FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_acct.c
1 /*-
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.
9 *
10 * Copyright (c) 1994 Christopher G. Demetriou
11 * Copyright (c) 2005 Robert N. M. Watson
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
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.
28 *
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
39 * SUCH DAMAGE.
40 *
41 * @(#)kern_acct.c 8.1 (Berkeley) 6/14/93
42 */
43
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD: releng/6.1/sys/kern/kern_acct.c 158179 2006-04-30 16:44:43Z cvs2svn $");
46
47 #include "opt_mac.h"
48
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/acct.h>
52 #include <sys/fcntl.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/lock.h>
56 #include <sys/mac.h>
57 #include <sys/mount.h>
58 #include <sys/mutex.h>
59 #include <sys/namei.h>
60 #include <sys/proc.h>
61 #include <sys/resourcevar.h>
62 #include <sys/sched.h>
63 #include <sys/sx.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
66 #include <sys/syslog.h>
67 #include <sys/sysproto.h>
68 #include <sys/tty.h>
69 #include <sys/vnode.h>
70
71 /*
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)
75 * on pages 62-63.
76 *
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.
80 */
81
82 /*
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
86 */
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 *);
91
92 /*
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.
96 */
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;
102
103 SX_SYSINIT(acct, &acct_sx, "acct_sx");
104
105 /*
106 * State of the accounting kthread.
107 */
108 static int acct_state;
109
110 #define ACCT_RUNNING 1 /* Accounting kthread is running. */
111 #define ACCT_EXITREQ 2 /* Accounting kthread should exit. */
112
113 /*
114 * Values associated with enabling and disabling accounting
115 */
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");
119
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");
123
124 static int acctchkfreq = 15; /* frequency (in seconds) to check space */
125
126 static int
127 sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS)
128 {
129 int error, value;
130
131 /* Write out the old value. */
132 error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int));
133 if (error || req->newptr == NULL)
134 return (error);
135
136 /* Read in and verify the new value. */
137 error = SYSCTL_IN(req, &value, sizeof(int));
138 if (error)
139 return (error);
140 if (value <= 0)
141 return (EINVAL);
142 acctchkfreq = value;
143 return (0);
144 }
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");
148
149 SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0,
150 "Accounting suspended or not");
151
152 /*
153 * Accounting system call. Written based on the specification and
154 * previous implementation done by Mark Tinguely.
155 *
156 * MPSAFE
157 */
158 int
159 acct(struct thread *td, struct acct_args *uap)
160 {
161 struct nameidata nd;
162 int error, flags;
163
164 /* Make sure that the caller is root. */
165 error = suser(td);
166 if (error)
167 return (error);
168
169 /*
170 * If accounting is to be started to a file, open that file for
171 * appending and make sure it's a 'normal'. While we could
172 * conditionally acquire Giant here, we're actually interacting with
173 * vnodes from possibly two file systems, making the logic a bit
174 * complicated. For now, use Giant unconditionally.
175 */
176 mtx_lock(&Giant);
177 if (uap->path != NULL) {
178 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->path, td);
179 flags = FWRITE | O_APPEND;
180 error = vn_open(&nd, &flags, 0, -1);
181 if (error)
182 goto done;
183 NDFREE(&nd, NDF_ONLY_PNBUF);
184 #ifdef MAC
185 error = mac_check_system_acct(td->td_ucred, nd.ni_vp);
186 if (error) {
187 VOP_UNLOCK(nd.ni_vp, 0, td);
188 vn_close(nd.ni_vp, flags, td->td_ucred, td);
189 goto done;
190 }
191 #endif
192 VOP_UNLOCK(nd.ni_vp, 0, td);
193 if (nd.ni_vp->v_type != VREG) {
194 vn_close(nd.ni_vp, flags, td->td_ucred, td);
195 error = EACCES;
196 goto done;
197 }
198 #ifdef MAC
199 } else {
200 error = mac_check_system_acct(td->td_ucred, NULL);
201 if (error)
202 goto done;
203 #endif
204 }
205
206 /*
207 * Disallow concurrent access to the accounting vnode while we swap
208 * it out, in order to prevent access after close.
209 */
210 sx_xlock(&acct_sx);
211
212 /*
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
216 * enabled.
217 */
218 acct_suspended = 0;
219 if (acct_vp != NULL)
220 error = acct_disable(td);
221 if (uap->path == NULL) {
222 if (acct_state & ACCT_RUNNING) {
223 acct_state |= ACCT_EXITREQ;
224 wakeup(&acct_state);
225 }
226 sx_xunlock(&acct_sx);
227 goto done;
228 }
229
230 /*
231 * Save the new accounting file vnode, and schedule the new
232 * free space watcher.
233 */
234 acct_vp = nd.ni_vp;
235 acct_cred = crhold(td->td_ucred);
236 acct_flags = flags;
237 if (acct_state & ACCT_RUNNING)
238 acct_state &= ~ACCT_EXITREQ;
239 else {
240 /*
241 * Try to start up an accounting kthread. We may start more
242 * than one, but if so the extras will commit suicide as
243 * soon as they start up.
244 */
245 error = kthread_create(acct_thread, NULL, NULL, 0, 0,
246 "accounting");
247 if (error) {
248 (void) vn_close(acct_vp, acct_flags, acct_cred, td);
249 crfree(acct_cred);
250 acct_vp = NULL;
251 acct_cred = NULL;
252 acct_flags = 0;
253 sx_xunlock(&acct_sx);
254 log(LOG_NOTICE, "Unable to start accounting thread\n");
255 goto done;
256 }
257 }
258 sx_xunlock(&acct_sx);
259 log(LOG_NOTICE, "Accounting enabled\n");
260 done:
261 mtx_unlock(&Giant);
262 return (error);
263 }
264
265 /*
266 * Disable currently in-progress accounting by closing the vnode, dropping
267 * our reference to the credential, and clearing the vnode's flags.
268 */
269 static int
270 acct_disable(struct thread *td)
271 {
272 int error;
273
274 sx_assert(&acct_sx, SX_XLOCKED);
275 error = vn_close(acct_vp, acct_flags, acct_cred, td);
276 crfree(acct_cred);
277 acct_vp = NULL;
278 acct_cred = NULL;
279 acct_flags = 0;
280 log(LOG_NOTICE, "Accounting disabled\n");
281 return (error);
282 }
283
284 /*
285 * Write out process accounting information, on process exit.
286 * Data to be written out is specified in Leffler, et al.
287 * and are enumerated below. (They're also noted in the system
288 * "acct.h" header file.)
289 */
290 int
291 acct_process(struct thread *td)
292 {
293 struct acct acct;
294 struct timeval ut, st, tmp;
295 struct plimit *newlim, *oldlim;
296 struct proc *p;
297 struct rusage *r;
298 int t, ret, vfslocked;
299
300 /*
301 * Lockless check of accounting condition before doing the hard
302 * work.
303 */
304 if (acct_vp == NULL || acct_suspended)
305 return (0);
306
307 sx_slock(&acct_sx);
308
309 /*
310 * If accounting isn't enabled, don't bother. Have to check again
311 * once we own the lock in case we raced with disabling of accounting
312 * by another thread.
313 */
314 if (acct_vp == NULL || acct_suspended) {
315 sx_sunlock(&acct_sx);
316 return (0);
317 }
318
319 p = td->td_proc;
320
321 /*
322 * Get process accounting information.
323 */
324
325 PROC_LOCK(p);
326 /* (1) The name of the command that ran */
327 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
328
329 /* (2) The amount of user and system time that was used */
330 calcru(p, &ut, &st);
331 acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
332 acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
333
334 /* (3) The elapsed time the command ran (and its starting time) */
335 tmp = boottime;
336 timevaladd(&tmp, &p->p_stats->p_start);
337 acct.ac_btime = tmp.tv_sec;
338 microuptime(&tmp);
339 timevalsub(&tmp, &p->p_stats->p_start);
340 acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
341
342 /* (4) The average amount of memory used */
343 r = &p->p_stats->p_ru;
344 tmp = ut;
345 timevaladd(&tmp, &st);
346 t = tmp.tv_sec * hz + tmp.tv_usec / tick;
347 if (t)
348 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
349 else
350 acct.ac_mem = 0;
351
352 /* (5) The number of disk I/O operations done */
353 acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
354
355 /* (6) The UID and GID of the process */
356 acct.ac_uid = p->p_ucred->cr_ruid;
357 acct.ac_gid = p->p_ucred->cr_rgid;
358
359 /* (7) The terminal from which the process was started */
360 SESS_LOCK(p->p_session);
361 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
362 acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev);
363 else
364 acct.ac_tty = NODEV;
365 SESS_UNLOCK(p->p_session);
366
367 /* (8) The boolean flags that tell how the process terminated, etc. */
368 acct.ac_flag = p->p_acflag;
369 PROC_UNLOCK(p);
370
371 /*
372 * Eliminate any file size rlimit.
373 */
374 newlim = lim_alloc();
375 PROC_LOCK(p);
376 oldlim = p->p_limit;
377 lim_copy(newlim, oldlim);
378 newlim->pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
379 p->p_limit = newlim;
380 PROC_UNLOCK(p);
381 lim_free(oldlim);
382
383 /*
384 * Write the accounting information to the file.
385 */
386 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
387 VOP_LEASE(acct_vp, td, acct_cred, LEASE_WRITE);
388 ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct),
389 (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED,
390 (int *)0, td);
391 VFS_UNLOCK_GIANT(vfslocked);
392 sx_sunlock(&acct_sx);
393 return (ret);
394 }
395
396 /*
397 * Encode_comp_t converts from ticks in seconds and microseconds
398 * to ticks in 1/AHZ seconds. The encoding is described in
399 * Leffler, et al., on page 63.
400 */
401
402 #define MANTSIZE 13 /* 13 bit mantissa. */
403 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
404 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
405
406 static comp_t
407 encode_comp_t(u_long s, u_long us)
408 {
409 int exp, rnd;
410
411 exp = 0;
412 rnd = 0;
413 s *= AHZ;
414 s += us / (1000000 / AHZ); /* Maximize precision. */
415
416 while (s > MAXFRACT) {
417 rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */
418 s >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */
419 exp++;
420 }
421
422 /* If we need to round up, do it (and handle overflow correctly). */
423 if (rnd && (++s > MAXFRACT)) {
424 s >>= EXPSIZE;
425 exp++;
426 }
427
428 /* Clean it up and polish it off. */
429 exp <<= MANTSIZE; /* Shift the exponent into place */
430 exp += s; /* and add on the mantissa. */
431 return (exp);
432 }
433
434 /*
435 * Periodically check the filesystem to see if accounting
436 * should be turned on or off. Beware the case where the vnode
437 * has been vgone()'d out from underneath us, e.g. when the file
438 * system containing the accounting file has been forcibly unmounted.
439 */
440 /* ARGSUSED */
441 static void
442 acctwatch(void)
443 {
444 struct statfs sb;
445 int vfslocked;
446
447 sx_assert(&acct_sx, SX_XLOCKED);
448
449 /*
450 * If accounting was disabled before our kthread was scheduled,
451 * then acct_vp might be NULL. If so, just ask our kthread to
452 * exit and return.
453 */
454 if (acct_vp == NULL) {
455 acct_state |= ACCT_EXITREQ;
456 return;
457 }
458
459 /*
460 * If our vnode is no longer valid, tear it down and signal the
461 * accounting thread to die.
462 */
463 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount);
464 if (acct_vp->v_type == VBAD) {
465 (void) acct_disable(NULL);
466 VFS_UNLOCK_GIANT(vfslocked);
467 acct_state |= ACCT_EXITREQ;
468 return;
469 }
470
471 /*
472 * Stopping here is better than continuing, maybe it will be VBAD
473 * next time around.
474 */
475 if (VFS_STATFS(acct_vp->v_mount, &sb, curthread) < 0) {
476 VFS_UNLOCK_GIANT(vfslocked);
477 return;
478 }
479 VFS_UNLOCK_GIANT(vfslocked);
480 if (acct_suspended) {
481 if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks /
482 100)) {
483 acct_suspended = 0;
484 log(LOG_NOTICE, "Accounting resumed\n");
485 }
486 } else {
487 if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks /
488 100)) {
489 acct_suspended = 1;
490 log(LOG_NOTICE, "Accounting suspended\n");
491 }
492 }
493 }
494
495 /*
496 * The main loop for the dedicated kernel thread that periodically calls
497 * acctwatch().
498 */
499 static void
500 acct_thread(void *dummy)
501 {
502 u_char pri;
503
504 /* This is a low-priority kernel thread. */
505 pri = PRI_MAX_KERN;
506 mtx_lock_spin(&sched_lock);
507 sched_prio(curthread, pri);
508 mtx_unlock_spin(&sched_lock);
509
510 /* If another accounting kthread is already running, just die. */
511 sx_xlock(&acct_sx);
512 if (acct_state & ACCT_RUNNING) {
513 sx_xunlock(&acct_sx);
514 kthread_exit(0);
515 }
516 acct_state |= ACCT_RUNNING;
517
518 /* Loop until we are asked to exit. */
519 while (!(acct_state & ACCT_EXITREQ)) {
520
521 /* Perform our periodic checks. */
522 acctwatch();
523
524 /*
525 * We check this flag again before sleeping since the
526 * acctwatch() might have shut down accounting and asked us
527 * to exit.
528 */
529 if (!(acct_state & ACCT_EXITREQ)) {
530 sx_xunlock(&acct_sx);
531 tsleep(&acct_state, pri, "-", acctchkfreq * hz);
532 sx_xlock(&acct_sx);
533 }
534 }
535
536 /*
537 * Acknowledge the exit request and shutdown. We clear both the
538 * exit request and running flags.
539 */
540 acct_state = 0;
541 sx_xunlock(&acct_sx);
542 kthread_exit(0);
543 }
Cache object: 17a026ee0df5d0e254697e884efcd391
|