bsd/kern/kern_acct.c

   1 /*
   2  * Copyright (c) 2000-2004 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_internal.h>
  71 #include <sys/kauth.h>
  72 #include <sys/mount_internal.h>
  73 #include <sys/vnode_internal.h>
  74 #include <sys/file_internal.h>
  75 #include <sys/syslog.h>
  76 #include <sys/kernel.h>
  77 #include <sys/namei.h>
  78 #include <sys/errno.h>
  79 #include <sys/acct.h>
  80 #include <sys/resourcevar.h>
  81 #include <sys/ioctl.h>
  82 #include <sys/tty.h>
  83 #include <sys/sysproto.h>
  84 #include <machine/spl.h>
  85
  86 /*
  87  * The routines implemented in this file are described in:
  88  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
  89  *          UNIX Operating System (Addison Welley, 1989)
  90  * on pages 62-63.
  91  *
  92  * Arguably, to simplify accounting operations, this mechanism should
  93  * be replaced by one in which an accounting log file (similar to /dev/klog)
  94  * is read by a user process, etc.  However, that has its own problems.
  95  */
  96
  97 /*
  98  * Internal accounting functions.
  99  * The former's operation is described in Leffler, et al., and the latter
 100  * was provided by UCB with the 4.4BSD-Lite release
 101  */
 102 comp_t  encode_comp_t(u_long, u_long);
 103 void    acctwatch(void *);
 104 void    acctwatch_funnel(void *);
 105
 106 /*
 107  * Accounting vnode pointer, and suspended accounting vnode pointer.  States
 108  * are as follows:
 109  *
 110  *      acctp           suspend_acctp   state
 111  *      -------------   ------------    ------------------------------
 112  *      NULL            NULL            Accounting disabled
 113  *      !NULL           NULL            Accounting enabled
 114  *      NULL            !NULL           Accounting enabled, but suspended
 115  *      !NULL           !NULL           <not allowed>
 116  */
 117 struct  vnode *acctp;
 118 struct  vnode *suspend_acctp;
 119
 120 /*
 121  * Values associated with enabling and disabling accounting
 122  */
 123 int     acctsuspend = 2;        /* stop accounting when < 2% free space left */
 124 int     acctresume = 4;         /* resume when free space risen to > 4% */
 125 int     acctchkfreq = 15;       /* frequency (in seconds) to check space */
 126
 127 /*
 128  * Accounting system call.  Written based on the specification and
 129  * previous implementation done by Mark Tinguely.
 130  */
 131 int
 132 acct(struct proc *p, struct acct_args *uap, __unused int *retval)
 133 {
 134         struct nameidata nd;
 135         int error;
 136         struct vfs_context context;
 137
 138         context.vc_proc = p;
 139         context.vc_ucred = kauth_cred_get();
 140
 141         /* Make sure that the caller is root. */
 142         if ((error = suser(kauth_cred_get(), &p->p_acflag)))
 143                 return (error);
 144
 145         /*
 146          * If accounting is to be started to a file, open that file for
 147          * writing and make sure it's a 'normal'.
 148          */
 149         if (uap->path != USER_ADDR_NULL) {
 150                 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->path, &context);
 151                 if ((error = vn_open(&nd, FWRITE, 0)))
 152                         return (error);
 153                 vnode_put(nd.ni_vp);
 154
 155                 if (nd.ni_vp->v_type != VREG) {
 156                         vn_close(nd.ni_vp, FWRITE, kauth_cred_get(), p);
 157                         return (EACCES);
 158                 }
 159         }
 160
 161         /*
 162          * If accounting was previously enabled, kill the old space-watcher,
 163          * close the file, and (if no new file was specified, leave).
 164          */
 165         if (acctp != NULLVP || suspend_acctp != NULLVP) {
 166                 untimeout(acctwatch_funnel, NULL);
 167                 error = vn_close((acctp != NULLVP ? acctp : suspend_acctp), FWRITE,
 168                     kauth_cred_get(), p);
 169
 170                 acctp = suspend_acctp = NULLVP;
 171         }
 172         if (uap->path == USER_ADDR_NULL)
 173                 return (error);
 174
 175         /*
 176          * Save the new accounting file vnode, and schedule the new
 177          * free space watcher.
 178          */
 179         acctp = nd.ni_vp;
 180         acctwatch(NULL);
 181         return (error);
 182 }
 183
 184 /*
 185  * Write out process accounting information, on process exit.
 186  * Data to be written out is specified in Leffler, et al.
 187  * and are enumerated below.  (They're also noted in the system
 188  * "acct.h" header file.)
 189  */
 190 int
 191 acct_process(p)
 192         struct proc *p;
 193 {
 194         struct acct an_acct;
 195         struct rusage *r;
 196         struct timeval ut, st, tmp;
 197         int t;
 198         int error;
 199         struct vnode *vp;
 200
 201         /* If accounting isn't enabled, don't bother */
 202         vp = acctp;
 203         if (vp == NULLVP)
 204                 return (0);
 205
 206         /*
 207          * Get process accounting information.
 208          */
 209
 210         /* (1) The name of the command that ran */
 211         bcopy(p->p_comm, an_acct.ac_comm, sizeof an_acct.ac_comm);
 212
 213         /* (2) The amount of user and system time that was used */
 214         calcru(p, &ut, &st, NULL);
 215         an_acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
 216         an_acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
 217
 218         /* (3) The elapsed time the commmand ran (and its starting time) */
 219         an_acct.ac_btime = p->p_stats->p_start.tv_sec;
 220         microtime(&tmp);
 221         timevalsub(&tmp, &p->p_stats->p_start);
 222         an_acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
 223
 224         /* (4) The average amount of memory used */
 225         r = &p->p_stats->p_ru;
 226         tmp = ut;
 227         timevaladd(&tmp, &st);
 228         t = tmp.tv_sec * hz + tmp.tv_usec / tick;
 229         if (t)
 230                 an_acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
 231         else
 232                 an_acct.ac_mem = 0;
 233
 234         /* (5) The number of disk I/O operations done */
 235         an_acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
 236
 237         /* (6) The UID and GID of the process */
 238         an_acct.ac_uid = p->p_ucred->cr_ruid;
 239         an_acct.ac_gid = p->p_ucred->cr_rgid;
 240
 241         /* (7) The terminal from which the process was started */
 242         if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
 243                 an_acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev;
 244         else
 245                 an_acct.ac_tty = NODEV;
 246
 247         /* (8) The boolean flags that tell how the process terminated, etc. */
 248         an_acct.ac_flag = p->p_acflag;
 249
 250         /*
 251          * Now, just write the accounting information to the file.
 252          */
 253         if ((error = vnode_getwithref(vp)) == 0) {
 254                 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)&an_acct, sizeof (an_acct),
 255                                 (off_t)0, UIO_SYSSPACE32, IO_APPEND|IO_UNIT, p->p_ucred,
 256                                 (int *)0, p);
 257                 vnode_put(vp);
 258         }
 259         return (error);
 260 }
 261
 262 /*
 263  * Encode_comp_t converts from ticks in seconds and microseconds
 264  * to ticks in 1/AHZ seconds.  The encoding is described in
 265  * Leffler, et al., on page 63.
 266  */
 267
 268 #define MANTSIZE        13                      /* 13 bit mantissa. */
 269 #define EXPSIZE         3                       /* Base 8 (3 bit) exponent. */
 270 #define MAXFRACT        ((1 << MANTSIZE) - 1)   /* Maximum fractional value. */
 271
 272 comp_t
 273 encode_comp_t(s, us)
 274         u_long s, us;
 275 {
 276         int exp, rnd;
 277
 278         exp = 0;
 279         rnd = 0;
 280         s *= AHZ;
 281         s += us / (1000000 / AHZ);      /* Maximize precision. */
 282
 283         while (s > MAXFRACT) {
 284         rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */
 285                 s >>= EXPSIZE;          /* Base 8 exponent == 3 bit shift. */
 286                 exp++;
 287         }
 288
 289         /* If we need to round up, do it (and handle overflow correctly). */
 290         if (rnd && (++s > MAXFRACT)) {
 291                 s >>= EXPSIZE;
 292                 exp++;
 293         }
 294
 295         /* Clean it up and polish it off. */
 296         exp <<= MANTSIZE;               /* Shift the exponent into place */
 297         exp += s;                       /* and add on the mantissa. */
 298         return (exp);
 299 }
 300
 301 void
 302 acctwatch_funnel(a)
 303         void *a;
 304 {
 305         thread_funnel_set(kernel_flock, TRUE);
 306         acctwatch(a);
 307         thread_funnel_set(kernel_flock, FALSE);
 308 }
 309
 310
 311 /*
 312  * Periodically check the file system to see if accounting
 313  * should be turned on or off.  Beware the case where the vnode
 314  * has been vgone()'d out from underneath us, e.g. when the file
 315  * system containing the accounting file has been forcibly unmounted.
 316  */
 317 /* ARGSUSED */
 318 void
 319 acctwatch(__unused void *a)
 320 {
 321         struct vfs_context context;
 322         struct vfs_attr va;
 323
 324         VFSATTR_INIT(&va);
 325         VFSATTR_WANTED(&va, f_blocks);
 326         VFSATTR_WANTED(&va, f_bavail);
 327         context.vc_proc = current_proc();
 328         context.vc_ucred = kauth_cred_get();
 329
 330         if (suspend_acctp != NULLVP) {
 331                 /*
 332                  * Resuming accounting when accounting is suspended, and the
 333                  * filesystem containing the suspended accounting file goes
 334                  * below a low watermark
 335                  */
 336                 if (suspend_acctp->v_type == VBAD) {
 337                         (void) vn_close(suspend_acctp, FWRITE, NOCRED, NULL);
 338                         suspend_acctp = NULLVP;
 339                         return;
 340                 }
 341                 (void)vfs_getattr(suspend_acctp->v_mount, &va, &context);
 342                 if (va.f_bavail > acctresume * va.f_blocks / 100) {
 343                         acctp = suspend_acctp;
 344                         suspend_acctp = NULLVP;
 345                         log(LOG_NOTICE, "Accounting resumed\n");
 346                 }
 347         } else if (acctp != NULLVP) {
 348                 /*
 349                  * Suspending accounting when accounting is currently active,
 350                  * and the filesystem containing the active accounting file
 351                  * goes over a high watermark
 352                  */
 353                 if (acctp->v_type == VBAD) {
 354                         (void) vn_close(acctp, FWRITE, NOCRED, NULL);
 355                         acctp = NULLVP;
 356                         return;
 357                 }
 358                 (void)vfs_getattr(acctp->v_mount, &va, &context);
 359                 if (va.f_bavail <= acctsuspend * va.f_blocks / 100) {
 360                         suspend_acctp = acctp;
 361                         acctp = NULLVP;
 362                         log(LOG_NOTICE, "Accounting suspended\n");
 363                 }
 364         } else {
 365                 return;
 366         }
 367
 368         timeout(acctwatch_funnel, NULL, acctchkfreq * hz);
 369 }