Home | History | Annotate | Line # | Download | only in kern
kern_acct.c revision 1.31
      1 /*	$NetBSD: kern_acct.c,v 1.31 1994/07/04 20:27:06 cgd Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1994 Christopher G. Demetriou
      5  * Copyright (c) 1982, 1986, 1989, 1993
      6  *	The Regents of the University of California.  All rights reserved.
      7  * (c) UNIX System Laboratories, Inc.
      8  * All or some portions of this file are derived from material licensed
      9  * to the University of California by American Telephone and Telegraph
     10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     11  * the permission of UNIX System Laboratories, Inc.
     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/param.h>
     45 #include <sys/proc.h>
     46 #include <sys/mount.h>
     47 #include <sys/vnode.h>
     48 #include <sys/file.h>
     49 #include <sys/syslog.h>
     50 #include <sys/kernel.h>
     51 #include <sys/namei.h>
     52 #include <sys/errno.h>
     53 #include <sys/acct.h>
     54 #include <sys/resourcevar.h>
     55 #include <sys/ioctl.h>
     56 #include <sys/tty.h>
     57 
     58 /*
     59  * The routines implemented in this file are described in:
     60  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
     61  *	    UNIX Operating System (Addison Welley, 1989)
     62  * on pages 62-63.
     63  *
     64  * Arguably, to simplify accounting operations, this mechanism should
     65  * be replaced by one in which an accounting log file (similar to /dev/klog)
     66  * is read by a user process, etc.  However, that has its own problems.
     67  */
     68 
     69 /*
     70  * Internal accounting functions.
     71  * The former's operation is described in Leffler, et al., and the latter
     72  * was provided by UCB with the 4.4BSD-Lite release
     73  */
     74 comp_t	encode_comp_t __P((u_long, u_long));
     75 void	acctwatch __P((void *));
     76 
     77 /*
     78  * Accounting vnode pointer, and saved vnode pointer.
     79  */
     80 struct	vnode *acctp;
     81 struct	vnode *savacctp;
     82 
     83 /*
     84  * Values associated with enabling and disabling accounting
     85  */
     86 int	acctsuspend = 2;	/* stop accounting when < 2% free space left */
     87 int	acctresume = 4;		/* resume when free space risen to > 4% */
     88 int	acctchkfreq = 15;	/* frequency (in seconds) to check space */
     89 
     90 /*
     91  * Accounting system call.  Written based on the specification and
     92  * previous implementation done by Mark Tinguely.
     93  */
     94 struct acct_args {
     95 	char	*path;
     96 };
     97 acct(p, uap, retval)
     98 	struct proc *p;
     99 	struct acct_args *uap;
    100 	int *retval;
    101 {
    102 	struct nameidata nd;
    103 	int error;
    104 
    105 	/* Make sure that the caller is root. */
    106 	if (error = suser(p->p_ucred, &p->p_acflag))
    107 		return (error);
    108 
    109 	/*
    110 	 * If accounting is to be started to a file, open that file for
    111 	 * writing and make sure it's a 'normal'.
    112 	 */
    113 	if (uap->path != NULL) {
    114 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->path, p);
    115 		if (error = vn_open(&nd, FWRITE, 0))
    116 			return (error);
    117 		VOP_UNLOCK(nd.ni_vp);
    118 		if (nd.ni_vp->v_type != VREG) {
    119 			vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
    120 			return (EACCES);
    121 		}
    122 	}
    123 
    124 	/*
    125 	 * If accounting was previously enabled, kill the old space-watcher,
    126 	 * close the file, and (if no new file was specified, leave).
    127 	 */
    128 	if (acctp != NULL || savacctp != NULL) {
    129 		untimeout(acctwatch, NULL);
    130 		error = vn_close((acctp != NULL ? acctp : savacctp), FWRITE,
    131 		    p->p_ucred, p);
    132 		acctp = savacctp = NULL;
    133 	}
    134 	if (uap->path == NULL)
    135 		return (error);
    136 
    137 	/*
    138 	 * Save the new accounting file vnode, and schedule the new
    139 	 * free space watcher.
    140 	 */
    141 	acctp = nd.ni_vp;
    142 	acctwatch(NULL);
    143 	return (error);
    144 }
    145 
    146 /*
    147  * Write out process accounting information, on process exit.
    148  * Data to be written out is specified in Leffler, et al.
    149  * and are enumerated below.  (They're also noted in the system
    150  * "acct.h" header file.)
    151  */
    152 acct_process(p)
    153 	struct proc *p;
    154 {
    155 	struct acct acct;
    156 	struct rusage *r;
    157 	struct timeval ut, st, tmp;
    158 	int s, t;
    159 	struct vnode *vp;
    160 
    161 	/* If accounting isn't enabled, don't bother */
    162 	vp = acctp;
    163 	if (vp == NULL)
    164 		return (0);
    165 
    166 	/*
    167 	 * Get process accounting information.
    168 	 */
    169 
    170 	/* (1) The name of the command that ran */
    171 	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
    172 
    173 	/* (2) The amount of user and system time that was used */
    174 	calcru(p, &ut, &st, NULL);
    175 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
    176 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
    177 
    178 	/* (3) The elapsed time the commmand ran (and its starting time) */
    179 	acct.ac_btime = p->p_stats->p_start.tv_sec;
    180 	s = splclock();
    181 	tmp = time;
    182 	splx(s);
    183 	timevalsub(&tmp, &p->p_stats->p_start);
    184 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
    185 
    186 	/* (4) The average amount of memory used */
    187 	r = &p->p_stats->p_ru;
    188 	tmp = ut;
    189 	timevaladd(&tmp, &st);
    190 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
    191 	if (t)
    192 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
    193 	else
    194 		acct.ac_mem = 0;
    195 
    196 	/* (5) The number of disk I/O operations done */
    197 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
    198 
    199 	/* (6) The UID and GID of the process */
    200 	acct.ac_uid = p->p_cred->p_ruid;
    201 	acct.ac_gid = p->p_cred->p_rgid;
    202 
    203 	/* (7) The terminal from which the process was started */
    204 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
    205 		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
    206 	else
    207 		acct.ac_tty = NODEV;
    208 
    209 	/* (8) The boolean flags that tell how the process terminated, etc. */
    210 	acct.ac_flag = p->p_acflag;
    211 
    212 	/*
    213 	 * Now, just write the accounting information to the file.
    214 	 */
    215 	LEASE_CHECK(vp, p, p->p_ucred, LEASE_WRITE);
    216 	return (vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
    217 	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, p->p_ucred,
    218 	    (int *)0, p));
    219 }
    220 
    221 /*
    222  * Encode_comp_t converts from ticks in seconds and microseconds
    223  * to ticks in 1/AHZ seconds.  The encoding is described in
    224  * Leffler, et al., on page 63.
    225  */
    226 
    227 #define	MANTSIZE	13			/* 13 bit mantissa. */
    228 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
    229 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
    230 
    231 comp_t
    232 encode_comp_t(s, us)
    233 	u_long s, us;
    234 {
    235 	int exp, rnd;
    236 
    237 	exp = 0;
    238 	rnd = 0;
    239 	s *= AHZ;
    240 	s += us / (1000000 / AHZ);	/* Maximize precision. */
    241 
    242 	while (s > MAXFRACT) {
    243 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
    244 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
    245 		exp++;
    246 	}
    247 
    248 	/* If we need to round up, do it (and handle overflow correctly). */
    249 	if (rnd && (++s > MAXFRACT)) {
    250 		s >>= EXPSIZE;
    251 		exp++;
    252 	}
    253 
    254 	/* Clean it up and polish it off. */
    255 	exp <<= MANTSIZE;		/* Shift the exponent into place */
    256 	exp += s;			/* and add on the mantissa. */
    257 	return (exp);
    258 }
    259 
    260 /*
    261  * Periodically check the file system to see if accounting
    262  * should be turned on or off.  Beware the case where the vnode
    263  * has been vgone()'d out from underneath us, e.g. when the file
    264  * system containing the accounting file has been forcibly unmounted.
    265  */
    266 /* ARGSUSED */
    267 void
    268 acctwatch(a)
    269 	void *a;
    270 {
    271 	struct statfs sb;
    272 
    273 	if (savacctp) {
    274 		if (savacctp->v_type == VBAD) {
    275 			(void) vn_close(savacctp, FWRITE, NOCRED, NULL);
    276 			savacctp = NULL;
    277 			return;
    278 		}
    279 		(void)VFS_STATFS(savacctp->v_mount, &sb, (struct proc *)0);
    280 		if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
    281 			acctp = savacctp;
    282 			savacctp = NULL;
    283 			log(LOG_NOTICE, "Accounting resumed\n");
    284 		}
    285 	} else {
    286 		if (acctp == NULL)
    287 			return;
    288 		if (acctp->v_type == VBAD) {
    289 			(void) vn_close(acctp, FWRITE, NOCRED, NULL);
    290 			acctp = NULL;
    291 			return;
    292 		}
    293 		(void)VFS_STATFS(acctp->v_mount, &sb, (struct proc *)0);
    294 		if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
    295 			savacctp = acctp;
    296 			acctp = NULL;
    297 			log(LOG_NOTICE, "Accounting suspended\n");
    298 		}
    299 	}
    300 	timeout(acctwatch, NULL, acctchkfreq * hz);
    301 }
    302