xref: /openbsd/sys/kern/kern_acct.c (revision 17df1aa7)
1 /*	$OpenBSD: kern_acct.c,v 1.22 2010/01/14 23:12:11 schwarze 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(struct proc *p, void *v, register_t *retval)
99 {
100 	struct sys_acct_args /* {
101 		syscallarg(const char *) path;
102 	} */ *uap = v;
103 	struct nameidata nd;
104 	int error;
105 
106 	/* Make sure that the caller is root. */
107 	if ((error = suser(p, 0)) != 0)
108 		return (error);
109 
110 	/*
111 	 * If accounting is to be started to a file, open that file for
112 	 * writing and make sure it's 'normal'.
113 	 */
114 	if (SCARG(uap, path) != NULL) {
115 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path),
116 		    p);
117 		if ((error = vn_open(&nd, FWRITE|O_APPEND, 0)) != 0)
118 			return (error);
119 		VOP_UNLOCK(nd.ni_vp, 0, p);
120 		if (nd.ni_vp->v_type != VREG) {
121 			vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
122 			return (EACCES);
123 		}
124 	}
125 
126 	/*
127 	 * If accounting was previously enabled, kill the old space-watcher,
128 	 * close the file, and (if no new file was specified, leave).
129 	 */
130 	if (acctp != NULL || savacctp != NULL) {
131 		wakeup(&acct_proc);
132 		error = vn_close((acctp != NULL ? acctp : savacctp), FWRITE,
133 		    p->p_ucred, p);
134 		acctp = savacctp = NULL;
135 	}
136 	if (SCARG(uap, path) == NULL)
137 		return (0);
138 
139 	/*
140 	 * Save the new accounting file vnode, and schedule the new
141 	 * free space watcher.
142 	 */
143 	acctp = nd.ni_vp;
144 	if ((error = acct_start()) != 0) {
145 		acctp = NULL;
146 		(void)vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
147 		return (error);
148 	}
149 	return (0);
150 }
151 
152 /*
153  * Write out process accounting information, on process exit.
154  * Data to be written out is specified in Leffler, et al.
155  * and are enumerated below.  (They're also noted in the system
156  * "acct.h" header file.)
157  */
158 int
159 acct_process(struct proc *p)
160 {
161 	struct acct acct;
162 	struct rusage *r;
163 	struct timeval ut, st, tmp;
164 	int t;
165 	struct vnode *vp;
166 	struct plimit *oplim = NULL;
167 	int error;
168 
169 	/* If accounting isn't enabled, don't bother */
170 	vp = acctp;
171 	if (vp == NULL)
172 		return (0);
173 
174 	/*
175 	 * Raise the file limit so that accounting can't be stopped by the
176 	 * user. (XXX - we should think about the cpu limit too).
177 	 */
178 	if (p->p_p->ps_limit->p_refcnt > 1) {
179 		oplim = p->p_p->ps_limit;
180 		p->p_p->ps_limit = limcopy(p->p_p->ps_limit);
181 	}
182 	p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
183 
184 	/*
185 	 * Get process accounting information.
186 	 */
187 
188 	/* (1) The name of the command that ran */
189 	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
190 
191 	/* (2) The amount of user and system time that was used */
192 	calcru(p, &ut, &st, NULL);
193 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
194 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
195 
196 	/* (3) The elapsed time the command ran (and its starting time) */
197 	acct.ac_btime = p->p_stats->p_start.tv_sec;
198 	getmicrotime(&tmp);
199 	timersub(&tmp, &p->p_stats->p_start, &tmp);
200 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
201 
202 	/* (4) The average amount of memory used */
203 	r = &p->p_stats->p_ru;
204 	timeradd(&ut, &st, &tmp);
205 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
206 	if (t)
207 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
208 	else
209 		acct.ac_mem = 0;
210 
211 	/* (5) The number of disk I/O operations done */
212 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
213 
214 	/* (6) The UID and GID of the process */
215 	acct.ac_uid = p->p_cred->p_ruid;
216 	acct.ac_gid = p->p_cred->p_rgid;
217 
218 	/* (7) The terminal from which the process was started */
219 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
220 		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
221 	else
222 		acct.ac_tty = NODEV;
223 
224 	/* (8) The boolean flags that tell how the process terminated, etc. */
225 	acct.ac_flag = p->p_acflag;
226 
227 	/*
228 	 * Now, just write the accounting information to the file.
229 	 */
230 	error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
231 	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, p->p_ucred, NULL, p);
232 
233 	if (oplim) {
234 		limfree(p->p_p->ps_limit);
235 		p->p_p->ps_limit = oplim;
236 	}
237 
238 	return error;
239 }
240 
241 /*
242  * Encode_comp_t converts from ticks in seconds and microseconds
243  * to ticks in 1/AHZ seconds.  The encoding is described in
244  * Leffler, et al., on page 63.
245  */
246 
247 #define	MANTSIZE	13			/* 13 bit mantissa. */
248 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
249 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
250 
251 comp_t
252 encode_comp_t(u_long s, u_long us)
253 {
254 	int exp, rnd;
255 
256 	exp = 0;
257 	rnd = 0;
258 	s *= AHZ;
259 	s += us / (1000000 / AHZ);	/* Maximize precision. */
260 
261 	while (s > MAXFRACT) {
262 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
263 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
264 		exp++;
265 	}
266 
267 	/* If we need to round up, do it (and handle overflow correctly). */
268 	if (rnd && (++s > MAXFRACT)) {
269 		s >>= EXPSIZE;
270 		exp++;
271 	}
272 
273 	/* Clean it up and polish it off. */
274 	exp <<= MANTSIZE;		/* Shift the exponent into place */
275 	exp += s;			/* and add on the mantissa. */
276 	return (exp);
277 }
278 
279 int
280 acct_start(void)
281 {
282 	/* Already running. */
283 	if (acct_proc != NULL)
284 		return (0);
285 
286 	return (kthread_create(acct_thread, NULL, &acct_proc, "acct"));
287 }
288 
289 /*
290  * Periodically check the file system to see if accounting
291  * should be turned on or off.  Beware the case where the vnode
292  * has been vgone()'d out from underneath us, e.g. when the file
293  * system containing the accounting file has been forcibly unmounted.
294  */
295 /* ARGSUSED */
296 void
297 acct_thread(void *arg)
298 {
299 	struct statfs sb;
300 	struct proc *p = curproc;
301 
302 	for (;;) {
303 		if (savacctp != NULL) {
304 			if (savacctp->v_type == VBAD) {
305 				(void) vn_close(savacctp, FWRITE, NOCRED, p);
306 				savacctp = NULL;
307 				break;
308 			}
309 			(void)VFS_STATFS(savacctp->v_mount, &sb, (struct proc *)0);
310 			if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
311 				acctp = savacctp;
312 				savacctp = NULL;
313 				log(LOG_NOTICE, "Accounting resumed\n");
314 			}
315 		} else if (acctp != NULL) {
316 			if (acctp->v_type == VBAD) {
317 				(void) vn_close(acctp, FWRITE, NOCRED, p);
318 				acctp = NULL;
319 				break;
320 			}
321 			(void)VFS_STATFS(acctp->v_mount, &sb, (struct proc *)0);
322 			if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
323 				savacctp = acctp;
324 				acctp = NULL;
325 				log(LOG_NOTICE, "Accounting suspended\n");
326 			}
327 		} else {
328 			break;
329 		}
330 		tsleep(&acct_proc, PPAUSE, "acct", acctchkfreq *hz);
331 	}
332 	acct_proc = NULL;
333 	kthread_exit(0);
334 }
335 
336 void
337 acct_shutdown(void)
338 {
339 
340 	struct proc *p = curproc;
341 
342 	if (acctp != NULL || savacctp != NULL) {
343 		vn_close((acctp != NULL ? acctp : savacctp), FWRITE,
344 		    NOCRED, p);
345 		acctp = savacctp = NULL;
346 	}
347 }
348