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
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.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
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.
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
37 * Copyright (c) 1994 Christopher G. Demetriou
39 * Redistribution and use in source and binary forms, with or without
40 * modification, are permitted provided that the following conditions
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.
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
67 * @(#)kern_acct.c 8.1 (Berkeley) 6/14/93
70 #include <sys/cdefs.h>
71 __FBSDID("$FreeBSD$");
73 #include <sys/param.h>
74 #include <sys/systm.h>
76 #include <sys/fcntl.h>
77 #include <sys/kernel.h>
78 #include <sys/kthread.h>
79 #include <sys/limits.h>
81 #include <sys/mount.h>
82 #include <sys/mutex.h>
83 #include <sys/namei.h>
86 #include <sys/resourcevar.h>
87 #include <sys/sched.h>
89 #include <sys/sysctl.h>
90 #include <sys/sysent.h>
91 #include <sys/syslog.h>
92 #include <sys/sysproto.h>
94 #include <sys/vnode.h>
96 #include <security/mac/mac_framework.h>
99 * The routines implemented in this file are described in:
100 * Leffler, et al.: The Design and Implementation of the 4.3BSD
101 * UNIX Operating System (Addison Welley, 1989)
103 * On May 2007 the historic 3 bits base 8 exponent, 13 bit fraction
104 * compt_t representation described in the above reference was replaced
105 * with that of IEEE-754 floats.
107 * Arguably, to simplify accounting operations, this mechanism should
108 * be replaced by one in which an accounting log file (similar to /dev/klog)
109 * is read by a user process, etc. However, that has its own problems.
112 /* Floating point definitions from <float.h>. */
113 #define FLT_MANT_DIG 24 /* p */
114 #define FLT_MAX_EXP 128 /* emax */
117 * Internal accounting functions.
118 * The former's operation is described in Leffler, et al., and the latter
119 * was provided by UCB with the 4.4BSD-Lite release
121 static uint32_t encode_timeval(struct timeval);
122 static uint32_t encode_long(long);
123 static void acctwatch(void);
124 static void acct_thread(void *);
125 static int acct_disable(struct thread *, int);
128 * Accounting vnode pointer, saved vnode pointer, and flags for each.
129 * acct_sx protects against changes to the active vnode and credentials
130 * while accounting records are being committed to disk.
132 static int acct_configured;
133 static int acct_suspended;
134 static struct vnode *acct_vp;
135 static struct ucred *acct_cred;
136 static struct plimit *acct_limit;
137 static int acct_flags;
138 static struct sx acct_sx;
140 SX_SYSINIT(acct, &acct_sx, "acct_sx");
143 * State of the accounting kthread.
145 static int acct_state;
147 #define ACCT_RUNNING 1 /* Accounting kthread is running. */
148 #define ACCT_EXITREQ 2 /* Accounting kthread should exit. */
151 * Values associated with enabling and disabling accounting
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");
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");
161 static int acctchkfreq = 15; /* frequency (in seconds) to check space */
164 sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS)
168 /* Write out the old value. */
169 error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int));
170 if (error || req->newptr == NULL)
173 /* Read in and verify the new value. */
174 error = SYSCTL_IN(req, &value, sizeof(int));
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");
186 SYSCTL_INT(_kern, OID_AUTO, acct_configured, CTLFLAG_RD, &acct_configured, 0,
187 "Accounting configured or not");
189 SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0,
190 "Accounting suspended or not");
193 * Accounting system call. Written based on the specification and previous
194 * implementation done by Mark Tinguely.
197 sys_acct(struct thread *td, struct acct_args *uap)
200 int error, flags, i, replacing;
202 error = priv_check(td, PRIV_ACCT);
207 * If accounting is to be started to a file, open that file for
208 * appending and make sure it's a 'normal'.
210 if (uap->path != NULL) {
211 NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1,
212 UIO_USERSPACE, uap->path, td);
213 flags = FWRITE | O_APPEND;
214 error = vn_open(&nd, &flags, 0, NULL);
217 NDFREE(&nd, NDF_ONLY_PNBUF);
219 error = mac_system_check_acct(td->td_ucred, nd.ni_vp);
221 VOP_UNLOCK(nd.ni_vp, 0);
222 vn_close(nd.ni_vp, flags, td->td_ucred, td);
226 VOP_UNLOCK(nd.ni_vp, 0);
227 if (nd.ni_vp->v_type != VREG) {
228 vn_close(nd.ni_vp, flags, td->td_ucred, td);
233 error = mac_system_check_acct(td->td_ucred, NULL);
240 * Disallow concurrent access to the accounting vnode while we swap
241 * it out, in order to prevent access after close.
246 * Don't log spurious disable/enable messages if we are
247 * switching from one accounting file to another due to log
250 replacing = (acct_vp != NULL && uap->path != NULL);
253 * If accounting was previously enabled, kill the old space-watcher,
254 * close the file, and (if no new file was specified, leave). Reset
255 * the suspended state regardless of whether accounting remains
260 error = acct_disable(td, !replacing);
261 if (uap->path == NULL) {
262 if (acct_state & ACCT_RUNNING) {
263 acct_state |= ACCT_EXITREQ;
266 sx_xunlock(&acct_sx);
271 * Create our own plimit object without limits. It will be assigned
272 * to exiting processes.
274 acct_limit = lim_alloc();
275 for (i = 0; i < RLIM_NLIMITS; i++)
276 acct_limit->pl_rlimit[i].rlim_cur =
277 acct_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
280 * Save the new accounting file vnode, and schedule the new
281 * free space watcher.
284 acct_cred = crhold(td->td_ucred);
286 if (acct_state & ACCT_RUNNING)
287 acct_state &= ~ACCT_EXITREQ;
290 * Try to start up an accounting kthread. We may start more
291 * than one, but if so the extras will commit suicide as
292 * soon as they start up.
294 error = kproc_create(acct_thread, NULL, NULL, 0, 0,
297 (void) acct_disable(td, 0);
298 sx_xunlock(&acct_sx);
299 log(LOG_NOTICE, "Unable to start accounting thread\n");
304 sx_xunlock(&acct_sx);
306 log(LOG_NOTICE, "Accounting enabled\n");
311 * Disable currently in-progress accounting by closing the vnode, dropping
312 * our reference to the credential, and clearing the vnode's flags.
315 acct_disable(struct thread *td, int logging)
319 sx_assert(&acct_sx, SX_XLOCKED);
320 error = vn_close(acct_vp, acct_flags, acct_cred, td);
322 lim_free(acct_limit);
328 log(LOG_NOTICE, "Accounting disabled\n");
333 * Write out process accounting information, on process exit.
334 * Data to be written out is specified in Leffler, et al.
335 * and are enumerated below. (They're also noted in the system
336 * "acct.h" header file.)
339 acct_process(struct thread *td)
342 struct timeval ut, st, tmp;
343 struct plimit *oldlim;
349 * Lockless check of accounting condition before doing the hard
352 if (acct_vp == NULL || acct_suspended)
358 * If accounting isn't enabled, don't bother. Have to check again
359 * once we own the lock in case we raced with disabling of accounting
362 if (acct_vp == NULL || acct_suspended) {
363 sx_sunlock(&acct_sx);
370 * Get process accounting information.
373 sx_slock(&proctree_lock);
376 /* (1) The terminal from which the process was started */
377 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
378 acct.ac_tty = tty_udev(p->p_pgrp->pg_session->s_ttyp);
381 sx_sunlock(&proctree_lock);
383 /* (2) The name of the command that ran */
384 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
386 /* (3) The amount of user and system time that was used */
387 rufetchcalc(p, &ru, &ut, &st);
388 acct.ac_utime = encode_timeval(ut);
389 acct.ac_stime = encode_timeval(st);
391 /* (4) The elapsed time the command ran (and its starting time) */
393 timevaladd(&tmp, &p->p_stats->p_start);
394 acct.ac_btime = tmp.tv_sec;
396 timevalsub(&tmp, &p->p_stats->p_start);
397 acct.ac_etime = encode_timeval(tmp);
399 /* (5) The average amount of memory used */
401 timevaladd(&tmp, &st);
402 /* Convert tmp (i.e. u + s) into hz units to match ru_i*. */
403 t = tmp.tv_sec * hz + tmp.tv_usec / tick;
405 acct.ac_mem = encode_long((ru.ru_ixrss + ru.ru_idrss +
410 /* (6) The number of disk I/O operations done */
411 acct.ac_io = encode_long(ru.ru_inblock + ru.ru_oublock);
413 /* (7) The UID and GID of the process */
414 acct.ac_uid = p->p_ucred->cr_ruid;
415 acct.ac_gid = p->p_ucred->cr_rgid;
417 /* (8) The boolean flags that tell how the process terminated, etc. */
418 acct.ac_flagx = p->p_acflag;
420 /* Setup ancillary structure fields. */
421 acct.ac_flagx |= ANVER;
424 acct.ac_len = acct.ac_len2 = sizeof(acct);
427 * Eliminate rlimits (file size limit in particular).
430 p->p_limit = lim_hold(acct_limit);
435 * Write the accounting information to the file.
437 ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct),
438 (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED,
440 sx_sunlock(&acct_sx);
444 /* FLOAT_CONVERSION_START (Regression testing; don't remove this line.) */
446 /* Convert timevals and longs into IEEE-754 bit patterns. */
448 /* Mantissa mask (MSB is implied, so subtract 1). */
449 #define MANT_MASK ((1 << (FLT_MANT_DIG - 1)) - 1)
452 * We calculate integer values to a precision of approximately
454 * This is high-enough precision to fill the 24 float bits
455 * and low-enough to avoid overflowing the 32 int bits.
459 /* log_2(1000000). */
463 * Convert the elements of a timeval into a 32-bit word holding
464 * the bits of a IEEE-754 float.
465 * The float value represents the timeval's value in microsecond units.
468 encode_timeval(struct timeval tv)
471 int val, exp; /* Unnormalized value and exponent */
472 int norm_exp; /* Normalized exponent */
476 * First calculate value and exponent to about CALC_BITS precision.
477 * Note that the following conditionals have been ordered so that
478 * the most common cases appear first.
480 if (tv.tv_sec == 0) {
487 * Calculate the value to a precision of approximately
490 log2_s = fls(tv.tv_sec) - 1;
491 if (log2_s + LOG2_1M < CALC_BITS) {
493 val = 1000000 * tv.tv_sec + tv.tv_usec;
495 exp = log2_s + LOG2_1M - CALC_BITS;
496 val = (unsigned int)(((uint64_t)1000000 * tv.tv_sec +
500 /* Now normalize and pack the value into an IEEE-754 float. */
501 norm_exp = fls(val) - 1;
502 shift = FLT_MANT_DIG - norm_exp - 1;
504 printf("val=%d exp=%d shift=%d log2(val)=%d\n",
505 val, exp, shift, norm_exp);
506 printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp,
507 ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK));
509 return (((FLT_MAX_EXP - 1 + exp + norm_exp) << (FLT_MANT_DIG - 1)) |
510 ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK));
514 * Convert a non-negative long value into the bit pattern of
515 * an IEEE-754 float value.
518 encode_long(long val)
520 int norm_exp; /* Normalized exponent */
527 "encode_long: negative value %ld in accounting record\n",
531 norm_exp = fls(val) - 1;
532 shift = FLT_MANT_DIG - norm_exp - 1;
534 printf("val=%d shift=%d log2(val)=%d\n",
535 val, shift, norm_exp);
536 printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp,
537 ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK));
539 return (((FLT_MAX_EXP - 1 + norm_exp) << (FLT_MANT_DIG - 1)) |
540 ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK));
543 /* FLOAT_CONVERSION_END (Regression testing; don't remove this line.) */
546 * Periodically check the filesystem to see if accounting
547 * should be turned on or off. Beware the case where the vnode
548 * has been vgone()'d out from underneath us, e.g. when the file
549 * system containing the accounting file has been forcibly unmounted.
557 sx_assert(&acct_sx, SX_XLOCKED);
560 * If accounting was disabled before our kthread was scheduled,
561 * then acct_vp might be NULL. If so, just ask our kthread to
564 if (acct_vp == NULL) {
565 acct_state |= ACCT_EXITREQ;
570 * If our vnode is no longer valid, tear it down and signal the
571 * accounting thread to die.
573 if (acct_vp->v_type == VBAD) {
574 (void) acct_disable(NULL, 1);
575 acct_state |= ACCT_EXITREQ;
580 * Stopping here is better than continuing, maybe it will be VBAD
583 if (VFS_STATFS(acct_vp->v_mount, &sb) < 0)
585 if (acct_suspended) {
586 if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks /
589 log(LOG_NOTICE, "Accounting resumed\n");
592 if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks /
595 log(LOG_NOTICE, "Accounting suspended\n");
601 * The main loop for the dedicated kernel thread that periodically calls
605 acct_thread(void *dummy)
609 /* This is a low-priority kernel thread. */
611 thread_lock(curthread);
612 sched_prio(curthread, pri);
613 thread_unlock(curthread);
615 /* If another accounting kthread is already running, just die. */
617 if (acct_state & ACCT_RUNNING) {
618 sx_xunlock(&acct_sx);
621 acct_state |= ACCT_RUNNING;
623 /* Loop until we are asked to exit. */
624 while (!(acct_state & ACCT_EXITREQ)) {
626 /* Perform our periodic checks. */
630 * We check this flag again before sleeping since the
631 * acctwatch() might have shut down accounting and asked us
634 if (!(acct_state & ACCT_EXITREQ)) {
635 sx_sleep(&acct_state, &acct_sx, 0, "-",
641 * Acknowledge the exit request and shutdown. We clear both the
642 * exit request and running flags.
645 sx_xunlock(&acct_sx);