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, vfslocked, 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 | MPSAFE | AUDITVNODE1,
212 UIO_USERSPACE, uap->path, td);
213 flags = FWRITE | O_APPEND;
214 error = vn_open(&nd, &flags, 0, NULL);
217 vfslocked = NDHASGIANT(&nd);
218 NDFREE(&nd, NDF_ONLY_PNBUF);
220 error = mac_system_check_acct(td->td_ucred, nd.ni_vp);
222 VOP_UNLOCK(nd.ni_vp, 0);
223 vn_close(nd.ni_vp, flags, td->td_ucred, td);
224 VFS_UNLOCK_GIANT(vfslocked);
228 VOP_UNLOCK(nd.ni_vp, 0);
229 if (nd.ni_vp->v_type != VREG) {
230 vn_close(nd.ni_vp, flags, td->td_ucred, td);
231 VFS_UNLOCK_GIANT(vfslocked);
234 VFS_UNLOCK_GIANT(vfslocked);
237 error = mac_system_check_acct(td->td_ucred, NULL);
244 * Disallow concurrent access to the accounting vnode while we swap
245 * it out, in order to prevent access after close.
250 * Don't log spurious disable/enable messages if we are
251 * switching from one accounting file to another due to log
254 replacing = (acct_vp != NULL && uap->path != NULL);
257 * If accounting was previously enabled, kill the old space-watcher,
258 * close the file, and (if no new file was specified, leave). Reset
259 * the suspended state regardless of whether accounting remains
263 if (acct_vp != NULL) {
264 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
265 error = acct_disable(td, !replacing);
266 VFS_UNLOCK_GIANT(vfslocked);
268 if (uap->path == NULL) {
269 if (acct_state & ACCT_RUNNING) {
270 acct_state |= ACCT_EXITREQ;
273 sx_xunlock(&acct_sx);
278 * Create our own plimit object without limits. It will be assigned
279 * to exiting processes.
281 acct_limit = lim_alloc();
282 for (i = 0; i < RLIM_NLIMITS; i++)
283 acct_limit->pl_rlimit[i].rlim_cur =
284 acct_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
287 * Save the new accounting file vnode, and schedule the new
288 * free space watcher.
291 acct_cred = crhold(td->td_ucred);
293 if (acct_state & ACCT_RUNNING)
294 acct_state &= ~ACCT_EXITREQ;
297 * Try to start up an accounting kthread. We may start more
298 * than one, but if so the extras will commit suicide as
299 * soon as they start up.
301 error = kproc_create(acct_thread, NULL, NULL, 0, 0,
304 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
305 (void) acct_disable(td, 0);
306 VFS_UNLOCK_GIANT(vfslocked);
307 sx_xunlock(&acct_sx);
308 log(LOG_NOTICE, "Unable to start accounting thread\n");
313 sx_xunlock(&acct_sx);
315 log(LOG_NOTICE, "Accounting enabled\n");
320 * Disable currently in-progress accounting by closing the vnode, dropping
321 * our reference to the credential, and clearing the vnode's flags.
324 acct_disable(struct thread *td, int logging)
328 sx_assert(&acct_sx, SX_XLOCKED);
329 error = vn_close(acct_vp, acct_flags, acct_cred, td);
331 lim_free(acct_limit);
337 log(LOG_NOTICE, "Accounting disabled\n");
342 * Write out process accounting information, on process exit.
343 * Data to be written out is specified in Leffler, et al.
344 * and are enumerated below. (They're also noted in the system
345 * "acct.h" header file.)
348 acct_process(struct thread *td)
351 struct timeval ut, st, tmp;
352 struct plimit *oldlim;
355 int t, ret, vfslocked;
358 * Lockless check of accounting condition before doing the hard
361 if (acct_vp == NULL || acct_suspended)
367 * If accounting isn't enabled, don't bother. Have to check again
368 * once we own the lock in case we raced with disabling of accounting
371 if (acct_vp == NULL || acct_suspended) {
372 sx_sunlock(&acct_sx);
379 * Get process accounting information.
382 sx_slock(&proctree_lock);
385 /* (1) The terminal from which the process was started */
386 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
387 acct.ac_tty = tty_udev(p->p_pgrp->pg_session->s_ttyp);
390 sx_sunlock(&proctree_lock);
392 /* (2) The name of the command that ran */
393 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
395 /* (3) The amount of user and system time that was used */
396 rufetchcalc(p, &ru, &ut, &st);
397 acct.ac_utime = encode_timeval(ut);
398 acct.ac_stime = encode_timeval(st);
400 /* (4) The elapsed time the command ran (and its starting time) */
402 timevaladd(&tmp, &p->p_stats->p_start);
403 acct.ac_btime = tmp.tv_sec;
405 timevalsub(&tmp, &p->p_stats->p_start);
406 acct.ac_etime = encode_timeval(tmp);
408 /* (5) The average amount of memory used */
410 timevaladd(&tmp, &st);
411 /* Convert tmp (i.e. u + s) into hz units to match ru_i*. */
412 t = tmp.tv_sec * hz + tmp.tv_usec / tick;
414 acct.ac_mem = encode_long((ru.ru_ixrss + ru.ru_idrss +
419 /* (6) The number of disk I/O operations done */
420 acct.ac_io = encode_long(ru.ru_inblock + ru.ru_oublock);
422 /* (7) The UID and GID of the process */
423 acct.ac_uid = p->p_ucred->cr_ruid;
424 acct.ac_gid = p->p_ucred->cr_rgid;
426 /* (8) The boolean flags that tell how the process terminated, etc. */
427 acct.ac_flagx = p->p_acflag;
429 /* Setup ancillary structure fields. */
430 acct.ac_flagx |= ANVER;
433 acct.ac_len = acct.ac_len2 = sizeof(acct);
436 * Eliminate rlimits (file size limit in particular).
439 p->p_limit = lim_hold(acct_limit);
444 * Write the accounting information to the file.
446 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
447 ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct),
448 (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED,
450 VFS_UNLOCK_GIANT(vfslocked);
451 sx_sunlock(&acct_sx);
455 /* FLOAT_CONVERSION_START (Regression testing; don't remove this line.) */
457 /* Convert timevals and longs into IEEE-754 bit patterns. */
459 /* Mantissa mask (MSB is implied, so subtract 1). */
460 #define MANT_MASK ((1 << (FLT_MANT_DIG - 1)) - 1)
463 * We calculate integer values to a precision of approximately
465 * This is high-enough precision to fill the 24 float bits
466 * and low-enough to avoid overflowing the 32 int bits.
470 /* log_2(1000000). */
474 * Convert the elements of a timeval into a 32-bit word holding
475 * the bits of a IEEE-754 float.
476 * The float value represents the timeval's value in microsecond units.
479 encode_timeval(struct timeval tv)
482 int val, exp; /* Unnormalized value and exponent */
483 int norm_exp; /* Normalized exponent */
487 * First calculate value and exponent to about CALC_BITS precision.
488 * Note that the following conditionals have been ordered so that
489 * the most common cases appear first.
491 if (tv.tv_sec == 0) {
498 * Calculate the value to a precision of approximately
501 log2_s = fls(tv.tv_sec) - 1;
502 if (log2_s + LOG2_1M < CALC_BITS) {
504 val = 1000000 * tv.tv_sec + tv.tv_usec;
506 exp = log2_s + LOG2_1M - CALC_BITS;
507 val = (unsigned int)(((uint64_t)1000000 * tv.tv_sec +
511 /* Now normalize and pack the value into an IEEE-754 float. */
512 norm_exp = fls(val) - 1;
513 shift = FLT_MANT_DIG - norm_exp - 1;
515 printf("val=%d exp=%d shift=%d log2(val)=%d\n",
516 val, exp, shift, norm_exp);
517 printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp,
518 ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK));
520 return (((FLT_MAX_EXP - 1 + exp + norm_exp) << (FLT_MANT_DIG - 1)) |
521 ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK));
525 * Convert a non-negative long value into the bit pattern of
526 * an IEEE-754 float value.
529 encode_long(long val)
531 int norm_exp; /* Normalized exponent */
538 "encode_long: negative value %ld in accounting record\n",
542 norm_exp = fls(val) - 1;
543 shift = FLT_MANT_DIG - norm_exp - 1;
545 printf("val=%d shift=%d log2(val)=%d\n",
546 val, shift, norm_exp);
547 printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp,
548 ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK));
550 return (((FLT_MAX_EXP - 1 + norm_exp) << (FLT_MANT_DIG - 1)) |
551 ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK));
554 /* FLOAT_CONVERSION_END (Regression testing; don't remove this line.) */
557 * Periodically check the filesystem to see if accounting
558 * should be turned on or off. Beware the case where the vnode
559 * has been vgone()'d out from underneath us, e.g. when the file
560 * system containing the accounting file has been forcibly unmounted.
569 sx_assert(&acct_sx, SX_XLOCKED);
572 * If accounting was disabled before our kthread was scheduled,
573 * then acct_vp might be NULL. If so, just ask our kthread to
576 if (acct_vp == NULL) {
577 acct_state |= ACCT_EXITREQ;
582 * If our vnode is no longer valid, tear it down and signal the
583 * accounting thread to die.
585 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
586 if (acct_vp->v_type == VBAD) {
587 (void) acct_disable(NULL, 1);
588 VFS_UNLOCK_GIANT(vfslocked);
589 acct_state |= ACCT_EXITREQ;
594 * Stopping here is better than continuing, maybe it will be VBAD
597 if (VFS_STATFS(acct_vp->v_mount, &sb) < 0) {
598 VFS_UNLOCK_GIANT(vfslocked);
601 VFS_UNLOCK_GIANT(vfslocked);
602 if (acct_suspended) {
603 if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks /
606 log(LOG_NOTICE, "Accounting resumed\n");
609 if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks /
612 log(LOG_NOTICE, "Accounting suspended\n");
618 * The main loop for the dedicated kernel thread that periodically calls
622 acct_thread(void *dummy)
626 /* This is a low-priority kernel thread. */
628 thread_lock(curthread);
629 sched_prio(curthread, pri);
630 thread_unlock(curthread);
632 /* If another accounting kthread is already running, just die. */
634 if (acct_state & ACCT_RUNNING) {
635 sx_xunlock(&acct_sx);
638 acct_state |= ACCT_RUNNING;
640 /* Loop until we are asked to exit. */
641 while (!(acct_state & ACCT_EXITREQ)) {
643 /* Perform our periodic checks. */
647 * We check this flag again before sleeping since the
648 * acctwatch() might have shut down accounting and asked us
651 if (!(acct_state & ACCT_EXITREQ)) {
652 sx_sleep(&acct_state, &acct_sx, 0, "-",
658 * Acknowledge the exit request and shutdown. We clear both the
659 * exit request and running flags.
662 sx_xunlock(&acct_sx);