Collect nropen for each process to monitor if fd leaks

This patch collects and displays the current number of opened file
descriptors for each process to monitor if fd leaks. As each thread
has the same fds with its process, there is no need to collect each
thread's fd numbers then.

Considering users may set the nr_open limit to a very large number,
we regard MAX_OPEN to be 10000 to avoid further performance
degradation.

Signed-off-by: Fei Li <lifei.shirley@bytedance.com>

Author: ShirleyFei

deviate.c

@@ -455,6 +455,7 @@ calcdiff(struct tstat *devstat, const struct tstat *curstat,
 	devstat->cpu.cgcpuweight = curstat->cpu.cgcpuweight;
 	devstat->cpu.cgcpumax    = curstat->cpu.cgcpumax;
 	devstat->cpu.cgcpumaxr   = curstat->cpu.cgcpumaxr;
+	devstat->cpu.nropen      = curstat->cpu.nropen;
 
 	if (curstat->cpu.wchan[0])
 		strcpy(devstat->cpu.wchan, curstat->cpu.wchan);

json.c

@@ -1030,6 +1030,7 @@ static void json_print_PRC(char *hp, struct sstat *ss, struct tstat *ps, int nac
 			"\"blkdelay\": %lld, "
 			"\"nvcsw\": %llu, "
 			"\"nivcsw\": %llu, "
+			"\"nropen\": %d, "
 			"\"sleepavg\": %d}",
 			ps->gen.pid,
 			ps->cpu.utime,
@@ -1043,6 +1044,7 @@ static void json_print_PRC(char *hp, struct sstat *ss, struct tstat *ps, int nac
 			ps->cpu.blkdelay*1000/hertz,
 			ps->cpu.nvcsw,
 			ps->cpu.nivcsw,
+			ps->cpu.nropen,
 			ps->cpu.sleepavg);
 	}
 

man/atop.1

@@ -1661,6 +1661,11 @@ Time that the process has been finished. If the process is still running,
 this field shows `active'.
 .PP
 .TP 9
+.B NROPEN
+Current number of opened file descriptors (fds) for each process, at least 0.
+As each thread has the same fds with its process, this filed shows `-`.
+.PP
+.TP 9
 .B ENVID
 Virtual environment identified (OpenVZ only).
 .PP

parseable.c

@@ -785,7 +785,7 @@ print_PRC(char *hp, struct sstat *ss, struct tstat *ps, int nact)
 	for (i=0; i < nact; i++, ps++)
 	{
 		printf("%s %d %s %c %u %lld %lld %d %d %d %d %d %d %d %c "
-		       "%llu %s %llu %d %d %llu %llu\n",
+		       "%llu %s %llu %d %d %d %llu %llu\n",
 			hp,
 			ps->gen.pid,
 			spaceformat(ps->gen.name, namout),
@@ -804,6 +804,7 @@ print_PRC(char *hp, struct sstat *ss, struct tstat *ps, int nact)
 			ps->cpu.rundelay,
 			spaceformat(ps->cpu.wchan, wchanout),
 			ps->cpu.blkdelay,
+			ps->cpu.nropen,
 			cgroupv2max(ps->gen.isproc, ps->cpu.cgcpumax),
 			cgroupv2max(ps->gen.isproc, ps->cpu.cgcpumaxr),
 			ps->cpu.nvcsw,

photoproc.c

@@ -63,6 +63,7 @@ static int	proccont(struct tstat *);
 static void	proccmd(struct tstat *);
 static void	procsmaps(struct tstat *);
 static void	procwchan(struct tstat *);
+static void	procfd(struct tstat *);
 static count_t	procschedstat(struct tstat *);
 static int	proccgroupv2(struct tstat *);
 static struct cgroupv2vals *
@@ -218,6 +219,8 @@ photoproc(struct tstat *tasklist, int maxtask)
                 if (getwchan)
                 	procwchan(curtask);
 
+		procfd(curtask);
+
 		// read network stats from netatop
 		netatop_gettask(curtask->gen.tgid, 'g', curtask);
 
@@ -981,6 +984,38 @@ procschedstat(struct tstat *curtask)
 	return curtask->cpu.rundelay;
 }
 
+/*
+** get current number of opened file descriptors for process
+** to monitor if fd leaks.
+** Considering users may set the max number of open files to
+** a very large number, we regard MAX_OPEN to be 10000.
+*/
+#define MAX_OPEN 10000
+
+static void
+procfd(struct tstat *curtask)
+{
+	DIR *dirp;
+	struct dirent *dentry;
+	int fd_num = 0;
+
+	if ( (dirp = opendir("fd")) )
+	{
+		while ( (dentry = readdir(dirp)) )
+		{
+			if ( isdigit(dentry->d_name[0]) )
+				fd_num++;
+
+			if ( fd_num >= MAX_OPEN )
+				break;
+		}
+
+		curtask->cpu.nropen = fd_num;
+
+		closedir(dirp);
+	}
+}
+
 /*
 ** CGROUP V2 specific items
 */

photoproc.h

@@ -82,7 +82,8 @@ struct tstat {
 		int	cgcpuweight;	/* cgroup cpu.weight            */
 		int	cgcpumax;	/* cgroup cpu.max percentage    */
 		int	cgcpumaxr;	/* restrictive percentage       */
-		int	ifuture[3];	/* reserved for future use	*/
+		int	nropen;		/* number of opened files       */
+		int	ifuture[2];	/* reserved for future use	*/
 		char	wchan[16];	/* wait channel string    	*/
 		count_t	rundelay;	/* schedstat rundelay (nanosec)	*/
 		count_t	blkdelay;	/* blkio delay (ticks)		*/

showlinux.c

@@ -423,6 +423,7 @@ proc_printdef *allprocpdefs[]=
 	&procprt_STTIME,
 	&procprt_ENDATE,
 	&procprt_ENTIME,
+	&procprt_NROPEN,
 	&procprt_THR,
 	&procprt_TRUN,
 	&procprt_TSLPI,
@@ -1300,7 +1301,7 @@ priphead(int curlist, int totlist, char *showtype, char *showorder,
 			"RUID:8 RGID:8 EUID:5 EGID:4 "
      			"SUID:3 SGID:2 FSUID:3 FSGID:2 "
                         "STDATE:7 STTIME:7 ENDATE:5 ENTIME:5 "
-			"ST:6 EXC:6 S:6 SORTITEM:10 CMD:10", 
+			"NROPEN:5 ST:6 EXC:6 S:6 SORTITEM:10 CMD:10",
                         "built-in varprocs");
 
                 make_proc_prints(cmdprocs, MAXITEMS,

showlinux.h

@@ -418,6 +418,7 @@ extern proc_printdef procprt_STDATE;
 extern proc_printdef procprt_STTIME;
 extern proc_printdef procprt_ENDATE;
 extern proc_printdef procprt_ENTIME;
+extern proc_printdef procprt_NROPEN;
 extern proc_printdef procprt_THR;
 extern proc_printdef procprt_TRUN;
 extern proc_printdef procprt_TSLPI;

showprocs.c

@@ -115,6 +115,8 @@ char *procprt_ENDATE_a(struct tstat *, int, int);
 char *procprt_ENDATE_e(struct tstat *, int, int);
 char *procprt_ENTIME_a(struct tstat *, int, int);
 char *procprt_ENTIME_e(struct tstat *, int, int);
+char *procprt_NROPEN_a(struct tstat *, int, int);
+char *procprt_NROPEN_e(struct tstat *, int, int);
 char *procprt_THR_a(struct tstat *, int, int);
 char *procprt_THR_e(struct tstat *, int, int);
 char *procprt_TRUN_a(struct tstat *, int, int);
@@ -1235,6 +1237,32 @@ proc_printdef procprt_ENTIME =
    { " ENTIME ", "ENTIME", procprt_ENTIME_a, procprt_ENTIME_e, 8 };
 /***************************************************************/
 char *
+procprt_NROPEN_a(struct tstat *curstat, int avgval, int nsecs)
+{
+	static char buf[64];
+
+	if (curstat->gen.isproc)
+		sprintf(buf, "%*d", procprt_NROPEN.width, curstat->cpu.nropen);
+	else
+		sprintf(buf, "%*s", procprt_NROPEN.width, "-");
+
+	return buf;
+}
+
+char *
+procprt_NROPEN_e(struct tstat *curstat, int avgval, int nsecs)
+{
+	static char buf[64];
+
+	sprintf(buf, "%*s", procprt_NROPEN.width, "-");
+
+	return buf;
+}
+
+proc_printdef procprt_NROPEN =
+   { " NROPEN", "NROPEN", procprt_NROPEN_a, procprt_NROPEN_e, 7 };
+/***************************************************************/
+char *
 procprt_THR_a(struct tstat *curstat, int avgval, int nsecs)
 {
         static char buf[15];