1 Per-task statistics interface
2 -----------------------------
5 Taskstats is a netlink-based interface for sending per-task and
6 per-process statistics from the kernel to userspace.
8 Taskstats was designed for the following benefits:
10 - efficiently provide statistics during lifetime of a task and on its exit
11 - unified interface for multiple accounting subsystems
12 - extensibility for use by future accounting patches
17 "pid", "tid" and "task" are used interchangeably and refer to the standard
18 Linux task defined by struct task_struct. per-pid stats are the same as
21 "tgid", "process" and "thread group" are used interchangeably and refer to the
22 tasks that share an mm_struct i.e. the traditional Unix process. Despite the
23 use of tgid, there is no special treatment for the task that is thread group
24 leader - a process is deemed alive as long as it has any task belonging to it.
29 To get statistics during a task's lifetime, userspace opens a unicast netlink
30 socket (NETLINK_GENERIC family) and sends commands specifying a pid or a tgid.
31 The response contains statistics for a task (if pid is specified) or the sum of
32 statistics for all tasks of the process (if tgid is specified).
34 To obtain statistics for tasks which are exiting, the userspace listener
35 sends a register command and specifies a cpumask. Whenever a task exits on
36 one of the cpus in the cpumask, its per-pid statistics are sent to the
37 registered listener. Using cpumasks allows the data received by one listener
38 to be limited and assists in flow control over the netlink interface and is
39 explained in more detail below.
41 If the exiting task is the last thread exiting its thread group,
42 an additional record containing the per-tgid stats is also sent to userspace.
43 The latter contains the sum of per-pid stats for all threads in the thread
44 group, both past and present.
46 getdelays.c is a simple utility demonstrating usage of the taskstats interface
47 for reporting delay accounting statistics. Users can register cpumasks,
48 send commands and process responses, listen for per-tid/tgid exit data,
49 write the data received to a file and do basic flow control by increasing
55 The user-kernel interface is encapsulated in include/linux/taskstats.h
57 To avoid this documentation becoming obsolete as the interface evolves, only
58 an outline of the current version is given. taskstats.h always overrides the
61 struct taskstats is the common accounting structure for both per-pid and
62 per-tgid data. It is versioned and can be extended by each accounting subsystem
63 that is added to the kernel. The fields and their semantics are defined in the
66 The data exchanged between user and kernel space is a netlink message belonging
67 to the NETLINK_GENERIC family and using the netlink attributes interface.
68 The messages are in the format
70 +----------+- - -+-------------+-------------------+
71 | nlmsghdr | Pad | genlmsghdr | taskstats payload |
72 +----------+- - -+-------------+-------------------+
75 The taskstats payload is one of the following three kinds:
77 1. Commands: Sent from user to kernel. Commands to get data on
78 a pid/tgid consist of one attribute, of type TASKSTATS_CMD_ATTR_PID/TGID,
79 containing a u32 pid or tgid in the attribute payload. The pid/tgid denotes
80 the task/process for which userspace wants statistics.
82 Commands to register/deregister interest in exit data from a set of cpus
83 consist of one attribute, of type
84 TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK and contain a cpumask in the
85 attribute payload. The cpumask is specified as an ascii string of
86 comma-separated cpu ranges e.g. to listen to exit data from cpus 1,2,3,5,7,8
87 the cpumask would be "1-3,5,7-8". If userspace forgets to deregister interest
88 in cpus before closing the listening socket, the kernel cleans up its interest
89 set over time. However, for the sake of efficiency, an explicit deregistration
92 2. Response for a command: sent from the kernel in response to a userspace
93 command. The payload is a series of three attributes of type:
95 a) TASKSTATS_TYPE_AGGR_PID/TGID : attribute containing no payload but indicates
96 a pid/tgid will be followed by some stats.
98 b) TASKSTATS_TYPE_PID/TGID: attribute whose payload is the pid/tgid whose stats
101 c) TASKSTATS_TYPE_STATS: attribute with a struct taskstsats as payload. The
102 same structure is used for both per-pid and per-tgid stats.
104 3. New message sent by kernel whenever a task exits. The payload consists of a
105 series of attributes of the following type:
107 a) TASKSTATS_TYPE_AGGR_PID: indicates next two attributes will be pid+stats
108 b) TASKSTATS_TYPE_PID: contains exiting task's pid
109 c) TASKSTATS_TYPE_STATS: contains the exiting task's per-pid stats
110 d) TASKSTATS_TYPE_AGGR_TGID: indicates next two attributes will be tgid+stats
111 e) TASKSTATS_TYPE_TGID: contains tgid of process to which task belongs
112 f) TASKSTATS_TYPE_STATS: contains the per-tgid stats for exiting task's process
118 Taskstats provides per-process stats, in addition to per-task stats, since
119 resource management is often done at a process granularity and aggregating task
120 stats in userspace alone is inefficient and potentially inaccurate (due to lack
123 However, maintaining per-process, in addition to per-task stats, within the
124 kernel has space and time overheads. To address this, the taskstats code
125 accumalates each exiting task's statistics into a process-wide data structure.
126 When the last task of a process exits, the process level data accumalated also
127 gets sent to userspace (along with the per-task data).
129 When a user queries to get per-tgid data, the sum of all other live threads in
130 the group is added up and added to the accumalated total for previously exited
131 threads of the same thread group.
136 There are two ways to extend the taskstats interface to export more
137 per-task/process stats as patches to collect them get added to the kernel
140 1. Adding more fields to the end of the existing struct taskstats. Backward
141 compatibility is ensured by the version number within the
142 structure. Userspace will use only the fields of the struct that correspond
143 to the version its using.
145 2. Defining separate statistic structs and using the netlink attributes
146 interface to return them. Since userspace processes each netlink attribute
147 independently, it can always ignore attributes whose type it does not
148 understand (because it is using an older version of the interface).
151 Choosing between 1. and 2. is a matter of trading off flexibility and
152 overhead. If only a few fields need to be added, then 1. is the preferable
153 path since the kernel and userspace don't need to incur the overhead of
154 processing new netlink attributes. But if the new fields expand the existing
155 struct too much, requiring disparate userspace accounting utilities to
156 unnecessarily receive large structures whose fields are of no interest, then
157 extending the attributes structure would be worthwhile.
159 Flow control for taskstats
160 --------------------------
162 When the rate of task exits becomes large, a listener may not be able to keep
163 up with the kernel's rate of sending per-tid/tgid exit data leading to data
164 loss. This possibility gets compounded when the taskstats structure gets
165 extended and the number of cpus grows large.
167 To avoid losing statistics, userspace should do one or more of the following:
169 - increase the receive buffer sizes for the netlink sockets opened by
170 listeners to receive exit data.
172 - create more listeners and reduce the number of cpus being listened to by
173 each listener. In the extreme case, there could be one listener for each cpu.
174 Users may also consider setting the cpu affinity of the listener to the subset
175 of cpus to which it listens, especially if they are listening to just one cpu.
177 Despite these measures, if the userspace receives ENOBUFS error messages
178 indicated overflow of receive buffers, it should take measures to handle the