Home | History | Annotate | Line # | Download | only in kern
kern_acct.c revision 1.43
      1 /*	$NetBSD: kern_acct.c,v 1.43 1997/10/19 02:00:25 mycroft 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/systm.h>
     46 #include <sys/proc.h>
     47 #include <sys/mount.h>
     48 #include <sys/vnode.h>
     49 #include <sys/file.h>
     50 #include <sys/syslog.h>
     51 #include <sys/kernel.h>
     52 #include <sys/namei.h>
     53 #include <sys/errno.h>
     54 #include <sys/acct.h>
     55 #include <sys/resourcevar.h>
     56 #include <sys/ioctl.h>
     57 #include <sys/tty.h>
     58 
     59 #include <sys/syscallargs.h>
     60 
     61 /*
     62  * The routines implemented in this file are described in:
     63  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
     64  *	    UNIX Operating System (Addison Welley, 1989)
     65  * on pages 62-63.
     66  *
     67  * Arguably, to simplify accounting operations, this mechanism should
     68  * be replaced by one in which an accounting log file (similar to /dev/klog)
     69  * is read by a user process, etc.  However, that has its own problems.
     70  */
     71 
     72 /*
     73  * Internal accounting functions.
     74  * The former's operation is described in Leffler, et al., and the latter
     75  * was provided by UCB with the 4.4BSD-Lite release
     76  */
     77 comp_t	encode_comp_t __P((u_long, u_long));
     78 void	acctwatch __P((void *));
     79 
     80 /*
     81  * Accounting vnode pointer, and saved vnode pointer.
     82  */
     83 struct	vnode *acctp;
     84 struct	vnode *savacctp;
     85 
     86 /*
     87  * Values associated with enabling and disabling accounting
     88  */
     89 int	acctsuspend = 2;	/* stop accounting when < 2% free space left */
     90 int	acctresume = 4;		/* resume when free space risen to > 4% */
     91 int	acctchkfreq = 15;	/* frequency (in seconds) to check space */
     92 
     93 /*
     94  * Accounting system call.  Written based on the specification and
     95  * previous implementation done by Mark Tinguely.
     96  */
     97 int
     98 sys_acct(p, v, retval)
     99 	struct proc *p;
    100 	void *v;
    101 	register_t *retval;
    102 {
    103 	struct sys_acct_args /* {
    104 		syscallarg(const char *) path;
    105 	} */ *uap = v;
    106 	struct nameidata nd;
    107 	int error;
    108 
    109 	/* Make sure that the caller is root. */
    110 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
    111 		return (error);
    112 
    113 	/*
    114 	 * If accounting is to be started to a file, open that file for
    115 	 * writing and make sure it's a 'normal'.
    116 	 */
    117 	if (SCARG(uap, path) != NULL) {
    118 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path),
    119 		    p);
    120 		if ((error = vn_open(&nd, FWRITE, 0)) != 0)
    121 			return (error);
    122 		VOP_UNLOCK(nd.ni_vp);
    123 		if (nd.ni_vp->v_type != VREG) {
    124 			vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
    125 			return (EACCES);
    126 		}
    127 	}
    128 
    129 	/*
    130 	 * If accounting was previously enabled, kill the old space-watcher,
    131 	 * close the file, and (if no new file was specified, leave).
    132 	 */
    133 	if (acctp != NULLVP || savacctp != NULLVP) {
    134 		untimeout(acctwatch, NULL);
    135 		error = vn_close((acctp != NULLVP ? acctp : savacctp), FWRITE,
    136 		    p->p_ucred, p);
    137 		acctp = savacctp = NULLVP;
    138 	}
    139 	if (SCARG(uap, path) == NULL)
    140 		return (error);
    141 
    142 	/*
    143 	 * Save the new accounting file vnode, and schedule the new
    144 	 * free space watcher.
    145 	 */
    146 	acctp = nd.ni_vp;
    147 	acctwatch(NULL);
    148 	return (error);
    149 }
    150 
    151 /*
    152  * Write out process accounting information, on process exit.
    153  * Data to be written out is specified in Leffler, et al.
    154  * and are enumerated below.  (They're also noted in the system
    155  * "acct.h" header file.)
    156  */
    157 int
    158 acct_process(p)
    159 	struct proc *p;
    160 {
    161 	struct acct acct;
    162 	struct rusage *r;
    163 	struct timeval ut, st, tmp;
    164 	int s, t;
    165 	struct vnode *vp;
    166 
    167 	/* If accounting isn't enabled, don't bother */
    168 	vp = acctp;
    169 	if (vp == NULLVP)
    170 		return (0);
    171 
    172 	/*
    173 	 * Get process accounting information.
    174 	 */
    175 
    176 	/* (1) The name of the command that ran */
    177 	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
    178 
    179 	/* (2) The amount of user and system time that was used */
    180 	calcru(p, &ut, &st, NULL);
    181 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
    182 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
    183 
    184 	/* (3) The elapsed time the commmand ran (and its starting time) */
    185 	acct.ac_btime = p->p_stats->p_start.tv_sec;
    186 	s = splclock();
    187 	timersub(&time, &p->p_stats->p_start, &tmp);
    188 	splx(s);
    189 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
    190 
    191 	/* (4) The average amount of memory used */
    192 	r = &p->p_stats->p_ru;
    193 	timeradd(&ut, &st, &tmp);
    194 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
    195 	if (t)
    196 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
    197 	else
    198 		acct.ac_mem = 0;
    199 
    200 	/* (5) The number of disk I/O operations done */
    201 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
    202 
    203 	/* (6) The UID and GID of the process */
    204 	acct.ac_uid = p->p_cred->p_ruid;
    205 	acct.ac_gid = p->p_cred->p_rgid;
    206 
    207 	/* (7) The terminal from which the process was started */
    208 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
    209 		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
    210 	else
    211 		acct.ac_tty = NODEV;
    212 
    213 	/* (8) The boolean flags that tell how the process terminated, etc. */
    214 	acct.ac_flag = p->p_acflag;
    215 
    216 	/*
    217 	 * Now, just write the accounting information to the file.
    218 	 */
    219 	VOP_LEASE(vp, p, p->p_ucred, LEASE_WRITE);
    220 	return (vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
    221 	    (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, p->p_ucred,
    222 	    (int *)0, p));
    223 }
    224 
    225 /*
    226  * Encode_comp_t converts from ticks in seconds and microseconds
    227  * to ticks in 1/AHZ seconds.  The encoding is described in
    228  * Leffler, et al., on page 63.
    229  */
    230 
    231 #define	MANTSIZE	13			/* 13 bit mantissa. */
    232 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
    233 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
    234 
    235 comp_t
    236 encode_comp_t(s, us)
    237 	u_long s, us;
    238 {
    239 	int exp, rnd;
    240 
    241 	exp = 0;
    242 	rnd = 0;
    243 	s *= AHZ;
    244 	s += us / (1000000 / AHZ);	/* Maximize precision. */
    245 
    246 	while (s > MAXFRACT) {
    247 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
    248 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
    249 		exp++;
    250 	}
    251 
    252 	/* If we need to round up, do it (and handle overflow correctly). */
    253 	if (rnd && (++s > MAXFRACT)) {
    254 		s >>= EXPSIZE;
    255 		exp++;
    256 	}
    257 
    258 	/* Clean it up and polish it off. */
    259 	exp <<= MANTSIZE;		/* Shift the exponent into place */
    260 	exp += s;			/* and add on the mantissa. */
    261 	return (exp);
    262 }
    263 
    264 /*
    265  * Periodically check the file system to see if accounting
    266  * should be turned on or off.  Beware the case where the vnode
    267  * has been vgone()'d out from underneath us, e.g. when the file
    268  * system containing the accounting file has been forcibly unmounted.
    269  */
    270 /* ARGSUSED */
    271 void
    272 acctwatch(a)
    273 	void *a;
    274 {
    275 	struct statfs sb;
    276 
    277 	if (savacctp != NULLVP) {
    278 		if (savacctp->v_type == VBAD) {
    279 			(void) vn_close(savacctp, FWRITE, NOCRED, NULL);
    280 			savacctp = NULLVP;
    281 			return;
    282 		}
    283 		(void)VFS_STATFS(savacctp->v_mount, &sb, (struct proc *)0);
    284 		if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
    285 			acctp = savacctp;
    286 			savacctp = NULLVP;
    287 			log(LOG_NOTICE, "Accounting resumed\n");
    288 		}
    289 	} else if (acctp != NULLVP) {
    290 		if (acctp->v_type == VBAD) {
    291 			(void) vn_close(acctp, FWRITE, NOCRED, NULL);
    292 			acctp = NULLVP;
    293 			return;
    294 		}
    295 		(void)VFS_STATFS(acctp->v_mount, &sb, (struct proc *)0);
    296 		if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
    297 			savacctp = acctp;
    298 			acctp = NULLVP;
    299 			log(LOG_NOTICE, "Accounting suspended\n");
    300 		}
    301 	} else
    302 		return;
    303 	timeout(acctwatch, NULL, acctchkfreq * hz);
    304 }
    305