1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/ltrack_stats.c
38 * Author: Milind Dumbare <milind@clusterfs.com>
41 #include <sys/types.h>
52 #define TRACK_BY_GID 0
53 #define TRACK_BY_PPID 1
54 #define TRACK_BY_PID 2
55 #define TRACK_FOR_ALL 3
57 /* We can have at the most 1024 llstats monitoring 1024 lustre clients
59 #define NO_OF_CLIENTS 1024
61 /* length of absolute path of vfs_ops_stats */
62 #define LEN_STATS 1024
64 /* Length of llstat command with all its switches and command line options */
65 #define LEN_LLSTAT (25 + LEN_STATS)
67 /* strlen of each lustre client entry in /proc/fs/lustre/llite/ */
68 #define LEN_CLIENT 1024
70 /* size of output of llstat command we read at a time */
73 /* max strlen of outfile we get on command line */
76 /* Length of command given on command line */
84 printf("ltrack_stats runs command given and does one of the "
86 "\t1. Writes its pid to "
87 "/proc/fs/lustre/llite/.../stats_track_pid\n"
88 " to collects stats for that process.\n"
89 "\t2. Writes its ppid to "
90 "/proc/fs/lustre/llite/.../stats_track_ppid\n"
91 " to collect stats of that process and all its children \n"
92 "\t3. Sets gid of process to some random gid (444) and also\n"
93 " writes that to/proc/fs/lustre/llite/.../stats_track_gid to"
94 " collect stats \nof all processes in that group\n\n"
95 " It also uses llstat to generate output with interval of 1 "
96 " second and duration\n of run of command for plot-llstat to "
97 "generate a graph\n\n");
98 printf("ltrack_stats [-l filename] [-g <gid> | -a | -i | -c | -h ]\n"
99 "\t<command with arguments ...>\n\n");
100 printf("-l: outputs the llstat.pl's output to given <filename>"
101 "with interval of 1 second \nbetween each output and flag for"
102 "graphable output. If -l flag is not given llstat \nwont be"
105 printf("-g: for displaying VFS operation statistics collected"
106 "for all processes having \ngroup id as given <gid> \n\n");
108 printf("-a: for displaying VFS operations statistics collected"
109 "for all processes\n\n");
111 printf("-i: for displaying VFS operation statistics collected"
112 "for only given <command's> \nPID.\n\n");
114 printf("-c: for displaying VFS operation statistics collected"
115 " for all processes whose \nparents' PID is same as pid of "
116 "<command> to be executed\n\n");
118 printf("-h: for showing this help\n");
121 /* - type: to which file data should be written to track VFS operations
123 * - id: we either need to write gid which is given on command line or pid
124 * to collect statistics of that process or its children. */
126 void write_track_xid(int type, unsigned short id, char* stats_path)
130 /* for loop is used if we have more than one lustre clients on same
131 * machine. glob() return /proc entry for each lustre client */
135 strcat(stats_path, "/stats_track_gid");
138 strcat(stats_path, "/stats_track_ppid");
141 strcat(stats_path, "/stats_track_pid");
145 fp = fopen(stats_path, "w+");
147 fprintf(stderr, "Error: Couldn't open /proc entry file: %s\n",
151 if (fprintf(fp, "%d", id) < 0) {
152 fprintf(stderr, "Error: Couldn't write id to tracking /proc"
153 "entry file: %s\n", stats_path);
157 fprintf(stderr, "Error: Couldn't close tracking /proc entry"
158 " file: %s\n", stats_path);
163 /* Get extra command lines concatenated to "command Function getopt scans
164 * one switch and its optional argument. So if command line is
165 * "-g 1234 make bzImage" 1234 is optarg and "make bzImage" will be collected
166 * with following function to exec it. */
167 char* get_command_from_argv(int optind, int argc, char** argv,char* optarg,
172 strcpy(command, optarg);
173 strcat(command, " ");
174 for (index = optind; index < argc; index++) {
175 strcat(command, argv[index]);
176 strcat(command, " ");
178 if (strlen(command) == 1) {
179 fprintf(stderr,"Error: command missing \n");
182 } else if (strlen(command) > COMM_LEN) {
183 fprintf(stderr,"Error: Too long command \n");
191 /* Check for the llstat command in $PATH env variable. */
196 status = system("which llstat.pl &> /dev/null");
198 fprintf(stderr,"Error: llstat.pl not found in PATH\n");
203 pid_t fork_llstat_command(char* llstat_file,char* stats_path)
205 char truncate_command[100];
206 char llstat_command[LEN_LLSTAT];
207 pid_t pid_llstat_command;
208 FILE *fp_popen, *fp_out;
212 /* Truncating llstat output file as it will be opened in while
213 * loop to append output */
214 sprintf(truncate_command,"> %s",llstat_file);
215 if ((ret = system(truncate_command)) != 0) {
216 ret = WEXITSTATUS(ret);
217 printf("error excuting truncate command: %d\n", ret);
221 strcat(stats_path, "/stats");
223 /* creating a string of llstat command to give to
225 sprintf(llstat_command, "llstat -i 1 -g %s ",
228 /* fork for llstat */
229 if ((pid_llstat_command = fork()) < 0)
230 fprintf(stderr, "Error: Fork error\n");
232 /* in child (llstat) */
233 if (pid_llstat_command == 0) {
234 /* Doing popen for llstat command */
235 fp_popen = popen(llstat_command, "r");
236 if (fp_popen == NULL) {
237 fprintf(stderr,"Couldn't popen the llstat command:"
238 "\"%s\"n", llstat_command);
241 while (fgets(buffer, 1024, fp_popen) != NULL) {
242 /* Following code should be in while loop as llstat
243 * will keep on sending output each second and will
244 * not exit on itself. It will be killed when we finsh
245 * with our command so we must make the output file
246 * consistent after writing each 1024 bytes chunk */
248 /* opening file where llstat will write its output */
249 fp_out = fopen(llstat_file, "a");
251 fprintf(stderr, "Error: Couldn't open llstat"
252 "outfile file: %s\n",
256 /* fgets reads the popen output and fprintf writes it to
259 if (fputs(buffer, fp_out) == EOF) {
260 fprintf(stderr, "Error: Couldn't write output"
261 "of llstat to out file\n");
265 /* closing file opened for storing llstat's output */
266 if (fclose(fp_out)) {
267 fprintf(stderr, "Error: Couldn't close llstat"
268 "outfile: %s\n", llstat_file);
272 /* closing popen for llstat */
273 if (pclose(fp_popen) < 0) {
274 fprintf(stderr, "Error: Couldn't pclos"
275 " llstat popen call\n");
279 return pid_llstat_command;
282 pid_t fork_actual_command(int type, unsigned short id, char* stats_path,
287 /* starting ltrack_stats functionality here */
288 if ((pid = fork()) < 0)
289 fprintf(stderr, "Error: Fork error\n");
291 /* fork for executing command */
295 if (setgid(id) < 0) {
296 fprintf(stderr, "Error: failed to"
308 /* 0 has to be written to vfs_track_pid to collect
309 * statistics of all processes */
315 write_track_xid(type, pid, stats_path);
316 execl("/bin/sh", "sh", "-c", command, (char *)0);
323 char* get_path_stats(int with_llstat, char* stats_path)
325 glob_t stats_glob_buffer;
330 /* No slots reserved in gl_pathv. Store the found path at 0 location */
331 stats_glob_buffer.gl_offs = 0;
333 /* doing glob() for attaching llstat to monitor each vfs_ops_stat for
334 * mulitiple lustre clients */
335 if (glob("/proc/fs/lustre/llite/*", GLOB_DOOFFS, NULL,
336 &stats_glob_buffer) != 0) {
337 fprintf(stderr,"Error: Couldn't find /proc entry for "
342 /* If multiple client entries found in /proc/fs/lustre/llite user will
343 * be prompted with choice of all */
344 if (stats_glob_buffer.gl_pathc > 1 && with_llstat) {
346 printf("Multiple lustre clients found, continuing... \n");
348 /* If flow is here again it means there was an error
349 * and notifying that to user */
352 if ((ret = system("clear")) != 0) {
353 ret = WEXITSTATUS(ret);
354 printf("error excuting clear command: %d\n", ret);
357 fprintf(stderr, "Error: Please give correct "
360 /* Simple menu based interface to avoid possible
361 * spelling mistakes */
362 printf("\t\tMenu.\n");
363 for (i = 0; i < stats_glob_buffer.gl_pathc; i++)
364 printf("\t\t%d. %s\n", i+1,
365 stats_glob_buffer.gl_pathv[i]);
367 printf("\nEnter the lustre client number you want to "
369 if (scanf(" %d", &choice) == EOF && ferror(stdin)) {
370 perror("reading from stdin");
374 } while (choice > stats_glob_buffer.gl_pathc || choice < 1);
375 strcpy(stats_path, stats_glob_buffer.gl_pathv[choice - 1]);
377 /*if only one client then simply copying the path from glob */
378 strcpy(stats_path, stats_glob_buffer.gl_pathv[0]);
380 /* this frees dynamically allocated space by glob() for storing found
382 globfree(&stats_glob_buffer);
387 /* Writes the id (gid/ pid/ ppid) value in appropriate tracking proc entry file
388 * and EXECs the command given */
389 void fork_command(int type, unsigned short id, char* command, char* llstat_file)
391 pid_t pid_actual_command = 0;
392 pid_t pid_llstat_command = 0;
397 char stats_path[1024];
398 char stats_path_temp[1024];
400 if (strlen(llstat_file) == 0)
403 get_path_stats(with_llstat, stats_path);
404 strcpy(stats_path_temp, stats_path);
406 /* llstat process attached to monitor given command */
408 pid_llstat_command = fork_llstat_command(llstat_file,
411 /* forking a process which will exec command given */
412 pid_actual_command = fork_actual_command(type, id, stats_path,
415 if (waitpid(pid_actual_command, NULL, 0) != pid_actual_command)
416 fprintf(stderr, "Error: waitpid error\n");
419 /* comment #25 of BUG 10968 */
422 /* sending kill to all llstat commands created for each
423 * lustre-client respectively */
424 kill(pid_llstat_command, 9);
425 waitpid(pid_llstat_command, &status, 0);
427 /* if llstat child is killed by KILL only then print note for
428 * plotting graph and if its exited normally with errornous
429 * status then it means there were some error and llstat was
431 if (!WIFEXITED(status))
432 printf("\n\t[Note: Do \"$plot-llstat %s\" to plot a graph"
433 " using GNU plot]\n", llstat_file);
439 int main(int argc, char **argv)
441 char gid_string[5] = "";
444 char command[COMM_LEN] = "";
445 char llstat_file[100] = "";
447 /* Checking for root*/
449 fprintf(stderr, "Error: You need to be root\n");
454 /* Parsing command line switches */
455 while ((c = getopt(argc, argv, "l:g:c:i:a:h")) != 1)
458 strcpy(llstat_file, optarg);
459 if (strlen(llstat_file) > LEN_OUT) {
460 fprintf(stderr, "length of outfile file"
466 /* When any value is written to vfs_track_gid, then VFS
467 * operation statistics are collected for all
468 * processes of that group ID.
469 * write_track_xid writes given <gid> in vfs_track_gid
472 strcpy(gid_string, optarg);
473 get_command_from_argv(optind, argc, argv, "",
475 gid = atoi(gid_string);
477 fork_command(TRACK_BY_GID, gid, command,
481 /* When any value is written to vfs_track_ppid, then VFS
482 * operation statistics are collected for all processes
483 * whose parents' PID is same as track_ppid.
484 *- write_track_xid writes pid to vfs_track_ppid here */
486 get_command_from_argv(optind, argc, argv,
488 fork_command(TRACK_BY_PPID, 0, command,
492 /* When a non-zero value is written to vfs_track_pid,
493 * then VFS operation statistics are collected for only
494 * that PID.Write_track_xid writes pid to vfs_track_pid
495 * here.Wei need to FORK a new process and EXEC it with
496 * given <command>. */
498 get_command_from_argv(optind, argc, argv,
500 fork_command(TRACK_BY_PID, 0, command,
504 /* When VFS operation statistics for all processes are
505 * to be collected, "0" is written to vfs_track_pid. */
507 get_command_from_argv(optind, argc, argv,
509 fork_command(TRACK_FOR_ALL, 0, command,