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 * 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.
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/malloc.h>
82 #include <sys/mount.h>
83 #include <sys/mutex.h>
84 #include <sys/namei.h>
87 #include <sys/resourcevar.h>
88 #include <sys/sched.h>
90 #include <sys/sysctl.h>
91 #include <sys/sysent.h>
92 #include <sys/syslog.h>
93 #include <sys/sysproto.h>
95 #include <sys/vnode.h>
97 #include <security/mac/mac_framework.h>
99 _Static_assert(sizeof(struct acctv3) - offsetof(struct acctv3, ac_trailer) ==
100 sizeof(struct acctv2) - offsetof(struct acctv2, ac_trailer), "trailer");
101 _Static_assert(sizeof(struct acctv3) - offsetof(struct acctv3, ac_len2) ==
102 sizeof(struct acctv2) - offsetof(struct acctv2, ac_len2), "len2");
105 * The routines implemented in this file are described in:
106 * Leffler, et al.: The Design and Implementation of the 4.3BSD
107 * UNIX Operating System (Addison Welley, 1989)
109 * On May 2007 the historic 3 bits base 8 exponent, 13 bit fraction
110 * compt_t representation described in the above reference was replaced
111 * with that of IEEE-754 floats.
113 * Arguably, to simplify accounting operations, this mechanism should
114 * be replaced by one in which an accounting log file (similar to /dev/klog)
115 * is read by a user process, etc. However, that has its own problems.
118 /* Floating point definitions from <float.h>. */
119 #define FLT_MANT_DIG 24 /* p */
120 #define FLT_MAX_EXP 128 /* emax */
123 * Internal accounting functions.
124 * The former's operation is described in Leffler, et al., and the latter
125 * was provided by UCB with the 4.4BSD-Lite release
127 static uint32_t encode_timeval(struct timeval);
128 static uint32_t encode_long(long);
129 static void acctwatch(void);
130 static void acct_thread(void *);
131 static int acct_disable(struct thread *, int);
134 * Accounting vnode pointer, saved vnode pointer, and flags for each.
135 * acct_sx protects against changes to the active vnode and credentials
136 * while accounting records are being committed to disk.
138 static int acct_configured;
139 static int acct_suspended;
140 static struct vnode *acct_vp;
141 static struct ucred *acct_cred;
142 static struct plimit *acct_limit;
143 static int acct_flags;
144 static struct sx acct_sx;
146 SX_SYSINIT(acct, &acct_sx, "acct_sx");
149 * State of the accounting kthread.
151 static int acct_state;
153 #define ACCT_RUNNING 1 /* Accounting kthread is running. */
154 #define ACCT_EXITREQ 2 /* Accounting kthread should exit. */
157 * Values associated with enabling and disabling accounting
159 static int acctsuspend = 2; /* stop accounting when < 2% free space left */
160 SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW,
161 &acctsuspend, 0, "percentage of free disk space below which accounting stops");
163 static int acctresume = 4; /* resume when free space risen to > 4% */
164 SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW,
165 &acctresume, 0, "percentage of free disk space above which accounting resumes");
167 static int acctchkfreq = 15; /* frequency (in seconds) to check space */
170 sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS)
174 /* Write out the old value. */
175 error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int));
176 if (error || req->newptr == NULL)
179 /* Read in and verify the new value. */
180 error = SYSCTL_IN(req, &value, sizeof(int));
188 SYSCTL_PROC(_kern, OID_AUTO, acct_chkfreq, CTLTYPE_INT|CTLFLAG_RW,
189 &acctchkfreq, 0, sysctl_acct_chkfreq, "I",
190 "frequency for checking the free space");
192 SYSCTL_INT(_kern, OID_AUTO, acct_configured, CTLFLAG_RD, &acct_configured, 0,
193 "Accounting configured or not");
195 SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0,
196 "Accounting suspended or not");
199 * Accounting system call. Written based on the specification and previous
200 * implementation done by Mark Tinguely.
203 sys_acct(struct thread *td, struct acct_args *uap)
206 int error, flags, i, replacing;
208 error = priv_check(td, PRIV_ACCT);
213 * If accounting is to be started to a file, open that file for
214 * appending and make sure it's a 'normal'.
216 if (uap->path != NULL) {
217 NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1,
218 UIO_USERSPACE, uap->path, td);
219 flags = FWRITE | O_APPEND;
220 error = vn_open(&nd, &flags, 0, NULL);
223 NDFREE(&nd, NDF_ONLY_PNBUF);
225 error = mac_system_check_acct(td->td_ucred, nd.ni_vp);
227 VOP_UNLOCK(nd.ni_vp, 0);
228 vn_close(nd.ni_vp, flags, td->td_ucred, td);
232 VOP_UNLOCK(nd.ni_vp, 0);
233 if (nd.ni_vp->v_type != VREG) {
234 vn_close(nd.ni_vp, flags, td->td_ucred, td);
239 error = mac_system_check_acct(td->td_ucred, NULL);
246 * Disallow concurrent access to the accounting vnode while we swap
247 * it out, in order to prevent access after close.
252 * Don't log spurious disable/enable messages if we are
253 * switching from one accounting file to another due to log
256 replacing = (acct_vp != NULL && uap->path != NULL);
259 * If accounting was previously enabled, kill the old space-watcher,
260 * close the file, and (if no new file was specified, leave). Reset
261 * the suspended state regardless of whether accounting remains
266 error = acct_disable(td, !replacing);
267 if (uap->path == NULL) {
268 if (acct_state & ACCT_RUNNING) {
269 acct_state |= ACCT_EXITREQ;
272 sx_xunlock(&acct_sx);
277 * Create our own plimit object without limits. It will be assigned
278 * to exiting processes.
280 acct_limit = lim_alloc();
281 for (i = 0; i < RLIM_NLIMITS; i++)
282 acct_limit->pl_rlimit[i].rlim_cur =
283 acct_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
286 * Save the new accounting file vnode, and schedule the new
287 * free space watcher.
290 acct_cred = crhold(td->td_ucred);
292 if (acct_state & ACCT_RUNNING)
293 acct_state &= ~ACCT_EXITREQ;
296 * Try to start up an accounting kthread. We may start more
297 * than one, but if so the extras will commit suicide as
298 * soon as they start up.
300 error = kproc_create(acct_thread, NULL, NULL, 0, 0,
303 (void) acct_disable(td, 0);
304 sx_xunlock(&acct_sx);
305 log(LOG_NOTICE, "Unable to start accounting thread\n");
310 sx_xunlock(&acct_sx);
312 log(LOG_NOTICE, "Accounting enabled\n");
317 * Disable currently in-progress accounting by closing the vnode, dropping
318 * our reference to the credential, and clearing the vnode's flags.
321 acct_disable(struct thread *td, int logging)
325 sx_assert(&acct_sx, SX_XLOCKED);
326 error = vn_close(acct_vp, acct_flags, acct_cred, td);
328 lim_free(acct_limit);
334 log(LOG_NOTICE, "Accounting disabled\n");
339 * Write out process accounting information, on process exit.
340 * Data to be written out is specified in Leffler, et al.
341 * and are enumerated below. (They're also noted in the system
342 * "acct.h" header file.)
345 acct_process(struct thread *td)
348 struct timeval ut, st, tmp;
349 struct plimit *oldlim;
355 * Lockless check of accounting condition before doing the hard
358 if (acct_vp == NULL || acct_suspended)
364 * If accounting isn't enabled, don't bother. Have to check again
365 * once we own the lock in case we raced with disabling of accounting
368 if (acct_vp == NULL || acct_suspended) {
369 sx_sunlock(&acct_sx);
376 * Get process accounting information.
379 sx_slock(&proctree_lock);
382 /* (1) The terminal from which the process was started */
383 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
384 acct.ac_tty = tty_udev(p->p_pgrp->pg_session->s_ttyp);
387 sx_sunlock(&proctree_lock);
389 /* (2) The name of the command that ran */
390 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
392 /* (3) The amount of user and system time that was used */
393 rufetchcalc(p, &ru, &ut, &st);
394 acct.ac_utime = encode_timeval(ut);
395 acct.ac_stime = encode_timeval(st);
397 /* (4) The elapsed time the command ran (and its starting time) */
399 timevaladd(&tmp, &p->p_stats->p_start);
400 acct.ac_btime = tmp.tv_sec;
402 timevalsub(&tmp, &p->p_stats->p_start);
403 acct.ac_etime = encode_timeval(tmp);
405 /* (5) The average amount of memory used */
407 timevaladd(&tmp, &st);
408 /* Convert tmp (i.e. u + s) into hz units to match ru_i*. */
409 t = tmp.tv_sec * hz + tmp.tv_usec / tick;
411 acct.ac_mem = encode_long((ru.ru_ixrss + ru.ru_idrss +
416 /* (6) The number of disk I/O operations done */
417 acct.ac_io = encode_long(ru.ru_inblock + ru.ru_oublock);
419 /* (7) The UID and GID of the process */
420 acct.ac_uid = p->p_ucred->cr_ruid;
421 acct.ac_gid = p->p_ucred->cr_rgid;
423 /* (8) The boolean flags that tell how the process terminated, etc. */
424 acct.ac_flagx = p->p_acflag;
426 /* Setup ancillary structure fields. */
427 acct.ac_flagx |= ANVER;
430 acct.ac_len = acct.ac_len2 = sizeof(acct);
433 * Eliminate rlimits (file size limit in particular).
436 p->p_limit = lim_hold(acct_limit);
441 * Write the accounting information to the file.
443 ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct),
444 (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED,
446 sx_sunlock(&acct_sx);
450 /* FLOAT_CONVERSION_START (Regression testing; don't remove this line.) */
452 /* Convert timevals and longs into IEEE-754 bit patterns. */
454 /* Mantissa mask (MSB is implied, so subtract 1). */
455 #define MANT_MASK ((1 << (FLT_MANT_DIG - 1)) - 1)
458 * We calculate integer values to a precision of approximately
460 * This is high-enough precision to fill the 24 float bits
461 * and low-enough to avoid overflowing the 32 int bits.
465 /* log_2(1000000). */
469 * Convert the elements of a timeval into a 32-bit word holding
470 * the bits of a IEEE-754 float.
471 * The float value represents the timeval's value in microsecond units.
474 encode_timeval(struct timeval tv)
477 int val, exp; /* Unnormalized value and exponent */
478 int norm_exp; /* Normalized exponent */
482 * First calculate value and exponent to about CALC_BITS precision.
483 * Note that the following conditionals have been ordered so that
484 * the most common cases appear first.
486 if (tv.tv_sec == 0) {
493 * Calculate the value to a precision of approximately
496 log2_s = fls(tv.tv_sec) - 1;
497 if (log2_s + LOG2_1M < CALC_BITS) {
499 val = 1000000 * tv.tv_sec + tv.tv_usec;
501 exp = log2_s + LOG2_1M - CALC_BITS;
502 val = (unsigned int)(((uint64_t)1000000 * tv.tv_sec +
506 /* Now normalize and pack the value into an IEEE-754 float. */
507 norm_exp = fls(val) - 1;
508 shift = FLT_MANT_DIG - norm_exp - 1;
510 printf("val=%d exp=%d shift=%d log2(val)=%d\n",
511 val, exp, shift, norm_exp);
512 printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp,
513 ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK));
515 return (((FLT_MAX_EXP - 1 + exp + norm_exp) << (FLT_MANT_DIG - 1)) |
516 ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK));
520 * Convert a non-negative long value into the bit pattern of
521 * an IEEE-754 float value.
524 encode_long(long val)
526 int norm_exp; /* Normalized exponent */
533 "encode_long: negative value %ld in accounting record\n",
537 norm_exp = fls(val) - 1;
538 shift = FLT_MANT_DIG - norm_exp - 1;
540 printf("val=%d shift=%d log2(val)=%d\n",
541 val, shift, norm_exp);
542 printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp,
543 ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK));
545 return (((FLT_MAX_EXP - 1 + norm_exp) << (FLT_MANT_DIG - 1)) |
546 ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK));
549 /* FLOAT_CONVERSION_END (Regression testing; don't remove this line.) */
552 * Periodically check the filesystem to see if accounting
553 * should be turned on or off. Beware the case where the vnode
554 * has been vgone()'d out from underneath us, e.g. when the file
555 * system containing the accounting file has been forcibly unmounted.
563 sx_assert(&acct_sx, SX_XLOCKED);
566 * If accounting was disabled before our kthread was scheduled,
567 * then acct_vp might be NULL. If so, just ask our kthread to
570 if (acct_vp == NULL) {
571 acct_state |= ACCT_EXITREQ;
576 * If our vnode is no longer valid, tear it down and signal the
577 * accounting thread to die.
579 if (acct_vp->v_type == VBAD) {
580 (void) acct_disable(NULL, 1);
581 acct_state |= ACCT_EXITREQ;
586 * Stopping here is better than continuing, maybe it will be VBAD
589 sp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
590 if (VFS_STATFS(acct_vp->v_mount, sp) < 0) {
594 if (acct_suspended) {
595 if (sp->f_bavail > (int64_t)(acctresume * sp->f_blocks /
598 log(LOG_NOTICE, "Accounting resumed\n");
601 if (sp->f_bavail <= (int64_t)(acctsuspend * sp->f_blocks /
604 log(LOG_NOTICE, "Accounting suspended\n");
611 * The main loop for the dedicated kernel thread that periodically calls
615 acct_thread(void *dummy)
619 /* This is a low-priority kernel thread. */
621 thread_lock(curthread);
622 sched_prio(curthread, pri);
623 thread_unlock(curthread);
625 /* If another accounting kthread is already running, just die. */
627 if (acct_state & ACCT_RUNNING) {
628 sx_xunlock(&acct_sx);
631 acct_state |= ACCT_RUNNING;
633 /* Loop until we are asked to exit. */
634 while (!(acct_state & ACCT_EXITREQ)) {
636 /* Perform our periodic checks. */
640 * We check this flag again before sleeping since the
641 * acctwatch() might have shut down accounting and asked us
644 if (!(acct_state & ACCT_EXITREQ)) {
645 sx_sleep(&acct_state, &acct_sx, 0, "-",
651 * Acknowledge the exit request and shutdown. We clear both the
652 * exit request and running flags.
655 sx_xunlock(&acct_sx);