sys/kern/kern_acct.c

   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  *      @(#)kern_acct.c 8.1 (Berkeley) 6/14/93
  40  * $FreeBSD: src/sys/kern/kern_acct.c,v 1.23.2.1 2002/07/24 18:33:55 johan Exp $
  41  * $DragonFly: src/sys/kern/kern_acct.c,v 1.19 2005/10/08 19:46:50 corecode Exp $
  42  */
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/sysproto.h>
  47 #include <sys/proc.h>
  48 #include <sys/mount.h>
  49 #include <sys/vnode.h>
  50 #include <sys/fcntl.h>
  51 #include <sys/syslog.h>
  52 #include <sys/kernel.h>
  53 #include <sys/sysent.h>
  54 #include <sys/sysctl.h>
  55 #include <sys/nlookup.h>
  56 #include <sys/acct.h>
  57 #include <sys/resourcevar.h>
  58 #include <sys/tty.h>
  59
  60 #include <vm/vm_zone.h>
  61
  62 /*
  63  * The routines implemented in this file are described in:
  64  *      Leffler, et al.: The Design and Implementation of the 4.3BSD
  65  *          UNIX Operating System (Addison Welley, 1989)
  66  * on pages 62-63.
  67  *
  68  * Arguably, to simplify accounting operations, this mechanism should
  69  * be replaced by one in which an accounting log file (similar to /dev/klog)
  70  * is read by a user process, etc.  However, that has its own problems.
  71  */
  72
  73 /*
  74  * Internal accounting functions.
  75  * The former's operation is described in Leffler, et al., and the latter
  76  * was provided by UCB with the 4.4BSD-Lite release
  77  */
  78 static comp_t   encode_comp_t (u_long, u_long);
  79 static void     acctwatch (void *);
  80
  81 /*
  82  * Accounting callout handle used for periodic scheduling of
  83  * acctwatch.
  84  */
  85 static struct   callout acctwatch_handle;
  86
  87 /*
  88  * Accounting vnode pointer, and saved vnode pointer.
  89  */
  90 static struct   vnode *acctp;
  91 static struct   vnode *savacctp;
  92
  93 /*
  94  * Values associated with enabling and disabling accounting
  95  */
  96 static int acctsuspend = 2;     /* stop accounting when < 2% free space left */
  97 SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW,
  98         &acctsuspend, 0, "Percentage of free disk space below which accounting will be suspended");
  99
 100 static int acctresume = 4;      /* resume when free space risen to > 4% */
 101 SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW,
 102         &acctresume, 0, "Minimum percentage of free disk space needed to resume accounting");
 103
 104 static int acctchkfreq = 15;    /* frequency (in seconds) to check space */
 105 SYSCTL_INT(_kern, OID_AUTO, acct_chkfreq, CTLFLAG_RW,
 106         &acctchkfreq, 0, "Frequency (in seconds) of free disk space checking");
 107
 108 static void
 109 acct_init(void *arg __unused)
 110 {
 111         callout_init(&acctwatch_handle);
 112 }
 113 SYSINIT(acct, SI_SUB_DRIVERS, SI_ORDER_ANY, acct_init, NULL);
 114
 115 /*
 116  * Accounting system call.  Written based on the specification and
 117  * previous implementation done by Mark Tinguely.
 118  *
 119  * acct(char *path)
 120  */
 121 int
 122 acct(struct acct_args *uap)
 123 {
 124         struct thread *td = curthread;
 125         struct nlookupdata nd;
 126         struct vnode *vp;
 127         int error;
 128
 129         /* Make sure that the caller is root. */
 130         error = suser(td);
 131         if (error)
 132                 return (error);
 133
 134         /*
 135          * If accounting is to be started to a file, open that file for
 136          * appending and make sure it's a 'normal'.
 137          */
 138         if (uap->path != NULL) {
 139                 error = nlookup_init(&nd, uap->path, UIO_USERSPACE,
 140                                         NLC_LOCKVP);
 141                 if (error == 0)
 142                         error = vn_open(&nd, NULL, FWRITE | O_APPEND, 0);
 143                 if (error == 0 && nd.nl_open_vp->v_type != VREG)
 144                         error = EACCES;
 145                 if (error) {
 146                         nlookup_done(&nd);
 147                         return (error);
 148                 }
 149                 vp = nd.nl_open_vp;
 150                 nd.nl_open_vp = NULL;
 151                 nlookup_done(&nd);
 152
 153                 VOP_UNLOCK(vp, 0, td);
 154         } else {
 155                 vp = NULL;
 156         }
 157
 158         /*
 159          * If accounting was previously enabled, kill the old space-watcher,
 160          * close the file.
 161          */
 162         if (acctp != NULLVP || savacctp != NULLVP) {
 163                 callout_stop(&acctwatch_handle);
 164                 error = vn_close((acctp != NULLVP ? acctp : savacctp),
 165                                 FWRITE | O_APPEND, td);
 166                 acctp = savacctp = NULLVP;
 167         }
 168
 169         /*
 170          * If no new file opened then leave.  We never did an nlookup so
 171          * don't try cleaning it up.
 172          */
 173         if (uap->path == NULL)
 174                 return (error);
 175
 176         /*
 177          * Save the new accounting file vnode, and schedule the new
 178          * free space watcher.
 179          */
 180         acctp = vp;
 181         acctwatch(NULL);
 182         return (error);
 183 }
 184
 185 /*
 186  * Write out process accounting information, on process exit.
 187  * Data to be written out is specified in Leffler, et al.
 188  * and are enumerated below.  (They're also noted in the system
 189  * "acct.h" header file.)
 190  */
 191
 192 int
 193 acct_process(struct proc *p)
 194 {
 195         struct acct acct;
 196         struct rusage *r;
 197         struct timeval ut, st, tmp;
 198         int t;
 199         struct vnode *vp;
 200         struct thread *td = p->p_thread;
 201
 202         /* If accounting isn't enabled, don't bother */
 203         vp = acctp;
 204         if (vp == NULLVP)
 205                 return (0);
 206
 207         /*
 208          * Get process accounting information.
 209          */
 210
 211         /* (1) The name of the command that ran */
 212         bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
 213
 214         /* (2) The amount of user and system time that was used */
 215         calcru(p, &ut, &st, NULL);
 216         acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec);
 217         acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec);
 218
 219         /* (3) The elapsed time the commmand ran (and its starting time) */
 220         acct.ac_btime = p->p_start.tv_sec;
 221         microtime(&tmp);
 222         timevalsub(&tmp, &p->p_start);
 223         acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec);
 224
 225         /* (4) The average amount of memory used */
 226         r = &p->p_stats->p_ru;
 227         tmp = ut;
 228         timevaladd(&tmp, &st);
 229         t = tmp.tv_sec * hz + tmp.tv_usec / tick;
 230         if (t)
 231                 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t;
 232         else
 233                 acct.ac_mem = 0;
 234
 235         /* (5) The number of disk I/O operations done */
 236         acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0);
 237
 238         /* (6) The UID and GID of the process */
 239         acct.ac_uid = p->p_ucred->cr_ruid;
 240         acct.ac_gid = p->p_ucred->cr_rgid;
 241
 242         /* (7) The terminal from which the process was started */
 243         if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
 244                 acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev);
 245         else
 246                 acct.ac_tty = NOUDEV;
 247
 248         /* (8) The boolean flags that tell how the process terminated, etc. */
 249         acct.ac_flag = p->p_acflag;
 250
 251         /*
 252          * Eliminate any file size rlimit.
 253          */
 254         if (p->p_limit->p_refcnt > 1 &&
 255             (p->p_limit->p_lflags & PL_SHAREMOD) == 0) {
 256                 p->p_limit->p_refcnt--;
 257                 p->p_limit = limcopy(p->p_limit);
 258         }
 259         p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
 260
 261         /*
 262          * Write the accounting information to the file.
 263          */
 264         VOP_LEASE(vp, td, p->p_ucred, LEASE_WRITE);
 265         return (vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct),
 266             (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, p->p_ucred,
 267             (int *)0, td));
 268 }
 269
 270 /*
 271  * Encode_comp_t converts from ticks in seconds and microseconds
 272  * to ticks in 1/AHZ seconds.  The encoding is described in
 273  * Leffler, et al., on page 63.
 274  */
 275
 276 #define MANTSIZE        13                      /* 13 bit mantissa. */
 277 #define EXPSIZE         3                       /* Base 8 (3 bit) exponent. */
 278 #define MAXFRACT        ((1 << MANTSIZE) - 1)   /* Maximum fractional value. */
 279
 280 static comp_t
 281 encode_comp_t(u_long s, u_long us)
 282 {
 283         int exp, rnd;
 284
 285         exp = 0;
 286         rnd = 0;
 287         s *= AHZ;
 288         s += us / (1000000 / AHZ);      /* Maximize precision. */
 289
 290         while (s > MAXFRACT) {
 291         rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */
 292                 s >>= EXPSIZE;          /* Base 8 exponent == 3 bit shift. */
 293                 exp++;
 294         }
 295
 296         /* If we need to round up, do it (and handle overflow correctly). */
 297         if (rnd && (++s > MAXFRACT)) {
 298                 s >>= EXPSIZE;
 299                 exp++;
 300         }
 301
 302         /* Clean it up and polish it off. */
 303         exp <<= MANTSIZE;               /* Shift the exponent into place */
 304         exp += s;                       /* and add on the mantissa. */
 305         return (exp);
 306 }
 307
 308 /*
 309  * Periodically check the file system to see if accounting
 310  * should be turned on or off.  Beware the case where the vnode
 311  * has been vgone()'d out from underneath us, e.g. when the file
 312  * system containing the accounting file has been forcibly unmounted.
 313  */
 314 /* ARGSUSED */
 315 static void
 316 acctwatch(void *a)
 317 {
 318         struct statfs sb;
 319
 320         if (savacctp != NULLVP) {
 321                 if (savacctp->v_type == VBAD) {
 322                         (void) vn_close(savacctp, FWRITE | O_APPEND, NULL);
 323                         savacctp = NULLVP;
 324                         return;
 325                 }
 326                 (void)VFS_STATFS(savacctp->v_mount, &sb, NULL);
 327                 if (sb.f_bavail > acctresume * sb.f_blocks / 100) {
 328                         acctp = savacctp;
 329                         savacctp = NULLVP;
 330                         log(LOG_NOTICE, "Accounting resumed\n");
 331                 }
 332         } else {
 333                 if (acctp == NULLVP)
 334                         return;
 335                 if (acctp->v_type == VBAD) {
 336                         (void) vn_close(acctp, FWRITE | O_APPEND, NULL);
 337                         acctp = NULLVP;
 338                         return;
 339                 }
 340                 (void)VFS_STATFS(acctp->v_mount, &sb, NULL);
 341                 if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) {
 342                         savacctp = acctp;
 343                         acctp = NULLVP;
 344                         log(LOG_NOTICE, "Accounting suspended\n");
 345                 }
 346         }
 347         callout_reset(&acctwatch_handle, acctchkfreq * hz, acctwatch, NULL);
 348 }