]>
Commit | Line | Data |
---|---|---|
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/kernel.h> | |
23 | #include <linux/ioprio.h> | |
24 | #include <linux/blkdev.h> | |
25 | #include <linux/capability.h> | |
26 | #include <linux/syscalls.h> | |
27 | #include <linux/security.h> | |
28 | #include <linux/pid_namespace.h> | |
29 | ||
30 | static int set_task_ioprio(struct task_struct *task, int ioprio) | |
31 | { | |
32 | int err; | |
33 | struct io_context *ioc; | |
34 | ||
35 | if (task->uid != current->euid && | |
36 | task->uid != current->uid && !capable(CAP_SYS_NICE)) | |
37 | return -EPERM; | |
38 | ||
39 | err = security_task_setioprio(task, ioprio); | |
40 | if (err) | |
41 | return err; | |
42 | ||
43 | task_lock(task); | |
44 | ||
45 | task->ioprio = ioprio; | |
46 | ||
47 | ioc = task->io_context; | |
48 | /* see wmb() in current_io_context() */ | |
49 | smp_read_barrier_depends(); | |
50 | ||
51 | if (ioc) | |
52 | ioc->ioprio_changed = 1; | |
53 | ||
54 | task_unlock(task); | |
55 | return 0; | |
56 | } | |
57 | ||
58 | asmlinkage long sys_ioprio_set(int which, int who, int ioprio) | |
59 | { | |
60 | int class = IOPRIO_PRIO_CLASS(ioprio); | |
61 | int data = IOPRIO_PRIO_DATA(ioprio); | |
62 | struct task_struct *p, *g; | |
63 | struct user_struct *user; | |
64 | struct pid *pgrp; | |
65 | int ret; | |
66 | ||
67 | switch (class) { | |
68 | case IOPRIO_CLASS_RT: | |
69 | if (!capable(CAP_SYS_ADMIN)) | |
70 | return -EPERM; | |
71 | /* fall through, rt has prio field too */ | |
72 | case IOPRIO_CLASS_BE: | |
73 | if (data >= IOPRIO_BE_NR || data < 0) | |
74 | return -EINVAL; | |
75 | ||
76 | break; | |
77 | case IOPRIO_CLASS_IDLE: | |
78 | if (!capable(CAP_SYS_ADMIN)) | |
79 | return -EPERM; | |
80 | break; | |
81 | case IOPRIO_CLASS_NONE: | |
82 | if (data) | |
83 | return -EINVAL; | |
84 | break; | |
85 | default: | |
86 | return -EINVAL; | |
87 | } | |
88 | ||
89 | ret = -ESRCH; | |
90 | /* | |
91 | * We want IOPRIO_WHO_PGRP/IOPRIO_WHO_USER to be "atomic", | |
92 | * so we can't use rcu_read_lock(). See re-copy of ->ioprio | |
93 | * in copy_process(). | |
94 | */ | |
95 | read_lock(&tasklist_lock); | |
96 | switch (which) { | |
97 | case IOPRIO_WHO_PROCESS: | |
98 | if (!who) | |
99 | p = current; | |
100 | else | |
101 | p = find_task_by_vpid(who); | |
102 | if (p) | |
103 | ret = set_task_ioprio(p, ioprio); | |
104 | break; | |
105 | case IOPRIO_WHO_PGRP: | |
106 | if (!who) | |
107 | pgrp = task_pgrp(current); | |
108 | else | |
109 | pgrp = find_vpid(who); | |
110 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) { | |
111 | ret = set_task_ioprio(p, ioprio); | |
112 | if (ret) | |
113 | break; | |
114 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p); | |
115 | break; | |
116 | case IOPRIO_WHO_USER: | |
117 | if (!who) | |
118 | user = current->user; | |
119 | else | |
120 | user = find_user(who); | |
121 | ||
122 | if (!user) | |
123 | break; | |
124 | ||
125 | do_each_thread(g, p) { | |
126 | if (p->uid != who) | |
127 | continue; | |
128 | ret = set_task_ioprio(p, ioprio); | |
129 | if (ret) | |
130 | goto free_uid; | |
131 | } while_each_thread(g, p); | |
132 | free_uid: | |
133 | if (who) | |
134 | free_uid(user); | |
135 | break; | |
136 | default: | |
137 | ret = -EINVAL; | |
138 | } | |
139 | ||
140 | read_unlock(&tasklist_lock); | |
141 | return ret; | |
142 | } | |
143 | ||
144 | static int get_task_ioprio(struct task_struct *p) | |
145 | { | |
146 | int ret; | |
147 | ||
148 | ret = security_task_getioprio(p); | |
149 | if (ret) | |
150 | goto out; | |
151 | ret = p->ioprio; | |
152 | out: | |
153 | return ret; | |
154 | } | |
155 | ||
156 | int ioprio_best(unsigned short aprio, unsigned short bprio) | |
157 | { | |
158 | unsigned short aclass = IOPRIO_PRIO_CLASS(aprio); | |
159 | unsigned short bclass = IOPRIO_PRIO_CLASS(bprio); | |
160 | ||
161 | if (aclass == IOPRIO_CLASS_NONE) | |
162 | aclass = IOPRIO_CLASS_BE; | |
163 | if (bclass == IOPRIO_CLASS_NONE) | |
164 | bclass = IOPRIO_CLASS_BE; | |
165 | ||
166 | if (aclass == bclass) | |
167 | return min(aprio, bprio); | |
168 | if (aclass > bclass) | |
169 | return bprio; | |
170 | else | |
171 | return aprio; | |
172 | } | |
173 | ||
174 | asmlinkage long sys_ioprio_get(int which, int who) | |
175 | { | |
176 | struct task_struct *g, *p; | |
177 | struct user_struct *user; | |
178 | struct pid *pgrp; | |
179 | int ret = -ESRCH; | |
180 | int tmpio; | |
181 | ||
182 | read_lock(&tasklist_lock); | |
183 | switch (which) { | |
184 | case IOPRIO_WHO_PROCESS: | |
185 | if (!who) | |
186 | p = current; | |
187 | else | |
188 | p = find_task_by_vpid(who); | |
189 | if (p) | |
190 | ret = get_task_ioprio(p); | |
191 | break; | |
192 | case IOPRIO_WHO_PGRP: | |
193 | if (!who) | |
194 | pgrp = task_pgrp(current); | |
195 | else | |
196 | pgrp = find_vpid(who); | |
197 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) { | |
198 | tmpio = get_task_ioprio(p); | |
199 | if (tmpio < 0) | |
200 | continue; | |
201 | if (ret == -ESRCH) | |
202 | ret = tmpio; | |
203 | else | |
204 | ret = ioprio_best(ret, tmpio); | |
205 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p); | |
206 | break; | |
207 | case IOPRIO_WHO_USER: | |
208 | if (!who) | |
209 | user = current->user; | |
210 | else | |
211 | user = find_user(who); | |
212 | ||
213 | if (!user) | |
214 | break; | |
215 | ||
216 | do_each_thread(g, p) { | |
217 | if (p->uid != user->uid) | |
218 | continue; | |
219 | tmpio = get_task_ioprio(p); | |
220 | if (tmpio < 0) | |
221 | continue; | |
222 | if (ret == -ESRCH) | |
223 | ret = tmpio; | |
224 | else | |
225 | ret = ioprio_best(ret, tmpio); | |
226 | } while_each_thread(g, p); | |
227 | ||
228 | if (who) | |
229 | free_uid(user); | |
230 | break; | |
231 | default: | |
232 | ret = -EINVAL; | |
233 | } | |
234 | ||
235 | read_unlock(&tasklist_lock); | |
236 | return ret; | |
237 | } | |
238 |