kern_acct.c revision 71498
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 71498 2001-01-24 00:28:07Z jhb $
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/fcntl.h>
50#include <sys/syslog.h>
51#include <sys/kernel.h>
52#include <sys/sysent.h>
53#include <sys/sysctl.h>
54#include <sys/namei.h>
55#include <sys/acct.h>
56#include <sys/resourcevar.h>
57#include <sys/tty.h>
58
59
60/*
61 * The routines implemented in this file are described in:
62 *      Leffler, et al.: The Design and Implementation of the 4.3BSD
63 *	    UNIX Operating System (Addison Welley, 1989)
64 * on pages 62-63.
65 *
66 * Arguably, to simplify accounting operations, this mechanism should
67 * be replaced by one in which an accounting log file (similar to /dev/klog)
68 * is read by a user process, etc.  However, that has its own problems.
69 */
70
71/*
72 * Internal accounting functions.
73 * The former's operation is described in Leffler, et al., and the latter
74 * was provided by UCB with the 4.4BSD-Lite release
75 */
76static comp_t	encode_comp_t __P((u_long, u_long));
77static void	acctwatch __P((void *));
78
79/*
80 * Accounting callout used for periodic scheduling of acctwatch.
81 */
82static struct	callout acctwatch_callout;
83
84/*
85 * Accounting vnode pointer, and saved vnode pointer.
86 */
87static struct	vnode *acctp;
88static struct	vnode *savacctp;
89
90/*
91 * Values associated with enabling and disabling accounting
92 */
93static int acctsuspend = 2;	/* stop accounting when < 2% free space left */
94SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW,
95	&acctsuspend, 0, "percentage of free disk space below which accounting stops");
96
97static int acctresume = 4;	/* resume when free space risen to > 4% */
98SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW,
99	&acctresume, 0, "percentage of free disk space above which accounting resumes");
100
101static int acctchkfreq = 15;	/* frequency (in seconds) to check space */
102SYSCTL_INT(_kern, OID_AUTO, acct_chkfreq, CTLFLAG_RW,
103	&acctchkfreq, 0, "frequency for checking the free space");
104
105/*
106 * Accounting system call.  Written based on the specification and
107 * previous implementation done by Mark Tinguely.
108 */
109int
110acct(a1, uap)
111	struct proc *a1;
112	struct acct_args /* {
113		syscallarg(char *) path;
114	} */ *uap;
115{
116	struct proc *p = curproc;	/* XXX */
117	struct ucred *uc;
118	struct nameidata nd;
119	int error, flags;
120
121	/* Make sure that the caller is root. */
122	error = suser(p);
123	if (error)
124		return (error);
125
126	/*
127	 * If accounting is to be started to a file, open that file for
128	 * writing and make sure it's a 'normal'.
129	 */
130	if (SCARG(uap, path) != NULL) {
131		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path),
132		       p);
133		flags = FWRITE;
134		error = vn_open(&nd, &flags, 0);
135		if (error)
136			return (error);
137		NDFREE(&nd, NDF_ONLY_PNBUF);
138		VOP_UNLOCK(nd.ni_vp, 0, p);
139		if (nd.ni_vp->v_type != VREG) {
140			PROC_LOCK(p);
141			uc = p->p_ucred;
142			crhold(uc);
143			PROC_UNLOCK(p);
144			vn_close(nd.ni_vp, FWRITE, uc, p);
145			crfree(uc);
146			return (EACCES);
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		PROC_LOCK(p);
157		uc = p->p_ucred;
158		crhold(uc);
159		PROC_UNLOCK(p);
160		error = vn_close((acctp != NULLVP ? acctp : savacctp), FWRITE,
161		    uc, p);
162		crfree(uc);
163		acctp = savacctp = NULLVP;
164	}
165	if (SCARG(uap, path) == NULL)
166		return (error);
167
168	/*
169	 * Save the new accounting file vnode, and schedule the new
170	 * free space watcher.
171	 */
172	acctp = nd.ni_vp;
173	callout_init(&acctwatch_callout, 0);
174	acctwatch(NULL);
175	return (error);
176}
177
178/*
179 * Write out process accounting information, on process exit.
180 * Data to be written out is specified in Leffler, et al.
181 * and are enumerated below.  (They're also noted in the system
182 * "acct.h" header file.)
183 */
184
185int
186acct_process(p)
187	struct proc *p;
188{
189	struct acct acct;
190	struct ucred *uc;
191	struct rusage *r;
192	struct timeval ut, st, tmp;
193	int t, error;
194	struct vnode *vp;
195
196	/* If accounting isn't enabled, don't bother */
197	vp = acctp;
198	if (vp == NULLVP)
199		return (0);
200
201	/*
202	 * Get process accounting information.
203	 */
204
205	/* (1) The name of the command that ran */
206	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
207
208	/* (2) The amount of user and system time that was used */
209	mtx_enter(&sched_lock, MTX_SPIN);
210	calcru(p, &ut, &st, NULL);
211	mtx_exit(&sched_lock, MTX_SPIN);
212	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
213	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
214
215	/* (3) The elapsed time the commmand ran (and its starting time) */
216	acct.ac_btime = p->p_stats->p_start.tv_sec;
217	microtime(&tmp);
218	timevalsub(&tmp, &p->p_stats->p_start);
219	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
220
221	/* (4) The average amount of memory used */
222	r = &p->p_stats->p_ru;
223	tmp = ut;
224	timevaladd(&tmp, &st);
225	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
226	if (t)
227		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
228	else
229		acct.ac_mem = 0;
230
231	/* (5) The number of disk I/O operations done */
232	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
233
234	PROC_LOCK(p);
235	/* (6) The UID and GID of the process */
236	acct.ac_uid = p->p_cred->p_ruid;
237	acct.ac_gid = p->p_cred->p_rgid;
238
239	/* (7) The terminal from which the process was started */
240	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
241		acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev);
242	else
243		acct.ac_tty = NOUDEV;
244
245	/* (8) The boolean flags that tell how the process terminated, etc. */
246	acct.ac_flag = p->p_acflag;
247
248	/*
249	 * Eliminate any file size rlimit.
250	 */
251	mtx_assert(&Giant, MA_OWNED);
252	if (p->p_limit->p_refcnt > 1 &&
253	    (p->p_limit->p_lflags & PL_SHAREMOD) == 0) {
254		p->p_limit->p_refcnt--;
255		p->p_limit = limcopy(p->p_limit);
256	}
257	p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
258
259	/*
260	 * Write the accounting information to the file.
261	 */
262	uc = p->p_ucred;
263	crhold(uc);
264	PROC_UNLOCK(p);
265	VOP_LEASE(vp, p, uc, LEASE_WRITE);
266	error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
267	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, uc, (int *)0, p);
268	crfree(uc);
269	return (error);
270}
271
272/*
273 * Encode_comp_t converts from ticks in seconds and microseconds
274 * to ticks in 1/AHZ seconds.  The encoding is described in
275 * Leffler, et al., on page 63.
276 */
277
278#define	MANTSIZE	13			/* 13 bit mantissa. */
279#define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
280#define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
281
282static comp_t
283encode_comp_t(s, us)
284	u_long s, us;
285{
286	int exp, rnd;
287
288	exp = 0;
289	rnd = 0;
290	s *= AHZ;
291	s += us / (1000000 / AHZ);	/* Maximize precision. */
292
293	while (s > MAXFRACT) {
294	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
295		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
296		exp++;
297	}
298
299	/* If we need to round up, do it (and handle overflow correctly). */
300	if (rnd && (++s > MAXFRACT)) {
301		s >>= EXPSIZE;
302		exp++;
303	}
304
305	/* Clean it up and polish it off. */
306	exp <<= MANTSIZE;		/* Shift the exponent into place */
307	exp += s;			/* and add on the mantissa. */
308	return (exp);
309}
310
311/*
312 * Periodically check the file system to see if accounting
313 * should be turned on or off.  Beware the case where the vnode
314 * has been vgone()'d out from underneath us, e.g. when the file
315 * system containing the accounting file has been forcibly unmounted.
316 */
317/* ARGSUSED */
318static void
319acctwatch(a)
320	void *a;
321{
322	struct statfs sb;
323
324	if (savacctp != NULLVP) {
325		if (savacctp->v_type == VBAD) {
326			(void) vn_close(savacctp, FWRITE, NOCRED, NULL);
327			savacctp = NULLVP;
328			return;
329		}
330		(void)VFS_STATFS(savacctp->v_mount, &sb, (struct proc *)0);
331		if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
332			acctp = savacctp;
333			savacctp = NULLVP;
334			log(LOG_NOTICE, "Accounting resumed\n");
335		}
336	} else {
337		if (acctp == NULLVP)
338			return;
339		if (acctp->v_type == VBAD) {
340			(void) vn_close(acctp, FWRITE, NOCRED, NULL);
341			acctp = NULLVP;
342			return;
343		}
344		(void)VFS_STATFS(acctp->v_mount, &sb, (struct proc *)0);
345		if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
346			savacctp = acctp;
347			acctp = NULLVP;
348			log(LOG_NOTICE, "Accounting suspended\n");
349		}
350	}
351	callout_reset(&acctwatch_callout, acctchkfreq * hz, acctwatch, NULL);
352}
353