xref: /freebsd/sys/kern/kern_acct.c (revision d056fa046c6a91b90cd98165face0e42a33a5173)
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  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Copyright (c) 1994 Christopher G. Demetriou
11  * Copyright (c) 2005 Robert N. M. Watson
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. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by the University of
24  *	California, Berkeley and its contributors.
25  * 4. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	@(#)kern_acct.c	8.1 (Berkeley) 6/14/93
42  */
43 
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46 
47 #include "opt_mac.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/acct.h>
52 #include <sys/fcntl.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/lock.h>
56 #include <sys/mac.h>
57 #include <sys/mount.h>
58 #include <sys/mutex.h>
59 #include <sys/namei.h>
60 #include <sys/proc.h>
61 #include <sys/resourcevar.h>
62 #include <sys/sched.h>
63 #include <sys/sx.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
66 #include <sys/syslog.h>
67 #include <sys/sysproto.h>
68 #include <sys/tty.h>
69 #include <sys/vnode.h>
70 
71 /*
72  * The routines implemented in this file are described in:
73  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
74  *	    UNIX Operating System (Addison Welley, 1989)
75  * on pages 62-63.
76  *
77  * Arguably, to simplify accounting operations, this mechanism should
78  * be replaced by one in which an accounting log file (similar to /dev/klog)
79  * is read by a user process, etc.  However, that has its own problems.
80  */
81 
82 /*
83  * Internal accounting functions.
84  * The former's operation is described in Leffler, et al., and the latter
85  * was provided by UCB with the 4.4BSD-Lite release
86  */
87 static comp_t	encode_comp_t(u_long, u_long);
88 static void	acctwatch(void);
89 static void	acct_thread(void *);
90 static int	acct_disable(struct thread *);
91 
92 /*
93  * Accounting vnode pointer, saved vnode pointer, and flags for each.
94  * acct_sx protects against changes to the active vnode and credentials
95  * while accounting records are being committed to disk.
96  */
97 static int		 acct_suspended;
98 static struct vnode	*acct_vp;
99 static struct ucred	*acct_cred;
100 static int		 acct_flags;
101 static struct sx	 acct_sx;
102 
103 SX_SYSINIT(acct, &acct_sx, "acct_sx");
104 
105 /*
106  * State of the accounting kthread.
107  */
108 static int		 acct_state;
109 
110 #define	ACCT_RUNNING	1	/* Accounting kthread is running. */
111 #define	ACCT_EXITREQ	2	/* Accounting kthread should exit. */
112 
113 /*
114  * Values associated with enabling and disabling accounting
115  */
116 static int acctsuspend = 2;	/* stop accounting when < 2% free space left */
117 SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW,
118 	&acctsuspend, 0, "percentage of free disk space below which accounting stops");
119 
120 static int acctresume = 4;	/* resume when free space risen to > 4% */
121 SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW,
122 	&acctresume, 0, "percentage of free disk space above which accounting resumes");
123 
124 static int acctchkfreq = 15;	/* frequency (in seconds) to check space */
125 
126 static int
127 sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS)
128 {
129 	int error, value;
130 
131 	/* Write out the old value. */
132 	error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int));
133 	if (error || req->newptr == NULL)
134 		return (error);
135 
136 	/* Read in and verify the new value. */
137 	error = SYSCTL_IN(req, &value, sizeof(int));
138 	if (error)
139 		return (error);
140 	if (value <= 0)
141 		return (EINVAL);
142 	acctchkfreq = value;
143 	return (0);
144 }
145 SYSCTL_PROC(_kern, OID_AUTO, acct_chkfreq, CTLTYPE_INT|CTLFLAG_RW,
146     &acctchkfreq, 0, sysctl_acct_chkfreq, "I",
147     "frequency for checking the free space");
148 
149 SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0,
150 	"Accounting suspended or not");
151 
152 /*
153  * Accounting system call.  Written based on the specification and
154  * previous implementation done by Mark Tinguely.
155  *
156  * MPSAFE
157  */
158 int
159 acct(struct thread *td, struct acct_args *uap)
160 {
161 	struct nameidata nd;
162 	int error, flags, vfslocked;
163 
164 	/* Make sure that the caller is root. */
165 	error = suser(td);
166 	if (error)
167 		return (error);
168 
169 	/*
170 	 * If accounting is to be started to a file, open that file for
171 	 * appending and make sure it's a 'normal'.
172 	 */
173 	if (uap->path != NULL) {
174 		NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE | AUDITVNODE1,
175 		    UIO_USERSPACE, uap->path, td);
176 		flags = FWRITE | O_APPEND;
177 		error = vn_open(&nd, &flags, 0, -1);
178 		if (error)
179 			return (error);
180 		vfslocked = NDHASGIANT(&nd);
181 		NDFREE(&nd, NDF_ONLY_PNBUF);
182 #ifdef MAC
183 		error = mac_check_system_acct(td->td_ucred, nd.ni_vp);
184 		if (error) {
185 			VOP_UNLOCK(nd.ni_vp, 0, td);
186 			vn_close(nd.ni_vp, flags, td->td_ucred, td);
187 			VFS_UNLOCK_GIANT(vfslocked);
188 			return (error);
189 		}
190 #endif
191 		VOP_UNLOCK(nd.ni_vp, 0, td);
192 		if (nd.ni_vp->v_type != VREG) {
193 			vn_close(nd.ni_vp, flags, td->td_ucred, td);
194 			VFS_UNLOCK_GIANT(vfslocked);
195 			return (EACCES);
196 		}
197 		VFS_UNLOCK_GIANT(vfslocked);
198 #ifdef MAC
199 	} else {
200 		error = mac_check_system_acct(td->td_ucred, NULL);
201 		if (error)
202 			return (error);
203 #endif
204 	}
205 
206 	/*
207 	 * Disallow concurrent access to the accounting vnode while we swap
208 	 * it out, in order to prevent access after close.
209 	 */
210 	sx_xlock(&acct_sx);
211 
212 	/*
213 	 * If accounting was previously enabled, kill the old space-watcher,
214 	 * close the file, and (if no new file was specified, leave).  Reset
215 	 * the suspended state regardless of whether accounting remains
216 	 * enabled.
217 	 */
218 	acct_suspended = 0;
219 	if (acct_vp != NULL) {
220 		vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
221 		error = acct_disable(td);
222 		VFS_UNLOCK_GIANT(vfslocked);
223 	}
224 	if (uap->path == NULL) {
225 		if (acct_state & ACCT_RUNNING) {
226 			acct_state |= ACCT_EXITREQ;
227 			wakeup(&acct_state);
228 		}
229 		sx_xunlock(&acct_sx);
230 		return (error);
231 	}
232 
233 	/*
234 	 * Save the new accounting file vnode, and schedule the new
235 	 * free space watcher.
236 	 */
237 	acct_vp = nd.ni_vp;
238 	acct_cred = crhold(td->td_ucred);
239 	acct_flags = flags;
240 	if (acct_state & ACCT_RUNNING)
241 		acct_state &= ~ACCT_EXITREQ;
242 	else {
243 		/*
244 		 * Try to start up an accounting kthread.  We may start more
245 		 * than one, but if so the extras will commit suicide as
246 		 * soon as they start up.
247 		 */
248 		error = kthread_create(acct_thread, NULL, NULL, 0, 0,
249 		    "accounting");
250 		if (error) {
251 			vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
252 			(void) vn_close(acct_vp, acct_flags, acct_cred, td);
253 			VFS_UNLOCK_GIANT(vfslocked);
254 			crfree(acct_cred);
255 			acct_vp = NULL;
256 			acct_cred = NULL;
257 			acct_flags = 0;
258 			sx_xunlock(&acct_sx);
259 			log(LOG_NOTICE, "Unable to start accounting thread\n");
260 			return (error);
261 		}
262 	}
263 	sx_xunlock(&acct_sx);
264 	log(LOG_NOTICE, "Accounting enabled\n");
265 	return (error);
266 }
267 
268 /*
269  * Disable currently in-progress accounting by closing the vnode, dropping
270  * our reference to the credential, and clearing the vnode's flags.
271  */
272 static int
273 acct_disable(struct thread *td)
274 {
275 	int error;
276 
277 	sx_assert(&acct_sx, SX_XLOCKED);
278 	error = vn_close(acct_vp, acct_flags, acct_cred, td);
279 	crfree(acct_cred);
280 	acct_vp = NULL;
281 	acct_cred = NULL;
282 	acct_flags = 0;
283 	log(LOG_NOTICE, "Accounting disabled\n");
284 	return (error);
285 }
286 
287 /*
288  * Write out process accounting information, on process exit.
289  * Data to be written out is specified in Leffler, et al.
290  * and are enumerated below.  (They're also noted in the system
291  * "acct.h" header file.)
292  */
293 int
294 acct_process(struct thread *td)
295 {
296 	struct acct acct;
297 	struct timeval ut, st, tmp;
298 	struct plimit *newlim, *oldlim;
299 	struct proc *p;
300 	struct rusage *r;
301 	int t, ret, vfslocked;
302 
303 	/*
304 	 * Lockless check of accounting condition before doing the hard
305 	 * work.
306 	 */
307 	if (acct_vp == NULL || acct_suspended)
308 		return (0);
309 
310 	sx_slock(&acct_sx);
311 
312 	/*
313 	 * If accounting isn't enabled, don't bother.  Have to check again
314 	 * once we own the lock in case we raced with disabling of accounting
315 	 * by another thread.
316 	 */
317 	if (acct_vp == NULL || acct_suspended) {
318 		sx_sunlock(&acct_sx);
319 		return (0);
320 	}
321 
322 	p = td->td_proc;
323 
324 	/*
325 	 * Get process accounting information.
326 	 */
327 
328 	PROC_LOCK(p);
329 	/* (1) The name of the command that ran */
330 	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
331 
332 	/* (2) The amount of user and system time that was used */
333 	calcru(p, &ut, &st);
334 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
335 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
336 
337 	/* (3) The elapsed time the command ran (and its starting time) */
338 	tmp = boottime;
339 	timevaladd(&tmp, &p->p_stats->p_start);
340 	acct.ac_btime = tmp.tv_sec;
341 	microuptime(&tmp);
342 	timevalsub(&tmp, &p->p_stats->p_start);
343 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
344 
345 	/* (4) The average amount of memory used */
346 	r = &p->p_stats->p_ru;
347 	tmp = ut;
348 	timevaladd(&tmp, &st);
349 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
350 	if (t)
351 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
352 	else
353 		acct.ac_mem = 0;
354 
355 	/* (5) The number of disk I/O operations done */
356 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
357 
358 	/* (6) The UID and GID of the process */
359 	acct.ac_uid = p->p_ucred->cr_ruid;
360 	acct.ac_gid = p->p_ucred->cr_rgid;
361 
362 	/* (7) The terminal from which the process was started */
363 	SESS_LOCK(p->p_session);
364 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
365 		acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev);
366 	else
367 		acct.ac_tty = NODEV;
368 	SESS_UNLOCK(p->p_session);
369 
370 	/* (8) The boolean flags that tell how the process terminated, etc. */
371 	acct.ac_flag = p->p_acflag;
372 	PROC_UNLOCK(p);
373 
374 	/*
375 	 * Eliminate any file size rlimit.
376 	 */
377 	newlim = lim_alloc();
378 	PROC_LOCK(p);
379 	oldlim = p->p_limit;
380 	lim_copy(newlim, oldlim);
381 	newlim->pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
382 	p->p_limit = newlim;
383 	PROC_UNLOCK(p);
384 	lim_free(oldlim);
385 
386 	/*
387 	 * Write the accounting information to the file.
388 	 */
389 	vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
390 	VOP_LEASE(acct_vp, td, acct_cred, LEASE_WRITE);
391 	ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct),
392 	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED,
393 	    (int *)0, td);
394 	VFS_UNLOCK_GIANT(vfslocked);
395 	sx_sunlock(&acct_sx);
396 	return (ret);
397 }
398 
399 /*
400  * Encode_comp_t converts from ticks in seconds and microseconds
401  * to ticks in 1/AHZ seconds.  The encoding is described in
402  * Leffler, et al., on page 63.
403  */
404 
405 #define	MANTSIZE	13			/* 13 bit mantissa. */
406 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
407 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
408 
409 static comp_t
410 encode_comp_t(u_long s, u_long us)
411 {
412 	int exp, rnd;
413 
414 	exp = 0;
415 	rnd = 0;
416 	s *= AHZ;
417 	s += us / (1000000 / AHZ);	/* Maximize precision. */
418 
419 	while (s > MAXFRACT) {
420 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
421 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
422 		exp++;
423 	}
424 
425 	/* If we need to round up, do it (and handle overflow correctly). */
426 	if (rnd && (++s > MAXFRACT)) {
427 		s >>= EXPSIZE;
428 		exp++;
429 	}
430 
431 	/* Clean it up and polish it off. */
432 	exp <<= MANTSIZE;		/* Shift the exponent into place */
433 	exp += s;			/* and add on the mantissa. */
434 	return (exp);
435 }
436 
437 /*
438  * Periodically check the filesystem to see if accounting
439  * should be turned on or off.  Beware the case where the vnode
440  * has been vgone()'d out from underneath us, e.g. when the file
441  * system containing the accounting file has been forcibly unmounted.
442  */
443 /* ARGSUSED */
444 static void
445 acctwatch(void)
446 {
447 	struct statfs sb;
448 	int vfslocked;
449 
450 	sx_assert(&acct_sx, SX_XLOCKED);
451 
452 	/*
453 	 * If accounting was disabled before our kthread was scheduled,
454 	 * then acct_vp might be NULL.  If so, just ask our kthread to
455 	 * exit and return.
456 	 */
457 	if (acct_vp == NULL) {
458 		acct_state |= ACCT_EXITREQ;
459 		return;
460 	}
461 
462 	/*
463 	 * If our vnode is no longer valid, tear it down and signal the
464 	 * accounting thread to die.
465 	 */
466 	vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
467 	if (acct_vp->v_type == VBAD) {
468 		(void) acct_disable(NULL);
469 		VFS_UNLOCK_GIANT(vfslocked);
470 		acct_state |= ACCT_EXITREQ;
471 		return;
472 	}
473 
474 	/*
475 	 * Stopping here is better than continuing, maybe it will be VBAD
476 	 * next time around.
477 	 */
478 	if (VFS_STATFS(acct_vp->v_mount, &sb, curthread) < 0) {
479 		VFS_UNLOCK_GIANT(vfslocked);
480 		return;
481 	}
482 	VFS_UNLOCK_GIANT(vfslocked);
483 	if (acct_suspended) {
484 		if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks /
485 		    100)) {
486 			acct_suspended = 0;
487 			log(LOG_NOTICE, "Accounting resumed\n");
488 		}
489 	} else {
490 		if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks /
491 		    100)) {
492 			acct_suspended = 1;
493 			log(LOG_NOTICE, "Accounting suspended\n");
494 		}
495 	}
496 }
497 
498 /*
499  * The main loop for the dedicated kernel thread that periodically calls
500  * acctwatch().
501  */
502 static void
503 acct_thread(void *dummy)
504 {
505 	u_char pri;
506 
507 	/* This is a low-priority kernel thread. */
508 	pri = PRI_MAX_KERN;
509 	mtx_lock_spin(&sched_lock);
510 	sched_prio(curthread, pri);
511 	mtx_unlock_spin(&sched_lock);
512 
513 	/* If another accounting kthread is already running, just die. */
514 	sx_xlock(&acct_sx);
515 	if (acct_state & ACCT_RUNNING) {
516 		sx_xunlock(&acct_sx);
517 		kthread_exit(0);
518 	}
519 	acct_state |= ACCT_RUNNING;
520 
521 	/* Loop until we are asked to exit. */
522 	while (!(acct_state & ACCT_EXITREQ)) {
523 
524 		/* Perform our periodic checks. */
525 		acctwatch();
526 
527 		/*
528 		 * We check this flag again before sleeping since the
529 		 * acctwatch() might have shut down accounting and asked us
530 		 * to exit.
531 		 */
532 		if (!(acct_state & ACCT_EXITREQ)) {
533 			sx_xunlock(&acct_sx);
534 			tsleep(&acct_state, pri, "-", acctchkfreq * hz);
535 			sx_xlock(&acct_sx);
536 		}
537 	}
538 
539 	/*
540 	 * Acknowledge the exit request and shutdown.  We clear both the
541 	 * exit request and running flags.
542 	 */
543 	acct_state = 0;
544 	sx_xunlock(&acct_sx);
545 	kthread_exit(0);
546 }
547