1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * fs/ioprio.c
4 *
5 * Copyright (C) 2004 Jens Axboe <axboe@kernel.dk>
6 *
7 * Helper functions for setting/querying io priorities of processes. The
8 * system calls closely mimmick getpriority/setpriority, see the man page for
9 * those. The prio argument is a composite of prio class and prio data, where
10 * the data argument has meaning within that class. The standard scheduling
11 * classes have 8 distinct prio levels, with 0 being the highest prio and 7
12 * being the lowest.
13 *
14 * IOW, setting BE scheduling class with prio 2 is done ala:
15 *
16 * unsigned int prio = (IOPRIO_CLASS_BE << IOPRIO_CLASS_SHIFT) | 2;
17 *
18 * ioprio_set(PRIO_PROCESS, pid, prio);
19 *
20 * See also Documentation/block/ioprio.rst
21 *
22 */
23 #include <linux/gfp.h>
24 #include <linux/kernel.h>
25 #include <linux/export.h>
26 #include <linux/ioprio.h>
27 #include <linux/cred.h>
28 #include <linux/blkdev.h>
29 #include <linux/capability.h>
30 #include <linux/sched/user.h>
31 #include <linux/sched/task.h>
32 #include <linux/syscalls.h>
33 #include <linux/security.h>
34 #include <linux/pid_namespace.h>
35
set_task_ioprio(struct task_struct * task,int ioprio)36 int set_task_ioprio(struct task_struct *task, int ioprio)
37 {
38 int err;
39 struct io_context *ioc;
40 const struct cred *cred = current_cred(), *tcred;
41
42 rcu_read_lock();
43 tcred = __task_cred(task);
44 if (!uid_eq(tcred->uid, cred->euid) &&
45 !uid_eq(tcred->uid, cred->uid) && !capable(CAP_SYS_NICE)) {
46 rcu_read_unlock();
47 return -EPERM;
48 }
49 rcu_read_unlock();
50
51 err = security_task_setioprio(task, ioprio);
52 if (err)
53 return err;
54
55 ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
56 if (ioc) {
57 ioc->ioprio = ioprio;
58 put_io_context(ioc);
59 }
60
61 return err;
62 }
63 EXPORT_SYMBOL_GPL(set_task_ioprio);
64
ioprio_check_cap(int ioprio)65 int ioprio_check_cap(int ioprio)
66 {
67 int class = IOPRIO_PRIO_CLASS(ioprio);
68 int data = IOPRIO_PRIO_DATA(ioprio);
69
70 switch (class) {
71 case IOPRIO_CLASS_RT:
72 if (!capable(CAP_SYS_NICE) && !capable(CAP_SYS_ADMIN))
73 return -EPERM;
74 fallthrough;
75 /* rt has prio field too */
76 case IOPRIO_CLASS_BE:
77 if (data >= IOPRIO_NR_LEVELS || data < 0)
78 return -EINVAL;
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
SYSCALL_DEFINE3(ioprio_set,int,which,int,who,int,ioprio)93 SYSCALL_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
122 read_lock(&tasklist_lock);
123 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
124 ret = set_task_ioprio(p, ioprio);
125 if (ret) {
126 read_unlock(&tasklist_lock);
127 goto out;
128 }
129 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
130 read_unlock(&tasklist_lock);
131
132 break;
133 case IOPRIO_WHO_USER:
134 uid = make_kuid(current_user_ns(), who);
135 if (!uid_valid(uid))
136 break;
137 if (!who)
138 user = current_user();
139 else
140 user = find_user(uid);
141
142 if (!user)
143 break;
144
145 for_each_process_thread(g, p) {
146 if (!uid_eq(task_uid(p), uid) ||
147 !task_pid_vnr(p))
148 continue;
149 ret = set_task_ioprio(p, ioprio);
150 if (ret)
151 goto free_uid;
152 }
153 free_uid:
154 if (who)
155 free_uid(user);
156 break;
157 default:
158 ret = -EINVAL;
159 }
160
161 out:
162 rcu_read_unlock();
163 return ret;
164 }
165
get_task_ioprio(struct task_struct * p)166 static int get_task_ioprio(struct task_struct *p)
167 {
168 int ret;
169
170 ret = security_task_getioprio(p);
171 if (ret)
172 goto out;
173 ret = IOPRIO_DEFAULT;
174 task_lock(p);
175 if (p->io_context)
176 ret = p->io_context->ioprio;
177 task_unlock(p);
178 out:
179 return ret;
180 }
181
ioprio_best(unsigned short aprio,unsigned short bprio)182 int ioprio_best(unsigned short aprio, unsigned short bprio)
183 {
184 if (!ioprio_valid(aprio))
185 aprio = IOPRIO_DEFAULT;
186 if (!ioprio_valid(bprio))
187 bprio = IOPRIO_DEFAULT;
188
189 return min(aprio, bprio);
190 }
191
SYSCALL_DEFINE2(ioprio_get,int,which,int,who)192 SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
193 {
194 struct task_struct *g, *p;
195 struct user_struct *user;
196 struct pid *pgrp;
197 kuid_t uid;
198 int ret = -ESRCH;
199 int tmpio;
200
201 rcu_read_lock();
202 switch (which) {
203 case IOPRIO_WHO_PROCESS:
204 if (!who)
205 p = current;
206 else
207 p = find_task_by_vpid(who);
208 if (p)
209 ret = get_task_ioprio(p);
210 break;
211 case IOPRIO_WHO_PGRP:
212 if (!who)
213 pgrp = task_pgrp(current);
214 else
215 pgrp = find_vpid(who);
216 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
217 tmpio = get_task_ioprio(p);
218 if (tmpio < 0)
219 continue;
220 if (ret == -ESRCH)
221 ret = tmpio;
222 else
223 ret = ioprio_best(ret, tmpio);
224 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
225 break;
226 case IOPRIO_WHO_USER:
227 uid = make_kuid(current_user_ns(), who);
228 if (!who)
229 user = current_user();
230 else
231 user = find_user(uid);
232
233 if (!user)
234 break;
235
236 for_each_process_thread(g, p) {
237 if (!uid_eq(task_uid(p), user->uid) ||
238 !task_pid_vnr(p))
239 continue;
240 tmpio = get_task_ioprio(p);
241 if (tmpio < 0)
242 continue;
243 if (ret == -ESRCH)
244 ret = tmpio;
245 else
246 ret = ioprio_best(ret, tmpio);
247 }
248
249 if (who)
250 free_uid(user);
251 break;
252 default:
253 ret = -EINVAL;
254 }
255
256 rcu_read_unlock();
257 return ret;
258 }
259
