kern_acct.c revision 1.54.2.4 1 /* $NetBSD: kern_acct.c,v 1.54.2.4 2004/09/21 13:35:03 skrll Exp $ */
2
3 /*-
4 * Copyright (c) 1982, 1986, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)kern_acct.c 8.8 (Berkeley) 5/14/95
37 */
38
39 /*-
40 * Copyright (c) 1994 Christopher G. Demetriou
41 *
42 * Redistribution and use in source and binary forms, with or without
43 * modification, are permitted provided that the following conditions
44 * are met:
45 * 1. Redistributions of source code must retain the above copyright
46 * notice, this list of conditions and the following disclaimer.
47 * 2. Redistributions in binary form must reproduce the above copyright
48 * notice, this list of conditions and the following disclaimer in the
49 * documentation and/or other materials provided with the distribution.
50 * 3. All advertising materials mentioning features or use of this software
51 * must display the following acknowledgement:
52 * This product includes software developed by the University of
53 * California, Berkeley and its contributors.
54 * 4. Neither the name of the University nor the names of its contributors
55 * may be used to endorse or promote products derived from this software
56 * without specific prior written permission.
57 *
58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68 * SUCH DAMAGE.
69 *
70 * @(#)kern_acct.c 8.8 (Berkeley) 5/14/95
71 */
72
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: kern_acct.c,v 1.54.2.4 2004/09/21 13:35:03 skrll Exp $");
75
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/proc.h>
79 #include <sys/mount.h>
80 #include <sys/vnode.h>
81 #include <sys/file.h>
82 #include <sys/syslog.h>
83 #include <sys/kernel.h>
84 #include <sys/kthread.h>
85 #include <sys/lock.h>
86 #include <sys/malloc.h>
87 #include <sys/namei.h>
88 #include <sys/errno.h>
89 #include <sys/acct.h>
90 #include <sys/resourcevar.h>
91 #include <sys/ioctl.h>
92 #include <sys/tty.h>
93
94 #include <sys/sa.h>
95 #include <sys/syscallargs.h>
96
97 /*
98 * The routines implemented in this file are described in:
99 * Leffler, et al.: The Design and Implementation of the 4.3BSD
100 * UNIX Operating System (Addison Welley, 1989)
101 * on pages 62-63.
102 *
103 * Arguably, to simplify accounting operations, this mechanism should
104 * be replaced by one in which an accounting log file (similar to /dev/klog)
105 * is read by a user process, etc. However, that has its own problems.
106 */
107
108 /*
109 * The global accounting state and related data. Gain the lock before
110 * accessing these variables.
111 */
112 enum {
113 ACCT_STOP,
114 ACCT_ACTIVE,
115 ACCT_SUSPENDED
116 } acct_state; /* The current accounting state. */
117 struct vnode *acct_vp; /* Accounting vnode pointer. */
118 struct ucred *acct_ucred; /* Credential of accounting file
119 owner (i.e root). Used when
120 accounting file i/o. */
121 struct proc *acct_dkwatcher; /* Free disk space checker. */
122
123 /*
124 * Lock to serialize system calls and kernel threads.
125 */
126 struct lock acct_lock;
127 #define ACCT_LOCK() \
128 do { \
129 (void) lockmgr(&acct_lock, LK_EXCLUSIVE, NULL); \
130 } while (/* CONSTCOND */0)
131 #define ACCT_UNLOCK() \
132 do { \
133 (void) lockmgr(&acct_lock, LK_RELEASE, NULL); \
134 } while (/* CONSTCOND */0)
135
136 /*
137 * Internal accounting functions.
138 * The former's operation is described in Leffler, et al., and the latter
139 * was provided by UCB with the 4.4BSD-Lite release
140 */
141 comp_t encode_comp_t(u_long, u_long);
142 void acctwatch(void *);
143 void acct_stop(void);
144 int acct_chkfree(void);
145
146 /*
147 * Values associated with enabling and disabling accounting
148 */
149 int acctsuspend = 2; /* stop accounting when < 2% free space left */
150 int acctresume = 4; /* resume when free space risen to > 4% */
151 int acctchkfreq = 15; /* frequency (in seconds) to check space */
152
153 void
154 acct_init()
155 {
156
157 acct_state = ACCT_STOP;
158 acct_vp = NULLVP;
159 acct_ucred = NULL;
160 lockinit(&acct_lock, PWAIT, "acctlk", 0, 0);
161 }
162
163 void
164 acct_stop()
165 {
166 int error;
167
168 if (acct_vp != NULLVP && acct_vp->v_type != VBAD) {
169 error = vn_close(acct_vp, FWRITE, acct_ucred, NULL);
170 #ifdef DIAGNOSTIC
171 if (error != 0)
172 printf("acct_stop: failed to close, errno = %d\n",
173 error);
174 #endif
175 acct_vp = NULLVP;
176 }
177 if (acct_ucred != NULL) {
178 crfree(acct_ucred);
179 acct_ucred = NULL;
180 }
181 acct_state = ACCT_STOP;
182 }
183
184 int
185 acct_chkfree()
186 {
187 int error;
188 struct statvfs sb;
189 int64_t bavail;
190
191 error = VFS_STATVFS(acct_vp->v_mount, &sb, NULL);
192 if (error != 0)
193 return (error);
194
195 bavail = sb.f_bfree - sb.f_bresvd;
196
197 switch (acct_state) {
198 case ACCT_SUSPENDED:
199 if (bavail > acctresume * sb.f_blocks / 100) {
200 acct_state = ACCT_ACTIVE;
201 log(LOG_NOTICE, "Accounting resumed\n");
202 }
203 break;
204 case ACCT_ACTIVE:
205 if (bavail <= acctsuspend * sb.f_blocks / 100) {
206 acct_state = ACCT_SUSPENDED;
207 log(LOG_NOTICE, "Accounting suspended\n");
208 }
209 break;
210 case ACCT_STOP:
211 break;
212 }
213 return (0);
214 }
215
216 /*
217 * Accounting system call. Written based on the specification and
218 * previous implementation done by Mark Tinguely.
219 */
220 int
221 sys_acct(l, v, retval)
222 struct lwp *l;
223 void *v;
224 register_t *retval;
225 {
226 struct sys_acct_args /* {
227 syscallarg(const char *) path;
228 } */ *uap = v;
229 struct nameidata nd;
230 int error;
231 struct proc *p = l->l_proc;
232
233 /* Make sure that the caller is root. */
234 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
235 return (error);
236
237 /*
238 * If accounting is to be started to a file, open that file for
239 * writing and make sure it's a 'normal'.
240 */
241 if (SCARG(uap, path) != NULL) {
242 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path),
243 l);
244 if ((error = vn_open(&nd, FWRITE|O_APPEND, 0)) != 0)
245 return (error);
246 VOP_UNLOCK(nd.ni_vp, 0);
247 if (nd.ni_vp->v_type != VREG) {
248 vn_close(nd.ni_vp, FWRITE, p->p_ucred, l);
249 return (EACCES);
250 }
251 }
252
253 ACCT_LOCK();
254
255 /*
256 * If accounting was previously enabled, kill the old space-watcher,
257 * free credential for accounting file i/o,
258 * ... (and, if no new file was specified, leave).
259 */
260 acct_stop();
261 if (SCARG(uap, path) == NULL)
262 goto out;
263
264 /*
265 * Save the new accounting file vnode and credential,
266 * and schedule the new free space watcher.
267 */
268 acct_state = ACCT_ACTIVE;
269 acct_vp = nd.ni_vp;
270 acct_ucred = p->p_ucred;
271 crhold(acct_ucred);
272
273 error = acct_chkfree(); /* Initial guess. */
274 if (error != 0) {
275 acct_stop();
276 goto out;
277 }
278
279 if (acct_dkwatcher == NULL) {
280 error = kthread_create1(acctwatch, NULL, &acct_dkwatcher,
281 "acctwatch");
282 if (error != 0)
283 acct_stop();
284 }
285
286 out:
287 ACCT_UNLOCK();
288 return (error);
289 }
290
291 /*
292 * Write out process accounting information, on process exit.
293 * Data to be written out is specified in Leffler, et al.
294 * and are enumerated below. (They're also noted in the system
295 * "acct.h" header file.)
296 */
297 int
298 acct_process(l)
299 struct lwp *l;
300 {
301 struct acct acct;
302 struct rusage *r;
303 struct timeval ut, st, tmp;
304 int s, t, error = 0;
305 struct plimit *oplim = NULL;
306 struct proc *p = l->l_proc;
307
308 ACCT_LOCK();
309
310 /* If accounting isn't enabled, don't bother */
311 if (acct_state != ACCT_ACTIVE)
312 goto out;
313
314 /*
315 * Raise the file limit so that accounting can't be stopped by
316 * the user.
317 *
318 * XXX We should think about the CPU limit, too.
319 */
320 if (p->p_limit->p_refcnt > 1) {
321 oplim = p->p_limit;
322 p->p_limit = limcopy(p->p_limit);
323 }
324 p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
325
326 /*
327 * Get process accounting information.
328 */
329
330 /* (1) The name of the command that ran */
331 memcpy(acct.ac_comm, p->p_comm, sizeof(acct.ac_comm));
332
333 /* (2) The amount of user and system time that was used */
334 calcru(p, &ut, &st, NULL);
335 acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
336 acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
337
338 /* (3) The elapsed time the commmand ran (and its starting time) */
339 acct.ac_btime = p->p_stats->p_start.tv_sec;
340 s = splclock();
341 timersub(&time, &p->p_stats->p_start, &tmp);
342 splx(s);
343 acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
344
345 /* (4) The average amount of memory used */
346 r = &p->p_stats->p_ru;
347 timeradd(&ut, &st, &tmp);
348 t = tmp.tv_sec * hz + tmp.tv_usec / tick;
349 if (t)
350 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
351 else
352 acct.ac_mem = 0;
353
354 /* (5) The number of disk I/O operations done */
355 acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
356
357 /* (6) The UID and GID of the process */
358 acct.ac_uid = p->p_cred->p_ruid;
359 acct.ac_gid = p->p_cred->p_rgid;
360
361 /* (7) The terminal from which the process was started */
362 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
363 acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
364 else
365 acct.ac_tty = NODEV;
366
367 /* (8) The boolean flags that tell how the process terminated, etc. */
368 acct.ac_flag = p->p_acflag;
369
370 /*
371 * Now, just write the accounting information to the file.
372 */
373 VOP_LEASE(acct_vp, l, p->p_ucred, LEASE_WRITE);
374 error = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct,
375 sizeof(acct), (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT,
376 acct_ucred, NULL, NULL);
377 if (error != 0)
378 log(LOG_ERR, "Accounting: write failed %d\n", error);
379
380 if (oplim) {
381 limfree(p->p_limit);
382 p->p_limit = oplim;
383 }
384
385 out:
386 ACCT_UNLOCK();
387 return (error);
388 }
389
390 /*
391 * Encode_comp_t converts from ticks in seconds and microseconds
392 * to ticks in 1/AHZ seconds. The encoding is described in
393 * Leffler, et al., on page 63.
394 */
395
396 #define MANTSIZE 13 /* 13 bit mantissa. */
397 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
398 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
399
400 comp_t
401 encode_comp_t(s, us)
402 u_long s, us;
403 {
404 int exp, rnd;
405
406 exp = 0;
407 rnd = 0;
408 s *= AHZ;
409 s += us / (1000000 / AHZ); /* Maximize precision. */
410
411 while (s > MAXFRACT) {
412 rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */
413 s >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */
414 exp++;
415 }
416
417 /* If we need to round up, do it (and handle overflow correctly). */
418 if (rnd && (++s > MAXFRACT)) {
419 s >>= EXPSIZE;
420 exp++;
421 }
422
423 /* Clean it up and polish it off. */
424 exp <<= MANTSIZE; /* Shift the exponent into place */
425 exp += s; /* and add on the mantissa. */
426 return (exp);
427 }
428
429 /*
430 * Periodically check the file system to see if accounting
431 * should be turned on or off. Beware the case where the vnode
432 * has been vgone()'d out from underneath us, e.g. when the file
433 * system containing the accounting file has been forcibly unmounted.
434 */
435 void
436 acctwatch(arg)
437 void *arg;
438 {
439 int error;
440
441 log(LOG_NOTICE, "Accounting started\n");
442 ACCT_LOCK();
443 while (acct_state != ACCT_STOP) {
444 if (acct_vp->v_type == VBAD) {
445 log(LOG_NOTICE, "Accounting terminated\n");
446 acct_stop();
447 continue;
448 }
449
450 error = acct_chkfree();
451 #ifdef DIAGNOSTIC
452 if (error != 0)
453 printf("acctwatch: failed to statvfs, error = %d\n",
454 error);
455 #endif
456
457 ACCT_UNLOCK();
458 error = tsleep(acctwatch, PSWP, "actwat", acctchkfreq * hz);
459 ACCT_LOCK();
460 #ifdef DIAGNOSTIC
461 if (error != 0 && error != EWOULDBLOCK)
462 printf("acctwatch: sleep error %d\n", error);
463 #endif
464 }
465 acct_dkwatcher = NULL;
466 ACCT_UNLOCK();
467
468 kthread_exit(0);
469 }
470