]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/ioprio.c
block: make ioc get/put interface more conventional and fix race on alloction
[mirror_ubuntu-zesty-kernel.git] / fs / ioprio.c
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/blkdev.h>
27 #include <linux/capability.h>
28 #include <linux/syscalls.h>
29 #include <linux/security.h>
30 #include <linux/pid_namespace.h>
31
32 int set_task_ioprio(struct task_struct *task, int ioprio)
33 {
34 int err;
35 struct io_context *ioc;
36 const struct cred *cred = current_cred(), *tcred;
37
38 rcu_read_lock();
39 tcred = __task_cred(task);
40 if (tcred->uid != cred->euid &&
41 tcred->uid != cred->uid && !capable(CAP_SYS_NICE)) {
42 rcu_read_unlock();
43 return -EPERM;
44 }
45 rcu_read_unlock();
46
47 err = security_task_setioprio(task, ioprio);
48 if (err)
49 return err;
50
51 ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
52 if (ioc) {
53 ioc->ioprio = ioprio;
54 ioc->ioprio_changed = 1;
55 put_io_context(ioc);
56 }
57
58 return err;
59 }
60 EXPORT_SYMBOL_GPL(set_task_ioprio);
61
62 SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
63 {
64 int class = IOPRIO_PRIO_CLASS(ioprio);
65 int data = IOPRIO_PRIO_DATA(ioprio);
66 struct task_struct *p, *g;
67 struct user_struct *user;
68 struct pid *pgrp;
69 int ret;
70
71 switch (class) {
72 case IOPRIO_CLASS_RT:
73 if (!capable(CAP_SYS_ADMIN))
74 return -EPERM;
75 /* fall through, rt has prio field too */
76 case IOPRIO_CLASS_BE:
77 if (data >= IOPRIO_BE_NR || data < 0)
78 return -EINVAL;
79
80 break;
81 case IOPRIO_CLASS_IDLE:
82 break;
83 case IOPRIO_CLASS_NONE:
84 if (data)
85 return -EINVAL;
86 break;
87 default:
88 return -EINVAL;
89 }
90
91 ret = -ESRCH;
92 rcu_read_lock();
93 switch (which) {
94 case IOPRIO_WHO_PROCESS:
95 if (!who)
96 p = current;
97 else
98 p = find_task_by_vpid(who);
99 if (p)
100 ret = set_task_ioprio(p, ioprio);
101 break;
102 case IOPRIO_WHO_PGRP:
103 if (!who)
104 pgrp = task_pgrp(current);
105 else
106 pgrp = find_vpid(who);
107 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
108 ret = set_task_ioprio(p, ioprio);
109 if (ret)
110 break;
111 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
112 break;
113 case IOPRIO_WHO_USER:
114 if (!who)
115 user = current_user();
116 else
117 user = find_user(who);
118
119 if (!user)
120 break;
121
122 do_each_thread(g, p) {
123 if (__task_cred(p)->uid != who)
124 continue;
125 ret = set_task_ioprio(p, ioprio);
126 if (ret)
127 goto free_uid;
128 } while_each_thread(g, p);
129 free_uid:
130 if (who)
131 free_uid(user);
132 break;
133 default:
134 ret = -EINVAL;
135 }
136
137 rcu_read_unlock();
138 return ret;
139 }
140
141 static int get_task_ioprio(struct task_struct *p)
142 {
143 int ret;
144
145 ret = security_task_getioprio(p);
146 if (ret)
147 goto out;
148 ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
149 if (p->io_context)
150 ret = p->io_context->ioprio;
151 out:
152 return ret;
153 }
154
155 int ioprio_best(unsigned short aprio, unsigned short bprio)
156 {
157 unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
158 unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
159
160 if (aclass == IOPRIO_CLASS_NONE)
161 aclass = IOPRIO_CLASS_BE;
162 if (bclass == IOPRIO_CLASS_NONE)
163 bclass = IOPRIO_CLASS_BE;
164
165 if (aclass == bclass)
166 return min(aprio, bprio);
167 if (aclass > bclass)
168 return bprio;
169 else
170 return aprio;
171 }
172
173 SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
174 {
175 struct task_struct *g, *p;
176 struct user_struct *user;
177 struct pid *pgrp;
178 int ret = -ESRCH;
179 int tmpio;
180
181 rcu_read_lock();
182 switch (which) {
183 case IOPRIO_WHO_PROCESS:
184 if (!who)
185 p = current;
186 else
187 p = find_task_by_vpid(who);
188 if (p)
189 ret = get_task_ioprio(p);
190 break;
191 case IOPRIO_WHO_PGRP:
192 if (!who)
193 pgrp = task_pgrp(current);
194 else
195 pgrp = find_vpid(who);
196 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
197 tmpio = get_task_ioprio(p);
198 if (tmpio < 0)
199 continue;
200 if (ret == -ESRCH)
201 ret = tmpio;
202 else
203 ret = ioprio_best(ret, tmpio);
204 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
205 break;
206 case IOPRIO_WHO_USER:
207 if (!who)
208 user = current_user();
209 else
210 user = find_user(who);
211
212 if (!user)
213 break;
214
215 do_each_thread(g, p) {
216 if (__task_cred(p)->uid != user->uid)
217 continue;
218 tmpio = get_task_ioprio(p);
219 if (tmpio < 0)
220 continue;
221 if (ret == -ESRCH)
222 ret = tmpio;
223 else
224 ret = ioprio_best(ret, tmpio);
225 } while_each_thread(g, p);
226
227 if (who)
228 free_uid(user);
229 break;
230 default:
231 ret = -EINVAL;
232 }
233
234 rcu_read_unlock();
235 return ret;
236 }