kern_acct.c revision 103208
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: head/sys/kern/kern_acct.c 103208 2002-09-11 04:10:41Z arr $ 41 */ 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/lock.h> 46#include <sys/mutex.h> 47#include <sys/sysproto.h> 48#include <sys/proc.h> 49#include <sys/mount.h> 50#include <sys/vnode.h> 51#include <sys/fcntl.h> 52#include <sys/syslog.h> 53#include <sys/kernel.h> 54#include <sys/sysent.h> 55#include <sys/sysctl.h> 56#include <sys/namei.h> 57#include <sys/acct.h> 58#include <sys/resourcevar.h> 59#include <sys/tty.h> 60 61/* 62 * The routines implemented in this file are described in: 63 * Leffler, et al.: The Design and Implementation of the 4.3BSD 64 * UNIX Operating System (Addison Welley, 1989) 65 * on pages 62-63. 66 * 67 * Arguably, to simplify accounting operations, this mechanism should 68 * be replaced by one in which an accounting log file (similar to /dev/klog) 69 * is read by a user process, etc. However, that has its own problems. 70 */ 71 72/* 73 * Internal accounting functions. 74 * The former's operation is described in Leffler, et al., and the latter 75 * was provided by UCB with the 4.4BSD-Lite release 76 */ 77static comp_t encode_comp_t(u_long, u_long); 78static void acctwatch(void *); 79 80/* 81 * Accounting callout used for periodic scheduling of acctwatch. 82 */ 83static struct callout acctwatch_callout; 84 85/* 86 * Accounting vnode pointer, saved vnode pointer, and flags for each. 87 */ 88static struct vnode *acctp; 89static struct ucred *acctcred; 90static int acctflags; 91static struct vnode *savacctp; 92static struct ucred *savacctcred; 93static int savacctflags; 94 95static struct mtx acct_mtx; 96MTX_SYSINIT(acct, &acct_mtx, "accounting", MTX_DEF); 97 98/* 99 * Values associated with enabling and disabling accounting 100 */ 101static int acctsuspend = 2; /* stop accounting when < 2% free space left */ 102SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW, 103 &acctsuspend, 0, "percentage of free disk space below which accounting stops"); 104 105static int acctresume = 4; /* resume when free space risen to > 4% */ 106SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW, 107 &acctresume, 0, "percentage of free disk space above which accounting resumes"); 108 109static int acctchkfreq = 15; /* frequency (in seconds) to check space */ 110SYSCTL_INT(_kern, OID_AUTO, acct_chkfreq, CTLFLAG_RW, 111 &acctchkfreq, 0, "frequency for checking the free space"); 112 113/* 114 * Accounting system call. Written based on the specification and 115 * previous implementation done by Mark Tinguely. 116 * 117 * MPSAFE 118 */ 119int 120acct(td, uap) 121 struct thread *td; 122 struct acct_args /* { 123 syscallarg(char *) path; 124 } */ *uap; 125{ 126 struct nameidata nd; 127 int error, flags; 128 129 /* Make sure that the caller is root. */ 130 error = suser(td); 131 if (error) 132 return (error); 133 134 mtx_lock(&Giant); 135 /* 136 * If accounting is to be started to a file, open that file for 137 * appending and make sure it's a 'normal'. 138 */ 139 if (SCARG(uap, path) != NULL) { 140 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), 141 td); 142 flags = FWRITE | O_APPEND; 143 error = vn_open(&nd, &flags, 0); 144 if (error) 145 goto done2; 146 NDFREE(&nd, NDF_ONLY_PNBUF); 147 VOP_UNLOCK(nd.ni_vp, 0, td); 148 if (nd.ni_vp->v_type != VREG) { 149 vn_close(nd.ni_vp, flags, td->td_ucred, td); 150 error = EACCES; 151 goto done2; 152 } 153 } 154 155 /* 156 * If accounting was previously enabled, kill the old space-watcher, 157 * close the file, and (if no new file was specified, leave). 158 */ 159 160 mtx_lock(&acct_mtx); 161 if (acctp != NULLVP || savacctp != NULLVP) { 162 callout_stop(&acctwatch_callout); 163 error = vn_close((acctp != NULLVP ? acctp : savacctp), 164 (acctp != NULLVP ? acctflags : savacctflags), 165 (acctcred != NOCRED ? acctcred : savacctcred), td); 166 acctp = savacctp = NULLVP; 167 crfree(acctcred != NOCRED ? acctcred : savacctcred); 168 acctcred = savacctcred = NOCRED; 169 } 170 if (SCARG(uap, path) == NULL) 171 goto done2; 172 173 /* 174 * Save the new accounting file vnode, and schedule the new 175 * free space watcher. 176 */ 177 acctp = nd.ni_vp; 178 acctcred = crhold(td->td_ucred); 179 acctflags = flags; 180 callout_init(&acctwatch_callout, 0); 181 acctwatch(NULL); 182done2: 183 mtx_unlock(&acct_mtx); 184 mtx_unlock(&Giant); 185 return (error); 186} 187 188/* 189 * Write out process accounting information, on process exit. 190 * Data to be written out is specified in Leffler, et al. 191 * and are enumerated below. (They're also noted in the system 192 * "acct.h" header file.) 193 */ 194 195int 196acct_process(td) 197 struct thread *td; 198{ 199 struct proc *p = td->td_proc; 200 struct acct acct; 201 struct rusage *r; 202 struct timeval ut, st, tmp; 203 int t, ret; 204 struct vnode *vp; 205 struct ucred *uc; 206 207 mtx_lock(&acct_mtx); 208 209 /* If accounting isn't enabled, don't bother */ 210 vp = acctp; 211 if (vp == NULLVP) { 212 mtx_unlock(&acct_mtx); 213 return (0); 214 } 215 216 /* 217 * Get process accounting information. 218 */ 219 220 /* (1) The name of the command that ran */ 221 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm); 222 223 /* (2) The amount of user and system time that was used */ 224 mtx_lock_spin(&sched_lock); 225 calcru(p, &ut, &st, NULL); 226 mtx_unlock_spin(&sched_lock); 227 acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec); 228 acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec); 229 230 /* (3) The elapsed time the commmand ran (and its starting time) */ 231 acct.ac_btime = p->p_stats->p_start.tv_sec; 232 microtime(&tmp); 233 timevalsub(&tmp, &p->p_stats->p_start); 234 acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec); 235 236 /* (4) The average amount of memory used */ 237 r = &p->p_stats->p_ru; 238 tmp = ut; 239 timevaladd(&tmp, &st); 240 t = tmp.tv_sec * hz + tmp.tv_usec / tick; 241 if (t) 242 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t; 243 else 244 acct.ac_mem = 0; 245 246 /* (5) The number of disk I/O operations done */ 247 acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0); 248 249 /* (6) The UID and GID of the process */ 250 acct.ac_uid = p->p_ucred->cr_ruid; 251 acct.ac_gid = p->p_ucred->cr_rgid; 252 253 /* (7) The terminal from which the process was started */ 254 PROC_LOCK(p); 255 SESS_LOCK(p->p_session); 256 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp) 257 acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev); 258 else 259 acct.ac_tty = NOUDEV; 260 SESS_UNLOCK(p->p_session); 261 PROC_UNLOCK(p); 262 263 /* (8) The boolean flags that tell how the process terminated, etc. */ 264 acct.ac_flag = p->p_acflag; 265 266 /* 267 * Eliminate any file size rlimit. 268 */ 269 if (p->p_limit->p_refcnt > 1 && 270 (p->p_limit->p_lflags & PL_SHAREMOD) == 0) { 271 p->p_limit->p_refcnt--; 272 p->p_limit = limcopy(p->p_limit); 273 } 274 p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; 275 276 /* 277 * Write the accounting information to the file. 278 */ 279 VOP_LEASE(vp, td, acctcred, LEASE_WRITE); 280 uc = crhold(acctcred); 281 vref(vp); 282 mtx_unlock(&acct_mtx); 283 ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)&acct, sizeof (acct), 284 (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, uc, NOCRED, 285 (int *)0, td); 286 vrele(vp); 287 crfree(uc); 288 return (ret); 289} 290 291/* 292 * Encode_comp_t converts from ticks in seconds and microseconds 293 * to ticks in 1/AHZ seconds. The encoding is described in 294 * Leffler, et al., on page 63. 295 */ 296 297#define MANTSIZE 13 /* 13 bit mantissa. */ 298#define EXPSIZE 3 /* Base 8 (3 bit) exponent. */ 299#define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */ 300 301static comp_t 302encode_comp_t(s, us) 303 u_long s, us; 304{ 305 int exp, rnd; 306 307 exp = 0; 308 rnd = 0; 309 s *= AHZ; 310 s += us / (1000000 / AHZ); /* Maximize precision. */ 311 312 while (s > MAXFRACT) { 313 rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */ 314 s >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */ 315 exp++; 316 } 317 318 /* If we need to round up, do it (and handle overflow correctly). */ 319 if (rnd && (++s > MAXFRACT)) { 320 s >>= EXPSIZE; 321 exp++; 322 } 323 324 /* Clean it up and polish it off. */ 325 exp <<= MANTSIZE; /* Shift the exponent into place */ 326 exp += s; /* and add on the mantissa. */ 327 return (exp); 328} 329 330/* 331 * Periodically check the filesystem to see if accounting 332 * should be turned on or off. Beware the case where the vnode 333 * has been vgone()'d out from underneath us, e.g. when the file 334 * system containing the accounting file has been forcibly unmounted. 335 */ 336/* ARGSUSED */ 337static void 338acctwatch(a) 339 void *a; 340{ 341 struct statfs sb; 342 343 mtx_lock(&acct_mtx); 344 345 if (savacctp != NULLVP) { 346 if (savacctp->v_type == VBAD) { 347 (void) vn_close(savacctp, savacctflags, savacctcred, 348 NULL); 349 savacctp = NULLVP; 350 savacctcred = NOCRED; 351 mtx_unlock(&acct_mtx); 352 return; 353 } 354 (void)VFS_STATFS(savacctp->v_mount, &sb, (struct thread *)0); 355 if (sb.f_bavail > acctresume * sb.f_blocks / 100) { 356 acctp = savacctp; 357 acctcred = savacctcred; 358 acctflags = savacctflags; 359 savacctp = NULLVP; 360 savacctcred = NOCRED; 361 log(LOG_NOTICE, "Accounting resumed\n"); 362 } 363 } else { 364 if (acctp == NULLVP) 365 return; 366 if (acctp->v_type == VBAD) { 367 (void) vn_close(acctp, acctflags, acctcred, NULL); 368 acctp = NULLVP; 369 crfree(acctcred); 370 acctcred = NOCRED; 371 mtx_unlock(&acct_mtx); 372 return; 373 } 374 (void)VFS_STATFS(acctp->v_mount, &sb, (struct thread *)0); 375 if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) { 376 savacctp = acctp; 377 savacctflags = acctflags; 378 acctp = NULLVP; 379 acctcred = NOCRED; 380 log(LOG_NOTICE, "Accounting suspended\n"); 381 } 382 } 383 callout_reset(&acctwatch_callout, acctchkfreq * hz, acctwatch, NULL); 384 mtx_unlock(&acct_mtx); 385} 386