1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Block rq-qos policy for assigning an I/O priority class to requests.
4 *
5 * Using an rq-qos policy for assigning I/O priority class has two advantages
6 * over using the ioprio_set() system call:
7 *
8 * - This policy is cgroup based so it has all the advantages of cgroups.
9 * - While ioprio_set() does not affect page cache writeback I/O, this rq-qos
10 * controller affects page cache writeback I/O for filesystems that support
11 * assiociating a cgroup with writeback I/O. See also
12 * Documentation/admin-guide/cgroup-v2.rst.
13 */
14
15 #include <linux/blk-cgroup.h>
16 #include <linux/blk-mq.h>
17 #include <linux/blk_types.h>
18 #include <linux/kernel.h>
19 #include <linux/module.h>
20 #include "blk-ioprio.h"
21 #include "blk-rq-qos.h"
22
23 /**
24 * enum prio_policy - I/O priority class policy.
25 * @POLICY_NO_CHANGE: (default) do not modify the I/O priority class.
26 * @POLICY_NONE_TO_RT: modify IOPRIO_CLASS_NONE into IOPRIO_CLASS_RT.
27 * @POLICY_RESTRICT_TO_BE: modify IOPRIO_CLASS_NONE and IOPRIO_CLASS_RT into
28 * IOPRIO_CLASS_BE.
29 * @POLICY_ALL_TO_IDLE: change the I/O priority class into IOPRIO_CLASS_IDLE.
30 *
31 * See also <linux/ioprio.h>.
32 */
33 enum prio_policy {
34 POLICY_NO_CHANGE = 0,
35 POLICY_NONE_TO_RT = 1,
36 POLICY_RESTRICT_TO_BE = 2,
37 POLICY_ALL_TO_IDLE = 3,
38 };
39
40 static const char *policy_name[] = {
41 [POLICY_NO_CHANGE] = "no-change",
42 [POLICY_NONE_TO_RT] = "none-to-rt",
43 [POLICY_RESTRICT_TO_BE] = "restrict-to-be",
44 [POLICY_ALL_TO_IDLE] = "idle",
45 };
46
47 static struct blkcg_policy ioprio_policy;
48
49 /**
50 * struct ioprio_blkg - Per (cgroup, request queue) data.
51 * @pd: blkg_policy_data structure.
52 */
53 struct ioprio_blkg {
54 struct blkg_policy_data pd;
55 };
56
57 /**
58 * struct ioprio_blkcg - Per cgroup data.
59 * @cpd: blkcg_policy_data structure.
60 * @prio_policy: One of the IOPRIO_CLASS_* values. See also <linux/ioprio.h>.
61 */
62 struct ioprio_blkcg {
63 struct blkcg_policy_data cpd;
64 enum prio_policy prio_policy;
65 };
66
pd_to_ioprio(struct blkg_policy_data * pd)67 static inline struct ioprio_blkg *pd_to_ioprio(struct blkg_policy_data *pd)
68 {
69 return pd ? container_of(pd, struct ioprio_blkg, pd) : NULL;
70 }
71
blkcg_to_ioprio_blkcg(struct blkcg * blkcg)72 static struct ioprio_blkcg *blkcg_to_ioprio_blkcg(struct blkcg *blkcg)
73 {
74 return container_of(blkcg_to_cpd(blkcg, &ioprio_policy),
75 struct ioprio_blkcg, cpd);
76 }
77
78 static struct ioprio_blkcg *
ioprio_blkcg_from_css(struct cgroup_subsys_state * css)79 ioprio_blkcg_from_css(struct cgroup_subsys_state *css)
80 {
81 return blkcg_to_ioprio_blkcg(css_to_blkcg(css));
82 }
83
ioprio_blkcg_from_bio(struct bio * bio)84 static struct ioprio_blkcg *ioprio_blkcg_from_bio(struct bio *bio)
85 {
86 struct blkg_policy_data *pd = blkg_to_pd(bio->bi_blkg, &ioprio_policy);
87
88 if (!pd)
89 return NULL;
90
91 return blkcg_to_ioprio_blkcg(pd->blkg->blkcg);
92 }
93
ioprio_show_prio_policy(struct seq_file * sf,void * v)94 static int ioprio_show_prio_policy(struct seq_file *sf, void *v)
95 {
96 struct ioprio_blkcg *blkcg = ioprio_blkcg_from_css(seq_css(sf));
97
98 seq_printf(sf, "%s\n", policy_name[blkcg->prio_policy]);
99 return 0;
100 }
101
ioprio_set_prio_policy(struct kernfs_open_file * of,char * buf,size_t nbytes,loff_t off)102 static ssize_t ioprio_set_prio_policy(struct kernfs_open_file *of, char *buf,
103 size_t nbytes, loff_t off)
104 {
105 struct ioprio_blkcg *blkcg = ioprio_blkcg_from_css(of_css(of));
106 int ret;
107
108 if (off != 0)
109 return -EIO;
110 /* kernfs_fop_write_iter() terminates 'buf' with '\0'. */
111 ret = sysfs_match_string(policy_name, buf);
112 if (ret < 0)
113 return ret;
114 blkcg->prio_policy = ret;
115
116 return nbytes;
117 }
118
119 static struct blkg_policy_data *
ioprio_alloc_pd(gfp_t gfp,struct request_queue * q,struct blkcg * blkcg)120 ioprio_alloc_pd(gfp_t gfp, struct request_queue *q, struct blkcg *blkcg)
121 {
122 struct ioprio_blkg *ioprio_blkg;
123
124 ioprio_blkg = kzalloc(sizeof(*ioprio_blkg), gfp);
125 if (!ioprio_blkg)
126 return NULL;
127
128 return &ioprio_blkg->pd;
129 }
130
ioprio_free_pd(struct blkg_policy_data * pd)131 static void ioprio_free_pd(struct blkg_policy_data *pd)
132 {
133 struct ioprio_blkg *ioprio_blkg = pd_to_ioprio(pd);
134
135 kfree(ioprio_blkg);
136 }
137
ioprio_alloc_cpd(gfp_t gfp)138 static struct blkcg_policy_data *ioprio_alloc_cpd(gfp_t gfp)
139 {
140 struct ioprio_blkcg *blkcg;
141
142 blkcg = kzalloc(sizeof(*blkcg), gfp);
143 if (!blkcg)
144 return NULL;
145 blkcg->prio_policy = POLICY_NO_CHANGE;
146 return &blkcg->cpd;
147 }
148
ioprio_free_cpd(struct blkcg_policy_data * cpd)149 static void ioprio_free_cpd(struct blkcg_policy_data *cpd)
150 {
151 struct ioprio_blkcg *blkcg = container_of(cpd, typeof(*blkcg), cpd);
152
153 kfree(blkcg);
154 }
155
156 #define IOPRIO_ATTRS \
157 { \
158 .name = "prio.class", \
159 .seq_show = ioprio_show_prio_policy, \
160 .write = ioprio_set_prio_policy, \
161 }, \
162 { } /* sentinel */
163
164 /* cgroup v2 attributes */
165 static struct cftype ioprio_files[] = {
166 IOPRIO_ATTRS
167 };
168
169 /* cgroup v1 attributes */
170 static struct cftype ioprio_legacy_files[] = {
171 IOPRIO_ATTRS
172 };
173
174 static struct blkcg_policy ioprio_policy = {
175 .dfl_cftypes = ioprio_files,
176 .legacy_cftypes = ioprio_legacy_files,
177
178 .cpd_alloc_fn = ioprio_alloc_cpd,
179 .cpd_free_fn = ioprio_free_cpd,
180
181 .pd_alloc_fn = ioprio_alloc_pd,
182 .pd_free_fn = ioprio_free_pd,
183 };
184
185 struct blk_ioprio {
186 struct rq_qos rqos;
187 };
188
blkcg_ioprio_track(struct rq_qos * rqos,struct request * rq,struct bio * bio)189 static void blkcg_ioprio_track(struct rq_qos *rqos, struct request *rq,
190 struct bio *bio)
191 {
192 struct ioprio_blkcg *blkcg = ioprio_blkcg_from_bio(bio);
193
194 /*
195 * Except for IOPRIO_CLASS_NONE, higher I/O priority numbers
196 * correspond to a lower priority. Hence, the max_t() below selects
197 * the lower priority of bi_ioprio and the cgroup I/O priority class.
198 * If the cgroup policy has been set to POLICY_NO_CHANGE == 0, the
199 * bio I/O priority is not modified. If the bio I/O priority equals
200 * IOPRIO_CLASS_NONE, the cgroup I/O priority is assigned to the bio.
201 */
202 bio->bi_ioprio = max_t(u16, bio->bi_ioprio,
203 IOPRIO_PRIO_VALUE(blkcg->prio_policy, 0));
204 }
205
blkcg_ioprio_exit(struct rq_qos * rqos)206 static void blkcg_ioprio_exit(struct rq_qos *rqos)
207 {
208 struct blk_ioprio *blkioprio_blkg =
209 container_of(rqos, typeof(*blkioprio_blkg), rqos);
210
211 blkcg_deactivate_policy(rqos->q, &ioprio_policy);
212 kfree(blkioprio_blkg);
213 }
214
215 static struct rq_qos_ops blkcg_ioprio_ops = {
216 .track = blkcg_ioprio_track,
217 .exit = blkcg_ioprio_exit,
218 };
219
blk_ioprio_init(struct request_queue * q)220 int blk_ioprio_init(struct request_queue *q)
221 {
222 struct blk_ioprio *blkioprio_blkg;
223 struct rq_qos *rqos;
224 int ret;
225
226 blkioprio_blkg = kzalloc(sizeof(*blkioprio_blkg), GFP_KERNEL);
227 if (!blkioprio_blkg)
228 return -ENOMEM;
229
230 ret = blkcg_activate_policy(q, &ioprio_policy);
231 if (ret) {
232 kfree(blkioprio_blkg);
233 return ret;
234 }
235
236 rqos = &blkioprio_blkg->rqos;
237 rqos->id = RQ_QOS_IOPRIO;
238 rqos->ops = &blkcg_ioprio_ops;
239 rqos->q = q;
240
241 /*
242 * Registering the rq-qos policy after activating the blk-cgroup
243 * policy guarantees that ioprio_blkcg_from_bio(bio) != NULL in the
244 * rq-qos callbacks.
245 */
246 rq_qos_add(q, rqos);
247
248 return 0;
249 }
250
ioprio_init(void)251 static int __init ioprio_init(void)
252 {
253 return blkcg_policy_register(&ioprio_policy);
254 }
255
ioprio_exit(void)256 static void __exit ioprio_exit(void)
257 {
258 blkcg_policy_unregister(&ioprio_policy);
259 }
260
261 module_init(ioprio_init);
262 module_exit(ioprio_exit);
263