1IOPRIO_SET(2) Linux Programmer's Manual IOPRIO_SET(2)
2
3
4
6 ioprio_get, ioprio_set - get/set I/O scheduling class and priority
7
9 int ioprio_get(int which, int who);
10 int ioprio_set(int which, int who, int ioprio);
11
13 The ioprio_get() and ioprio_set() system calls respectively get and set
14 the I/O scheduling class and priority of one or more processes.
15
16 The which and who arguments identify the process(es) on which the sys‐
17 tem calls operate. The which argument determines how who is inter‐
18 preted, and has one of the following values:
19
20 IOPRIO_WHO_PROCESS
21 who is a process ID identifying a single process.
22
23 IOPRIO_WHO_PGRP
24 who is a process group ID identifying all the members of a
25 process group.
26
27 IOPRIO_WHO_USER
28 who is a user ID identifying all of the processes that have a
29 matching real UID.
30
31 If which is specified as IOPRIO_WHO_PGRP or IOPRIO_WHO_USER when call‐
32 ing ioprio_get(), and more than one process matches who, then the
33 returned priority will be the highest one found among all of the match‐
34 ing processes. One priority is said to be higher than another one if
35 it belongs to a higher priority class (IOPRIO_CLASS_RT is the highest
36 priority class; IOPRIO_CLASS_IDLE is the lowest) or if it belongs to
37 the same priority class as the other process but has a higher priority
38 level (a lower priority number means a higher priority level).
39
40 The ioprio argument given to ioprio_set() is a bit mask that specifies
41 both the scheduling class and the priority to be assigned to the target
42 process(es). The following macros are used for assembling and dissect‐
43 ing ioprio values:
44
45 IOPRIO_PRIO_VALUE(class, data)
46 Given a scheduling class and priority (data), this macro com‐
47 bines the two values to produce an ioprio value, which is
48 returned as the result of the macro.
49
50 IOPRIO_PRIO_CLASS(mask)
51 Given mask (an ioprio value), this macro returns its I/O class
52 component, that is, one of the values IOPRIO_CLASS_RT,
53 IOPRIO_CLASS_BE, or IOPRIO_CLASS_IDLE.
54
55 IOPRIO_PRIO_DATA(mask)
56 Given mask (an ioprio value), this macro returns its priority
57 (data) component.
58
59 See the NOTES section for more information on scheduling classes and
60 priorities.
61
62 I/O priorities are supported for reads and for synchronous (O_DIRECT,
63 O_SYNC) writes. I/O priorities are not supported for asynchronous
64 writes because they are issued outside the context of the program
65 dirtying the memory, and thus program-specific priorities do not apply.
66
68 On success, ioprio_get() returns the ioprio value of the process with
69 highest I/O priority of any of the processes that match the criteria
70 specified in which and who. On error, -1 is returned, and errno is set
71 to indicate the error.
72
73 On success, ioprio_set() returns 0. On error, -1 is returned, and
74 errno is set to indicate the error.
75
77 EINVAL Invalid value for which or ioprio. Refer to the NOTES section
78 for available scheduler classes and priority levels for ioprio.
79
80 EPERM The calling process does not have the privilege needed to assign
81 this ioprio to the specified process(es). See the NOTES section
82 for more information on required privileges for ioprio_set().
83
84 ESRCH No process(es) could be found that matched the specification in
85 which and who.
86
88 These system calls have been available on Linux since kernel 2.6.13.
89
91 These system calls are Linux-specific.
92
94 Glibc does not provide wrapper for these system calls; call them using
95 syscall(2).
96
97 These system calls only have an effect when used in conjunction with an
98 I/O scheduler that supports I/O priorities. As at kernel 2.6.17 the
99 only such scheduler is the Completely Fair Queuing (CFQ) I/O scheduler.
100
101 Selecting an I/O Scheduler
102 I/O Schedulers are selected on a per-device basis via the special file
103 /sys/block/<device>/queue/scheduler.
104
105 One can view the current I/O scheduler via the /sys file system. For
106 example, the following command displays a list of all schedulers cur‐
107 rently loaded in the kernel:
108
109 $ cat /sys/block/hda/queue/scheduler
110 noop anticipatory deadline [cfq]
111
112 The scheduler surrounded by brackets is the one actually in use for the
113 device (hda in the example). Setting another scheduler is done by
114 writing the name of the new scheduler to this file. For example, the
115 following command will set the scheduler for the hda device to cfq:
116
117 $ su
118 Password:
119 # echo cfq > /sys/block/hda/queue/scheduler
120
121 The Completely Fair Queuing (CFQ) I/O Scheduler
122 Since v3 (aka CFQ Time Sliced) CFQ implements I/O nice levels similar
123 to those of CPU scheduling. These nice levels are grouped in three
124 scheduling classes each one containing one or more priority levels:
125
126 IOPRIO_CLASS_RT (1)
127 This is the real-time I/O class. This scheduling class is given
128 higher priority than any other class: processes from this class
129 are given first access to the disk every time. Thus this I/O
130 class needs to be used with some care: one I/O real-time process
131 can starve the entire system. Within the real-time class, there
132 are 8 levels of class data (priority) that determine exactly how
133 much time this process needs the disk for on each service. The
134 highest real-time priority level is 0; the lowest is 7. In the
135 future this might change to be more directly mappable to perfor‐
136 mance, by passing in a desired data rate instead.
137
138 IOPRIO_CLASS_BE (2)
139 This is the best-effort scheduling class, which is the default
140 for any process that hasn't set a specific I/O priority. The
141 class data (priority) determines how much I/O bandwidth the
142 process will get. Best-effort priority levels are analogous to
143 CPU nice values (see getpriority(2)). The priority level deter‐
144 mines a priority relative to other processes in the best-effort
145 scheduling class. Priority levels range from 0 (highest) to 7
146 (lowest).
147
148 IOPRIO_CLASS_IDLE (3)
149 This is the idle scheduling class. Processes running at this
150 level only get I/O time when no-one else needs the disk. The
151 idle class has no class data. Attention is required when
152 assigning this priority class to a process, since it may become
153 starved if higher priority processes are constantly accessing
154 the disk.
155
156 Refer to Documentation/block/ioprio.txt for more information on the CFQ
157 I/O Scheduler and an example program.
158
159 Required permissions to set I/O priorities
160 Permission to change a process's priority is granted or denied based on
161 two assertions:
162
163 Process ownership
164 An unprivileged process may only set the I/O priority of a
165 process whose real UID matches the real or effective UID of the
166 calling process. A process which has the CAP_SYS_NICE capabil‐
167 ity can change the priority of any process.
168
169 What is the desired priority
170 Attempts to set very high priorities (IOPRIO_CLASS_RT) require
171 the CAP_SYS_ADMIN capability. Kernel versions up to 2.6.24 also
172 required CAP_SYS_ADMIN to set a very low priority
173 (IOPRIO_CLASS_IDLE), but since Linux 2.6.25, this is no longer
174 required.
175
176 A call to ioprio_set() must follow both rules, or the call will fail
177 with the error EPERM.
178
180 Glibc does not yet provide a suitable header file defining the function
181 prototypes and macros described on this page. Suitable definitions can
182 be found in linux/ioprio.h.
183
185 getpriority(2), open(2), capabilities(7)
186
187 Documentation/block/ioprio.txt in the kernel source tree.
188
190 This page is part of release 3.25 of the Linux man-pages project. A
191 description of the project, and information about reporting bugs, can
192 be found at http://www.kernel.org/doc/man-pages/.
193
194
195
196Linux 2008-07-09 IOPRIO_SET(2)