kern_acct.c revision 29041
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 * $Id: kern_acct.c,v 1.15 1997/03/24 11:24:34 bde Exp $ 41 */ 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/sysproto.h> 46#include <sys/proc.h> 47#include <sys/mount.h> 48#include <sys/vnode.h> 49#include <sys/fcntl.h> 50#include <sys/syslog.h> 51#include <sys/kernel.h> 52#include <sys/sysent.h> 53#include <sys/sysctl.h> 54#include <sys/namei.h> 55#include <sys/acct.h> 56#include <sys/resourcevar.h> 57#include <sys/tty.h> 58 59/* 60 * The routines implemented in this file are described in: 61 * Leffler, et al.: The Design and Implementation of the 4.3BSD 62 * UNIX Operating System (Addison Welley, 1989) 63 * on pages 62-63. 64 * 65 * Arguably, to simplify accounting operations, this mechanism should 66 * be replaced by one in which an accounting log file (similar to /dev/klog) 67 * is read by a user process, etc. However, that has its own problems. 68 */ 69 70/* 71 * Internal accounting functions. 72 * The former's operation is described in Leffler, et al., and the latter 73 * was provided by UCB with the 4.4BSD-Lite release 74 */ 75static comp_t encode_comp_t __P((u_long, u_long)); 76static void acctwatch __P((void *)); 77 78/* 79 * Accounting vnode pointer, and saved vnode pointer. 80 */ 81static struct vnode *acctp; 82static struct vnode *savacctp; 83 84/* 85 * Values associated with enabling and disabling accounting 86 */ 87static int acctsuspend = 2; /* stop accounting when < 2% free space left */ 88SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW, 89 &acctsuspend, 0, ""); 90 91static int acctresume = 4; /* resume when free space risen to > 4% */ 92SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW, 93 &acctresume, 0, ""); 94 95static int acctchkfreq = 15; /* frequency (in seconds) to check space */ 96SYSCTL_INT(_kern, OID_AUTO, acct_chkfreq, CTLFLAG_RW, 97 &acctchkfreq, 0, ""); 98 99/* 100 * Accounting system call. Written based on the specification and 101 * previous implementation done by Mark Tinguely. 102 */ 103int 104acct(a1, uap, a3) 105 struct proc *a1; 106 struct acct_args /* { 107 syscallarg(char *) path; 108 } */ *uap; 109 int *a3; 110{ 111 struct proc *p = curproc; /* XXX */ 112 struct nameidata nd; 113 int error; 114 115 /* Make sure that the caller is root. */ 116 error = suser(p->p_ucred, &p->p_acflag); 117 if (error) 118 return (error); 119 120 /* 121 * If accounting is to be started to a file, open that file for 122 * writing and make sure it's a 'normal'. 123 */ 124 if (SCARG(uap, path) != NULL) { 125 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), 126 p); 127 error = vn_open(&nd, FWRITE, 0); 128 if (error) 129 return (error); 130 VOP_UNLOCK(nd.ni_vp, 0, p); 131 if (nd.ni_vp->v_type != VREG) { 132 vn_close(nd.ni_vp, FWRITE, p->p_ucred, p); 133 return (EACCES); 134 } 135 } 136 137 /* 138 * If accounting was previously enabled, kill the old space-watcher, 139 * close the file, and (if no new file was specified, leave). 140 */ 141 if (acctp != NULLVP || savacctp != NULLVP) { 142 untimeout(acctwatch, NULL); 143 error = vn_close((acctp != NULLVP ? acctp : savacctp), FWRITE, 144 p->p_ucred, p); 145 acctp = savacctp = NULLVP; 146 } 147 if (SCARG(uap, path) == NULL) 148 return (error); 149 150 /* 151 * Save the new accounting file vnode, and schedule the new 152 * free space watcher. 153 */ 154 acctp = nd.ni_vp; 155 acctwatch(NULL); 156 return (error); 157} 158 159/* 160 * Write out process accounting information, on process exit. 161 * Data to be written out is specified in Leffler, et al. 162 * and are enumerated below. (They're also noted in the system 163 * "acct.h" header file.) 164 */ 165 166int 167acct_process(p) 168 struct proc *p; 169{ 170 struct acct acct; 171 struct rusage *r; 172 struct timeval ut, st, tmp; 173 int t; 174 struct vnode *vp; 175 176 /* If accounting isn't enabled, don't bother */ 177 vp = acctp; 178 if (vp == NULLVP) 179 return (0); 180 181 /* 182 * Get process accounting information. 183 */ 184 185 /* (1) The name of the command that ran */ 186 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm); 187 188 /* (2) The amount of user and system time that was used */ 189 calcru(p, &ut, &st, NULL); 190 acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec); 191 acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec); 192 193 /* (3) The elapsed time the commmand ran (and its starting time) */ 194 acct.ac_btime = p->p_stats->p_start.tv_sec; 195 microtime(&tmp); 196 timevalsub(&tmp, &p->p_stats->p_start); 197 acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec); 198 199 /* (4) The average amount of memory used */ 200 r = &p->p_stats->p_ru; 201 tmp = ut; 202 timevaladd(&tmp, &st); 203 t = tmp.tv_sec * hz + tmp.tv_usec / tick; 204 if (t) 205 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t; 206 else 207 acct.ac_mem = 0; 208 209 /* (5) The number of disk I/O operations done */ 210 acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0); 211 212 /* (6) The UID and GID of the process */ 213 acct.ac_uid = p->p_cred->p_ruid; 214 acct.ac_gid = p->p_cred->p_rgid; 215 216 /* (7) The terminal from which the process was started */ 217 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp) 218 acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev; 219 else 220 acct.ac_tty = NODEV; 221 222 /* (8) The boolean flags that tell how the process terminated, etc. */ 223 acct.ac_flag = p->p_acflag; 224 225 /* 226 * Now, just write the accounting information to the file. 227 */ 228 VOP_LEASE(vp, p, p->p_ucred, LEASE_WRITE); 229 return (vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct), 230 (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, p->p_ucred, 231 (int *)0, p)); 232} 233 234/* 235 * Encode_comp_t converts from ticks in seconds and microseconds 236 * to ticks in 1/AHZ seconds. The encoding is described in 237 * Leffler, et al., on page 63. 238 */ 239 240#define MANTSIZE 13 /* 13 bit mantissa. */ 241#define EXPSIZE 3 /* Base 8 (3 bit) exponent. */ 242#define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */ 243 244static comp_t 245encode_comp_t(s, us) 246 u_long s, us; 247{ 248 int exp, rnd; 249 250 exp = 0; 251 rnd = 0; 252 s *= AHZ; 253 s += us / (1000000 / AHZ); /* Maximize precision. */ 254 255 while (s > MAXFRACT) { 256 rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */ 257 s >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */ 258 exp++; 259 } 260 261 /* If we need to round up, do it (and handle overflow correctly). */ 262 if (rnd && (++s > MAXFRACT)) { 263 s >>= EXPSIZE; 264 exp++; 265 } 266 267 /* Clean it up and polish it off. */ 268 exp <<= MANTSIZE; /* Shift the exponent into place */ 269 exp += s; /* and add on the mantissa. */ 270 return (exp); 271} 272 273/* 274 * Periodically check the file system to see if accounting 275 * should be turned on or off. Beware the case where the vnode 276 * has been vgone()'d out from underneath us, e.g. when the file 277 * system containing the accounting file has been forcibly unmounted. 278 */ 279/* ARGSUSED */ 280static void 281acctwatch(a) 282 void *a; 283{ 284 struct statfs sb; 285 286 if (savacctp != NULLVP) { 287 if (savacctp->v_type == VBAD) { 288 (void) vn_close(savacctp, FWRITE, NOCRED, NULL); 289 savacctp = NULLVP; 290 return; 291 } 292 (void)VFS_STATFS(savacctp->v_mount, &sb, (struct proc *)0); 293 if (sb.f_bavail > acctresume * sb.f_blocks / 100) { 294 acctp = savacctp; 295 savacctp = NULLVP; 296 log(LOG_NOTICE, "Accounting resumed\n"); 297 } 298 } else { 299 if (acctp == NULLVP) 300 return; 301 if (acctp->v_type == VBAD) { 302 (void) vn_close(acctp, FWRITE, NOCRED, NULL); 303 acctp = NULLVP; 304 return; 305 } 306 (void)VFS_STATFS(acctp->v_mount, &sb, (struct proc *)0); 307 if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) { 308 savacctp = acctp; 309 acctp = NULLVP; 310 log(LOG_NOTICE, "Accounting suspended\n"); 311 } 312 } 313 timeout(acctwatch, NULL, acctchkfreq * hz); 314} 315