bsd/kern/kern_acct.c

   1 /*
   2  * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * The contents of this file constitute Original Code as defined in and
   7  * are subject to the Apple Public Source License Version 1.1 (the
   8  * "License").  You may not use this file except in compliance with the
   9  * License.  Please obtain a copy of the License at
  10  * http://www.apple.com/publicsource and read it before using this file.
  11  *
  12  * This Original Code and all software distributed under the License are
  13  * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  14  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  15  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
  17  * License for the specific language governing rights and limitations
  18  * under the License.
  19  *
  20  * @APPLE_LICENSE_HEADER_END@
  21  */
  22 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
  23 /*-
  24  * Copyright (c) 1982, 1986, 1989, 1993
  25  *      The Regents of the University of California.  All rights reserved.
  26  * (c) UNIX System Laboratories, Inc.
  27  * All or some portions of this file are derived from material licensed
  28  * to the University of California by American Telephone and Telegraph
  29  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
  30  * the permission of UNIX System Laboratories, Inc.
  31  *
  32  * Redistribution and use in source and binary forms, with or without
  33  * modification, are permitted provided that the following conditions
  34  * are met:
  35  * 1. Redistributions of source code must retain the above copyright
  36  *    notice, this list of conditions and the following disclaimer.
  37  * 2. Redistributions in binary form must reproduce the above copyright
  38  *    notice, this list of conditions and the following disclaimer in the
  39  *    documentation and/or other materials provided with the distribution.
  40  * 3. All advertising materials mentioning features or use of this software
  41  *    must display the following acknowledgement:
  42  *      This product includes software developed by the University of
  43  *      California, Berkeley and its contributors.
  44  * 4. Neither the name of the University nor the names of its contributors
  45  *    may be used to endorse or promote products derived from this software
  46  *    without specific prior written permission.
  47  *
  48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  58  * SUCH DAMAGE.
  59  *
  60  *      @(#)kern_acct.c 8.1 (Berkeley) 6/14/93
  61  */
  62 /* HISTORY
  63  * 08-May-95  Mac Gillon (mgillon) at NeXT
  64  *      Purged old history
  65  *      New version based on 4.4
  66  */
  67
  68
  69 #include <sys/param.h>
  70 #include <sys/proc.h>
  71 #include <sys/mount.h>
  72 #include <sys/vnode.h>
  73 #include <sys/file.h>
  74 #include <sys/syslog.h>
  75 #include <sys/kernel.h>
  76 #include <sys/namei.h>
  77 #include <sys/errno.h>
  78 #include <sys/acct.h>
  79 #include <sys/resourcevar.h>
  80 #include <sys/ioctl.h>
  81 #include <sys/tty.h>
  82
  83 /*
  84  * The routines implemented in this file are described in:
  85  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
  86  *          UNIX Operating System (Addison Welley, 1989)
  87  * on pages 62-63.
  88  *
  89  * Arguably, to simplify accounting operations, this mechanism should
  90  * be replaced by one in which an accounting log file (similar to /dev/klog)
  91  * is read by a user process, etc.  However, that has its own problems.
  92  */
  93
  94 /*
  95  * Internal accounting functions.
  96  * The former's operation is described in Leffler, et al., and the latter
  97  * was provided by UCB with the 4.4BSD-Lite release
  98  */
  99 comp_t  encode_comp_t __P((u_long, u_long));
 100 void    acctwatch __P((void *));
 101 void    acctwatch_funnel __P((void *));
 102
 103 /*
 104  * Accounting vnode pointer, and saved vnode pointer.
 105  */
 106 struct  vnode *acctp;
 107 struct  vnode *savacctp;
 108
 109 /*
 110  * Values associated with enabling and disabling accounting
 111  */
 112 int     acctsuspend = 2;        /* stop accounting when < 2% free space left */
 113 int     acctresume = 4;         /* resume when free space risen to > 4% */
 114 int     acctchkfreq = 15;       /* frequency (in seconds) to check space */
 115
 116 /*
 117  * Accounting system call.  Written based on the specification and
 118  * previous implementation done by Mark Tinguely.
 119  */
 120 struct acct_args {
 121         char    *path;
 122 };
 123 acct(p, uap, retval)
 124         struct proc *p;
 125         struct acct_args *uap;
 126         int *retval;
 127 {
 128         struct nameidata nd;
 129         int error;
 130
 131         /* Make sure that the caller is root. */
 132         if (error = suser(p->p_ucred, &p->p_acflag))
 133                 return (error);
 134
 135         /*
 136          * If accounting is to be started to a file, open that file for
 137          * writing and make sure it's a 'normal'.
 138          */
 139         if (uap->path != NULL) {
 140                 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->path, p);
 141                 if (error = vn_open(&nd, FWRITE, 0))
 142                         return (error);
 143                 VOP_UNLOCK(nd.ni_vp, 0, p);
 144                 if (nd.ni_vp->v_type != VREG) {
 145                         vn_close(nd.ni_vp, FWRITE, p->p_ucred, p);
 146                         return (EACCES);
 147                 }
 148         }
 149
 150         /*
 151          * If accounting was previously enabled, kill the old space-watcher,
 152          * close the file, and (if no new file was specified, leave).
 153          */
 154         if (acctp != NULLVP || savacctp != NULLVP) {
 155                 untimeout(acctwatch_funnel, NULL);
 156                 error = vn_close((acctp != NULLVP ? acctp : savacctp), FWRITE,
 157                     p->p_ucred, p);
 158                 acctp = savacctp = NULLVP;
 159         }
 160         if (uap->path == NULL)
 161                 return (error);
 162
 163         /*
 164          * Save the new accounting file vnode, and schedule the new
 165          * free space watcher.
 166          */
 167         acctp = nd.ni_vp;
 168         acctwatch(NULL);
 169         return (error);
 170 }
 171
 172 /*
 173  * Write out process accounting information, on process exit.
 174  * Data to be written out is specified in Leffler, et al.
 175  * and are enumerated below.  (They're also noted in the system
 176  * "acct.h" header file.)
 177  */
 178 acct_process(p)
 179         struct proc *p;
 180 {
 181         struct acct acct;
 182         struct rusage *r;
 183         struct timeval ut, st, tmp;
 184         int s, t;
 185         struct vnode *vp;
 186
 187         /* If accounting isn't enabled, don't bother */
 188         vp = acctp;
 189         if (vp == NULLVP)
 190                 return (0);
 191
 192         /*
 193          * Get process accounting information.
 194          */
 195
 196         /* (1) The name of the command that ran */
 197         bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
 198
 199         /* (2) The amount of user and system time that was used */
 200         calcru(p, &ut, &st, NULL);
 201         acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
 202         acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
 203
 204         /* (3) The elapsed time the commmand ran (and its starting time) */
 205         acct.ac_btime = p->p_stats->p_start.tv_sec;
 206         s = splclock();
 207         tmp = time;
 208         splx(s);
 209         timevalsub(&tmp, &p->p_stats->p_start);
 210         acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
 211
 212         /* (4) The average amount of memory used */
 213         r = &p->p_stats->p_ru;
 214         tmp = ut;
 215         timevaladd(&tmp, &st);
 216         t = tmp.tv_sec * hz + tmp.tv_usec / tick;
 217         if (t)
 218                 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
 219         else
 220                 acct.ac_mem = 0;
 221
 222         /* (5) The number of disk I/O operations done */
 223         acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
 224
 225         /* (6) The UID and GID of the process */
 226         acct.ac_uid = p->p_cred->p_ruid;
 227         acct.ac_gid = p->p_cred->p_rgid;
 228
 229         /* (7) The terminal from which the process was started */
 230         if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
 231                 acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
 232         else
 233                 acct.ac_tty = NODEV;
 234
 235         /* (8) The boolean flags that tell how the process terminated, etc. */
 236         acct.ac_flag = p->p_acflag;
 237
 238         /*
 239          * Now, just write the accounting information to the file.
 240          */
 241         VOP_LEASE(vp, p, p->p_ucred, LEASE_WRITE);
 242         return (vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
 243             (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, p->p_ucred,
 244             (int *)0, p));
 245 }
 246
 247 /*
 248  * Encode_comp_t converts from ticks in seconds and microseconds
 249  * to ticks in 1/AHZ seconds.  The encoding is described in
 250  * Leffler, et al., on page 63.
 251  */
 252
 253 #define MANTSIZE        13                      /* 13 bit mantissa. */
 254 #define EXPSIZE         3                       /* Base 8 (3 bit) exponent. */
 255 #define MAXFRACT        ((1 << MANTSIZE) - 1)   /* Maximum fractional value. */
 256
 257 comp_t
 258 encode_comp_t(s, us)
 259         u_long s, us;
 260 {
 261         int exp, rnd;
 262
 263         exp = 0;
 264         rnd = 0;
 265         s *= AHZ;
 266         s += us / (1000000 / AHZ);      /* Maximize precision. */
 267
 268         while (s > MAXFRACT) {
 269         rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */
 270                 s >>= EXPSIZE;          /* Base 8 exponent == 3 bit shift. */
 271                 exp++;
 272         }
 273
 274         /* If we need to round up, do it (and handle overflow correctly). */
 275         if (rnd && (++s > MAXFRACT)) {
 276                 s >>= EXPSIZE;
 277                 exp++;
 278         }
 279
 280         /* Clean it up and polish it off. */
 281         exp <<= MANTSIZE;               /* Shift the exponent into place */
 282         exp += s;                       /* and add on the mantissa. */
 283         return (exp);
 284 }
 285
 286 void
 287 acctwatch_funnel(a)
 288         void *a;
 289 {
 290         thread_funnel_set(kernel_flock, TRUE);
 291         acctwatch(a);
 292         thread_funnel_set(kernel_flock, FALSE);
 293 }
 294
 295
 296 /*
 297  * Periodically check the file system to see if accounting
 298  * should be turned on or off.  Beware the case where the vnode
 299  * has been vgone()'d out from underneath us, e.g. when the file
 300  * system containing the accounting file has been forcibly unmounted.
 301  */
 302 /* ARGSUSED */
 303 void
 304 acctwatch(a)
 305         void *a;
 306 {
 307         struct statfs sb;
 308
 309         if (savacctp != NULLVP) {
 310                 if (savacctp->v_type == VBAD) {
 311                         (void) vn_close(savacctp, FWRITE, NOCRED, NULL);
 312                         savacctp = NULLVP;
 313                         return;
 314                 }
 315                 (void)VFS_STATFS(savacctp->v_mount, &sb, (struct proc *)0);
 316                 if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
 317                         acctp = savacctp;
 318                         savacctp = NULLVP;
 319                         log(LOG_NOTICE, "Accounting resumed\n");
 320                 }
 321         } else if (acctp != NULLVP) {
 322                 if (acctp->v_type == VBAD) {
 323                         (void) vn_close(acctp, FWRITE, NOCRED, NULL);
 324                         acctp = NULLVP;
 325                         return;
 326                 }
 327                 (void)VFS_STATFS(acctp->v_mount, &sb, (struct proc *)0);
 328                 if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
 329                         savacctp = acctp;
 330                         acctp = NULLVP;
 331                         log(LOG_NOTICE, "Accounting suspended\n");
 332                 }
 333         } else {
 334                 return;
 335         }
 336
 337         timeout(acctwatch_funnel, NULL, acctchkfreq * hz);
 338 }