xref: /freebsd/sys/kern/kern_acct.c (revision 64a0982bee3db2236df43357e70ce8dddbc21d48)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * Copyright (c) 2005 Robert N. M. Watson
6  * All rights reserved.
7  *
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  * Copyright (c) 1994 Christopher G. Demetriou
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. All advertising materials mentioning features or use of this software
48  *    must display the following acknowledgement:
49  *	This product includes software developed by the University of
50  *	California, Berkeley and its contributors.
51  * 4. Neither the name of the University nor the names of its contributors
52  *    may be used to endorse or promote products derived from this software
53  *    without specific prior written permission.
54  *
55  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65  * SUCH DAMAGE.
66  *
67  *	@(#)kern_acct.c	8.1 (Berkeley) 6/14/93
68  */
69 
70 #include <sys/cdefs.h>
71 __FBSDID("$FreeBSD$");
72 
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/acct.h>
76 #include <sys/fcntl.h>
77 #include <sys/kernel.h>
78 #include <sys/kthread.h>
79 #include <sys/limits.h>
80 #include <sys/lock.h>
81 #include <sys/malloc.h>
82 #include <sys/mount.h>
83 #include <sys/mutex.h>
84 #include <sys/namei.h>
85 #include <sys/priv.h>
86 #include <sys/proc.h>
87 #include <sys/resourcevar.h>
88 #include <sys/sched.h>
89 #include <sys/sx.h>
90 #include <sys/sysctl.h>
91 #include <sys/sysent.h>
92 #include <sys/syslog.h>
93 #include <sys/sysproto.h>
94 #include <sys/tty.h>
95 #include <sys/vnode.h>
96 
97 #include <security/mac/mac_framework.h>
98 
99 /*
100  * The routines implemented in this file are described in:
101  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
102  *	    UNIX Operating System (Addison Welley, 1989)
103  * on pages 62-63.
104  * On May 2007 the historic 3 bits base 8 exponent, 13 bit fraction
105  * compt_t representation described in the above reference was replaced
106  * with that of IEEE-754 floats.
107  *
108  * Arguably, to simplify accounting operations, this mechanism should
109  * be replaced by one in which an accounting log file (similar to /dev/klog)
110  * is read by a user process, etc.  However, that has its own problems.
111  */
112 
113 /* Floating point definitions from <float.h>. */
114 #define FLT_MANT_DIG    24              /* p */
115 #define FLT_MAX_EXP     128             /* emax */
116 
117 /*
118  * Internal accounting functions.
119  * The former's operation is described in Leffler, et al., and the latter
120  * was provided by UCB with the 4.4BSD-Lite release
121  */
122 static uint32_t	encode_timeval(struct timeval);
123 static uint32_t	encode_long(long);
124 static void	acctwatch(void);
125 static void	acct_thread(void *);
126 static int	acct_disable(struct thread *, int);
127 
128 /*
129  * Accounting vnode pointer, saved vnode pointer, and flags for each.
130  * acct_sx protects against changes to the active vnode and credentials
131  * while accounting records are being committed to disk.
132  */
133 static int		 acct_configured;
134 static int		 acct_suspended;
135 static struct vnode	*acct_vp;
136 static struct ucred	*acct_cred;
137 static struct plimit	*acct_limit;
138 static int		 acct_flags;
139 static struct sx	 acct_sx;
140 
141 SX_SYSINIT(acct, &acct_sx, "acct_sx");
142 
143 /*
144  * State of the accounting kthread.
145  */
146 static int		 acct_state;
147 
148 #define	ACCT_RUNNING	1	/* Accounting kthread is running. */
149 #define	ACCT_EXITREQ	2	/* Accounting kthread should exit. */
150 
151 /*
152  * Values associated with enabling and disabling accounting
153  */
154 static int acctsuspend = 2;	/* stop accounting when < 2% free space left */
155 SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW,
156 	&acctsuspend, 0, "percentage of free disk space below which accounting stops");
157 
158 static int acctresume = 4;	/* resume when free space risen to > 4% */
159 SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW,
160 	&acctresume, 0, "percentage of free disk space above which accounting resumes");
161 
162 static int acctchkfreq = 15;	/* frequency (in seconds) to check space */
163 
164 static int
165 sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS)
166 {
167 	int error, value;
168 
169 	/* Write out the old value. */
170 	error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int));
171 	if (error || req->newptr == NULL)
172 		return (error);
173 
174 	/* Read in and verify the new value. */
175 	error = SYSCTL_IN(req, &value, sizeof(int));
176 	if (error)
177 		return (error);
178 	if (value <= 0)
179 		return (EINVAL);
180 	acctchkfreq = value;
181 	return (0);
182 }
183 SYSCTL_PROC(_kern, OID_AUTO, acct_chkfreq, CTLTYPE_INT|CTLFLAG_RW,
184     &acctchkfreq, 0, sysctl_acct_chkfreq, "I",
185     "frequency for checking the free space");
186 
187 SYSCTL_INT(_kern, OID_AUTO, acct_configured, CTLFLAG_RD, &acct_configured, 0,
188 	"Accounting configured or not");
189 
190 SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0,
191 	"Accounting suspended or not");
192 
193 /*
194  * Accounting system call.  Written based on the specification and previous
195  * implementation done by Mark Tinguely.
196  */
197 int
198 sys_acct(struct thread *td, struct acct_args *uap)
199 {
200 	struct nameidata nd;
201 	int error, flags, i, replacing;
202 
203 	error = priv_check(td, PRIV_ACCT);
204 	if (error)
205 		return (error);
206 
207 	/*
208 	 * If accounting is to be started to a file, open that file for
209 	 * appending and make sure it's a 'normal'.
210 	 */
211 	if (uap->path != NULL) {
212 		NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1,
213 		    UIO_USERSPACE, uap->path, td);
214 		flags = FWRITE | O_APPEND;
215 		error = vn_open(&nd, &flags, 0, NULL);
216 		if (error)
217 			return (error);
218 		NDFREE(&nd, NDF_ONLY_PNBUF);
219 #ifdef MAC
220 		error = mac_system_check_acct(td->td_ucred, nd.ni_vp);
221 		if (error) {
222 			VOP_UNLOCK(nd.ni_vp, 0);
223 			vn_close(nd.ni_vp, flags, td->td_ucred, td);
224 			return (error);
225 		}
226 #endif
227 		VOP_UNLOCK(nd.ni_vp, 0);
228 		if (nd.ni_vp->v_type != VREG) {
229 			vn_close(nd.ni_vp, flags, td->td_ucred, td);
230 			return (EACCES);
231 		}
232 #ifdef MAC
233 	} else {
234 		error = mac_system_check_acct(td->td_ucred, NULL);
235 		if (error)
236 			return (error);
237 #endif
238 	}
239 
240 	/*
241 	 * Disallow concurrent access to the accounting vnode while we swap
242 	 * it out, in order to prevent access after close.
243 	 */
244 	sx_xlock(&acct_sx);
245 
246 	/*
247 	 * Don't log spurious disable/enable messages if we are
248 	 * switching from one accounting file to another due to log
249 	 * rotation.
250 	 */
251 	replacing = (acct_vp != NULL && uap->path != NULL);
252 
253 	/*
254 	 * If accounting was previously enabled, kill the old space-watcher,
255 	 * close the file, and (if no new file was specified, leave).  Reset
256 	 * the suspended state regardless of whether accounting remains
257 	 * enabled.
258 	 */
259 	acct_suspended = 0;
260 	if (acct_vp != NULL)
261 		error = acct_disable(td, !replacing);
262 	if (uap->path == NULL) {
263 		if (acct_state & ACCT_RUNNING) {
264 			acct_state |= ACCT_EXITREQ;
265 			wakeup(&acct_state);
266 		}
267 		sx_xunlock(&acct_sx);
268 		return (error);
269 	}
270 
271 	/*
272 	 * Create our own plimit object without limits. It will be assigned
273 	 * to exiting processes.
274 	 */
275 	acct_limit = lim_alloc();
276 	for (i = 0; i < RLIM_NLIMITS; i++)
277 		acct_limit->pl_rlimit[i].rlim_cur =
278 		    acct_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
279 
280 	/*
281 	 * Save the new accounting file vnode, and schedule the new
282 	 * free space watcher.
283 	 */
284 	acct_vp = nd.ni_vp;
285 	acct_cred = crhold(td->td_ucred);
286 	acct_flags = flags;
287 	if (acct_state & ACCT_RUNNING)
288 		acct_state &= ~ACCT_EXITREQ;
289 	else {
290 		/*
291 		 * Try to start up an accounting kthread.  We may start more
292 		 * than one, but if so the extras will commit suicide as
293 		 * soon as they start up.
294 		 */
295 		error = kproc_create(acct_thread, NULL, NULL, 0, 0,
296 		    "accounting");
297 		if (error) {
298 			(void) acct_disable(td, 0);
299 			sx_xunlock(&acct_sx);
300 			log(LOG_NOTICE, "Unable to start accounting thread\n");
301 			return (error);
302 		}
303 	}
304 	acct_configured = 1;
305 	sx_xunlock(&acct_sx);
306 	if (!replacing)
307 		log(LOG_NOTICE, "Accounting enabled\n");
308 	return (error);
309 }
310 
311 /*
312  * Disable currently in-progress accounting by closing the vnode, dropping
313  * our reference to the credential, and clearing the vnode's flags.
314  */
315 static int
316 acct_disable(struct thread *td, int logging)
317 {
318 	int error;
319 
320 	sx_assert(&acct_sx, SX_XLOCKED);
321 	error = vn_close(acct_vp, acct_flags, acct_cred, td);
322 	crfree(acct_cred);
323 	lim_free(acct_limit);
324 	acct_configured = 0;
325 	acct_vp = NULL;
326 	acct_cred = NULL;
327 	acct_flags = 0;
328 	if (logging)
329 		log(LOG_NOTICE, "Accounting disabled\n");
330 	return (error);
331 }
332 
333 /*
334  * Write out process accounting information, on process exit.
335  * Data to be written out is specified in Leffler, et al.
336  * and are enumerated below.  (They're also noted in the system
337  * "acct.h" header file.)
338  */
339 int
340 acct_process(struct thread *td)
341 {
342 	struct acctv2 acct;
343 	struct timeval ut, st, tmp;
344 	struct plimit *oldlim;
345 	struct proc *p;
346 	struct rusage ru;
347 	int t, ret;
348 
349 	/*
350 	 * Lockless check of accounting condition before doing the hard
351 	 * work.
352 	 */
353 	if (acct_vp == NULL || acct_suspended)
354 		return (0);
355 
356 	sx_slock(&acct_sx);
357 
358 	/*
359 	 * If accounting isn't enabled, don't bother.  Have to check again
360 	 * once we own the lock in case we raced with disabling of accounting
361 	 * by another thread.
362 	 */
363 	if (acct_vp == NULL || acct_suspended) {
364 		sx_sunlock(&acct_sx);
365 		return (0);
366 	}
367 
368 	p = td->td_proc;
369 
370 	/*
371 	 * Get process accounting information.
372 	 */
373 
374 	sx_slock(&proctree_lock);
375 	PROC_LOCK(p);
376 
377 	/* (1) The terminal from which the process was started */
378 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
379 		acct.ac_tty = tty_udev(p->p_pgrp->pg_session->s_ttyp);
380 	else
381 		acct.ac_tty = NODEV;
382 	sx_sunlock(&proctree_lock);
383 
384 	/* (2) The name of the command that ran */
385 	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
386 
387 	/* (3) The amount of user and system time that was used */
388 	rufetchcalc(p, &ru, &ut, &st);
389 	acct.ac_utime = encode_timeval(ut);
390 	acct.ac_stime = encode_timeval(st);
391 
392 	/* (4) The elapsed time the command ran (and its starting time) */
393 	getboottime(&tmp);
394 	timevaladd(&tmp, &p->p_stats->p_start);
395 	acct.ac_btime = tmp.tv_sec;
396 	microuptime(&tmp);
397 	timevalsub(&tmp, &p->p_stats->p_start);
398 	acct.ac_etime = encode_timeval(tmp);
399 
400 	/* (5) The average amount of memory used */
401 	tmp = ut;
402 	timevaladd(&tmp, &st);
403 	/* Convert tmp (i.e. u + s) into hz units to match ru_i*. */
404 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
405 	if (t)
406 		acct.ac_mem = encode_long((ru.ru_ixrss + ru.ru_idrss +
407 		    + ru.ru_isrss) / t);
408 	else
409 		acct.ac_mem = 0;
410 
411 	/* (6) The number of disk I/O operations done */
412 	acct.ac_io = encode_long(ru.ru_inblock + ru.ru_oublock);
413 
414 	/* (7) The UID and GID of the process */
415 	acct.ac_uid = p->p_ucred->cr_ruid;
416 	acct.ac_gid = p->p_ucred->cr_rgid;
417 
418 	/* (8) The boolean flags that tell how the process terminated, etc. */
419 	acct.ac_flagx = p->p_acflag;
420 
421 	/* Setup ancillary structure fields. */
422 	acct.ac_flagx |= ANVER;
423 	acct.ac_zero = 0;
424 	acct.ac_version = 2;
425 	acct.ac_len = acct.ac_len2 = sizeof(acct);
426 
427 	/*
428 	 * Eliminate rlimits (file size limit in particular).
429 	 */
430 	oldlim = p->p_limit;
431 	p->p_limit = lim_hold(acct_limit);
432 	PROC_UNLOCK(p);
433 	lim_free(oldlim);
434 
435 	/*
436 	 * Write the accounting information to the file.
437 	 */
438 	ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct),
439 	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED,
440 	    NULL, td);
441 	sx_sunlock(&acct_sx);
442 	return (ret);
443 }
444 
445 /* FLOAT_CONVERSION_START (Regression testing; don't remove this line.) */
446 
447 /* Convert timevals and longs into IEEE-754 bit patterns. */
448 
449 /* Mantissa mask (MSB is implied, so subtract 1). */
450 #define MANT_MASK ((1 << (FLT_MANT_DIG - 1)) - 1)
451 
452 /*
453  * We calculate integer values to a precision of approximately
454  * 28 bits.
455  * This is high-enough precision to fill the 24 float bits
456  * and low-enough to avoid overflowing the 32 int bits.
457  */
458 #define CALC_BITS 28
459 
460 /* log_2(1000000). */
461 #define LOG2_1M 20
462 
463 /*
464  * Convert the elements of a timeval into a 32-bit word holding
465  * the bits of a IEEE-754 float.
466  * The float value represents the timeval's value in microsecond units.
467  */
468 static uint32_t
469 encode_timeval(struct timeval tv)
470 {
471 	int log2_s;
472 	int val, exp;	/* Unnormalized value and exponent */
473 	int norm_exp;	/* Normalized exponent */
474 	int shift;
475 
476 	/*
477 	 * First calculate value and exponent to about CALC_BITS precision.
478 	 * Note that the following conditionals have been ordered so that
479 	 * the most common cases appear first.
480 	 */
481 	if (tv.tv_sec == 0) {
482 		if (tv.tv_usec == 0)
483 			return (0);
484 		exp = 0;
485 		val = tv.tv_usec;
486 	} else {
487 		/*
488 		 * Calculate the value to a precision of approximately
489 		 * CALC_BITS.
490 		 */
491 		log2_s = fls(tv.tv_sec) - 1;
492 		if (log2_s + LOG2_1M < CALC_BITS) {
493 			exp = 0;
494 			val = 1000000 * tv.tv_sec + tv.tv_usec;
495 		} else {
496 			exp = log2_s + LOG2_1M - CALC_BITS;
497 			val = (unsigned int)(((uint64_t)1000000 * tv.tv_sec +
498 			    tv.tv_usec) >> exp);
499 		}
500 	}
501 	/* Now normalize and pack the value into an IEEE-754 float. */
502 	norm_exp = fls(val) - 1;
503 	shift = FLT_MANT_DIG - norm_exp - 1;
504 #ifdef ACCT_DEBUG
505 	printf("val=%d exp=%d shift=%d log2(val)=%d\n",
506 	    val, exp, shift, norm_exp);
507 	printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp,
508 	    ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK));
509 #endif
510 	return (((FLT_MAX_EXP - 1 + exp + norm_exp) << (FLT_MANT_DIG - 1)) |
511 	    ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK));
512 }
513 
514 /*
515  * Convert a non-negative long value into the bit pattern of
516  * an IEEE-754 float value.
517  */
518 static uint32_t
519 encode_long(long val)
520 {
521 	int norm_exp;	/* Normalized exponent */
522 	int shift;
523 
524 	if (val == 0)
525 		return (0);
526 	if (val < 0) {
527 		log(LOG_NOTICE,
528 		    "encode_long: negative value %ld in accounting record\n",
529 		    val);
530 		val = LONG_MAX;
531 	}
532 	norm_exp = fls(val) - 1;
533 	shift = FLT_MANT_DIG - norm_exp - 1;
534 #ifdef ACCT_DEBUG
535 	printf("val=%d shift=%d log2(val)=%d\n",
536 	    val, shift, norm_exp);
537 	printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp,
538 	    ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK));
539 #endif
540 	return (((FLT_MAX_EXP - 1 + norm_exp) << (FLT_MANT_DIG - 1)) |
541 	    ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK));
542 }
543 
544 /* FLOAT_CONVERSION_END (Regression testing; don't remove this line.) */
545 
546 /*
547  * Periodically check the filesystem to see if accounting
548  * should be turned on or off.  Beware the case where the vnode
549  * has been vgone()'d out from underneath us, e.g. when the file
550  * system containing the accounting file has been forcibly unmounted.
551  */
552 /* ARGSUSED */
553 static void
554 acctwatch(void)
555 {
556 	struct statfs *sp;
557 
558 	sx_assert(&acct_sx, SX_XLOCKED);
559 
560 	/*
561 	 * If accounting was disabled before our kthread was scheduled,
562 	 * then acct_vp might be NULL.  If so, just ask our kthread to
563 	 * exit and return.
564 	 */
565 	if (acct_vp == NULL) {
566 		acct_state |= ACCT_EXITREQ;
567 		return;
568 	}
569 
570 	/*
571 	 * If our vnode is no longer valid, tear it down and signal the
572 	 * accounting thread to die.
573 	 */
574 	if (acct_vp->v_type == VBAD) {
575 		(void) acct_disable(NULL, 1);
576 		acct_state |= ACCT_EXITREQ;
577 		return;
578 	}
579 
580 	/*
581 	 * Stopping here is better than continuing, maybe it will be VBAD
582 	 * next time around.
583 	 */
584 	sp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
585 	if (VFS_STATFS(acct_vp->v_mount, sp) < 0) {
586 		free(sp, M_STATFS);
587 		return;
588 	}
589 	if (acct_suspended) {
590 		if (sp->f_bavail > (int64_t)(acctresume * sp->f_blocks /
591 		    100)) {
592 			acct_suspended = 0;
593 			log(LOG_NOTICE, "Accounting resumed\n");
594 		}
595 	} else {
596 		if (sp->f_bavail <= (int64_t)(acctsuspend * sp->f_blocks /
597 		    100)) {
598 			acct_suspended = 1;
599 			log(LOG_NOTICE, "Accounting suspended\n");
600 		}
601 	}
602 	free(sp, M_STATFS);
603 }
604 
605 /*
606  * The main loop for the dedicated kernel thread that periodically calls
607  * acctwatch().
608  */
609 static void
610 acct_thread(void *dummy)
611 {
612 	u_char pri;
613 
614 	/* This is a low-priority kernel thread. */
615 	pri = PRI_MAX_KERN;
616 	thread_lock(curthread);
617 	sched_prio(curthread, pri);
618 	thread_unlock(curthread);
619 
620 	/* If another accounting kthread is already running, just die. */
621 	sx_xlock(&acct_sx);
622 	if (acct_state & ACCT_RUNNING) {
623 		sx_xunlock(&acct_sx);
624 		kproc_exit(0);
625 	}
626 	acct_state |= ACCT_RUNNING;
627 
628 	/* Loop until we are asked to exit. */
629 	while (!(acct_state & ACCT_EXITREQ)) {
630 
631 		/* Perform our periodic checks. */
632 		acctwatch();
633 
634 		/*
635 		 * We check this flag again before sleeping since the
636 		 * acctwatch() might have shut down accounting and asked us
637 		 * to exit.
638 		 */
639 		if (!(acct_state & ACCT_EXITREQ)) {
640 			sx_sleep(&acct_state, &acct_sx, 0, "-",
641 			    acctchkfreq * hz);
642 		}
643 	}
644 
645 	/*
646 	 * Acknowledge the exit request and shutdown.  We clear both the
647 	 * exit request and running flags.
648 	 */
649 	acct_state = 0;
650 	sx_xunlock(&acct_sx);
651 	kproc_exit(0);
652 }
653