Commit | Line | Data |
---|---|---|
556910e3 BVA |
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 | ||
556910e3 BVA |
15 | #include <linux/blk-mq.h> |
16 | #include <linux/blk_types.h> | |
17 | #include <linux/kernel.h> | |
18 | #include <linux/module.h> | |
672fdcf0 | 19 | #include "blk-cgroup.h" |
556910e3 BVA |
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 | ||
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 | ||
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 * | |
79 | ioprio_blkcg_from_css(struct cgroup_subsys_state *css) | |
80 | { | |
81 | return blkcg_to_ioprio_blkcg(css_to_blkcg(css)); | |
82 | } | |
83 | ||
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 | ||
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 | ||
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; | |
556910e3 BVA |
115 | return nbytes; |
116 | } | |
117 | ||
118 | static struct blkg_policy_data * | |
119 | ioprio_alloc_pd(gfp_t gfp, struct request_queue *q, struct blkcg *blkcg) | |
120 | { | |
121 | struct ioprio_blkg *ioprio_blkg; | |
122 | ||
123 | ioprio_blkg = kzalloc(sizeof(*ioprio_blkg), gfp); | |
124 | if (!ioprio_blkg) | |
125 | return NULL; | |
126 | ||
127 | return &ioprio_blkg->pd; | |
128 | } | |
129 | ||
130 | static void ioprio_free_pd(struct blkg_policy_data *pd) | |
131 | { | |
132 | struct ioprio_blkg *ioprio_blkg = pd_to_ioprio(pd); | |
133 | ||
134 | kfree(ioprio_blkg); | |
135 | } | |
136 | ||
137 | static struct blkcg_policy_data *ioprio_alloc_cpd(gfp_t gfp) | |
138 | { | |
139 | struct ioprio_blkcg *blkcg; | |
140 | ||
141 | blkcg = kzalloc(sizeof(*blkcg), gfp); | |
142 | if (!blkcg) | |
143 | return NULL; | |
144 | blkcg->prio_policy = POLICY_NO_CHANGE; | |
145 | return &blkcg->cpd; | |
146 | } | |
147 | ||
148 | static void ioprio_free_cpd(struct blkcg_policy_data *cpd) | |
149 | { | |
150 | struct ioprio_blkcg *blkcg = container_of(cpd, typeof(*blkcg), cpd); | |
151 | ||
152 | kfree(blkcg); | |
153 | } | |
154 | ||
155 | #define IOPRIO_ATTRS \ | |
156 | { \ | |
157 | .name = "prio.class", \ | |
158 | .seq_show = ioprio_show_prio_policy, \ | |
159 | .write = ioprio_set_prio_policy, \ | |
160 | }, \ | |
161 | { } /* sentinel */ | |
162 | ||
163 | /* cgroup v2 attributes */ | |
164 | static struct cftype ioprio_files[] = { | |
165 | IOPRIO_ATTRS | |
166 | }; | |
167 | ||
168 | /* cgroup v1 attributes */ | |
169 | static struct cftype ioprio_legacy_files[] = { | |
170 | IOPRIO_ATTRS | |
171 | }; | |
172 | ||
173 | static struct blkcg_policy ioprio_policy = { | |
174 | .dfl_cftypes = ioprio_files, | |
175 | .legacy_cftypes = ioprio_legacy_files, | |
176 | ||
177 | .cpd_alloc_fn = ioprio_alloc_cpd, | |
178 | .cpd_free_fn = ioprio_free_cpd, | |
179 | ||
180 | .pd_alloc_fn = ioprio_alloc_pd, | |
181 | .pd_free_fn = ioprio_free_pd, | |
182 | }; | |
183 | ||
82b74cac | 184 | void blkcg_set_ioprio(struct bio *bio) |
556910e3 BVA |
185 | { |
186 | struct ioprio_blkcg *blkcg = ioprio_blkcg_from_bio(bio); | |
25c4b5e0 JA |
187 | u16 prio; |
188 | ||
82b74cac | 189 | if (!blkcg || blkcg->prio_policy == POLICY_NO_CHANGE) |
25c4b5e0 | 190 | return; |
556910e3 BVA |
191 | |
192 | /* | |
193 | * Except for IOPRIO_CLASS_NONE, higher I/O priority numbers | |
194 | * correspond to a lower priority. Hence, the max_t() below selects | |
195 | * the lower priority of bi_ioprio and the cgroup I/O priority class. | |
f2586544 JK |
196 | * If the bio I/O priority equals IOPRIO_CLASS_NONE, the cgroup I/O |
197 | * priority is assigned to the bio. | |
556910e3 | 198 | */ |
25c4b5e0 JA |
199 | prio = max_t(u16, bio->bi_ioprio, |
200 | IOPRIO_PRIO_VALUE(blkcg->prio_policy, 0)); | |
201 | if (prio > bio->bi_ioprio) | |
202 | bio->bi_ioprio = prio; | |
556910e3 BVA |
203 | } |
204 | ||
b0dde3f5 | 205 | void blk_ioprio_exit(struct gendisk *disk) |
556910e3 | 206 | { |
b0dde3f5 | 207 | blkcg_deactivate_policy(disk->queue, &ioprio_policy); |
556910e3 BVA |
208 | } |
209 | ||
b0dde3f5 | 210 | int blk_ioprio_init(struct gendisk *disk) |
556910e3 | 211 | { |
b0dde3f5 | 212 | return blkcg_activate_policy(disk->queue, &ioprio_policy); |
556910e3 BVA |
213 | } |
214 | ||
215 | static int __init ioprio_init(void) | |
216 | { | |
217 | return blkcg_policy_register(&ioprio_policy); | |
218 | } | |
219 | ||
220 | static void __exit ioprio_exit(void) | |
221 | { | |
222 | blkcg_policy_unregister(&ioprio_policy); | |
223 | } | |
224 | ||
225 | module_init(ioprio_init); | |
226 | module_exit(ioprio_exit); |