Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6
[linux-2.6/openmoko-kernel/knife-kernel.git] / fs / ioprio.c
blobc4a1c3c65aac35ba999a4739fb3ed6549b5e1b75
1 /*
2 * fs/ioprio.c
4 * Copyright (C) 2004 Jens Axboe <axboe@kernel.dk>
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.
13 * IOW, setting BE scheduling class with prio 2 is done ala:
15 * unsigned int prio = (IOPRIO_CLASS_BE << IOPRIO_CLASS_SHIFT) | 2;
17 * ioprio_set(PRIO_PROCESS, pid, prio);
19 * See also Documentation/block/ioprio.txt
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>
30 static int set_task_ioprio(struct task_struct *task, int ioprio)
32 int err;
33 struct io_context *ioc;
35 if (task->uid != current->euid &&
36 task->uid != current->uid && !capable(CAP_SYS_NICE))
37 return -EPERM;
39 err = security_task_setioprio(task, ioprio);
40 if (err)
41 return err;
43 task_lock(task);
44 do {
45 ioc = task->io_context;
46 /* see wmb() in current_io_context() */
47 smp_read_barrier_depends();
48 if (ioc)
49 break;
51 ioc = alloc_io_context(GFP_ATOMIC, -1);
52 if (!ioc) {
53 err = -ENOMEM;
54 break;
56 task->io_context = ioc;
57 } while (1);
59 if (!err) {
60 ioc->ioprio = ioprio;
61 ioc->ioprio_changed = 1;
64 task_unlock(task);
65 return err;
68 asmlinkage long sys_ioprio_set(int which, int who, int ioprio)
70 int class = IOPRIO_PRIO_CLASS(ioprio);
71 int data = IOPRIO_PRIO_DATA(ioprio);
72 struct task_struct *p, *g;
73 struct user_struct *user;
74 struct pid *pgrp;
75 int ret;
77 switch (class) {
78 case IOPRIO_CLASS_RT:
79 if (!capable(CAP_SYS_ADMIN))
80 return -EPERM;
81 /* fall through, rt has prio field too */
82 case IOPRIO_CLASS_BE:
83 if (data >= IOPRIO_BE_NR || data < 0)
84 return -EINVAL;
86 break;
87 case IOPRIO_CLASS_IDLE:
88 break;
89 case IOPRIO_CLASS_NONE:
90 if (data)
91 return -EINVAL;
92 break;
93 default:
94 return -EINVAL;
97 ret = -ESRCH;
99 * We want IOPRIO_WHO_PGRP/IOPRIO_WHO_USER to be "atomic",
100 * so we can't use rcu_read_lock(). See re-copy of ->ioprio
101 * in copy_process().
103 read_lock(&tasklist_lock);
104 switch (which) {
105 case IOPRIO_WHO_PROCESS:
106 if (!who)
107 p = current;
108 else
109 p = find_task_by_vpid(who);
110 if (p)
111 ret = set_task_ioprio(p, ioprio);
112 break;
113 case IOPRIO_WHO_PGRP:
114 if (!who)
115 pgrp = task_pgrp(current);
116 else
117 pgrp = find_vpid(who);
118 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
119 ret = set_task_ioprio(p, ioprio);
120 if (ret)
121 break;
122 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
123 break;
124 case IOPRIO_WHO_USER:
125 if (!who)
126 user = current->user;
127 else
128 user = find_user(who);
130 if (!user)
131 break;
133 do_each_thread(g, p) {
134 if (p->uid != who)
135 continue;
136 ret = set_task_ioprio(p, ioprio);
137 if (ret)
138 goto free_uid;
139 } while_each_thread(g, p);
140 free_uid:
141 if (who)
142 free_uid(user);
143 break;
144 default:
145 ret = -EINVAL;
148 read_unlock(&tasklist_lock);
149 return ret;
152 static int get_task_ioprio(struct task_struct *p)
154 int ret;
156 ret = security_task_getioprio(p);
157 if (ret)
158 goto out;
159 ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
160 if (p->io_context)
161 ret = p->io_context->ioprio;
162 out:
163 return ret;
166 int ioprio_best(unsigned short aprio, unsigned short bprio)
168 unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
169 unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
171 if (aclass == IOPRIO_CLASS_NONE)
172 aclass = IOPRIO_CLASS_BE;
173 if (bclass == IOPRIO_CLASS_NONE)
174 bclass = IOPRIO_CLASS_BE;
176 if (aclass == bclass)
177 return min(aprio, bprio);
178 if (aclass > bclass)
179 return bprio;
180 else
181 return aprio;
184 asmlinkage long sys_ioprio_get(int which, int who)
186 struct task_struct *g, *p;
187 struct user_struct *user;
188 struct pid *pgrp;
189 int ret = -ESRCH;
190 int tmpio;
192 read_lock(&tasklist_lock);
193 switch (which) {
194 case IOPRIO_WHO_PROCESS:
195 if (!who)
196 p = current;
197 else
198 p = find_task_by_vpid(who);
199 if (p)
200 ret = get_task_ioprio(p);
201 break;
202 case IOPRIO_WHO_PGRP:
203 if (!who)
204 pgrp = task_pgrp(current);
205 else
206 pgrp = find_vpid(who);
207 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
208 tmpio = get_task_ioprio(p);
209 if (tmpio < 0)
210 continue;
211 if (ret == -ESRCH)
212 ret = tmpio;
213 else
214 ret = ioprio_best(ret, tmpio);
215 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
216 break;
217 case IOPRIO_WHO_USER:
218 if (!who)
219 user = current->user;
220 else
221 user = find_user(who);
223 if (!user)
224 break;
226 do_each_thread(g, p) {
227 if (p->uid != user->uid)
228 continue;
229 tmpio = get_task_ioprio(p);
230 if (tmpio < 0)
231 continue;
232 if (ret == -ESRCH)
233 ret = tmpio;
234 else
235 ret = ioprio_best(ret, tmpio);
236 } while_each_thread(g, p);
238 if (who)
239 free_uid(user);
240 break;
241 default:
242 ret = -EINVAL;
245 read_unlock(&tasklist_lock);
246 return ret;