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>
57 #include <sys/mount.h>
58 #include <sys/mutex.h>
59 #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>
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)
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.
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
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 *);
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.
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;
103 SX_SYSINIT(acct, &acct_sx, "acct_sx");
106 * State of the accounting kthread.
108 static int acct_state;
110 #define ACCT_RUNNING 1 /* Accounting kthread is running. */
111 #define ACCT_EXITREQ 2 /* Accounting kthread should exit. */
114 * Values associated with enabling and disabling accounting
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");
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");
124 static int acctchkfreq = 15; /* frequency (in seconds) to check space */
127 sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS)
131 /* Write out the old value. */
132 error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int));
133 if (error || req->newptr == NULL)
136 /* Read in and verify the new value. */
137 error = SYSCTL_IN(req, &value, sizeof(int));
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");
149 SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0,
150 "Accounting suspended or not");
153 * Accounting system call. Written based on the specification and
154 * previous implementation done by Mark Tinguely.
159 acct(struct thread *td, struct acct_args *uap)
162 int error, flags, vfslocked;
164 /* Make sure that the caller is root. */
170 * If accounting is to be started to a file, open that file for
171 * appending and make sure it's a 'normal'.
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);
180 vfslocked = NDHASGIANT(&nd);
181 NDFREE(&nd, NDF_ONLY_PNBUF);
183 error = mac_check_system_acct(td->td_ucred, nd.ni_vp);
185 VOP_UNLOCK(nd.ni_vp, 0, td);
186 vn_close(nd.ni_vp, flags, td->td_ucred, td);
187 VFS_UNLOCK_GIANT(vfslocked);
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);
197 VFS_UNLOCK_GIANT(vfslocked);
200 error = mac_check_system_acct(td->td_ucred, NULL);
207 * Disallow concurrent access to the accounting vnode while we swap
208 * it out, in order to prevent access after close.
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
219 if (acct_vp != NULL) {
220 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
221 error = acct_disable(td);
222 VFS_UNLOCK_GIANT(vfslocked);
224 if (uap->path == NULL) {
225 if (acct_state & ACCT_RUNNING) {
226 acct_state |= ACCT_EXITREQ;
229 sx_xunlock(&acct_sx);
234 * Save the new accounting file vnode, and schedule the new
235 * free space watcher.
238 acct_cred = crhold(td->td_ucred);
240 if (acct_state & ACCT_RUNNING)
241 acct_state &= ~ACCT_EXITREQ;
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.
248 error = kthread_create(acct_thread, NULL, NULL, 0, 0,
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);
258 sx_xunlock(&acct_sx);
259 log(LOG_NOTICE, "Unable to start accounting thread\n");
263 sx_xunlock(&acct_sx);
264 log(LOG_NOTICE, "Accounting enabled\n");
269 * Disable currently in-progress accounting by closing the vnode, dropping
270 * our reference to the credential, and clearing the vnode's flags.
273 acct_disable(struct thread *td)
277 sx_assert(&acct_sx, SX_XLOCKED);
278 error = vn_close(acct_vp, acct_flags, acct_cred, td);
283 log(LOG_NOTICE, "Accounting disabled\n");
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.)
294 acct_process(struct thread *td)
297 struct timeval ut, st, tmp;
298 struct plimit *newlim, *oldlim;
301 int t, ret, vfslocked;
304 * Lockless check of accounting condition before doing the hard
307 if (acct_vp == NULL || acct_suspended)
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
317 if (acct_vp == NULL || acct_suspended) {
318 sx_sunlock(&acct_sx);
325 * Get process accounting information.
329 /* (1) The name of the command that ran */
330 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
332 /* (2) The amount of user and system time that was used */
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);
337 /* (3) The elapsed time the command ran (and its starting time) */
339 timevaladd(&tmp, &p->p_stats->p_start);
340 acct.ac_btime = tmp.tv_sec;
342 timevalsub(&tmp, &p->p_stats->p_start);
343 acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
345 /* (4) The average amount of memory used */
346 r = &p->p_stats->p_ru;
348 timevaladd(&tmp, &st);
349 t = tmp.tv_sec * hz + tmp.tv_usec / tick;
351 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
355 /* (5) The number of disk I/O operations done */
356 acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
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;
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);
368 SESS_UNLOCK(p->p_session);
370 /* (8) The boolean flags that tell how the process terminated, etc. */
371 acct.ac_flag = p->p_acflag;
375 * Eliminate any file size rlimit.
377 newlim = lim_alloc();
380 lim_copy(newlim, oldlim);
381 newlim->pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
387 * Write the accounting information to the file.
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,
394 VFS_UNLOCK_GIANT(vfslocked);
395 sx_sunlock(&acct_sx);
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.
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. */
410 encode_comp_t(u_long s, u_long us)
417 s += us / (1000000 / AHZ); /* Maximize precision. */
419 while (s > MAXFRACT) {
420 rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */
421 s >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */
425 /* If we need to round up, do it (and handle overflow correctly). */
426 if (rnd && (++s > MAXFRACT)) {
431 /* Clean it up and polish it off. */
432 exp <<= MANTSIZE; /* Shift the exponent into place */
433 exp += s; /* and add on the mantissa. */
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.
450 sx_assert(&acct_sx, SX_XLOCKED);
453 * If accounting was disabled before our kthread was scheduled,
454 * then acct_vp might be NULL. If so, just ask our kthread to
457 if (acct_vp == NULL) {
458 acct_state |= ACCT_EXITREQ;
463 * If our vnode is no longer valid, tear it down and signal the
464 * accounting thread to die.
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;
475 * Stopping here is better than continuing, maybe it will be VBAD
478 if (VFS_STATFS(acct_vp->v_mount, &sb, curthread) < 0) {
479 VFS_UNLOCK_GIANT(vfslocked);
482 VFS_UNLOCK_GIANT(vfslocked);
483 if (acct_suspended) {
484 if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks /
487 log(LOG_NOTICE, "Accounting resumed\n");
490 if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks /
493 log(LOG_NOTICE, "Accounting suspended\n");
499 * The main loop for the dedicated kernel thread that periodically calls
503 acct_thread(void *dummy)
507 /* This is a low-priority kernel thread. */
509 mtx_lock_spin(&sched_lock);
510 sched_prio(curthread, pri);
511 mtx_unlock_spin(&sched_lock);
513 /* If another accounting kthread is already running, just die. */
515 if (acct_state & ACCT_RUNNING) {
516 sx_xunlock(&acct_sx);
519 acct_state |= ACCT_RUNNING;
521 /* Loop until we are asked to exit. */
522 while (!(acct_state & ACCT_EXITREQ)) {
524 /* Perform our periodic checks. */
528 * We check this flag again before sleeping since the
529 * acctwatch() might have shut down accounting and asked us
532 if (!(acct_state & ACCT_EXITREQ)) {
533 sx_xunlock(&acct_sx);
534 tsleep(&acct_state, pri, "-", acctchkfreq * hz);
540 * Acknowledge the exit request and shutdown. We clear both the
541 * exit request and running flags.
544 sx_xunlock(&acct_sx);