1/*
2 * fs/ioprio.c
3 *
4 * Copyright (C) 2004 Jens Axboe <axboe@kernel.dk>
5 *
6 * Helper functions for setting/querying io priorities of processes. The
7 * system calls closely mimmick getpriority/setpriority, see the man page for
8 * those. The prio argument is a composite of prio class and prio data, where
9 * the data argument has meaning within that class. The standard scheduling
10 * classes have 8 distinct prio levels, with 0 being the highest prio and 7
11 * being the lowest.
12 *
13 * IOW, setting BE scheduling class with prio 2 is done ala:
14 *
15 * unsigned int prio = (IOPRIO_CLASS_BE << IOPRIO_CLASS_SHIFT) | 2;
16 *
17 * ioprio_set(PRIO_PROCESS, pid, prio);
18 *
19 * See also Documentation/block/ioprio.txt
20 *
21 */
22#include <linux/gfp.h>
23#include <linux/kernel.h>
24#include <linux/export.h>
25#include <linux/ioprio.h>
26#include <linux/cred.h>
27#include <linux/blkdev.h>
28#include <linux/capability.h>
29#include <linux/sched/user.h>
30#include <linux/sched/task.h>
31#include <linux/syscalls.h>
32#include <linux/security.h>
33#include <linux/pid_namespace.h>
34
35int set_task_ioprio(struct task_struct *task, int ioprio)
36{
37 int err;
38 struct io_context *ioc;
39 const struct cred *cred = current_cred(), *tcred;
40
41 rcu_read_lock();
42 tcred = __task_cred(task);
43 if (!uid_eq(tcred->uid, cred->euid) &&
44 !uid_eq(tcred->uid, cred->uid) && !capable(CAP_SYS_NICE)) {
45 rcu_read_unlock();
46 return -EPERM;
47 }
48 rcu_read_unlock();
49
50 err = security_task_setioprio(task, ioprio);
51 if (err)
52 return err;
53
54 ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
55 if (ioc) {
56 ioc->ioprio = ioprio;
57 put_io_context(ioc);
58 }
59
60 return err;
61}
62EXPORT_SYMBOL_GPL(set_task_ioprio);
63
64int ioprio_check_cap(int ioprio)
65{
66 int class = IOPRIO_PRIO_CLASS(ioprio);
67 int data = IOPRIO_PRIO_DATA(ioprio);
68
69 switch (class) {
70 case IOPRIO_CLASS_RT:
71 if (!capable(CAP_SYS_ADMIN))
72 return -EPERM;
73 /* fall through */
74 /* rt has prio field too */
75 case IOPRIO_CLASS_BE:
76 if (data >= IOPRIO_BE_NR || data < 0)
77 return -EINVAL;
78
79 break;
80 case IOPRIO_CLASS_IDLE:
81 break;
82 case IOPRIO_CLASS_NONE:
83 if (data)
84 return -EINVAL;
85 break;
86 default:
87 return -EINVAL;
88 }
89
90 return 0;
91}
92
93SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
94{
95 struct task_struct *p, *g;
96 struct user_struct *user;
97 struct pid *pgrp;
98 kuid_t uid;
99 int ret;
100
101 ret = ioprio_check_cap(ioprio);
102 if (ret)
103 return ret;
104
105 ret = -ESRCH;
106 rcu_read_lock();
107 switch (which) {
108 case IOPRIO_WHO_PROCESS:
109 if (!who)
110 p = current;
111 else
112 p = find_task_by_vpid(who);
113 if (p)
114 ret = set_task_ioprio(p, ioprio);
115 break;
116 case IOPRIO_WHO_PGRP:
117 if (!who)
118 pgrp = task_pgrp(current);
119 else
120 pgrp = find_vpid(who);
121 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
122 ret = set_task_ioprio(p, ioprio);
123 if (ret)
124 break;
125 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
126 break;
127 case IOPRIO_WHO_USER:
128 uid = make_kuid(current_user_ns(), who);
129 if (!uid_valid(uid))
130 break;
131 if (!who)
132 user = current_user();
133 else
134 user = find_user(uid);
135
136 if (!user)
137 break;
138
139 for_each_process_thread(g, p) {
140 if (!uid_eq(task_uid(p), uid) ||
141 !task_pid_vnr(p))
142 continue;
143 ret = set_task_ioprio(p, ioprio);
144 if (ret)
145 goto free_uid;
146 }
147free_uid:
148 if (who)
149 free_uid(user);
150 break;
151 default:
152 ret = -EINVAL;
153 }
154
155 rcu_read_unlock();
156 return ret;
157}
158
159static int get_task_ioprio(struct task_struct *p)
160{
161 int ret;
162
163 ret = security_task_getioprio(p);
164 if (ret)
165 goto out;
166 ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
167 task_lock(p);
168 if (p->io_context)
169 ret = p->io_context->ioprio;
170 task_unlock(p);
171out:
172 return ret;
173}
174
175int ioprio_best(unsigned short aprio, unsigned short bprio)
176{
177 if (!ioprio_valid(aprio))
178 aprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
179 if (!ioprio_valid(bprio))
180 bprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
181
182 return min(aprio, bprio);
183}
184
185SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
186{
187 struct task_struct *g, *p;
188 struct user_struct *user;
189 struct pid *pgrp;
190 kuid_t uid;
191 int ret = -ESRCH;
192 int tmpio;
193
194 rcu_read_lock();
195 switch (which) {
196 case IOPRIO_WHO_PROCESS:
197 if (!who)
198 p = current;
199 else
200 p = find_task_by_vpid(who);
201 if (p)
202 ret = get_task_ioprio(p);
203 break;
204 case IOPRIO_WHO_PGRP:
205 if (!who)
206 pgrp = task_pgrp(current);
207 else
208 pgrp = find_vpid(who);
209 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
210 tmpio = get_task_ioprio(p);
211 if (tmpio < 0)
212 continue;
213 if (ret == -ESRCH)
214 ret = tmpio;
215 else
216 ret = ioprio_best(ret, tmpio);
217 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
218 break;
219 case IOPRIO_WHO_USER:
220 uid = make_kuid(current_user_ns(), who);
221 if (!who)
222 user = current_user();
223 else
224 user = find_user(uid);
225
226 if (!user)
227 break;
228
229 for_each_process_thread(g, p) {
230 if (!uid_eq(task_uid(p), user->uid) ||
231 !task_pid_vnr(p))
232 continue;
233 tmpio = get_task_ioprio(p);
234 if (tmpio < 0)
235 continue;
236 if (ret == -ESRCH)
237 ret = tmpio;
238 else
239 ret = ioprio_best(ret, tmpio);
240 }
241
242 if (who)
243 free_uid(user);
244 break;
245 default:
246 ret = -EINVAL;
247 }
248
249 rcu_read_unlock();
250 return ret;
251}
252