4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * This file is part of Lustre, http://www.lustre.org/
32 * Lustre is a trademark of Sun Microsystems, Inc.
34 * lustre/utils/ltrack_stats.c
36 * Author: Milind Dumbare <milind@clusterfs.com>
39 #include <sys/types.h>
50 #define TRACK_BY_GID 0
51 #define TRACK_BY_PPID 1
52 #define TRACK_BY_PID 2
53 #define TRACK_FOR_ALL 3
55 /* We can have at the most 1024 llstats monitoring 1024 lustre clients
57 #define NO_OF_CLIENTS 1024
59 /* length of absolute path of vfs_ops_stats */
60 #define LEN_STATS 1024
62 /* Length of llstat command with all its switches and command line options */
63 #define LEN_LLSTAT (25 + LEN_STATS)
65 /* strlen of each lustre client entry in /proc/fs/lustre/llite/ */
66 #define LEN_CLIENT 1024
68 /* size of output of llstat command we read at a time */
71 /* max strlen of outfile we get on command line */
74 /* Length of command given on command line */
82 printf("ltrack_stats runs command given and does one of the "
84 "\t1. Writes its pid to "
85 "/proc/fs/lustre/llite/.../stats_track_pid\n"
86 " to collects stats for that process.\n"
87 "\t2. Writes its ppid to "
88 "/proc/fs/lustre/llite/.../stats_track_ppid\n"
89 " to collect stats of that process and all its children \n"
90 "\t3. Sets gid of process to some random gid (444) and also\n"
91 " writes that to/proc/fs/lustre/llite/.../stats_track_gid to"
92 " collect stats \nof all processes in that group\n\n"
93 " It also uses llstat to generate output with interval of 1 "
94 " second and duration\n of run of command for plot-llstat to "
95 "generate a graph\n\n");
96 printf("ltrack_stats [-l filename] [-g <gid> | -a | -i | -c | -h ]\n"
97 "\t<command with arguments ...>\n\n");
98 printf("-l: outputs the llstat.pl's output to given <filename>"
99 "with interval of 1 second \nbetween each output and flag for"
100 "graphable output. If -l flag is not given llstat \nwont be"
103 printf("-g: for displaying VFS operation statistics collected"
104 "for all processes having \ngroup id as given <gid> \n\n");
106 printf("-a: for displaying VFS operations statistics collected"
107 "for all processes\n\n");
109 printf("-i: for displaying VFS operation statistics collected"
110 "for only given <command's> \nPID.\n\n");
112 printf("-c: for displaying VFS operation statistics collected"
113 " for all processes whose \nparents' PID is same as pid of "
114 "<command> to be executed\n\n");
116 printf("-h: for showing this help\n");
119 /* - type: to which file data should be written to track VFS operations
121 * - id: we either need to write gid which is given on command line or pid
122 * to collect statistics of that process or its children. */
124 void write_track_xid(int type, unsigned short id, char* stats_path)
128 /* for loop is used if we have more than one lustre clients on same
129 * machine. glob() return /proc entry for each lustre client */
133 strcat(stats_path, "/stats_track_gid");
136 strcat(stats_path, "/stats_track_ppid");
139 strcat(stats_path, "/stats_track_pid");
143 fp = fopen(stats_path, "w+");
145 fprintf(stderr, "Error: Couldn't open /proc entry file: %s\n",
149 if (fprintf(fp, "%d", id) < 0) {
150 fprintf(stderr, "Error: Couldn't write id to tracking /proc"
151 "entry file: %s\n", stats_path);
155 fprintf(stderr, "Error: Couldn't close tracking /proc entry"
156 " file: %s\n", stats_path);
161 /* Get extra command lines concatenated to "command Function getopt scans
162 * one switch and its optional argument. So if command line is
163 * "-g 1234 make bzImage" 1234 is optarg and "make bzImage" will be collected
164 * with following function to exec it. */
165 char* get_command_from_argv(int optind, int argc, char** argv,char* optarg,
170 strcpy(command, optarg);
171 strcat(command, " ");
172 for (index = optind; index < argc; index++) {
173 strcat(command, argv[index]);
174 strcat(command, " ");
176 if (strlen(command) == 1) {
177 fprintf(stderr,"Error: command missing \n");
180 } else if (strlen(command) > COMM_LEN) {
181 fprintf(stderr,"Error: Too long command \n");
189 /* Check for the llstat command in $PATH env variable. */
194 status = system("which llstat.pl &> /dev/null");
196 fprintf(stderr,"Error: llstat.pl not found in PATH\n");
201 pid_t fork_llstat_command(char* llstat_file,char* stats_path)
203 char truncate_command[100];
204 char llstat_command[LEN_LLSTAT];
205 pid_t pid_llstat_command;
206 FILE *fp_popen, *fp_out;
210 /* Truncating llstat output file as it will be opened in while
211 * loop to append output */
212 sprintf(truncate_command,"> %s",llstat_file);
213 if ((ret = system(truncate_command)) != 0) {
214 ret = WEXITSTATUS(ret);
215 printf("error excuting truncate command: %d\n", ret);
219 strcat(stats_path, "/stats");
221 /* creating a string of llstat command to give to
223 sprintf(llstat_command, "llstat -i 1 -g %s ",
226 /* fork for llstat */
227 if ((pid_llstat_command = fork()) < 0)
228 fprintf(stderr, "Error: Fork error\n");
230 /* in child (llstat) */
231 if (pid_llstat_command == 0) {
232 /* Doing popen for llstat command */
233 fp_popen = popen(llstat_command, "r");
234 if (fp_popen == NULL) {
235 fprintf(stderr,"Couldn't popen the llstat command:"
236 "\"%s\"n", llstat_command);
239 while (fgets(buffer, 1024, fp_popen) != NULL) {
240 /* Following code should be in while loop as llstat
241 * will keep on sending output each second and will
242 * not exit on itself. It will be killed when we finsh
243 * with our command so we must make the output file
244 * consistent after writing each 1024 bytes chunk */
246 /* opening file where llstat will write its output */
247 fp_out = fopen(llstat_file, "a");
249 fprintf(stderr, "Error: Couldn't open llstat"
250 "outfile file: %s\n",
254 /* fgets reads the popen output and fprintf writes it to
257 if (fputs(buffer, fp_out) == EOF) {
258 fprintf(stderr, "Error: Couldn't write output"
259 "of llstat to out file\n");
263 /* closing file opened for storing llstat's output */
264 if (fclose(fp_out)) {
265 fprintf(stderr, "Error: Couldn't close llstat"
266 "outfile: %s\n", llstat_file);
270 /* closing popen for llstat */
271 if (pclose(fp_popen) < 0) {
272 fprintf(stderr, "Error: Couldn't pclos"
273 " llstat popen call\n");
277 return pid_llstat_command;
280 pid_t fork_actual_command(int type, unsigned short id, char* stats_path,
285 /* starting ltrack_stats functionality here */
286 if ((pid = fork()) < 0)
287 fprintf(stderr, "Error: Fork error\n");
289 /* fork for executing command */
293 if (setgid(id) < 0) {
294 fprintf(stderr, "Error: failed to"
306 /* 0 has to be written to vfs_track_pid to collect
307 * statistics of all processes */
313 write_track_xid(type, pid, stats_path);
314 execl("/bin/sh", "sh", "-c", command, (char *)0);
321 char* get_path_stats(int with_llstat, char* stats_path)
323 glob_t stats_glob_buffer;
328 /* No slots reserved in gl_pathv. Store the found path at 0 location */
329 stats_glob_buffer.gl_offs = 0;
331 /* doing glob() for attaching llstat to monitor each vfs_ops_stat for
332 * mulitiple lustre clients */
333 if (glob("/proc/fs/lustre/llite/*", GLOB_DOOFFS, NULL,
334 &stats_glob_buffer) != 0) {
335 fprintf(stderr,"Error: Couldn't find /proc entry for "
340 /* If multiple client entries found in /proc/fs/lustre/llite user will
341 * be prompted with choice of all */
342 if (stats_glob_buffer.gl_pathc > 1 && with_llstat) {
344 printf("Multiple lustre clients found, continuing... \n");
346 /* If flow is here again it means there was an error
347 * and notifying that to user */
350 if ((ret = system("clear")) != 0) {
351 ret = WEXITSTATUS(ret);
352 printf("error excuting clear command: %d\n", ret);
355 fprintf(stderr, "Error: Please give correct "
358 /* Simple menu based interface to avoid possible
359 * spelling mistakes */
360 printf("\t\tMenu.\n");
361 for (i = 0; i < stats_glob_buffer.gl_pathc; i++)
362 printf("\t\t%d. %s\n", i+1,
363 stats_glob_buffer.gl_pathv[i]);
365 printf("\nEnter the lustre client number you want to "
367 if (scanf(" %d", &choice) == EOF && ferror(stdin)) {
368 perror("reading from stdin");
372 } while (choice > stats_glob_buffer.gl_pathc || choice < 1);
373 strcpy(stats_path, stats_glob_buffer.gl_pathv[choice - 1]);
375 /*if only one client then simply copying the path from glob */
376 strcpy(stats_path, stats_glob_buffer.gl_pathv[0]);
378 /* this frees dynamically allocated space by glob() for storing found
380 globfree(&stats_glob_buffer);
385 /* Writes the id (gid/ pid/ ppid) value in appropriate tracking proc entry file
386 * and EXECs the command given */
387 void fork_command(int type, unsigned short id, char* command, char* llstat_file)
389 pid_t pid_actual_command = 0;
390 pid_t pid_llstat_command = 0;
395 char stats_path[1024];
396 char stats_path_temp[1024];
398 if (strlen(llstat_file) == 0)
401 get_path_stats(with_llstat, stats_path);
402 strcpy(stats_path_temp, stats_path);
404 /* llstat process attached to monitor given command */
406 pid_llstat_command = fork_llstat_command(llstat_file,
409 /* forking a process which will exec command given */
410 pid_actual_command = fork_actual_command(type, id, stats_path,
413 if (waitpid(pid_actual_command, NULL, 0) != pid_actual_command)
414 fprintf(stderr, "Error: waitpid error\n");
417 /* comment #25 of BUG 10968 */
420 /* sending kill to all llstat commands created for each
421 * lustre-client respectively */
422 kill(pid_llstat_command, 9);
423 waitpid(pid_llstat_command, &status, 0);
425 /* if llstat child is killed by KILL only then print note for
426 * plotting graph and if its exited normally with errornous
427 * status then it means there were some error and llstat was
429 if (!WIFEXITED(status))
430 printf("\n\t[Note: Do \"$plot-llstat %s\" to plot a graph"
431 " using GNU plot]\n", llstat_file);
437 int main(int argc, char **argv)
439 char gid_string[5] = "";
442 char command[COMM_LEN] = "";
443 char llstat_file[100] = "";
445 /* Checking for root*/
447 fprintf(stderr, "Error: You need to be root\n");
452 /* Parsing command line switches */
453 while ((c = getopt(argc, argv, "l:g:c:i:a:h")) != 1)
456 strcpy(llstat_file, optarg);
457 if (strlen(llstat_file) > LEN_OUT) {
458 fprintf(stderr, "length of outfile file"
464 /* When any value is written to vfs_track_gid, then VFS
465 * operation statistics are collected for all
466 * processes of that group ID.
467 * write_track_xid writes given <gid> in vfs_track_gid
470 strcpy(gid_string, optarg);
471 get_command_from_argv(optind, argc, argv, "",
473 gid = atoi(gid_string);
475 fork_command(TRACK_BY_GID, gid, command,
479 /* When any value is written to vfs_track_ppid, then VFS
480 * operation statistics are collected for all processes
481 * whose parents' PID is same as track_ppid.
482 *- write_track_xid writes pid to vfs_track_ppid here */
484 get_command_from_argv(optind, argc, argv,
486 fork_command(TRACK_BY_PPID, 0, command,
490 /* When a non-zero value is written to vfs_track_pid,
491 * then VFS operation statistics are collected for only
492 * that PID.Write_track_xid writes pid to vfs_track_pid
493 * here.Wei need to FORK a new process and EXEC it with
494 * given <command>. */
496 get_command_from_argv(optind, argc, argv,
498 fork_command(TRACK_BY_PID, 0, command,
502 /* When VFS operation statistics for all processes are
503 * to be collected, "0" is written to vfs_track_pid. */
505 get_command_from_argv(optind, argc, argv,
507 fork_command(TRACK_FOR_ALL, 0, command,