kern_acct.c revision 3308
156383Smjacob/*-
247993Sken * Copyright (c) 1994 Christopher G. Demetriou
347993Sken * Copyright (c) 1982, 1986, 1989, 1993
439214Sgibbs *	The Regents of the University of California.  All rights reserved.
539214Sgibbs * (c) UNIX System Laboratories, Inc.
639214Sgibbs * All or some portions of this file are derived from material licensed
756383Smjacob * to the University of California by American Telephone and Telegraph
839214Sgibbs * Co. or Unix System Laboratories, Inc. and are reproduced herein with
956383Smjacob * the permission of UNIX System Laboratories, Inc.
1039214Sgibbs *
1139214Sgibbs * Redistribution and use in source and binary forms, with or without
1239214Sgibbs * modification, are permitted provided that the following conditions
1339214Sgibbs * 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 1.5 1994/09/26 21:09:00 davidg Exp $
41 */
42
43#include <sys/param.h>
44#include <sys/systm.h>
45#include <sys/proc.h>
46#include <sys/mount.h>
47#include <sys/vnode.h>
48#include <sys/file.h>
49#include <sys/syslog.h>
50#include <sys/kernel.h>
51#include <sys/namei.h>
52#include <sys/errno.h>
53#include <sys/acct.h>
54#include <sys/resourcevar.h>
55#include <sys/ioctl.h>
56#include <sys/tty.h>
57
58/*
59 * The routines implemented in this file are described in:
60 *      Leffler, et al.: The Design and Implementation of the 4.3BSD
61 *	    UNIX Operating System (Addison Welley, 1989)
62 * on pages 62-63.
63 *
64 * Arguably, to simplify accounting operations, this mechanism should
65 * be replaced by one in which an accounting log file (similar to /dev/klog)
66 * is read by a user process, etc.  However, that has its own problems.
67 */
68
69/*
70 * Internal accounting functions.
71 * The former's operation is described in Leffler, et al., and the latter
72 * was provided by UCB with the 4.4BSD-Lite release
73 */
74comp_t	encode_comp_t __P((u_long, u_long));
75void	acctwatch __P((void *));
76
77/*
78 * Accounting vnode pointer, and saved vnode pointer.
79 */
80struct	vnode *acctp;
81struct	vnode *savacctp;
82
83/*
84 * Values associated with enabling and disabling accounting
85 */
86int	acctsuspend = 2;	/* stop accounting when < 2% free space left */
87int	acctresume = 4;		/* resume when free space risen to > 4% */
88int	acctchkfreq = 15;	/* frequency (in seconds) to check space */
89
90/*
91 * Accounting system call.  Written based on the specification and
92 * previous implementation done by Mark Tinguely.
93 */
94struct acct_args {
95	char	*path;
96};
97
98int
99acct(p, uap, retval)
100	struct proc *p;
101	struct acct_args *uap;
102	int *retval;
103{
104	struct nameidata nd;
105	int error;
106
107	/* Make sure that the caller is root. */
108	error = suser(p->p_ucred, &p->p_acflag);
109	if (error)
110		return (error);
111
112	/*
113	 * If accounting is to be started to a file, open that file for
114	 * writing and make sure it's a 'normal'.
115	 */
116	if (uap->path != NULL) {
117		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->path, p);
118		error = vn_open(&nd, FWRITE, 0);
119		if (error)
120			return (error);
121		VOP_UNLOCK(nd.ni_vp);
122		if (nd.ni_vp->v_type != VREG) {
123			vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
124			return (EACCES);
125		}
126	}
127
128	/*
129	 * If accounting was previously enabled, kill the old space-watcher,
130	 * close the file, and (if no new file was specified, leave).
131	 */
132	if (acctp != NULLVP || savacctp != NULLVP) {
133		untimeout(acctwatch, NULL);
134		error = vn_close((acctp != NULLVP ? acctp : savacctp), FWRITE,
135		    p->p_ucred, p);
136		acctp = savacctp = NULLVP;
137	}
138	if (uap->path == NULL)
139		return (error);
140
141	/*
142	 * Save the new accounting file vnode, and schedule the new
143	 * free space watcher.
144	 */
145	acctp = nd.ni_vp;
146	acctwatch(NULL);
147	return (error);
148}
149
150/*
151 * Write out process accounting information, on process exit.
152 * Data to be written out is specified in Leffler, et al.
153 * and are enumerated below.  (They're also noted in the system
154 * "acct.h" header file.)
155 */
156
157int
158acct_process(p)
159	struct proc *p;
160{
161	struct acct acct;
162	struct rusage *r;
163	struct timeval ut, st, tmp;
164	int s, t;
165	struct vnode *vp;
166
167	/* If accounting isn't enabled, don't bother */
168	vp = acctp;
169	if (vp == NULLVP)
170		return (0);
171
172	/*
173	 * Get process accounting information.
174	 */
175
176	/* (1) The name of the command that ran */
177	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
178
179	/* (2) The amount of user and system time that was used */
180	calcru(p, &ut, &st, NULL);
181	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
182	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
183
184	/* (3) The elapsed time the commmand ran (and its starting time) */
185	acct.ac_btime = p->p_stats->p_start.tv_sec;
186	s = splclock();
187	tmp = time;
188	splx(s);
189	timevalsub(&tmp, &p->p_stats->p_start);
190	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
191
192	/* (4) The average amount of memory used */
193	r = &p->p_stats->p_ru;
194	tmp = ut;
195	timevaladd(&tmp, &st);
196	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
197	if (t)
198		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
199	else
200		acct.ac_mem = 0;
201
202	/* (5) The number of disk I/O operations done */
203	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
204
205	/* (6) The UID and GID of the process */
206	acct.ac_uid = p->p_cred->p_ruid;
207	acct.ac_gid = p->p_cred->p_rgid;
208
209	/* (7) The terminal from which the process was started */
210	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
211		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
212	else
213		acct.ac_tty = NODEV;
214
215	/* (8) The boolean flags that tell how the process terminated, etc. */
216	acct.ac_flag = p->p_acflag;
217
218	/*
219	 * Now, just write the accounting information to the file.
220	 */
221	LEASE_CHECK(vp, p, p->p_ucred, LEASE_WRITE);
222	return (vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
223	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, p->p_ucred,
224	    (int *)0, p));
225}
226
227/*
228 * Encode_comp_t converts from ticks in seconds and microseconds
229 * to ticks in 1/AHZ seconds.  The encoding is described in
230 * Leffler, et al., on page 63.
231 */
232
233#define	MANTSIZE	13			/* 13 bit mantissa. */
234#define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
235#define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
236
237comp_t
238encode_comp_t(s, us)
239	u_long s, us;
240{
241	int exp, rnd;
242
243	exp = 0;
244	rnd = 0;
245	s *= AHZ;
246	s += us / (1000000 / AHZ);	/* Maximize precision. */
247
248	while (s > MAXFRACT) {
249	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
250		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
251		exp++;
252	}
253
254	/* If we need to round up, do it (and handle overflow correctly). */
255	if (rnd && (++s > MAXFRACT)) {
256		s >>= EXPSIZE;
257		exp++;
258	}
259
260	/* Clean it up and polish it off. */
261	exp <<= MANTSIZE;		/* Shift the exponent into place */
262	exp += s;			/* and add on the mantissa. */
263	return (exp);
264}
265
266/*
267 * Periodically check the file system to see if accounting
268 * should be turned on or off.  Beware the case where the vnode
269 * has been vgone()'d out from underneath us, e.g. when the file
270 * system containing the accounting file has been forcibly unmounted.
271 */
272/* ARGSUSED */
273void
274acctwatch(a)
275	void *a;
276{
277	struct statfs sb;
278
279	if (savacctp != NULLVP) {
280		if (savacctp->v_type == VBAD) {
281			(void) vn_close(savacctp, FWRITE, NOCRED, NULL);
282			savacctp = NULLVP;
283			return;
284		}
285		(void)VFS_STATFS(savacctp->v_mount, &sb, (struct proc *)0);
286		if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
287			acctp = savacctp;
288			savacctp = NULLVP;
289			log(LOG_NOTICE, "Accounting resumed\n");
290		}
291	} else if (acctp != NULLVP) {
292		if (acctp->v_type == VBAD) {
293			(void) vn_close(acctp, FWRITE, NOCRED, NULL);
294			acctp = NULLVP;
295			return;
296		}
297		(void)VFS_STATFS(acctp->v_mount, &sb, (struct proc *)0);
298		if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
299			savacctp = acctp;
300			acctp = NULLVP;
301			log(LOG_NOTICE, "Accounting suspended\n");
302		}
303	} else
304		return;
305	timeout(acctwatch, NULL, acctchkfreq * hz);
306}
307