xref: /titanic_41/usr/src/uts/common/os/acct.c (revision 5aefb6555731130ca4fd295960123d71f2d21fe8)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
28 /*	  All Rights Reserved  	*/
29 
30 
31 #pragma ident	"%Z%%M%	%I%	%E% SMI"
32 
33 #include <sys/types.h>
34 #include <sys/sysmacros.h>
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/acct.h>
38 #include <sys/cred.h>
39 #include <sys/user.h>
40 #include <sys/errno.h>
41 #include <sys/file.h>
42 #include <sys/vnode.h>
43 #include <sys/debug.h>
44 #include <sys/proc.h>
45 #include <sys/resource.h>
46 #include <sys/session.h>
47 #include <sys/modctl.h>
48 #include <sys/syscall.h>
49 #include <sys/policy.h>
50 #include <sys/list.h>
51 #include <sys/time.h>
52 #include <sys/msacct.h>
53 #include <sys/zone.h>
54 
55 /*
56  * Each zone has its own accounting settings (on or off) and associated
57  * file.  The global zone is not special in this aspect; it will only
58  * generate records for processes that ran in the global zone.  We could
59  * allow the global zone to record all activity on the system, but there
60  * would be no way of knowing the zone in which the processes executed.
61  * sysacct() is thus virtualized to only act on the caller's zone.
62  */
63 struct acct_globals {
64 	struct acct	acctbuf;
65 	kmutex_t	aclock;
66 	struct vnode	*acctvp;
67 	list_node_t	aclink;
68 };
69 
70 /*
71  * We need a list of all accounting settings for all zones, so we can
72  * accurately determine if a file is in use for accounting (possibly by
73  * another zone).
74  */
75 static zone_key_t acct_zone_key;
76 static list_t acct_list;
77 kmutex_t acct_list_lock;
78 
79 static struct sysent acctsysent = {
80 	1,
81 	SE_NOUNLOAD | SE_ARGC | SE_32RVAL1,
82 	sysacct
83 };
84 
85 static struct modlsys modlsys = {
86 	&mod_syscallops, "acct(2) syscall", &acctsysent
87 };
88 
89 #ifdef _SYSCALL32_IMPL
90 static struct modlsys modlsys32 = {
91 	&mod_syscallops32, "32-bit acct(2) syscall", &acctsysent
92 };
93 #endif
94 
95 static struct modlinkage modlinkage = {
96 	MODREV_1,
97 	&modlsys,
98 #ifdef _SYSCALL32_IMPL
99 	&modlsys32,
100 #endif
101 	NULL
102 };
103 
104 /*ARGSUSED*/
105 static void *
106 acct_init(zoneid_t zoneid)
107 {
108 	struct acct_globals *ag;
109 
110 	ag = kmem_alloc(sizeof (*ag), KM_SLEEP);
111 	bzero(&ag->acctbuf, sizeof (ag->acctbuf));
112 	mutex_init(&ag->aclock, NULL, MUTEX_DEFAULT, NULL);
113 	ag->acctvp = NULL;
114 
115 	mutex_enter(&acct_list_lock);
116 	list_insert_tail(&acct_list, ag);
117 	mutex_exit(&acct_list_lock);
118 	return (ag);
119 }
120 
121 /* ARGSUSED */
122 static void
123 acct_shutdown(zoneid_t zoneid, void *arg)
124 {
125 	struct acct_globals *ag = arg;
126 
127 	mutex_enter(&ag->aclock);
128 	if (ag->acctvp) {
129 		/*
130 		 * This needs to be done as a shutdown callback, otherwise this
131 		 * held vnode may cause filesystems to be busy, and the zone
132 		 * shutdown operation to fail.
133 		 */
134 		(void) VOP_CLOSE(ag->acctvp, FWRITE, 1, (offset_t)0, kcred);
135 		VN_RELE(ag->acctvp);
136 	}
137 	ag->acctvp = NULL;
138 	mutex_exit(&ag->aclock);
139 }
140 
141 /*ARGSUSED*/
142 static void
143 acct_fini(zoneid_t zoneid, void *arg)
144 {
145 	struct acct_globals *ag = arg;
146 
147 	mutex_enter(&acct_list_lock);
148 	list_remove(&acct_list, ag);
149 	mutex_exit(&acct_list_lock);
150 
151 	mutex_destroy(&ag->aclock);
152 	kmem_free(ag, sizeof (*ag));
153 }
154 
155 int
156 _init(void)
157 {
158 	int error;
159 
160 	mutex_init(&acct_list_lock, NULL, MUTEX_DEFAULT, NULL);
161 	list_create(&acct_list, sizeof (struct acct_globals),
162 	    offsetof(struct acct_globals, aclink));
163 	/*
164 	 * Using an initializer here wastes a bit of memory for zones that
165 	 * don't use accounting, but vastly simplifies the locking.
166 	 */
167 	zone_key_create(&acct_zone_key, acct_init, acct_shutdown, acct_fini);
168 	if ((error = mod_install(&modlinkage)) != 0) {
169 		(void) zone_key_delete(acct_zone_key);
170 		list_destroy(&acct_list);
171 		mutex_destroy(&acct_list_lock);
172 	}
173 	return (error);
174 }
175 
176 int
177 _info(struct modinfo *modinfop)
178 {
179 	return (mod_info(&modlinkage, modinfop));
180 }
181 
182 /*
183  * acct() is a "weak stub" routine called from exit().
184  * Once this module has been loaded, we refuse to allow
185  * it to unload - otherwise accounting would quietly
186  * cease.  See 1211661.  It's possible to make this module
187  * unloadable but it's substantially safer not to bother.
188  */
189 int
190 _fini(void)
191 {
192 	return (EBUSY);
193 }
194 
195 /*
196  * See if vp is in use by the accounting system on any zone.  This does a deep
197  * comparison of vnodes such that a file and a lofs "shadow" node of it will
198  * appear to be the same.
199  *
200  * If 'compare_vfs' is true, the function will do a comparison of vfs_t's
201  * instead (ie, is the vfs_t on which the vnode resides in use by the
202  * accounting system in any zone).
203  *
204  * Returns 1 if found (in use), 0 otherwise.
205  */
206 static int
207 acct_find(vnode_t *vp, boolean_t compare_vfs)
208 {
209 	struct acct_globals *ag;
210 	vnode_t *realvp;
211 
212 	ASSERT(MUTEX_HELD(&acct_list_lock));
213 	ASSERT(vp != NULL);
214 
215 	if (VOP_REALVP(vp, &realvp))
216 		realvp = vp;
217 	for (ag = list_head(&acct_list); ag != NULL;
218 	    ag = list_next(&acct_list, ag)) {
219 		vnode_t *racctvp;
220 		boolean_t found = B_FALSE;
221 
222 		mutex_enter(&ag->aclock);
223 		if (ag->acctvp == NULL) {
224 			mutex_exit(&ag->aclock);
225 			continue;
226 		}
227 		if (VOP_REALVP(ag->acctvp, &racctvp))
228 			racctvp = ag->acctvp;
229 		if (compare_vfs) {
230 			if (racctvp->v_vfsp == realvp->v_vfsp)
231 				found = B_TRUE;
232 		} else {
233 			if (VN_CMP(realvp, racctvp))
234 				found = B_TRUE;
235 		}
236 		mutex_exit(&ag->aclock);
237 		if (found)
238 			return (1);
239 	}
240 	return (0);
241 }
242 
243 /*
244  * Returns 1 if the vfs that vnode resides on is in use for the accounting
245  * subsystem, 0 otherwise.
246  */
247 int
248 acct_fs_in_use(vnode_t *vp)
249 {
250 	int found;
251 
252 	if (vp == NULL)
253 		return (0);
254 	mutex_enter(&acct_list_lock);
255 	found = acct_find(vp, B_TRUE);
256 	mutex_exit(&acct_list_lock);
257 	return (found);
258 }
259 
260 /*
261  * Perform process accounting functions.
262  */
263 int
264 sysacct(char *fname)
265 {
266 	struct acct_globals *ag;
267 	struct vnode *vp;
268 	int error = 0;
269 
270 	if (secpolicy_acct(CRED()) != 0)
271 		return (set_errno(EPERM));
272 
273 	ag = zone_getspecific(acct_zone_key, curproc->p_zone);
274 	ASSERT(ag != NULL);
275 
276 	if (fname == NULL) {
277 		/*
278 		 * Close the file and stop accounting.
279 		 */
280 		mutex_enter(&ag->aclock);
281 		vp = ag->acctvp;
282 		ag->acctvp = NULL;
283 		mutex_exit(&ag->aclock);
284 		if (vp) {
285 			error = VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, CRED());
286 			VN_RELE(vp);
287 		}
288 		return (error == 0 ? 0 : set_errno(error));
289 	}
290 
291 	/*
292 	 * Either (a) open a new file and begin accounting -or- (b)
293 	 * switch accounting from an old to a new file.
294 	 *
295 	 * (Open the file without holding aclock in case it
296 	 * sleeps (holding the lock prevents process exit).)
297 	 */
298 	if ((error = vn_open(fname, UIO_USERSPACE, FWRITE,
299 	    0, &vp, (enum create)0, 0)) != 0) {
300 		/* SVID  compliance */
301 		if (error == EISDIR)
302 			error = EACCES;
303 		return (set_errno(error));
304 	}
305 
306 	if (vp->v_type != VREG) {
307 		error = EACCES;
308 	} else {
309 		mutex_enter(&acct_list_lock);
310 		if (acct_find(vp, B_FALSE)) {
311 			error = EBUSY;
312 		} else {
313 			mutex_enter(&ag->aclock);
314 			if (ag->acctvp) {
315 				vnode_t *oldvp;
316 
317 				/*
318 				 * close old acctvp, and point acct()
319 				 * at new file by swapping vp and acctvp
320 				 */
321 				oldvp = ag->acctvp;
322 				ag->acctvp = vp;
323 				vp = oldvp;
324 			} else {
325 				/*
326 				 * no existing file, start accounting ..
327 				 */
328 				ag->acctvp = vp;
329 				vp = NULL;
330 			}
331 			mutex_exit(&ag->aclock);
332 		}
333 		mutex_exit(&acct_list_lock);
334 	}
335 
336 	if (vp) {
337 		(void) VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, CRED());
338 		VN_RELE(vp);
339 	}
340 	return (error == 0 ? 0 : set_errno(error));
341 }
342 
343 /*
344  * Produce a pseudo-floating point representation
345  * with 3 bits base-8 exponent, 13 bits fraction.
346  */
347 static comp_t
348 acct_compress(ulong_t t)
349 {
350 	int exp = 0, round = 0;
351 
352 	while (t >= 8192) {
353 		exp++;
354 		round = t & 04;
355 		t >>= 3;
356 	}
357 	if (round) {
358 		t++;
359 		if (t >= 8192) {
360 			t >>= 3;
361 			exp++;
362 		}
363 	}
364 #ifdef _LP64
365 	if (exp > 7) {
366 		/* prevent wraparound */
367 		t = 8191;
368 		exp = 7;
369 	}
370 #endif
371 	return ((exp << 13) + t);
372 }
373 
374 /*
375  * On exit, write a record on the accounting file.
376  */
377 void
378 acct(char st)
379 {
380 	struct vnode *vp;
381 	struct cred *cr;
382 	struct proc *p;
383 	struct vattr va;
384 	ssize_t resid = 0;
385 	int error;
386 	struct acct_globals *ag;
387 
388 	ag = zone_getspecific(acct_zone_key, curproc->p_zone);
389 
390 	mutex_enter(&ag->aclock);
391 	if ((vp = ag->acctvp) == NULL) {
392 		mutex_exit(&ag->aclock);
393 		return;
394 	}
395 
396 	/*
397 	 * This only gets called from exit after all lwp's have exited so no
398 	 * cred locking is needed.
399 	 */
400 	p = curproc;
401 	bcopy(u.u_comm, ag->acctbuf.ac_comm, sizeof (ag->acctbuf.ac_comm));
402 	ag->acctbuf.ac_btime = u.u_start.tv_sec;
403 	ag->acctbuf.ac_utime = acct_compress(NSEC_TO_TICK(p->p_acct[LMS_USER]));
404 	ag->acctbuf.ac_stime = acct_compress(
405 	    NSEC_TO_TICK(p->p_acct[LMS_SYSTEM] + p->p_acct[LMS_TRAP]));
406 	ag->acctbuf.ac_etime = acct_compress(lbolt - u.u_ticks);
407 	ag->acctbuf.ac_mem = acct_compress((ulong_t)u.u_mem);
408 	ag->acctbuf.ac_io = acct_compress((ulong_t)p->p_ru.ioch);
409 	ag->acctbuf.ac_rw = acct_compress((ulong_t)(p->p_ru.inblock +
410 	    p->p_ru.oublock));
411 	cr = CRED();
412 	ag->acctbuf.ac_uid = crgetruid(cr);
413 	ag->acctbuf.ac_gid = crgetrgid(cr);
414 	(void) cmpldev(&ag->acctbuf.ac_tty, cttydev(p));
415 	ag->acctbuf.ac_stat = st;
416 	ag->acctbuf.ac_flag = (u.u_acflag | AEXPND);
417 
418 	/*
419 	 * Save the size. If the write fails, reset the size to avoid
420 	 * corrupted acct files.
421 	 *
422 	 * Large Files: We deliberately prevent accounting files from
423 	 * exceeding the 2GB limit as none of the accounting commands are
424 	 * currently large file aware.
425 	 */
426 	va.va_mask = AT_SIZE;
427 	if (VOP_GETATTR(vp, &va, 0, kcred) == 0) {
428 		error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&ag->acctbuf,
429 		    sizeof (ag->acctbuf), 0LL, UIO_SYSSPACE, FAPPEND,
430 		    (rlim64_t)MAXOFF32_T, kcred, &resid);
431 		if (error || resid)
432 			(void) VOP_SETATTR(vp, &va, 0, kcred, NULL);
433 	}
434 	mutex_exit(&ag->aclock);
435 }
436