1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * f2fs iostat support
4  *
5  * Copyright 2021 Google LLC
6  * Author: Daeho Jeong <daehojeong@google.com>
7  */
8 
9 #include <linux/fs.h>
10 #include <linux/f2fs_fs.h>
11 #include <linux/seq_file.h>
12 
13 #include "f2fs.h"
14 #include "iostat.h"
15 #include <trace/events/f2fs.h>
16 
17 #define NUM_PREALLOC_IOSTAT_CTXS	128
18 static struct kmem_cache *bio_iostat_ctx_cache;
19 static mempool_t *bio_iostat_ctx_pool;
20 
iostat_info_seq_show(struct seq_file * seq,void * offset)21 int __maybe_unused iostat_info_seq_show(struct seq_file *seq, void *offset)
22 {
23 	struct super_block *sb = seq->private;
24 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
25 	time64_t now = ktime_get_real_seconds();
26 
27 	if (!sbi->iostat_enable)
28 		return 0;
29 
30 	seq_printf(seq, "time:		%-16llu\n", now);
31 
32 	/* print app write IOs */
33 	seq_puts(seq, "[WRITE]\n");
34 	seq_printf(seq, "app buffered:	%-16llu\n",
35 				sbi->rw_iostat[APP_BUFFERED_IO]);
36 	seq_printf(seq, "app direct:	%-16llu\n",
37 				sbi->rw_iostat[APP_DIRECT_IO]);
38 	seq_printf(seq, "app mapped:	%-16llu\n",
39 				sbi->rw_iostat[APP_MAPPED_IO]);
40 
41 	/* print fs write IOs */
42 	seq_printf(seq, "fs data:	%-16llu\n",
43 				sbi->rw_iostat[FS_DATA_IO]);
44 	seq_printf(seq, "fs node:	%-16llu\n",
45 				sbi->rw_iostat[FS_NODE_IO]);
46 	seq_printf(seq, "fs meta:	%-16llu\n",
47 				sbi->rw_iostat[FS_META_IO]);
48 	seq_printf(seq, "fs gc data:	%-16llu\n",
49 				sbi->rw_iostat[FS_GC_DATA_IO]);
50 	seq_printf(seq, "fs gc node:	%-16llu\n",
51 				sbi->rw_iostat[FS_GC_NODE_IO]);
52 	seq_printf(seq, "fs cp data:	%-16llu\n",
53 				sbi->rw_iostat[FS_CP_DATA_IO]);
54 	seq_printf(seq, "fs cp node:	%-16llu\n",
55 				sbi->rw_iostat[FS_CP_NODE_IO]);
56 	seq_printf(seq, "fs cp meta:	%-16llu\n",
57 				sbi->rw_iostat[FS_CP_META_IO]);
58 
59 	/* print app read IOs */
60 	seq_puts(seq, "[READ]\n");
61 	seq_printf(seq, "app buffered:	%-16llu\n",
62 				sbi->rw_iostat[APP_BUFFERED_READ_IO]);
63 	seq_printf(seq, "app direct:	%-16llu\n",
64 				sbi->rw_iostat[APP_DIRECT_READ_IO]);
65 	seq_printf(seq, "app mapped:	%-16llu\n",
66 				sbi->rw_iostat[APP_MAPPED_READ_IO]);
67 
68 	/* print fs read IOs */
69 	seq_printf(seq, "fs data:	%-16llu\n",
70 				sbi->rw_iostat[FS_DATA_READ_IO]);
71 	seq_printf(seq, "fs gc data:	%-16llu\n",
72 				sbi->rw_iostat[FS_GDATA_READ_IO]);
73 	seq_printf(seq, "fs compr_data:	%-16llu\n",
74 				sbi->rw_iostat[FS_CDATA_READ_IO]);
75 	seq_printf(seq, "fs node:	%-16llu\n",
76 				sbi->rw_iostat[FS_NODE_READ_IO]);
77 	seq_printf(seq, "fs meta:	%-16llu\n",
78 				sbi->rw_iostat[FS_META_READ_IO]);
79 
80 	/* print other IOs */
81 	seq_puts(seq, "[OTHER]\n");
82 	seq_printf(seq, "fs discard:	%-16llu\n",
83 				sbi->rw_iostat[FS_DISCARD]);
84 
85 	return 0;
86 }
87 
__record_iostat_latency(struct f2fs_sb_info * sbi)88 static inline void __record_iostat_latency(struct f2fs_sb_info *sbi)
89 {
90 	int io, idx = 0;
91 	unsigned int cnt;
92 	struct f2fs_iostat_latency iostat_lat[MAX_IO_TYPE][NR_PAGE_TYPE];
93 	struct iostat_lat_info *io_lat = sbi->iostat_io_lat;
94 	unsigned long flags;
95 
96 	spin_lock_irqsave(&sbi->iostat_lat_lock, flags);
97 	for (idx = 0; idx < MAX_IO_TYPE; idx++) {
98 		for (io = 0; io < NR_PAGE_TYPE; io++) {
99 			cnt = io_lat->bio_cnt[idx][io];
100 			iostat_lat[idx][io].peak_lat =
101 			   jiffies_to_msecs(io_lat->peak_lat[idx][io]);
102 			iostat_lat[idx][io].cnt = cnt;
103 			iostat_lat[idx][io].avg_lat = cnt ?
104 			   jiffies_to_msecs(io_lat->sum_lat[idx][io]) / cnt : 0;
105 			io_lat->sum_lat[idx][io] = 0;
106 			io_lat->peak_lat[idx][io] = 0;
107 			io_lat->bio_cnt[idx][io] = 0;
108 		}
109 	}
110 	spin_unlock_irqrestore(&sbi->iostat_lat_lock, flags);
111 
112 	trace_f2fs_iostat_latency(sbi, iostat_lat);
113 }
114 
f2fs_record_iostat(struct f2fs_sb_info * sbi)115 static inline void f2fs_record_iostat(struct f2fs_sb_info *sbi)
116 {
117 	unsigned long long iostat_diff[NR_IO_TYPE];
118 	int i;
119 	unsigned long flags;
120 
121 	if (time_is_after_jiffies(sbi->iostat_next_period))
122 		return;
123 
124 	/* Need double check under the lock */
125 	spin_lock_irqsave(&sbi->iostat_lock, flags);
126 	if (time_is_after_jiffies(sbi->iostat_next_period)) {
127 		spin_unlock_irqrestore(&sbi->iostat_lock, flags);
128 		return;
129 	}
130 	sbi->iostat_next_period = jiffies +
131 				msecs_to_jiffies(sbi->iostat_period_ms);
132 
133 	for (i = 0; i < NR_IO_TYPE; i++) {
134 		iostat_diff[i] = sbi->rw_iostat[i] -
135 				sbi->prev_rw_iostat[i];
136 		sbi->prev_rw_iostat[i] = sbi->rw_iostat[i];
137 	}
138 	spin_unlock_irqrestore(&sbi->iostat_lock, flags);
139 
140 	trace_f2fs_iostat(sbi, iostat_diff);
141 
142 	__record_iostat_latency(sbi);
143 }
144 
f2fs_reset_iostat(struct f2fs_sb_info * sbi)145 void f2fs_reset_iostat(struct f2fs_sb_info *sbi)
146 {
147 	struct iostat_lat_info *io_lat = sbi->iostat_io_lat;
148 	int i;
149 
150 	spin_lock_irq(&sbi->iostat_lock);
151 	for (i = 0; i < NR_IO_TYPE; i++) {
152 		sbi->rw_iostat[i] = 0;
153 		sbi->prev_rw_iostat[i] = 0;
154 	}
155 	spin_unlock_irq(&sbi->iostat_lock);
156 
157 	spin_lock_irq(&sbi->iostat_lat_lock);
158 	memset(io_lat, 0, sizeof(struct iostat_lat_info));
159 	spin_unlock_irq(&sbi->iostat_lat_lock);
160 }
161 
f2fs_update_iostat(struct f2fs_sb_info * sbi,enum iostat_type type,unsigned long long io_bytes)162 void f2fs_update_iostat(struct f2fs_sb_info *sbi,
163 			enum iostat_type type, unsigned long long io_bytes)
164 {
165 	unsigned long flags;
166 
167 	if (!sbi->iostat_enable)
168 		return;
169 
170 	spin_lock_irqsave(&sbi->iostat_lock, flags);
171 	sbi->rw_iostat[type] += io_bytes;
172 
173 	if (type == APP_BUFFERED_IO || type == APP_DIRECT_IO)
174 		sbi->rw_iostat[APP_WRITE_IO] += io_bytes;
175 
176 	if (type == APP_BUFFERED_READ_IO || type == APP_DIRECT_READ_IO)
177 		sbi->rw_iostat[APP_READ_IO] += io_bytes;
178 
179 	spin_unlock_irqrestore(&sbi->iostat_lock, flags);
180 
181 	f2fs_record_iostat(sbi);
182 }
183 
__update_iostat_latency(struct bio_iostat_ctx * iostat_ctx,int rw,bool is_sync)184 static inline void __update_iostat_latency(struct bio_iostat_ctx *iostat_ctx,
185 				int rw, bool is_sync)
186 {
187 	unsigned long ts_diff;
188 	unsigned int iotype = iostat_ctx->type;
189 	struct f2fs_sb_info *sbi = iostat_ctx->sbi;
190 	struct iostat_lat_info *io_lat = sbi->iostat_io_lat;
191 	int idx;
192 	unsigned long flags;
193 
194 	if (!sbi->iostat_enable)
195 		return;
196 
197 	ts_diff = jiffies - iostat_ctx->submit_ts;
198 	if (iotype >= META_FLUSH)
199 		iotype = META;
200 
201 	if (rw == 0) {
202 		idx = READ_IO;
203 	} else {
204 		if (is_sync)
205 			idx = WRITE_SYNC_IO;
206 		else
207 			idx = WRITE_ASYNC_IO;
208 	}
209 
210 	spin_lock_irqsave(&sbi->iostat_lat_lock, flags);
211 	io_lat->sum_lat[idx][iotype] += ts_diff;
212 	io_lat->bio_cnt[idx][iotype]++;
213 	if (ts_diff > io_lat->peak_lat[idx][iotype])
214 		io_lat->peak_lat[idx][iotype] = ts_diff;
215 	spin_unlock_irqrestore(&sbi->iostat_lat_lock, flags);
216 }
217 
iostat_update_and_unbind_ctx(struct bio * bio,int rw)218 void iostat_update_and_unbind_ctx(struct bio *bio, int rw)
219 {
220 	struct bio_iostat_ctx *iostat_ctx = bio->bi_private;
221 	bool is_sync = bio->bi_opf & REQ_SYNC;
222 
223 	if (rw == 0)
224 		bio->bi_private = iostat_ctx->post_read_ctx;
225 	else
226 		bio->bi_private = iostat_ctx->sbi;
227 	__update_iostat_latency(iostat_ctx, rw, is_sync);
228 	mempool_free(iostat_ctx, bio_iostat_ctx_pool);
229 }
230 
iostat_alloc_and_bind_ctx(struct f2fs_sb_info * sbi,struct bio * bio,struct bio_post_read_ctx * ctx)231 void iostat_alloc_and_bind_ctx(struct f2fs_sb_info *sbi,
232 		struct bio *bio, struct bio_post_read_ctx *ctx)
233 {
234 	struct bio_iostat_ctx *iostat_ctx;
235 	/* Due to the mempool, this never fails. */
236 	iostat_ctx = mempool_alloc(bio_iostat_ctx_pool, GFP_NOFS);
237 	iostat_ctx->sbi = sbi;
238 	iostat_ctx->submit_ts = 0;
239 	iostat_ctx->type = 0;
240 	iostat_ctx->post_read_ctx = ctx;
241 	bio->bi_private = iostat_ctx;
242 }
243 
f2fs_init_iostat_processing(void)244 int __init f2fs_init_iostat_processing(void)
245 {
246 	bio_iostat_ctx_cache =
247 		kmem_cache_create("f2fs_bio_iostat_ctx",
248 				  sizeof(struct bio_iostat_ctx), 0, 0, NULL);
249 	if (!bio_iostat_ctx_cache)
250 		goto fail;
251 	bio_iostat_ctx_pool =
252 		mempool_create_slab_pool(NUM_PREALLOC_IOSTAT_CTXS,
253 					 bio_iostat_ctx_cache);
254 	if (!bio_iostat_ctx_pool)
255 		goto fail_free_cache;
256 	return 0;
257 
258 fail_free_cache:
259 	kmem_cache_destroy(bio_iostat_ctx_cache);
260 fail:
261 	return -ENOMEM;
262 }
263 
f2fs_destroy_iostat_processing(void)264 void f2fs_destroy_iostat_processing(void)
265 {
266 	mempool_destroy(bio_iostat_ctx_pool);
267 	kmem_cache_destroy(bio_iostat_ctx_cache);
268 }
269 
f2fs_init_iostat(struct f2fs_sb_info * sbi)270 int f2fs_init_iostat(struct f2fs_sb_info *sbi)
271 {
272 	/* init iostat info */
273 	spin_lock_init(&sbi->iostat_lock);
274 	spin_lock_init(&sbi->iostat_lat_lock);
275 	sbi->iostat_enable = false;
276 	sbi->iostat_period_ms = DEFAULT_IOSTAT_PERIOD_MS;
277 	sbi->iostat_io_lat = f2fs_kzalloc(sbi, sizeof(struct iostat_lat_info),
278 					GFP_KERNEL);
279 	if (!sbi->iostat_io_lat)
280 		return -ENOMEM;
281 
282 	return 0;
283 }
284 
f2fs_destroy_iostat(struct f2fs_sb_info * sbi)285 void f2fs_destroy_iostat(struct f2fs_sb_info *sbi)
286 {
287 	kfree(sbi->iostat_io_lat);
288 }
289