 1209 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; |
4375a336 | 1210 | |
107a805d CY |
1211 | if (lfs_mode) |
1212 | down_write(&fio.sbi->io_order_lock); | |
1213 | ||
543b8c46 JK |
1214 | mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi), |
1215 | fio.old_blkaddr, false); | |
d7cd3702 CY |
1216 | if (!mpage) { |
1217 | err = -ENOMEM; | |
543b8c46 | 1218 | goto up_out; |
d7cd3702 | 1219 | } |
543b8c46 JK |
1220 | |
1221 | fio.encrypted_page = mpage; | |
1222 | ||
1223 | /* read source block in mpage */ | |
1224 | if (!PageUptodate(mpage)) { | |
1225 | err = f2fs_submit_page_bio(&fio); | |
1226 | if (err) { | |
1227 | f2fs_put_page(mpage, 1); | |
1228 | goto up_out; | |
1229 | } | |
8b83ac81 CY |
1230 | |
1231 | f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE); | |
9c122384 | 1232 | f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE); |
8b83ac81 | 1233 | |
543b8c46 JK |
1234 | lock_page(mpage); |
1235 | if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) || | |
1236 | !PageUptodate(mpage))) { | |
1237 | err = -EIO; | |
1238 | f2fs_put_page(mpage, 1); | |
1239 | goto up_out; | |
1240 | } | |
1241 | } | |
1242 | ||
cf740403 CY |
1243 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); |
1244 | ||
1245 | /* allocate block address */ | |
4d57b86d | 1246 | f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, |
093749e2 | 1247 | &sum, type, NULL); |
4356e48e | 1248 | |
01eccef7 CY |
1249 | fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), |
1250 | newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
4356e48e CY |
1251 | if (!fio.encrypted_page) { |
1252 | err = -ENOMEM; | |
6aa58d8a | 1253 | f2fs_put_page(mpage, 1); |
543b8c46 | 1254 | goto recover_block; |
4356e48e | 1255 | } |
548aedac | 1256 | |
543b8c46 | 1257 | /* write target block */ |
bae0ee7a | 1258 | f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true); |
543b8c46 JK |
1259 | memcpy(page_address(fio.encrypted_page), |
1260 | page_address(mpage), PAGE_SIZE); | |
1261 | f2fs_put_page(mpage, 1); | |
1262 | invalidate_mapping_pages(META_MAPPING(fio.sbi), | |
1263 | fio.old_blkaddr, fio.old_blkaddr); | |
1264 | ||
8d64d365 | 1265 | set_page_dirty(fio.encrypted_page); |
6282adbf JK |
1266 | if (clear_page_dirty_for_io(fio.encrypted_page)) |
1267 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
1268 | ||
548aedac | 1269 | set_page_writeback(fio.encrypted_page); |
17c50035 | 1270 | ClearPageError(page); |
4375a336 | 1271 | |
04d328de | 1272 | fio.op = REQ_OP_WRITE; |
70fd7614 | 1273 | fio.op_flags = REQ_SYNC; |
4356e48e | 1274 | fio.new_blkaddr = newaddr; |
fe16efe6 CY |
1275 | f2fs_submit_page_write(&fio); |
1276 | if (fio.retry) { | |
48018b4c | 1277 | err = -EAGAIN; |
a9d572c7 SY |
1278 | if (PageWriteback(fio.encrypted_page)) |
1279 | end_page_writeback(fio.encrypted_page); | |
1280 | goto put_page_out; | |
1281 | } | |
4375a336 | 1282 | |
b0af6d49 CY |
1283 | f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE); |
1284 | ||
f28b3434 | 1285 | f2fs_update_data_blkaddr(&dn, newaddr); |
91942321 | 1286 | set_inode_flag(inode, FI_APPEND_WRITE); |
4375a336 | 1287 | if (page->index == 0) |
91942321 | 1288 | set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); |
548aedac | 1289 | put_page_out: |
4375a336 | 1290 | f2fs_put_page(fio.encrypted_page, 1); |
4356e48e CY |
1291 | recover_block: |
1292 | if (err) | |
4d57b86d | 1293 | f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, |
c5d02785 | 1294 | true, true, true); |
543b8c46 JK |
1295 | up_out: |
1296 | if (lfs_mode) | |
1297 | up_write(&fio.sbi->io_order_lock); | |
4375a336 JK |
1298 | put_out: |
1299 | f2fs_put_dnode(&dn); | |
1300 | out: | |
1301 | f2fs_put_page(page, 1); | |
48018b4c | 1302 | return err; |
4375a336 JK |
1303 | } |
1304 | ||
48018b4c | 1305 | static int move_data_page(struct inode *inode, block_t bidx, int gc_type, |
20614711 | 1306 | unsigned int segno, int off) |
7bc09003 | 1307 | { |
c879f90d | 1308 | struct page *page; |
48018b4c | 1309 | int err = 0; |
c879f90d | 1310 | |
4d57b86d | 1311 | page = f2fs_get_lock_data_page(inode, bidx, true); |
c879f90d | 1312 | if (IS_ERR(page)) |
48018b4c | 1313 | return PTR_ERR(page); |
63a0b7cb | 1314 | |
48018b4c CY |
1315 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { |
1316 | err = -ENOENT; | |
20614711 | 1317 | goto out; |
48018b4c | 1318 | } |
20614711 | 1319 | |
2ef79ecb CY |
1320 | if (f2fs_is_atomic_file(inode)) { |
1321 | F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; | |
1322 | F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; | |
48018b4c | 1323 | err = -EAGAIN; |
5fe45743 | 1324 | goto out; |
2ef79ecb | 1325 | } |
1ad71a27 JK |
1326 | if (f2fs_is_pinned_file(inode)) { |
1327 | if (gc_type == FG_GC) | |
1328 | f2fs_pin_file_control(inode, true); | |
48018b4c | 1329 | err = -EAGAIN; |
1ad71a27 JK |
1330 | goto out; |
1331 | } | |
5fe45743 | 1332 | |
7bc09003 | 1333 | if (gc_type == BG_GC) { |
48018b4c CY |
1334 | if (PageWriteback(page)) { |
1335 | err = -EAGAIN; | |
4ebefc44 | 1336 | goto out; |
48018b4c | 1337 | } |
7bc09003 JK |
1338 | set_page_dirty(page); |
1339 | set_cold_data(page); | |
1340 | } else { | |
c879f90d JK |
1341 | struct f2fs_io_info fio = { |
1342 | .sbi = F2FS_I_SB(inode), | |
39d787be | 1343 | .ino = inode->i_ino, |
c879f90d | 1344 | .type = DATA, |
a912b54d | 1345 | .temp = COLD, |
04d328de | 1346 | .op = REQ_OP_WRITE, |
70fd7614 | 1347 | .op_flags = REQ_SYNC, |
e959c8f5 | 1348 | .old_blkaddr = NULL_ADDR, |
c879f90d | 1349 | .page = page, |
4375a336 | 1350 | .encrypted_page = NULL, |
cc15620b | 1351 | .need_lock = LOCK_REQ, |
b0af6d49 | 1352 | .io_type = FS_GC_DATA_IO, |
c879f90d | 1353 | }; |
72e1c797 | 1354 | bool is_dirty = PageDirty(page); |
72e1c797 CY |
1355 | |
1356 | retry: | |
bae0ee7a | 1357 | f2fs_wait_on_page_writeback(page, DATA, true, true); |
8d64d365 CY |
1358 | |
1359 | set_page_dirty(page); | |
933439c8 | 1360 | if (clear_page_dirty_for_io(page)) { |
a7ffdbe2 | 1361 | inode_dec_dirty_pages(inode); |
4d57b86d | 1362 | f2fs_remove_dirty_inode(inode); |
933439c8 | 1363 | } |
72e1c797 | 1364 | |
7bc09003 | 1365 | set_cold_data(page); |
72e1c797 | 1366 | |
4d57b86d | 1367 | err = f2fs_do_write_data_page(&fio); |
14a28559 CY |
1368 | if (err) { |
1369 | clear_cold_data(page); | |
1370 | if (err == -ENOMEM) { | |
5df7731f CY |
1371 | congestion_wait(BLK_RW_ASYNC, |
1372 | DEFAULT_IO_TIMEOUT); | |
14a28559 CY |
1373 | goto retry; |
1374 | } | |
1375 | if (is_dirty) | |
1376 | set_page_dirty(page); | |
72e1c797 | 1377 | } |
7bc09003 JK |
1378 | } |
1379 | out: | |
1380 | f2fs_put_page(page, 1); | |
48018b4c | 1381 | return err; |
7bc09003 JK |
1382 | } |
1383 | ||
0a8165d7 | 1384 | /* |
7bc09003 JK |
1385 | * This function tries to get parent node of victim data block, and identifies |
1386 | * data block validity. If the block is valid, copy that with cold status and | |
1387 | * modify parent node. | |
1388 | * If the parent node is not valid or the data block address is different, | |
1389 | * the victim data block is ignored. | |
1390 | */ | |
48018b4c | 1391 | static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dede886 CY |
1392 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type, |
1393 | bool force_migrate) | |
7bc09003 JK |
1394 | { |
1395 | struct super_block *sb = sbi->sb; | |
1396 | struct f2fs_summary *entry; | |
1397 | block_t start_addr; | |
43727527 | 1398 | int off; |
7bc09003 | 1399 | int phase = 0; |
48018b4c | 1400 | int submitted = 0; |
de881df9 | 1401 | unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); |
7bc09003 JK |
1402 | |
1403 | start_addr = START_BLOCK(sbi, segno); | |
1404 | ||
1405 | next_step: | |
1406 | entry = sum; | |
c718379b | 1407 | |
de881df9 | 1408 | for (off = 0; off < usable_blks_in_seg; off++, entry++) { |
7bc09003 JK |
1409 | struct page *data_page; |
1410 | struct inode *inode; | |
1411 | struct node_info dni; /* dnode info for the data */ | |
1412 | unsigned int ofs_in_node, nofs; | |
1413 | block_t start_bidx; | |
7ea984b0 | 1414 | nid_t nid = le32_to_cpu(entry->nid); |
7bc09003 | 1415 | |
803e74be JK |
1416 | /* |
1417 | * stop BG_GC if there is not enough free sections. | |
1418 | * Or, stop GC if the segment becomes fully valid caused by | |
1419 | * race condition along with SSR block allocation. | |
1420 | */ | |
1421 | if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) || | |
7dede886 CY |
1422 | (!force_migrate && get_valid_blocks(sbi, segno, true) == |
1423 | BLKS_PER_SEC(sbi))) | |
48018b4c | 1424 | return submitted; |
7bc09003 | 1425 | |
43727527 | 1426 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
1427 | continue; |
1428 | ||
1429 | if (phase == 0) { | |
4d57b86d | 1430 | f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
7ea984b0 CY |
1431 | META_NAT, true); |
1432 | continue; | |
1433 | } | |
1434 | ||
1435 | if (phase == 1) { | |
4d57b86d | 1436 | f2fs_ra_node_page(sbi, nid); |
7bc09003 JK |
1437 | continue; |
1438 | } | |
1439 | ||
1440 | /* Get an inode by ino with checking validity */ | |
c1079892 | 1441 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
1442 | continue; |
1443 | ||
7ea984b0 | 1444 | if (phase == 2) { |
4d57b86d | 1445 | f2fs_ra_node_page(sbi, dni.ino); |
7bc09003 JK |
1446 | continue; |
1447 | } | |
1448 | ||
7bc09003 JK |
1449 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
1450 | ||
7ea984b0 | 1451 | if (phase == 3) { |
d4686d56 | 1452 | inode = f2fs_iget(sb, dni.ino); |
4eea93e3 JK |
1453 | if (IS_ERR(inode) || is_bad_inode(inode)) { |
1454 | set_sbi_flag(sbi, SBI_NEED_FSCK); | |
7bc09003 | 1455 | continue; |
4eea93e3 | 1456 | } |
7bc09003 | 1457 | |
bb06664a | 1458 | if (!down_write_trylock( |
b2532c69 | 1459 | &F2FS_I(inode)->i_gc_rwsem[WRITE])) { |
bb06664a | 1460 | iput(inode); |
6f8d4455 | 1461 | sbi->skipped_gc_rwsem++; |
bb06664a CY |
1462 | continue; |
1463 | } | |
1464 | ||
6aa58d8a CY |
1465 | start_bidx = f2fs_start_bidx_of_node(nofs, inode) + |
1466 | ofs_in_node; | |
1467 | ||
1468 | if (f2fs_post_read_required(inode)) { | |
1469 | int err = ra_data_block(inode, start_bidx); | |
1470 | ||
1471 | up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); | |
1472 | if (err) { | |
1473 | iput(inode); | |
1474 | continue; | |
1475 | } | |
1476 | add_gc_inode(gc_list, inode); | |
1477 | continue; | |
1478 | } | |
1479 | ||
4d57b86d | 1480 | data_page = f2fs_get_read_data_page(inode, |
6aa58d8a | 1481 | start_bidx, REQ_RAHEAD, true); |
b2532c69 | 1482 | up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); |
31a32688 CL |
1483 | if (IS_ERR(data_page)) { |
1484 | iput(inode); | |
1485 | continue; | |
1486 | } | |
7bc09003 JK |
1487 | |
1488 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 1489 | add_gc_inode(gc_list, inode); |
31a32688 CL |
1490 | continue; |
1491 | } | |
1492 | ||
7ea984b0 | 1493 | /* phase 4 */ |
7dda2af8 | 1494 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 1495 | if (inode) { |
82e0a5aa CY |
1496 | struct f2fs_inode_info *fi = F2FS_I(inode); |
1497 | bool locked = false; | |
48018b4c | 1498 | int err; |
82e0a5aa CY |
1499 | |
1500 | if (S_ISREG(inode->i_mode)) { | |
b2532c69 | 1501 | if (!down_write_trylock(&fi->i_gc_rwsem[READ])) |
82e0a5aa CY |
1502 | continue; |
1503 | if (!down_write_trylock( | |
b2532c69 | 1504 | &fi->i_gc_rwsem[WRITE])) { |
6f8d4455 | 1505 | sbi->skipped_gc_rwsem++; |
b2532c69 | 1506 | up_write(&fi->i_gc_rwsem[READ]); |
82e0a5aa CY |
1507 | continue; |
1508 | } | |
1509 | locked = true; | |
73ac2f4e CY |
1510 | |
1511 | /* wait for all inflight aio data */ | |
1512 | inode_dio_wait(inode); | |
82e0a5aa CY |
1513 | } |
1514 | ||
4d57b86d | 1515 | start_bidx = f2fs_start_bidx_of_node(nofs, inode) |
c879f90d | 1516 | + ofs_in_node; |
6dbb1796 | 1517 | if (f2fs_post_read_required(inode)) |
48018b4c CY |
1518 | err = move_data_block(inode, start_bidx, |
1519 | gc_type, segno, off); | |
4375a336 | 1520 | else |
48018b4c | 1521 | err = move_data_page(inode, start_bidx, gc_type, |
d4c759ee | 1522 | segno, off); |
82e0a5aa | 1523 | |
48018b4c CY |
1524 | if (!err && (gc_type == FG_GC || |
1525 | f2fs_post_read_required(inode))) | |
1526 | submitted++; | |
1527 | ||
82e0a5aa | 1528 | if (locked) { |
b2532c69 CY |
1529 | up_write(&fi->i_gc_rwsem[WRITE]); |
1530 | up_write(&fi->i_gc_rwsem[READ]); | |
82e0a5aa CY |
1531 | } |
1532 | ||
e1235983 | 1533 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 1534 | } |
7bc09003 | 1535 | } |
c718379b | 1536 | |
7ea984b0 | 1537 | if (++phase < 5) |
7bc09003 | 1538 | goto next_step; |
48018b4c CY |
1539 | |
1540 | return submitted; | |
7bc09003 JK |
1541 | } |
1542 | ||
1543 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 1544 | int gc_type) |
7bc09003 JK |
1545 | { |
1546 | struct sit_info *sit_i = SIT_I(sbi); | |
1547 | int ret; | |
8a2d0ace | 1548 | |
3d26fa6b | 1549 | down_write(&sit_i->sentry_lock); |
8a2d0ace | 1550 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
093749e2 | 1551 | NO_CHECK_TYPE, LFS, 0); |
3d26fa6b | 1552 | up_write(&sit_i->sentry_lock); |
7bc09003 JK |
1553 | return ret; |
1554 | } | |
1555 | ||
718e53fa CY |
1556 | static int do_garbage_collect(struct f2fs_sb_info *sbi, |
1557 | unsigned int start_segno, | |
7dede886 CY |
1558 | struct gc_inode_list *gc_list, int gc_type, |
1559 | bool force_migrate) | |
7bc09003 JK |
1560 | { |
1561 | struct page *sum_page; | |
1562 | struct f2fs_summary_block *sum; | |
c718379b | 1563 | struct blk_plug plug; |
718e53fa CY |
1564 | unsigned int segno = start_segno; |
1565 | unsigned int end_segno = start_segno + sbi->segs_per_sec; | |
e3080b01 | 1566 | int seg_freed = 0, migrated = 0; |
718e53fa CY |
1567 | unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? |
1568 | SUM_TYPE_DATA : SUM_TYPE_NODE; | |
48018b4c | 1569 | int submitted = 0; |
7bc09003 | 1570 | |
e3080b01 CY |
1571 | if (__is_large_section(sbi)) |
1572 | end_segno = rounddown(end_segno, sbi->segs_per_sec); | |
1573 | ||
de881df9 AR |
1574 | /* |
1575 | * zone-capacity can be less than zone-size in zoned devices, | |
1576 | * resulting in less than expected usable segments in the zone, | |
1577 | * calculate the end segno in the zone which can be garbage collected | |
1578 | */ | |
1579 | if (f2fs_sb_has_blkzoned(sbi)) | |
1580 | end_segno -= sbi->segs_per_sec - | |
1581 | f2fs_usable_segs_in_sec(sbi, segno); | |
1582 | ||
093749e2 CY |
1583 | sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type); |
1584 | ||
718e53fa | 1585 | /* readahead multi ssa blocks those have contiguous address */ |
2c70c5e3 | 1586 | if (__is_large_section(sbi)) |
4d57b86d | 1587 | f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), |
e3080b01 | 1588 | end_segno - segno, META_SSA, true); |
718e53fa CY |
1589 | |
1590 | /* reference all summary page */ | |
1591 | while (segno < end_segno) { | |
4d57b86d | 1592 | sum_page = f2fs_get_sum_page(sbi, segno++); |
edc55aaf JK |
1593 | if (IS_ERR(sum_page)) { |
1594 | int err = PTR_ERR(sum_page); | |
1595 | ||
1596 | end_segno = segno - 1; | |
1597 | for (segno = start_segno; segno < end_segno; segno++) { | |
1598 | sum_page = find_get_page(META_MAPPING(sbi), | |
1599 | GET_SUM_BLOCK(sbi, segno)); | |
1600 | f2fs_put_page(sum_page, 0); | |
1601 | f2fs_put_page(sum_page, 0); | |
1602 | } | |
1603 | return err; | |
1604 | } | |
718e53fa CY |
1605 | unlock_page(sum_page); |
1606 | } | |
7bc09003 | 1607 | |
c718379b JK |
1608 | blk_start_plug(&plug); |
1609 | ||
718e53fa | 1610 | for (segno = start_segno; segno < end_segno; segno++) { |
aa987273 | 1611 | |
718e53fa CY |
1612 | /* find segment summary of victim */ |
1613 | sum_page = find_get_page(META_MAPPING(sbi), | |
1614 | GET_SUM_BLOCK(sbi, segno)); | |
718e53fa | 1615 | f2fs_put_page(sum_page, 0); |
7bc09003 | 1616 | |
d6c66cd1 YS |
1617 | if (get_valid_blocks(sbi, segno, false) == 0) |
1618 | goto freed; | |
dabfbbc8 | 1619 | if (gc_type == BG_GC && __is_large_section(sbi) && |
e3080b01 CY |
1620 | migrated >= sbi->migration_granularity) |
1621 | goto skip; | |
d6c66cd1 | 1622 | if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi))) |
e3080b01 | 1623 | goto skip; |
de0dcc40 | 1624 | |
718e53fa | 1625 | sum = page_address(sum_page); |
10d255c3 | 1626 | if (type != GET_SUM_TYPE((&sum->footer))) { |
dcbb4c10 JP |
1627 | f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT", |
1628 | segno, type, GET_SUM_TYPE((&sum->footer))); | |
10d255c3 | 1629 | set_sbi_flag(sbi, SBI_NEED_FSCK); |
793ab1c8 | 1630 | f2fs_stop_checkpoint(sbi, false); |
e3080b01 | 1631 | goto skip; |
10d255c3 | 1632 | } |
718e53fa CY |
1633 | |
1634 | /* | |
1635 | * this is to avoid deadlock: | |
1636 | * - lock_page(sum_page) - f2fs_replace_block | |
3d26fa6b CY |
1637 | * - check_valid_map() - down_write(sentry_lock) |
1638 | * - down_read(sentry_lock) - change_curseg() | |
718e53fa CY |
1639 | * - lock_page(sum_page) |
1640 | */ | |
718e53fa | 1641 | if (type == SUM_TYPE_NODE) |
48018b4c | 1642 | submitted += gc_node_segment(sbi, sum->entries, segno, |
718e53fa | 1643 | gc_type); |
48018b4c CY |
1644 | else |
1645 | submitted += gc_data_segment(sbi, sum->entries, gc_list, | |
7dede886 CY |
1646 | segno, gc_type, |
1647 | force_migrate); | |
718e53fa CY |
1648 | |
1649 | stat_inc_seg_count(sbi, type, gc_type); | |
8c7b9ac1 | 1650 | migrated++; |
c56f16da | 1651 | |
d6c66cd1 | 1652 | freed: |
c56f16da CY |
1653 | if (gc_type == FG_GC && |
1654 | get_valid_blocks(sbi, segno, false) == 0) | |
1655 | seg_freed++; | |
e3080b01 CY |
1656 | |
1657 | if (__is_large_section(sbi) && segno + 1 < end_segno) | |
1658 | sbi->next_victim_seg[gc_type] = segno + 1; | |
1659 | skip: | |
718e53fa CY |
1660 | f2fs_put_page(sum_page, 0); |
1661 | } | |
1662 | ||
48018b4c | 1663 | if (submitted) |
b9109b0e JK |
1664 | f2fs_submit_merged_write(sbi, |
1665 | (type == SUM_TYPE_NODE) ? NODE : DATA); | |
c718379b | 1666 | |
718e53fa | 1667 | blk_finish_plug(&plug); |
7bc09003 | 1668 | |
17d899df CY |
1669 | stat_inc_call_count(sbi->stat_info); |
1670 | ||
c56f16da | 1671 | return seg_freed; |
7bc09003 JK |
1672 | } |
1673 | ||
e066b83c | 1674 | int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, |
7dede886 | 1675 | bool background, bool force, unsigned int segno) |
7bc09003 | 1676 | { |
d530d4d8 | 1677 | int gc_type = sync ? FG_GC : BG_GC; |
c56f16da CY |
1678 | int sec_freed = 0, seg_freed = 0, total_freed = 0; |
1679 | int ret = 0; | |
d5053a34 | 1680 | struct cp_control cpc; |
e066b83c | 1681 | unsigned int init_segno = segno; |
7dda2af8 CL |
1682 | struct gc_inode_list gc_list = { |
1683 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
f6bb2a2c | 1684 | .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), |
7dda2af8 | 1685 | }; |
2ef79ecb | 1686 | unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC]; |
6f8d4455 | 1687 | unsigned long long first_skipped; |
2ef79ecb | 1688 | unsigned int skipped_round = 0, round = 0; |
d5053a34 | 1689 | |
c56f16da CY |
1690 | trace_f2fs_gc_begin(sbi->sb, sync, background, |
1691 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1692 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1693 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1694 | free_sections(sbi), | |
1695 | free_segments(sbi), | |
1696 | reserved_segments(sbi), | |
1697 | prefree_segments(sbi)); | |
1698 | ||
119ee914 | 1699 | cpc.reason = __get_cp_reason(sbi); |
6f8d4455 JK |
1700 | sbi->skipped_gc_rwsem = 0; |
1701 | first_skipped = last_skipped; | |
7bc09003 | 1702 | gc_more: |
1751e8a6 | 1703 | if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { |
e5dbd956 | 1704 | ret = -EINVAL; |
408e9375 | 1705 | goto stop; |
e5dbd956 | 1706 | } |
6d5a1495 CY |
1707 | if (unlikely(f2fs_cp_error(sbi))) { |
1708 | ret = -EIO; | |
203681f6 | 1709 | goto stop; |
6d5a1495 | 1710 | } |
7bc09003 | 1711 | |
19f4e688 | 1712 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) { |
6e17bfbc | 1713 | /* |
19f4e688 HP |
1714 | * For example, if there are many prefree_segments below given |
1715 | * threshold, we can make them free by checkpoint. Then, we | |
1716 | * secure free segments which doesn't need fggc any more. | |
6e17bfbc | 1717 | */ |
4354994f DR |
1718 | if (prefree_segments(sbi) && |
1719 | !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { | |
4d57b86d | 1720 | ret = f2fs_write_checkpoint(sbi, &cpc); |
8fd5a37e JK |
1721 | if (ret) |
1722 | goto stop; | |
1723 | } | |
19f4e688 HP |
1724 | if (has_not_enough_free_secs(sbi, 0, 0)) |
1725 | gc_type = FG_GC; | |
d64f8047 | 1726 | } |
7bc09003 | 1727 | |
19f4e688 | 1728 | /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ |
c56f16da CY |
1729 | if (gc_type == BG_GC && !background) { |
1730 | ret = -EINVAL; | |
19f4e688 | 1731 | goto stop; |
c56f16da | 1732 | } |
97767500 QZ |
1733 | ret = __get_victim(sbi, &segno, gc_type); |
1734 | if (ret) | |
408e9375 | 1735 | goto stop; |
7bc09003 | 1736 | |
7dede886 | 1737 | seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, force); |
de881df9 AR |
1738 | if (gc_type == FG_GC && |
1739 | seg_freed == f2fs_usable_segs_in_sec(sbi, segno)) | |
45fe8492 | 1740 | sec_freed++; |
c56f16da | 1741 | total_freed += seg_freed; |
43727527 | 1742 | |
2ef79ecb | 1743 | if (gc_type == FG_GC) { |
6f8d4455 JK |
1744 | if (sbi->skipped_atomic_files[FG_GC] > last_skipped || |
1745 | sbi->skipped_gc_rwsem) | |
2ef79ecb CY |
1746 | skipped_round++; |
1747 | last_skipped = sbi->skipped_atomic_files[FG_GC]; | |
1748 | round++; | |
1749 | } | |
1750 | ||
957fa478 | 1751 | if (gc_type == FG_GC && seg_freed) |
5ec4e49f | 1752 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 1753 | |
6f8d4455 JK |
1754 | if (sync) |
1755 | goto stop; | |
1756 | ||
1757 | if (has_not_enough_free_secs(sbi, sec_freed, 0)) { | |
1758 | if (skipped_round <= MAX_SKIP_GC_COUNT || | |
1759 | skipped_round * 2 < round) { | |
e066b83c | 1760 | segno = NULL_SEGNO; |
d530d4d8 | 1761 | goto gc_more; |
e066b83c | 1762 | } |
43727527 | 1763 | |
6f8d4455 JK |
1764 | if (first_skipped < last_skipped && |
1765 | (last_skipped - first_skipped) > | |
1766 | sbi->skipped_gc_rwsem) { | |
1767 | f2fs_drop_inmem_pages_all(sbi, true); | |
1768 | segno = NULL_SEGNO; | |
1769 | goto gc_more; | |
1770 | } | |
4354994f | 1771 | if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) |
4d57b86d | 1772 | ret = f2fs_write_checkpoint(sbi, &cpc); |
d530d4d8 | 1773 | } |
408e9375 | 1774 | stop: |
e066b83c JK |
1775 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; |
1776 | SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno; | |
c56f16da CY |
1777 | |
1778 | trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed, | |
1779 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1780 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1781 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1782 | free_sections(sbi), | |
1783 | free_segments(sbi), | |
1784 | reserved_segments(sbi), | |
1785 | prefree_segments(sbi)); | |
1786 | ||
fb24fea7 | 1787 | up_write(&sbi->gc_lock); |
7bc09003 | 1788 | |
7dda2af8 | 1789 | put_gc_inode(&gc_list); |
d530d4d8 | 1790 | |
61f7725a | 1791 | if (sync && !ret) |
d530d4d8 | 1792 | ret = sec_freed ? 0 : -EAGAIN; |
43727527 | 1793 | return ret; |
7bc09003 JK |
1794 | } |
1795 | ||
093749e2 CY |
1796 | int __init f2fs_create_garbage_collection_cache(void) |
1797 | { | |
1798 | victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry", | |
1799 | sizeof(struct victim_entry)); | |
1800 | if (!victim_entry_slab) | |
1801 | return -ENOMEM; | |
1802 | return 0; | |
1803 | } | |
1804 | ||
1805 | void f2fs_destroy_garbage_collection_cache(void) | |
1806 | { | |
1807 | kmem_cache_destroy(victim_entry_slab); | |
1808 | } | |
1809 | ||
1810 | static void init_atgc_management(struct f2fs_sb_info *sbi) | |
1811 | { | |
1812 | struct atgc_management *am = &sbi->am; | |
1813 | ||
1814 | if (test_opt(sbi, ATGC) && | |
1815 | SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD) | |
1816 | am->atgc_enabled = true; | |
1817 | ||
1818 | am->root = RB_ROOT_CACHED; | |
1819 | INIT_LIST_HEAD(&am->victim_list); | |
1820 | am->victim_count = 0; | |
1821 | ||
1822 | am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO; | |
1823 | am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT; | |
1824 | am->age_weight = DEF_GC_THREAD_AGE_WEIGHT; | |
1825 | } | |
1826 | ||
4d57b86d | 1827 | void f2fs_build_gc_manager(struct f2fs_sb_info *sbi) |
7bc09003 JK |
1828 | { |
1829 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
e93b9865 | 1830 | |
1ad71a27 | 1831 | sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES; |
d5793249 JK |
1832 | |
1833 | /* give warm/cold data area from slower device */ | |
0916878d | 1834 | if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi)) |
d5793249 JK |
1835 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = |
1836 | GET_SEGNO(sbi, FDEV(0).end_blk) + 1; | |
093749e2 CY |
1837 | |
1838 | init_atgc_management(sbi); | |
7bc09003 | 1839 | } |
04f0b2ea | 1840 | |
b4b10061 JK |
1841 | static int free_segment_range(struct f2fs_sb_info *sbi, |
1842 | unsigned int secs, bool gc_only) | |
04f0b2ea | 1843 | { |
b4b10061 JK |
1844 | unsigned int segno, next_inuse, start, end; |
1845 | struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; | |
1846 | int gc_mode, gc_type; | |
04f0b2ea | 1847 | int err = 0; |
b4b10061 JK |
1848 | int type; |
1849 | ||
1850 | /* Force block allocation for GC */ | |
1851 | MAIN_SECS(sbi) -= secs; | |
1852 | start = MAIN_SECS(sbi) * sbi->segs_per_sec; | |
1853 | end = MAIN_SEGS(sbi) - 1; | |
1854 | ||
1855 | mutex_lock(&DIRTY_I(sbi)->seglist_lock); | |
1856 | for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++) | |
1857 | if (SIT_I(sbi)->last_victim[gc_mode] >= start) | |
1858 | SIT_I(sbi)->last_victim[gc_mode] = 0; | |
1859 | ||
1860 | for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++) | |
1861 | if (sbi->next_victim_seg[gc_type] >= start) | |
1862 | sbi->next_victim_seg[gc_type] = NULL_SEGNO; | |
1863 | mutex_unlock(&DIRTY_I(sbi)->seglist_lock); | |
04f0b2ea QS |
1864 | |
1865 | /* Move out cursegs from the target range */ | |
d0b9e42a | 1866 | for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++) |
0ef81833 | 1867 | f2fs_allocate_segment_for_resize(sbi, type, start, end); |
04f0b2ea QS |
1868 | |
1869 | /* do GC to move out valid blocks in the range */ | |
1870 | for (segno = start; segno <= end; segno += sbi->segs_per_sec) { | |
1871 | struct gc_inode_list gc_list = { | |
1872 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
1873 | .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), | |
1874 | }; | |
1875 | ||
7dede886 | 1876 | do_garbage_collect(sbi, segno, &gc_list, FG_GC, true); |
04f0b2ea QS |
1877 | put_gc_inode(&gc_list); |
1878 | ||
b4b10061 JK |
1879 | if (!gc_only && get_valid_blocks(sbi, segno, true)) { |
1880 | err = -EAGAIN; | |
1881 | goto out; | |
1882 | } | |
1883 | if (fatal_signal_pending(current)) { | |
1884 | err = -ERESTARTSYS; | |
1885 | goto out; | |
1886 | } | |
04f0b2ea | 1887 | } |
b4b10061 JK |
1888 | if (gc_only) |
1889 | goto out; | |
04f0b2ea | 1890 | |
b4b10061 | 1891 | err = f2fs_write_checkpoint(sbi, &cpc); |
04f0b2ea | 1892 | if (err) |
b4b10061 | 1893 | goto out; |
04f0b2ea QS |
1894 | |
1895 | next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start); | |
1896 | if (next_inuse <= end) { | |
dcbb4c10 JP |
1897 | f2fs_err(sbi, "segno %u should be free but still inuse!", |
1898 | next_inuse); | |
04f0b2ea QS |
1899 | f2fs_bug_on(sbi, 1); |
1900 | } | |
b4b10061 JK |
1901 | out: |
1902 | MAIN_SECS(sbi) += secs; | |
04f0b2ea QS |
1903 | return err; |
1904 | } | |
1905 | ||
1906 | static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs) | |
1907 | { | |
1908 | struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi); | |
a4ba5dfc CY |
1909 | int section_count; |
1910 | int segment_count; | |
1911 | int segment_count_main; | |
1912 | long long block_count; | |
04f0b2ea QS |
1913 | int segs = secs * sbi->segs_per_sec; |
1914 | ||
a4ba5dfc CY |
1915 | down_write(&sbi->sb_lock); |
1916 | ||
1917 | section_count = le32_to_cpu(raw_sb->section_count); | |
1918 | segment_count = le32_to_cpu(raw_sb->segment_count); | |
1919 | segment_count_main = le32_to_cpu(raw_sb->segment_count_main); | |
1920 | block_count = le64_to_cpu(raw_sb->block_count); | |
1921 | ||
04f0b2ea QS |
1922 | raw_sb->section_count = cpu_to_le32(section_count + secs); |
1923 | raw_sb->segment_count = cpu_to_le32(segment_count + segs); | |
1924 | raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs); | |
1925 | raw_sb->block_count = cpu_to_le64(block_count + | |
1926 | (long long)segs * sbi->blocks_per_seg); | |
46d9ce19 QS |
1927 | if (f2fs_is_multi_device(sbi)) { |
1928 | int last_dev = sbi->s_ndevs - 1; | |
1929 | int dev_segs = | |
1930 | le32_to_cpu(raw_sb->devs[last_dev].total_segments); | |
1931 | ||
1932 | raw_sb->devs[last_dev].total_segments = | |
1933 | cpu_to_le32(dev_segs + segs); | |
1934 | } | |
a4ba5dfc CY |
1935 | |
1936 | up_write(&sbi->sb_lock); | |
04f0b2ea QS |
1937 | } |
1938 | ||
1939 | static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs) | |
1940 | { | |
1941 | int segs = secs * sbi->segs_per_sec; | |
46d9ce19 | 1942 | long long blks = (long long)segs * sbi->blocks_per_seg; |
04f0b2ea QS |
1943 | long long user_block_count = |
1944 | le64_to_cpu(F2FS_CKPT(sbi)->user_block_count); | |
1945 | ||
1946 | SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs; | |
1947 | MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs; | |
b4b10061 | 1948 | MAIN_SECS(sbi) += secs; |
04f0b2ea QS |
1949 | FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs; |
1950 | FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs; | |
46d9ce19 QS |
1951 | F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks); |
1952 | ||
1953 | if (f2fs_is_multi_device(sbi)) { | |
1954 | int last_dev = sbi->s_ndevs - 1; | |
1955 | ||
1956 | FDEV(last_dev).total_segments = | |
1957 | (int)FDEV(last_dev).total_segments + segs; | |
1958 | FDEV(last_dev).end_blk = | |
1959 | (long long)FDEV(last_dev).end_blk + blks; | |
1960 | #ifdef CONFIG_BLK_DEV_ZONED | |
1961 | FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz + | |
1962 | (int)(blks >> sbi->log_blocks_per_blkz); | |
1963 | #endif | |
1964 | } | |
04f0b2ea QS |
1965 | } |
1966 | ||
1967 | int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count) | |
1968 | { | |
1969 | __u64 old_block_count, shrunk_blocks; | |
b4b10061 | 1970 | struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; |
04f0b2ea | 1971 | unsigned int secs; |
04f0b2ea QS |
1972 | int err = 0; |
1973 | __u32 rem; | |
1974 | ||
1975 | old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count); | |
1976 | if (block_count > old_block_count) | |
1977 | return -EINVAL; | |
1978 | ||
46d9ce19 QS |
1979 | if (f2fs_is_multi_device(sbi)) { |
1980 | int last_dev = sbi->s_ndevs - 1; | |
1981 | __u64 last_segs = FDEV(last_dev).total_segments; | |
1982 | ||
1983 | if (block_count + last_segs * sbi->blocks_per_seg <= | |
1984 | old_block_count) | |
1985 | return -EINVAL; | |
1986 | } | |
1987 | ||
04f0b2ea QS |
1988 | /* new fs size should align to section size */ |
1989 | div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem); | |
1990 | if (rem) | |
1991 | return -EINVAL; | |
1992 | ||
1993 | if (block_count == old_block_count) | |
1994 | return 0; | |
1995 | ||
1996 | if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { | |
dcbb4c10 | 1997 | f2fs_err(sbi, "Should run fsck to repair first."); |
10f966bb | 1998 | return -EFSCORRUPTED; |
04f0b2ea QS |
1999 | } |
2000 | ||
2001 | if (test_opt(sbi, DISABLE_CHECKPOINT)) { | |
dcbb4c10 | 2002 | f2fs_err(sbi, "Checkpoint should be enabled."); |
04f0b2ea QS |
2003 | return -EINVAL; |
2004 | } | |
2005 | ||
04f0b2ea QS |
2006 | shrunk_blocks = old_block_count - block_count; |
2007 | secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi)); | |
b4b10061 JK |
2008 | |
2009 | /* stop other GC */ | |
2010 | if (!down_write_trylock(&sbi->gc_lock)) | |
2011 | return -EAGAIN; | |
2012 | ||
2013 | /* stop CP to protect MAIN_SEC in free_segment_range */ | |
2014 | f2fs_lock_op(sbi); | |
3ab0598e CY |
2015 | |
2016 | spin_lock(&sbi->stat_lock); | |
2017 | if (shrunk_blocks + valid_user_blocks(sbi) + | |
2018 | sbi->current_reserved_blocks + sbi->unusable_block_count + | |
2019 | F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) | |
2020 | err = -ENOSPC; | |
2021 | spin_unlock(&sbi->stat_lock); | |
2022 | ||
2023 | if (err) | |
2024 | goto out_unlock; | |
2025 | ||
b4b10061 | 2026 | err = free_segment_range(sbi, secs, true); |
3ab0598e CY |
2027 | |
2028 | out_unlock: | |
b4b10061 JK |
2029 | f2fs_unlock_op(sbi); |
2030 | up_write(&sbi->gc_lock); | |
2031 | if (err) | |
2032 | return err; | |
2033 | ||
2034 | set_sbi_flag(sbi, SBI_IS_RESIZEFS); | |
2035 | ||
2036 | freeze_super(sbi->sb); | |
2037 | down_write(&sbi->gc_lock); | |
8769918b | 2038 | down_write(&sbi->cp_global_sem); |
b4b10061 | 2039 | |
04f0b2ea QS |
2040 | spin_lock(&sbi->stat_lock); |
2041 | if (shrunk_blocks + valid_user_blocks(sbi) + | |
2042 | sbi->current_reserved_blocks + sbi->unusable_block_count + | |
2043 | F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) | |
2044 | err = -ENOSPC; | |
2045 | else | |
2046 | sbi->user_block_count -= shrunk_blocks; | |
2047 | spin_unlock(&sbi->stat_lock); | |
b4b10061 JK |
2048 | if (err) |
2049 | goto out_err; | |
04f0b2ea | 2050 | |
b4b10061 | 2051 | err = free_segment_range(sbi, secs, false); |
04f0b2ea | 2052 | if (err) |
b4b10061 | 2053 | goto recover_out; |
04f0b2ea QS |
2054 | |
2055 | update_sb_metadata(sbi, -secs); | |
2056 | ||
2057 | err = f2fs_commit_super(sbi, false); | |
2058 | if (err) { | |
2059 | update_sb_metadata(sbi, secs); | |
b4b10061 | 2060 | goto recover_out; |
04f0b2ea QS |
2061 | } |
2062 | ||
2063 | update_fs_metadata(sbi, -secs); | |
2064 | clear_sbi_flag(sbi, SBI_IS_RESIZEFS); | |
68275682 | 2065 | set_sbi_flag(sbi, SBI_IS_DIRTY); |
68275682 | 2066 | |
b4b10061 | 2067 | err = f2fs_write_checkpoint(sbi, &cpc); |
04f0b2ea QS |
2068 | if (err) { |
2069 | update_fs_metadata(sbi, secs); | |
2070 | update_sb_metadata(sbi, secs); | |
2071 | f2fs_commit_super(sbi, false); | |
2072 | } | |
b4b10061 | 2073 | recover_out: |
04f0b2ea QS |
2074 | if (err) { |
2075 | set_sbi_flag(sbi, SBI_NEED_FSCK); | |
dcbb4c10 | 2076 | f2fs_err(sbi, "resize_fs failed, should run fsck to repair!"); |
04f0b2ea | 2077 | |
04f0b2ea QS |
2078 | spin_lock(&sbi->stat_lock); |
2079 | sbi->user_block_count += shrunk_blocks; | |
2080 | spin_unlock(&sbi->stat_lock); | |
2081 | } | |
b4b10061 | 2082 | out_err: |
8769918b | 2083 | up_write(&sbi->cp_global_sem); |
b4b10061 JK |
2084 | up_write(&sbi->gc_lock); |
2085 | thaw_super(sbi->sb); | |
04f0b2ea | 2086 | clear_sbi_flag(sbi, SBI_IS_RESIZEFS); |
04f0b2ea QS |
2087 | return err; |
2088 | } |