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 data: %-16llu\n",
35 sbi->rw_iostat[APP_BUFFERED_IO]);
36 seq_printf(seq, "app direct data: %-16llu\n",
37 sbi->rw_iostat[APP_DIRECT_IO]);
38 seq_printf(seq, "app mapped data: %-16llu\n",
39 sbi->rw_iostat[APP_MAPPED_IO]);
40 seq_printf(seq, "app buffered cdata: %-16llu\n",
41 sbi->rw_iostat[APP_BUFFERED_CDATA_IO]);
42 seq_printf(seq, "app mapped cdata: %-16llu\n",
43 sbi->rw_iostat[APP_MAPPED_CDATA_IO]);
44
45 /* print fs write IOs */
46 seq_printf(seq, "fs data: %-16llu\n",
47 sbi->rw_iostat[FS_DATA_IO]);
48 seq_printf(seq, "fs cdata: %-16llu\n",
49 sbi->rw_iostat[FS_CDATA_IO]);
50 seq_printf(seq, "fs node: %-16llu\n",
51 sbi->rw_iostat[FS_NODE_IO]);
52 seq_printf(seq, "fs meta: %-16llu\n",
53 sbi->rw_iostat[FS_META_IO]);
54 seq_printf(seq, "fs gc data: %-16llu\n",
55 sbi->rw_iostat[FS_GC_DATA_IO]);
56 seq_printf(seq, "fs gc node: %-16llu\n",
57 sbi->rw_iostat[FS_GC_NODE_IO]);
58 seq_printf(seq, "fs cp data: %-16llu\n",
59 sbi->rw_iostat[FS_CP_DATA_IO]);
60 seq_printf(seq, "fs cp node: %-16llu\n",
61 sbi->rw_iostat[FS_CP_NODE_IO]);
62 seq_printf(seq, "fs cp meta: %-16llu\n",
63 sbi->rw_iostat[FS_CP_META_IO]);
64
65 /* print app read IOs */
66 seq_puts(seq, "[READ]\n");
67 seq_printf(seq, "app buffered data: %-16llu\n",
68 sbi->rw_iostat[APP_BUFFERED_READ_IO]);
69 seq_printf(seq, "app direct data: %-16llu\n",
70 sbi->rw_iostat[APP_DIRECT_READ_IO]);
71 seq_printf(seq, "app mapped data: %-16llu\n",
72 sbi->rw_iostat[APP_MAPPED_READ_IO]);
73 seq_printf(seq, "app buffered cdata: %-16llu\n",
74 sbi->rw_iostat[APP_BUFFERED_CDATA_READ_IO]);
75 seq_printf(seq, "app mapped cdata: %-16llu\n",
76 sbi->rw_iostat[APP_MAPPED_CDATA_READ_IO]);
77
78 /* print fs read IOs */
79 seq_printf(seq, "fs data: %-16llu\n",
80 sbi->rw_iostat[FS_DATA_READ_IO]);
81 seq_printf(seq, "fs gc data: %-16llu\n",
82 sbi->rw_iostat[FS_GDATA_READ_IO]);
83 seq_printf(seq, "fs cdata: %-16llu\n",
84 sbi->rw_iostat[FS_CDATA_READ_IO]);
85 seq_printf(seq, "fs node: %-16llu\n",
86 sbi->rw_iostat[FS_NODE_READ_IO]);
87 seq_printf(seq, "fs meta: %-16llu\n",
88 sbi->rw_iostat[FS_META_READ_IO]);
89
90 /* print other IOs */
91 seq_puts(seq, "[OTHER]\n");
92 seq_printf(seq, "fs discard: %-16llu\n",
93 sbi->rw_iostat[FS_DISCARD]);
94
95 return 0;
96 }
97
__record_iostat_latency(struct f2fs_sb_info * sbi)98 static inline void __record_iostat_latency(struct f2fs_sb_info *sbi)
99 {
100 int io, idx = 0;
101 unsigned int cnt;
102 struct f2fs_iostat_latency iostat_lat[MAX_IO_TYPE][NR_PAGE_TYPE];
103 struct iostat_lat_info *io_lat = sbi->iostat_io_lat;
104 unsigned long flags;
105
106 spin_lock_irqsave(&sbi->iostat_lat_lock, flags);
107 for (idx = 0; idx < MAX_IO_TYPE; idx++) {
108 for (io = 0; io < NR_PAGE_TYPE; io++) {
109 cnt = io_lat->bio_cnt[idx][io];
110 iostat_lat[idx][io].peak_lat =
111 jiffies_to_msecs(io_lat->peak_lat[idx][io]);
112 iostat_lat[idx][io].cnt = cnt;
113 iostat_lat[idx][io].avg_lat = cnt ?
114 jiffies_to_msecs(io_lat->sum_lat[idx][io]) / cnt : 0;
115 io_lat->sum_lat[idx][io] = 0;
116 io_lat->peak_lat[idx][io] = 0;
117 io_lat->bio_cnt[idx][io] = 0;
118 }
119 }
120 spin_unlock_irqrestore(&sbi->iostat_lat_lock, flags);
121
122 trace_f2fs_iostat_latency(sbi, iostat_lat);
123 }
124
f2fs_record_iostat(struct f2fs_sb_info * sbi)125 static inline void f2fs_record_iostat(struct f2fs_sb_info *sbi)
126 {
127 unsigned long long iostat_diff[NR_IO_TYPE];
128 int i;
129 unsigned long flags;
130
131 if (time_is_after_jiffies(sbi->iostat_next_period))
132 return;
133
134 /* Need double check under the lock */
135 spin_lock_irqsave(&sbi->iostat_lock, flags);
136 if (time_is_after_jiffies(sbi->iostat_next_period)) {
137 spin_unlock_irqrestore(&sbi->iostat_lock, flags);
138 return;
139 }
140 sbi->iostat_next_period = jiffies +
141 msecs_to_jiffies(sbi->iostat_period_ms);
142
143 for (i = 0; i < NR_IO_TYPE; i++) {
144 iostat_diff[i] = sbi->rw_iostat[i] -
145 sbi->prev_rw_iostat[i];
146 sbi->prev_rw_iostat[i] = sbi->rw_iostat[i];
147 }
148 spin_unlock_irqrestore(&sbi->iostat_lock, flags);
149
150 trace_f2fs_iostat(sbi, iostat_diff);
151
152 __record_iostat_latency(sbi);
153 }
154
f2fs_reset_iostat(struct f2fs_sb_info * sbi)155 void f2fs_reset_iostat(struct f2fs_sb_info *sbi)
156 {
157 struct iostat_lat_info *io_lat = sbi->iostat_io_lat;
158 int i;
159
160 spin_lock_irq(&sbi->iostat_lock);
161 for (i = 0; i < NR_IO_TYPE; i++) {
162 sbi->rw_iostat[i] = 0;
163 sbi->prev_rw_iostat[i] = 0;
164 }
165 spin_unlock_irq(&sbi->iostat_lock);
166
167 spin_lock_irq(&sbi->iostat_lat_lock);
168 memset(io_lat, 0, sizeof(struct iostat_lat_info));
169 spin_unlock_irq(&sbi->iostat_lat_lock);
170 }
171
f2fs_update_iostat(struct f2fs_sb_info * sbi,struct inode * inode,enum iostat_type type,unsigned long long io_bytes)172 void f2fs_update_iostat(struct f2fs_sb_info *sbi, struct inode *inode,
173 enum iostat_type type, unsigned long long io_bytes)
174 {
175 unsigned long flags;
176
177 if (!sbi->iostat_enable)
178 return;
179
180 spin_lock_irqsave(&sbi->iostat_lock, flags);
181 sbi->rw_iostat[type] += io_bytes;
182
183 if (type == APP_BUFFERED_IO || type == APP_DIRECT_IO)
184 sbi->rw_iostat[APP_WRITE_IO] += io_bytes;
185
186 if (type == APP_BUFFERED_READ_IO || type == APP_DIRECT_READ_IO)
187 sbi->rw_iostat[APP_READ_IO] += io_bytes;
188
189 #ifdef CONFIG_F2FS_FS_COMPRESSION
190 if (inode && f2fs_compressed_file(inode)) {
191 if (type == APP_BUFFERED_IO)
192 sbi->rw_iostat[APP_BUFFERED_CDATA_IO] += io_bytes;
193
194 if (type == APP_BUFFERED_READ_IO)
195 sbi->rw_iostat[APP_BUFFERED_CDATA_READ_IO] += io_bytes;
196
197 if (type == APP_MAPPED_READ_IO)
198 sbi->rw_iostat[APP_MAPPED_CDATA_READ_IO] += io_bytes;
199
200 if (type == APP_MAPPED_IO)
201 sbi->rw_iostat[APP_MAPPED_CDATA_IO] += io_bytes;
202
203 if (type == FS_DATA_READ_IO)
204 sbi->rw_iostat[FS_CDATA_READ_IO] += io_bytes;
205
206 if (type == FS_DATA_IO)
207 sbi->rw_iostat[FS_CDATA_IO] += io_bytes;
208 }
209 #endif
210
211 spin_unlock_irqrestore(&sbi->iostat_lock, flags);
212
213 f2fs_record_iostat(sbi);
214 }
215
__update_iostat_latency(struct bio_iostat_ctx * iostat_ctx,int rw,bool is_sync)216 static inline void __update_iostat_latency(struct bio_iostat_ctx *iostat_ctx,
217 int rw, bool is_sync)
218 {
219 unsigned long ts_diff;
220 unsigned int iotype = iostat_ctx->type;
221 struct f2fs_sb_info *sbi = iostat_ctx->sbi;
222 struct iostat_lat_info *io_lat = sbi->iostat_io_lat;
223 int idx;
224 unsigned long flags;
225
226 if (!sbi->iostat_enable)
227 return;
228
229 ts_diff = jiffies - iostat_ctx->submit_ts;
230 if (iotype >= META_FLUSH)
231 iotype = META;
232
233 if (rw == 0) {
234 idx = READ_IO;
235 } else {
236 if (is_sync)
237 idx = WRITE_SYNC_IO;
238 else
239 idx = WRITE_ASYNC_IO;
240 }
241
242 spin_lock_irqsave(&sbi->iostat_lat_lock, flags);
243 io_lat->sum_lat[idx][iotype] += ts_diff;
244 io_lat->bio_cnt[idx][iotype]++;
245 if (ts_diff > io_lat->peak_lat[idx][iotype])
246 io_lat->peak_lat[idx][iotype] = ts_diff;
247 spin_unlock_irqrestore(&sbi->iostat_lat_lock, flags);
248 }
249
iostat_update_and_unbind_ctx(struct bio * bio,int rw)250 void iostat_update_and_unbind_ctx(struct bio *bio, int rw)
251 {
252 struct bio_iostat_ctx *iostat_ctx = bio->bi_private;
253 bool is_sync = bio->bi_opf & REQ_SYNC;
254
255 if (rw == 0)
256 bio->bi_private = iostat_ctx->post_read_ctx;
257 else
258 bio->bi_private = iostat_ctx->sbi;
259 __update_iostat_latency(iostat_ctx, rw, is_sync);
260 mempool_free(iostat_ctx, bio_iostat_ctx_pool);
261 }
262
iostat_alloc_and_bind_ctx(struct f2fs_sb_info * sbi,struct bio * bio,struct bio_post_read_ctx * ctx)263 void iostat_alloc_and_bind_ctx(struct f2fs_sb_info *sbi,
264 struct bio *bio, struct bio_post_read_ctx *ctx)
265 {
266 struct bio_iostat_ctx *iostat_ctx;
267 /* Due to the mempool, this never fails. */
268 iostat_ctx = mempool_alloc(bio_iostat_ctx_pool, GFP_NOFS);
269 iostat_ctx->sbi = sbi;
270 iostat_ctx->submit_ts = 0;
271 iostat_ctx->type = 0;
272 iostat_ctx->post_read_ctx = ctx;
273 bio->bi_private = iostat_ctx;
274 }
275
f2fs_init_iostat_processing(void)276 int __init f2fs_init_iostat_processing(void)
277 {
278 bio_iostat_ctx_cache =
279 kmem_cache_create("f2fs_bio_iostat_ctx",
280 sizeof(struct bio_iostat_ctx), 0, 0, NULL);
281 if (!bio_iostat_ctx_cache)
282 goto fail;
283 bio_iostat_ctx_pool =
284 mempool_create_slab_pool(NUM_PREALLOC_IOSTAT_CTXS,
285 bio_iostat_ctx_cache);
286 if (!bio_iostat_ctx_pool)
287 goto fail_free_cache;
288 return 0;
289
290 fail_free_cache:
291 kmem_cache_destroy(bio_iostat_ctx_cache);
292 fail:
293 return -ENOMEM;
294 }
295
f2fs_destroy_iostat_processing(void)296 void f2fs_destroy_iostat_processing(void)
297 {
298 mempool_destroy(bio_iostat_ctx_pool);
299 kmem_cache_destroy(bio_iostat_ctx_cache);
300 }
301
f2fs_init_iostat(struct f2fs_sb_info * sbi)302 int f2fs_init_iostat(struct f2fs_sb_info *sbi)
303 {
304 /* init iostat info */
305 spin_lock_init(&sbi->iostat_lock);
306 spin_lock_init(&sbi->iostat_lat_lock);
307 sbi->iostat_enable = false;
308 sbi->iostat_period_ms = DEFAULT_IOSTAT_PERIOD_MS;
309 sbi->iostat_io_lat = f2fs_kzalloc(sbi, sizeof(struct iostat_lat_info),
310 GFP_KERNEL);
311 if (!sbi->iostat_io_lat)
312 return -ENOMEM;
313
314 return 0;
315 }
316
f2fs_destroy_iostat(struct f2fs_sb_info * sbi)317 void f2fs_destroy_iostat(struct f2fs_sb_info *sbi)
318 {
319 kfree(sbi->iostat_io_lat);
320 }
321