Home | History | Annotate | Line # | Download | only in kern
kern_acct.c revision 1.66.4.1
      1 /*	$NetBSD: kern_acct.c,v 1.66.4.1 2006/09/11 18:07:25 ad 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 1.66.4.1 2006/09/11 18:07:25 ad 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/malloc.h>
     86 #include <sys/namei.h>
     87 #include <sys/errno.h>
     88 #include <sys/acct.h>
     89 #include <sys/resourcevar.h>
     90 #include <sys/ioctl.h>
     91 #include <sys/tty.h>
     92 #include <sys/kauth.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  * Mutex to serialize system calls and kernel threads.
    110  */
    111 kmutex_t acct_mutex;
    112 
    113 /*
    114  * The global accounting state and related data.  Gain the mutex before
    115  * accessing these variables.
    116  */
    117 static enum {
    118 	ACCT_STOP,
    119 	ACCT_ACTIVE,
    120 	ACCT_SUSPENDED
    121 } acct_state;				/* The current accounting state. */
    122 static struct vnode *acct_vp;		/* Accounting vnode pointer. */
    123 static kauth_cred_t acct_cred;		/* Credential of accounting file
    124 					   owner (i.e root).  Used when
    125  					   accounting file i/o.  */
    126 static struct proc *acct_dkwatcher;	/* Free disk space checker. */
    127 
    128 /*
    129  * Values associated with enabling and disabling accounting
    130  */
    131 int	acctsuspend = 2;	/* stop accounting when < 2% free space left */
    132 int	acctresume = 4;		/* resume when free space risen to > 4% */
    133 int	acctchkfreq = 15;	/* frequency (in seconds) to check space */
    134 
    135 /*
    136  * Encode_comp_t converts from ticks in seconds and microseconds
    137  * to ticks in 1/AHZ seconds.  The encoding is described in
    138  * Leffler, et al., on page 63.
    139  */
    140 
    141 #define	MANTSIZE	13			/* 13 bit mantissa. */
    142 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
    143 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
    144 
    145 static comp_t
    146 encode_comp_t(u_long s, u_long us)
    147 {
    148 	int exp, rnd;
    149 
    150 	exp = 0;
    151 	rnd = 0;
    152 	s *= AHZ;
    153 	s += us / (1000000 / AHZ);	/* Maximize precision. */
    154 
    155 	while (s > MAXFRACT) {
    156 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
    157 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
    158 		exp++;
    159 	}
    160 
    161 	/* If we need to round up, do it (and handle overflow correctly). */
    162 	if (rnd && (++s > MAXFRACT)) {
    163 		s >>= EXPSIZE;
    164 		exp++;
    165 	}
    166 
    167 	/* Clean it up and polish it off. */
    168 	exp <<= MANTSIZE;		/* Shift the exponent into place */
    169 	exp += s;			/* and add on the mantissa. */
    170 	return (exp);
    171 }
    172 
    173 static int
    174 acct_chkfree(void)
    175 {
    176 	int error;
    177 	struct statvfs *sb;
    178 	int64_t bavail;
    179 
    180 	sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK);
    181 	error = VFS_STATVFS(acct_vp->v_mount, sb, NULL);
    182 	if (error != 0)
    183 		return (error);
    184 
    185 	bavail = sb->f_bfree - sb->f_bresvd;
    186 
    187 	switch (acct_state) {
    188 	case ACCT_SUSPENDED:
    189 		if (bavail > acctresume * sb->f_blocks / 100) {
    190 			acct_state = ACCT_ACTIVE;
    191 			log(LOG_NOTICE, "Accounting resumed\n");
    192 		}
    193 		break;
    194 	case ACCT_ACTIVE:
    195 		if (bavail <= acctsuspend * sb->f_blocks / 100) {
    196 			acct_state = ACCT_SUSPENDED;
    197 			log(LOG_NOTICE, "Accounting suspended\n");
    198 		}
    199 		break;
    200 	case ACCT_STOP:
    201 		break;
    202 	}
    203 	free(sb, M_TEMP);
    204 	return (0);
    205 }
    206 
    207 static void
    208 acct_stop(void)
    209 {
    210 	int error;
    211 
    212 	if (acct_vp != NULLVP && acct_vp->v_type != VBAD) {
    213 		error = vn_close(acct_vp, FWRITE, acct_cred, NULL);
    214 #ifdef DIAGNOSTIC
    215 		if (error != 0)
    216 			printf("acct_stop: failed to close, errno = %d\n",
    217 			    error);
    218 #endif
    219 		acct_vp = NULLVP;
    220 	}
    221 	if (acct_cred != NULL) {
    222 		kauth_cred_free(acct_cred);
    223 		acct_cred = NULL;
    224 	}
    225 	acct_state = ACCT_STOP;
    226 }
    227 
    228 /*
    229  * Periodically check the file system to see if accounting
    230  * should be turned on or off.  Beware the case where the vnode
    231  * has been vgone()'d out from underneath us, e.g. when the file
    232  * system containing the accounting file has been forcibly unmounted.
    233  */
    234 static void
    235 acctwatch(void *arg)
    236 {
    237 	int error;
    238 
    239 	log(LOG_NOTICE, "Accounting started\n");
    240 	mutex_enter(&acct_mutex);
    241 	while (acct_state != ACCT_STOP) {
    242 		if (acct_vp->v_type == VBAD) {
    243 			log(LOG_NOTICE, "Accounting terminated\n");
    244 			acct_stop();
    245 			continue;
    246 		}
    247 
    248 		error = acct_chkfree();
    249 #ifdef DIAGNOSTIC
    250 		if (error != 0)
    251 			printf("acctwatch: failed to statvfs, error = %d\n",
    252 			    error);
    253 #endif
    254 		error = mtsleep(acctwatch, PSWP, "actwat", acctchkfreq * hz,
    255 		    &acct_mutex);
    256 #ifdef DIAGNOSTIC
    257 		if (error != 0 && error != EWOULDBLOCK)
    258 			printf("acctwatch: sleep error %d\n", error);
    259 #endif
    260 	}
    261 	acct_dkwatcher = NULL;
    262 	mutex_exit(&acct_mutex);
    263 
    264 	kthread_exit(0);
    265 }
    266 
    267 void
    268 acct_init(void)
    269 {
    270 
    271 	acct_state = ACCT_STOP;
    272 	acct_vp = NULLVP;
    273 	acct_cred = NULL;
    274 	mutex_init(&acct_mutex, MUTEX_DEFAULT, IPL_NONE);
    275 }
    276 
    277 /*
    278  * Accounting system call.  Written based on the specification and
    279  * previous implementation done by Mark Tinguely.
    280  */
    281 int
    282 sys_acct(struct lwp *l, void *v, register_t *retval)
    283 {
    284 	struct sys_acct_args /* {
    285 		syscallarg(const char *) path;
    286 	} */ *uap = v;
    287 	struct nameidata nd;
    288 	int error;
    289 
    290 	/* Make sure that the caller is root. */
    291 	if ((error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER,
    292 	    &l->l_acflag)) != 0)
    293 		return (error);
    294 
    295 	/*
    296 	 * If accounting is to be started to a file, open that file for
    297 	 * writing and make sure it's a 'normal'.
    298 	 */
    299 	if (SCARG(uap, path) != NULL) {
    300 		struct vattr va;
    301 		size_t pad;
    302 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path),
    303 		    l);
    304 		if ((error = vn_open(&nd, FWRITE|O_APPEND, 0)) != 0)
    305 			return (error);
    306 		if (nd.ni_vp->v_type != VREG) {
    307 			VOP_UNLOCK(nd.ni_vp, 0);
    308 			error = EACCES;
    309 			goto bad;
    310 		}
    311 		if ((error = VOP_GETATTR(nd.ni_vp, &va, l->l_cred, l)) != 0) {
    312 			VOP_UNLOCK(nd.ni_vp, 0);
    313 			goto bad;
    314 		}
    315 
    316 		if ((pad = (va.va_size % sizeof(struct acct))) != 0) {
    317 			u_quad_t size = va.va_size - pad;
    318 #ifdef DIAGNOSTIC
    319 			printf("Size of accounting file not a multiple of "
    320 			    "%lu - incomplete record truncated\n",
    321 			    (unsigned long)sizeof(struct acct));
    322 #endif
    323 			VATTR_NULL(&va);
    324 			va.va_size = size;
    325 			error = VOP_SETATTR(nd.ni_vp, &va, l->l_cred, l);
    326 			if (error != 0) {
    327 				VOP_UNLOCK(nd.ni_vp, 0);
    328 				goto bad;
    329 			}
    330 		}
    331 		VOP_UNLOCK(nd.ni_vp, 0);
    332 	}
    333 
    334 	mutex_enter(&acct_mutex);
    335 
    336 	/*
    337 	 * If accounting was previously enabled, kill the old space-watcher,
    338 	 * free credential for accounting file i/o,
    339 	 * ... (and, if no new file was specified, leave).
    340 	 */
    341 	acct_stop();
    342 	if (SCARG(uap, path) == NULL)
    343 		goto out;
    344 
    345 	/*
    346 	 * Save the new accounting file vnode and credential,
    347 	 * and schedule the new free space watcher.
    348 	 */
    349 	acct_state = ACCT_ACTIVE;
    350 	acct_vp = nd.ni_vp;
    351 	acct_cred = l->l_cred;
    352 	kauth_cred_hold(acct_cred);
    353 
    354 	error = acct_chkfree();		/* Initial guess. */
    355 	if (error != 0) {
    356 		acct_stop();
    357 		goto out;
    358 	}
    359 
    360 	if (acct_dkwatcher == NULL) {
    361 		error = kthread_create1(acctwatch, NULL, &acct_dkwatcher,
    362 		    "acctwatch");
    363 		if (error != 0)
    364 			acct_stop();
    365 	}
    366 
    367  out:
    368 	mutex_exit(&acct_mutex);
    369 	return (error);
    370  bad:
    371 	vn_close(nd.ni_vp, FWRITE, l->l_cred, l);
    372 	return error;
    373 }
    374 
    375 /*
    376  * Write out process accounting information, on process exit.
    377  * Data to be written out is specified in Leffler, et al.
    378  * and are enumerated below.  (They're also noted in the system
    379  * "acct.h" header file.)
    380  */
    381 int
    382 acct_process(struct lwp *l)
    383 {
    384 	struct acct acct;
    385 	struct timeval ut, st, tmp;
    386 	struct rusage *r;
    387 	int t, error = 0;
    388 	struct plimit *oplim = NULL;
    389 	struct proc *p = l->l_proc;
    390 
    391 	mutex_enter(&acct_mutex);
    392 
    393 	/* If accounting isn't enabled, don't bother */
    394 	if (acct_state != ACCT_ACTIVE)
    395 		goto out;
    396 
    397 	/*
    398 	 * Raise the file limit so that accounting can't be stopped by
    399 	 * the user.
    400 	 *
    401 	 * XXX We should think about the CPU limit, too.
    402 	 */
    403 	if (p->p_limit->p_refcnt > 1) {
    404 		oplim = p->p_limit;
    405 		p->p_limit = limcopy(p->p_limit);
    406 	}
    407 	p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
    408 
    409 	/*
    410 	 * Get process accounting information.
    411 	 */
    412 
    413 	/* (1) The name of the command that ran */
    414 	memcpy(acct.ac_comm, p->p_comm, sizeof(acct.ac_comm));
    415 
    416 	/* (2) The amount of user and system time that was used */
    417 	calcru(p, &ut, &st, NULL);
    418 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
    419 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
    420 
    421 	/* (3) The elapsed time the commmand ran (and its starting time) */
    422 	acct.ac_btime = p->p_stats->p_start.tv_sec;
    423 	getmicrotime(&tmp);
    424 	timersub(&tmp, &p->p_stats->p_start, &tmp);
    425 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
    426 
    427 	/* (4) The average amount of memory used */
    428 	r = &p->p_stats->p_ru;
    429 	timeradd(&ut, &st, &tmp);
    430 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
    431 	if (t)
    432 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
    433 	else
    434 		acct.ac_mem = 0;
    435 
    436 	/* (5) The number of disk I/O operations done */
    437 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
    438 
    439 	/* (6) The UID and GID of the process */
    440 	acct.ac_uid = kauth_cred_getuid(l->l_cred);
    441 	acct.ac_gid = kauth_cred_getgid(l->l_cred);
    442 
    443 	/* (7) The terminal from which the process was started */
    444 	rw_enter(&proclist_lock, RW_READER);
    445 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
    446 		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
    447 	else
    448 		acct.ac_tty = NODEV;
    449 	rw_exit(&proclist_lock);
    450 
    451 	/* (8) The boolean flags that tell how the process terminated, etc. */
    452 	acct.ac_flag = p->p_acflag;
    453 
    454 	/*
    455 	 * Now, just write the accounting information to the file.
    456 	 */
    457 	VOP_LEASE(acct_vp, l, l->l_cred, LEASE_WRITE);
    458 	error = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct,
    459 	    sizeof(acct), (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT,
    460 	    acct_cred, NULL, NULL);
    461 	if (error != 0)
    462 		log(LOG_ERR, "Accounting: write failed %d\n", error);
    463 
    464 	if (oplim) {
    465 		limfree(p->p_limit);
    466 		p->p_limit = oplim;
    467 	}
    468 
    469  out:
    470 	mutex_exit(&acct_mutex);
    471 	return (error);
    472 }
    473