Home | History | Annotate | Line # | Download | only in kern
kern_acct.c revision 1.55
      1 /*	$NetBSD: kern_acct.c,v 1.55 2003/08/07 16:31:42 agc 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 1.55 2003/08/07 16:31:42 agc 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/lock.h>
     86 #include <sys/malloc.h>
     87 #include <sys/namei.h>
     88 #include <sys/errno.h>
     89 #include <sys/acct.h>
     90 #include <sys/resourcevar.h>
     91 #include <sys/ioctl.h>
     92 #include <sys/tty.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  * The global accounting state and related data.  Gain the lock before
    110  * accessing these variables.
    111  */
    112 enum {
    113 	ACCT_STOP,
    114 	ACCT_ACTIVE,
    115 	ACCT_SUSPENDED
    116 } acct_state;				/* The current accounting state. */
    117 struct vnode *acct_vp;			/* Accounting vnode pointer. */
    118 struct ucred *acct_ucred;		/* Credential of accounting file
    119 					   owner (i.e root).  Used when
    120  					   accounting file i/o.  */
    121 struct proc *acct_dkwatcher;		/* Free disk space checker. */
    122 
    123 /*
    124  * Lock to serialize system calls and kernel threads.
    125  */
    126 struct	lock acct_lock;
    127 #define	ACCT_LOCK()						\
    128 do {								\
    129 	(void) lockmgr(&acct_lock, LK_EXCLUSIVE, NULL);		\
    130 } while (/* CONSTCOND */0)
    131 #define	ACCT_UNLOCK()						\
    132 do {								\
    133 	(void) lockmgr(&acct_lock, LK_RELEASE, NULL);		\
    134 } while (/* CONSTCOND */0)
    135 
    136 /*
    137  * Internal accounting functions.
    138  * The former's operation is described in Leffler, et al., and the latter
    139  * was provided by UCB with the 4.4BSD-Lite release
    140  */
    141 comp_t	encode_comp_t __P((u_long, u_long));
    142 void	acctwatch __P((void *));
    143 void	acct_stop __P((void));
    144 int	acct_chkfree __P((void));
    145 
    146 /*
    147  * Values associated with enabling and disabling accounting
    148  */
    149 int	acctsuspend = 2;	/* stop accounting when < 2% free space left */
    150 int	acctresume = 4;		/* resume when free space risen to > 4% */
    151 int	acctchkfreq = 15;	/* frequency (in seconds) to check space */
    152 
    153 void
    154 acct_init()
    155 {
    156 
    157 	acct_state = ACCT_STOP;
    158 	acct_vp = NULLVP;
    159 	acct_ucred = NULL;
    160 	lockinit(&acct_lock, PWAIT, "acctlk", 0, 0);
    161 }
    162 
    163 void
    164 acct_stop()
    165 {
    166 	int error;
    167 
    168 	if (acct_vp != NULLVP && acct_vp->v_type != VBAD) {
    169 		error = vn_close(acct_vp, FWRITE, acct_ucred, NULL);
    170 #ifdef DIAGNOSTIC
    171 		if (error != 0)
    172 			printf("acct_stop: failed to close, errno = %d\n",
    173 			    error);
    174 #endif
    175 		acct_vp = NULLVP;
    176 	}
    177 	if (acct_ucred != NULL) {
    178 		crfree(acct_ucred);
    179 		acct_ucred = NULL;
    180 	}
    181 	acct_state = ACCT_STOP;
    182 }
    183 
    184 int
    185 acct_chkfree()
    186 {
    187 	int error;
    188 	struct statfs sb;
    189 
    190 	error = VFS_STATFS(acct_vp->v_mount, &sb, NULL);
    191 	if (error != 0)
    192 		return (error);
    193 
    194 	switch (acct_state) {
    195 	case ACCT_SUSPENDED:
    196 		if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
    197 			acct_state = ACCT_ACTIVE;
    198 			log(LOG_NOTICE, "Accounting resumed\n");
    199 		}
    200 		break;
    201 	case ACCT_ACTIVE:
    202 		if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
    203 			acct_state = ACCT_SUSPENDED;
    204 			log(LOG_NOTICE, "Accounting suspended\n");
    205 		}
    206 		break;
    207 	case ACCT_STOP:
    208 		break;
    209 	}
    210 	return (0);
    211 }
    212 
    213 /*
    214  * Accounting system call.  Written based on the specification and
    215  * previous implementation done by Mark Tinguely.
    216  */
    217 int
    218 sys_acct(l, v, retval)
    219 	struct lwp *l;
    220 	void *v;
    221 	register_t *retval;
    222 {
    223 	struct sys_acct_args /* {
    224 		syscallarg(const char *) path;
    225 	} */ *uap = v;
    226 	struct nameidata nd;
    227 	int error;
    228 	struct proc *p = l->l_proc;
    229 
    230 	/* Make sure that the caller is root. */
    231 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
    232 		return (error);
    233 
    234 	/*
    235 	 * If accounting is to be started to a file, open that file for
    236 	 * writing and make sure it's a 'normal'.
    237 	 */
    238 	if (SCARG(uap, path) != NULL) {
    239 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path),
    240 		    p);
    241 		if ((error = vn_open(&nd, FWRITE|O_APPEND, 0)) != 0)
    242 			return (error);
    243 		VOP_UNLOCK(nd.ni_vp, 0);
    244 		if (nd.ni_vp->v_type != VREG) {
    245 			vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
    246 			return (EACCES);
    247 		}
    248 	}
    249 
    250 	ACCT_LOCK();
    251 
    252 	/*
    253 	 * If accounting was previously enabled, kill the old space-watcher,
    254 	 * free credential for accounting file i/o,
    255 	 * ... (and, if no new file was specified, leave).
    256 	 */
    257 	acct_stop();
    258 	if (SCARG(uap, path) == NULL)
    259 		goto out;
    260 
    261 	/*
    262 	 * Save the new accounting file vnode and credential,
    263 	 * and schedule the new free space watcher.
    264 	 */
    265 	acct_state = ACCT_ACTIVE;
    266 	acct_vp = nd.ni_vp;
    267 	acct_ucred = p->p_ucred;
    268 	crhold(acct_ucred);
    269 
    270 	error = acct_chkfree();		/* Initial guess. */
    271 	if (error != 0) {
    272 		acct_stop();
    273 		goto out;
    274 	}
    275 
    276 	if (acct_dkwatcher == NULL) {
    277 		error = kthread_create1(acctwatch, NULL, &acct_dkwatcher,
    278 		    "acctwatch");
    279 		if (error != 0)
    280 			acct_stop();
    281 	}
    282 
    283  out:
    284 	ACCT_UNLOCK();
    285 	return (error);
    286 }
    287 
    288 /*
    289  * Write out process accounting information, on process exit.
    290  * Data to be written out is specified in Leffler, et al.
    291  * and are enumerated below.  (They're also noted in the system
    292  * "acct.h" header file.)
    293  */
    294 int
    295 acct_process(p)
    296 	struct proc *p;
    297 {
    298 	struct acct acct;
    299 	struct rusage *r;
    300 	struct timeval ut, st, tmp;
    301 	int s, t, error = 0;
    302 	struct plimit *oplim = NULL;
    303 
    304 	ACCT_LOCK();
    305 
    306 	/* If accounting isn't enabled, don't bother */
    307 	if (acct_state != ACCT_ACTIVE)
    308 		goto out;
    309 
    310 	/*
    311 	 * Raise the file limit so that accounting can't be stopped by
    312 	 * the user.
    313 	 *
    314 	 * XXX We should think about the CPU limit, too.
    315 	 */
    316 	if (p->p_limit->p_refcnt > 1) {
    317 		oplim = p->p_limit;
    318 		p->p_limit = limcopy(p->p_limit);
    319 	}
    320 	p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
    321 
    322 	/*
    323 	 * Get process accounting information.
    324 	 */
    325 
    326 	/* (1) The name of the command that ran */
    327 	memcpy(acct.ac_comm, p->p_comm, sizeof(acct.ac_comm));
    328 
    329 	/* (2) The amount of user and system time that was used */
    330 	calcru(p, &ut, &st, NULL);
    331 	acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
    332 	acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
    333 
    334 	/* (3) The elapsed time the commmand ran (and its starting time) */
    335 	acct.ac_btime = p->p_stats->p_start.tv_sec;
    336 	s = splclock();
    337 	timersub(&time, &p->p_stats->p_start, &tmp);
    338 	splx(s);
    339 	acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
    340 
    341 	/* (4) The average amount of memory used */
    342 	r = &p->p_stats->p_ru;
    343 	timeradd(&ut, &st, &tmp);
    344 	t = tmp.tv_sec * hz + tmp.tv_usec / tick;
    345 	if (t)
    346 		acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
    347 	else
    348 		acct.ac_mem = 0;
    349 
    350 	/* (5) The number of disk I/O operations done */
    351 	acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
    352 
    353 	/* (6) The UID and GID of the process */
    354 	acct.ac_uid = p->p_cred->p_ruid;
    355 	acct.ac_gid = p->p_cred->p_rgid;
    356 
    357 	/* (7) The terminal from which the process was started */
    358 	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
    359 		acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
    360 	else
    361 		acct.ac_tty = NODEV;
    362 
    363 	/* (8) The boolean flags that tell how the process terminated, etc. */
    364 	acct.ac_flag = p->p_acflag;
    365 
    366 	/*
    367 	 * Now, just write the accounting information to the file.
    368 	 */
    369 	VOP_LEASE(acct_vp, p, p->p_ucred, LEASE_WRITE);
    370 	error = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct,
    371 	    sizeof(acct), (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT,
    372 	    acct_ucred, NULL, p);
    373 	if (error != 0)
    374 		log(LOG_ERR, "Accounting: write failed %d\n", error);
    375 
    376 	if (oplim) {
    377 		limfree(p->p_limit);
    378 		p->p_limit = oplim;
    379 	}
    380 
    381  out:
    382 	ACCT_UNLOCK();
    383 	return (error);
    384 }
    385 
    386 /*
    387  * Encode_comp_t converts from ticks in seconds and microseconds
    388  * to ticks in 1/AHZ seconds.  The encoding is described in
    389  * Leffler, et al., on page 63.
    390  */
    391 
    392 #define	MANTSIZE	13			/* 13 bit mantissa. */
    393 #define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
    394 #define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
    395 
    396 comp_t
    397 encode_comp_t(s, us)
    398 	u_long s, us;
    399 {
    400 	int exp, rnd;
    401 
    402 	exp = 0;
    403 	rnd = 0;
    404 	s *= AHZ;
    405 	s += us / (1000000 / AHZ);	/* Maximize precision. */
    406 
    407 	while (s > MAXFRACT) {
    408 	rnd = s & (1 << (EXPSIZE - 1));	/* Round up? */
    409 		s >>= EXPSIZE;		/* Base 8 exponent == 3 bit shift. */
    410 		exp++;
    411 	}
    412 
    413 	/* If we need to round up, do it (and handle overflow correctly). */
    414 	if (rnd && (++s > MAXFRACT)) {
    415 		s >>= EXPSIZE;
    416 		exp++;
    417 	}
    418 
    419 	/* Clean it up and polish it off. */
    420 	exp <<= MANTSIZE;		/* Shift the exponent into place */
    421 	exp += s;			/* and add on the mantissa. */
    422 	return (exp);
    423 }
    424 
    425 /*
    426  * Periodically check the file system to see if accounting
    427  * should be turned on or off.  Beware the case where the vnode
    428  * has been vgone()'d out from underneath us, e.g. when the file
    429  * system containing the accounting file has been forcibly unmounted.
    430  */
    431 void
    432 acctwatch(arg)
    433 	void *arg;
    434 {
    435 	int error;
    436 
    437 	log(LOG_NOTICE, "Accounting started\n");
    438 	ACCT_LOCK();
    439 	while (acct_state != ACCT_STOP) {
    440 		if (acct_vp->v_type == VBAD) {
    441 			log(LOG_NOTICE, "Accounting terminated\n");
    442 			acct_stop();
    443 			continue;
    444 		}
    445 
    446 		error = acct_chkfree();
    447 #ifdef DIAGNOSTIC
    448 		if (error != 0)
    449 			printf("acctwatch: failed to statfs, error = %d\n",
    450 			    error);
    451 #endif
    452 
    453 		ACCT_UNLOCK();
    454 		error = tsleep(acctwatch, PSWP, "actwat", acctchkfreq * hz);
    455 		ACCT_LOCK();
    456 #ifdef DIAGNOSTIC
    457 		if (error != 0 && error != EWOULDBLOCK)
    458 			printf("acctwatch: sleep error %d\n", error);
    459 #endif
    460 	}
    461 	acct_dkwatcher = NULL;
    462 	ACCT_UNLOCK();
    463 
    464 	kthread_exit(0);
    465 }
    466