taskstats: version 12 with thread group and exe info

The task exit struct needs some crucial information to be able to provide
an enhanced version of process and thread accounting.  This change
provides:

1. ac_tgid in additon to ac_pid
2. thread group execution walltime in ac_tgetime
3. flag AGROUP in ac_flag to indicate the last task
   in a thread group / process
4. device ID and inode of task's /proc/self/exe in
   ac_exe_dev and ac_exe_inode
5. tools/accounting/procacct as demonstrator

When a task exits, taskstats are reported to userspace including the
task's pid and ppid, but without the id of the thread group this task is
part of.  Without the tgid, the stats of single tasks cannot be correlated
to each other as a thread group (process).

The taskstats documentation suggests that on process exit a data set
consisting of accumulated stats for the whole group is produced.  But such
an additional set of stats is only produced for actually multithreaded
processes, not groups that had only one thread, and also those stats only
contain data about delay accounting and not the more basic information
about CPU and memory resource usage.  Adding the AGROUP flag to be set
when the last task of a group exited enables determination of process end
also for single-threaded processes.

My applicaton basically does enhanced process accounting with summed
cputime, biggest maxrss, tasks per process.  The data is not available
with the traditional BSD process accounting (which is not designed to be
extensible) and the taskstats interface allows more efficient on-the-fly
grouping and summing of the stats, anyway, without intermediate disk
writes.

Furthermore, I do carry statistics on which exact program binary is used
how often with associated resources, getting a picture on how important
which parts of a collection of installed scientific software in different
versions are, and how well they put load on the machine.  This is enabled
by providing information on /proc/self/exe for each task.  I assume the
two 64-bit fields for device ID and inode are more appropriate than the
possibly large resolved path to keep the data volume down.

Add the tgid to the stats to complete task identification, the flag AGROUP
to mark the last task of a group, the group wallclock time, and
inode-based identification of the associated executable file.

Add tools/accounting/procacct.c as a simplified fork of getdelays.c to
demonstrate process and thread accounting.

[thomas.orgis@uni-hamburg.de: fix version number in comment]
  Link: https://lkml.kernel.org/r/20220405003601.7a5f6008@plasteblaster
Link: https://lkml.kernel.org/r/20220331004106.64e5616b@plasteblasterSigned-off-by: Dr. Thomas Orgis <thomas.orgis@uni-hamburg.de>
Reviewed-by: Ismael Luceno <ismael@iodev.co.uk>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: xu xin <xu.xin16@zte.com.cn>
Cc: Yang Yang <yang.yang29@zte.com.cn>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

include/uapi/linux/acct.h

@@ -103,12 +103,13 @@ struct acct_v3
 /*
  *  accounting flags
  */
-				/* bit set when the process ... */
+				/* bit set when the process/task ... */
 #define AFORK		0x01	/* ... executed fork, but did not exec */
 #define ASU		0x02	/* ... used super-user privileges */
 #define ACOMPAT		0x04	/* ... used compatibility mode (VAX only not used) */
 #define ACORE		0x08	/* ... dumped core */
 #define AXSIG		0x10	/* ... was killed by a signal */
+#define AGROUP		0x20	/* ... was the last task of the process (task group) */
 
 #if defined(__BYTE_ORDER) ? __BYTE_ORDER == __BIG_ENDIAN : defined(__BIG_ENDIAN)
 #define ACCT_BYTEORDER	0x80	/* accounting file is big endian */

include/uapi/linux/taskstats.h

@@ -34,7 +34,7 @@
  */
 
 
-#define TASKSTATS_VERSION	11
+#define TASKSTATS_VERSION	12
 #define TS_COMM_LEN		32	/* should be >= TASK_COMM_LEN
 					 * in linux/sched.h */
 
@@ -48,7 +48,8 @@ struct taskstats {
 	__u32	ac_exitcode;		/* Exit status */
 
 	/* The accounting flags of a task as defined in <linux/acct.h>
-	 * Defined values are AFORK, ASU, ACOMPAT, ACORE, and AXSIG.
+	 * Defined values are AFORK, ASU, ACOMPAT, ACORE, AXSIG, and AGROUP.
+	 * (AGROUP since version 12).
 	 */
 	__u8	ac_flag;		/* Record flags */
 	__u8	ac_nice;		/* task_nice */
@@ -173,9 +174,26 @@ struct taskstats {
 	/* v10: 64-bit btime to avoid overflow */
 	__u64	ac_btime64;		/* 64-bit begin time */
 
-	/* Delay waiting for memory compact */
+	/* v11: Delay waiting for memory compact */
 	__u64	compact_count;
 	__u64	compact_delay_total;
+
+	/* v12 begin */
+	__u32   ac_tgid;	/* thread group ID */
+	/* Thread group walltime up to now. This is total process walltime if
+	 * AGROUP flag is set.
+	 */
+	__u64	ac_tgetime __attribute__((aligned(8)));
+	/* Lightweight information to identify process binary files.
+	 * This leaves userspace to match this to a file system path, using
+	 * MAJOR() and MINOR() macros to identify a device and mount point,
+	 * the inode to identify the executable file. This is /proc/self/exe
+	 * at the end, so matching the most recent exec(). Values are zero
+	 * for kernel threads.
+	 */
+	__u64   ac_exe_dev;     /* program binary device ID */
+	__u64   ac_exe_inode;   /* program binary inode number */
+	/* v12 end */
 };
 
 

kernel/taskstats.c

@@ -9,6 +9,7 @@
 #include <linux/kernel.h>
 #include <linux/taskstats_kern.h>
 #include <linux/tsacct_kern.h>
+#include <linux/acct.h>
 #include <linux/delayacct.h>
 #include <linux/cpumask.h>
 #include <linux/percpu.h>
@@ -153,6 +154,23 @@ static void send_cpu_listeners(struct sk_buff *skb,
 	up_write(&listeners->sem);
 }
 
+static void exe_add_tsk(struct taskstats *stats, struct task_struct *tsk)
+{
+	/* No idea if I'm allowed to access that here, now. */
+	struct file *exe_file = get_task_exe_file(tsk);
+
+	if (exe_file) {
+		/* Following cp_new_stat64() in stat.c . */
+		stats->ac_exe_dev =
+			huge_encode_dev(exe_file->f_inode->i_sb->s_dev);
+		stats->ac_exe_inode = exe_file->f_inode->i_ino;
+		fput(exe_file);
+	} else {
+		stats->ac_exe_dev = 0;
+		stats->ac_exe_inode = 0;
+	}
+}
+
 static void fill_stats(struct user_namespace *user_ns,
 		       struct pid_namespace *pid_ns,
 		       struct task_struct *tsk, struct taskstats *stats)
@@ -175,6 +193,9 @@ static void fill_stats(struct user_namespace *user_ns,
 
 	/* fill in extended acct fields */
 	xacct_add_tsk(stats, tsk);
+
+	/* add executable info */
+	exe_add_tsk(stats, tsk);
 }
 
 static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
@@ -620,6 +641,8 @@ void taskstats_exit(struct task_struct *tsk, int group_dead)
 		goto err;
 
 	fill_stats(&init_user_ns, &init_pid_ns, tsk, stats);
+	if (group_dead)
+		stats->ac_flag |= AGROUP;
 
 	/*
 	 * Doesn't matter if tsk is the leader or the last group member leaving

kernel/tsacct.c

@@ -23,15 +23,20 @@ void bacct_add_tsk(struct user_namespace *user_ns,
 {
 	const struct cred *tcred;
 	u64 utime, stime, utimescaled, stimescaled;
-	u64 delta;
+	u64 now_ns, delta;
 	time64_t btime;
 
 	BUILD_BUG_ON(TS_COMM_LEN < TASK_COMM_LEN);
 
 	/* calculate task elapsed time in nsec */
-	delta = ktime_get_ns() - tsk->start_time;
+	now_ns = ktime_get_ns();
+	/* store whole group time first */
+	delta = now_ns - tsk->group_leader->start_time;
 	/* Convert to micro seconds */
 	do_div(delta, NSEC_PER_USEC);
+	stats->ac_tgetime = delta;
+	delta = now_ns - tsk->start_time;
+	do_div(delta, NSEC_PER_USEC);
 	stats->ac_etime = delta;
 	/* Convert to seconds for btime (note y2106 limit) */
 	btime = ktime_get_real_seconds() - div_u64(delta, USEC_PER_SEC);
@@ -51,6 +56,7 @@ void bacct_add_tsk(struct user_namespace *user_ns,
 	stats->ac_nice	 = task_nice(tsk);
 	stats->ac_sched	 = tsk->policy;
 	stats->ac_pid	 = task_pid_nr_ns(tsk, pid_ns);
+	stats->ac_tgid   = task_tgid_nr_ns(tsk, pid_ns);
 	rcu_read_lock();
 	tcred = __task_cred(tsk);
 	stats->ac_uid	 = from_kuid_munged(user_ns, tcred->uid);

tools/accounting/.gitignore

 # SPDX-License-Identifier: GPL-2.0-only
 getdelays
+procacct

tools/accounting/Makefile

@@ -2,7 +2,7 @@
 CC := $(CROSS_COMPILE)gcc
 CFLAGS := -I../../usr/include
 
-PROGS := getdelays
+PROGS := getdelays procacct
 
 all: $(PROGS)