xref: /netbsd-src/sys/kern/kern_acct.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d)
1 /*	$NetBSD: kern_acct.c,v 1.90 2010/11/19 06:44:42 dholland 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 1.90 2010/11/19 06:44:42 dholland 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/kmem.h>
86 #include <sys/namei.h>
87 #include <sys/errno.h>
88 #include <sys/acct.h>
89 #include <sys/resourcevar.h>
90 #include <sys/ioctl.h>
91 #include <sys/tty.h>
92 #include <sys/kauth.h>
93 
94 #include <sys/syscallargs.h>
95 
96 /*
97  * The routines implemented in this file are described in:
98  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
99  *	    UNIX Operating System (Addison Welley, 1989)
100  * on pages 62-63.
101  *
102  * Arguably, to simplify accounting operations, this mechanism should
103  * be replaced by one in which an accounting log file (similar to /dev/klog)
104  * is read by a user process, etc.  However, that has its own problems.
105  */
106 
107 /*
108  * Lock to serialize system calls and kernel threads.
109  */
110 krwlock_t	acct_lock;
111 
112 /*
113  * The global accounting state and related data.  Gain the mutex before
114  * accessing these variables.
115  */
116 static enum {
117 	ACCT_STOP,
118 	ACCT_ACTIVE,
119 	ACCT_SUSPENDED
120 } acct_state;				/* The current accounting state. */
121 static struct vnode *acct_vp;		/* Accounting vnode pointer. */
122 static kauth_cred_t acct_cred;		/* Credential of accounting file
123 					   owner (i.e root).  Used when
124  					   accounting file i/o.  */
125 static struct lwp *acct_dkwatcher;	/* Free disk space checker. */
126 
127 /*
128  * Values associated with enabling and disabling accounting
129  */
130 int	acctsuspend = 2;	/* stop accounting when < 2% free space left */
131 int	acctresume = 4;		/* resume when free space risen to > 4% */
132 int	acctchkfreq = 15;	/* frequency (in seconds) to check space */
133 
134 /*
135  * Encode_comp_t converts from ticks in seconds and microseconds
136  * to ticks in 1/AHZ seconds.  The encoding is described in
137  * Leffler, et al., on page 63.
138  */
139 
140 #define	MANTSIZE	13			/* 13 bit mantissa. */
141 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
142 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
143 
144 static comp_t
145 encode_comp_t(u_long s, u_long us)
146 {
147 	int exp, rnd;
148 
149 	exp = 0;
150 	rnd = 0;
151 	s *= AHZ;
152 	s += us / (1000000 / AHZ);	/* Maximize precision. */
153 
154 	while (s > MAXFRACT) {
155 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
156 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
157 		exp++;
158 	}
159 
160 	/* If we need to round up, do it (and handle overflow correctly). */
161 	if (rnd && (++s > MAXFRACT)) {
162 		s >>= EXPSIZE;
163 		exp++;
164 	}
165 
166 	/* Clean it up and polish it off. */
167 	exp <<= MANTSIZE;		/* Shift the exponent into place */
168 	exp += s;			/* and add on the mantissa. */
169 	return (exp);
170 }
171 
172 static int
173 acct_chkfree(void)
174 {
175 	int error;
176 	struct statvfs *sb;
177 	int64_t bavail;
178 
179 	sb = kmem_alloc(sizeof(*sb), KM_SLEEP);
180 	if (sb == NULL)
181 		return (ENOMEM);
182 	error = VFS_STATVFS(acct_vp->v_mount, sb);
183 	if (error != 0) {
184 		kmem_free(sb, sizeof(*sb));
185 		return (error);
186 	}
187 
188 	bavail = sb->f_bfree - sb->f_bresvd;
189 
190 	switch (acct_state) {
191 	case ACCT_SUSPENDED:
192 		if (bavail > acctresume * sb->f_blocks / 100) {
193 			acct_state = ACCT_ACTIVE;
194 			log(LOG_NOTICE, "Accounting resumed\n");
195 		}
196 		break;
197 	case ACCT_ACTIVE:
198 		if (bavail <= acctsuspend * sb->f_blocks / 100) {
199 			acct_state = ACCT_SUSPENDED;
200 			log(LOG_NOTICE, "Accounting suspended\n");
201 		}
202 		break;
203 	case ACCT_STOP:
204 		break;
205 	}
206 	kmem_free(sb, sizeof(*sb));
207 	return (0);
208 }
209 
210 static void
211 acct_stop(void)
212 {
213 	int error;
214 
215 	KASSERT(rw_write_held(&acct_lock));
216 
217 	if (acct_vp != NULLVP && acct_vp->v_type != VBAD) {
218 		error = vn_close(acct_vp, FWRITE, acct_cred);
219 #ifdef DIAGNOSTIC
220 		if (error != 0)
221 			printf("acct_stop: failed to close, errno = %d\n",
222 			    error);
223 #endif
224 		acct_vp = NULLVP;
225 	}
226 	if (acct_cred != NULL) {
227 		kauth_cred_free(acct_cred);
228 		acct_cred = NULL;
229 	}
230 	acct_state = ACCT_STOP;
231 }
232 
233 /*
234  * Periodically check the file system to see if accounting
235  * should be turned on or off.  Beware the case where the vnode
236  * has been vgone()'d out from underneath us, e.g. when the file
237  * system containing the accounting file has been forcibly unmounted.
238  */
239 static void
240 acctwatch(void *arg)
241 {
242 	int error;
243 
244 	log(LOG_NOTICE, "Accounting started\n");
245 	rw_enter(&acct_lock, RW_WRITER);
246 	while (acct_state != ACCT_STOP) {
247 		if (acct_vp->v_type == VBAD) {
248 			log(LOG_NOTICE, "Accounting terminated\n");
249 			acct_stop();
250 			continue;
251 		}
252 
253 		error = acct_chkfree();
254 #ifdef DIAGNOSTIC
255 		if (error != 0)
256 			printf("acctwatch: failed to statvfs, error = %d\n",
257 			    error);
258 #endif
259 		rw_exit(&acct_lock);
260 		error = kpause("actwat", false, acctchkfreq * hz, NULL);
261 		rw_enter(&acct_lock, RW_WRITER);
262 #ifdef DIAGNOSTIC
263 		if (error != 0 && error != EWOULDBLOCK)
264 			printf("acctwatch: sleep error %d\n", error);
265 #endif
266 	}
267 	acct_dkwatcher = NULL;
268 	rw_exit(&acct_lock);
269 
270 	kthread_exit(0);
271 }
272 
273 void
274 acct_init(void)
275 {
276 
277 	acct_state = ACCT_STOP;
278 	acct_vp = NULLVP;
279 	acct_cred = NULL;
280 	rw_init(&acct_lock);
281 }
282 
283 /*
284  * Accounting system call.  Written based on the specification and
285  * previous implementation done by Mark Tinguely.
286  */
287 int
288 sys_acct(struct lwp *l, const struct sys_acct_args *uap, register_t *retval)
289 {
290 	/* {
291 		syscallarg(const char *) path;
292 	} */
293 	struct pathbuf *pb;
294 	struct nameidata nd;
295 	int error;
296 
297 	/* Make sure that the caller is root. */
298 	if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_ACCOUNTING,
299 	    0, NULL, NULL, NULL)))
300 		return (error);
301 
302 	/*
303 	 * If accounting is to be started to a file, open that file for
304 	 * writing and make sure it's a 'normal'.
305 	 */
306 	if (SCARG(uap, path) != NULL) {
307 		struct vattr va;
308 		size_t pad;
309 
310 		error = pathbuf_copyin(SCARG(uap, path), &pb);
311 		if (error) {
312 			return error;
313 		}
314 		NDINIT(&nd, LOOKUP, FOLLOW | TRYEMULROOT, pb);
315 		if ((error = vn_open(&nd, FWRITE|O_APPEND, 0)) != 0) {
316 			pathbuf_destroy(pb);
317 			return error;
318 		}
319 		if (nd.ni_vp->v_type != VREG) {
320 			VOP_UNLOCK(nd.ni_vp);
321 			error = EACCES;
322 			goto bad;
323 		}
324 		if ((error = VOP_GETATTR(nd.ni_vp, &va, l->l_cred)) != 0) {
325 			VOP_UNLOCK(nd.ni_vp);
326 			goto bad;
327 		}
328 
329 		if ((pad = (va.va_size % sizeof(struct acct))) != 0) {
330 			u_quad_t size = va.va_size - pad;
331 #ifdef DIAGNOSTIC
332 			printf("Size of accounting file not a multiple of "
333 			    "%lu - incomplete record truncated\n",
334 			    (unsigned long)sizeof(struct acct));
335 #endif
336 			vattr_null(&va);
337 			va.va_size = size;
338 			error = VOP_SETATTR(nd.ni_vp, &va, l->l_cred);
339 			if (error != 0) {
340 				VOP_UNLOCK(nd.ni_vp);
341 				goto bad;
342 			}
343 		}
344 		VOP_UNLOCK(nd.ni_vp);
345 	}
346 
347 	rw_enter(&acct_lock, RW_WRITER);
348 
349 	/*
350 	 * If accounting was previously enabled, kill the old space-watcher,
351 	 * free credential for accounting file i/o,
352 	 * ... (and, if no new file was specified, leave).
353 	 */
354 	acct_stop();
355 	if (SCARG(uap, path) == NULL)
356 		goto out;
357 
358 	/*
359 	 * Save the new accounting file vnode and credential,
360 	 * and schedule the new free space watcher.
361 	 */
362 	acct_state = ACCT_ACTIVE;
363 	acct_vp = nd.ni_vp;
364 	acct_cred = l->l_cred;
365 	kauth_cred_hold(acct_cred);
366 
367 	pathbuf_destroy(pb);
368 
369 	error = acct_chkfree();		/* Initial guess. */
370 	if (error != 0) {
371 		acct_stop();
372 		goto out;
373 	}
374 
375 	if (acct_dkwatcher == NULL) {
376 		error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
377 		    acctwatch, NULL, &acct_dkwatcher, "acctwatch");
378 		if (error != 0)
379 			acct_stop();
380 	}
381 
382  out:
383 	rw_exit(&acct_lock);
384 	return (error);
385  bad:
386 	vn_close(nd.ni_vp, FWRITE, l->l_cred);
387 	pathbuf_destroy(pb);
388 	return error;
389 }
390 
391 /*
392  * Write out process accounting information, on process exit.
393  * Data to be written out is specified in Leffler, et al.
394  * and are enumerated below.  (They're also noted in the system
395  * "acct.h" header file.)
396  */
397 int
398 acct_process(struct lwp *l)
399 {
400 	struct acct acct;
401 	struct timeval ut, st, tmp;
402 	struct rusage *r;
403 	int t, error = 0;
404 	struct rlimit orlim;
405 	struct proc *p = l->l_proc;
406 
407 	if (acct_state != ACCT_ACTIVE)
408 		return 0;
409 
410 	rw_enter(&acct_lock, RW_READER);
411 
412 	/* If accounting isn't enabled, don't bother */
413 	if (acct_state != ACCT_ACTIVE)
414 		goto out;
415 
416 	/*
417 	 * Temporarily raise the file limit so that accounting can't
418 	 * be stopped by the user.
419 	 *
420 	 * XXX We should think about the CPU limit, too.
421 	 */
422 	lim_privatise(p, false);
423 	orlim = p->p_rlimit[RLIMIT_FSIZE];
424 	/* Set current and max to avoid illegal values */
425 	p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
426 	p->p_rlimit[RLIMIT_FSIZE].rlim_max = RLIM_INFINITY;
427 
428 	/*
429 	 * Get process accounting information.
430 	 */
431 
432 	/* (1) The name of the command that ran */
433 	memcpy(acct.ac_comm, p->p_comm, sizeof(acct.ac_comm));
434 
435 	/* (2) The amount of user and system time that was used */
436 	mutex_enter(p->p_lock);
437 	calcru(p, &ut, &st, NULL, NULL);
438 	mutex_exit(p->p_lock);
439 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
440 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
441 
442 	/* (3) The elapsed time the commmand ran (and its starting time) */
443 	acct.ac_btime = p->p_stats->p_start.tv_sec;
444 	getmicrotime(&tmp);
445 	timersub(&tmp, &p->p_stats->p_start, &tmp);
446 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
447 
448 	/* (4) The average amount of memory used */
449 	r = &p->p_stats->p_ru;
450 	timeradd(&ut, &st, &tmp);
451 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
452 	if (t)
453 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
454 	else
455 		acct.ac_mem = 0;
456 
457 	/* (5) The number of disk I/O operations done */
458 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
459 
460 	/* (6) The UID and GID of the process */
461 	acct.ac_uid = kauth_cred_getuid(l->l_cred);
462 	acct.ac_gid = kauth_cred_getgid(l->l_cred);
463 
464 	/* (7) The terminal from which the process was started */
465 	mutex_enter(proc_lock);
466 	if ((p->p_lflag & PL_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
467 		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
468 	else
469 		acct.ac_tty = NODEV;
470 	mutex_exit(proc_lock);
471 
472 	/* (8) The boolean flags that tell how the process terminated, etc. */
473 	acct.ac_flag = p->p_acflag;
474 
475 	/*
476 	 * Now, just write the accounting information to the file.
477 	 */
478 	error = vn_rdwr(UIO_WRITE, acct_vp, (void *)&acct,
479 	    sizeof(acct), (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT,
480 	    acct_cred, NULL, NULL);
481 	if (error != 0)
482 		log(LOG_ERR, "Accounting: write failed %d\n", error);
483 
484 	/* Restore limit - rather pointless since process is about to exit */
485 	p->p_rlimit[RLIMIT_FSIZE] = orlim;
486 
487  out:
488 	rw_exit(&acct_lock);
489 	return (error);
490 }
491