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> | |
7bc09003 JK |
16 | |
17 | #include "f2fs.h" | |
18 | #include "node.h" | |
19 | #include "segment.h" | |
20 | #include "gc.h" | |
8e46b3ed | 21 | #include <trace/events/f2fs.h> |
7bc09003 | 22 | |
7bc09003 JK |
23 | static int gc_thread_func(void *data) |
24 | { | |
25 | struct f2fs_sb_info *sbi = data; | |
b59d0bae | 26 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; |
7bc09003 | 27 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; |
b8c502b8 | 28 | unsigned int wait_ms; |
7bc09003 | 29 | |
b59d0bae | 30 | wait_ms = gc_th->min_sleep_time; |
7bc09003 | 31 | |
1d7be270 | 32 | set_freezable(); |
7bc09003 | 33 | do { |
1d7be270 | 34 | wait_event_interruptible_timeout(*wq, |
d9872a69 JK |
35 | kthread_should_stop() || freezing(current) || |
36 | gc_th->gc_wake, | |
1d7be270 JK |
37 | msecs_to_jiffies(wait_ms)); |
38 | ||
d9872a69 JK |
39 | /* give it a try one time */ |
40 | if (gc_th->gc_wake) | |
41 | gc_th->gc_wake = 0; | |
42 | ||
274bd9ba CY |
43 | if (try_to_freeze()) { |
44 | stat_other_skip_bggc_count(sbi); | |
7bc09003 | 45 | continue; |
274bd9ba | 46 | } |
7bc09003 JK |
47 | if (kthread_should_stop()) |
48 | break; | |
49 | ||
d6212a5f | 50 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
88dd8934 | 51 | increase_sleep_time(gc_th, &wait_ms); |
274bd9ba | 52 | stat_other_skip_bggc_count(sbi); |
d6212a5f CL |
53 | continue; |
54 | } | |
55 | ||
55523519 CY |
56 | if (time_to_inject(sbi, FAULT_CHECKPOINT)) { |
57 | f2fs_show_injection_info(FAULT_CHECKPOINT); | |
0f348028 | 58 | f2fs_stop_checkpoint(sbi, false); |
55523519 | 59 | } |
0f348028 | 60 | |
274bd9ba CY |
61 | if (!sb_start_write_trylock(sbi->sb)) { |
62 | stat_other_skip_bggc_count(sbi); | |
dc6febb6 | 63 | continue; |
274bd9ba | 64 | } |
dc6febb6 | 65 | |
7bc09003 JK |
66 | /* |
67 | * [GC triggering condition] | |
68 | * 0. GC is not conducted currently. | |
69 | * 1. There are enough dirty segments. | |
70 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
71 | * 3. IO subsystem is idle by checking the # of requests in | |
72 | * bdev's request list. | |
73 | * | |
e1c42045 | 74 | * Note) We have to avoid triggering GCs frequently. |
7bc09003 JK |
75 | * Because it is possible that some segments can be |
76 | * invalidated soon after by user update or deletion. | |
77 | * So, I'd like to wait some time to collect dirty segments. | |
78 | */ | |
5b0e9539 | 79 | if (sbi->gc_mode == GC_URGENT) { |
d9872a69 | 80 | wait_ms = gc_th->urgent_sleep_time; |
69babac0 | 81 | mutex_lock(&sbi->gc_mutex); |
d9872a69 JK |
82 | goto do_gc; |
83 | } | |
84 | ||
274bd9ba CY |
85 | if (!mutex_trylock(&sbi->gc_mutex)) { |
86 | stat_other_skip_bggc_count(sbi); | |
69babac0 | 87 | goto next; |
274bd9ba | 88 | } |
69babac0 | 89 | |
a7d10cf3 | 90 | if (!is_idle(sbi, GC_TIME)) { |
88dd8934 | 91 | increase_sleep_time(gc_th, &wait_ms); |
7bc09003 | 92 | mutex_unlock(&sbi->gc_mutex); |
274bd9ba | 93 | stat_io_skip_bggc_count(sbi); |
dc6febb6 | 94 | goto next; |
7bc09003 JK |
95 | } |
96 | ||
97 | if (has_enough_invalid_blocks(sbi)) | |
88dd8934 | 98 | decrease_sleep_time(gc_th, &wait_ms); |
7bc09003 | 99 | else |
88dd8934 | 100 | increase_sleep_time(gc_th, &wait_ms); |
d9872a69 | 101 | do_gc: |
dcdfff65 | 102 | stat_inc_bggc_count(sbi); |
7bc09003 | 103 | |
43727527 | 104 | /* if return value is not zero, no victim was selected */ |
e066b83c | 105 | if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO)) |
b59d0bae | 106 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 107 | |
84e4214f JK |
108 | trace_f2fs_background_gc(sbi->sb, wait_ms, |
109 | prefree_segments(sbi), free_segments(sbi)); | |
110 | ||
4660f9c0 JK |
111 | /* balancing f2fs's metadata periodically */ |
112 | f2fs_balance_fs_bg(sbi); | |
dc6febb6 CY |
113 | next: |
114 | sb_end_write(sbi->sb); | |
81eb8d6e | 115 | |
7bc09003 JK |
116 | } while (!kthread_should_stop()); |
117 | return 0; | |
118 | } | |
119 | ||
4d57b86d | 120 | int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) |
7bc09003 | 121 | { |
1042d60f | 122 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 123 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 124 | int err = 0; |
7bc09003 | 125 | |
1ecc0c5c | 126 | gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
7a267f8d NJ |
127 | if (!gc_th) { |
128 | err = -ENOMEM; | |
129 | goto out; | |
130 | } | |
7bc09003 | 131 | |
d9872a69 | 132 | gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME; |
b59d0bae NJ |
133 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
134 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
135 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
136 | ||
d9872a69 | 137 | gc_th->gc_wake= 0; |
d2dc095f | 138 | |
7bc09003 JK |
139 | sbi->gc_thread = gc_th; |
140 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
141 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 142 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 143 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 144 | err = PTR_ERR(gc_th->f2fs_gc_task); |
5222595d | 145 | kvfree(gc_th); |
25718423 | 146 | sbi->gc_thread = NULL; |
7bc09003 | 147 | } |
7a267f8d NJ |
148 | out: |
149 | return err; | |
7bc09003 JK |
150 | } |
151 | ||
4d57b86d | 152 | void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) |
7bc09003 JK |
153 | { |
154 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
155 | if (!gc_th) | |
156 | return; | |
157 | kthread_stop(gc_th->f2fs_gc_task); | |
5222595d | 158 | kvfree(gc_th); |
7bc09003 JK |
159 | sbi->gc_thread = NULL; |
160 | } | |
161 | ||
5b0e9539 | 162 | static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) |
7bc09003 | 163 | { |
d2dc095f NJ |
164 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
165 | ||
5b0e9539 JK |
166 | switch (sbi->gc_mode) { |
167 | case GC_IDLE_CB: | |
168 | gc_mode = GC_CB; | |
169 | break; | |
170 | case GC_IDLE_GREEDY: | |
171 | case GC_URGENT: | |
b27bc809 | 172 | gc_mode = GC_GREEDY; |
5b0e9539 JK |
173 | break; |
174 | } | |
d2dc095f | 175 | return gc_mode; |
7bc09003 JK |
176 | } |
177 | ||
178 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
179 | int type, struct victim_sel_policy *p) | |
180 | { | |
181 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
182 | ||
4ebefc44 | 183 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
184 | p->gc_mode = GC_GREEDY; |
185 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
a26b7c8a | 186 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 JK |
187 | p->ofs_unit = 1; |
188 | } else { | |
5b0e9539 | 189 | p->gc_mode = select_gc_type(sbi, gc_type); |
7bc09003 | 190 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
a26b7c8a | 191 | p->max_search = dirty_i->nr_dirty[DIRTY]; |
7bc09003 JK |
192 | p->ofs_unit = sbi->segs_per_sec; |
193 | } | |
a26b7c8a | 194 | |
e93b9865 | 195 | /* we need to check every dirty segments in the FG_GC case */ |
b27bc809 | 196 | if (gc_type != FG_GC && |
5b0e9539 | 197 | (sbi->gc_mode != GC_URGENT) && |
b27bc809 | 198 | p->max_search > sbi->max_victim_search) |
b1c57c1c | 199 | p->max_search = sbi->max_victim_search; |
a26b7c8a | 200 | |
b94929d9 YS |
201 | /* let's select beginning hot/small space first in no_heap mode*/ |
202 | if (test_opt(sbi, NOHEAP) && | |
203 | (type == CURSEG_HOT_DATA || IS_NODESEG(type))) | |
7a20b8a6 JK |
204 | p->offset = 0; |
205 | else | |
e066b83c | 206 | p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; |
7bc09003 JK |
207 | } |
208 | ||
209 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
210 | struct victim_sel_policy *p) | |
211 | { | |
b7250d2d JK |
212 | /* SSR allocates in a segment unit */ |
213 | if (p->alloc_mode == SSR) | |
3519e3f9 | 214 | return sbi->blocks_per_seg; |
7bc09003 | 215 | if (p->gc_mode == GC_GREEDY) |
c541a51b | 216 | return 2 * sbi->blocks_per_seg * p->ofs_unit; |
7bc09003 JK |
217 | else if (p->gc_mode == GC_CB) |
218 | return UINT_MAX; | |
219 | else /* No other gc_mode */ | |
220 | return 0; | |
221 | } | |
222 | ||
223 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
224 | { | |
225 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 226 | unsigned int secno; |
7bc09003 JK |
227 | |
228 | /* | |
229 | * If the gc_type is FG_GC, we can select victim segments | |
230 | * selected by background GC before. | |
231 | * Those segments guarantee they have small valid blocks. | |
232 | */ | |
7cd8558b | 233 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 234 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 235 | continue; |
5ec4e49f | 236 | clear_bit(secno, dirty_i->victim_secmap); |
4ddb1a4d | 237 | return GET_SEG_FROM_SEC(sbi, secno); |
7bc09003 JK |
238 | } |
239 | return NULL_SEGNO; | |
240 | } | |
241 | ||
242 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
243 | { | |
244 | struct sit_info *sit_i = SIT_I(sbi); | |
4ddb1a4d JK |
245 | unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); |
246 | unsigned int start = GET_SEG_FROM_SEC(sbi, secno); | |
7bc09003 JK |
247 | unsigned long long mtime = 0; |
248 | unsigned int vblocks; | |
249 | unsigned char age = 0; | |
250 | unsigned char u; | |
251 | unsigned int i; | |
252 | ||
253 | for (i = 0; i < sbi->segs_per_sec; i++) | |
254 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
302bd348 | 255 | vblocks = get_valid_blocks(sbi, segno, true); |
7bc09003 JK |
256 | |
257 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
258 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
259 | ||
260 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
261 | ||
e1c42045 | 262 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
263 | if (mtime < sit_i->min_mtime) |
264 | sit_i->min_mtime = mtime; | |
265 | if (mtime > sit_i->max_mtime) | |
266 | sit_i->max_mtime = mtime; | |
267 | if (sit_i->max_mtime != sit_i->min_mtime) | |
268 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
269 | sit_i->max_mtime - sit_i->min_mtime); | |
270 | ||
271 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
272 | } | |
273 | ||
a57e564d JX |
274 | static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, |
275 | unsigned int segno, struct victim_sel_policy *p) | |
7bc09003 JK |
276 | { |
277 | if (p->alloc_mode == SSR) | |
2afce76a | 278 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; |
7bc09003 JK |
279 | |
280 | /* alloc_mode == LFS */ | |
281 | if (p->gc_mode == GC_GREEDY) | |
91f4382b | 282 | return get_valid_blocks(sbi, segno, true); |
7bc09003 JK |
283 | else |
284 | return get_cb_cost(sbi, segno); | |
285 | } | |
286 | ||
688159b6 FL |
287 | static unsigned int count_bits(const unsigned long *addr, |
288 | unsigned int offset, unsigned int len) | |
289 | { | |
290 | unsigned int end = offset + len, sum = 0; | |
291 | ||
292 | while (offset < end) { | |
293 | if (test_bit(offset++, addr)) | |
294 | ++sum; | |
295 | } | |
296 | return sum; | |
297 | } | |
298 | ||
0a8165d7 | 299 | /* |
111d2495 | 300 | * This function is called from two paths. |
7bc09003 JK |
301 | * One is garbage collection and the other is SSR segment selection. |
302 | * When it is called during GC, it just gets a victim segment | |
303 | * and it does not remove it from dirty seglist. | |
304 | * When it is called from SSR segment selection, it finds a segment | |
305 | * which has minimum valid blocks and removes it from dirty seglist. | |
306 | */ | |
307 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
308 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
309 | { | |
310 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
e066b83c | 311 | struct sit_info *sm = SIT_I(sbi); |
7bc09003 | 312 | struct victim_sel_policy p; |
3fa56503 | 313 | unsigned int secno, last_victim; |
04f0b2ea | 314 | unsigned int last_segment; |
688159b6 | 315 | unsigned int nsearched = 0; |
7bc09003 | 316 | |
210f41bc | 317 | mutex_lock(&dirty_i->seglist_lock); |
04f0b2ea | 318 | last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec; |
210f41bc | 319 | |
7bc09003 JK |
320 | p.alloc_mode = alloc_mode; |
321 | select_policy(sbi, gc_type, type, &p); | |
322 | ||
323 | p.min_segno = NULL_SEGNO; | |
3fa56503 | 324 | p.min_cost = get_max_cost(sbi, &p); |
7bc09003 | 325 | |
e066b83c | 326 | if (*result != NULL_SEGNO) { |
08ac9a38 | 327 | if (get_valid_blocks(sbi, *result, false) && |
e066b83c JK |
328 | !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) |
329 | p.min_segno = *result; | |
330 | goto out; | |
331 | } | |
332 | ||
3342bb30 CY |
333 | if (p.max_search == 0) |
334 | goto out; | |
335 | ||
e3080b01 CY |
336 | if (__is_large_section(sbi) && p.alloc_mode == LFS) { |
337 | if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) { | |
338 | p.min_segno = sbi->next_victim_seg[BG_GC]; | |
339 | *result = p.min_segno; | |
340 | sbi->next_victim_seg[BG_GC] = NULL_SEGNO; | |
341 | goto got_result; | |
342 | } | |
343 | if (gc_type == FG_GC && | |
344 | sbi->next_victim_seg[FG_GC] != NULL_SEGNO) { | |
345 | p.min_segno = sbi->next_victim_seg[FG_GC]; | |
346 | *result = p.min_segno; | |
347 | sbi->next_victim_seg[FG_GC] = NULL_SEGNO; | |
348 | goto got_result; | |
349 | } | |
350 | } | |
351 | ||
e066b83c | 352 | last_victim = sm->last_victim[p.gc_mode]; |
7bc09003 JK |
353 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
354 | p.min_segno = check_bg_victims(sbi); | |
355 | if (p.min_segno != NULL_SEGNO) | |
356 | goto got_it; | |
357 | } | |
358 | ||
359 | while (1) { | |
360 | unsigned long cost; | |
5ec4e49f | 361 | unsigned int segno; |
7bc09003 | 362 | |
a43f7ec3 CY |
363 | segno = find_next_bit(p.dirty_segmap, last_segment, p.offset); |
364 | if (segno >= last_segment) { | |
e066b83c JK |
365 | if (sm->last_victim[p.gc_mode]) { |
366 | last_segment = | |
367 | sm->last_victim[p.gc_mode]; | |
368 | sm->last_victim[p.gc_mode] = 0; | |
7bc09003 JK |
369 | p.offset = 0; |
370 | continue; | |
371 | } | |
372 | break; | |
373 | } | |
a57e564d JX |
374 | |
375 | p.offset = segno + p.ofs_unit; | |
688159b6 | 376 | if (p.ofs_unit > 1) { |
a57e564d | 377 | p.offset -= segno % p.ofs_unit; |
688159b6 FL |
378 | nsearched += count_bits(p.dirty_segmap, |
379 | p.offset - p.ofs_unit, | |
380 | p.ofs_unit); | |
381 | } else { | |
382 | nsearched++; | |
383 | } | |
384 | ||
bbf9f7d9 ST |
385 | #ifdef CONFIG_F2FS_CHECK_FS |
386 | /* | |
387 | * skip selecting the invalid segno (that is failed due to block | |
388 | * validity check failure during GC) to avoid endless GC loop in | |
389 | * such cases. | |
390 | */ | |
391 | if (test_bit(segno, sm->invalid_segmap)) | |
392 | goto next; | |
393 | #endif | |
394 | ||
4ddb1a4d | 395 | secno = GET_SEC_FROM_SEG(sbi, segno); |
7bc09003 | 396 | |
5ec4e49f | 397 | if (sec_usage_check(sbi, secno)) |
688159b6 | 398 | goto next; |
4354994f DR |
399 | /* Don't touch checkpointed data */ |
400 | if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) && | |
49dd883c JK |
401 | get_ckpt_valid_blocks(sbi, segno) && |
402 | p.alloc_mode != SSR)) | |
4354994f | 403 | goto next; |
5ec4e49f | 404 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
688159b6 | 405 | goto next; |
7bc09003 JK |
406 | |
407 | cost = get_gc_cost(sbi, segno, &p); | |
408 | ||
409 | if (p.min_cost > cost) { | |
410 | p.min_segno = segno; | |
411 | p.min_cost = cost; | |
a57e564d | 412 | } |
688159b6 FL |
413 | next: |
414 | if (nsearched >= p.max_search) { | |
e066b83c JK |
415 | if (!sm->last_victim[p.gc_mode] && segno <= last_victim) |
416 | sm->last_victim[p.gc_mode] = last_victim + 1; | |
4ce53776 | 417 | else |
e066b83c | 418 | sm->last_victim[p.gc_mode] = segno + 1; |
04f0b2ea QS |
419 | sm->last_victim[p.gc_mode] %= |
420 | (MAIN_SECS(sbi) * sbi->segs_per_sec); | |
7bc09003 JK |
421 | break; |
422 | } | |
423 | } | |
7bc09003 | 424 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 425 | got_it: |
e3080b01 CY |
426 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
427 | got_result: | |
7bc09003 | 428 | if (p.alloc_mode == LFS) { |
4ddb1a4d | 429 | secno = GET_SEC_FROM_SEG(sbi, p.min_segno); |
5ec4e49f JK |
430 | if (gc_type == FG_GC) |
431 | sbi->cur_victim_sec = secno; | |
432 | else | |
433 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 434 | } |
8e46b3ed | 435 | |
e3c59108 ST |
436 | } |
437 | out: | |
438 | if (p.min_segno != NULL_SEGNO) | |
8e46b3ed NJ |
439 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, |
440 | sbi->cur_victim_sec, | |
441 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 JK |
442 | mutex_unlock(&dirty_i->seglist_lock); |
443 | ||
444 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
445 | } | |
446 | ||
447 | static const struct victim_selection default_v_ops = { | |
448 | .get_victim = get_victim_by_default, | |
449 | }; | |
450 | ||
7dda2af8 | 451 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 452 | { |
7bc09003 JK |
453 | struct inode_entry *ie; |
454 | ||
7dda2af8 CL |
455 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
456 | if (ie) | |
457 | return ie->inode; | |
7bc09003 JK |
458 | return NULL; |
459 | } | |
460 | ||
7dda2af8 | 461 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 462 | { |
6cc4af56 GZ |
463 | struct inode_entry *new_ie; |
464 | ||
7dda2af8 | 465 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
466 | iput(inode); |
467 | return; | |
7bc09003 | 468 | } |
4d57b86d | 469 | new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS); |
7bc09003 | 470 | new_ie->inode = inode; |
f28e5034 CY |
471 | |
472 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 473 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
474 | } |
475 | ||
7dda2af8 | 476 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
477 | { |
478 | struct inode_entry *ie, *next_ie; | |
7dda2af8 CL |
479 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
480 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
481 | iput(ie->inode); |
482 | list_del(&ie->list); | |
4d57b86d | 483 | kmem_cache_free(f2fs_inode_entry_slab, ie); |
7bc09003 JK |
484 | } |
485 | } | |
486 | ||
487 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
488 | unsigned int segno, int offset) | |
489 | { | |
490 | struct sit_info *sit_i = SIT_I(sbi); | |
491 | struct seg_entry *sentry; | |
492 | int ret; | |
493 | ||
3d26fa6b | 494 | down_read(&sit_i->sentry_lock); |
7bc09003 JK |
495 | sentry = get_seg_entry(sbi, segno); |
496 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
3d26fa6b | 497 | up_read(&sit_i->sentry_lock); |
43727527 | 498 | return ret; |
7bc09003 JK |
499 | } |
500 | ||
0a8165d7 | 501 | /* |
7bc09003 JK |
502 | * This function compares node address got in summary with that in NAT. |
503 | * On validity, copy that node with cold status, otherwise (invalid node) | |
504 | * ignore that. | |
505 | */ | |
48018b4c | 506 | static int gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
507 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
508 | { | |
7bc09003 | 509 | struct f2fs_summary *entry; |
26d58599 | 510 | block_t start_addr; |
7bc09003 | 511 | int off; |
7ea984b0 | 512 | int phase = 0; |
c29fd0c0 | 513 | bool fggc = (gc_type == FG_GC); |
48018b4c | 514 | int submitted = 0; |
7bc09003 | 515 | |
26d58599 JK |
516 | start_addr = START_BLOCK(sbi, segno); |
517 | ||
7bc09003 JK |
518 | next_step: |
519 | entry = sum; | |
c718379b | 520 | |
c29fd0c0 CY |
521 | if (fggc && phase == 2) |
522 | atomic_inc(&sbi->wb_sync_req[NODE]); | |
523 | ||
7bc09003 JK |
524 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
525 | nid_t nid = le32_to_cpu(entry->nid); | |
526 | struct page *node_page; | |
26d58599 | 527 | struct node_info ni; |
48018b4c | 528 | int err; |
7bc09003 | 529 | |
43727527 | 530 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 531 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
48018b4c | 532 | return submitted; |
7bc09003 | 533 | |
43727527 | 534 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
535 | continue; |
536 | ||
7ea984b0 | 537 | if (phase == 0) { |
4d57b86d | 538 | f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
7ea984b0 CY |
539 | META_NAT, true); |
540 | continue; | |
541 | } | |
542 | ||
543 | if (phase == 1) { | |
4d57b86d | 544 | f2fs_ra_node_page(sbi, nid); |
7bc09003 JK |
545 | continue; |
546 | } | |
7ea984b0 CY |
547 | |
548 | /* phase == 2 */ | |
4d57b86d | 549 | node_page = f2fs_get_node_page(sbi, nid); |
7bc09003 JK |
550 | if (IS_ERR(node_page)) |
551 | continue; | |
552 | ||
4d57b86d | 553 | /* block may become invalid during f2fs_get_node_page */ |
9a01b56b HY |
554 | if (check_valid_map(sbi, segno, off) == 0) { |
555 | f2fs_put_page(node_page, 1); | |
556 | continue; | |
26d58599 JK |
557 | } |
558 | ||
7735730d CY |
559 | if (f2fs_get_node_info(sbi, nid, &ni)) { |
560 | f2fs_put_page(node_page, 1); | |
561 | continue; | |
562 | } | |
563 | ||
26d58599 JK |
564 | if (ni.blk_addr != start_addr + off) { |
565 | f2fs_put_page(node_page, 1); | |
566 | continue; | |
9a01b56b HY |
567 | } |
568 | ||
48018b4c CY |
569 | err = f2fs_move_node_page(node_page, gc_type); |
570 | if (!err && gc_type == FG_GC) | |
571 | submitted++; | |
e1235983 | 572 | stat_inc_node_blk_count(sbi, 1, gc_type); |
7bc09003 | 573 | } |
c718379b | 574 | |
7ea984b0 | 575 | if (++phase < 3) |
7bc09003 | 576 | goto next_step; |
c29fd0c0 CY |
577 | |
578 | if (fggc) | |
579 | atomic_dec(&sbi->wb_sync_req[NODE]); | |
48018b4c | 580 | return submitted; |
7bc09003 JK |
581 | } |
582 | ||
0a8165d7 | 583 | /* |
9af45ef5 JK |
584 | * Calculate start block index indicating the given node offset. |
585 | * Be careful, caller should give this node offset only indicating direct node | |
586 | * blocks. If any node offsets, which point the other types of node blocks such | |
587 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
588 | * bug. | |
7bc09003 | 589 | */ |
4d57b86d | 590 | block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode) |
7bc09003 | 591 | { |
ce19a5d4 JK |
592 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
593 | unsigned int bidx; | |
7bc09003 | 594 | |
ce19a5d4 JK |
595 | if (node_ofs == 0) |
596 | return 0; | |
7bc09003 | 597 | |
ce19a5d4 | 598 | if (node_ofs <= 2) { |
7bc09003 JK |
599 | bidx = node_ofs - 1; |
600 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 601 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
602 | bidx = node_ofs - 2 - dec; |
603 | } else { | |
ce19a5d4 | 604 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
605 | bidx = node_ofs - 5 - dec; |
606 | } | |
d02a6e61 | 607 | return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode); |
7bc09003 JK |
608 | } |
609 | ||
c1079892 | 610 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
611 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
612 | { | |
613 | struct page *node_page; | |
614 | nid_t nid; | |
615 | unsigned int ofs_in_node; | |
616 | block_t source_blkaddr; | |
617 | ||
618 | nid = le32_to_cpu(sum->nid); | |
619 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
620 | ||
4d57b86d | 621 | node_page = f2fs_get_node_page(sbi, nid); |
7bc09003 | 622 | if (IS_ERR(node_page)) |
c1079892 | 623 | return false; |
7bc09003 | 624 | |
7735730d CY |
625 | if (f2fs_get_node_info(sbi, nid, dni)) { |
626 | f2fs_put_page(node_page, 1); | |
627 | return false; | |
628 | } | |
7bc09003 JK |
629 | |
630 | if (sum->version != dni->version) { | |
dcbb4c10 JP |
631 | f2fs_warn(sbi, "%s: valid data with mismatched node version.", |
632 | __func__); | |
c13ff37e | 633 | set_sbi_flag(sbi, SBI_NEED_FSCK); |
7bc09003 JK |
634 | } |
635 | ||
636 | *nofs = ofs_of_node(node_page); | |
7a2af766 | 637 | source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node); |
7bc09003 JK |
638 | f2fs_put_page(node_page, 1); |
639 | ||
bbf9f7d9 ST |
640 | if (source_blkaddr != blkaddr) { |
641 | #ifdef CONFIG_F2FS_CHECK_FS | |
642 | unsigned int segno = GET_SEGNO(sbi, blkaddr); | |
643 | unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); | |
644 | ||
645 | if (unlikely(check_valid_map(sbi, segno, offset))) { | |
646 | if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) { | |
647 | f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n", | |
648 | blkaddr, source_blkaddr, segno); | |
649 | f2fs_bug_on(sbi, 1); | |
650 | } | |
651 | } | |
652 | #endif | |
c1079892 | 653 | return false; |
bbf9f7d9 | 654 | } |
c1079892 | 655 | return true; |
7bc09003 JK |
656 | } |
657 | ||
6aa58d8a CY |
658 | static int ra_data_block(struct inode *inode, pgoff_t index) |
659 | { | |
660 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
661 | struct address_space *mapping = inode->i_mapping; | |
662 | struct dnode_of_data dn; | |
663 | struct page *page; | |
664 | struct extent_info ei = {0, 0, 0}; | |
665 | struct f2fs_io_info fio = { | |
666 | .sbi = sbi, | |
667 | .ino = inode->i_ino, | |
668 | .type = DATA, | |
669 | .temp = COLD, | |
670 | .op = REQ_OP_READ, | |
671 | .op_flags = 0, | |
672 | .encrypted_page = NULL, | |
673 | .in_list = false, | |
674 | .retry = false, | |
675 | }; | |
676 | int err; | |
677 | ||
678 | page = f2fs_grab_cache_page(mapping, index, true); | |
679 | if (!page) | |
680 | return -ENOMEM; | |
681 | ||
682 | if (f2fs_lookup_extent_cache(inode, index, &ei)) { | |
683 | dn.data_blkaddr = ei.blk + index - ei.fofs; | |
93770ab7 CY |
684 | if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, |
685 | DATA_GENERIC_ENHANCE_READ))) { | |
10f966bb | 686 | err = -EFSCORRUPTED; |
93770ab7 CY |
687 | goto put_page; |
688 | } | |
6aa58d8a CY |
689 | goto got_it; |
690 | } | |
691 | ||
692 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
693 | err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); | |
694 | if (err) | |
695 | goto put_page; | |
696 | f2fs_put_dnode(&dn); | |
697 | ||
93770ab7 CY |
698 | if (!__is_valid_data_blkaddr(dn.data_blkaddr)) { |
699 | err = -ENOENT; | |
700 | goto put_page; | |
701 | } | |
6aa58d8a | 702 | if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, |
93770ab7 | 703 | DATA_GENERIC_ENHANCE))) { |
10f966bb | 704 | err = -EFSCORRUPTED; |
6aa58d8a CY |
705 | goto put_page; |
706 | } | |
707 | got_it: | |
708 | /* read page */ | |
709 | fio.page = page; | |
710 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; | |
711 | ||
9bf1a3f7 YS |
712 | /* |
713 | * don't cache encrypted data into meta inode until previous dirty | |
714 | * data were writebacked to avoid racing between GC and flush. | |
715 | */ | |
bae0ee7a | 716 | f2fs_wait_on_page_writeback(page, DATA, true, true); |
9bf1a3f7 YS |
717 | |
718 | f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); | |
719 | ||
6aa58d8a CY |
720 | fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi), |
721 | dn.data_blkaddr, | |
722 | FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
723 | if (!fio.encrypted_page) { | |
724 | err = -ENOMEM; | |
725 | goto put_page; | |
726 | } | |
727 | ||
728 | err = f2fs_submit_page_bio(&fio); | |
729 | if (err) | |
730 | goto put_encrypted_page; | |
731 | f2fs_put_page(fio.encrypted_page, 0); | |
732 | f2fs_put_page(page, 1); | |
733 | return 0; | |
734 | put_encrypted_page: | |
735 | f2fs_put_page(fio.encrypted_page, 1); | |
736 | put_page: | |
737 | f2fs_put_page(page, 1); | |
738 | return err; | |
739 | } | |
740 | ||
d4c759ee JK |
741 | /* |
742 | * Move data block via META_MAPPING while keeping locked data page. | |
743 | * This can be used to move blocks, aka LBAs, directly on disk. | |
744 | */ | |
48018b4c | 745 | static int move_data_block(struct inode *inode, block_t bidx, |
2ef79ecb | 746 | int gc_type, unsigned int segno, int off) |
4375a336 JK |
747 | { |
748 | struct f2fs_io_info fio = { | |
749 | .sbi = F2FS_I_SB(inode), | |
39d787be | 750 | .ino = inode->i_ino, |
4375a336 | 751 | .type = DATA, |
a912b54d | 752 | .temp = COLD, |
04d328de | 753 | .op = REQ_OP_READ, |
70fd7614 | 754 | .op_flags = 0, |
4375a336 | 755 | .encrypted_page = NULL, |
fb830fc5 | 756 | .in_list = false, |
fe16efe6 | 757 | .retry = false, |
4375a336 JK |
758 | }; |
759 | struct dnode_of_data dn; | |
760 | struct f2fs_summary sum; | |
761 | struct node_info ni; | |
6aa58d8a | 762 | struct page *page, *mpage; |
4356e48e | 763 | block_t newaddr; |
48018b4c | 764 | int err = 0; |
107a805d | 765 | bool lfs_mode = test_opt(fio.sbi, LFS); |
4375a336 JK |
766 | |
767 | /* do not read out */ | |
a56c7c6f | 768 | page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); |
4375a336 | 769 | if (!page) |
48018b4c | 770 | return -ENOMEM; |
4375a336 | 771 | |
48018b4c CY |
772 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { |
773 | err = -ENOENT; | |
20614711 | 774 | goto out; |
48018b4c | 775 | } |
20614711 | 776 | |
2ef79ecb CY |
777 | if (f2fs_is_atomic_file(inode)) { |
778 | F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; | |
779 | F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; | |
48018b4c | 780 | err = -EAGAIN; |
5fe45743 | 781 | goto out; |
2ef79ecb | 782 | } |
5fe45743 | 783 | |
1ad71a27 JK |
784 | if (f2fs_is_pinned_file(inode)) { |
785 | f2fs_pin_file_control(inode, true); | |
48018b4c | 786 | err = -EAGAIN; |
1ad71a27 JK |
787 | goto out; |
788 | } | |
789 | ||
4375a336 | 790 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
4d57b86d | 791 | err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE); |
4375a336 JK |
792 | if (err) |
793 | goto out; | |
794 | ||
08b39fbd CY |
795 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
796 | ClearPageUptodate(page); | |
48018b4c | 797 | err = -ENOENT; |
4375a336 | 798 | goto put_out; |
08b39fbd CY |
799 | } |
800 | ||
801 | /* | |
802 | * don't cache encrypted data into meta inode until previous dirty | |
803 | * data were writebacked to avoid racing between GC and flush. | |
804 | */ | |
bae0ee7a | 805 | f2fs_wait_on_page_writeback(page, DATA, true, true); |
4375a336 | 806 | |
9bf1a3f7 YS |
807 | f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); |
808 | ||
7735730d CY |
809 | err = f2fs_get_node_info(fio.sbi, dn.nid, &ni); |
810 | if (err) | |
811 | goto put_out; | |
812 | ||
4375a336 JK |
813 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); |
814 | ||
815 | /* read page */ | |
816 | fio.page = page; | |
7a9d7548 | 817 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; |
4375a336 | 818 | |
107a805d CY |
819 | if (lfs_mode) |
820 | down_write(&fio.sbi->io_order_lock); | |
821 | ||
543b8c46 JK |
822 | mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi), |
823 | fio.old_blkaddr, false); | |
824 | if (!mpage) | |
825 | goto up_out; | |
826 | ||
827 | fio.encrypted_page = mpage; | |
828 | ||
829 | /* read source block in mpage */ | |
830 | if (!PageUptodate(mpage)) { | |
831 | err = f2fs_submit_page_bio(&fio); | |
832 | if (err) { | |
833 | f2fs_put_page(mpage, 1); | |
834 | goto up_out; | |
835 | } | |
836 | lock_page(mpage); | |
837 | if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) || | |
838 | !PageUptodate(mpage))) { | |
839 | err = -EIO; | |
840 | f2fs_put_page(mpage, 1); | |
841 | goto up_out; | |
842 | } | |
843 | } | |
844 | ||
4d57b86d | 845 | f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, |
fb830fc5 | 846 | &sum, CURSEG_COLD_DATA, NULL, false); |
4356e48e | 847 | |
01eccef7 CY |
848 | fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), |
849 | newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
4356e48e CY |
850 | if (!fio.encrypted_page) { |
851 | err = -ENOMEM; | |
6aa58d8a | 852 | f2fs_put_page(mpage, 1); |
543b8c46 | 853 | goto recover_block; |
4356e48e | 854 | } |
548aedac | 855 | |
543b8c46 | 856 | /* write target block */ |
bae0ee7a | 857 | f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true); |
543b8c46 JK |
858 | memcpy(page_address(fio.encrypted_page), |
859 | page_address(mpage), PAGE_SIZE); | |
860 | f2fs_put_page(mpage, 1); | |
861 | invalidate_mapping_pages(META_MAPPING(fio.sbi), | |
862 | fio.old_blkaddr, fio.old_blkaddr); | |
863 | ||
8d64d365 | 864 | set_page_dirty(fio.encrypted_page); |
6282adbf JK |
865 | if (clear_page_dirty_for_io(fio.encrypted_page)) |
866 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
867 | ||
548aedac | 868 | set_page_writeback(fio.encrypted_page); |
17c50035 | 869 | ClearPageError(page); |
4375a336 JK |
870 | |
871 | /* allocate block address */ | |
bae0ee7a | 872 | f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true); |
4356e48e | 873 | |
04d328de | 874 | fio.op = REQ_OP_WRITE; |
70fd7614 | 875 | fio.op_flags = REQ_SYNC; |
4356e48e | 876 | fio.new_blkaddr = newaddr; |
fe16efe6 CY |
877 | f2fs_submit_page_write(&fio); |
878 | if (fio.retry) { | |
48018b4c | 879 | err = -EAGAIN; |
a9d572c7 SY |
880 | if (PageWriteback(fio.encrypted_page)) |
881 | end_page_writeback(fio.encrypted_page); | |
882 | goto put_page_out; | |
883 | } | |
4375a336 | 884 | |
b0af6d49 CY |
885 | f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE); |
886 | ||
f28b3434 | 887 | f2fs_update_data_blkaddr(&dn, newaddr); |
91942321 | 888 | set_inode_flag(inode, FI_APPEND_WRITE); |
4375a336 | 889 | if (page->index == 0) |
91942321 | 890 | set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); |
548aedac | 891 | put_page_out: |
4375a336 | 892 | f2fs_put_page(fio.encrypted_page, 1); |
4356e48e CY |
893 | recover_block: |
894 | if (err) | |
4d57b86d | 895 | f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, |
4356e48e | 896 | true, true); |
543b8c46 JK |
897 | up_out: |
898 | if (lfs_mode) | |
899 | up_write(&fio.sbi->io_order_lock); | |
4375a336 JK |
900 | put_out: |
901 | f2fs_put_dnode(&dn); | |
902 | out: | |
903 | f2fs_put_page(page, 1); | |
48018b4c | 904 | return err; |
4375a336 JK |
905 | } |
906 | ||
48018b4c | 907 | static int move_data_page(struct inode *inode, block_t bidx, int gc_type, |
20614711 | 908 | unsigned int segno, int off) |
7bc09003 | 909 | { |
c879f90d | 910 | struct page *page; |
48018b4c | 911 | int err = 0; |
c879f90d | 912 | |
4d57b86d | 913 | page = f2fs_get_lock_data_page(inode, bidx, true); |
c879f90d | 914 | if (IS_ERR(page)) |
48018b4c | 915 | return PTR_ERR(page); |
63a0b7cb | 916 | |
48018b4c CY |
917 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { |
918 | err = -ENOENT; | |
20614711 | 919 | goto out; |
48018b4c | 920 | } |
20614711 | 921 | |
2ef79ecb CY |
922 | if (f2fs_is_atomic_file(inode)) { |
923 | F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; | |
924 | F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; | |
48018b4c | 925 | err = -EAGAIN; |
5fe45743 | 926 | goto out; |
2ef79ecb | 927 | } |
1ad71a27 JK |
928 | if (f2fs_is_pinned_file(inode)) { |
929 | if (gc_type == FG_GC) | |
930 | f2fs_pin_file_control(inode, true); | |
48018b4c | 931 | err = -EAGAIN; |
1ad71a27 JK |
932 | goto out; |
933 | } | |
5fe45743 | 934 | |
7bc09003 | 935 | if (gc_type == BG_GC) { |
48018b4c CY |
936 | if (PageWriteback(page)) { |
937 | err = -EAGAIN; | |
4ebefc44 | 938 | goto out; |
48018b4c | 939 | } |
7bc09003 JK |
940 | set_page_dirty(page); |
941 | set_cold_data(page); | |
942 | } else { | |
c879f90d JK |
943 | struct f2fs_io_info fio = { |
944 | .sbi = F2FS_I_SB(inode), | |
39d787be | 945 | .ino = inode->i_ino, |
c879f90d | 946 | .type = DATA, |
a912b54d | 947 | .temp = COLD, |
04d328de | 948 | .op = REQ_OP_WRITE, |
70fd7614 | 949 | .op_flags = REQ_SYNC, |
e959c8f5 | 950 | .old_blkaddr = NULL_ADDR, |
c879f90d | 951 | .page = page, |
4375a336 | 952 | .encrypted_page = NULL, |
cc15620b | 953 | .need_lock = LOCK_REQ, |
b0af6d49 | 954 | .io_type = FS_GC_DATA_IO, |
c879f90d | 955 | }; |
72e1c797 | 956 | bool is_dirty = PageDirty(page); |
72e1c797 CY |
957 | |
958 | retry: | |
bae0ee7a | 959 | f2fs_wait_on_page_writeback(page, DATA, true, true); |
8d64d365 CY |
960 | |
961 | set_page_dirty(page); | |
933439c8 | 962 | if (clear_page_dirty_for_io(page)) { |
a7ffdbe2 | 963 | inode_dec_dirty_pages(inode); |
4d57b86d | 964 | f2fs_remove_dirty_inode(inode); |
933439c8 | 965 | } |
72e1c797 | 966 | |
7bc09003 | 967 | set_cold_data(page); |
72e1c797 | 968 | |
4d57b86d | 969 | err = f2fs_do_write_data_page(&fio); |
14a28559 CY |
970 | if (err) { |
971 | clear_cold_data(page); | |
972 | if (err == -ENOMEM) { | |
973 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
974 | goto retry; | |
975 | } | |
976 | if (is_dirty) | |
977 | set_page_dirty(page); | |
72e1c797 | 978 | } |
7bc09003 JK |
979 | } |
980 | out: | |
981 | f2fs_put_page(page, 1); | |
48018b4c | 982 | return err; |
7bc09003 JK |
983 | } |
984 | ||
0a8165d7 | 985 | /* |
7bc09003 JK |
986 | * This function tries to get parent node of victim data block, and identifies |
987 | * data block validity. If the block is valid, copy that with cold status and | |
988 | * modify parent node. | |
989 | * If the parent node is not valid or the data block address is different, | |
990 | * the victim data block is ignored. | |
991 | */ | |
48018b4c | 992 | static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dda2af8 | 993 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type) |
7bc09003 JK |
994 | { |
995 | struct super_block *sb = sbi->sb; | |
996 | struct f2fs_summary *entry; | |
997 | block_t start_addr; | |
43727527 | 998 | int off; |
7bc09003 | 999 | int phase = 0; |
48018b4c | 1000 | int submitted = 0; |
7bc09003 JK |
1001 | |
1002 | start_addr = START_BLOCK(sbi, segno); | |
1003 | ||
1004 | next_step: | |
1005 | entry = sum; | |
c718379b | 1006 | |
7bc09003 JK |
1007 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
1008 | struct page *data_page; | |
1009 | struct inode *inode; | |
1010 | struct node_info dni; /* dnode info for the data */ | |
1011 | unsigned int ofs_in_node, nofs; | |
1012 | block_t start_bidx; | |
7ea984b0 | 1013 | nid_t nid = le32_to_cpu(entry->nid); |
7bc09003 | 1014 | |
43727527 | 1015 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 1016 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
48018b4c | 1017 | return submitted; |
7bc09003 | 1018 | |
43727527 | 1019 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
1020 | continue; |
1021 | ||
1022 | if (phase == 0) { | |
4d57b86d | 1023 | f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
7ea984b0 CY |
1024 | META_NAT, true); |
1025 | continue; | |
1026 | } | |
1027 | ||
1028 | if (phase == 1) { | |
4d57b86d | 1029 | f2fs_ra_node_page(sbi, nid); |
7bc09003 JK |
1030 | continue; |
1031 | } | |
1032 | ||
1033 | /* Get an inode by ino with checking validity */ | |
c1079892 | 1034 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
1035 | continue; |
1036 | ||
7ea984b0 | 1037 | if (phase == 2) { |
4d57b86d | 1038 | f2fs_ra_node_page(sbi, dni.ino); |
7bc09003 JK |
1039 | continue; |
1040 | } | |
1041 | ||
7bc09003 JK |
1042 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
1043 | ||
7ea984b0 | 1044 | if (phase == 3) { |
d4686d56 | 1045 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 1046 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
1047 | continue; |
1048 | ||
bb06664a | 1049 | if (!down_write_trylock( |
b2532c69 | 1050 | &F2FS_I(inode)->i_gc_rwsem[WRITE])) { |
bb06664a | 1051 | iput(inode); |
6f8d4455 | 1052 | sbi->skipped_gc_rwsem++; |
bb06664a CY |
1053 | continue; |
1054 | } | |
1055 | ||
6aa58d8a CY |
1056 | start_bidx = f2fs_start_bidx_of_node(nofs, inode) + |
1057 | ofs_in_node; | |
1058 | ||
1059 | if (f2fs_post_read_required(inode)) { | |
1060 | int err = ra_data_block(inode, start_bidx); | |
1061 | ||
1062 | up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); | |
1063 | if (err) { | |
1064 | iput(inode); | |
1065 | continue; | |
1066 | } | |
1067 | add_gc_inode(gc_list, inode); | |
1068 | continue; | |
1069 | } | |
1070 | ||
4d57b86d | 1071 | data_page = f2fs_get_read_data_page(inode, |
6aa58d8a | 1072 | start_bidx, REQ_RAHEAD, true); |
b2532c69 | 1073 | up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); |
31a32688 CL |
1074 | if (IS_ERR(data_page)) { |
1075 | iput(inode); | |
1076 | continue; | |
1077 | } | |
7bc09003 JK |
1078 | |
1079 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 1080 | add_gc_inode(gc_list, inode); |
31a32688 CL |
1081 | continue; |
1082 | } | |
1083 | ||
7ea984b0 | 1084 | /* phase 4 */ |
7dda2af8 | 1085 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 1086 | if (inode) { |
82e0a5aa CY |
1087 | struct f2fs_inode_info *fi = F2FS_I(inode); |
1088 | bool locked = false; | |
48018b4c | 1089 | int err; |
82e0a5aa CY |
1090 | |
1091 | if (S_ISREG(inode->i_mode)) { | |
b2532c69 | 1092 | if (!down_write_trylock(&fi->i_gc_rwsem[READ])) |
82e0a5aa CY |
1093 | continue; |
1094 | if (!down_write_trylock( | |
b2532c69 | 1095 | &fi->i_gc_rwsem[WRITE])) { |
6f8d4455 | 1096 | sbi->skipped_gc_rwsem++; |
b2532c69 | 1097 | up_write(&fi->i_gc_rwsem[READ]); |
82e0a5aa CY |
1098 | continue; |
1099 | } | |
1100 | locked = true; | |
73ac2f4e CY |
1101 | |
1102 | /* wait for all inflight aio data */ | |
1103 | inode_dio_wait(inode); | |
82e0a5aa CY |
1104 | } |
1105 | ||
4d57b86d | 1106 | start_bidx = f2fs_start_bidx_of_node(nofs, inode) |
c879f90d | 1107 | + ofs_in_node; |
6dbb1796 | 1108 | if (f2fs_post_read_required(inode)) |
48018b4c CY |
1109 | err = move_data_block(inode, start_bidx, |
1110 | gc_type, segno, off); | |
4375a336 | 1111 | else |
48018b4c | 1112 | err = move_data_page(inode, start_bidx, gc_type, |
d4c759ee | 1113 | segno, off); |
82e0a5aa | 1114 | |
48018b4c CY |
1115 | if (!err && (gc_type == FG_GC || |
1116 | f2fs_post_read_required(inode))) | |
1117 | submitted++; | |
1118 | ||
82e0a5aa | 1119 | if (locked) { |
b2532c69 CY |
1120 | up_write(&fi->i_gc_rwsem[WRITE]); |
1121 | up_write(&fi->i_gc_rwsem[READ]); | |
82e0a5aa CY |
1122 | } |
1123 | ||
e1235983 | 1124 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 1125 | } |
7bc09003 | 1126 | } |
c718379b | 1127 | |
7ea984b0 | 1128 | if (++phase < 5) |
7bc09003 | 1129 | goto next_step; |
48018b4c CY |
1130 | |
1131 | return submitted; | |
7bc09003 JK |
1132 | } |
1133 | ||
1134 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 1135 | int gc_type) |
7bc09003 JK |
1136 | { |
1137 | struct sit_info *sit_i = SIT_I(sbi); | |
1138 | int ret; | |
8a2d0ace | 1139 | |
3d26fa6b | 1140 | down_write(&sit_i->sentry_lock); |
8a2d0ace GZ |
1141 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
1142 | NO_CHECK_TYPE, LFS); | |
3d26fa6b | 1143 | up_write(&sit_i->sentry_lock); |
7bc09003 JK |
1144 | return ret; |
1145 | } | |
1146 | ||
718e53fa CY |
1147 | static int do_garbage_collect(struct f2fs_sb_info *sbi, |
1148 | unsigned int start_segno, | |
7dda2af8 | 1149 | struct gc_inode_list *gc_list, int gc_type) |
7bc09003 JK |
1150 | { |
1151 | struct page *sum_page; | |
1152 | struct f2fs_summary_block *sum; | |
c718379b | 1153 | struct blk_plug plug; |
718e53fa CY |
1154 | unsigned int segno = start_segno; |
1155 | unsigned int end_segno = start_segno + sbi->segs_per_sec; | |
e3080b01 | 1156 | int seg_freed = 0, migrated = 0; |
718e53fa CY |
1157 | unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? |
1158 | SUM_TYPE_DATA : SUM_TYPE_NODE; | |
48018b4c | 1159 | int submitted = 0; |
7bc09003 | 1160 | |
e3080b01 CY |
1161 | if (__is_large_section(sbi)) |
1162 | end_segno = rounddown(end_segno, sbi->segs_per_sec); | |
1163 | ||
718e53fa | 1164 | /* readahead multi ssa blocks those have contiguous address */ |
2c70c5e3 | 1165 | if (__is_large_section(sbi)) |
4d57b86d | 1166 | f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), |
e3080b01 | 1167 | end_segno - segno, META_SSA, true); |
718e53fa CY |
1168 | |
1169 | /* reference all summary page */ | |
1170 | while (segno < end_segno) { | |
4d57b86d | 1171 | sum_page = f2fs_get_sum_page(sbi, segno++); |
edc55aaf JK |
1172 | if (IS_ERR(sum_page)) { |
1173 | int err = PTR_ERR(sum_page); | |
1174 | ||
1175 | end_segno = segno - 1; | |
1176 | for (segno = start_segno; segno < end_segno; segno++) { | |
1177 | sum_page = find_get_page(META_MAPPING(sbi), | |
1178 | GET_SUM_BLOCK(sbi, segno)); | |
1179 | f2fs_put_page(sum_page, 0); | |
1180 | f2fs_put_page(sum_page, 0); | |
1181 | } | |
1182 | return err; | |
1183 | } | |
718e53fa CY |
1184 | unlock_page(sum_page); |
1185 | } | |
7bc09003 | 1186 | |
c718379b JK |
1187 | blk_start_plug(&plug); |
1188 | ||
718e53fa | 1189 | for (segno = start_segno; segno < end_segno; segno++) { |
aa987273 | 1190 | |
718e53fa CY |
1191 | /* find segment summary of victim */ |
1192 | sum_page = find_get_page(META_MAPPING(sbi), | |
1193 | GET_SUM_BLOCK(sbi, segno)); | |
718e53fa | 1194 | f2fs_put_page(sum_page, 0); |
7bc09003 | 1195 | |
d6c66cd1 YS |
1196 | if (get_valid_blocks(sbi, segno, false) == 0) |
1197 | goto freed; | |
e3080b01 CY |
1198 | if (__is_large_section(sbi) && |
1199 | migrated >= sbi->migration_granularity) | |
1200 | goto skip; | |
d6c66cd1 | 1201 | if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi))) |
e3080b01 | 1202 | goto skip; |
de0dcc40 | 1203 | |
718e53fa | 1204 | sum = page_address(sum_page); |
10d255c3 | 1205 | if (type != GET_SUM_TYPE((&sum->footer))) { |
dcbb4c10 JP |
1206 | f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT", |
1207 | segno, type, GET_SUM_TYPE((&sum->footer))); | |
10d255c3 | 1208 | set_sbi_flag(sbi, SBI_NEED_FSCK); |
793ab1c8 | 1209 | f2fs_stop_checkpoint(sbi, false); |
e3080b01 | 1210 | goto skip; |
10d255c3 | 1211 | } |
718e53fa CY |
1212 | |
1213 | /* | |
1214 | * this is to avoid deadlock: | |
1215 | * - lock_page(sum_page) - f2fs_replace_block | |
3d26fa6b CY |
1216 | * - check_valid_map() - down_write(sentry_lock) |
1217 | * - down_read(sentry_lock) - change_curseg() | |
718e53fa CY |
1218 | * - lock_page(sum_page) |
1219 | */ | |
718e53fa | 1220 | if (type == SUM_TYPE_NODE) |
48018b4c | 1221 | submitted += gc_node_segment(sbi, sum->entries, segno, |
718e53fa | 1222 | gc_type); |
48018b4c CY |
1223 | else |
1224 | submitted += gc_data_segment(sbi, sum->entries, gc_list, | |
1225 | segno, gc_type); | |
718e53fa CY |
1226 | |
1227 | stat_inc_seg_count(sbi, type, gc_type); | |
c56f16da | 1228 | |
d6c66cd1 | 1229 | freed: |
c56f16da CY |
1230 | if (gc_type == FG_GC && |
1231 | get_valid_blocks(sbi, segno, false) == 0) | |
1232 | seg_freed++; | |
e3080b01 CY |
1233 | migrated++; |
1234 | ||
1235 | if (__is_large_section(sbi) && segno + 1 < end_segno) | |
1236 | sbi->next_victim_seg[gc_type] = segno + 1; | |
1237 | skip: | |
718e53fa CY |
1238 | f2fs_put_page(sum_page, 0); |
1239 | } | |
1240 | ||
48018b4c | 1241 | if (submitted) |
b9109b0e JK |
1242 | f2fs_submit_merged_write(sbi, |
1243 | (type == SUM_TYPE_NODE) ? NODE : DATA); | |
c718379b | 1244 | |
718e53fa | 1245 | blk_finish_plug(&plug); |
7bc09003 | 1246 | |
17d899df CY |
1247 | stat_inc_call_count(sbi->stat_info); |
1248 | ||
c56f16da | 1249 | return seg_freed; |
7bc09003 JK |
1250 | } |
1251 | ||
e066b83c JK |
1252 | int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, |
1253 | bool background, unsigned int segno) | |
7bc09003 | 1254 | { |
d530d4d8 | 1255 | int gc_type = sync ? FG_GC : BG_GC; |
c56f16da CY |
1256 | int sec_freed = 0, seg_freed = 0, total_freed = 0; |
1257 | int ret = 0; | |
d5053a34 | 1258 | struct cp_control cpc; |
e066b83c | 1259 | unsigned int init_segno = segno; |
7dda2af8 CL |
1260 | struct gc_inode_list gc_list = { |
1261 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
f6bb2a2c | 1262 | .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), |
7dda2af8 | 1263 | }; |
2ef79ecb | 1264 | unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC]; |
6f8d4455 | 1265 | unsigned long long first_skipped; |
2ef79ecb | 1266 | unsigned int skipped_round = 0, round = 0; |
d5053a34 | 1267 | |
c56f16da CY |
1268 | trace_f2fs_gc_begin(sbi->sb, sync, background, |
1269 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1270 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1271 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1272 | free_sections(sbi), | |
1273 | free_segments(sbi), | |
1274 | reserved_segments(sbi), | |
1275 | prefree_segments(sbi)); | |
1276 | ||
119ee914 | 1277 | cpc.reason = __get_cp_reason(sbi); |
6f8d4455 JK |
1278 | sbi->skipped_gc_rwsem = 0; |
1279 | first_skipped = last_skipped; | |
7bc09003 | 1280 | gc_more: |
1751e8a6 | 1281 | if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { |
e5dbd956 | 1282 | ret = -EINVAL; |
408e9375 | 1283 | goto stop; |
e5dbd956 | 1284 | } |
6d5a1495 CY |
1285 | if (unlikely(f2fs_cp_error(sbi))) { |
1286 | ret = -EIO; | |
203681f6 | 1287 | goto stop; |
6d5a1495 | 1288 | } |
7bc09003 | 1289 | |
19f4e688 | 1290 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) { |
6e17bfbc | 1291 | /* |
19f4e688 HP |
1292 | * For example, if there are many prefree_segments below given |
1293 | * threshold, we can make them free by checkpoint. Then, we | |
1294 | * secure free segments which doesn't need fggc any more. | |
6e17bfbc | 1295 | */ |
4354994f DR |
1296 | if (prefree_segments(sbi) && |
1297 | !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { | |
4d57b86d | 1298 | ret = f2fs_write_checkpoint(sbi, &cpc); |
8fd5a37e JK |
1299 | if (ret) |
1300 | goto stop; | |
1301 | } | |
19f4e688 HP |
1302 | if (has_not_enough_free_secs(sbi, 0, 0)) |
1303 | gc_type = FG_GC; | |
d64f8047 | 1304 | } |
7bc09003 | 1305 | |
19f4e688 | 1306 | /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ |
c56f16da CY |
1307 | if (gc_type == BG_GC && !background) { |
1308 | ret = -EINVAL; | |
19f4e688 | 1309 | goto stop; |
c56f16da CY |
1310 | } |
1311 | if (!__get_victim(sbi, &segno, gc_type)) { | |
1312 | ret = -ENODATA; | |
408e9375 | 1313 | goto stop; |
c56f16da | 1314 | } |
7bc09003 | 1315 | |
c56f16da CY |
1316 | seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type); |
1317 | if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec) | |
45fe8492 | 1318 | sec_freed++; |
c56f16da | 1319 | total_freed += seg_freed; |
43727527 | 1320 | |
2ef79ecb | 1321 | if (gc_type == FG_GC) { |
6f8d4455 JK |
1322 | if (sbi->skipped_atomic_files[FG_GC] > last_skipped || |
1323 | sbi->skipped_gc_rwsem) | |
2ef79ecb CY |
1324 | skipped_round++; |
1325 | last_skipped = sbi->skipped_atomic_files[FG_GC]; | |
1326 | round++; | |
1327 | } | |
1328 | ||
957fa478 | 1329 | if (gc_type == FG_GC && seg_freed) |
5ec4e49f | 1330 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 1331 | |
6f8d4455 JK |
1332 | if (sync) |
1333 | goto stop; | |
1334 | ||
1335 | if (has_not_enough_free_secs(sbi, sec_freed, 0)) { | |
1336 | if (skipped_round <= MAX_SKIP_GC_COUNT || | |
1337 | skipped_round * 2 < round) { | |
e066b83c | 1338 | segno = NULL_SEGNO; |
d530d4d8 | 1339 | goto gc_more; |
e066b83c | 1340 | } |
43727527 | 1341 | |
6f8d4455 JK |
1342 | if (first_skipped < last_skipped && |
1343 | (last_skipped - first_skipped) > | |
1344 | sbi->skipped_gc_rwsem) { | |
1345 | f2fs_drop_inmem_pages_all(sbi, true); | |
1346 | segno = NULL_SEGNO; | |
1347 | goto gc_more; | |
1348 | } | |
4354994f | 1349 | if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) |
4d57b86d | 1350 | ret = f2fs_write_checkpoint(sbi, &cpc); |
d530d4d8 | 1351 | } |
408e9375 | 1352 | stop: |
e066b83c JK |
1353 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; |
1354 | SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno; | |
c56f16da CY |
1355 | |
1356 | trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed, | |
1357 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1358 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1359 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1360 | free_sections(sbi), | |
1361 | free_segments(sbi), | |
1362 | reserved_segments(sbi), | |
1363 | prefree_segments(sbi)); | |
1364 | ||
7bc09003 JK |
1365 | mutex_unlock(&sbi->gc_mutex); |
1366 | ||
7dda2af8 | 1367 | put_gc_inode(&gc_list); |
d530d4d8 | 1368 | |
61f7725a | 1369 | if (sync && !ret) |
d530d4d8 | 1370 | ret = sec_freed ? 0 : -EAGAIN; |
43727527 | 1371 | return ret; |
7bc09003 JK |
1372 | } |
1373 | ||
4d57b86d | 1374 | void f2fs_build_gc_manager(struct f2fs_sb_info *sbi) |
7bc09003 JK |
1375 | { |
1376 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
e93b9865 | 1377 | |
1ad71a27 | 1378 | sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES; |
d5793249 JK |
1379 | |
1380 | /* give warm/cold data area from slower device */ | |
0916878d | 1381 | if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi)) |
d5793249 JK |
1382 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = |
1383 | GET_SEGNO(sbi, FDEV(0).end_blk) + 1; | |
7bc09003 | 1384 | } |
04f0b2ea QS |
1385 | |
1386 | static int free_segment_range(struct f2fs_sb_info *sbi, unsigned int start, | |
1387 | unsigned int end) | |
1388 | { | |
1389 | int type; | |
1390 | unsigned int segno, next_inuse; | |
1391 | int err = 0; | |
1392 | ||
1393 | /* Move out cursegs from the target range */ | |
1394 | for (type = CURSEG_HOT_DATA; type < NR_CURSEG_TYPE; type++) | |
1395 | allocate_segment_for_resize(sbi, type, start, end); | |
1396 | ||
1397 | /* do GC to move out valid blocks in the range */ | |
1398 | for (segno = start; segno <= end; segno += sbi->segs_per_sec) { | |
1399 | struct gc_inode_list gc_list = { | |
1400 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
1401 | .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), | |
1402 | }; | |
1403 | ||
1404 | mutex_lock(&sbi->gc_mutex); | |
1405 | do_garbage_collect(sbi, segno, &gc_list, FG_GC); | |
1406 | mutex_unlock(&sbi->gc_mutex); | |
1407 | put_gc_inode(&gc_list); | |
1408 | ||
1409 | if (get_valid_blocks(sbi, segno, true)) | |
1410 | return -EAGAIN; | |
1411 | } | |
1412 | ||
1413 | err = f2fs_sync_fs(sbi->sb, 1); | |
1414 | if (err) | |
1415 | return err; | |
1416 | ||
1417 | next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start); | |
1418 | if (next_inuse <= end) { | |
dcbb4c10 JP |
1419 | f2fs_err(sbi, "segno %u should be free but still inuse!", |
1420 | next_inuse); | |
04f0b2ea QS |
1421 | f2fs_bug_on(sbi, 1); |
1422 | } | |
1423 | return err; | |
1424 | } | |
1425 | ||
1426 | static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs) | |
1427 | { | |
1428 | struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi); | |
1429 | int section_count = le32_to_cpu(raw_sb->section_count); | |
1430 | int segment_count = le32_to_cpu(raw_sb->segment_count); | |
1431 | int segment_count_main = le32_to_cpu(raw_sb->segment_count_main); | |
1432 | long long block_count = le64_to_cpu(raw_sb->block_count); | |
1433 | int segs = secs * sbi->segs_per_sec; | |
1434 | ||
1435 | raw_sb->section_count = cpu_to_le32(section_count + secs); | |
1436 | raw_sb->segment_count = cpu_to_le32(segment_count + segs); | |
1437 | raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs); | |
1438 | raw_sb->block_count = cpu_to_le64(block_count + | |
1439 | (long long)segs * sbi->blocks_per_seg); | |
1440 | } | |
1441 | ||
1442 | static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs) | |
1443 | { | |
1444 | int segs = secs * sbi->segs_per_sec; | |
1445 | long long user_block_count = | |
1446 | le64_to_cpu(F2FS_CKPT(sbi)->user_block_count); | |
1447 | ||
1448 | SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs; | |
1449 | MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs; | |
1450 | FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs; | |
1451 | FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs; | |
1452 | F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + | |
1453 | (long long)segs * sbi->blocks_per_seg); | |
1454 | } | |
1455 | ||
1456 | int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count) | |
1457 | { | |
1458 | __u64 old_block_count, shrunk_blocks; | |
1459 | unsigned int secs; | |
1460 | int gc_mode, gc_type; | |
1461 | int err = 0; | |
1462 | __u32 rem; | |
1463 | ||
1464 | old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count); | |
1465 | if (block_count > old_block_count) | |
1466 | return -EINVAL; | |
1467 | ||
1468 | /* new fs size should align to section size */ | |
1469 | div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem); | |
1470 | if (rem) | |
1471 | return -EINVAL; | |
1472 | ||
1473 | if (block_count == old_block_count) | |
1474 | return 0; | |
1475 | ||
1476 | if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { | |
dcbb4c10 | 1477 | f2fs_err(sbi, "Should run fsck to repair first."); |
10f966bb | 1478 | return -EFSCORRUPTED; |
04f0b2ea QS |
1479 | } |
1480 | ||
1481 | if (test_opt(sbi, DISABLE_CHECKPOINT)) { | |
dcbb4c10 | 1482 | f2fs_err(sbi, "Checkpoint should be enabled."); |
04f0b2ea QS |
1483 | return -EINVAL; |
1484 | } | |
1485 | ||
1486 | freeze_bdev(sbi->sb->s_bdev); | |
1487 | ||
1488 | shrunk_blocks = old_block_count - block_count; | |
1489 | secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi)); | |
1490 | spin_lock(&sbi->stat_lock); | |
1491 | if (shrunk_blocks + valid_user_blocks(sbi) + | |
1492 | sbi->current_reserved_blocks + sbi->unusable_block_count + | |
1493 | F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) | |
1494 | err = -ENOSPC; | |
1495 | else | |
1496 | sbi->user_block_count -= shrunk_blocks; | |
1497 | spin_unlock(&sbi->stat_lock); | |
1498 | if (err) { | |
1499 | thaw_bdev(sbi->sb->s_bdev, sbi->sb); | |
1500 | return err; | |
1501 | } | |
1502 | ||
1503 | mutex_lock(&sbi->resize_mutex); | |
1504 | set_sbi_flag(sbi, SBI_IS_RESIZEFS); | |
1505 | ||
1506 | mutex_lock(&DIRTY_I(sbi)->seglist_lock); | |
1507 | ||
1508 | MAIN_SECS(sbi) -= secs; | |
1509 | ||
1510 | for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++) | |
1511 | if (SIT_I(sbi)->last_victim[gc_mode] >= | |
1512 | MAIN_SECS(sbi) * sbi->segs_per_sec) | |
1513 | SIT_I(sbi)->last_victim[gc_mode] = 0; | |
1514 | ||
1515 | for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++) | |
1516 | if (sbi->next_victim_seg[gc_type] >= | |
1517 | MAIN_SECS(sbi) * sbi->segs_per_sec) | |
1518 | sbi->next_victim_seg[gc_type] = NULL_SEGNO; | |
1519 | ||
1520 | mutex_unlock(&DIRTY_I(sbi)->seglist_lock); | |
1521 | ||
1522 | err = free_segment_range(sbi, MAIN_SECS(sbi) * sbi->segs_per_sec, | |
1523 | MAIN_SEGS(sbi) - 1); | |
1524 | if (err) | |
1525 | goto out; | |
1526 | ||
1527 | update_sb_metadata(sbi, -secs); | |
1528 | ||
1529 | err = f2fs_commit_super(sbi, false); | |
1530 | if (err) { | |
1531 | update_sb_metadata(sbi, secs); | |
1532 | goto out; | |
1533 | } | |
1534 | ||
1535 | update_fs_metadata(sbi, -secs); | |
1536 | clear_sbi_flag(sbi, SBI_IS_RESIZEFS); | |
1537 | err = f2fs_sync_fs(sbi->sb, 1); | |
1538 | if (err) { | |
1539 | update_fs_metadata(sbi, secs); | |
1540 | update_sb_metadata(sbi, secs); | |
1541 | f2fs_commit_super(sbi, false); | |
1542 | } | |
1543 | out: | |
1544 | if (err) { | |
1545 | set_sbi_flag(sbi, SBI_NEED_FSCK); | |
dcbb4c10 | 1546 | f2fs_err(sbi, "resize_fs failed, should run fsck to repair!"); |
04f0b2ea QS |
1547 | |
1548 | MAIN_SECS(sbi) += secs; | |
1549 | spin_lock(&sbi->stat_lock); | |
1550 | sbi->user_block_count += shrunk_blocks; | |
1551 | spin_unlock(&sbi->stat_lock); | |
1552 | } | |
1553 | clear_sbi_flag(sbi, SBI_IS_RESIZEFS); | |
1554 | mutex_unlock(&sbi->resize_mutex); | |
1555 | thaw_bdev(sbi->sb->s_bdev, sbi->sb); | |
1556 | return err; | |
1557 | } |