xref: /freebsd/sys/kern/kern_acct.c (revision 5773cccf19ef7b97e56c1101aa481c43149224da)
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  * $FreeBSD$
41  */
42 
43 #include "opt_mac.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/lock.h>
48 #include <sys/mutex.h>
49 #include <sys/sysproto.h>
50 #include <sys/proc.h>
51 #include <sys/mac.h>
52 #include <sys/mount.h>
53 #include <sys/vnode.h>
54 #include <sys/fcntl.h>
55 #include <sys/syslog.h>
56 #include <sys/kernel.h>
57 #include <sys/sysent.h>
58 #include <sys/sysctl.h>
59 #include <sys/namei.h>
60 #include <sys/acct.h>
61 #include <sys/resourcevar.h>
62 #include <sys/tty.h>
63 
64 /*
65  * The routines implemented in this file are described in:
66  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
67  *	    UNIX Operating System (Addison Welley, 1989)
68  * on pages 62-63.
69  *
70  * Arguably, to simplify accounting operations, this mechanism should
71  * be replaced by one in which an accounting log file (similar to /dev/klog)
72  * is read by a user process, etc.  However, that has its own problems.
73  */
74 
75 /*
76  * Internal accounting functions.
77  * The former's operation is described in Leffler, et al., and the latter
78  * was provided by UCB with the 4.4BSD-Lite release
79  */
80 static comp_t	encode_comp_t(u_long, u_long);
81 static void	acctwatch(void *);
82 
83 /*
84  * Accounting callout used for periodic scheduling of acctwatch.
85  */
86 static struct	callout acctwatch_callout;
87 
88 /*
89  * Accounting vnode pointer, saved vnode pointer, and flags for each.
90  */
91 static struct	vnode *acctp;
92 static struct	ucred *acctcred;
93 static int	acctflags;
94 static struct	vnode *savacctp;
95 static struct	ucred *savacctcred;
96 static int	savacctflags;
97 
98 static struct mtx	acct_mtx;
99 MTX_SYSINIT(acct, &acct_mtx, "accounting", MTX_DEF);
100 
101 /*
102  * Values associated with enabling and disabling accounting
103  */
104 static int acctsuspend = 2;	/* stop accounting when < 2% free space left */
105 SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW,
106 	&acctsuspend, 0, "percentage of free disk space below which accounting stops");
107 
108 static int acctresume = 4;	/* resume when free space risen to > 4% */
109 SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW,
110 	&acctresume, 0, "percentage of free disk space above which accounting resumes");
111 
112 static int acctchkfreq = 15;	/* frequency (in seconds) to check space */
113 SYSCTL_INT(_kern, OID_AUTO, acct_chkfreq, CTLFLAG_RW,
114 	&acctchkfreq, 0, "frequency for checking the free space");
115 
116 /*
117  * Accounting system call.  Written based on the specification and
118  * previous implementation done by Mark Tinguely.
119  *
120  * MPSAFE
121  */
122 int
123 acct(td, uap)
124 	struct thread *td;
125 	struct acct_args /* {
126 		syscallarg(char *) path;
127 	} */ *uap;
128 {
129 	struct nameidata nd;
130 	int error, flags;
131 
132 	/* Make sure that the caller is root. */
133 	error = suser(td);
134 	if (error)
135 		return (error);
136 
137 	mtx_lock(&Giant);
138 	/*
139 	 * If accounting is to be started to a file, open that file for
140 	 * appending and make sure it's a 'normal'.
141 	 */
142 	if (SCARG(uap, path) != NULL) {
143 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path),
144 		       td);
145 		flags = FWRITE | O_APPEND;
146 		error = vn_open(&nd, &flags, 0);
147 		if (error)
148 			goto done2;
149 		NDFREE(&nd, NDF_ONLY_PNBUF);
150 #ifdef MAC
151 		error = mac_check_system_acct(td->td_ucred, nd.ni_vp);
152 		if (error) {
153 			vn_close(nd.ni_vp, flags, td->td_ucred, td);
154 			goto done2;
155 		}
156 #endif
157 		VOP_UNLOCK(nd.ni_vp, 0, td);
158 		if (nd.ni_vp->v_type != VREG) {
159 			vn_close(nd.ni_vp, flags, td->td_ucred, td);
160 			error = EACCES;
161 			goto done2;
162 		}
163 #ifdef MAC
164 	} else {
165 		error = mac_check_system_acct(td->td_ucred, NULL);
166 		if (error)
167 			goto done2;
168 #endif
169 	}
170 
171 	/*
172 	 * If accounting was previously enabled, kill the old space-watcher,
173 	 * close the file, and (if no new file was specified, leave).
174 	 */
175 
176 	/*
177 	 * XXX arr: Should not hold lock over vnode operation.
178 	 */
179 
180 	mtx_lock(&acct_mtx);
181 	if (acctp != NULLVP || savacctp != NULLVP) {
182 		callout_stop(&acctwatch_callout);
183 		error = vn_close((acctp != NULLVP ? acctp : savacctp),
184 		    (acctp != NULLVP ? acctflags : savacctflags),
185 		    (acctcred != NOCRED ? acctcred : savacctcred), td);
186 		acctp = savacctp = NULLVP;
187 		crfree(acctcred != NOCRED ? acctcred : savacctcred);
188 		acctcred = savacctcred = NOCRED;
189 	}
190 	if (SCARG(uap, path) == NULL) {
191 		mtx_unlock(&acct_mtx);
192 		goto done2;
193 	}
194 
195 	/*
196 	 * Save the new accounting file vnode, and schedule the new
197 	 * free space watcher.
198 	 */
199 	acctp = nd.ni_vp;
200 	acctcred = crhold(td->td_ucred);
201 	acctflags = flags;
202 	callout_init(&acctwatch_callout, 0);
203 	mtx_unlock(&acct_mtx);
204 	acctwatch(NULL);
205 done2:
206 	mtx_unlock(&Giant);
207 	return (error);
208 }
209 
210 /*
211  * Write out process accounting information, on process exit.
212  * Data to be written out is specified in Leffler, et al.
213  * and are enumerated below.  (They're also noted in the system
214  * "acct.h" header file.)
215  */
216 
217 int
218 acct_process(td)
219 	struct thread *td;
220 {
221 	struct proc *p = td->td_proc;
222 	struct acct acct;
223 	struct rusage *r;
224 	struct timeval ut, st, tmp;
225 	int t, ret;
226 	struct vnode *vp;
227 	struct ucred *uc;
228 
229 	mtx_lock(&acct_mtx);
230 
231 	/* If accounting isn't enabled, don't bother */
232 	vp = acctp;
233 	if (vp == NULLVP) {
234 		mtx_unlock(&acct_mtx);
235 		return (0);
236 	}
237 
238 	/*
239 	 * Get process accounting information.
240 	 */
241 
242 	/* (1) The name of the command that ran */
243 	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
244 
245 	/* (2) The amount of user and system time that was used */
246 	mtx_lock_spin(&sched_lock);
247 	calcru(p, &ut, &st, NULL);
248 	mtx_unlock_spin(&sched_lock);
249 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
250 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
251 
252 	/* (3) The elapsed time the command ran (and its starting time) */
253 	acct.ac_btime = p->p_stats->p_start.tv_sec;
254 	microtime(&tmp);
255 	timevalsub(&tmp, &p->p_stats->p_start);
256 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
257 
258 	/* (4) The average amount of memory used */
259 	r = &p->p_stats->p_ru;
260 	tmp = ut;
261 	timevaladd(&tmp, &st);
262 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
263 	if (t)
264 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
265 	else
266 		acct.ac_mem = 0;
267 
268 	/* (5) The number of disk I/O operations done */
269 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
270 
271 	/* (6) The UID and GID of the process */
272 	acct.ac_uid = p->p_ucred->cr_ruid;
273 	acct.ac_gid = p->p_ucred->cr_rgid;
274 
275 	/* (7) The terminal from which the process was started */
276 	PROC_LOCK(p);
277 	SESS_LOCK(p->p_session);
278 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
279 		acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev);
280 	else
281 		acct.ac_tty = NOUDEV;
282 	SESS_UNLOCK(p->p_session);
283 	PROC_UNLOCK(p);
284 
285 	/* (8) The boolean flags that tell how the process terminated, etc. */
286 	acct.ac_flag = p->p_acflag;
287 
288 	/*
289 	 * Eliminate any file size rlimit.
290 	 */
291 	if (p->p_limit->p_refcnt > 1 &&
292 	    (p->p_limit->p_lflags & PL_SHAREMOD) == 0) {
293 		p->p_limit->p_refcnt--;
294 		p->p_limit = limcopy(p->p_limit);
295 	}
296 	p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
297 
298 	/*
299 	 * Write the accounting information to the file.
300 	 */
301 	uc = crhold(acctcred);
302 	vref(vp);
303 	mtx_unlock(&acct_mtx);
304 	VOP_LEASE(vp, td, uc, LEASE_WRITE);
305 	ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
306 	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, uc, NOCRED,
307 	    (int *)0, td);
308 	vrele(vp);
309 	crfree(uc);
310 	return (ret);
311 }
312 
313 /*
314  * Encode_comp_t converts from ticks in seconds and microseconds
315  * to ticks in 1/AHZ seconds.  The encoding is described in
316  * Leffler, et al., on page 63.
317  */
318 
319 #define	MANTSIZE	13			/* 13 bit mantissa. */
320 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
321 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
322 
323 static comp_t
324 encode_comp_t(s, us)
325 	u_long s, us;
326 {
327 	int exp, rnd;
328 
329 	exp = 0;
330 	rnd = 0;
331 	s *= AHZ;
332 	s += us / (1000000 / AHZ);	/* Maximize precision. */
333 
334 	while (s > MAXFRACT) {
335 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
336 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
337 		exp++;
338 	}
339 
340 	/* If we need to round up, do it (and handle overflow correctly). */
341 	if (rnd && (++s > MAXFRACT)) {
342 		s >>= EXPSIZE;
343 		exp++;
344 	}
345 
346 	/* Clean it up and polish it off. */
347 	exp <<= MANTSIZE;		/* Shift the exponent into place */
348 	exp += s;			/* and add on the mantissa. */
349 	return (exp);
350 }
351 
352 /*
353  * Periodically check the filesystem to see if accounting
354  * should be turned on or off.  Beware the case where the vnode
355  * has been vgone()'d out from underneath us, e.g. when the file
356  * system containing the accounting file has been forcibly unmounted.
357  */
358 /* ARGSUSED */
359 static void
360 acctwatch(a)
361 	void *a;
362 {
363 	struct statfs sb;
364 
365 	mtx_lock(&acct_mtx);
366 
367 	/*
368 	 * XXX arr: Need to fix the issue of holding acct_mtx over
369 	 * the below vnode operations.
370 	 */
371 
372 	if (savacctp != NULLVP) {
373 		if (savacctp->v_type == VBAD) {
374 			(void) vn_close(savacctp, savacctflags, savacctcred,
375 			    NULL);
376 			savacctp = NULLVP;
377 			savacctcred = NOCRED;
378 			mtx_unlock(&acct_mtx);
379 			return;
380 		}
381 		(void)VFS_STATFS(savacctp->v_mount, &sb, (struct thread *)0);
382 		if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
383 			acctp = savacctp;
384 			acctcred = savacctcred;
385 			acctflags = savacctflags;
386 			savacctp = NULLVP;
387 			savacctcred = NOCRED;
388 			log(LOG_NOTICE, "Accounting resumed\n");
389 		}
390 	} else {
391 		if (acctp == NULLVP) {
392 			mtx_unlock(&acct_mtx);
393 			return;
394 		}
395 		if (acctp->v_type == VBAD) {
396 			(void) vn_close(acctp, acctflags, acctcred, NULL);
397 			acctp = NULLVP;
398 			crfree(acctcred);
399 			acctcred = NOCRED;
400 			mtx_unlock(&acct_mtx);
401 			return;
402 		}
403 		(void)VFS_STATFS(acctp->v_mount, &sb, (struct thread *)0);
404 		if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
405 			savacctp = acctp;
406 			savacctflags = acctflags;
407 			savacctcred = acctcred;
408 			acctp = NULLVP;
409 			acctcred = NOCRED;
410 			log(LOG_NOTICE, "Accounting suspended\n");
411 		}
412 	}
413 	callout_reset(&acctwatch_callout, acctchkfreq * hz, acctwatch, NULL);
414 	mtx_unlock(&acct_mtx);
415 }
416