kern_acct.c revision 83366
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 * $FreeBSD: head/sys/kern/kern_acct.c 83366 2001-09-12 08:38:13Z julian $
41 */
42
43#include <sys/param.h>
44#include <sys/systm.h>
45#include <sys/lock.h>
46#include <sys/mutex.h>
47#include <sys/sysproto.h>
48#include <sys/proc.h>
49#include <sys/mount.h>
50#include <sys/vnode.h>
51#include <sys/fcntl.h>
52#include <sys/syslog.h>
53#include <sys/kernel.h>
54#include <sys/sysent.h>
55#include <sys/sysctl.h>
56#include <sys/namei.h>
57#include <sys/acct.h>
58#include <sys/resourcevar.h>
59#include <sys/tty.h>
60
61/*
62 * The routines implemented in this file are described in:
63 *      Leffler, et al.: The Design and Implementation of the 4.3BSD
64 *	    UNIX Operating System (Addison Welley, 1989)
65 * on pages 62-63.
66 *
67 * Arguably, to simplify accounting operations, this mechanism should
68 * be replaced by one in which an accounting log file (similar to /dev/klog)
69 * is read by a user process, etc.  However, that has its own problems.
70 */
71
72/*
73 * Internal accounting functions.
74 * The former's operation is described in Leffler, et al., and the latter
75 * was provided by UCB with the 4.4BSD-Lite release
76 */
77static comp_t	encode_comp_t __P((u_long, u_long));
78static void	acctwatch __P((void *));
79
80/*
81 * Accounting callout used for periodic scheduling of acctwatch.
82 */
83static struct	callout acctwatch_callout;
84
85/*
86 * Accounting vnode pointer, and saved vnode pointer.
87 */
88static struct	vnode *acctp;
89static struct	vnode *savacctp;
90
91/*
92 * Values associated with enabling and disabling accounting
93 */
94static int acctsuspend = 2;	/* stop accounting when < 2% free space left */
95SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW,
96	&acctsuspend, 0, "percentage of free disk space below which accounting stops");
97
98static int acctresume = 4;	/* resume when free space risen to > 4% */
99SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW,
100	&acctresume, 0, "percentage of free disk space above which accounting resumes");
101
102static int acctchkfreq = 15;	/* frequency (in seconds) to check space */
103SYSCTL_INT(_kern, OID_AUTO, acct_chkfreq, CTLFLAG_RW,
104	&acctchkfreq, 0, "frequency for checking the free space");
105
106/*
107 * Accounting system call.  Written based on the specification and
108 * previous implementation done by Mark Tinguely.
109 *
110 * MPSAFE
111 */
112int
113acct(td, uap)
114	struct thread *td;
115	struct acct_args /* {
116		syscallarg(char *) path;
117	} */ *uap;
118{
119	struct nameidata nd;
120	int error, flags;
121
122	mtx_lock(&Giant);
123	if (td != curthread)
124		panic("acct");		/* XXXKSE DIAGNOSTIC */
125	/* Make sure that the caller is root. */
126	error = suser(td->td_proc);
127	if (error)
128		goto done2;
129
130	/*
131	 * If accounting is to be started to a file, open that file for
132	 * writing and make sure it's a 'normal'.
133	 */
134	if (SCARG(uap, path) != NULL) {
135		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path),
136		       td);
137		flags = FWRITE;
138		error = vn_open(&nd, &flags, 0);
139		if (error)
140			goto done2;
141		NDFREE(&nd, NDF_ONLY_PNBUF);
142		VOP_UNLOCK(nd.ni_vp, 0, td);
143		if (nd.ni_vp->v_type != VREG) {
144			vn_close(nd.ni_vp, FWRITE, td->td_proc->p_ucred, td);
145			error = EACCES;
146			goto done2;
147		}
148	}
149
150	/*
151	 * If accounting was previously enabled, kill the old space-watcher,
152	 * close the file, and (if no new file was specified, leave).
153	 */
154	if (acctp != NULLVP || savacctp != NULLVP) {
155		callout_stop(&acctwatch_callout);
156		error = vn_close((acctp != NULLVP ? acctp : savacctp), FWRITE,
157		    td->td_proc->p_ucred, td);
158		acctp = savacctp = NULLVP;
159	}
160	if (SCARG(uap, path) == NULL)
161		goto done2;
162
163	/*
164	 * Save the new accounting file vnode, and schedule the new
165	 * free space watcher.
166	 */
167	acctp = nd.ni_vp;
168	callout_init(&acctwatch_callout, 0);
169	acctwatch(NULL);
170done2:
171	mtx_unlock(&Giant);
172	return (error);
173}
174
175/*
176 * Write out process accounting information, on process exit.
177 * Data to be written out is specified in Leffler, et al.
178 * and are enumerated below.  (They're also noted in the system
179 * "acct.h" header file.)
180 */
181
182int
183acct_process(td)
184	struct thread *td;
185{
186	struct proc *p = td->td_proc;
187	struct acct acct;
188	struct rusage *r;
189	struct timeval ut, st, tmp;
190	int t;
191	struct vnode *vp;
192
193	/* If accounting isn't enabled, don't bother */
194	vp = acctp;
195	if (vp == NULLVP)
196		return (0);
197
198	/*
199	 * Get process accounting information.
200	 */
201
202	/* (1) The name of the command that ran */
203	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
204
205	/* (2) The amount of user and system time that was used */
206	mtx_lock_spin(&sched_lock);
207	calcru(p, &ut, &st, NULL);
208	mtx_unlock_spin(&sched_lock);
209	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
210	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
211
212	/* (3) The elapsed time the commmand ran (and its starting time) */
213	acct.ac_btime = p->p_stats->p_start.tv_sec;
214	microtime(&tmp);
215	timevalsub(&tmp, &p->p_stats->p_start);
216	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
217
218	/* (4) The average amount of memory used */
219	r = &p->p_stats->p_ru;
220	tmp = ut;
221	timevaladd(&tmp, &st);
222	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
223	if (t)
224		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
225	else
226		acct.ac_mem = 0;
227
228	/* (5) The number of disk I/O operations done */
229	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
230
231	/* (6) The UID and GID of the process */
232	acct.ac_uid = p->p_ucred->cr_ruid;
233	acct.ac_gid = p->p_ucred->cr_rgid;
234
235	/* (7) The terminal from which the process was started */
236	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
237		acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev);
238	else
239		acct.ac_tty = NOUDEV;
240
241	/* (8) The boolean flags that tell how the process terminated, etc. */
242	acct.ac_flag = p->p_acflag;
243
244	/*
245	 * Eliminate any file size rlimit.
246	 */
247	if (p->p_limit->p_refcnt > 1 &&
248	    (p->p_limit->p_lflags & PL_SHAREMOD) == 0) {
249		p->p_limit->p_refcnt--;
250		p->p_limit = limcopy(p->p_limit);
251	}
252	p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
253
254	/*
255	 * Write the accounting information to the file.
256	 */
257	VOP_LEASE(vp, td, td->td_proc->p_ucred, LEASE_WRITE);
258	return (vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
259	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, td->td_proc->p_ucred,
260	    (int *)0, td));
261}
262
263/*
264 * Encode_comp_t converts from ticks in seconds and microseconds
265 * to ticks in 1/AHZ seconds.  The encoding is described in
266 * Leffler, et al., on page 63.
267 */
268
269#define	MANTSIZE	13			/* 13 bit mantissa. */
270#define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
271#define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
272
273static comp_t
274encode_comp_t(s, us)
275	u_long s, us;
276{
277	int exp, rnd;
278
279	exp = 0;
280	rnd = 0;
281	s *= AHZ;
282	s += us / (1000000 / AHZ);	/* Maximize precision. */
283
284	while (s > MAXFRACT) {
285	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
286		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
287		exp++;
288	}
289
290	/* If we need to round up, do it (and handle overflow correctly). */
291	if (rnd && (++s > MAXFRACT)) {
292		s >>= EXPSIZE;
293		exp++;
294	}
295
296	/* Clean it up and polish it off. */
297	exp <<= MANTSIZE;		/* Shift the exponent into place */
298	exp += s;			/* and add on the mantissa. */
299	return (exp);
300}
301
302/*
303 * Periodically check the file system to see if accounting
304 * should be turned on or off.  Beware the case where the vnode
305 * has been vgone()'d out from underneath us, e.g. when the file
306 * system containing the accounting file has been forcibly unmounted.
307 */
308/* ARGSUSED */
309static void
310acctwatch(a)
311	void *a;
312{
313	struct statfs sb;
314
315	if (savacctp != NULLVP) {
316		if (savacctp->v_type == VBAD) {
317			(void) vn_close(savacctp, FWRITE, NOCRED, NULL);
318			savacctp = NULLVP;
319			return;
320		}
321		(void)VFS_STATFS(savacctp->v_mount, &sb, (struct thread *)0);
322		if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
323			acctp = savacctp;
324			savacctp = NULLVP;
325			log(LOG_NOTICE, "Accounting resumed\n");
326		}
327	} else {
328		if (acctp == NULLVP)
329			return;
330		if (acctp->v_type == VBAD) {
331			(void) vn_close(acctp, FWRITE, NOCRED, NULL);
332			acctp = NULLVP;
333			return;
334		}
335		(void)VFS_STATFS(acctp->v_mount, &sb, (struct thread *)0);
336		if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
337			savacctp = acctp;
338			acctp = NULLVP;
339			log(LOG_NOTICE, "Accounting suspended\n");
340		}
341	}
342	callout_reset(&acctwatch_callout, acctchkfreq * hz, acctwatch, NULL);
343}
344