xref: /freebsd/sys/kern/kern_acct.c (revision 282a3889ebf826db9839be296ff1dd903f6d6d6e)
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  * 4. 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 "opt_mac.h"
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/acct.h>
78 #include <sys/fcntl.h>
79 #include <sys/kernel.h>
80 #include <sys/kthread.h>
81 #include <sys/lock.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 *);
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 int		 acct_flags;
138 static struct sx	 acct_sx;
139 
140 SX_SYSINIT(acct, &acct_sx, "acct_sx");
141 
142 /*
143  * State of the accounting kthread.
144  */
145 static int		 acct_state;
146 
147 #define	ACCT_RUNNING	1	/* Accounting kthread is running. */
148 #define	ACCT_EXITREQ	2	/* Accounting kthread should exit. */
149 
150 /*
151  * Values associated with enabling and disabling accounting
152  */
153 static int acctsuspend = 2;	/* stop accounting when < 2% free space left */
154 SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW,
155 	&acctsuspend, 0, "percentage of free disk space below which accounting stops");
156 
157 static int acctresume = 4;	/* resume when free space risen to > 4% */
158 SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW,
159 	&acctresume, 0, "percentage of free disk space above which accounting resumes");
160 
161 static int acctchkfreq = 15;	/* frequency (in seconds) to check space */
162 
163 static int
164 sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS)
165 {
166 	int error, value;
167 
168 	/* Write out the old value. */
169 	error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int));
170 	if (error || req->newptr == NULL)
171 		return (error);
172 
173 	/* Read in and verify the new value. */
174 	error = SYSCTL_IN(req, &value, sizeof(int));
175 	if (error)
176 		return (error);
177 	if (value <= 0)
178 		return (EINVAL);
179 	acctchkfreq = value;
180 	return (0);
181 }
182 SYSCTL_PROC(_kern, OID_AUTO, acct_chkfreq, CTLTYPE_INT|CTLFLAG_RW,
183     &acctchkfreq, 0, sysctl_acct_chkfreq, "I",
184     "frequency for checking the free space");
185 
186 SYSCTL_INT(_kern, OID_AUTO, acct_configured, CTLFLAG_RD, &acct_configured, 0,
187 	"Accounting configured or not");
188 
189 SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0,
190 	"Accounting suspended or not");
191 
192 /*
193  * Accounting system call.  Written based on the specification and previous
194  * implementation done by Mark Tinguely.
195  */
196 int
197 acct(struct thread *td, struct acct_args *uap)
198 {
199 	struct nameidata nd;
200 	int error, flags, vfslocked;
201 
202 	error = priv_check(td, PRIV_ACCT);
203 	if (error)
204 		return (error);
205 
206 	/*
207 	 * If accounting is to be started to a file, open that file for
208 	 * appending and make sure it's a 'normal'.
209 	 */
210 	if (uap->path != NULL) {
211 		NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE | AUDITVNODE1,
212 		    UIO_USERSPACE, uap->path, td);
213 		flags = FWRITE | O_APPEND;
214 		error = vn_open(&nd, &flags, 0, NULL);
215 		if (error)
216 			return (error);
217 		vfslocked = NDHASGIANT(&nd);
218 		NDFREE(&nd, NDF_ONLY_PNBUF);
219 #ifdef MAC
220 		error = mac_check_system_acct(td->td_ucred, nd.ni_vp);
221 		if (error) {
222 			VOP_UNLOCK(nd.ni_vp, 0, td);
223 			vn_close(nd.ni_vp, flags, td->td_ucred, td);
224 			VFS_UNLOCK_GIANT(vfslocked);
225 			return (error);
226 		}
227 #endif
228 		VOP_UNLOCK(nd.ni_vp, 0, td);
229 		if (nd.ni_vp->v_type != VREG) {
230 			vn_close(nd.ni_vp, flags, td->td_ucred, td);
231 			VFS_UNLOCK_GIANT(vfslocked);
232 			return (EACCES);
233 		}
234 		VFS_UNLOCK_GIANT(vfslocked);
235 #ifdef MAC
236 	} else {
237 		error = mac_check_system_acct(td->td_ucred, NULL);
238 		if (error)
239 			return (error);
240 #endif
241 	}
242 
243 	/*
244 	 * Disallow concurrent access to the accounting vnode while we swap
245 	 * it out, in order to prevent access after close.
246 	 */
247 	sx_xlock(&acct_sx);
248 
249 	/*
250 	 * If accounting was previously enabled, kill the old space-watcher,
251 	 * close the file, and (if no new file was specified, leave).  Reset
252 	 * the suspended state regardless of whether accounting remains
253 	 * enabled.
254 	 */
255 	acct_suspended = 0;
256 	if (acct_vp != NULL) {
257 		vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
258 		error = acct_disable(td);
259 		VFS_UNLOCK_GIANT(vfslocked);
260 	}
261 	if (uap->path == NULL) {
262 		if (acct_state & ACCT_RUNNING) {
263 			acct_state |= ACCT_EXITREQ;
264 			wakeup(&acct_state);
265 		}
266 		sx_xunlock(&acct_sx);
267 		return (error);
268 	}
269 
270 	/*
271 	 * Save the new accounting file vnode, and schedule the new
272 	 * free space watcher.
273 	 */
274 	acct_vp = nd.ni_vp;
275 	acct_cred = crhold(td->td_ucred);
276 	acct_flags = flags;
277 	if (acct_state & ACCT_RUNNING)
278 		acct_state &= ~ACCT_EXITREQ;
279 	else {
280 		/*
281 		 * Try to start up an accounting kthread.  We may start more
282 		 * than one, but if so the extras will commit suicide as
283 		 * soon as they start up.
284 		 */
285 		error = kthread_create(acct_thread, NULL, NULL, 0, 0,
286 		    "accounting");
287 		if (error) {
288 			vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
289 			(void) vn_close(acct_vp, acct_flags, acct_cred, td);
290 			VFS_UNLOCK_GIANT(vfslocked);
291 			crfree(acct_cred);
292 			acct_configured = 0;
293 			acct_vp = NULL;
294 			acct_cred = NULL;
295 			acct_flags = 0;
296 			sx_xunlock(&acct_sx);
297 			log(LOG_NOTICE, "Unable to start accounting thread\n");
298 			return (error);
299 		}
300 	}
301 	acct_configured = 1;
302 	sx_xunlock(&acct_sx);
303 	log(LOG_NOTICE, "Accounting enabled\n");
304 	return (error);
305 }
306 
307 /*
308  * Disable currently in-progress accounting by closing the vnode, dropping
309  * our reference to the credential, and clearing the vnode's flags.
310  */
311 static int
312 acct_disable(struct thread *td)
313 {
314 	int error;
315 
316 	sx_assert(&acct_sx, SX_XLOCKED);
317 	error = vn_close(acct_vp, acct_flags, acct_cred, td);
318 	crfree(acct_cred);
319 	acct_configured = 0;
320 	acct_vp = NULL;
321 	acct_cred = NULL;
322 	acct_flags = 0;
323 	log(LOG_NOTICE, "Accounting disabled\n");
324 	return (error);
325 }
326 
327 /*
328  * Write out process accounting information, on process exit.
329  * Data to be written out is specified in Leffler, et al.
330  * and are enumerated below.  (They're also noted in the system
331  * "acct.h" header file.)
332  */
333 int
334 acct_process(struct thread *td)
335 {
336 	struct acctv2 acct;
337 	struct timeval ut, st, tmp;
338 	struct plimit *newlim, *oldlim;
339 	struct proc *p;
340 	struct rusage ru;
341 	int t, ret, vfslocked;
342 
343 	/*
344 	 * Lockless check of accounting condition before doing the hard
345 	 * work.
346 	 */
347 	if (acct_vp == NULL || acct_suspended)
348 		return (0);
349 
350 	sx_slock(&acct_sx);
351 
352 	/*
353 	 * If accounting isn't enabled, don't bother.  Have to check again
354 	 * once we own the lock in case we raced with disabling of accounting
355 	 * by another thread.
356 	 */
357 	if (acct_vp == NULL || acct_suspended) {
358 		sx_sunlock(&acct_sx);
359 		return (0);
360 	}
361 
362 	p = td->td_proc;
363 
364 	/*
365 	 * Get process accounting information.
366 	 */
367 
368 	PROC_LOCK(p);
369 	/* (1) The name of the command that ran */
370 	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
371 
372 	/* (2) The amount of user and system time that was used */
373 	rufetchcalc(p, &ru, &ut, &st);
374 	acct.ac_utime = encode_timeval(ut);
375 	acct.ac_stime = encode_timeval(st);
376 
377 	/* (3) The elapsed time the command ran (and its starting time) */
378 	tmp = boottime;
379 	timevaladd(&tmp, &p->p_stats->p_start);
380 	acct.ac_btime = tmp.tv_sec;
381 	microuptime(&tmp);
382 	timevalsub(&tmp, &p->p_stats->p_start);
383 	acct.ac_etime = encode_timeval(tmp);
384 
385 	/* (4) The average amount of memory used */
386 	tmp = ut;
387 	timevaladd(&tmp, &st);
388 	/* Convert tmp (i.e. u + s) into hz units to match ru_i*. */
389 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
390 	if (t)
391 		acct.ac_mem = encode_long((ru.ru_ixrss + ru.ru_idrss +
392 		    + ru.ru_isrss) / t);
393 	else
394 		acct.ac_mem = 0;
395 
396 	/* (5) The number of disk I/O operations done */
397 	acct.ac_io = encode_long(ru.ru_inblock + ru.ru_oublock);
398 
399 	/* (6) The UID and GID of the process */
400 	acct.ac_uid = p->p_ucred->cr_ruid;
401 	acct.ac_gid = p->p_ucred->cr_rgid;
402 
403 	/* (7) The terminal from which the process was started */
404 	SESS_LOCK(p->p_session);
405 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
406 		acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev);
407 	else
408 		acct.ac_tty = NODEV;
409 	SESS_UNLOCK(p->p_session);
410 
411 	/* (8) The boolean flags that tell how the process terminated, etc. */
412 	acct.ac_flagx = p->p_acflag;
413 	PROC_UNLOCK(p);
414 
415 	/* Setup ancillary structure fields. */
416 	acct.ac_flagx |= ANVER;
417 	acct.ac_zero = 0;
418 	acct.ac_version = 2;
419 	acct.ac_len = acct.ac_len2 = sizeof(acct);
420 
421 	/*
422 	 * Eliminate any file size rlimit.
423 	 */
424 	newlim = lim_alloc();
425 	PROC_LOCK(p);
426 	oldlim = p->p_limit;
427 	lim_copy(newlim, oldlim);
428 	newlim->pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
429 	p->p_limit = newlim;
430 	PROC_UNLOCK(p);
431 	lim_free(oldlim);
432 
433 	/*
434 	 * Write the accounting information to the file.
435 	 */
436 	vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
437 	VOP_LEASE(acct_vp, td, acct_cred, LEASE_WRITE);
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 	    (int *)0, td);
441 	VFS_UNLOCK_GIANT(vfslocked);
442 	sx_sunlock(&acct_sx);
443 	return (ret);
444 }
445 
446 /* FLOAT_CONVERSION_START (Regression testing; don't remove this line.) */
447 
448 /* Convert timevals and longs into IEEE-754 bit patterns. */
449 
450 /* Mantissa mask (MSB is implied, so subtract 1). */
451 #define MANT_MASK ((1 << (FLT_MANT_DIG - 1)) - 1)
452 
453 /*
454  * We calculate integer values to a precision of approximately
455  * 28 bits.
456  * This is high-enough precision to fill the 24 float bits
457  * and low-enough to avoid overflowing the 32 int bits.
458  */
459 #define CALC_BITS 28
460 
461 /* log_2(1000000). */
462 #define LOG2_1M 20
463 
464 /*
465  * Convert the elements of a timeval into a 32-bit word holding
466  * the bits of a IEEE-754 float.
467  * The float value represents the timeval's value in microsecond units.
468  */
469 static uint32_t
470 encode_timeval(struct timeval tv)
471 {
472 	int log2_s;
473 	int val, exp;	/* Unnormalized value and exponent */
474 	int norm_exp;	/* Normalized exponent */
475 	int shift;
476 
477 	/*
478 	 * First calculate value and exponent to about CALC_BITS precision.
479 	 * Note that the following conditionals have been ordered so that
480 	 * the most common cases appear first.
481 	 */
482 	if (tv.tv_sec == 0) {
483 		if (tv.tv_usec == 0)
484 			return (0);
485 		exp = 0;
486 		val = tv.tv_usec;
487 	} else {
488 		/*
489 		 * Calculate the value to a precision of approximately
490 		 * CALC_BITS.
491 		 */
492 		log2_s = fls(tv.tv_sec) - 1;
493 		if (log2_s + LOG2_1M < CALC_BITS) {
494 			exp = 0;
495 			val = 1000000 * tv.tv_sec + tv.tv_usec;
496 		} else {
497 			exp = log2_s + LOG2_1M - CALC_BITS;
498 			val = (unsigned int)(((u_int64_t)1000000 * tv.tv_sec +
499 			    tv.tv_usec) >> exp);
500 		}
501 	}
502 	/* Now normalize and pack the value into an IEEE-754 float. */
503 	norm_exp = fls(val) - 1;
504 	shift = FLT_MANT_DIG - norm_exp - 1;
505 #ifdef ACCT_DEBUG
506 	printf("val=%d exp=%d shift=%d log2(val)=%d\n",
507 	    val, exp, shift, norm_exp);
508 	printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp,
509 	    ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK));
510 #endif
511 	return (((FLT_MAX_EXP - 1 + exp + norm_exp) << (FLT_MANT_DIG - 1)) |
512 	    ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK));
513 }
514 
515 /*
516  * Convert a non-negative long value into the bit pattern of
517  * an IEEE-754 float value.
518  */
519 static uint32_t
520 encode_long(long val)
521 {
522 	int norm_exp;	/* Normalized exponent */
523 	int shift;
524 
525 	KASSERT(val >= 0,  ("encode_long: -ve value %ld", val));
526 	if (val == 0)
527 		return (0);
528 	norm_exp = fls(val) - 1;
529 	shift = FLT_MANT_DIG - norm_exp - 1;
530 #ifdef ACCT_DEBUG
531 	printf("val=%d shift=%d log2(val)=%d\n",
532 	    val, shift, norm_exp);
533 	printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp,
534 	    ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK));
535 #endif
536 	return (((FLT_MAX_EXP - 1 + norm_exp) << (FLT_MANT_DIG - 1)) |
537 	    ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK));
538 }
539 
540 /* FLOAT_CONVERSION_END (Regression testing; don't remove this line.) */
541 
542 /*
543  * Periodically check the filesystem to see if accounting
544  * should be turned on or off.  Beware the case where the vnode
545  * has been vgone()'d out from underneath us, e.g. when the file
546  * system containing the accounting file has been forcibly unmounted.
547  */
548 /* ARGSUSED */
549 static void
550 acctwatch(void)
551 {
552 	struct statfs sb;
553 	int vfslocked;
554 
555 	sx_assert(&acct_sx, SX_XLOCKED);
556 
557 	/*
558 	 * If accounting was disabled before our kthread was scheduled,
559 	 * then acct_vp might be NULL.  If so, just ask our kthread to
560 	 * exit and return.
561 	 */
562 	if (acct_vp == NULL) {
563 		acct_state |= ACCT_EXITREQ;
564 		return;
565 	}
566 
567 	/*
568 	 * If our vnode is no longer valid, tear it down and signal the
569 	 * accounting thread to die.
570 	 */
571 	vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
572 	if (acct_vp->v_type == VBAD) {
573 		(void) acct_disable(NULL);
574 		VFS_UNLOCK_GIANT(vfslocked);
575 		acct_state |= ACCT_EXITREQ;
576 		return;
577 	}
578 
579 	/*
580 	 * Stopping here is better than continuing, maybe it will be VBAD
581 	 * next time around.
582 	 */
583 	if (VFS_STATFS(acct_vp->v_mount, &sb, curthread) < 0) {
584 		VFS_UNLOCK_GIANT(vfslocked);
585 		return;
586 	}
587 	VFS_UNLOCK_GIANT(vfslocked);
588 	if (acct_suspended) {
589 		if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks /
590 		    100)) {
591 			acct_suspended = 0;
592 			log(LOG_NOTICE, "Accounting resumed\n");
593 		}
594 	} else {
595 		if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks /
596 		    100)) {
597 			acct_suspended = 1;
598 			log(LOG_NOTICE, "Accounting suspended\n");
599 		}
600 	}
601 }
602 
603 /*
604  * The main loop for the dedicated kernel thread that periodically calls
605  * acctwatch().
606  */
607 static void
608 acct_thread(void *dummy)
609 {
610 	u_char pri;
611 
612 	/* This is a low-priority kernel thread. */
613 	pri = PRI_MAX_KERN;
614 	thread_lock(curthread);
615 	sched_prio(curthread, pri);
616 	thread_unlock(curthread);
617 
618 	/* If another accounting kthread is already running, just die. */
619 	sx_xlock(&acct_sx);
620 	if (acct_state & ACCT_RUNNING) {
621 		sx_xunlock(&acct_sx);
622 		kthread_exit(0);
623 	}
624 	acct_state |= ACCT_RUNNING;
625 
626 	/* Loop until we are asked to exit. */
627 	while (!(acct_state & ACCT_EXITREQ)) {
628 
629 		/* Perform our periodic checks. */
630 		acctwatch();
631 
632 		/*
633 		 * We check this flag again before sleeping since the
634 		 * acctwatch() might have shut down accounting and asked us
635 		 * to exit.
636 		 */
637 		if (!(acct_state & ACCT_EXITREQ)) {
638 			sx_sleep(&acct_state, &acct_sx, 0, "-",
639 			    acctchkfreq * hz);
640 		}
641 	}
642 
643 	/*
644 	 * Acknowledge the exit request and shutdown.  We clear both the
645 	 * exit request and running flags.
646 	 */
647 	acct_state = 0;
648 	sx_xunlock(&acct_sx);
649 	kthread_exit(0);
650 }
651