xref: /openbsd-src/sys/kern/kern_acct.c (revision 11efff7f3ac2b3cfeff0c0cddc14294d9b3aca4f)
1 /*	$OpenBSD: kern_acct.c,v 1.15 2004/06/24 19:35:24 tholo Exp $	*/
2 /*	$NetBSD: kern_acct.c,v 1.42 1996/02/04 02:15:12 christos Exp $	*/
3 
4 /*-
5  * Copyright (c) 1994 Christopher G. Demetriou
6  * Copyright (c) 1982, 1986, 1989, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_acct.c	8.1 (Berkeley) 6/14/93
39  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/proc.h>
44 #include <sys/mount.h>
45 #include <sys/vnode.h>
46 #include <sys/file.h>
47 #include <sys/syslog.h>
48 #include <sys/kernel.h>
49 #include <sys/namei.h>
50 #include <sys/errno.h>
51 #include <sys/acct.h>
52 #include <sys/resourcevar.h>
53 #include <sys/ioctl.h>
54 #include <sys/tty.h>
55 #include <sys/kthread.h>
56 
57 #include <sys/syscallargs.h>
58 
59 /*
60  * The routines implemented in this file are described in:
61  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
62  *	    UNIX Operating System (Addison Welley, 1989)
63  * on pages 62-63.
64  *
65  * Arguably, to simplify accounting operations, this mechanism should
66  * be replaced by one in which an accounting log file (similar to /dev/klog)
67  * is read by a user process, etc.  However, that has its own problems.
68  */
69 
70 /*
71  * Internal accounting functions.
72  */
73 comp_t	encode_comp_t(u_long, u_long);
74 int	acct_start(void);
75 void	acct_thread(void *);
76 void	acct_shutdown(void);
77 
78 /*
79  * Accounting vnode pointer, and saved vnode pointer.
80  */
81 struct	vnode *acctp;
82 struct	vnode *savacctp;
83 
84 /*
85  * Values associated with enabling and disabling accounting
86  */
87 int	acctsuspend = 2;	/* stop accounting when < 2% free space left */
88 int	acctresume = 4;		/* resume when free space risen to > 4% */
89 int	acctchkfreq = 15;	/* frequency (in seconds) to check space */
90 
91 struct proc *acct_proc;
92 
93 /*
94  * Accounting system call.  Written based on the specification and
95  * previous implementation done by Mark Tinguely.
96  */
97 int
98 sys_acct(p, v, retval)
99 	struct proc *p;
100 	void *v;
101 	register_t *retval;
102 {
103 	struct sys_acct_args /* {
104 		syscallarg(const char *) path;
105 	} */ *uap = v;
106 	struct nameidata nd;
107 	int error;
108 
109 	/* Make sure that the caller is root. */
110 	if ((error = suser(p, 0)) != 0)
111 		return (error);
112 
113 	/*
114 	 * If accounting is to be started to a file, open that file for
115 	 * writing and make sure it's a 'normal'.
116 	 */
117 	if (SCARG(uap, path) != NULL) {
118 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path),
119 		    p);
120 		if ((error = vn_open(&nd, FWRITE|O_APPEND, 0)) != 0)
121 			return (error);
122 		VOP_UNLOCK(nd.ni_vp, 0, p);
123 		if (nd.ni_vp->v_type != VREG) {
124 			vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
125 			return (EACCES);
126 		}
127 	}
128 
129 	/*
130 	 * If accounting was previously enabled, kill the old space-watcher,
131 	 * close the file, and (if no new file was specified, leave).
132 	 */
133 	if (acctp != NULL || savacctp != NULL) {
134 		wakeup(&acct_proc);
135 		error = vn_close((acctp != NULL ? acctp : savacctp), FWRITE,
136 		    p->p_ucred, p);
137 		acctp = savacctp = NULL;
138 	}
139 	if (SCARG(uap, path) == NULL)
140 		return (0);
141 
142 	/*
143 	 * Save the new accounting file vnode, and schedule the new
144 	 * free space watcher.
145 	 */
146 	acctp = nd.ni_vp;
147 	if ((error = acct_start()) != 0) {
148 		acctp = NULL;
149 		(void)vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
150 		return (error);
151 	}
152 	return (0);
153 }
154 
155 /*
156  * Write out process accounting information, on process exit.
157  * Data to be written out is specified in Leffler, et al.
158  * and are enumerated below.  (They're also noted in the system
159  * "acct.h" header file.)
160  */
161 int
162 acct_process(struct proc *p)
163 {
164 	struct acct acct;
165 	struct rusage *r;
166 	struct timeval ut, st, tmp;
167 	int t;
168 	struct vnode *vp;
169 	struct plimit *oplim = NULL;
170 	int error;
171 
172 	/* If accounting isn't enabled, don't bother */
173 	vp = acctp;
174 	if (vp == NULL)
175 		return (0);
176 
177 	/*
178 	 * Raise the file limit so that accounting can't be stopped by the
179 	 * user. (XXX - we should think about the cpu limit too).
180 	 */
181 	if (p->p_limit->p_refcnt > 1) {
182 		oplim = p->p_limit;
183 		p->p_limit = limcopy(p->p_limit);
184 	}
185 	p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
186 
187 	/*
188 	 * Get process accounting information.
189 	 */
190 
191 	/* (1) The name of the command that ran */
192 	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
193 
194 	/* (2) The amount of user and system time that was used */
195 	calcru(p, &ut, &st, NULL);
196 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
197 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
198 
199 	/* (3) The elapsed time the commmand ran (and its starting time) */
200 	acct.ac_btime = p->p_stats->p_start.tv_sec;
201 	getmicrotime(&tmp);
202 	timersub(&tmp, &p->p_stats->p_start, &tmp);
203 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
204 
205 	/* (4) The average amount of memory used */
206 	r = &p->p_stats->p_ru;
207 	timeradd(&ut, &st, &tmp);
208 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
209 	if (t)
210 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
211 	else
212 		acct.ac_mem = 0;
213 
214 	/* (5) The number of disk I/O operations done */
215 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
216 
217 	/* (6) The UID and GID of the process */
218 	acct.ac_uid = p->p_cred->p_ruid;
219 	acct.ac_gid = p->p_cred->p_rgid;
220 
221 	/* (7) The terminal from which the process was started */
222 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
223 		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
224 	else
225 		acct.ac_tty = NODEV;
226 
227 	/* (8) The boolean flags that tell how the process terminated, etc. */
228 	acct.ac_flag = p->p_acflag;
229 
230 	/*
231 	 * Now, just write the accounting information to the file.
232 	 */
233 	VOP_LEASE(vp, p, p->p_ucred, LEASE_WRITE);
234 	error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
235 	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, p->p_ucred, NULL, p);
236 
237 	if (oplim) {
238 		limfree(p->p_limit);
239 		p->p_limit = oplim;
240 	}
241 
242 	return error;
243 }
244 
245 /*
246  * Encode_comp_t converts from ticks in seconds and microseconds
247  * to ticks in 1/AHZ seconds.  The encoding is described in
248  * Leffler, et al., on page 63.
249  */
250 
251 #define	MANTSIZE	13			/* 13 bit mantissa. */
252 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
253 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
254 
255 comp_t
256 encode_comp_t(u_long s, u_long us)
257 {
258 	int exp, rnd;
259 
260 	exp = 0;
261 	rnd = 0;
262 	s *= AHZ;
263 	s += us / (1000000 / AHZ);	/* Maximize precision. */
264 
265 	while (s > MAXFRACT) {
266 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
267 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
268 		exp++;
269 	}
270 
271 	/* If we need to round up, do it (and handle overflow correctly). */
272 	if (rnd && (++s > MAXFRACT)) {
273 		s >>= EXPSIZE;
274 		exp++;
275 	}
276 
277 	/* Clean it up and polish it off. */
278 	exp <<= MANTSIZE;		/* Shift the exponent into place */
279 	exp += s;			/* and add on the mantissa. */
280 	return (exp);
281 }
282 
283 int
284 acct_start(void)
285 {
286 	/* Already running. */
287 	if (acct_proc != NULL)
288 		return (0);
289 
290 	return (kthread_create(acct_thread, NULL, &acct_proc, "acct"));
291 }
292 
293 /*
294  * Periodically check the file system to see if accounting
295  * should be turned on or off.  Beware the case where the vnode
296  * has been vgone()'d out from underneath us, e.g. when the file
297  * system containing the accounting file has been forcibly unmounted.
298  */
299 /* ARGSUSED */
300 void
301 acct_thread(void *arg)
302 {
303 	struct statfs sb;
304 
305 	for (;;) {
306 		if (savacctp != NULL) {
307 			if (savacctp->v_type == VBAD) {
308 				(void) vn_close(savacctp, FWRITE, NOCRED, NULL);
309 				savacctp = NULL;
310 				return;
311 			}
312 			(void)VFS_STATFS(savacctp->v_mount, &sb, (struct proc *)0);
313 			if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
314 				acctp = savacctp;
315 				savacctp = NULL;
316 				log(LOG_NOTICE, "Accounting resumed\n");
317 			}
318 		} else if (acctp != NULL) {
319 			if (acctp->v_type == VBAD) {
320 				(void) vn_close(acctp, FWRITE, NOCRED, NULL);
321 				acctp = NULL;
322 				return;
323 			}
324 			(void)VFS_STATFS(acctp->v_mount, &sb, (struct proc *)0);
325 			if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
326 				savacctp = acctp;
327 				acctp = NULL;
328 				log(LOG_NOTICE, "Accounting suspended\n");
329 			}
330 		} else {
331 			acct_proc = NULL;
332 			kthread_exit(0);
333 		}
334 		tsleep(&acct_proc, PPAUSE, "acct", acctchkfreq *hz);
335 	}
336 }
337 
338 void
339 acct_shutdown(void)
340 {
341 
342 	if (acctp != NULL || savacctp != NULL) {
343 		vn_close((acctp != NULL ? acctp : savacctp), FWRITE,
344 		    NOCRED, NULL);
345 		acctp = savacctp = NULL;
346 	}
347 }
348