xref: /netbsd-src/sys/kern/kern_acct.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: kern_acct.c,v 1.59 2004/11/13 20:41:41 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_acct.c	8.8 (Berkeley) 5/14/95
37  */
38 
39 /*-
40  * Copyright (c) 1994 Christopher G. Demetriou
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. All advertising materials mentioning features or use of this software
51  *    must display the following acknowledgement:
52  *	This product includes software developed by the University of
53  *	California, Berkeley and its contributors.
54  * 4. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)kern_acct.c	8.8 (Berkeley) 5/14/95
71  */
72 
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: kern_acct.c,v 1.59 2004/11/13 20:41:41 christos Exp $");
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/proc.h>
79 #include <sys/mount.h>
80 #include <sys/vnode.h>
81 #include <sys/file.h>
82 #include <sys/syslog.h>
83 #include <sys/kernel.h>
84 #include <sys/kthread.h>
85 #include <sys/lock.h>
86 #include <sys/malloc.h>
87 #include <sys/namei.h>
88 #include <sys/errno.h>
89 #include <sys/acct.h>
90 #include <sys/resourcevar.h>
91 #include <sys/ioctl.h>
92 #include <sys/tty.h>
93 
94 #include <sys/sa.h>
95 #include <sys/syscallargs.h>
96 
97 /*
98  * The routines implemented in this file are described in:
99  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
100  *	    UNIX Operating System (Addison Welley, 1989)
101  * on pages 62-63.
102  *
103  * Arguably, to simplify accounting operations, this mechanism should
104  * be replaced by one in which an accounting log file (similar to /dev/klog)
105  * is read by a user process, etc.  However, that has its own problems.
106  */
107 
108 /*
109  * The global accounting state and related data.  Gain the lock before
110  * accessing these variables.
111  */
112 enum {
113 	ACCT_STOP,
114 	ACCT_ACTIVE,
115 	ACCT_SUSPENDED
116 } acct_state;				/* The current accounting state. */
117 struct vnode *acct_vp;			/* Accounting vnode pointer. */
118 struct ucred *acct_ucred;		/* Credential of accounting file
119 					   owner (i.e root).  Used when
120  					   accounting file i/o.  */
121 struct proc *acct_dkwatcher;		/* Free disk space checker. */
122 
123 /*
124  * Lock to serialize system calls and kernel threads.
125  */
126 struct	lock acct_lock;
127 #define	ACCT_LOCK()						\
128 do {								\
129 	(void) lockmgr(&acct_lock, LK_EXCLUSIVE, NULL);		\
130 } while (/* CONSTCOND */0)
131 #define	ACCT_UNLOCK()						\
132 do {								\
133 	(void) lockmgr(&acct_lock, LK_RELEASE, NULL);		\
134 } while (/* CONSTCOND */0)
135 
136 /*
137  * Internal accounting functions.
138  * The former's operation is described in Leffler, et al., and the latter
139  * was provided by UCB with the 4.4BSD-Lite release
140  */
141 comp_t	encode_comp_t(u_long, u_long);
142 void	acctwatch(void *);
143 void	acct_stop(void);
144 int	acct_chkfree(void);
145 
146 /*
147  * Values associated with enabling and disabling accounting
148  */
149 int	acctsuspend = 2;	/* stop accounting when < 2% free space left */
150 int	acctresume = 4;		/* resume when free space risen to > 4% */
151 int	acctchkfreq = 15;	/* frequency (in seconds) to check space */
152 
153 void
154 acct_init()
155 {
156 
157 	acct_state = ACCT_STOP;
158 	acct_vp = NULLVP;
159 	acct_ucred = NULL;
160 	lockinit(&acct_lock, PWAIT, "acctlk", 0, 0);
161 }
162 
163 void
164 acct_stop()
165 {
166 	int error;
167 
168 	if (acct_vp != NULLVP && acct_vp->v_type != VBAD) {
169 		error = vn_close(acct_vp, FWRITE, acct_ucred, NULL);
170 #ifdef DIAGNOSTIC
171 		if (error != 0)
172 			printf("acct_stop: failed to close, errno = %d\n",
173 			    error);
174 #endif
175 		acct_vp = NULLVP;
176 	}
177 	if (acct_ucred != NULL) {
178 		crfree(acct_ucred);
179 		acct_ucred = NULL;
180 	}
181 	acct_state = ACCT_STOP;
182 }
183 
184 int
185 acct_chkfree()
186 {
187 	int error;
188 	struct statvfs sb;
189 	int64_t bavail;
190 
191 	error = VFS_STATVFS(acct_vp->v_mount, &sb, NULL);
192 	if (error != 0)
193 		return (error);
194 
195 	bavail = sb.f_bfree - sb.f_bresvd;
196 
197 	switch (acct_state) {
198 	case ACCT_SUSPENDED:
199 		if (bavail > acctresume * sb.f_blocks / 100) {
200 			acct_state = ACCT_ACTIVE;
201 			log(LOG_NOTICE, "Accounting resumed\n");
202 		}
203 		break;
204 	case ACCT_ACTIVE:
205 		if (bavail <= acctsuspend * sb.f_blocks / 100) {
206 			acct_state = ACCT_SUSPENDED;
207 			log(LOG_NOTICE, "Accounting suspended\n");
208 		}
209 		break;
210 	case ACCT_STOP:
211 		break;
212 	}
213 	return (0);
214 }
215 
216 /*
217  * Accounting system call.  Written based on the specification and
218  * previous implementation done by Mark Tinguely.
219  */
220 int
221 sys_acct(l, v, retval)
222 	struct lwp *l;
223 	void *v;
224 	register_t *retval;
225 {
226 	struct sys_acct_args /* {
227 		syscallarg(const char *) path;
228 	} */ *uap = v;
229 	struct nameidata nd;
230 	int error;
231 	struct proc *p = l->l_proc;
232 
233 	/* Make sure that the caller is root. */
234 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
235 		return (error);
236 
237 	/*
238 	 * If accounting is to be started to a file, open that file for
239 	 * writing and make sure it's a 'normal'.
240 	 */
241 	if (SCARG(uap, path) != NULL) {
242 		struct vattr va;
243 		size_t pad;
244 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path),
245 		    p);
246 		if ((error = vn_open(&nd, FWRITE|O_APPEND, 0)) != 0)
247 			return (error);
248 		VOP_UNLOCK(nd.ni_vp, 0);
249 		if (nd.ni_vp->v_type != VREG) {
250 			error = EACCES;
251 			goto bad;
252 		}
253 		if ((error = VOP_GETATTR(nd.ni_vp, &va, p->p_ucred, p)) != 0)
254 			goto bad;
255 
256 		if ((pad = (va.va_size % sizeof(struct acct))) != 0) {
257 #ifdef DIAGNOSTIC
258 			printf("Size of accounting file not a multiple of "
259 			    "%lu - incomplete record truncated\n",
260 			    (unsigned long)sizeof(struct acct));
261 #endif
262 			va.va_size -= pad;
263 			if ((error = VOP_TRUNCATE(nd.ni_vp, va.va_size, 0,
264 			     p->p_ucred, p)) != 0)
265 				goto bad;
266 		}
267 	}
268 
269 	ACCT_LOCK();
270 
271 	/*
272 	 * If accounting was previously enabled, kill the old space-watcher,
273 	 * free credential for accounting file i/o,
274 	 * ... (and, if no new file was specified, leave).
275 	 */
276 	acct_stop();
277 	if (SCARG(uap, path) == NULL)
278 		goto out;
279 
280 	/*
281 	 * Save the new accounting file vnode and credential,
282 	 * and schedule the new free space watcher.
283 	 */
284 	acct_state = ACCT_ACTIVE;
285 	acct_vp = nd.ni_vp;
286 	acct_ucred = p->p_ucred;
287 	crhold(acct_ucred);
288 
289 	error = acct_chkfree();		/* Initial guess. */
290 	if (error != 0) {
291 		acct_stop();
292 		goto out;
293 	}
294 
295 	if (acct_dkwatcher == NULL) {
296 		error = kthread_create1(acctwatch, NULL, &acct_dkwatcher,
297 		    "acctwatch");
298 		if (error != 0)
299 			acct_stop();
300 	}
301 
302  out:
303 	ACCT_UNLOCK();
304 	return (error);
305  bad:
306 	vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
307 	return error;
308 }
309 
310 /*
311  * Write out process accounting information, on process exit.
312  * Data to be written out is specified in Leffler, et al.
313  * and are enumerated below.  (They're also noted in the system
314  * "acct.h" header file.)
315  */
316 int
317 acct_process(p)
318 	struct proc *p;
319 {
320 	struct acct acct;
321 	struct rusage *r;
322 	struct timeval ut, st, tmp;
323 	int s, t, error = 0;
324 	struct plimit *oplim = NULL;
325 
326 	ACCT_LOCK();
327 
328 	/* If accounting isn't enabled, don't bother */
329 	if (acct_state != ACCT_ACTIVE)
330 		goto out;
331 
332 	/*
333 	 * Raise the file limit so that accounting can't be stopped by
334 	 * the user.
335 	 *
336 	 * XXX We should think about the CPU limit, too.
337 	 */
338 	if (p->p_limit->p_refcnt > 1) {
339 		oplim = p->p_limit;
340 		p->p_limit = limcopy(p->p_limit);
341 	}
342 	p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
343 
344 	/*
345 	 * Get process accounting information.
346 	 */
347 
348 	/* (1) The name of the command that ran */
349 	memcpy(acct.ac_comm, p->p_comm, sizeof(acct.ac_comm));
350 
351 	/* (2) The amount of user and system time that was used */
352 	calcru(p, &ut, &st, NULL);
353 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
354 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
355 
356 	/* (3) The elapsed time the commmand ran (and its starting time) */
357 	acct.ac_btime = p->p_stats->p_start.tv_sec;
358 	s = splclock();
359 	timersub(&time, &p->p_stats->p_start, &tmp);
360 	splx(s);
361 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
362 
363 	/* (4) The average amount of memory used */
364 	r = &p->p_stats->p_ru;
365 	timeradd(&ut, &st, &tmp);
366 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
367 	if (t)
368 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
369 	else
370 		acct.ac_mem = 0;
371 
372 	/* (5) The number of disk I/O operations done */
373 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
374 
375 	/* (6) The UID and GID of the process */
376 	acct.ac_uid = p->p_cred->p_ruid;
377 	acct.ac_gid = p->p_cred->p_rgid;
378 
379 	/* (7) The terminal from which the process was started */
380 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
381 		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
382 	else
383 		acct.ac_tty = NODEV;
384 
385 	/* (8) The boolean flags that tell how the process terminated, etc. */
386 	acct.ac_flag = p->p_acflag;
387 
388 	/*
389 	 * Now, just write the accounting information to the file.
390 	 */
391 	VOP_LEASE(acct_vp, p, p->p_ucred, LEASE_WRITE);
392 	error = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct,
393 	    sizeof(acct), (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT,
394 	    acct_ucred, NULL, NULL);
395 	if (error != 0)
396 		log(LOG_ERR, "Accounting: write failed %d\n", error);
397 
398 	if (oplim) {
399 		limfree(p->p_limit);
400 		p->p_limit = oplim;
401 	}
402 
403  out:
404 	ACCT_UNLOCK();
405 	return (error);
406 }
407 
408 /*
409  * Encode_comp_t converts from ticks in seconds and microseconds
410  * to ticks in 1/AHZ seconds.  The encoding is described in
411  * Leffler, et al., on page 63.
412  */
413 
414 #define	MANTSIZE	13			/* 13 bit mantissa. */
415 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
416 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
417 
418 comp_t
419 encode_comp_t(s, us)
420 	u_long s, us;
421 {
422 	int exp, rnd;
423 
424 	exp = 0;
425 	rnd = 0;
426 	s *= AHZ;
427 	s += us / (1000000 / AHZ);	/* Maximize precision. */
428 
429 	while (s > MAXFRACT) {
430 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
431 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
432 		exp++;
433 	}
434 
435 	/* If we need to round up, do it (and handle overflow correctly). */
436 	if (rnd && (++s > MAXFRACT)) {
437 		s >>= EXPSIZE;
438 		exp++;
439 	}
440 
441 	/* Clean it up and polish it off. */
442 	exp <<= MANTSIZE;		/* Shift the exponent into place */
443 	exp += s;			/* and add on the mantissa. */
444 	return (exp);
445 }
446 
447 /*
448  * Periodically check the file system to see if accounting
449  * should be turned on or off.  Beware the case where the vnode
450  * has been vgone()'d out from underneath us, e.g. when the file
451  * system containing the accounting file has been forcibly unmounted.
452  */
453 void
454 acctwatch(arg)
455 	void *arg;
456 {
457 	int error;
458 
459 	log(LOG_NOTICE, "Accounting started\n");
460 	ACCT_LOCK();
461 	while (acct_state != ACCT_STOP) {
462 		if (acct_vp->v_type == VBAD) {
463 			log(LOG_NOTICE, "Accounting terminated\n");
464 			acct_stop();
465 			continue;
466 		}
467 
468 		error = acct_chkfree();
469 #ifdef DIAGNOSTIC
470 		if (error != 0)
471 			printf("acctwatch: failed to statvfs, error = %d\n",
472 			    error);
473 #endif
474 
475 		ACCT_UNLOCK();
476 		error = tsleep(acctwatch, PSWP, "actwat", acctchkfreq * hz);
477 		ACCT_LOCK();
478 #ifdef DIAGNOSTIC
479 		if (error != 0 && error != EWOULDBLOCK)
480 			printf("acctwatch: sleep error %d\n", error);
481 #endif
482 	}
483 	acct_dkwatcher = NULL;
484 	ACCT_UNLOCK();
485 
486 	kthread_exit(0);
487 }
488