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.
10 * Copyright (c) 1994 Christopher G. Demetriou
11 * Copyright (c) 2005 Robert N. M. Watson
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. 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.
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
41 * @(#)kern_acct.c 8.1 (Berkeley) 6/14/93
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
49 #include <sys/param.h>
50 #include <sys/systm.h>
52 #include <sys/fcntl.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
56 #include <sys/mount.h>
57 #include <sys/mutex.h>
58 #include <sys/namei.h>
61 #include <sys/resourcevar.h>
62 #include <sys/sched.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
66 #include <sys/syslog.h>
67 #include <sys/sysproto.h>
69 #include <sys/vnode.h>
71 #include <security/mac/mac_framework.h>
74 * The routines implemented in this file are described in:
75 * Leffler, et al.: The Design and Implementation of the 4.3BSD
76 * UNIX Operating System (Addison Welley, 1989)
79 * Arguably, to simplify accounting operations, this mechanism should
80 * be replaced by one in which an accounting log file (similar to /dev/klog)
81 * is read by a user process, etc. However, that has its own problems.
85 * Internal accounting functions.
86 * The former's operation is described in Leffler, et al., and the latter
87 * was provided by UCB with the 4.4BSD-Lite release
89 static comp_t encode_comp_t(u_long, u_long);
90 static void acctwatch(void);
91 static void acct_thread(void *);
92 static int acct_disable(struct thread *);
95 * Accounting vnode pointer, saved vnode pointer, and flags for each.
96 * acct_sx protects against changes to the active vnode and credentials
97 * while accounting records are being committed to disk.
99 static int acct_configured;
100 static int acct_suspended;
101 static struct vnode *acct_vp;
102 static struct ucred *acct_cred;
103 static int acct_flags;
104 static struct sx acct_sx;
106 SX_SYSINIT(acct, &acct_sx, "acct_sx");
109 * State of the accounting kthread.
111 static int acct_state;
113 #define ACCT_RUNNING 1 /* Accounting kthread is running. */
114 #define ACCT_EXITREQ 2 /* Accounting kthread should exit. */
117 * Values associated with enabling and disabling accounting
119 static int acctsuspend = 2; /* stop accounting when < 2% free space left */
120 SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW,
121 &acctsuspend, 0, "percentage of free disk space below which accounting stops");
123 static int acctresume = 4; /* resume when free space risen to > 4% */
124 SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW,
125 &acctresume, 0, "percentage of free disk space above which accounting resumes");
127 static int acctchkfreq = 15; /* frequency (in seconds) to check space */
130 sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS)
134 /* Write out the old value. */
135 error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int));
136 if (error || req->newptr == NULL)
139 /* Read in and verify the new value. */
140 error = SYSCTL_IN(req, &value, sizeof(int));
148 SYSCTL_PROC(_kern, OID_AUTO, acct_chkfreq, CTLTYPE_INT|CTLFLAG_RW,
149 &acctchkfreq, 0, sysctl_acct_chkfreq, "I",
150 "frequency for checking the free space");
152 SYSCTL_INT(_kern, OID_AUTO, acct_configured, CTLFLAG_RD, &acct_configured, 0,
153 "Accounting configured or not");
155 SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0,
156 "Accounting suspended or not");
159 * Accounting system call. Written based on the specification and
160 * previous implementation done by Mark Tinguely.
165 acct(struct thread *td, struct acct_args *uap)
168 int error, flags, vfslocked;
170 error = priv_check(td, PRIV_ACCT);
175 * If accounting is to be started to a file, open that file for
176 * appending and make sure it's a 'normal'.
178 if (uap->path != NULL) {
179 NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE | AUDITVNODE1,
180 UIO_USERSPACE, uap->path, td);
181 flags = FWRITE | O_APPEND;
182 error = vn_open(&nd, &flags, 0, -1);
185 vfslocked = NDHASGIANT(&nd);
186 NDFREE(&nd, NDF_ONLY_PNBUF);
188 error = mac_check_system_acct(td->td_ucred, nd.ni_vp);
190 VOP_UNLOCK(nd.ni_vp, 0, td);
191 vn_close(nd.ni_vp, flags, td->td_ucred, td);
192 VFS_UNLOCK_GIANT(vfslocked);
196 VOP_UNLOCK(nd.ni_vp, 0, td);
197 if (nd.ni_vp->v_type != VREG) {
198 vn_close(nd.ni_vp, flags, td->td_ucred, td);
199 VFS_UNLOCK_GIANT(vfslocked);
202 VFS_UNLOCK_GIANT(vfslocked);
205 error = mac_check_system_acct(td->td_ucred, NULL);
212 * Disallow concurrent access to the accounting vnode while we swap
213 * it out, in order to prevent access after close.
218 * If accounting was previously enabled, kill the old space-watcher,
219 * close the file, and (if no new file was specified, leave). Reset
220 * the suspended state regardless of whether accounting remains
224 if (acct_vp != NULL) {
225 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
226 error = acct_disable(td);
227 VFS_UNLOCK_GIANT(vfslocked);
229 if (uap->path == NULL) {
230 if (acct_state & ACCT_RUNNING) {
231 acct_state |= ACCT_EXITREQ;
234 sx_xunlock(&acct_sx);
239 * Save the new accounting file vnode, and schedule the new
240 * free space watcher.
243 acct_cred = crhold(td->td_ucred);
245 if (acct_state & ACCT_RUNNING)
246 acct_state &= ~ACCT_EXITREQ;
249 * Try to start up an accounting kthread. We may start more
250 * than one, but if so the extras will commit suicide as
251 * soon as they start up.
253 error = kthread_create(acct_thread, NULL, NULL, 0, 0,
256 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
257 (void) vn_close(acct_vp, acct_flags, acct_cred, td);
258 VFS_UNLOCK_GIANT(vfslocked);
264 sx_xunlock(&acct_sx);
265 log(LOG_NOTICE, "Unable to start accounting thread\n");
270 sx_xunlock(&acct_sx);
271 log(LOG_NOTICE, "Accounting enabled\n");
276 * Disable currently in-progress accounting by closing the vnode, dropping
277 * our reference to the credential, and clearing the vnode's flags.
280 acct_disable(struct thread *td)
284 sx_assert(&acct_sx, SX_XLOCKED);
285 error = vn_close(acct_vp, acct_flags, acct_cred, td);
291 log(LOG_NOTICE, "Accounting disabled\n");
296 * Write out process accounting information, on process exit.
297 * Data to be written out is specified in Leffler, et al.
298 * and are enumerated below. (They're also noted in the system
299 * "acct.h" header file.)
302 acct_process(struct thread *td)
305 struct timeval ut, st, tmp;
306 struct plimit *newlim, *oldlim;
309 int t, ret, vfslocked;
312 * Lockless check of accounting condition before doing the hard
315 if (acct_vp == NULL || acct_suspended)
321 * If accounting isn't enabled, don't bother. Have to check again
322 * once we own the lock in case we raced with disabling of accounting
325 if (acct_vp == NULL || acct_suspended) {
326 sx_sunlock(&acct_sx);
333 * Get process accounting information.
337 /* (1) The name of the command that ran */
338 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
340 /* (2) The amount of user and system time that was used */
342 acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
343 acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
345 /* (3) The elapsed time the command ran (and its starting time) */
347 timevaladd(&tmp, &p->p_stats->p_start);
348 acct.ac_btime = tmp.tv_sec;
350 timevalsub(&tmp, &p->p_stats->p_start);
351 acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
353 /* (4) The average amount of memory used */
354 r = &p->p_stats->p_ru;
356 timevaladd(&tmp, &st);
357 t = tmp.tv_sec * hz + tmp.tv_usec / tick;
359 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
363 /* (5) The number of disk I/O operations done */
364 acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
366 /* (6) The UID and GID of the process */
367 acct.ac_uid = p->p_ucred->cr_ruid;
368 acct.ac_gid = p->p_ucred->cr_rgid;
370 /* (7) The terminal from which the process was started */
371 SESS_LOCK(p->p_session);
372 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
373 acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev);
376 SESS_UNLOCK(p->p_session);
378 /* (8) The boolean flags that tell how the process terminated, etc. */
379 acct.ac_flag = p->p_acflag;
383 * Eliminate any file size rlimit.
385 newlim = lim_alloc();
388 lim_copy(newlim, oldlim);
389 newlim->pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
395 * Write the accounting information to the file.
397 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
398 VOP_LEASE(acct_vp, td, acct_cred, LEASE_WRITE);
399 ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct),
400 (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED,
402 VFS_UNLOCK_GIANT(vfslocked);
403 sx_sunlock(&acct_sx);
408 * Encode_comp_t converts from ticks in seconds and microseconds
409 * to ticks in 1/AHZ seconds. The encoding is described in
410 * Leffler, et al., on page 63.
413 #define MANTSIZE 13 /* 13 bit mantissa. */
414 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
415 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
418 encode_comp_t(u_long s, u_long us)
425 s += us / (1000000 / AHZ); /* Maximize precision. */
427 while (s > MAXFRACT) {
428 rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */
429 s >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */
433 /* If we need to round up, do it (and handle overflow correctly). */
434 if (rnd && (++s > MAXFRACT)) {
439 /* Clean it up and polish it off. */
440 exp <<= MANTSIZE; /* Shift the exponent into place */
441 exp += s; /* and add on the mantissa. */
446 * Periodically check the filesystem to see if accounting
447 * should be turned on or off. Beware the case where the vnode
448 * has been vgone()'d out from underneath us, e.g. when the file
449 * system containing the accounting file has been forcibly unmounted.
458 sx_assert(&acct_sx, SX_XLOCKED);
461 * If accounting was disabled before our kthread was scheduled,
462 * then acct_vp might be NULL. If so, just ask our kthread to
465 if (acct_vp == NULL) {
466 acct_state |= ACCT_EXITREQ;
471 * If our vnode is no longer valid, tear it down and signal the
472 * accounting thread to die.
474 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
475 if (acct_vp->v_type == VBAD) {
476 (void) acct_disable(NULL);
477 VFS_UNLOCK_GIANT(vfslocked);
478 acct_state |= ACCT_EXITREQ;
483 * Stopping here is better than continuing, maybe it will be VBAD
486 if (VFS_STATFS(acct_vp->v_mount, &sb, curthread) < 0) {
487 VFS_UNLOCK_GIANT(vfslocked);
490 VFS_UNLOCK_GIANT(vfslocked);
491 if (acct_suspended) {
492 if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks /
495 log(LOG_NOTICE, "Accounting resumed\n");
498 if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks /
501 log(LOG_NOTICE, "Accounting suspended\n");
507 * The main loop for the dedicated kernel thread that periodically calls
511 acct_thread(void *dummy)
515 /* This is a low-priority kernel thread. */
517 mtx_lock_spin(&sched_lock);
518 sched_prio(curthread, pri);
519 mtx_unlock_spin(&sched_lock);
521 /* If another accounting kthread is already running, just die. */
523 if (acct_state & ACCT_RUNNING) {
524 sx_xunlock(&acct_sx);
527 acct_state |= ACCT_RUNNING;
529 /* Loop until we are asked to exit. */
530 while (!(acct_state & ACCT_EXITREQ)) {
532 /* Perform our periodic checks. */
536 * We check this flag again before sleeping since the
537 * acctwatch() might have shut down accounting and asked us
540 if (!(acct_state & ACCT_EXITREQ)) {
541 sx_xunlock(&acct_sx);
542 tsleep(&acct_state, pri, "-", acctchkfreq * hz);
548 * Acknowledge the exit request and shutdown. We clear both the
549 * exit request and running flags.
552 sx_xunlock(&acct_sx);