kern_acct.c revision 12221
1/*-
2 * Copyright (c) 1994 Christopher G. Demetriou
3 * Copyright (c) 1982, 1986, 1989, 1993
4 *	The Regents of the University of California.  All rights reserved.
5 * (c) UNIX System Laboratories, Inc.
6 * All or some portions of this file are derived from material licensed
7 * to the University of California by American Telephone and Telegraph
8 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
9 * the permission of UNIX System Laboratories, Inc.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 *    notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 *    notice, this list of conditions and the following disclaimer in the
18 *    documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 *    must display the following acknowledgement:
21 *	This product includes software developed by the University of
22 *	California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 *    may be used to endorse or promote products derived from this software
25 *    without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * SUCH DAMAGE.
38 *
39 *	@(#)kern_acct.c	8.1 (Berkeley) 6/14/93
40 *	$Id: kern_acct.c,v 1.8 1995/10/29 15:30:56 phk Exp $
41 */
42
43#include <sys/param.h>
44#include <sys/systm.h>
45#include <sys/sysproto.h>
46#include <sys/proc.h>
47#include <sys/mount.h>
48#include <sys/vnode.h>
49#include <sys/file.h>
50#include <sys/syslog.h>
51#include <sys/kernel.h>
52#include <sys/namei.h>
53#include <sys/errno.h>
54#include <sys/acct.h>
55#include <sys/resourcevar.h>
56#include <sys/ioctl.h>
57#include <sys/tty.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 * The former's operation is described in Leffler, et al., and the latter
73 * was provided by UCB with the 4.4BSD-Lite release
74 */
75comp_t	encode_comp_t __P((u_long, u_long));
76void	acctwatch __P((void *));
77
78/*
79 * Accounting vnode pointer, and saved vnode pointer.
80 */
81struct	vnode *acctp;
82struct	vnode *savacctp;
83
84/*
85 * Values associated with enabling and disabling accounting
86 */
87int	acctsuspend = 2;	/* stop accounting when < 2% free space left */
88int	acctresume = 4;		/* resume when free space risen to > 4% */
89int	acctchkfreq = 15;	/* frequency (in seconds) to check space */
90
91/*
92 * Accounting system call.  Written based on the specification and
93 * previous implementation done by Mark Tinguely.
94 */
95#ifndef _SYS_SYSPROTO_H_
96struct acct_args {
97	char	*path;
98};
99
100#endif
101int
102acct(p, uap, retval)
103	struct proc *p;
104	struct acct_args *uap;
105	int *retval;
106{
107	struct nameidata nd;
108	int error;
109
110	/* Make sure that the caller is root. */
111	error = suser(p->p_ucred, &p->p_acflag);
112	if (error)
113		return (error);
114
115	/*
116	 * If accounting is to be started to a file, open that file for
117	 * writing and make sure it's a 'normal'.
118	 */
119	if (uap->path != NULL) {
120		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->path, p);
121		error = vn_open(&nd, FWRITE, 0);
122		if (error)
123			return (error);
124		VOP_UNLOCK(nd.ni_vp);
125		if (nd.ni_vp->v_type != VREG) {
126			vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
127			return (EACCES);
128		}
129	}
130
131	/*
132	 * If accounting was previously enabled, kill the old space-watcher,
133	 * close the file, and (if no new file was specified, leave).
134	 */
135	if (acctp != NULLVP || savacctp != NULLVP) {
136		untimeout(acctwatch, NULL);
137		error = vn_close((acctp != NULLVP ? acctp : savacctp), FWRITE,
138		    p->p_ucred, p);
139		acctp = savacctp = NULLVP;
140	}
141	if (uap->path == NULL)
142		return (error);
143
144	/*
145	 * Save the new accounting file vnode, and schedule the new
146	 * free space watcher.
147	 */
148	acctp = nd.ni_vp;
149	acctwatch(NULL);
150	return (error);
151}
152
153/*
154 * Write out process accounting information, on process exit.
155 * Data to be written out is specified in Leffler, et al.
156 * and are enumerated below.  (They're also noted in the system
157 * "acct.h" header file.)
158 */
159
160int
161acct_process(p)
162	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
170	/* If accounting isn't enabled, don't bother */
171	vp = acctp;
172	if (vp == NULLVP)
173		return (0);
174
175	/*
176	 * Get process accounting information.
177	 */
178
179	/* (1) The name of the command that ran */
180	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
181
182	/* (2) The amount of user and system time that was used */
183	calcru(p, &ut, &st, NULL);
184	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
185	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
186
187	/* (3) The elapsed time the commmand ran (and its starting time) */
188	acct.ac_btime = p->p_stats->p_start.tv_sec;
189	microtime(&tmp);
190	timevalsub(&tmp, &p->p_stats->p_start);
191	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
192
193	/* (4) The average amount of memory used */
194	r = &p->p_stats->p_ru;
195	tmp = ut;
196	timevaladd(&tmp, &st);
197	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
198	if (t)
199		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
200	else
201		acct.ac_mem = 0;
202
203	/* (5) The number of disk I/O operations done */
204	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
205
206	/* (6) The UID and GID of the process */
207	acct.ac_uid = p->p_cred->p_ruid;
208	acct.ac_gid = p->p_cred->p_rgid;
209
210	/* (7) The terminal from which the process was started */
211	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
212		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
213	else
214		acct.ac_tty = NODEV;
215
216	/* (8) The boolean flags that tell how the process terminated, etc. */
217	acct.ac_flag = p->p_acflag;
218
219	/*
220	 * Now, just write the accounting information to the file.
221	 */
222	LEASE_CHECK(vp, p, p->p_ucred, LEASE_WRITE);
223	return (vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
224	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, p->p_ucred,
225	    (int *)0, p));
226}
227
228/*
229 * Encode_comp_t converts from ticks in seconds and microseconds
230 * to ticks in 1/AHZ seconds.  The encoding is described in
231 * Leffler, et al., on page 63.
232 */
233
234#define	MANTSIZE	13			/* 13 bit mantissa. */
235#define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
236#define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
237
238comp_t
239encode_comp_t(s, us)
240	u_long s, us;
241{
242	int exp, rnd;
243
244	exp = 0;
245	rnd = 0;
246	s *= AHZ;
247	s += us / (1000000 / AHZ);	/* Maximize precision. */
248
249	while (s > MAXFRACT) {
250	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
251		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
252		exp++;
253	}
254
255	/* If we need to round up, do it (and handle overflow correctly). */
256	if (rnd && (++s > MAXFRACT)) {
257		s >>= EXPSIZE;
258		exp++;
259	}
260
261	/* Clean it up and polish it off. */
262	exp <<= MANTSIZE;		/* Shift the exponent into place */
263	exp += s;			/* and add on the mantissa. */
264	return (exp);
265}
266
267/*
268 * Periodically check the file system to see if accounting
269 * should be turned on or off.  Beware the case where the vnode
270 * has been vgone()'d out from underneath us, e.g. when the file
271 * system containing the accounting file has been forcibly unmounted.
272 */
273/* ARGSUSED */
274void
275acctwatch(a)
276	void *a;
277{
278	struct statfs sb;
279
280	if (savacctp != NULLVP) {
281		if (savacctp->v_type == VBAD) {
282			(void) vn_close(savacctp, FWRITE, NOCRED, NULL);
283			savacctp = NULLVP;
284			return;
285		}
286		(void)VFS_STATFS(savacctp->v_mount, &sb, (struct proc *)0);
287		if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
288			acctp = savacctp;
289			savacctp = NULLVP;
290			log(LOG_NOTICE, "Accounting resumed\n");
291		}
292	} else if (acctp != NULLVP) {
293		if (acctp->v_type == VBAD) {
294			(void) vn_close(acctp, FWRITE, NOCRED, NULL);
295			acctp = NULLVP;
296			return;
297		}
298		(void)VFS_STATFS(acctp->v_mount, &sb, (struct proc *)0);
299		if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
300			savacctp = acctp;
301			acctp = NULLVP;
302			log(LOG_NOTICE, "Accounting suspended\n");
303		}
304	} else
305		return;
306	timeout(acctwatch, NULL, acctchkfreq * hz);
307}
308