Home | History | Annotate | Line # | Download | only in kern
kern_acct.c revision 1.29
      1 /*-
      2  * Copyright (c) 1994 Christopher G. Demetriou
      3  * Copyright (c) 1982, 1986, 1989, 1993
      4  *	The Regents of the University of California.  All rights reserved.
      5  * (c) UNIX System Laboratories, Inc.
      6  * All or some portions of this file are derived from material licensed
      7  * to the University of California by American Telephone and Telegraph
      8  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
      9  * the permission of UNIX System Laboratories, Inc.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *	This product includes software developed by the University of
     22  *	California, Berkeley and its contributors.
     23  * 4. Neither the name of the University nor the names of its contributors
     24  *    may be used to endorse or promote products derived from this software
     25  *    without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     37  * SUCH DAMAGE.
     38  *
     39  *	from: @(#)kern_acct.c	8.1 (Berkeley) 6/14/93
     40  *	$Id: kern_acct.c,v 1.29 1994/06/19 22:58:29 cgd Exp $
     41  */
     42 
     43 #include <sys/param.h>
     44 #include <sys/proc.h>
     45 #include <sys/mount.h>
     46 #include <sys/vnode.h>
     47 #include <sys/file.h>
     48 #include <sys/syslog.h>
     49 #include <sys/kernel.h>
     50 #include <sys/namei.h>
     51 #include <sys/errno.h>
     52 #include <sys/acct.h>
     53 #include <sys/resourcevar.h>
     54 #include <sys/ioctl.h>
     55 #include <sys/tty.h>
     56 
     57 /*
     58  * The routines implemented in this file are described in:
     59  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
     60  *	    UNIX Operating System (Addison Welley, 1989)
     61  * on pages 62-63.
     62  *
     63  * Arguably, to simplify accounting operations, this mechanism should
     64  * be replaced by one in which an accounting log file (similar to /dev/klog)
     65  * is read by a user process, etc.  However, that has its own problems.
     66  */
     67 
     68 /*
     69  * Internal accounting functions.
     70  * The former's operation is described in Leffler, et al., and the latter
     71  * was provided by UCB with the 4.4BSD-Lite release
     72  */
     73 comp_t	encode_comp_t __P((u_long, u_long));
     74 void	acctwatch __P((void *));
     75 
     76 /*
     77  * Accounting vnode pointer, and saved vnode pointer.
     78  */
     79 struct	vnode *acctp;
     80 struct	vnode *savacctp;
     81 
     82 /*
     83  * Values associated with enabling and disabling accounting
     84  */
     85 int	acctsuspend = 2;	/* stop accounting when < 2% free space left */
     86 int	acctresume = 4;		/* resume when free space risen to > 4% */
     87 int	acctchkfreq = 15;	/* frequency (in seconds) to check space */
     88 
     89 /*
     90  * Accounting system call.  Written based on the specification and
     91  * previous implementation done by Mark Tinguely.
     92  */
     93 struct acct_args {
     94 	char	*path;
     95 };
     96 acct(p, uap, retval)
     97 	struct proc *p;
     98 	struct acct_args *uap;
     99 	int *retval;
    100 {
    101 	struct nameidata nd;
    102 	int error;
    103 
    104 	/* Make sure that the caller is root. */
    105 	if (error = suser(p->p_ucred, &p->p_acflag))
    106 		return (error);
    107 
    108 	/*
    109 	 * If accounting is to be started to a file, open that file for
    110 	 * writing and make sure it's a 'normal'.
    111 	 */
    112 	if (uap->path != NULL) {
    113 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->path, p);
    114 		if (error = vn_open(&nd, FWRITE, 0))
    115 			return (error);
    116 		VOP_UNLOCK(nd.ni_vp);
    117 		if (nd.ni_vp->v_type != VREG) {
    118 			vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
    119 			return (EACCES);
    120 		}
    121 	}
    122 
    123 	/*
    124 	 * If accounting was previously enabled, kill the old space-watcher,
    125 	 * close the file, and (if no new file was specified, leave).
    126 	 */
    127 	if (acctp != NULLVP || savacctp != NULLVP) {
    128 		untimeout(acctwatch, NULL);
    129 		error = vn_close((acctp != NULLVP ? acctp : savacctp), FWRITE,
    130 		    p->p_ucred, p);
    131 		acctp = savacctp = NULLVP;
    132 	}
    133 	if (uap->path == NULL)
    134 		return (error);
    135 
    136 	/*
    137 	 * Save the new accounting file vnode, and schedule the new
    138 	 * free space watcher.
    139 	 */
    140 	acctp = nd.ni_vp;
    141 	acctwatch(NULL);
    142 	return (error);
    143 }
    144 
    145 /*
    146  * Write out process accounting information, on process exit.
    147  * Data to be written out is specified in Leffler, et al.
    148  * and are enumerated below.  (They're also noted in the system
    149  * "acct.h" header file.)
    150  */
    151 acct_process(p)
    152 	struct proc *p;
    153 {
    154 	struct acct acct;
    155 	struct rusage *r;
    156 	struct timeval ut, st, tmp;
    157 	int s, t;
    158 	struct vnode *vp;
    159 
    160 	/* If accounting isn't enabled, don't bother */
    161 	vp = acctp;
    162 	if (vp == NULLVP)
    163 		return (0);
    164 
    165 	/*
    166 	 * Get process accounting information.
    167 	 */
    168 
    169 	/* (1) The name of the command that ran */
    170 	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
    171 
    172 	/* (2) The amount of user and system time that was used */
    173 	calcru(p, &ut, &st, NULL);
    174 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
    175 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
    176 
    177 	/* (3) The elapsed time the commmand ran (and its starting time) */
    178 	acct.ac_btime = p->p_stats->p_start.tv_sec;
    179 	s = splclock();
    180 	tmp = time;
    181 	splx(s);
    182 	timevalsub(&tmp, &p->p_stats->p_start);
    183 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
    184 
    185 	/* (4) The average amount of memory used */
    186 	r = &p->p_stats->p_ru;
    187 	tmp = ut;
    188 	timevaladd(&tmp, &st);
    189 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
    190 	if (t)
    191 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
    192 	else
    193 		acct.ac_mem = 0;
    194 
    195 	/* (5) The number of disk I/O operations done */
    196 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
    197 
    198 	/* (6) The UID and GID of the process */
    199 	acct.ac_uid = p->p_cred->p_ruid;
    200 	acct.ac_gid = p->p_cred->p_rgid;
    201 
    202 	/* (7) The terminal from which the process was started */
    203 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
    204 		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
    205 	else
    206 		acct.ac_tty = NODEV;
    207 
    208 	/* (8) The boolean flags that tell how the process terminated, etc. */
    209 	acct.ac_flag = p->p_acflag;
    210 
    211 	/*
    212 	 * Now, just write the accounting information to the file.
    213 	 */
    214 	LEASE_CHECK(vp, p, p->p_ucred, LEASE_WRITE);
    215 	return (vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
    216 	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, p->p_ucred,
    217 	    (int *)0, p));
    218 }
    219 
    220 /*
    221  * Encode_comp_t converts from ticks in seconds and microseconds
    222  * to ticks in 1/AHZ seconds.  The encoding is described in
    223  * Leffler, et al., on page 63.
    224  */
    225 
    226 #define	MANTSIZE	13			/* 13 bit mantissa. */
    227 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
    228 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
    229 
    230 comp_t
    231 encode_comp_t(s, us)
    232 	u_long s, us;
    233 {
    234 	int exp, rnd;
    235 
    236 	exp = 0;
    237 	rnd = 0;
    238 	s *= AHZ;
    239 	s += us / (1000000 / AHZ);	/* Maximize precision. */
    240 
    241 	while (s > MAXFRACT) {
    242 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
    243 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
    244 		exp++;
    245 	}
    246 
    247 	/* If we need to round up, do it (and handle overflow correctly). */
    248 	if (rnd && (++s > MAXFRACT)) {
    249 		s >>= EXPSIZE;
    250 		exp++;
    251 	}
    252 
    253 	/* Clean it up and polish it off. */
    254 	exp <<= MANTSIZE;		/* Shift the exponent into place */
    255 	exp += s;			/* and add on the mantissa. */
    256 	return (exp);
    257 }
    258 
    259 /*
    260  * Periodically check the file system to see if accounting
    261  * should be turned on or off.  Beware the case where the vnode
    262  * has been vgone()'d out from underneath us, e.g. when the file
    263  * system containing the accounting file has been forcibly unmounted.
    264  */
    265 /* ARGSUSED */
    266 void
    267 acctwatch(a)
    268 	void *a;
    269 {
    270 	struct statfs sb;
    271 
    272 	if (savacctp != NULLVP && savacctp->v_type != VBAD) {
    273 		(void)VFS_STATFS(savacctp->v_mount, &sb, (struct proc *)0);
    274 		if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
    275 			acctp = savacctp;
    276 			savacctp = NULLVP;
    277 			log(LOG_NOTICE, "Accounting resumed\n");
    278 		}
    279 	} else if (acctp != NULLVP && acctp->v_type != VBAD) {
    280 		(void)VFS_STATFS(acctp->v_mount, &sb, (struct proc *)0);
    281 		if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
    282 			savacctp = acctp;
    283 			acctp = NULLVP;
    284 			log(LOG_NOTICE, "Accounting suspended\n");
    285 		}
    286 	} else {
    287 		/* Accounting was turned off behind our back. */
    288 		return;
    289 	}
    290 	timeout(acctwatch, NULL, acctchkfreq * hz);
    291 }
    292